maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/lib/Target/Mips/MipsSEISelLowering.cpp

3797 lines
138 KiB
C++
Raw Normal View History

[Mips] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC). llvm-svn: 309993
2017-08-04 06:12:30 +08:00
//===- MipsSEISelLowering.cpp - MipsSE DAG Lowering Interface -------------===//
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// Subclass of MipsTargetLowering specialized for mips32/64.
//
//===----------------------------------------------------------------------===//
[Mips] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC). llvm-svn: 309993
2017-08-04 06:12:30 +08:00
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
#include "MipsSEISelLowering.h"
Reinstate "Nuke the old JIT." Approved by Jim Grosbach, Lang Hames, Rafael Espindola. This reinstates commits r215111, 215115, 215116, 215117, 215136. llvm-svn: 216982
2014-09-03 06:28:02 +08:00
#include "MipsMachineFunction.h"
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
#include "MipsRegisterInfo.h"
[Mips] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC). llvm-svn: 309993
2017-08-04 06:12:30 +08:00
#include "MipsSubtarget.h"
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
#include "llvm/ADT/APInt.h"
[Mips] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC). llvm-svn: 309993
2017-08-04 06:12:30 +08:00
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Triple.h"
#include "llvm/CodeGen/CallingConvLower.h"
#include "llvm/CodeGen/ISDOpcodes.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstr.h"
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
#include "llvm/CodeGen/MachineInstrBuilder.h"
[Mips] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC). llvm-svn: 309993
2017-08-04 06:12:30 +08:00
#include "llvm/CodeGen/MachineMemOperand.h"
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
#include "llvm/CodeGen/MachineRegisterInfo.h"
[Mips] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC). llvm-svn: 309993
2017-08-04 06:12:30 +08:00
#include "llvm/CodeGen/MachineValueType.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/SelectionDAGNodes.h"
Target/TargetInstrInfo.h -> CodeGen/TargetInstrInfo.h to match layering This header includes CodeGen headers, and is not, itself, included by any Target headers, so move it into CodeGen to match the layering of its implementation. llvm-svn: 317647
2017-11-08 09:01:31 +08:00
#include "llvm/CodeGen/TargetInstrInfo.h"
[Mips] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC). llvm-svn: 309993
2017-08-04 06:12:30 +08:00
#include "llvm/CodeGen/ValueTypes.h"
#include "llvm/IR/DebugLoc.h"
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
#include "llvm/IR/Intrinsics.h"
[Mips] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC). llvm-svn: 309993
2017-08-04 06:12:30 +08:00
#include "llvm/Support/Casting.h"
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
#include "llvm/Support/CommandLine.h"
[mips] MipsSETargetLowering now reports DAGCombiner changes when using -debug-only=mips-isel No test since -debug output is intended for developers and not end-users. llvm-svn: 193681
2013-10-30 21:31:27 +08:00
#include "llvm/Support/Debug.h"
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
#include "llvm/Support/ErrorHandling.h"
[Mips] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC). llvm-svn: 309993
2017-08-04 06:12:30 +08:00
#include "llvm/Support/MathExtras.h"
Add #include of raw_ostream.h to MipsSEISelLowering.cpp Fixing this Windows build error: ..\lib\Target\Mips\MipsSEISelLowering.cpp(997) : error C2027: use of undefined type 'llvm::raw_ostream' llvm-svn: 193696
2013-10-31 00:10:10 +08:00
#include "llvm/Support/raw_ostream.h"
[Mips] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC). llvm-svn: 309993
2017-08-04 06:12:30 +08:00
#include "llvm/Target/TargetSubtargetInfo.h"
#include <algorithm>
#include <cassert>
#include <cstdint>
#include <iterator>
#include <utility>
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
using namespace llvm;
[Modules] Fix potential ODR violations by sinking the DEBUG_TYPE definition below all of the header #include lines, lib/Target/... edition. llvm-svn: 206842
2014-04-22 10:41:26 +08:00
#define DEBUG_TYPE "mips-isel"
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
static cl::opt<bool>
[mips] Enable tail calls by default Enable tail calls by default for (micro)MIPS(64). microMIPS is slightly more tricky than doing it for MIPS(R6) or microMIPSR6. microMIPS has two instruction encodings: 16bit and 32bit along with some restrictions on the size of the instruction that can fill the delay slot. For safe tail calls for microMIPS, the delay slot filler attempts to find a correct size instruction for the delay slot of TAILCALL pseudos. Reviewers: dsanders, vkalintris Subscribers: jfb, dsanders, sdardis, llvm-commits Differential Revision: https://reviews.llvm.org/D21138 llvm-svn: 277708
2016-08-04 17:17:07 +08:00
UseMipsTailCalls("mips-tail-calls", cl::Hidden,
[mips] Disable tail calls temporarily Disable tail calls while the remaining bugs are fixed. Enable only for tests. Reviewers: vkalintiris Differential Review: https://reviews.llvm.org/D24912 llvm-svn: 282487
2016-09-27 21:15:54 +08:00
cl::desc("MIPS: permit tail calls."), cl::init(false));
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
[mips] Enhance command line option "-mno-ldc1-sdc1" to expand base+index double precision loads and stores as well as reg+imm double precision loads and stores. Previously, expansion of loads and stores was done after register allocation, but now it takes place during legalization. As a result, users will see double precision stores and loads being emitted to spill and restore 64-bit FP registers. llvm-svn: 190235
2013-09-07 08:52:30 +08:00
static cl::opt<bool> NoDPLoadStore("mno-ldc1-sdc1", cl::init(false),
cl::desc("Expand double precision loads and "
"stores to their single precision "
"counterparts"));
constify the TargetMachine being passed through the Mips subtarget creation. llvm-svn: 218169
2014-09-20 07:30:42 +08:00
MipsSETargetLowering::MipsSETargetLowering(const MipsTargetMachine &TM,
In preparation for replacing the whole subtarget on the target machine, have target lowering take the subtarget explicitly. llvm-svn: 213426
2014-07-19 07:25:04 +08:00
const MipsSubtarget &STI)
: MipsTargetLowering(TM, STI) {
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
// Set up the register classes
[mips] Rename register classes CPURegs and CPU64Regs. llvm-svn: 187832
2013-08-07 07:08:38 +08:00
addRegisterClass(MVT::i32, &Mips::GPR32RegClass);
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (Subtarget.isGP64bit())
[mips] Rename register classes CPURegs and CPU64Regs. llvm-svn: 187832
2013-08-07 07:08:38 +08:00
addRegisterClass(MVT::i64, &Mips::GPR64RegClass);
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (Subtarget.hasDSP() || Subtarget.hasMSA()) {
[mips][msa] Expand all truncstores and loadexts for MSA as well as DSP llvm-svn: 191496
2013-09-27 17:44:59 +08:00
// Expand all truncating stores and extending loads.
[CodeGen] Use MVT iterator_ranges in legality loops. NFC intended. A few loops do trickier things than just iterating on an MVT subset, so I'll leave them be for now. Follow-up of r225387. llvm-svn: 225392
2015-01-08 05:27:10 +08:00
for (MVT VT0 : MVT::vector_valuetypes()) {
[SelectionDAG] Allow targets to specify legality of extloads' result type (in addition to the memory type). The *LoadExt* legalization handling used to only have one type, the memory type. This forced users to assume that as long as the extload for the memory type was declared legal, and the result type was legal, the whole extload was legal. However, this isn't always the case. For instance, on X86, with AVX, this is legal: v4i32 load, zext from v4i8 but this isn't: v4i64 load, zext from v4i8 Whereas v4i64 is (arguably) legal, even without AVX2. Note that the same thing was done a while ago for truncstores (r46140), but I assume no one needed it yet for extloads, so here we go. Calls to getLoadExtAction were changed to add the value type, found manually in the surrounding code. Calls to setLoadExtAction were mechanically changed, by wrapping the call in a loop, to match previous behavior. The loop iterates over the MVT subrange corresponding to the memory type (FP vectors, etc...). I also pulled neighboring setTruncStoreActions into some of the loops; those shouldn't make a difference, as the additional types are illegal. (e.g., i128->i1 truncstores on PPC.) No functional change intended. Differential Revision: http://reviews.llvm.org/D6532 llvm-svn: 225421
2015-01-08 08:51:32 +08:00
for (MVT VT1 : MVT::vector_valuetypes()) {
[CodeGen] Use MVT iterator_ranges in legality loops. NFC intended. A few loops do trickier things than just iterating on an MVT subset, so I'll leave them be for now. Follow-up of r225387. llvm-svn: 225392
2015-01-08 05:27:10 +08:00
setTruncStoreAction(VT0, VT1, Expand);
[SelectionDAG] Allow targets to specify legality of extloads' result type (in addition to the memory type). The *LoadExt* legalization handling used to only have one type, the memory type. This forced users to assume that as long as the extload for the memory type was declared legal, and the result type was legal, the whole extload was legal. However, this isn't always the case. For instance, on X86, with AVX, this is legal: v4i32 load, zext from v4i8 but this isn't: v4i64 load, zext from v4i8 Whereas v4i64 is (arguably) legal, even without AVX2. Note that the same thing was done a while ago for truncstores (r46140), but I assume no one needed it yet for extloads, so here we go. Calls to getLoadExtAction were changed to add the value type, found manually in the surrounding code. Calls to setLoadExtAction were mechanically changed, by wrapping the call in a loop, to match previous behavior. The loop iterates over the MVT subrange corresponding to the memory type (FP vectors, etc...). I also pulled neighboring setTruncStoreActions into some of the loops; those shouldn't make a difference, as the additional types are illegal. (e.g., i128->i1 truncstores on PPC.) No functional change intended. Differential Revision: http://reviews.llvm.org/D6532 llvm-svn: 225421
2015-01-08 08:51:32 +08:00
setLoadExtAction(ISD::SEXTLOAD, VT0, VT1, Expand);
setLoadExtAction(ISD::ZEXTLOAD, VT0, VT1, Expand);
setLoadExtAction(ISD::EXTLOAD, VT0, VT1, Expand);
}
[mips][msa] Expand all truncstores and loadexts for MSA as well as DSP llvm-svn: 191496
2013-09-27 17:44:59 +08:00
}
}
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (Subtarget.hasDSP()) {
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
MVT::SimpleValueType VecTys[2] = {MVT::v2i16, MVT::v4i8};
for (unsigned i = 0; i < array_lengthof(VecTys); ++i) {
[mips] Rename DSPRegs. llvm-svn: 188342
2013-08-14 08:53:38 +08:00
addRegisterClass(VecTys[i], &Mips::DSPRRegClass);
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
// Expand all builtin opcodes.
for (unsigned Opc = 0; Opc < ISD::BUILTIN_OP_END; ++Opc)
setOperationAction(Opc, VecTys[i], Expand);
[mips] Reapply r179420 and r179421. llvm-svn: 179434
2013-04-13 08:55:41 +08:00
setOperationAction(ISD::ADD, VecTys[i], Legal);
setOperationAction(ISD::SUB, VecTys[i], Legal);
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
setOperationAction(ISD::LOAD, VecTys[i], Legal);
setOperationAction(ISD::STORE, VecTys[i], Legal);
setOperationAction(ISD::BITCAST, VecTys[i], Legal);
}
[mips] Instruction selection patterns for DSP-ASE vector shifts. llvm-svn: 179906
2013-04-20 07:21:32 +08:00
setTargetDAGCombine(ISD::SHL);
setTargetDAGCombine(ISD::SRA);
setTargetDAGCombine(ISD::SRL);
[mips] Instruction selection patterns for DSP-ASE vector select and compare instructions. llvm-svn: 180820
2013-05-01 06:37:26 +08:00
setTargetDAGCombine(ISD::SETCC);
setTargetDAGCombine(ISD::VSELECT);
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
}
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (Subtarget.hasDSPR2())
[mips] Reapply r179420 and r179421. llvm-svn: 179434
2013-04-13 08:55:41 +08:00
setOperationAction(ISD::MUL, MVT::v2i16, Legal);
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (Subtarget.hasMSA()) {
[mips][msa] Separate the configuration of int/float vector types since they will diverge soon No functional change llvm-svn: 190506
2013-09-11 18:15:48 +08:00
addMSAIntType(MVT::v16i8, &Mips::MSA128BRegClass);
addMSAIntType(MVT::v8i16, &Mips::MSA128HRegClass);
addMSAIntType(MVT::v4i32, &Mips::MSA128WRegClass);
addMSAIntType(MVT::v2i64, &Mips::MSA128DRegClass);
addMSAFloatType(MVT::v8f16, &Mips::MSA128HRegClass);
addMSAFloatType(MVT::v4f32, &Mips::MSA128WRegClass);
addMSAFloatType(MVT::v2f64, &Mips::MSA128DRegClass);
[mips][msa] Added support for matching nor from normal IR (i.e. not intrinsics) llvm-svn: 191195
2013-09-23 21:22:24 +08:00
[mips][msa] Implement f16 support The MIPS MSA ASE provides instructions to convert to and from half precision floating point. This patch teaches the MIPS backend to treat f16 as a legal type and how to promote such values to f32 for the usual set of operations. As a result of this, the fexup[lr].w intrinsics no longer crash LLVM during type legalization. Reviewers: zoran.jovanvoic, vkalintiris Differential Revision: https://reviews.llvm.org/D26398 llvm-svn: 287349
2016-11-19 00:17:44 +08:00
// f16 is a storage-only type, always promote it to f32.
addRegisterClass(MVT::f16, &Mips::MSA128HRegClass);
setOperationAction(ISD::SETCC, MVT::f16, Promote);
setOperationAction(ISD::BR_CC, MVT::f16, Promote);
setOperationAction(ISD::SELECT_CC, MVT::f16, Promote);
setOperationAction(ISD::SELECT, MVT::f16, Promote);
setOperationAction(ISD::FADD, MVT::f16, Promote);
setOperationAction(ISD::FSUB, MVT::f16, Promote);
setOperationAction(ISD::FMUL, MVT::f16, Promote);
setOperationAction(ISD::FDIV, MVT::f16, Promote);
setOperationAction(ISD::FREM, MVT::f16, Promote);
setOperationAction(ISD::FMA, MVT::f16, Promote);
setOperationAction(ISD::FNEG, MVT::f16, Promote);
setOperationAction(ISD::FABS, MVT::f16, Promote);
setOperationAction(ISD::FCEIL, MVT::f16, Promote);
setOperationAction(ISD::FCOPYSIGN, MVT::f16, Promote);
setOperationAction(ISD::FCOS, MVT::f16, Promote);
setOperationAction(ISD::FP_EXTEND, MVT::f16, Promote);
setOperationAction(ISD::FFLOOR, MVT::f16, Promote);
setOperationAction(ISD::FNEARBYINT, MVT::f16, Promote);
setOperationAction(ISD::FPOW, MVT::f16, Promote);
setOperationAction(ISD::FPOWI, MVT::f16, Promote);
setOperationAction(ISD::FRINT, MVT::f16, Promote);
setOperationAction(ISD::FSIN, MVT::f16, Promote);
setOperationAction(ISD::FSINCOS, MVT::f16, Promote);
setOperationAction(ISD::FSQRT, MVT::f16, Promote);
setOperationAction(ISD::FEXP, MVT::f16, Promote);
setOperationAction(ISD::FEXP2, MVT::f16, Promote);
setOperationAction(ISD::FLOG, MVT::f16, Promote);
setOperationAction(ISD::FLOG2, MVT::f16, Promote);
setOperationAction(ISD::FLOG10, MVT::f16, Promote);
setOperationAction(ISD::FROUND, MVT::f16, Promote);
setOperationAction(ISD::FTRUNC, MVT::f16, Promote);
setOperationAction(ISD::FMINNUM, MVT::f16, Promote);
setOperationAction(ISD::FMAXNUM, MVT::f16, Promote);
setOperationAction(ISD::FMINNAN, MVT::f16, Promote);
setOperationAction(ISD::FMAXNAN, MVT::f16, Promote);
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
setTargetDAGCombine(ISD::AND);
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
setTargetDAGCombine(ISD::OR);
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
setTargetDAGCombine(ISD::SRA);
[mips][msa] Added support for matching bsel and bseli from normal IR (i.e. not intrinsics) This required correcting the definition of the bsel and bseli intrinsics. llvm-svn: 191290
2013-09-24 20:04:44 +08:00
setTargetDAGCombine(ISD::VSELECT);
[mips][msa] Added support for matching nor from normal IR (i.e. not intrinsics) llvm-svn: 191195
2013-09-23 21:22:24 +08:00
setTargetDAGCombine(ISD::XOR);
[Mips][msa] Added initial MSA support. * msa SubtargetFeature * registers * ld.[bhwd], and st.[bhwd] instructions Does not correctly prohibit use of both 32-bit FPU registers and MSA together. Patch by Daniel Sanders llvm-svn: 188313
2013-08-14 04:54:07 +08:00
}
Rename the MIPS routine abiUsesSoftFloat -> useSoftFloat to match some incoming changes and the general scheme used by features (use/has). llvm-svn: 236794
2015-05-08 07:10:21 +08:00
if (!Subtarget.useSoftFloat()) {
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
addRegisterClass(MVT::f32, &Mips::FGR32RegClass);
// When dealing with single precision only, use libcalls
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (!Subtarget.isSingleFloat()) {
if (Subtarget.isFP64bit())
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
addRegisterClass(MVT::f64, &Mips::FGR64RegClass);
else
addRegisterClass(MVT::f64, &Mips::AFGR64RegClass);
}
}
[mips] Fix definitions of multiply, multiply-add/sub and divide instructions. The new instructions have explicit register output operands and use table-gen patterns instead of C++ code to do instruction selection. Mips16's instructions are unaffected by this change. llvm-svn: 178403
2013-03-30 09:36:35 +08:00
setOperationAction(ISD::SMUL_LOHI, MVT::i32, Custom);
setOperationAction(ISD::UMUL_LOHI, MVT::i32, Custom);
setOperationAction(ISD::MULHS, MVT::i32, Custom);
setOperationAction(ISD::MULHU, MVT::i32, Custom);
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (Subtarget.hasCnMips())
[MIPS] Add cpu octeon and some instructions The Octeon cpu from Cavium Networks is mips64r2 based and has an extended instruction set. In order to utilize this with LLVM, a new cpu feature "octeon" and a subtarget feature "cnmips" is added. A small set of new instructions (baddu, dmul, pop, dpop, seq, sne) is also added. LLVM generates dmul, pop and dpop instructions with option -mcpu=octeon or -mattr=+cnmips. llvm-svn: 204337
2014-03-20 19:51:58 +08:00
setOperationAction(ISD::MUL, MVT::i64, Legal);
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
else if (Subtarget.isGP64bit())
[MIPS] Add cpu octeon and some instructions The Octeon cpu from Cavium Networks is mips64r2 based and has an extended instruction set. In order to utilize this with LLVM, a new cpu feature "octeon" and a subtarget feature "cnmips" is added. A small set of new instructions (baddu, dmul, pop, dpop, seq, sne) is also added. LLVM generates dmul, pop and dpop instructions with option -mcpu=octeon or -mattr=+cnmips. llvm-svn: 204337
2014-03-20 19:51:58 +08:00
setOperationAction(ISD::MUL, MVT::i64, Custom);
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (Subtarget.isGP64bit()) {
[mips] Enable arithmetic and binary operations for the i128 data type. Summary: This patch adds support for some operations that were missing from 128-bit integer types (add/sub/mul/sdiv/udiv... etc.). With these changes we can support the __int128_t and __uint128_t data types from C/C++. Depends on D7125 Reviewers: dsanders Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D7143 llvm-svn: 227089
2015-01-26 20:33:22 +08:00
setOperationAction(ISD::SMUL_LOHI, MVT::i64, Custom);
setOperationAction(ISD::UMUL_LOHI, MVT::i64, Custom);
[mips] Custom-lower i64 MULHS and MULHU nodes. Remove the code which selects multiply instructions in MipsSEDAGToDAGISel. This patch was supposed to be part of r178403. llvm-svn: 179314
2013-04-12 03:29:26 +08:00
setOperationAction(ISD::MULHS, MVT::i64, Custom);
setOperationAction(ISD::MULHU, MVT::i64, Custom);
Mips: Only set divrem i64 to custom on 64bit Reviewed-by: Daniel Sanders <daniel.sanders@imgtec.com> Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu> llvm-svn: 220046
2014-10-17 22:45:28 +08:00
setOperationAction(ISD::SDIVREM, MVT::i64, Custom);
setOperationAction(ISD::UDIVREM, MVT::i64, Custom);
[mips] Custom-lower i64 MULHS and MULHU nodes. Remove the code which selects multiply instructions in MipsSEDAGToDAGISel. This patch was supposed to be part of r178403. llvm-svn: 179314
2013-04-12 03:29:26 +08:00
}
[mips] Fix definitions of multiply, multiply-add/sub and divide instructions. The new instructions have explicit register output operands and use table-gen patterns instead of C++ code to do instruction selection. Mips16's instructions are unaffected by this change. llvm-svn: 178403
2013-03-30 09:36:35 +08:00
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::i64, Custom);
setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::i64, Custom);
[mips] Fix definitions of multiply, multiply-add/sub and divide instructions. The new instructions have explicit register output operands and use table-gen patterns instead of C++ code to do instruction selection. Mips16's instructions are unaffected by this change. llvm-svn: 178403
2013-03-30 09:36:35 +08:00
setOperationAction(ISD::SDIVREM, MVT::i32, Custom);
setOperationAction(ISD::UDIVREM, MVT::i32, Custom);
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Custom);
setOperationAction(ISD::LOAD, MVT::i32, Custom);
setOperationAction(ISD::STORE, MVT::i32, Custom);
[mips] Improve code generation for constant multiplication using shifts, adds and subs. llvm-svn: 185011
2013-06-27 02:48:17 +08:00
setTargetDAGCombine(ISD::MUL);
[mips] Move the code which does dag-combine for multiply-add/sub nodes to derived class MipsSETargetLowering. We shouldn't be generating madd/msub nodes if target is Mips16, since Mips16 doesn't have support for multipy-add/sub instructions. llvm-svn: 178404
2013-03-30 09:42:24 +08:00
[mips][msa] Added bnz.df, bnz.v, bz.df, and bz.v These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes, are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as a branch/mov sequence to evaluate to 0 or 1. Note: The resulting code is sub-optimal since it doesnt seem to be possible to feed the result of an intrinsic directly into a brcond. At the moment it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily evaluates the boolean twice. llvm-svn: 189478
2013-08-28 20:14:50 +08:00
setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
[mips][msa] Added load/store intrinsics. llvm-svn: 189476
2013-08-28 20:04:29 +08:00
setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::Other, Custom);
setOperationAction(ISD::INTRINSIC_VOID, MVT::Other, Custom);
[mips] Enhance command line option "-mno-ldc1-sdc1" to expand base+index double precision loads and stores as well as reg+imm double precision loads and stores. Previously, expansion of loads and stores was done after register allocation, but now it takes place during legalization. As a result, users will see double precision stores and loads being emitted to spill and restore 64-bit FP registers. llvm-svn: 190235
2013-09-07 08:52:30 +08:00
if (NoDPLoadStore) {
setOperationAction(ISD::LOAD, MVT::f64, Custom);
setOperationAction(ISD::STORE, MVT::f64, Custom);
}
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (Subtarget.hasMips32r6()) {
[mips][mips64r6] Replace m[tf]hi, m[tf]lo, mult, multu, dmult, dmultu, div, ddiv, divu, ddivu for MIPS32r6/MIPS64. Summary: The accumulator-based (HI/LO) multiplies and divides from earlier ISA's have been removed and replaced with GPR-based equivalents. For example: div $1, $2 mflo $3 is now: div $3, $1, $2 This patch disables the accumulator-based multiplies and divides for MIPS32r6/MIPS64r6 and uses the GPR-based equivalents instead. Renamed expandPseudoDiv to insertDivByZeroTrap to better describe the behaviour of the function. MipsDelaySlotFiller now invalidates the liveness information when moving instructions to the delay slot. Without this, divrem.ll will abort since %GP ends up used before it is defined. Reviewers: vmedic, zoran.jovanovic, jkolek Reviewed By: jkolek Differential Revision: http://reviews.llvm.org/D3896 llvm-svn: 210760
2014-06-12 18:44:10 +08:00
// MIPS32r6 replaces the accumulator-based multiplies with a three register
// instruction
[mips][mips64r6] madd/maddu/msub/msubu are not available on MIPS32r6/MIPS64r6 Summary: This patch disables madd/maddu/msub/msubu in both the assembler and code generator. Depends on D3896 Reviewers: jkolek, zoran.jovanovic, vmedic Reviewed By: vmedic Differential Revision: http://reviews.llvm.org/D3955 llvm-svn: 210762
2014-06-12 18:54:16 +08:00
setOperationAction(ISD::SMUL_LOHI, MVT::i32, Expand);
setOperationAction(ISD::UMUL_LOHI, MVT::i32, Expand);
[mips][mips64r6] Replace m[tf]hi, m[tf]lo, mult, multu, dmult, dmultu, div, ddiv, divu, ddivu for MIPS32r6/MIPS64. Summary: The accumulator-based (HI/LO) multiplies and divides from earlier ISA's have been removed and replaced with GPR-based equivalents. For example: div $1, $2 mflo $3 is now: div $3, $1, $2 This patch disables the accumulator-based multiplies and divides for MIPS32r6/MIPS64r6 and uses the GPR-based equivalents instead. Renamed expandPseudoDiv to insertDivByZeroTrap to better describe the behaviour of the function. MipsDelaySlotFiller now invalidates the liveness information when moving instructions to the delay slot. Without this, divrem.ll will abort since %GP ends up used before it is defined. Reviewers: vmedic, zoran.jovanovic, jkolek Reviewed By: jkolek Differential Revision: http://reviews.llvm.org/D3896 llvm-svn: 210760
2014-06-12 18:44:10 +08:00
setOperationAction(ISD::MUL, MVT::i32, Legal);
setOperationAction(ISD::MULHS, MVT::i32, Legal);
setOperationAction(ISD::MULHU, MVT::i32, Legal);
// MIPS32r6 replaces the accumulator-based division/remainder with separate
// three register division and remainder instructions.
setOperationAction(ISD::SDIVREM, MVT::i32, Expand);
setOperationAction(ISD::UDIVREM, MVT::i32, Expand);
setOperationAction(ISD::SDIV, MVT::i32, Legal);
setOperationAction(ISD::UDIV, MVT::i32, Legal);
setOperationAction(ISD::SREM, MVT::i32, Legal);
setOperationAction(ISD::UREM, MVT::i32, Legal);
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
// MIPS32r6 replaces conditional moves with an equivalent that removes the
// need for three GPR read ports.
setOperationAction(ISD::SETCC, MVT::i32, Legal);
setOperationAction(ISD::SELECT, MVT::i32, Legal);
setOperationAction(ISD::SELECT_CC, MVT::i32, Expand);
setOperationAction(ISD::SETCC, MVT::f32, Legal);
setOperationAction(ISD::SELECT, MVT::f32, Legal);
setOperationAction(ISD::SELECT_CC, MVT::f32, Expand);
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
assert(Subtarget.isFP64bit() && "FR=1 is required for MIPS32r6");
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
setOperationAction(ISD::SETCC, MVT::f64, Legal);
Reland r308585 Builder clang-x86_64-linux-abi-test apparently failed due to a spurious error unrelated to the changes r308585 introduced. llvm-svn: 308612
2017-07-20 21:08:18 +08:00
setOperationAction(ISD::SELECT, MVT::f64, Custom);
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
setOperationAction(ISD::SELECT_CC, MVT::f64, Expand);
[mips][mips64r6] bc1[tf] are not available on MIPS32r6/MIPS64r6 Summary: Also tightened up the acceptable condition operand for these instructions on MIPS-I to MIPS-III. Support for $fcc[1-7] was added in MIPS-IV. Prior to that only $fcc0 is acceptable. We currently don't optimize (BEQZ (NOT $a), $target) and similar. It's probably best to do this in InstCombine. Depends on D4111 Reviewers: jkolek, zoran.jovanovic, vmedic Reviewed By: vmedic Differential Revision: http://reviews.llvm.org/D4112 llvm-svn: 210787
2014-06-12 23:00:17 +08:00
setOperationAction(ISD::BRCOND, MVT::Other, Legal);
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
// Floating point > and >= are supported via < and <=
setCondCodeAction(ISD::SETOGE, MVT::f32, Expand);
setCondCodeAction(ISD::SETOGT, MVT::f32, Expand);
setCondCodeAction(ISD::SETUGE, MVT::f32, Expand);
setCondCodeAction(ISD::SETUGT, MVT::f32, Expand);
setCondCodeAction(ISD::SETOGE, MVT::f64, Expand);
setCondCodeAction(ISD::SETOGT, MVT::f64, Expand);
setCondCodeAction(ISD::SETUGE, MVT::f64, Expand);
setCondCodeAction(ISD::SETUGT, MVT::f64, Expand);
[mips][mips64r6] Replace m[tf]hi, m[tf]lo, mult, multu, dmult, dmultu, div, ddiv, divu, ddivu for MIPS32r6/MIPS64. Summary: The accumulator-based (HI/LO) multiplies and divides from earlier ISA's have been removed and replaced with GPR-based equivalents. For example: div $1, $2 mflo $3 is now: div $3, $1, $2 This patch disables the accumulator-based multiplies and divides for MIPS32r6/MIPS64r6 and uses the GPR-based equivalents instead. Renamed expandPseudoDiv to insertDivByZeroTrap to better describe the behaviour of the function. MipsDelaySlotFiller now invalidates the liveness information when moving instructions to the delay slot. Without this, divrem.ll will abort since %GP ends up used before it is defined. Reviewers: vmedic, zoran.jovanovic, jkolek Reviewed By: jkolek Differential Revision: http://reviews.llvm.org/D3896 llvm-svn: 210760
2014-06-12 18:44:10 +08:00
}
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (Subtarget.hasMips64r6()) {
[mips][mips64r6] Replace m[tf]hi, m[tf]lo, mult, multu, dmult, dmultu, div, ddiv, divu, ddivu for MIPS32r6/MIPS64. Summary: The accumulator-based (HI/LO) multiplies and divides from earlier ISA's have been removed and replaced with GPR-based equivalents. For example: div $1, $2 mflo $3 is now: div $3, $1, $2 This patch disables the accumulator-based multiplies and divides for MIPS32r6/MIPS64r6 and uses the GPR-based equivalents instead. Renamed expandPseudoDiv to insertDivByZeroTrap to better describe the behaviour of the function. MipsDelaySlotFiller now invalidates the liveness information when moving instructions to the delay slot. Without this, divrem.ll will abort since %GP ends up used before it is defined. Reviewers: vmedic, zoran.jovanovic, jkolek Reviewed By: jkolek Differential Revision: http://reviews.llvm.org/D3896 llvm-svn: 210760
2014-06-12 18:44:10 +08:00
// MIPS64r6 replaces the accumulator-based multiplies with a three register
// instruction
[mips] Enable arithmetic and binary operations for the i128 data type. Summary: This patch adds support for some operations that were missing from 128-bit integer types (add/sub/mul/sdiv/udiv... etc.). With these changes we can support the __int128_t and __uint128_t data types from C/C++. Depends on D7125 Reviewers: dsanders Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D7143 llvm-svn: 227089
2015-01-26 20:33:22 +08:00
setOperationAction(ISD::SMUL_LOHI, MVT::i64, Expand);
setOperationAction(ISD::UMUL_LOHI, MVT::i64, Expand);
[mips][mips64r6] Replace m[tf]hi, m[tf]lo, mult, multu, dmult, dmultu, div, ddiv, divu, ddivu for MIPS32r6/MIPS64. Summary: The accumulator-based (HI/LO) multiplies and divides from earlier ISA's have been removed and replaced with GPR-based equivalents. For example: div $1, $2 mflo $3 is now: div $3, $1, $2 This patch disables the accumulator-based multiplies and divides for MIPS32r6/MIPS64r6 and uses the GPR-based equivalents instead. Renamed expandPseudoDiv to insertDivByZeroTrap to better describe the behaviour of the function. MipsDelaySlotFiller now invalidates the liveness information when moving instructions to the delay slot. Without this, divrem.ll will abort since %GP ends up used before it is defined. Reviewers: vmedic, zoran.jovanovic, jkolek Reviewed By: jkolek Differential Revision: http://reviews.llvm.org/D3896 llvm-svn: 210760
2014-06-12 18:44:10 +08:00
setOperationAction(ISD::MUL, MVT::i64, Legal);
setOperationAction(ISD::MULHS, MVT::i64, Legal);
setOperationAction(ISD::MULHU, MVT::i64, Legal);
// MIPS32r6 replaces the accumulator-based division/remainder with separate
// three register division and remainder instructions.
setOperationAction(ISD::SDIVREM, MVT::i64, Expand);
setOperationAction(ISD::UDIVREM, MVT::i64, Expand);
setOperationAction(ISD::SDIV, MVT::i64, Legal);
setOperationAction(ISD::UDIV, MVT::i64, Legal);
setOperationAction(ISD::SREM, MVT::i64, Legal);
setOperationAction(ISD::UREM, MVT::i64, Legal);
[mips][mips64r6] c.cond.fmt, mov[fntz], and mov[fntz].[ds] are not available on MIPS32r6/MIPS64r6 Summary: c.cond.fmt has been replaced by cmp.cond.fmt. Where c.cond.fmt wrote to dedicated condition registers, cmp.cond.fmt writes 1 or 0 to normal FGR's (like the GPR comparisons). mov[fntz] have been replaced by seleqz and selnez. These instructions conditionally zero a register based on a bool in a GPR. The results can then be or'd together to act as a select without, for example, requiring a third register read port. mov[fntz].[ds] have been replaced with sel.[ds] MIPS64r6 currently generates unnecessary sign-extensions for most selects. This is because the result of a SETCC is currently an i32. Bits 32-63 are undefined in i32 and the behaviour of seleqz/selnez would otherwise depend on undefined bits. Later, we will fix this by making the result of SETCC an i64 on MIPS64 targets. Depends on D3958 Reviewers: jkolek, vmedic, zoran.jovanovic Reviewed By: vmedic, zoran.jovanovic Differential Revision: http://reviews.llvm.org/D4003 llvm-svn: 210777
2014-06-12 21:39:06 +08:00
// MIPS64r6 replaces conditional moves with an equivalent that removes the
// need for three GPR read ports.
setOperationAction(ISD::SETCC, MVT::i64, Legal);
setOperationAction(ISD::SELECT, MVT::i64, Legal);
setOperationAction(ISD::SELECT_CC, MVT::i64, Expand);
[mips][mips64r6] Replace m[tf]hi, m[tf]lo, mult, multu, dmult, dmultu, div, ddiv, divu, ddivu for MIPS32r6/MIPS64. Summary: The accumulator-based (HI/LO) multiplies and divides from earlier ISA's have been removed and replaced with GPR-based equivalents. For example: div $1, $2 mflo $3 is now: div $3, $1, $2 This patch disables the accumulator-based multiplies and divides for MIPS32r6/MIPS64r6 and uses the GPR-based equivalents instead. Renamed expandPseudoDiv to insertDivByZeroTrap to better describe the behaviour of the function. MipsDelaySlotFiller now invalidates the liveness information when moving instructions to the delay slot. Without this, divrem.ll will abort since %GP ends up used before it is defined. Reviewers: vmedic, zoran.jovanovic, jkolek Reviewed By: jkolek Differential Revision: http://reviews.llvm.org/D3896 llvm-svn: 210760
2014-06-12 18:44:10 +08:00
}
Remove an argument-less call to getSubtargetImpl from TargetLoweringBase. This required plumbing a TargetRegisterInfo through computeRegisterProperties and into findRepresentativeClass which uses it for register class iteration. This required passing a subtarget into a few target specific initializations of TargetLowering. llvm-svn: 230583
2015-02-26 08:00:24 +08:00
computeRegisterProperties(Subtarget.getRegisterInfo());
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
}
const MipsTargetLowering *
constify the TargetMachine being passed through the Mips subtarget creation. llvm-svn: 218169
2014-09-20 07:30:42 +08:00
llvm::createMipsSETargetLowering(const MipsTargetMachine &TM,
In preparation for replacing the whole subtarget on the target machine, have target lowering take the subtarget explicitly. llvm-svn: 213426
2014-07-19 07:25:04 +08:00
const MipsSubtarget &STI) {
return new MipsSETargetLowering(TM, STI);
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
}
So that we can include target lowering in the subtarget, remove include circular dependency with the subtarget by inlining accessor methods and outlining a routine. llvm-svn: 212234
2014-07-03 07:18:40 +08:00
const TargetRegisterClass *
MipsSETargetLowering::getRepRegClassFor(MVT VT) const {
if (VT == MVT::Untyped)
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
return Subtarget.hasDSP() ? &Mips::ACC64DSPRegClass : &Mips::ACC64RegClass;
So that we can include target lowering in the subtarget, remove include circular dependency with the subtarget by inlining accessor methods and outlining a routine. llvm-svn: 212234
2014-07-03 07:18:40 +08:00
return TargetLowering::getRepRegClassFor(VT);
}
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
// Enable MSA support for the given integer type and Register class.
[mips][msa] Split MSA128 regset into size-specific sets containing the same registers. llvm-svn: 189095
2013-08-23 18:10:13 +08:00
void MipsSETargetLowering::
[mips][msa] Separate the configuration of int/float vector types since they will diverge soon No functional change llvm-svn: 190506
2013-09-11 18:15:48 +08:00
addMSAIntType(MVT::SimpleValueType Ty, const TargetRegisterClass *RC) {
addRegisterClass(Ty, RC);
// Expand all builtin opcodes.
for (unsigned Opc = 0; Opc < ISD::BUILTIN_OP_END; ++Opc)
setOperationAction(Opc, Ty, Expand);
setOperationAction(ISD::BITCAST, Ty, Legal);
setOperationAction(ISD::LOAD, Ty, Legal);
setOperationAction(ISD::STORE, Ty, Legal);
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
setOperationAction(ISD::EXTRACT_VECTOR_ELT, Ty, Custom);
setOperationAction(ISD::INSERT_VECTOR_ELT, Ty, Legal);
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
setOperationAction(ISD::BUILD_VECTOR, Ty, Custom);
[mips][msa] Separate the configuration of int/float vector types since they will diverge soon No functional change llvm-svn: 190506
2013-09-11 18:15:48 +08:00
[mips][msa] Added support for matching addv from normal IR (i.e. not intrinsics) The corresponding intrinsic is now lowered into equivalent IR (ISD::ADD) before instruction selection. llvm-svn: 190507
2013-09-11 18:28:16 +08:00
setOperationAction(ISD::ADD, Ty, Legal);
[mips][msa] Added support for matching and, or, and xor from normal IR (i.e. not intrinsics) llvm-svn: 191194
2013-09-23 20:57:42 +08:00
setOperationAction(ISD::AND, Ty, Legal);
[mips][msa] Added support for matching mulv, nlzc, sll, sra, srl, and subv from normal IR (i.e. not intrinsics) llvm-svn: 190518
2013-09-11 19:58:30 +08:00
setOperationAction(ISD::CTLZ, Ty, Legal);
[mips][msa] Added support for matching pcnt from normal IR (i.e. not intrinsics) llvm-svn: 191198
2013-09-23 21:40:21 +08:00
setOperationAction(ISD::CTPOP, Ty, Legal);
[mips][msa] Added support for matching mulv, nlzc, sll, sra, srl, and subv from normal IR (i.e. not intrinsics) llvm-svn: 190518
2013-09-11 19:58:30 +08:00
setOperationAction(ISD::MUL, Ty, Legal);
[mips][msa] Added support for matching and, or, and xor from normal IR (i.e. not intrinsics) llvm-svn: 191194
2013-09-23 20:57:42 +08:00
setOperationAction(ISD::OR, Ty, Legal);
[mips][msa] Added support for matching div_[su] from normal IR (i.e. not intrinsics) llvm-svn: 190509
2013-09-11 18:38:58 +08:00
setOperationAction(ISD::SDIV, Ty, Legal);
[mips][msa] Added support for matching mod_[us] from normal IR (i.e. not intrinsics) llvm-svn: 191737
2013-10-01 18:22:35 +08:00
setOperationAction(ISD::SREM, Ty, Legal);
[mips][msa] Added support for matching mulv, nlzc, sll, sra, srl, and subv from normal IR (i.e. not intrinsics) llvm-svn: 190518
2013-09-11 19:58:30 +08:00
setOperationAction(ISD::SHL, Ty, Legal);
setOperationAction(ISD::SRA, Ty, Legal);
setOperationAction(ISD::SRL, Ty, Legal);
setOperationAction(ISD::SUB, Ty, Legal);
[mips][msa] Added support for matching div_[su] from normal IR (i.e. not intrinsics) llvm-svn: 190509
2013-09-11 18:38:58 +08:00
setOperationAction(ISD::UDIV, Ty, Legal);
[mips][msa] Added support for matching mod_[us] from normal IR (i.e. not intrinsics) llvm-svn: 191737
2013-10-01 18:22:35 +08:00
setOperationAction(ISD::UREM, Ty, Legal);
[mips][msa] Added support for matching vshf from normal IR (i.e. not intrinsics) llvm-svn: 191301
2013-09-24 22:02:15 +08:00
setOperationAction(ISD::VECTOR_SHUFFLE, Ty, Custom);
[mips][msa] Added support for matching bsel and bseli from normal IR (i.e. not intrinsics) This required correcting the definition of the bsel and bseli intrinsics. llvm-svn: 191290
2013-09-24 20:04:44 +08:00
setOperationAction(ISD::VSELECT, Ty, Legal);
[mips][msa] Added support for matching and, or, and xor from normal IR (i.e. not intrinsics) llvm-svn: 191194
2013-09-23 20:57:42 +08:00
setOperationAction(ISD::XOR, Ty, Legal);
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
[mips][msa] Added support for matching ffint_[us].[wd], and ftrunc_[us].[wd] from normal IR (i.e. not intrinsics) llvm-svn: 192429
2013-10-11 18:00:06 +08:00
if (Ty == MVT::v4i32 || Ty == MVT::v2i64) {
setOperationAction(ISD::FP_TO_SINT, Ty, Legal);
setOperationAction(ISD::FP_TO_UINT, Ty, Legal);
setOperationAction(ISD::SINT_TO_FP, Ty, Legal);
setOperationAction(ISD::UINT_TO_FP, Ty, Legal);
}
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
setOperationAction(ISD::SETCC, Ty, Legal);
setCondCodeAction(ISD::SETNE, Ty, Expand);
setCondCodeAction(ISD::SETGE, Ty, Expand);
setCondCodeAction(ISD::SETGT, Ty, Expand);
setCondCodeAction(ISD::SETUGE, Ty, Expand);
setCondCodeAction(ISD::SETUGT, Ty, Expand);
[mips][msa] Separate the configuration of int/float vector types since they will diverge soon No functional change llvm-svn: 190506
2013-09-11 18:15:48 +08:00
}
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
// Enable MSA support for the given floating-point type and Register class.
[mips][msa] Separate the configuration of int/float vector types since they will diverge soon No functional change llvm-svn: 190506
2013-09-11 18:15:48 +08:00
void MipsSETargetLowering::
addMSAFloatType(MVT::SimpleValueType Ty, const TargetRegisterClass *RC) {
[mips][msa] Split MSA128 regset into size-specific sets containing the same registers. llvm-svn: 189095
2013-08-23 18:10:13 +08:00
addRegisterClass(Ty, RC);
[Mips][msa] Added the simple builtins (add_a to dpsub[su], ilvev to ldi) Includes: add_a, adds_[asu], addv, addvi, andi.b, asub_[su].[bhwd], aver?_[su]_[bhwd], bclr, bclri, bins[lr], bins[lr]i, bmnzi, bmzi, bneg, bnegi, bseli, bset, bseti, c(eq|ne), c(eq|ne)i, cl[et]_[su], cl[et]i_[su], copy_[su].[bhw], div_[su], dotp_[su], dpadd_[su], dpsub_[su], ilvev, ilvl, ilvod, ilvr, insv, insve, ldi Patch by Daniel Sanders llvm-svn: 188457
2013-08-15 20:24:57 +08:00
// Expand all builtin opcodes.
for (unsigned Opc = 0; Opc < ISD::BUILTIN_OP_END; ++Opc)
setOperationAction(Opc, Ty, Expand);
setOperationAction(ISD::LOAD, Ty, Legal);
setOperationAction(ISD::STORE, Ty, Legal);
setOperationAction(ISD::BITCAST, Ty, Legal);
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
setOperationAction(ISD::EXTRACT_VECTOR_ELT, Ty, Legal);
[mips][msa] Implemented insert_vector_elt for v4f32 and v2f64. For v4f32 and v2f64, INSERT_VECTOR_ELT is matched by a pseudo-insn which is later expanded to appropriate insve.[wd] insns. llvm-svn: 191515
2013-09-27 20:31:32 +08:00
setOperationAction(ISD::INSERT_VECTOR_ELT, Ty, Legal);
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
setOperationAction(ISD::BUILD_VECTOR, Ty, Custom);
[mips][msa] Added support for matching fadd, fdiv, flog2, fmul, frint, fsqrt, and fsub from normal IR (i.e. not intrinsics) llvm-svn: 190512
2013-09-11 18:51:30 +08:00
if (Ty != MVT::v8f16) {
[mips][msa] Added partial support for matching fmax_a from normal IR (i.e. not intrinsics) This covers the case where fmax_a can be used to implement ISD::FABS. llvm-svn: 191296
2013-09-24 21:02:08 +08:00
setOperationAction(ISD::FABS, Ty, Legal);
[mips][msa] Added support for matching fadd, fdiv, flog2, fmul, frint, fsqrt, and fsub from normal IR (i.e. not intrinsics) llvm-svn: 190512
2013-09-11 18:51:30 +08:00
setOperationAction(ISD::FADD, Ty, Legal);
setOperationAction(ISD::FDIV, Ty, Legal);
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
setOperationAction(ISD::FEXP2, Ty, Legal);
[mips][msa] Added support for matching fadd, fdiv, flog2, fmul, frint, fsqrt, and fsub from normal IR (i.e. not intrinsics) llvm-svn: 190512
2013-09-11 18:51:30 +08:00
setOperationAction(ISD::FLOG2, Ty, Legal);
[mips][msa] Added support for matching fmadd.[wd] from normal IR (i.e. not intrinsics) llvm-svn: 192430
2013-10-11 18:14:25 +08:00
setOperationAction(ISD::FMA, Ty, Legal);
[mips][msa] Added support for matching fadd, fdiv, flog2, fmul, frint, fsqrt, and fsub from normal IR (i.e. not intrinsics) llvm-svn: 190512
2013-09-11 18:51:30 +08:00
setOperationAction(ISD::FMUL, Ty, Legal);
setOperationAction(ISD::FRINT, Ty, Legal);
setOperationAction(ISD::FSQRT, Ty, Legal);
setOperationAction(ISD::FSUB, Ty, Legal);
[mips][msa] Added support for matching bsel and bseli from normal IR (i.e. not intrinsics) This required correcting the definition of the bsel and bseli intrinsics. llvm-svn: 191290
2013-09-24 20:04:44 +08:00
setOperationAction(ISD::VSELECT, Ty, Legal);
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
setOperationAction(ISD::SETCC, Ty, Legal);
setCondCodeAction(ISD::SETOGE, Ty, Expand);
setCondCodeAction(ISD::SETOGT, Ty, Expand);
setCondCodeAction(ISD::SETUGE, Ty, Expand);
setCondCodeAction(ISD::SETUGT, Ty, Expand);
setCondCodeAction(ISD::SETGE, Ty, Expand);
setCondCodeAction(ISD::SETGT, Ty, Expand);
[mips][msa] Added support for matching fadd, fdiv, flog2, fmul, frint, fsqrt, and fsub from normal IR (i.e. not intrinsics) llvm-svn: 190512
2013-09-11 18:51:30 +08:00
}
[Mips][msa] Added the simple builtins (add_a to dpsub[su], ilvev to ldi) Includes: add_a, adds_[asu], addv, addvi, andi.b, asub_[su].[bhwd], aver?_[su]_[bhwd], bclr, bclri, bins[lr], bins[lr]i, bmnzi, bmzi, bneg, bnegi, bseli, bset, bseti, c(eq|ne), c(eq|ne)i, cl[et]_[su], cl[et]i_[su], copy_[su].[bhw], div_[su], dotp_[su], dpadd_[su], dpsub_[su], ilvev, ilvl, ilvod, ilvr, insv, insve, ldi Patch by Daniel Sanders llvm-svn: 188457
2013-08-15 20:24:57 +08:00
}
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
Reland r308585 Builder clang-x86_64-linux-abi-test apparently failed due to a spurious error unrelated to the changes r308585 introduced. llvm-svn: 308612
2017-07-20 21:08:18 +08:00
SDValue MipsSETargetLowering::lowerSELECT(SDValue Op, SelectionDAG &DAG) const {
if(!Subtarget.hasMips32r6())
return MipsTargetLowering::LowerOperation(Op, DAG);
EVT ResTy = Op->getValueType(0);
SDLoc DL(Op);
// Although MTC1_D64 takes an i32 and writes an f64, the upper 32 bits of the
// floating point register are undefined. Not really an issue as sel.d, which
// is produced from an FSELECT node, only looks at bit 0.
SDValue Tmp = DAG.getNode(MipsISD::MTC1_D64, DL, MVT::f64, Op->getOperand(0));
return DAG.getNode(MipsISD::FSELECT, DL, ResTy, Tmp, Op->getOperand(1),
Op->getOperand(2));
}
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
bool
Add alignment value to allowsUnalignedMemoryAccess Rename to allowsMisalignedMemoryAccess. On R600, 8 and 16 byte accesses are mostly OK with 4-byte alignment, and don't need to be split into multiple accesses. Vector loads with an alignment of the element type are not uncommon in OpenCL code. llvm-svn: 214055
2014-07-28 01:46:40 +08:00
MipsSETargetLowering::allowsMisalignedMemoryAccesses(EVT VT,
unsigned,
unsigned,
bool *Fast) const {
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
MVT::SimpleValueType SVT = VT.getSimpleVT().SimpleTy;
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (Subtarget.systemSupportsUnalignedAccess()) {
[mips][mips64r6] [ls][dw][lr] are not available in MIPS32r6/MIPS64r6 Summary: Instead the system is required to provide some means of handling unaligned load/store without special instructions. Options include full hardware support, full trap-and-emulate, and hybrids such as hardware support within a cache line and trap-and-emulate for multi-line accesses. MipsSETargetLowering::allowsUnalignedMemoryAccesses() has been configured to assume that unaligned accesses are 'fast' on the basis that I expect few hardware implementations will opt for pure-software handling of unaligned accesses. The ones that do handle it purely in software can override this. mips64-load-store-left-right.ll has been merged into load-store-left-right.ll The stricter testing revealed a Bits!=Bytes bug in passByValArg(). This has been fixed and the variables renamed to clarify the units they hold. Reviewers: zoran.jovanovic, jkolek, vmedic Reviewed By: vmedic Differential Revision: http://reviews.llvm.org/D3872 llvm-svn: 209512
2014-05-23 21:18:02 +08:00
// MIPS32r6/MIPS64r6 is required to support unaligned access. It's
// implementation defined whether this is handled by hardware, software, or
// a hybrid of the two but it's expected that most implementations will
// handle the majority of cases in hardware.
if (Fast)
*Fast = true;
return true;
}
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
switch (SVT) {
case MVT::i64:
case MVT::i32:
if (Fast)
*Fast = true;
return true;
default:
return false;
}
}
[mips] Fix definitions of multiply, multiply-add/sub and divide instructions. The new instructions have explicit register output operands and use table-gen patterns instead of C++ code to do instruction selection. Mips16's instructions are unaffected by this change. llvm-svn: 178403
2013-03-30 09:36:35 +08:00
SDValue MipsSETargetLowering::LowerOperation(SDValue Op,
SelectionDAG &DAG) const {
switch(Op.getOpcode()) {
[mips] Enhance command line option "-mno-ldc1-sdc1" to expand base+index double precision loads and stores as well as reg+imm double precision loads and stores. Previously, expansion of loads and stores was done after register allocation, but now it takes place during legalization. As a result, users will see double precision stores and loads being emitted to spill and restore 64-bit FP registers. llvm-svn: 190235
2013-09-07 08:52:30 +08:00
case ISD::LOAD: return lowerLOAD(Op, DAG);
case ISD::STORE: return lowerSTORE(Op, DAG);
[mips] Fix definitions of multiply, multiply-add/sub and divide instructions. The new instructions have explicit register output operands and use table-gen patterns instead of C++ code to do instruction selection. Mips16's instructions are unaffected by this change. llvm-svn: 178403
2013-03-30 09:36:35 +08:00
case ISD::SMUL_LOHI: return lowerMulDiv(Op, MipsISD::Mult, true, true, DAG);
case ISD::UMUL_LOHI: return lowerMulDiv(Op, MipsISD::Multu, true, true, DAG);
case ISD::MULHS: return lowerMulDiv(Op, MipsISD::Mult, false, true, DAG);
case ISD::MULHU: return lowerMulDiv(Op, MipsISD::Multu, false, true, DAG);
case ISD::MUL: return lowerMulDiv(Op, MipsISD::Mult, true, false, DAG);
case ISD::SDIVREM: return lowerMulDiv(Op, MipsISD::DivRem, true, true, DAG);
80 columns. llvm-svn: 180040
2013-04-23 04:13:37 +08:00
case ISD::UDIVREM: return lowerMulDiv(Op, MipsISD::DivRemU, true, true,
DAG);
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
case ISD::INTRINSIC_WO_CHAIN: return lowerINTRINSIC_WO_CHAIN(Op, DAG);
case ISD::INTRINSIC_W_CHAIN: return lowerINTRINSIC_W_CHAIN(Op, DAG);
[mips][msa] Added load/store intrinsics. llvm-svn: 189476
2013-08-28 20:04:29 +08:00
case ISD::INTRINSIC_VOID: return lowerINTRINSIC_VOID(Op, DAG);
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
case ISD::EXTRACT_VECTOR_ELT: return lowerEXTRACT_VECTOR_ELT(Op, DAG);
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
case ISD::BUILD_VECTOR: return lowerBUILD_VECTOR(Op, DAG);
[mips][msa] Added support for matching vshf from normal IR (i.e. not intrinsics) llvm-svn: 191301
2013-09-24 22:02:15 +08:00
case ISD::VECTOR_SHUFFLE: return lowerVECTOR_SHUFFLE(Op, DAG);
Reland r308585 Builder clang-x86_64-linux-abi-test apparently failed due to a spurious error unrelated to the changes r308585 introduced. llvm-svn: 308612
2017-07-20 21:08:18 +08:00
case ISD::SELECT: return lowerSELECT(Op, DAG);
[mips] Fix definitions of multiply, multiply-add/sub and divide instructions. The new instructions have explicit register output operands and use table-gen patterns instead of C++ code to do instruction selection. Mips16's instructions are unaffected by this change. llvm-svn: 178403
2013-03-30 09:36:35 +08:00
}
return MipsTargetLowering::LowerOperation(Op, DAG);
}
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
// Fold zero extensions into MipsISD::VEXTRACT_[SZ]EXT_ELT
//
// Performs the following transformations:
// - Changes MipsISD::VEXTRACT_[SZ]EXT_ELT to zero extension if its
// sign/zero-extension is completely overwritten by the new one performed by
// the ISD::AND.
// - Removes redundant zero extensions performed by an ISD::AND.
static SDValue performANDCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
const MipsSubtarget &Subtarget) {
if (!Subtarget.hasMSA())
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
return SDValue();
SDValue Op0 = N->getOperand(0);
SDValue Op1 = N->getOperand(1);
unsigned Op0Opcode = Op0->getOpcode();
// (and (MipsVExtract[SZ]Ext $a, $b, $c), imm:$d)
// where $d + 1 == 2^n and n == 32
// or $d + 1 == 2^n and n <= 32 and ZExt
// -> (MipsVExtractZExt $a, $b, $c)
if (Op0Opcode == MipsISD::VEXTRACT_SEXT_ELT ||
Op0Opcode == MipsISD::VEXTRACT_ZEXT_ELT) {
ConstantSDNode *Mask = dyn_cast<ConstantSDNode>(Op1);
if (!Mask)
return SDValue();
int32_t Log2IfPositive = (Mask->getAPIntValue() + 1).exactLogBase2();
if (Log2IfPositive <= 0)
return SDValue(); // Mask+1 is not a power of 2
SDValue Op0Op2 = Op0->getOperand(2);
EVT ExtendTy = cast<VTSDNode>(Op0Op2)->getVT();
unsigned ExtendTySize = ExtendTy.getSizeInBits();
unsigned Log2 = Log2IfPositive;
if ((Op0Opcode == MipsISD::VEXTRACT_ZEXT_ELT && Log2 >= ExtendTySize) ||
Log2 == ExtendTySize) {
SDValue Ops[] = { Op0->getOperand(0), Op0->getOperand(1), Op0Op2 };
[SDAG] MorphNodeTo recursively deletes dead operands of the old fromulation of the node, which isn't really the desired behavior from within the combiner or legalizer, but is necessary within ISel. I've added a hopefully helpful comment and fixed the only two places where this took place. Yet another step toward the combiner and legalizer not needing to use update listeners with virtual calls to manage the worklists behind legalization and combining. llvm-svn: 214574
2014-08-02 06:09:43 +08:00
return DAG.getNode(MipsISD::VEXTRACT_ZEXT_ELT, SDLoc(Op0),
Op0->getVTList(),
makeArrayRef(Ops, Op0->getNumOperands()));
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
}
}
return SDValue();
}
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
// Determine if the specified node is a constant vector splat.
//
// Returns true and sets Imm if:
// * N is a ISD::BUILD_VECTOR representing a constant splat
//
// This function is quite similar to MipsSEDAGToDAGISel::selectVSplat. The
// differences are that it assumes the MSA has already been checked and the
// arbitrary requirement for a maximum of 32-bit integers isn't applied (and
// must not be in order for binsri.d to be selectable).
static bool isVSplat(SDValue N, APInt &Imm, bool IsLittleEndian) {
BuildVectorSDNode *Node = dyn_cast<BuildVectorSDNode>(N.getNode());
[C++] Use 'nullptr'. Target edition. llvm-svn: 207197
2014-04-25 13:30:21 +08:00
if (!Node)
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
return false;
APInt SplatValue, SplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
if (!Node->isConstantSplat(SplatValue, SplatUndef, SplatBitSize, HasAnyUndefs,
8, !IsLittleEndian))
return false;
Imm = SplatValue;
return true;
}
[mips][msa] Added support for matching bmnz, bmnzi, bmz, and bmzi from normal IR (i.e. not intrinsics) Also corrected the definition of the intrinsics for these instructions (the result register is also the first operand), and added intrinsics for bsel and bseli to clang (they already existed in the backend). These four operations are mostly equivalent to bsel, and bseli (the difference is which operand is tied to the result). As a result some of the tests changed as described below. bitwise.ll: - bsel.v test adapted so that the mask is unknown at compile-time. This stops it emitting bmnzi.b instead of the intended bsel.v. - The bseli.b test now tests the right thing. Namely the case when one of the values is an uimm8, rather than when the condition is a uimm8 (which is covered by bmnzi.b) compare.ll: - bsel.v tests now (correctly) emits bmnz.v instead of bsel.v because this is the same operation (see MSA.txt). i8.ll - CHECK-DAG-ized test. - bmzi.b test now (correctly) emits equivalent bmnzi.b with swapped operands because this is the same operation (see MSA.txt). - bseli.b still emits bseli.b though because the immediate makes it distinguishable from bmnzi.b. vec.ll: - CHECK-DAG-ized test. - bmz.v tests now (correctly) emits bmnz.v with swapped operands (see MSA.txt). - bsel.v tests now (correctly) emits bmnz.v with swapped operands (see MSA.txt). llvm-svn: 193693
2013-10-30 23:20:38 +08:00
// Test whether the given node is an all-ones build_vector.
static bool isVectorAllOnes(SDValue N) {
// Look through bitcasts. Endianness doesn't matter because we are looking
// for an all-ones value.
if (N->getOpcode() == ISD::BITCAST)
N = N->getOperand(0);
BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(N);
if (!BVN)
return false;
APInt SplatValue, SplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
// Endianness doesn't matter in this context because we are looking for
// an all-ones value.
if (BVN->isConstantSplat(SplatValue, SplatUndef, SplatBitSize, HasAnyUndefs))
return SplatValue.isAllOnesValue();
return false;
}
// Test whether N is the bitwise inverse of OfNode.
static bool isBitwiseInverse(SDValue N, SDValue OfNode) {
if (N->getOpcode() != ISD::XOR)
return false;
if (isVectorAllOnes(N->getOperand(0)))
return N->getOperand(1) == OfNode;
if (isVectorAllOnes(N->getOperand(1)))
return N->getOperand(0) == OfNode;
return false;
}
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
// Perform combines where ISD::OR is the root node.
//
// Performs the following transformations:
// - (or (and $a, $mask), (and $b, $inv_mask)) => (vselect $mask, $a, $b)
// where $inv_mask is the bitwise inverse of $mask and the 'or' has a 128-bit
// vector type.
static SDValue performORCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
const MipsSubtarget &Subtarget) {
if (!Subtarget.hasMSA())
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
return SDValue();
EVT Ty = N->getValueType(0);
if (!Ty.is128BitVector())
return SDValue();
SDValue Op0 = N->getOperand(0);
SDValue Op1 = N->getOperand(1);
if (Op0->getOpcode() == ISD::AND && Op1->getOpcode() == ISD::AND) {
SDValue Op0Op0 = Op0->getOperand(0);
SDValue Op0Op1 = Op0->getOperand(1);
SDValue Op1Op0 = Op1->getOperand(0);
SDValue Op1Op1 = Op1->getOperand(1);
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
bool IsLittleEndian = !Subtarget.isLittle();
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
SDValue IfSet, IfClr, Cond;
[mips][msa] Added support for matching bmnz, bmnzi, bmz, and bmzi from normal IR (i.e. not intrinsics) Also corrected the definition of the intrinsics for these instructions (the result register is also the first operand), and added intrinsics for bsel and bseli to clang (they already existed in the backend). These four operations are mostly equivalent to bsel, and bseli (the difference is which operand is tied to the result). As a result some of the tests changed as described below. bitwise.ll: - bsel.v test adapted so that the mask is unknown at compile-time. This stops it emitting bmnzi.b instead of the intended bsel.v. - The bseli.b test now tests the right thing. Namely the case when one of the values is an uimm8, rather than when the condition is a uimm8 (which is covered by bmnzi.b) compare.ll: - bsel.v tests now (correctly) emits bmnz.v instead of bsel.v because this is the same operation (see MSA.txt). i8.ll - CHECK-DAG-ized test. - bmzi.b test now (correctly) emits equivalent bmnzi.b with swapped operands because this is the same operation (see MSA.txt). - bseli.b still emits bseli.b though because the immediate makes it distinguishable from bmnzi.b. vec.ll: - CHECK-DAG-ized test. - bmz.v tests now (correctly) emits bmnz.v with swapped operands (see MSA.txt). - bsel.v tests now (correctly) emits bmnz.v with swapped operands (see MSA.txt). llvm-svn: 193693
2013-10-30 23:20:38 +08:00
bool IsConstantMask = false;
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
APInt Mask, InvMask;
// If Op0Op0 is an appropriate mask, try to find it's inverse in either
// Op1Op0, or Op1Op1. Keep track of the Cond, IfSet, and IfClr nodes, while
// looking.
// IfClr will be set if we find a valid match.
if (isVSplat(Op0Op0, Mask, IsLittleEndian)) {
Cond = Op0Op0;
IfSet = Op0Op1;
[mips][msa] Fix a corner case in performORCombine() when combining nodes into VSELECT. Mask == ~InvMask asserts if the width of Mask and InvMask differ. The combine isn't valid (with two exceptions, see below) if the widths differ so test for this before testing Mask == ~InvMask. In the specific cases of Mask=~0 and InvMask=0, as well as Mask=0 and InvMask=~0, the combine is still valid. However, there are more appropriate combines that could be used in these cases such as folding x & 0 to 0, or x & ~0 to x. llvm-svn: 195364
2013-11-22 00:11:31 +08:00
if (isVSplat(Op1Op0, InvMask, IsLittleEndian) &&
Mask.getBitWidth() == InvMask.getBitWidth() && Mask == ~InvMask)
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
IfClr = Op1Op1;
[mips][msa] Fix a corner case in performORCombine() when combining nodes into VSELECT. Mask == ~InvMask asserts if the width of Mask and InvMask differ. The combine isn't valid (with two exceptions, see below) if the widths differ so test for this before testing Mask == ~InvMask. In the specific cases of Mask=~0 and InvMask=0, as well as Mask=0 and InvMask=~0, the combine is still valid. However, there are more appropriate combines that could be used in these cases such as folding x & 0 to 0, or x & ~0 to x. llvm-svn: 195364
2013-11-22 00:11:31 +08:00
else if (isVSplat(Op1Op1, InvMask, IsLittleEndian) &&
Mask.getBitWidth() == InvMask.getBitWidth() && Mask == ~InvMask)
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
IfClr = Op1Op0;
[mips][msa] Added support for matching bmnz, bmnzi, bmz, and bmzi from normal IR (i.e. not intrinsics) Also corrected the definition of the intrinsics for these instructions (the result register is also the first operand), and added intrinsics for bsel and bseli to clang (they already existed in the backend). These four operations are mostly equivalent to bsel, and bseli (the difference is which operand is tied to the result). As a result some of the tests changed as described below. bitwise.ll: - bsel.v test adapted so that the mask is unknown at compile-time. This stops it emitting bmnzi.b instead of the intended bsel.v. - The bseli.b test now tests the right thing. Namely the case when one of the values is an uimm8, rather than when the condition is a uimm8 (which is covered by bmnzi.b) compare.ll: - bsel.v tests now (correctly) emits bmnz.v instead of bsel.v because this is the same operation (see MSA.txt). i8.ll - CHECK-DAG-ized test. - bmzi.b test now (correctly) emits equivalent bmnzi.b with swapped operands because this is the same operation (see MSA.txt). - bseli.b still emits bseli.b though because the immediate makes it distinguishable from bmnzi.b. vec.ll: - CHECK-DAG-ized test. - bmz.v tests now (correctly) emits bmnz.v with swapped operands (see MSA.txt). - bsel.v tests now (correctly) emits bmnz.v with swapped operands (see MSA.txt). llvm-svn: 193693
2013-10-30 23:20:38 +08:00
IsConstantMask = true;
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
}
// If IfClr is not yet set, and Op0Op1 is an appropriate mask, try the same
// thing again using this mask.
// IfClr will be set if we find a valid match.
if (!IfClr.getNode() && isVSplat(Op0Op1, Mask, IsLittleEndian)) {
Cond = Op0Op1;
IfSet = Op0Op0;
[mips][msa] Fix a corner case in performORCombine() when combining nodes into VSELECT. Mask == ~InvMask asserts if the width of Mask and InvMask differ. The combine isn't valid (with two exceptions, see below) if the widths differ so test for this before testing Mask == ~InvMask. In the specific cases of Mask=~0 and InvMask=0, as well as Mask=0 and InvMask=~0, the combine is still valid. However, there are more appropriate combines that could be used in these cases such as folding x & 0 to 0, or x & ~0 to x. llvm-svn: 195364
2013-11-22 00:11:31 +08:00
if (isVSplat(Op1Op0, InvMask, IsLittleEndian) &&
Mask.getBitWidth() == InvMask.getBitWidth() && Mask == ~InvMask)
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
IfClr = Op1Op1;
[mips][msa] Fix a corner case in performORCombine() when combining nodes into VSELECT. Mask == ~InvMask asserts if the width of Mask and InvMask differ. The combine isn't valid (with two exceptions, see below) if the widths differ so test for this before testing Mask == ~InvMask. In the specific cases of Mask=~0 and InvMask=0, as well as Mask=0 and InvMask=~0, the combine is still valid. However, there are more appropriate combines that could be used in these cases such as folding x & 0 to 0, or x & ~0 to x. llvm-svn: 195364
2013-11-22 00:11:31 +08:00
else if (isVSplat(Op1Op1, InvMask, IsLittleEndian) &&
Mask.getBitWidth() == InvMask.getBitWidth() && Mask == ~InvMask)
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
IfClr = Op1Op0;
[mips][msa] Added support for matching bmnz, bmnzi, bmz, and bmzi from normal IR (i.e. not intrinsics) Also corrected the definition of the intrinsics for these instructions (the result register is also the first operand), and added intrinsics for bsel and bseli to clang (they already existed in the backend). These four operations are mostly equivalent to bsel, and bseli (the difference is which operand is tied to the result). As a result some of the tests changed as described below. bitwise.ll: - bsel.v test adapted so that the mask is unknown at compile-time. This stops it emitting bmnzi.b instead of the intended bsel.v. - The bseli.b test now tests the right thing. Namely the case when one of the values is an uimm8, rather than when the condition is a uimm8 (which is covered by bmnzi.b) compare.ll: - bsel.v tests now (correctly) emits bmnz.v instead of bsel.v because this is the same operation (see MSA.txt). i8.ll - CHECK-DAG-ized test. - bmzi.b test now (correctly) emits equivalent bmnzi.b with swapped operands because this is the same operation (see MSA.txt). - bseli.b still emits bseli.b though because the immediate makes it distinguishable from bmnzi.b. vec.ll: - CHECK-DAG-ized test. - bmz.v tests now (correctly) emits bmnz.v with swapped operands (see MSA.txt). - bsel.v tests now (correctly) emits bmnz.v with swapped operands (see MSA.txt). llvm-svn: 193693
2013-10-30 23:20:38 +08:00
IsConstantMask = true;
}
// If IfClr is not yet set, try looking for a non-constant match.
// IfClr will be set if we find a valid match amongst the eight
// possibilities.
if (!IfClr.getNode()) {
if (isBitwiseInverse(Op0Op0, Op1Op0)) {
Cond = Op1Op0;
IfSet = Op1Op1;
IfClr = Op0Op1;
} else if (isBitwiseInverse(Op0Op1, Op1Op0)) {
Cond = Op1Op0;
IfSet = Op1Op1;
IfClr = Op0Op0;
} else if (isBitwiseInverse(Op0Op0, Op1Op1)) {
Cond = Op1Op1;
IfSet = Op1Op0;
IfClr = Op0Op1;
} else if (isBitwiseInverse(Op0Op1, Op1Op1)) {
Cond = Op1Op1;
IfSet = Op1Op0;
IfClr = Op0Op0;
} else if (isBitwiseInverse(Op1Op0, Op0Op0)) {
Cond = Op0Op0;
IfSet = Op0Op1;
IfClr = Op1Op1;
} else if (isBitwiseInverse(Op1Op1, Op0Op0)) {
Cond = Op0Op0;
IfSet = Op0Op1;
IfClr = Op1Op0;
} else if (isBitwiseInverse(Op1Op0, Op0Op1)) {
Cond = Op0Op1;
IfSet = Op0Op0;
IfClr = Op1Op1;
} else if (isBitwiseInverse(Op1Op1, Op0Op1)) {
Cond = Op0Op1;
IfSet = Op0Op0;
IfClr = Op1Op0;
}
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
}
// At this point, IfClr will be set if we have a valid match.
if (!IfClr.getNode())
return SDValue();
assert(Cond.getNode() && IfSet.getNode());
// Fold degenerate cases.
[mips][msa] Added support for matching bmnz, bmnzi, bmz, and bmzi from normal IR (i.e. not intrinsics) Also corrected the definition of the intrinsics for these instructions (the result register is also the first operand), and added intrinsics for bsel and bseli to clang (they already existed in the backend). These four operations are mostly equivalent to bsel, and bseli (the difference is which operand is tied to the result). As a result some of the tests changed as described below. bitwise.ll: - bsel.v test adapted so that the mask is unknown at compile-time. This stops it emitting bmnzi.b instead of the intended bsel.v. - The bseli.b test now tests the right thing. Namely the case when one of the values is an uimm8, rather than when the condition is a uimm8 (which is covered by bmnzi.b) compare.ll: - bsel.v tests now (correctly) emits bmnz.v instead of bsel.v because this is the same operation (see MSA.txt). i8.ll - CHECK-DAG-ized test. - bmzi.b test now (correctly) emits equivalent bmnzi.b with swapped operands because this is the same operation (see MSA.txt). - bseli.b still emits bseli.b though because the immediate makes it distinguishable from bmnzi.b. vec.ll: - CHECK-DAG-ized test. - bmz.v tests now (correctly) emits bmnz.v with swapped operands (see MSA.txt). - bsel.v tests now (correctly) emits bmnz.v with swapped operands (see MSA.txt). llvm-svn: 193693
2013-10-30 23:20:38 +08:00
if (IsConstantMask) {
if (Mask.isAllOnesValue())
return IfSet;
else if (Mask == 0)
return IfClr;
}
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
// Transform the DAG into an equivalent VSELECT.
[mips] BSEL's and BINS[RL] operands are reversed compared to the vselect node used in the pattern. Summary: Correct the match patterns and the lowerings that made the CodeGen tests pass despite the mistakes. The original testcase that discovered the problem was SingleSource/UnitTests/SignlessType/factor.c in test-suite. During review, we also found that some of the existing CodeGen tests were incorrect and fixed them: * bitwise.ll: In bsel_v16i8 the IfSet/IfClear were reversed because bsel and bmnz have different operand orders and the test didn't correctly account for this. bmnz goes 'IfClear, IfSet, CondMask', while bsel goes 'CondMask, IfClear, IfSet'. * vec.ll: In the cases where a bsel is emitted as a bmnz (they are the same operation with a different input tied to the result) the operands were in the wrong order. * compare.ll and compare_float.ll: The bsel operand order was correct for a greater-than comparison, but a greater-than comparison instruction doesn't exist. Lowering this operation inverts the condition so the IfSet/IfClear need to be swapped to match. The differences between BSEL, BMNZ, and BMZ and how they map to/from vselect are rather confusing. I've therefore added a note to MSA.txt to explain this in a single place in addition to the comments that explain each case. Reviewers: matheusalmeida, jacksprat Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D3028 llvm-svn: 203657
2014-03-12 19:54:00 +08:00
return DAG.getNode(ISD::VSELECT, SDLoc(N), Ty, Cond, IfSet, IfClr);
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
}
return SDValue();
}
[mips] Improve genConstMult() to work with arbitrary precision APInt is now used instead of uint64_t in function genConstMult() allowing multiplication optimizations with constants of arbitrary length. Patch by Milos Stojanovic. Differential Revision: https://reviews.llvm.org/D38130 llvm-svn: 318296
2017-11-15 23:24:04 +08:00
static SDValue genConstMult(SDValue X, APInt C, const SDLoc &DL, EVT VT,
[mips] Improve code generation for constant multiplication using shifts, adds and subs. llvm-svn: 185011
2013-06-27 02:48:17 +08:00
EVT ShiftTy, SelectionDAG &DAG) {
// Return 0.
if (C == 0)
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
return DAG.getConstant(0, DL, VT);
[mips] Improve code generation for constant multiplication using shifts, adds and subs. llvm-svn: 185011
2013-06-27 02:48:17 +08:00
// Return x.
if (C == 1)
return X;
// If c is power of 2, return (shl x, log2(c)).
[mips] Improve genConstMult() to work with arbitrary precision APInt is now used instead of uint64_t in function genConstMult() allowing multiplication optimizations with constants of arbitrary length. Patch by Milos Stojanovic. Differential Revision: https://reviews.llvm.org/D38130 llvm-svn: 318296
2017-11-15 23:24:04 +08:00
if (C.isPowerOf2())
[mips] Improve code generation for constant multiplication using shifts, adds and subs. llvm-svn: 185011
2013-06-27 02:48:17 +08:00
return DAG.getNode(ISD::SHL, DL, VT, X,
[mips] Improve genConstMult() to work with arbitrary precision APInt is now used instead of uint64_t in function genConstMult() allowing multiplication optimizations with constants of arbitrary length. Patch by Milos Stojanovic. Differential Revision: https://reviews.llvm.org/D38130 llvm-svn: 318296
2017-11-15 23:24:04 +08:00
DAG.getConstant(C.logBase2(), DL, ShiftTy));
[mips] Improve code generation for constant multiplication using shifts, adds and subs. llvm-svn: 185011
2013-06-27 02:48:17 +08:00
[mips] Improve genConstMult() to work with arbitrary precision APInt is now used instead of uint64_t in function genConstMult() allowing multiplication optimizations with constants of arbitrary length. Patch by Milos Stojanovic. Differential Revision: https://reviews.llvm.org/D38130 llvm-svn: 318296
2017-11-15 23:24:04 +08:00
unsigned BitWidth = C.getBitWidth();
APInt Floor = APInt(BitWidth, 1) << C.logBase2();
APInt Ceil = C.isNegative() ? APInt(BitWidth, 0) :
APInt(BitWidth, 1) << C.ceilLogBase2();
[mips] Improve code generation for constant multiplication using shifts, adds and subs. llvm-svn: 185011
2013-06-27 02:48:17 +08:00
// If |c - floor_c| <= |c - ceil_c|,
// where floor_c = pow(2, floor(log2(c))) and ceil_c = pow(2, ceil(log2(c))),
// return (add constMult(x, floor_c), constMult(x, c - floor_c)).
[mips] Improve genConstMult() to work with arbitrary precision APInt is now used instead of uint64_t in function genConstMult() allowing multiplication optimizations with constants of arbitrary length. Patch by Milos Stojanovic. Differential Revision: https://reviews.llvm.org/D38130 llvm-svn: 318296
2017-11-15 23:24:04 +08:00
if ((C - Floor).ule(Ceil - C)) {
[mips] Improve code generation for constant multiplication using shifts, adds and subs. llvm-svn: 185011
2013-06-27 02:48:17 +08:00
SDValue Op0 = genConstMult(X, Floor, DL, VT, ShiftTy, DAG);
SDValue Op1 = genConstMult(X, C - Floor, DL, VT, ShiftTy, DAG);
return DAG.getNode(ISD::ADD, DL, VT, Op0, Op1);
}
// If |c - floor_c| > |c - ceil_c|,
// return (sub constMult(x, ceil_c), constMult(x, ceil_c - c)).
SDValue Op0 = genConstMult(X, Ceil, DL, VT, ShiftTy, DAG);
SDValue Op1 = genConstMult(X, Ceil - C, DL, VT, ShiftTy, DAG);
return DAG.getNode(ISD::SUB, DL, VT, Op0, Op1);
}
static SDValue performMULCombine(SDNode *N, SelectionDAG &DAG,
const TargetLowering::DAGCombinerInfo &DCI,
const MipsSETargetLowering *TL) {
EVT VT = N->getValueType(0);
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(1)))
if (!VT.isVector())
[mips] Improve genConstMult() to work with arbitrary precision APInt is now used instead of uint64_t in function genConstMult() allowing multiplication optimizations with constants of arbitrary length. Patch by Milos Stojanovic. Differential Revision: https://reviews.llvm.org/D38130 llvm-svn: 318296
2017-11-15 23:24:04 +08:00
return genConstMult(N->getOperand(0), C->getAPIntValue(), SDLoc(N), VT,
Re-instate the EVT parameter to getScalarShiftAmountTy() for OOT user A documentation for this function would be nice by the way. From: Mehdi Amini <mehdi.amini@apple.com> llvm-svn: 241807
2015-07-09 23:12:23 +08:00
TL->getScalarShiftAmountTy(DAG.getDataLayout(), VT),
DAG);
[mips] Improve code generation for constant multiplication using shifts, adds and subs. llvm-svn: 185011
2013-06-27 02:48:17 +08:00
return SDValue(N, 0);
}
[mips] Instruction selection patterns for DSP-ASE vector shifts. llvm-svn: 179906
2013-04-20 07:21:32 +08:00
static SDValue performDSPShiftCombine(unsigned Opc, SDNode *N, EVT Ty,
SelectionDAG &DAG,
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
const MipsSubtarget &Subtarget) {
[mips] Instruction selection patterns for DSP-ASE vector shifts. llvm-svn: 179906
2013-04-20 07:21:32 +08:00
// See if this is a vector splat immediate node.
APInt SplatValue, SplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
getVectorElementType().getSizeInBits() -> getScalarSizeInBits() ; NFCI llvm-svn: 281495
2016-09-15 00:37:15 +08:00
unsigned EltSize = Ty.getScalarSizeInBits();
[mips] Instruction selection patterns for DSP-ASE vector shifts. llvm-svn: 179906
2013-04-20 07:21:32 +08:00
BuildVectorSDNode *BV = dyn_cast<BuildVectorSDNode>(N->getOperand(1));
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (!Subtarget.hasDSP())
[mips][msa/dsp] Only do DSP combines if DSP is enabled. Fixes a crash (null pointer dereferenced) when MSA is enabled. llvm-svn: 195343
2013-11-21 19:40:14 +08:00
return SDValue();
[mips] In performDSPShiftCombine, check that all elements in the vector are shifted by the same amount and the shift amount is smaller than the element size. llvm-svn: 180039
2013-04-23 03:58:23 +08:00
if (!BV ||
80 columns. llvm-svn: 180040
2013-04-23 04:13:37 +08:00
!BV->isConstantSplat(SplatValue, SplatUndef, SplatBitSize, HasAnyUndefs,
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
EltSize, !Subtarget.isLittle()) ||
[mips] In performDSPShiftCombine, check that all elements in the vector are shifted by the same amount and the shift amount is smaller than the element size. llvm-svn: 180039
2013-04-23 03:58:23 +08:00
(SplatBitSize != EltSize) ||
[mips] Compare splat value with element size instead of calling isUIntN. No intended changes in functionality. llvm-svn: 180130
2013-04-24 02:09:42 +08:00
(SplatValue.getZExtValue() >= EltSize))
[mips] Instruction selection patterns for DSP-ASE vector shifts. llvm-svn: 179906
2013-04-20 07:21:32 +08:00
return SDValue();
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
SDLoc DL(N);
return DAG.getNode(Opc, DL, Ty, N->getOperand(0),
DAG.getConstant(SplatValue.getZExtValue(), DL, MVT::i32));
[mips] Instruction selection patterns for DSP-ASE vector shifts. llvm-svn: 179906
2013-04-20 07:21:32 +08:00
}
static SDValue performSHLCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
const MipsSubtarget &Subtarget) {
[mips] Instruction selection patterns for DSP-ASE vector shifts. llvm-svn: 179906
2013-04-20 07:21:32 +08:00
EVT Ty = N->getValueType(0);
if ((Ty != MVT::v2i16) && (Ty != MVT::v4i8))
return SDValue();
return performDSPShiftCombine(MipsISD::SHLL_DSP, N, Ty, DAG, Subtarget);
}
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
// Fold sign-extensions into MipsISD::VEXTRACT_[SZ]EXT_ELT for MSA and fold
// constant splats into MipsISD::SHRA_DSP for DSPr2.
//
// Performs the following transformations:
// - Changes MipsISD::VEXTRACT_[SZ]EXT_ELT to sign extension if its
// sign/zero-extension is completely overwritten by the new one performed by
// the ISD::SRA and ISD::SHL nodes.
// - Removes redundant sign extensions performed by an ISD::SRA and ISD::SHL
// sequence.
//
// See performDSPShiftCombine for more information about the transformation
// used for DSPr2.
[mips] Instruction selection patterns for DSP-ASE vector shifts. llvm-svn: 179906
2013-04-20 07:21:32 +08:00
static SDValue performSRACombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
const MipsSubtarget &Subtarget) {
[mips] Instruction selection patterns for DSP-ASE vector shifts. llvm-svn: 179906
2013-04-20 07:21:32 +08:00
EVT Ty = N->getValueType(0);
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (Subtarget.hasMSA()) {
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
SDValue Op0 = N->getOperand(0);
SDValue Op1 = N->getOperand(1);
// (sra (shl (MipsVExtract[SZ]Ext $a, $b, $c), imm:$d), imm:$d)
// where $d + sizeof($c) == 32
// or $d + sizeof($c) <= 32 and SExt
// -> (MipsVExtractSExt $a, $b, $c)
if (Op0->getOpcode() == ISD::SHL && Op1 == Op0->getOperand(1)) {
SDValue Op0Op0 = Op0->getOperand(0);
ConstantSDNode *ShAmount = dyn_cast<ConstantSDNode>(Op1);
if (!ShAmount)
return SDValue();
[mips][msa] Added missing check in performSRACombine Reviewers: jacksprat, dsanders Reviewed By: dsanders Differential Revision: http://llvm-reviews.chandlerc.com/D1755 llvm-svn: 191495
2013-09-27 17:25:29 +08:00
if (Op0Op0->getOpcode() != MipsISD::VEXTRACT_SEXT_ELT &&
Op0Op0->getOpcode() != MipsISD::VEXTRACT_ZEXT_ELT)
return SDValue();
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
EVT ExtendTy = cast<VTSDNode>(Op0Op0->getOperand(2))->getVT();
unsigned TotalBits = ShAmount->getZExtValue() + ExtendTy.getSizeInBits();
if (TotalBits == 32 ||
(Op0Op0->getOpcode() == MipsISD::VEXTRACT_SEXT_ELT &&
TotalBits <= 32)) {
SDValue Ops[] = { Op0Op0->getOperand(0), Op0Op0->getOperand(1),
Op0Op0->getOperand(2) };
[SDAG] MorphNodeTo recursively deletes dead operands of the old fromulation of the node, which isn't really the desired behavior from within the combiner or legalizer, but is necessary within ISel. I've added a hopefully helpful comment and fixed the only two places where this took place. Yet another step toward the combiner and legalizer not needing to use update listeners with virtual calls to manage the worklists behind legalization and combining. llvm-svn: 214574
2014-08-02 06:09:43 +08:00
return DAG.getNode(MipsISD::VEXTRACT_SEXT_ELT, SDLoc(Op0Op0),
Op0Op0->getVTList(),
makeArrayRef(Ops, Op0Op0->getNumOperands()));
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
}
}
}
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if ((Ty != MVT::v2i16) && ((Ty != MVT::v4i8) || !Subtarget.hasDSPR2()))
[mips] Instruction selection patterns for DSP-ASE vector shifts. llvm-svn: 179906
2013-04-20 07:21:32 +08:00
return SDValue();
return performDSPShiftCombine(MipsISD::SHRA_DSP, N, Ty, DAG, Subtarget);
}
static SDValue performSRLCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
const MipsSubtarget &Subtarget) {
[mips] Instruction selection patterns for DSP-ASE vector shifts. llvm-svn: 179906
2013-04-20 07:21:32 +08:00
EVT Ty = N->getValueType(0);
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (((Ty != MVT::v2i16) || !Subtarget.hasDSPR2()) && (Ty != MVT::v4i8))
[mips] Instruction selection patterns for DSP-ASE vector shifts. llvm-svn: 179906
2013-04-20 07:21:32 +08:00
return SDValue();
return performDSPShiftCombine(MipsISD::SHRL_DSP, N, Ty, DAG, Subtarget);
}
[mips] Instruction selection patterns for DSP-ASE vector select and compare instructions. llvm-svn: 180820
2013-05-01 06:37:26 +08:00
static bool isLegalDSPCondCode(EVT Ty, ISD::CondCode CC) {
bool IsV216 = (Ty == MVT::v2i16);
switch (CC) {
case ISD::SETEQ:
case ISD::SETNE: return true;
case ISD::SETLT:
case ISD::SETLE:
case ISD::SETGT:
case ISD::SETGE: return IsV216;
case ISD::SETULT:
case ISD::SETULE:
case ISD::SETUGT:
case ISD::SETUGE: return !IsV216;
default: return false;
}
}
static SDValue performSETCCCombine(SDNode *N, SelectionDAG &DAG) {
EVT Ty = N->getValueType(0);
if ((Ty != MVT::v2i16) && (Ty != MVT::v4i8))
return SDValue();
if (!isLegalDSPCondCode(Ty, cast<CondCodeSDNode>(N->getOperand(2))->get()))
return SDValue();
Track IR ordering of SelectionDAG nodes 2/4. Change SelectionDAG::getXXXNode() interfaces as well as call sites of these functions to pass in SDLoc instead of DebugLoc. llvm-svn: 182703
2013-05-25 10:42:55 +08:00
return DAG.getNode(MipsISD::SETCC_DSP, SDLoc(N), Ty, N->getOperand(0),
[mips] Instruction selection patterns for DSP-ASE vector select and compare instructions. llvm-svn: 180820
2013-05-01 06:37:26 +08:00
N->getOperand(1), N->getOperand(2));
}
static SDValue performVSELECTCombine(SDNode *N, SelectionDAG &DAG) {
EVT Ty = N->getValueType(0);
[mips][msa] Added support for matching max, maxi, min, mini from normal IR (i.e. not intrinsics) llvm-svn: 191291
2013-09-24 20:18:31 +08:00
if (Ty.is128BitVector() && Ty.isInteger()) {
// Try the following combines:
// (vselect (setcc $a, $b, SETLT), $b, $a)) -> (vsmax $a, $b)
// (vselect (setcc $a, $b, SETLE), $b, $a)) -> (vsmax $a, $b)
// (vselect (setcc $a, $b, SETLT), $a, $b)) -> (vsmin $a, $b)
// (vselect (setcc $a, $b, SETLE), $a, $b)) -> (vsmin $a, $b)
// (vselect (setcc $a, $b, SETULT), $b, $a)) -> (vumax $a, $b)
// (vselect (setcc $a, $b, SETULE), $b, $a)) -> (vumax $a, $b)
// (vselect (setcc $a, $b, SETULT), $a, $b)) -> (vumin $a, $b)
// (vselect (setcc $a, $b, SETULE), $a, $b)) -> (vumin $a, $b)
// SETGT/SETGE/SETUGT/SETUGE variants of these will show up initially but
// will be expanded to equivalent SETLT/SETLE/SETULT/SETULE versions by the
// legalizer.
SDValue Op0 = N->getOperand(0);
if (Op0->getOpcode() != ISD::SETCC)
return SDValue();
[mips] Instruction selection patterns for DSP-ASE vector select and compare instructions. llvm-svn: 180820
2013-05-01 06:37:26 +08:00
[mips][msa] Added support for matching max, maxi, min, mini from normal IR (i.e. not intrinsics) llvm-svn: 191291
2013-09-24 20:18:31 +08:00
ISD::CondCode CondCode = cast<CondCodeSDNode>(Op0->getOperand(2))->get();
bool Signed;
[mips] Instruction selection patterns for DSP-ASE vector select and compare instructions. llvm-svn: 180820
2013-05-01 06:37:26 +08:00
[mips][msa] Added support for matching max, maxi, min, mini from normal IR (i.e. not intrinsics) llvm-svn: 191291
2013-09-24 20:18:31 +08:00
if (CondCode == ISD::SETLT || CondCode == ISD::SETLE)
Signed = true;
else if (CondCode == ISD::SETULT || CondCode == ISD::SETULE)
Signed = false;
else
return SDValue();
[mips] Instruction selection patterns for DSP-ASE vector select and compare instructions. llvm-svn: 180820
2013-05-01 06:37:26 +08:00
[mips][msa] Added support for matching max, maxi, min, mini from normal IR (i.e. not intrinsics) llvm-svn: 191291
2013-09-24 20:18:31 +08:00
SDValue Op1 = N->getOperand(1);
SDValue Op2 = N->getOperand(2);
SDValue Op0Op0 = Op0->getOperand(0);
SDValue Op0Op1 = Op0->getOperand(1);
if (Op1 == Op0Op0 && Op2 == Op0Op1)
return DAG.getNode(Signed ? MipsISD::VSMIN : MipsISD::VUMIN, SDLoc(N),
Ty, Op1, Op2);
else if (Op1 == Op0Op1 && Op2 == Op0Op0)
return DAG.getNode(Signed ? MipsISD::VSMAX : MipsISD::VUMAX, SDLoc(N),
Ty, Op1, Op2);
} else if ((Ty == MVT::v2i16) || (Ty == MVT::v4i8)) {
SDValue SetCC = N->getOperand(0);
if (SetCC.getOpcode() != MipsISD::SETCC_DSP)
return SDValue();
return DAG.getNode(MipsISD::SELECT_CC_DSP, SDLoc(N), Ty,
SetCC.getOperand(0), SetCC.getOperand(1),
N->getOperand(1), N->getOperand(2), SetCC.getOperand(2));
}
return SDValue();
[mips] Instruction selection patterns for DSP-ASE vector select and compare instructions. llvm-svn: 180820
2013-05-01 06:37:26 +08:00
}
[mips][msa] Added support for matching nor from normal IR (i.e. not intrinsics) llvm-svn: 191195
2013-09-23 21:22:24 +08:00
static SDValue performXORCombine(SDNode *N, SelectionDAG &DAG,
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
const MipsSubtarget &Subtarget) {
[mips][msa] Added support for matching nor from normal IR (i.e. not intrinsics) llvm-svn: 191195
2013-09-23 21:22:24 +08:00
EVT Ty = N->getValueType(0);
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (Subtarget.hasMSA() && Ty.is128BitVector() && Ty.isInteger()) {
[mips][msa] Added support for matching nor from normal IR (i.e. not intrinsics) llvm-svn: 191195
2013-09-23 21:22:24 +08:00
// Try the following combines:
// (xor (or $a, $b), (build_vector allones))
// (xor (or $a, $b), (bitcast (build_vector allones)))
SDValue Op0 = N->getOperand(0);
SDValue Op1 = N->getOperand(1);
SDValue NotOp;
if (ISD::isBuildVectorAllOnes(Op0.getNode()))
NotOp = Op1;
else if (ISD::isBuildVectorAllOnes(Op1.getNode()))
NotOp = Op0;
else
return SDValue();
if (NotOp->getOpcode() == ISD::OR)
return DAG.getNode(MipsISD::VNOR, SDLoc(N), Ty, NotOp->getOperand(0),
NotOp->getOperand(1));
}
return SDValue();
}
[mips] Move the code which does dag-combine for multiply-add/sub nodes to derived class MipsSETargetLowering. We shouldn't be generating madd/msub nodes if target is Mips16, since Mips16 doesn't have support for multipy-add/sub instructions. llvm-svn: 178404
2013-03-30 09:42:24 +08:00
SDValue
MipsSETargetLowering::PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const {
SelectionDAG &DAG = DCI.DAG;
[mips] Instruction selection patterns for DSP-ASE vector select and compare instructions. llvm-svn: 180820
2013-05-01 06:37:26 +08:00
SDValue Val;
[mips] Move the code which does dag-combine for multiply-add/sub nodes to derived class MipsSETargetLowering. We shouldn't be generating madd/msub nodes if target is Mips16, since Mips16 doesn't have support for multipy-add/sub instructions. llvm-svn: 178404
2013-03-30 09:42:24 +08:00
switch (N->getOpcode()) {
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
case ISD::AND:
Val = performANDCombine(N, DAG, DCI, Subtarget);
break;
[mips][msa] Combine binsri-like DAG of AND and OR into equivalent VSELECT (or (and $a, $mask), (and $b, $inverse_mask)) => (vselect $mask, $a, $b). where $mask is a constant splat. This allows bitwise operations to make use of bsel. It's also a stepping stone towards matching bins[lr], and bins[lr]i from normal IR. Two sets of similar tests have been added in this commit. The bsel_* functions test the case where binsri cannot be used. The binsr_*_i functions will start to use the binsri instruction in the next commit. llvm-svn: 193682
2013-10-30 21:51:01 +08:00
case ISD::OR:
Val = performORCombine(N, DAG, DCI, Subtarget);
break;
[mips] Improve code generation for constant multiplication using shifts, adds and subs. llvm-svn: 185011
2013-06-27 02:48:17 +08:00
case ISD::MUL:
return performMULCombine(N, DAG, DCI, this);
[mips] Instruction selection patterns for DSP-ASE vector shifts. llvm-svn: 179906
2013-04-20 07:21:32 +08:00
case ISD::SHL:
[Mips] Add support to match more patterns for DEXT and CINS This patch adds support for recognizing more patterns to match to DEXT and CINS instructions. It finds cases where multiple instructions could be replaced with a single DEXT or CINS instruction. For example, for the following: define i64 @dext_and32(i64 zeroext %a) { entry: %and = and i64 %a, 4294967295 ret i64 %and } instead of generating: 0000000000000088 <dext_and32>: 88: 64010001 daddiu at,zero,1 8c: 0001083c dsll32 at,at,0x0 90: 6421ffff daddiu at,at,-1 94: 03e00008 jr ra 98: 00811024 and v0,a0,at 9c: 00000000 nop the following gets generated: 0000000000000068 <dext_and32>: 68: 03e00008 jr ra 6c: 7c82f803 dext v0,a0,0x0,0x20 Cases that are covered: DEXT: 1. and $src, mask where mask > 0xffff 2. zext $src zero extend from i32 to i64 CINS: 1. and (shl $src, pos), mask 2. shl (and $src, mask), pos 3. zext (shl $src, pos) zero extend from i32 to i64 Patch by Violeta Vukobrat. Differential Revision: https://reviews.llvm.org/D30464 llvm-svn: 297832
2017-03-15 21:10:08 +08:00
Val = performSHLCombine(N, DAG, DCI, Subtarget);
break;
[mips] Instruction selection patterns for DSP-ASE vector shifts. llvm-svn: 179906
2013-04-20 07:21:32 +08:00
case ISD::SRA:
return performSRACombine(N, DAG, DCI, Subtarget);
case ISD::SRL:
return performSRLCombine(N, DAG, DCI, Subtarget);
[mips] Instruction selection patterns for DSP-ASE vector select and compare instructions. llvm-svn: 180820
2013-05-01 06:37:26 +08:00
case ISD::VSELECT:
return performVSELECTCombine(N, DAG);
[mips][msa] Added support for matching nor from normal IR (i.e. not intrinsics) llvm-svn: 191195
2013-09-23 21:22:24 +08:00
case ISD::XOR:
Val = performXORCombine(N, DAG, Subtarget);
break;
case ISD::SETCC:
[mips] Instruction selection patterns for DSP-ASE vector select and compare instructions. llvm-svn: 180820
2013-05-01 06:37:26 +08:00
Val = performSETCCCombine(N, DAG);
break;
[mips] Move the code which does dag-combine for multiply-add/sub nodes to derived class MipsSETargetLowering. We shouldn't be generating madd/msub nodes if target is Mips16, since Mips16 doesn't have support for multipy-add/sub instructions. llvm-svn: 178404
2013-03-30 09:42:24 +08:00
}
[mips] Instruction selection patterns for DSP-ASE vector select and compare instructions. llvm-svn: 180820
2013-05-01 06:37:26 +08:00
[mips] MipsSETargetLowering now reports DAGCombiner changes when using -debug-only=mips-isel No test since -debug output is intended for developers and not end-users. llvm-svn: 193681
2013-10-30 21:31:27 +08:00
if (Val.getNode()) {
DEBUG(dbgs() << "\nMipsSE DAG Combine:\n";
N->printrWithDepth(dbgs(), &DAG);
dbgs() << "\n=> \n";
Val.getNode()->printrWithDepth(dbgs(), &DAG);
dbgs() << "\n");
[mips] Instruction selection patterns for DSP-ASE vector select and compare instructions. llvm-svn: 180820
2013-05-01 06:37:26 +08:00
return Val;
[mips] MipsSETargetLowering now reports DAGCombiner changes when using -debug-only=mips-isel No test since -debug output is intended for developers and not end-users. llvm-svn: 193681
2013-10-30 21:31:27 +08:00
}
[mips] Instruction selection patterns for DSP-ASE vector select and compare instructions. llvm-svn: 180820
2013-05-01 06:37:26 +08:00
return MipsTargetLowering::PerformDAGCombine(N, DCI);
[mips] Move the code which does dag-combine for multiply-add/sub nodes to derived class MipsSETargetLowering. We shouldn't be generating madd/msub nodes if target is Mips16, since Mips16 doesn't have support for multipy-add/sub instructions. llvm-svn: 178404
2013-03-30 09:42:24 +08:00
}
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
MachineBasicBlock *
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MipsSETargetLowering::EmitInstrWithCustomInserter(MachineInstr &MI,
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
MachineBasicBlock *BB) const {
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
switch (MI.getOpcode()) {
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
default:
return MipsTargetLowering::EmitInstrWithCustomInserter(MI, BB);
case Mips::BPOSGE32_PSEUDO:
return emitBPOSGE32(MI, BB);
[mips][msa] Added bnz.df, bnz.v, bz.df, and bz.v These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes, are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as a branch/mov sequence to evaluate to 0 or 1. Note: The resulting code is sub-optimal since it doesnt seem to be possible to feed the result of an intrinsic directly into a brcond. At the moment it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily evaluates the boolean twice. llvm-svn: 189478
2013-08-28 20:14:50 +08:00
case Mips::SNZ_B_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BNZ_B);
case Mips::SNZ_H_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BNZ_H);
case Mips::SNZ_W_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BNZ_W);
case Mips::SNZ_D_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BNZ_D);
case Mips::SNZ_V_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BNZ_V);
case Mips::SZ_B_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BZ_B);
case Mips::SZ_H_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BZ_H);
case Mips::SZ_W_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BZ_W);
case Mips::SZ_D_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BZ_D);
case Mips::SZ_V_PSEUDO:
return emitMSACBranchPseudo(MI, BB, Mips::BZ_V);
[mips][msa] Implemented extract_vector_elt for v4f32 or v2f64 For v4f32 and v2f64, EXTRACT_VECTOR_ELT is matched by a pseudo-insn which may be expanded to subregister copies and/or instructions as appropriate. llvm-svn: 191514
2013-09-27 20:17:32 +08:00
case Mips::COPY_FW_PSEUDO:
return emitCOPY_FW(MI, BB);
case Mips::COPY_FD_PSEUDO:
return emitCOPY_FD(MI, BB);
[mips][msa] Implemented insert_vector_elt for v4f32 and v2f64. For v4f32 and v2f64, INSERT_VECTOR_ELT is matched by a pseudo-insn which is later expanded to appropriate insve.[wd] insns. llvm-svn: 191515
2013-09-27 20:31:32 +08:00
case Mips::INSERT_FW_PSEUDO:
return emitINSERT_FW(MI, BB);
case Mips::INSERT_FD_PSEUDO:
return emitINSERT_FD(MI, BB);
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
case Mips::INSERT_B_VIDX_PSEUDO:
[mips] Generate code for insert/extract operations when using the N64 ABI and MSA. Summary: When using the N64 ABI, element-indices use the i64 type instead of i32. In many cases, we can use iPTR to account for this but additional patterns and pseudo's are also required. This fixes most (but not quite all) failures in the test-suite when using N64 and MSA together. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9342 llvm-svn: 236494
2015-05-05 18:32:24 +08:00
case Mips::INSERT_B_VIDX64_PSEUDO:
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
return emitINSERT_DF_VIDX(MI, BB, 1, false);
case Mips::INSERT_H_VIDX_PSEUDO:
[mips] Generate code for insert/extract operations when using the N64 ABI and MSA. Summary: When using the N64 ABI, element-indices use the i64 type instead of i32. In many cases, we can use iPTR to account for this but additional patterns and pseudo's are also required. This fixes most (but not quite all) failures in the test-suite when using N64 and MSA together. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9342 llvm-svn: 236494
2015-05-05 18:32:24 +08:00
case Mips::INSERT_H_VIDX64_PSEUDO:
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
return emitINSERT_DF_VIDX(MI, BB, 2, false);
case Mips::INSERT_W_VIDX_PSEUDO:
[mips] Generate code for insert/extract operations when using the N64 ABI and MSA. Summary: When using the N64 ABI, element-indices use the i64 type instead of i32. In many cases, we can use iPTR to account for this but additional patterns and pseudo's are also required. This fixes most (but not quite all) failures in the test-suite when using N64 and MSA together. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9342 llvm-svn: 236494
2015-05-05 18:32:24 +08:00
case Mips::INSERT_W_VIDX64_PSEUDO:
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
return emitINSERT_DF_VIDX(MI, BB, 4, false);
case Mips::INSERT_D_VIDX_PSEUDO:
[mips] Generate code for insert/extract operations when using the N64 ABI and MSA. Summary: When using the N64 ABI, element-indices use the i64 type instead of i32. In many cases, we can use iPTR to account for this but additional patterns and pseudo's are also required. This fixes most (but not quite all) failures in the test-suite when using N64 and MSA together. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9342 llvm-svn: 236494
2015-05-05 18:32:24 +08:00
case Mips::INSERT_D_VIDX64_PSEUDO:
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
return emitINSERT_DF_VIDX(MI, BB, 8, false);
case Mips::INSERT_FW_VIDX_PSEUDO:
[mips] Generate code for insert/extract operations when using the N64 ABI and MSA. Summary: When using the N64 ABI, element-indices use the i64 type instead of i32. In many cases, we can use iPTR to account for this but additional patterns and pseudo's are also required. This fixes most (but not quite all) failures in the test-suite when using N64 and MSA together. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9342 llvm-svn: 236494
2015-05-05 18:32:24 +08:00
case Mips::INSERT_FW_VIDX64_PSEUDO:
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
return emitINSERT_DF_VIDX(MI, BB, 4, true);
case Mips::INSERT_FD_VIDX_PSEUDO:
[mips] Generate code for insert/extract operations when using the N64 ABI and MSA. Summary: When using the N64 ABI, element-indices use the i64 type instead of i32. In many cases, we can use iPTR to account for this but additional patterns and pseudo's are also required. This fixes most (but not quite all) failures in the test-suite when using N64 and MSA together. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9342 llvm-svn: 236494
2015-05-05 18:32:24 +08:00
case Mips::INSERT_FD_VIDX64_PSEUDO:
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
return emitINSERT_DF_VIDX(MI, BB, 8, true);
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
case Mips::FILL_FW_PSEUDO:
return emitFILL_FW(MI, BB);
case Mips::FILL_FD_PSEUDO:
return emitFILL_FD(MI, BB);
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
case Mips::FEXP2_W_1_PSEUDO:
return emitFEXP2_W_1(MI, BB);
case Mips::FEXP2_D_1_PSEUDO:
return emitFEXP2_D_1(MI, BB);
[mips][msa] Implement f16 support The MIPS MSA ASE provides instructions to convert to and from half precision floating point. This patch teaches the MIPS backend to treat f16 as a legal type and how to promote such values to f32 for the usual set of operations. As a result of this, the fexup[lr].w intrinsics no longer crash LLVM during type legalization. Reviewers: zoran.jovanvoic, vkalintiris Differential Revision: https://reviews.llvm.org/D26398 llvm-svn: 287349
2016-11-19 00:17:44 +08:00
case Mips::ST_F16:
return emitST_F16_PSEUDO(MI, BB);
case Mips::LD_F16:
return emitLD_F16_PSEUDO(MI, BB);
case Mips::MSA_FP_EXTEND_W_PSEUDO:
return emitFPEXTEND_PSEUDO(MI, BB, false);
case Mips::MSA_FP_ROUND_W_PSEUDO:
return emitFPROUND_PSEUDO(MI, BB, false);
case Mips::MSA_FP_EXTEND_D_PSEUDO:
return emitFPEXTEND_PSEUDO(MI, BB, true);
case Mips::MSA_FP_ROUND_D_PSEUDO:
return emitFPROUND_PSEUDO(MI, BB, true);
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
}
}
Re-commit r221056 and others with fix, "[mips] Move F128 argument handling into MipsCCState as we did for returns. NFC." sret arguments can never originate from an f128 argument so we detect sret arguments and push false into OriginalArgWasF128. llvm-svn: 221102
2014-11-03 00:09:29 +08:00
bool MipsSETargetLowering::isEligibleForTailCallOptimization(
const CCState &CCInfo, unsigned NextStackOffset,
const MipsFunctionInfo &FI) const {
[mips] Enable tail calls by default Enable tail calls by default for (micro)MIPS(64). microMIPS is slightly more tricky than doing it for MIPS(R6) or microMIPSR6. microMIPS has two instruction encodings: 16bit and 32bit along with some restrictions on the size of the instruction that can fill the delay slot. For safe tail calls for microMIPS, the delay slot filler attempts to find a correct size instruction for the delay slot of TAILCALL pseudos. Reviewers: dsanders, vkalintris Subscribers: jfb, dsanders, sdardis, llvm-commits Differential Revision: https://reviews.llvm.org/D21138 llvm-svn: 277708
2016-08-04 17:17:07 +08:00
if (!UseMipsTailCalls)
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
return false;
[mips] Interrupt attribute support for mips32r2+. Summary: This patch adds support for using the "interrupt" attribute on Mips for interrupt handling functions. At this time only mips32r2+ with the o32 ABI with the static relocation model is supported. Unsupported configurations will be rejected Patch by Simon Dardis (+ clang-format & some trivial changes to follow the LLVM coding standards by me). Reviewers: mpf, dsanders Subscribers: dsanders, vkalintiris, llvm-commits Differential Revision: http://reviews.llvm.org/D10768 llvm-svn: 251286
2015-10-26 20:38:43 +08:00
// Exception has to be cleared with eret.
if (FI.isISR())
return false;
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
// Return false if either the callee or caller has a byval argument.
Re-commit r221056 and others with fix, "[mips] Move F128 argument handling into MipsCCState as we did for returns. NFC." sret arguments can never originate from an f128 argument so we detect sret arguments and push false into OriginalArgWasF128. llvm-svn: 221102
2014-11-03 00:09:29 +08:00
if (CCInfo.getInRegsParamsCount() > 0 || FI.hasByvalArg())
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
return false;
// Return true if the callee's argument area is no larger than the
// caller's.
return NextStackOffset <= FI.getIncomingArgSize();
}
void MipsSETargetLowering::
getOpndList(SmallVectorImpl<SDValue> &Ops,
[Mips] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC). llvm-svn: 309993
2017-08-04 06:12:30 +08:00
std::deque<std::pair<unsigned, SDValue>> &RegsToPass,
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
bool IsPICCall, bool GlobalOrExternal, bool InternalLinkage,
[mips] For indirect calls we don't need $gp to point to .got. Mips linker doesn't generate lazy binding stub for a function whose address is taken in the program. Differential Revision: http://reviews.llvm.org/D5067 llvm-svn: 218744
2014-10-01 16:22:21 +08:00
bool IsCallReloc, CallLoweringInfo &CLI, SDValue Callee,
SDValue Chain) const {
[mips] Implement the following optimizations using dominance information to make PIC calls a little more efficient: 1. Remove instructions setting up $gp if it is known that a function has been called at least once. 2. Save the address of a called function in a register instead of loading it from the GOT at every call site. llvm-svn: 195892
2013-11-28 07:38:42 +08:00
Ops.push_back(Callee);
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
MipsTargetLowering::getOpndList(Ops, RegsToPass, IsPICCall, GlobalOrExternal,
[mips] For indirect calls we don't need $gp to point to .got. Mips linker doesn't generate lazy binding stub for a function whose address is taken in the program. Differential Revision: http://reviews.llvm.org/D5067 llvm-svn: 218744
2014-10-01 16:22:21 +08:00
InternalLinkage, IsCallReloc, CLI, Callee,
Chain);
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
}
[mips] Enhance command line option "-mno-ldc1-sdc1" to expand base+index double precision loads and stores as well as reg+imm double precision loads and stores. Previously, expansion of loads and stores was done after register allocation, but now it takes place during legalization. As a result, users will see double precision stores and loads being emitted to spill and restore 64-bit FP registers. llvm-svn: 190235
2013-09-07 08:52:30 +08:00
SDValue MipsSETargetLowering::lowerLOAD(SDValue Op, SelectionDAG &DAG) const {
LoadSDNode &Nd = *cast<LoadSDNode>(Op);
if (Nd.getMemoryVT() != MVT::f64 || !NoDPLoadStore)
return MipsTargetLowering::lowerLOAD(Op, DAG);
// Replace a double precision load with two i32 loads and a buildpair64.
SDLoc DL(Op);
SDValue Ptr = Nd.getBasePtr(), Chain = Nd.getChain();
EVT PtrVT = Ptr.getValueType();
// i32 load from lower address.
[SelectionDAG] Get rid of bool parameters in SelectionDAG::getLoad, getStore, and friends. Summary: Instead, we take a single flags arg (a bitset). Also add a default 0 alignment, and change the order of arguments so the alignment comes before the flags. This greatly simplifies many callsites, and fixes a bug in AMDGPUISelLowering, wherein the order of the args to getLoad was inverted. It also greatly simplifies the process of adding another flag to getLoad. Reviewers: chandlerc, tstellarAMD Subscribers: jholewinski, arsenm, jyknight, dsanders, nemanjai, llvm-commits Differential Revision: http://reviews.llvm.org/D22249 llvm-svn: 275592
2016-07-16 02:27:10 +08:00
SDValue Lo = DAG.getLoad(MVT::i32, DL, Chain, Ptr, MachinePointerInfo(),
Nd.getAlignment(), Nd.getMemOperand()->getFlags());
[mips] Enhance command line option "-mno-ldc1-sdc1" to expand base+index double precision loads and stores as well as reg+imm double precision loads and stores. Previously, expansion of loads and stores was done after register allocation, but now it takes place during legalization. As a result, users will see double precision stores and loads being emitted to spill and restore 64-bit FP registers. llvm-svn: 190235
2013-09-07 08:52:30 +08:00
// i32 load from higher address.
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
Ptr = DAG.getNode(ISD::ADD, DL, PtrVT, Ptr, DAG.getConstant(4, DL, PtrVT));
[SelectionDAG] Get rid of bool parameters in SelectionDAG::getLoad, getStore, and friends. Summary: Instead, we take a single flags arg (a bitset). Also add a default 0 alignment, and change the order of arguments so the alignment comes before the flags. This greatly simplifies many callsites, and fixes a bug in AMDGPUISelLowering, wherein the order of the args to getLoad was inverted. It also greatly simplifies the process of adding another flag to getLoad. Reviewers: chandlerc, tstellarAMD Subscribers: jholewinski, arsenm, jyknight, dsanders, nemanjai, llvm-commits Differential Revision: http://reviews.llvm.org/D22249 llvm-svn: 275592
2016-07-16 02:27:10 +08:00
SDValue Hi = DAG.getLoad(
MVT::i32, DL, Lo.getValue(1), Ptr, MachinePointerInfo(),
std::min(Nd.getAlignment(), 4U), Nd.getMemOperand()->getFlags());
[mips] Enhance command line option "-mno-ldc1-sdc1" to expand base+index double precision loads and stores as well as reg+imm double precision loads and stores. Previously, expansion of loads and stores was done after register allocation, but now it takes place during legalization. As a result, users will see double precision stores and loads being emitted to spill and restore 64-bit FP registers. llvm-svn: 190235
2013-09-07 08:52:30 +08:00
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (!Subtarget.isLittle())
[mips] Enhance command line option "-mno-ldc1-sdc1" to expand base+index double precision loads and stores as well as reg+imm double precision loads and stores. Previously, expansion of loads and stores was done after register allocation, but now it takes place during legalization. As a result, users will see double precision stores and loads being emitted to spill and restore 64-bit FP registers. llvm-svn: 190235
2013-09-07 08:52:30 +08:00
std::swap(Lo, Hi);
SDValue BP = DAG.getNode(MipsISD::BuildPairF64, DL, MVT::f64, Lo, Hi);
SDValue Ops[2] = {BP, Hi.getValue(1)};
Convert SelectionDAG::getMergeValues to use ArrayRef. llvm-svn: 207374
2014-04-28 03:20:57 +08:00
return DAG.getMergeValues(Ops, DL);
[mips] Enhance command line option "-mno-ldc1-sdc1" to expand base+index double precision loads and stores as well as reg+imm double precision loads and stores. Previously, expansion of loads and stores was done after register allocation, but now it takes place during legalization. As a result, users will see double precision stores and loads being emitted to spill and restore 64-bit FP registers. llvm-svn: 190235
2013-09-07 08:52:30 +08:00
}
SDValue MipsSETargetLowering::lowerSTORE(SDValue Op, SelectionDAG &DAG) const {
StoreSDNode &Nd = *cast<StoreSDNode>(Op);
if (Nd.getMemoryVT() != MVT::f64 || !NoDPLoadStore)
return MipsTargetLowering::lowerSTORE(Op, DAG);
// Replace a double precision store with two extractelement64s and i32 stores.
SDLoc DL(Op);
SDValue Val = Nd.getValue(), Ptr = Nd.getBasePtr(), Chain = Nd.getChain();
EVT PtrVT = Ptr.getValueType();
SDValue Lo = DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32,
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
Val, DAG.getConstant(0, DL, MVT::i32));
[mips] Enhance command line option "-mno-ldc1-sdc1" to expand base+index double precision loads and stores as well as reg+imm double precision loads and stores. Previously, expansion of loads and stores was done after register allocation, but now it takes place during legalization. As a result, users will see double precision stores and loads being emitted to spill and restore 64-bit FP registers. llvm-svn: 190235
2013-09-07 08:52:30 +08:00
SDValue Hi = DAG.getNode(MipsISD::ExtractElementF64, DL, MVT::i32,
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
Val, DAG.getConstant(1, DL, MVT::i32));
[mips] Enhance command line option "-mno-ldc1-sdc1" to expand base+index double precision loads and stores as well as reg+imm double precision loads and stores. Previously, expansion of loads and stores was done after register allocation, but now it takes place during legalization. As a result, users will see double precision stores and loads being emitted to spill and restore 64-bit FP registers. llvm-svn: 190235
2013-09-07 08:52:30 +08:00
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (!Subtarget.isLittle())
[mips] Enhance command line option "-mno-ldc1-sdc1" to expand base+index double precision loads and stores as well as reg+imm double precision loads and stores. Previously, expansion of loads and stores was done after register allocation, but now it takes place during legalization. As a result, users will see double precision stores and loads being emitted to spill and restore 64-bit FP registers. llvm-svn: 190235
2013-09-07 08:52:30 +08:00
std::swap(Lo, Hi);
// i32 store to lower address.
[SelectionDAG] Get rid of bool parameters in SelectionDAG::getLoad, getStore, and friends. Summary: Instead, we take a single flags arg (a bitset). Also add a default 0 alignment, and change the order of arguments so the alignment comes before the flags. This greatly simplifies many callsites, and fixes a bug in AMDGPUISelLowering, wherein the order of the args to getLoad was inverted. It also greatly simplifies the process of adding another flag to getLoad. Reviewers: chandlerc, tstellarAMD Subscribers: jholewinski, arsenm, jyknight, dsanders, nemanjai, llvm-commits Differential Revision: http://reviews.llvm.org/D22249 llvm-svn: 275592
2016-07-16 02:27:10 +08:00
Chain =
DAG.getStore(Chain, DL, Lo, Ptr, MachinePointerInfo(), Nd.getAlignment(),
Nd.getMemOperand()->getFlags(), Nd.getAAInfo());
[mips] Enhance command line option "-mno-ldc1-sdc1" to expand base+index double precision loads and stores as well as reg+imm double precision loads and stores. Previously, expansion of loads and stores was done after register allocation, but now it takes place during legalization. As a result, users will see double precision stores and loads being emitted to spill and restore 64-bit FP registers. llvm-svn: 190235
2013-09-07 08:52:30 +08:00
// i32 store to higher address.
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
Ptr = DAG.getNode(ISD::ADD, DL, PtrVT, Ptr, DAG.getConstant(4, DL, PtrVT));
[mips] Enhance command line option "-mno-ldc1-sdc1" to expand base+index double precision loads and stores as well as reg+imm double precision loads and stores. Previously, expansion of loads and stores was done after register allocation, but now it takes place during legalization. As a result, users will see double precision stores and loads being emitted to spill and restore 64-bit FP registers. llvm-svn: 190235
2013-09-07 08:52:30 +08:00
return DAG.getStore(Chain, DL, Hi, Ptr, MachinePointerInfo(),
[SelectionDAG] Get rid of bool parameters in SelectionDAG::getLoad, getStore, and friends. Summary: Instead, we take a single flags arg (a bitset). Also add a default 0 alignment, and change the order of arguments so the alignment comes before the flags. This greatly simplifies many callsites, and fixes a bug in AMDGPUISelLowering, wherein the order of the args to getLoad was inverted. It also greatly simplifies the process of adding another flag to getLoad. Reviewers: chandlerc, tstellarAMD Subscribers: jholewinski, arsenm, jyknight, dsanders, nemanjai, llvm-commits Differential Revision: http://reviews.llvm.org/D22249 llvm-svn: 275592
2016-07-16 02:27:10 +08:00
std::min(Nd.getAlignment(), 4U),
Nd.getMemOperand()->getFlags(), Nd.getAAInfo());
[mips] Enhance command line option "-mno-ldc1-sdc1" to expand base+index double precision loads and stores as well as reg+imm double precision loads and stores. Previously, expansion of loads and stores was done after register allocation, but now it takes place during legalization. As a result, users will see double precision stores and loads being emitted to spill and restore 64-bit FP registers. llvm-svn: 190235
2013-09-07 08:52:30 +08:00
}
[mips] Fix definitions of multiply, multiply-add/sub and divide instructions. The new instructions have explicit register output operands and use table-gen patterns instead of C++ code to do instruction selection. Mips16's instructions are unaffected by this change. llvm-svn: 178403
2013-03-30 09:36:35 +08:00
SDValue MipsSETargetLowering::lowerMulDiv(SDValue Op, unsigned NewOpc,
bool HasLo, bool HasHi,
SelectionDAG &DAG) const {
[mips][mips64r6] Replace m[tf]hi, m[tf]lo, mult, multu, dmult, dmultu, div, ddiv, divu, ddivu for MIPS32r6/MIPS64. Summary: The accumulator-based (HI/LO) multiplies and divides from earlier ISA's have been removed and replaced with GPR-based equivalents. For example: div $1, $2 mflo $3 is now: div $3, $1, $2 This patch disables the accumulator-based multiplies and divides for MIPS32r6/MIPS64r6 and uses the GPR-based equivalents instead. Renamed expandPseudoDiv to insertDivByZeroTrap to better describe the behaviour of the function. MipsDelaySlotFiller now invalidates the liveness information when moving instructions to the delay slot. Without this, divrem.ll will abort since %GP ends up used before it is defined. Reviewers: vmedic, zoran.jovanovic, jkolek Reviewed By: jkolek Differential Revision: http://reviews.llvm.org/D3896 llvm-svn: 210760
2014-06-12 18:44:10 +08:00
// MIPS32r6/MIPS64r6 removed accumulator based multiplies.
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
assert(!Subtarget.hasMips32r6());
[mips][mips64r6] Replace m[tf]hi, m[tf]lo, mult, multu, dmult, dmultu, div, ddiv, divu, ddivu for MIPS32r6/MIPS64. Summary: The accumulator-based (HI/LO) multiplies and divides from earlier ISA's have been removed and replaced with GPR-based equivalents. For example: div $1, $2 mflo $3 is now: div $3, $1, $2 This patch disables the accumulator-based multiplies and divides for MIPS32r6/MIPS64r6 and uses the GPR-based equivalents instead. Renamed expandPseudoDiv to insertDivByZeroTrap to better describe the behaviour of the function. MipsDelaySlotFiller now invalidates the liveness information when moving instructions to the delay slot. Without this, divrem.ll will abort since %GP ends up used before it is defined. Reviewers: vmedic, zoran.jovanovic, jkolek Reviewed By: jkolek Differential Revision: http://reviews.llvm.org/D3896 llvm-svn: 210760
2014-06-12 18:44:10 +08:00
[mips] Fix definitions of multiply, multiply-add/sub and divide instructions. The new instructions have explicit register output operands and use table-gen patterns instead of C++ code to do instruction selection. Mips16's instructions are unaffected by this change. llvm-svn: 178403
2013-03-30 09:36:35 +08:00
EVT Ty = Op.getOperand(0).getValueType();
Track IR ordering of SelectionDAG nodes 2/4. Change SelectionDAG::getXXXNode() interfaces as well as call sites of these functions to pass in SDLoc instead of DebugLoc. llvm-svn: 182703
2013-05-25 10:42:55 +08:00
SDLoc DL(Op);
[mips] Fix definitions of multiply, multiply-add/sub and divide instructions. The new instructions have explicit register output operands and use table-gen patterns instead of C++ code to do instruction selection. Mips16's instructions are unaffected by this change. llvm-svn: 178403
2013-03-30 09:36:35 +08:00
SDValue Mult = DAG.getNode(NewOpc, DL, MVT::Untyped,
Op.getOperand(0), Op.getOperand(1));
SDValue Lo, Hi;
if (HasLo)
[mips] Rename isel nodes. llvm-svn: 192663
2013-10-15 09:12:50 +08:00
Lo = DAG.getNode(MipsISD::MFLO, DL, Ty, Mult);
[mips] Fix definitions of multiply, multiply-add/sub and divide instructions. The new instructions have explicit register output operands and use table-gen patterns instead of C++ code to do instruction selection. Mips16's instructions are unaffected by this change. llvm-svn: 178403
2013-03-30 09:36:35 +08:00
if (HasHi)
[mips] Rename isel nodes. llvm-svn: 192663
2013-10-15 09:12:50 +08:00
Hi = DAG.getNode(MipsISD::MFHI, DL, Ty, Mult);
[mips] Fix definitions of multiply, multiply-add/sub and divide instructions. The new instructions have explicit register output operands and use table-gen patterns instead of C++ code to do instruction selection. Mips16's instructions are unaffected by this change. llvm-svn: 178403
2013-03-30 09:36:35 +08:00
if (!HasLo || !HasHi)
return HasLo ? Lo : Hi;
SDValue Vals[] = { Lo, Hi };
Convert SelectionDAG::getMergeValues to use ArrayRef. llvm-svn: 207374
2014-04-28 03:20:57 +08:00
return DAG.getMergeValues(Vals, DL);
[mips] Fix definitions of multiply, multiply-add/sub and divide instructions. The new instructions have explicit register output operands and use table-gen patterns instead of C++ code to do instruction selection. Mips16's instructions are unaffected by this change. llvm-svn: 178403
2013-03-30 09:36:35 +08:00
}
Pass DebugLoc and SDLoc by const ref. This used to be free, copying and moving DebugLocs became expensive after the metadata rewrite. Passing by reference eliminates a ton of track/untrack operations. No functionality change intended. llvm-svn: 272512
2016-06-12 23:39:02 +08:00
static SDValue initAccumulator(SDValue In, const SDLoc &DL, SelectionDAG &DAG) {
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
SDValue InLo = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, In,
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
DAG.getConstant(0, DL, MVT::i32));
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
SDValue InHi = DAG.getNode(ISD::EXTRACT_ELEMENT, DL, MVT::i32, In,
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
DAG.getConstant(1, DL, MVT::i32));
[mips] Rename isel nodes. llvm-svn: 192663
2013-10-15 09:12:50 +08:00
return DAG.getNode(MipsISD::MTLOHI, DL, MVT::Untyped, InLo, InHi);
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
}
Pass DebugLoc and SDLoc by const ref. This used to be free, copying and moving DebugLocs became expensive after the metadata rewrite. Passing by reference eliminates a ton of track/untrack operations. No functionality change intended. llvm-svn: 272512
2016-06-12 23:39:02 +08:00
static SDValue extractLOHI(SDValue Op, const SDLoc &DL, SelectionDAG &DAG) {
[mips] Rename isel nodes. llvm-svn: 192663
2013-10-15 09:12:50 +08:00
SDValue Lo = DAG.getNode(MipsISD::MFLO, DL, MVT::i32, Op);
SDValue Hi = DAG.getNode(MipsISD::MFHI, DL, MVT::i32, Op);
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
return DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, Lo, Hi);
}
// This function expands mips intrinsic nodes which have 64-bit input operands
// or output values.
//
// out64 = intrinsic-node in64
// =>
// lo = copy (extract-element (in64, 0))
// hi = copy (extract-element (in64, 1))
// mips-specific-node
// v0 = copy lo
// v1 = copy hi
// out64 = merge-values (v0, v1)
//
static SDValue lowerDSPIntr(SDValue Op, SelectionDAG &DAG, unsigned Opc) {
Track IR ordering of SelectionDAG nodes 2/4. Change SelectionDAG::getXXXNode() interfaces as well as call sites of these functions to pass in SDLoc instead of DebugLoc. llvm-svn: 182703
2013-05-25 10:42:55 +08:00
SDLoc DL(Op);
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
bool HasChainIn = Op->getOperand(0).getValueType() == MVT::Other;
SmallVector<SDValue, 3> Ops;
unsigned OpNo = 0;
// See if Op has a chain input.
if (HasChainIn)
Ops.push_back(Op->getOperand(OpNo++));
// The next operand is the intrinsic opcode.
assert(Op->getOperand(OpNo).getOpcode() == ISD::TargetConstant);
// See if the next operand has type i64.
SDValue Opnd = Op->getOperand(++OpNo), In64;
if (Opnd.getValueType() == MVT::i64)
In64 = initAccumulator(Opnd, DL, DAG);
else
Ops.push_back(Opnd);
// Push the remaining operands.
for (++OpNo ; OpNo < Op->getNumOperands(); ++OpNo)
Ops.push_back(Op->getOperand(OpNo));
// Add In64 to the end of the list.
if (In64.getNode())
Ops.push_back(In64);
// Scan output.
SmallVector<EVT, 2> ResTys;
for (SDNode::value_iterator I = Op->value_begin(), E = Op->value_end();
I != E; ++I)
ResTys.push_back((*I == MVT::i64) ? MVT::Untyped : *I);
// Create node.
Convert SelectionDAG::getNode methods to use ArrayRef<SDValue>. llvm-svn: 207327
2014-04-27 02:35:24 +08:00
SDValue Val = DAG.getNode(Opc, DL, ResTys, Ops);
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
SDValue Out = (ResTys[0] == MVT::Untyped) ? extractLOHI(Val, DL, DAG) : Val;
if (!HasChainIn)
return Out;
assert(Val->getValueType(1) == MVT::Other);
SDValue Vals[] = { Out, SDValue(Val.getNode(), 1) };
Convert SelectionDAG::getMergeValues to use ArrayRef. llvm-svn: 207374
2014-04-28 03:20:57 +08:00
return DAG.getMergeValues(Vals, DL);
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
}
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
// Lower an MSA copy intrinsic into the specified SelectionDAG node
static SDValue lowerMSACopyIntr(SDValue Op, SelectionDAG &DAG, unsigned Opc) {
SDLoc DL(Op);
SDValue Vec = Op->getOperand(1);
SDValue Idx = Op->getOperand(2);
EVT ResTy = Op->getValueType(0);
EVT EltTy = Vec->getValueType(0).getVectorElementType();
SDValue Result = DAG.getNode(Opc, DL, ResTy, Vec, Idx,
DAG.getValueType(EltTy));
return Result;
}
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
static SDValue lowerMSASplatZExt(SDValue Op, unsigned OpNr, SelectionDAG &DAG) {
EVT ResVecTy = Op->getValueType(0);
EVT ViaVecTy = ResVecTy;
[mips][msa] Splat.d endianness check Before this change, it was always the first element of a vector that got splatted since the lower 6 bits of vshf.d $wd were always zero for little endian. Additionally, masking has been performed for vshf via which splat.d is created. Vshf has a property where if its first operand's elements have either bit 6 or 7 set, destination element is set to zero. Initially masked with 63 to avoid this property, which would result in generation of and.v + vshf.d in all cases. Masking with one results in generating a single splati.d instruction when possible. Differential Revision: https://reviews.llvm.org/D32216 llvm-svn: 306090
2017-06-23 17:09:31 +08:00
bool BigEndian = !DAG.getSubtarget().getTargetTriple().isLittleEndian();
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
SDLoc DL(Op);
// When ResVecTy == MVT::v2i64, LaneA is the upper 32 bits of the lane and
// LaneB is the lower 32-bits. Otherwise LaneA and LaneB are alternating
// lanes.
[mips][msa] Splat.d endianness check Before this change, it was always the first element of a vector that got splatted since the lower 6 bits of vshf.d $wd were always zero for little endian. Additionally, masking has been performed for vshf via which splat.d is created. Vshf has a property where if its first operand's elements have either bit 6 or 7 set, destination element is set to zero. Initially masked with 63 to avoid this property, which would result in generation of and.v + vshf.d in all cases. Masking with one results in generating a single splati.d instruction when possible. Differential Revision: https://reviews.llvm.org/D32216 llvm-svn: 306090
2017-06-23 17:09:31 +08:00
SDValue LaneA = Op->getOperand(OpNr);
SDValue LaneB;
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
if (ResVecTy == MVT::v2i64) {
[mips][msa] Splat.d endianness check Before this change, it was always the first element of a vector that got splatted since the lower 6 bits of vshf.d $wd were always zero for little endian. Additionally, masking has been performed for vshf via which splat.d is created. Vshf has a property where if its first operand's elements have either bit 6 or 7 set, destination element is set to zero. Initially masked with 63 to avoid this property, which would result in generation of and.v + vshf.d in all cases. Masking with one results in generating a single splati.d instruction when possible. Differential Revision: https://reviews.llvm.org/D32216 llvm-svn: 306090
2017-06-23 17:09:31 +08:00
LaneB = DAG.getConstant(0, DL, MVT::i32);
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
ViaVecTy = MVT::v4i32;
[mips][msa] Splat.d endianness check Before this change, it was always the first element of a vector that got splatted since the lower 6 bits of vshf.d $wd were always zero for little endian. Additionally, masking has been performed for vshf via which splat.d is created. Vshf has a property where if its first operand's elements have either bit 6 or 7 set, destination element is set to zero. Initially masked with 63 to avoid this property, which would result in generation of and.v + vshf.d in all cases. Masking with one results in generating a single splati.d instruction when possible. Differential Revision: https://reviews.llvm.org/D32216 llvm-svn: 306090
2017-06-23 17:09:31 +08:00
if(BigEndian)
std::swap(LaneA, LaneB);
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
} else
[mips][msa] Splat.d endianness check Before this change, it was always the first element of a vector that got splatted since the lower 6 bits of vshf.d $wd were always zero for little endian. Additionally, masking has been performed for vshf via which splat.d is created. Vshf has a property where if its first operand's elements have either bit 6 or 7 set, destination element is set to zero. Initially masked with 63 to avoid this property, which would result in generation of and.v + vshf.d in all cases. Masking with one results in generating a single splati.d instruction when possible. Differential Revision: https://reviews.llvm.org/D32216 llvm-svn: 306090
2017-06-23 17:09:31 +08:00
LaneB = LaneA;
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
SDValue Ops[16] = { LaneA, LaneB, LaneA, LaneB, LaneA, LaneB, LaneA, LaneB,
LaneA, LaneB, LaneA, LaneB, LaneA, LaneB, LaneA, LaneB };
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
[CodeGen] Add getBuildVector and getSplatBuildVector helpers. NFCI. Differential Revision: http://reviews.llvm.org/D17176 llvm-svn: 267606
2016-04-27 05:15:30 +08:00
SDValue Result = DAG.getBuildVector(
ViaVecTy, DL, makeArrayRef(Ops, ViaVecTy.getVectorNumElements()));
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
[mips][msa] Splat.d endianness check Before this change, it was always the first element of a vector that got splatted since the lower 6 bits of vshf.d $wd were always zero for little endian. Additionally, masking has been performed for vshf via which splat.d is created. Vshf has a property where if its first operand's elements have either bit 6 or 7 set, destination element is set to zero. Initially masked with 63 to avoid this property, which would result in generation of and.v + vshf.d in all cases. Masking with one results in generating a single splati.d instruction when possible. Differential Revision: https://reviews.llvm.org/D32216 llvm-svn: 306090
2017-06-23 17:09:31 +08:00
if (ViaVecTy != ResVecTy) {
SDValue One = DAG.getConstant(1, DL, ViaVecTy);
Result = DAG.getNode(ISD::BITCAST, DL, ResVecTy,
DAG.getNode(ISD::AND, DL, ViaVecTy, Result, One));
}
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
return Result;
}
[mips][msa] Added support for matching addvi, and subvi from normal IR (i.e. not intrinsics) llvm-svn: 191203
2013-09-23 22:29:55 +08:00
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
static SDValue lowerMSASplatImm(SDValue Op, unsigned ImmOp, SelectionDAG &DAG,
bool IsSigned = false) {
return DAG.getConstant(
APInt(Op->getValueType(0).getScalarType().getSizeInBits(),
Op->getConstantOperandVal(ImmOp), IsSigned),
SDLoc(Op), Op->getValueType(0));
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
}
static SDValue getBuildVectorSplat(EVT VecTy, SDValue SplatValue,
bool BigEndian, SelectionDAG &DAG) {
EVT ViaVecTy = VecTy;
SDValue SplatValueA = SplatValue;
SDValue SplatValueB = SplatValue;
SDLoc DL(SplatValue);
if (VecTy == MVT::v2i64) {
// v2i64 BUILD_VECTOR must be performed via v4i32 so split into i32's.
ViaVecTy = MVT::v4i32;
SplatValueA = DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, SplatValue);
SplatValueB = DAG.getNode(ISD::SRL, DL, MVT::i64, SplatValue,
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
DAG.getConstant(32, DL, MVT::i32));
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
SplatValueB = DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, SplatValueB);
}
// We currently hold the parts in little endian order. Swap them if
// necessary.
if (BigEndian)
std::swap(SplatValueA, SplatValueB);
SDValue Ops[16] = { SplatValueA, SplatValueB, SplatValueA, SplatValueB,
SplatValueA, SplatValueB, SplatValueA, SplatValueB,
SplatValueA, SplatValueB, SplatValueA, SplatValueB,
SplatValueA, SplatValueB, SplatValueA, SplatValueB };
[CodeGen] Add getBuildVector and getSplatBuildVector helpers. NFCI. Differential Revision: http://reviews.llvm.org/D17176 llvm-svn: 267606
2016-04-27 05:15:30 +08:00
SDValue Result = DAG.getBuildVector(
ViaVecTy, DL, makeArrayRef(Ops, ViaVecTy.getVectorNumElements()));
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
if (VecTy != ViaVecTy)
Result = DAG.getNode(ISD::BITCAST, DL, VecTy, Result);
return Result;
[mips][msa] Added support for matching fadd, fdiv, flog2, fmul, frint, fsqrt, and fsub from normal IR (i.e. not intrinsics) llvm-svn: 190512
2013-09-11 18:51:30 +08:00
}
[mips][msa] Added support for matching bset, bseti, bneg, and bnegi from normal IR (i.e. not intrinsics) llvm-svn: 194469
2013-11-12 18:31:49 +08:00
static SDValue lowerMSABinaryBitImmIntr(SDValue Op, SelectionDAG &DAG,
unsigned Opc, SDValue Imm,
bool BigEndian) {
EVT VecTy = Op->getValueType(0);
SDValue Exp2Imm;
SDLoc DL(Op);
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
// The DAG Combiner can't constant fold bitcasted vectors yet so we must do it
// here for now.
[mips][msa] Added support for matching bset, bseti, bneg, and bnegi from normal IR (i.e. not intrinsics) llvm-svn: 194469
2013-11-12 18:31:49 +08:00
if (VecTy == MVT::v2i64) {
if (ConstantSDNode *CImm = dyn_cast<ConstantSDNode>(Imm)) {
APInt BitImm = APInt(64, 1) << CImm->getAPIntValue();
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
SDValue BitImmHiOp = DAG.getConstant(BitImm.lshr(32).trunc(32), DL,
MVT::i32);
SDValue BitImmLoOp = DAG.getConstant(BitImm.trunc(32), DL, MVT::i32);
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
if (BigEndian)
std::swap(BitImmLoOp, BitImmHiOp);
[CodeGen] Add getBuildVector and getSplatBuildVector helpers. NFCI. Differential Revision: http://reviews.llvm.org/D17176 llvm-svn: 267606
2016-04-27 05:15:30 +08:00
Exp2Imm = DAG.getNode(
ISD::BITCAST, DL, MVT::v2i64,
DAG.getBuildVector(MVT::v4i32, DL,
{BitImmLoOp, BitImmHiOp, BitImmLoOp, BitImmHiOp}));
[mips][msa] Added support for matching bset, bseti, bneg, and bnegi from normal IR (i.e. not intrinsics) llvm-svn: 194469
2013-11-12 18:31:49 +08:00
}
}
[C++] Use 'nullptr'. Target edition. llvm-svn: 207197
2014-04-25 13:30:21 +08:00
if (!Exp2Imm.getNode()) {
[mips][msa] Added support for matching bset, bseti, bneg, and bnegi from normal IR (i.e. not intrinsics) llvm-svn: 194469
2013-11-12 18:31:49 +08:00
// We couldnt constant fold, do a vector shift instead
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
// Extend i32 to i64 if necessary. Sign or zero extend doesn't matter since
// only values 0-63 are valid.
if (VecTy == MVT::v2i64)
Imm = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i64, Imm);
Exp2Imm = getBuildVectorSplat(VecTy, Imm, BigEndian, DAG);
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
Exp2Imm = DAG.getNode(ISD::SHL, DL, VecTy, DAG.getConstant(1, DL, VecTy),
Exp2Imm);
[mips][msa] Added support for matching bset, bseti, bneg, and bnegi from normal IR (i.e. not intrinsics) llvm-svn: 194469
2013-11-12 18:31:49 +08:00
}
return DAG.getNode(Opc, DL, VecTy, Op->getOperand(1), Exp2Imm);
}
[mips][msa] Mask vectors holding shift amounts Masked vectors which hold shift amounts when creating the following nodes: ISD::SHL, ISD::SRL or ISD::SRA. Instructions that use said nodes, which have had their arguments altered are sll, srl, sra, bneg, bclr and bset. For said instructions, the shift amount or the bit position that is specified in the corresponding vector elements will be interpreted as the shift amount/bit position modulo the size of the element in bits. The problem lies in compiling with -O2 enabled, where the instructions for formats .w and .d are not generated, but are instead optimized away. In this case, having shift amounts that are either negative or greater than the element bit size results in generation of incorrect results when constant folding. We remedy this by masking the operands for the nodes mentioned above before actually creating them, so that the final result is correct before placed into the constant pool. Patch by Stefan Maksimovic. Differential Revision: https://reviews.llvm.org/D31331 llvm-svn: 300839
2017-04-20 21:26:46 +08:00
static SDValue truncateVecElts(SDValue Op, SelectionDAG &DAG) {
SDLoc DL(Op);
EVT ResTy = Op->getValueType(0);
SDValue Vec = Op->getOperand(2);
bool BigEndian = !DAG.getSubtarget().getTargetTriple().isLittleEndian();
MVT ResEltTy = ResTy == MVT::v2i64 ? MVT::i64 : MVT::i32;
SDValue ConstValue = DAG.getConstant(Vec.getScalarValueSizeInBits() - 1,
DL, ResEltTy);
SDValue SplatVec = getBuildVectorSplat(ResTy, ConstValue, BigEndian, DAG);
return DAG.getNode(ISD::AND, DL, ResTy, Vec, SplatVec);
}
[mips][msa] Added support for matching bclr, and bclri from normal IR (i.e. not intrinsics) llvm-svn: 194471
2013-11-12 18:45:18 +08:00
static SDValue lowerMSABitClear(SDValue Op, SelectionDAG &DAG) {
EVT ResTy = Op->getValueType(0);
SDLoc DL(Op);
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
SDValue One = DAG.getConstant(1, DL, ResTy);
[mips][msa] Mask vectors holding shift amounts Masked vectors which hold shift amounts when creating the following nodes: ISD::SHL, ISD::SRL or ISD::SRA. Instructions that use said nodes, which have had their arguments altered are sll, srl, sra, bneg, bclr and bset. For said instructions, the shift amount or the bit position that is specified in the corresponding vector elements will be interpreted as the shift amount/bit position modulo the size of the element in bits. The problem lies in compiling with -O2 enabled, where the instructions for formats .w and .d are not generated, but are instead optimized away. In this case, having shift amounts that are either negative or greater than the element bit size results in generation of incorrect results when constant folding. We remedy this by masking the operands for the nodes mentioned above before actually creating them, so that the final result is correct before placed into the constant pool. Patch by Stefan Maksimovic. Differential Revision: https://reviews.llvm.org/D31331 llvm-svn: 300839
2017-04-20 21:26:46 +08:00
SDValue Bit = DAG.getNode(ISD::SHL, DL, ResTy, One, truncateVecElts(Op, DAG));
[mips][msa] Added support for matching bclr, and bclri from normal IR (i.e. not intrinsics) llvm-svn: 194471
2013-11-12 18:45:18 +08:00
[mips][msa] lowerMSABitClear() should use SelectionDAG::getNOT() instead of using a long-winded equivalent. Now that getConstant(-1, MVT::v2i64) works correctly on MIPS32 we can use SelectionDAG::getNOT() to produce the bitmask. llvm-svn: 194819
2013-11-16 00:02:04 +08:00
return DAG.getNode(ISD::AND, DL, ResTy, Op->getOperand(1),
DAG.getNOT(DL, Bit, ResTy));
[mips][msa] Added support for matching bclr, and bclri from normal IR (i.e. not intrinsics) llvm-svn: 194471
2013-11-12 18:45:18 +08:00
}
static SDValue lowerMSABitClearImm(SDValue Op, SelectionDAG &DAG) {
SDLoc DL(Op);
EVT ResTy = Op->getValueType(0);
getVectorElementType().getSizeInBits() -> getScalarSizeInBits() ; NFCI llvm-svn: 281495
2016-09-15 00:37:15 +08:00
APInt BitImm = APInt(ResTy.getScalarSizeInBits(), 1)
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
<< cast<ConstantSDNode>(Op->getOperand(2))->getAPIntValue();
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
SDValue BitMask = DAG.getConstant(~BitImm, DL, ResTy);
[mips][msa] Added support for matching bclr, and bclri from normal IR (i.e. not intrinsics) llvm-svn: 194471
2013-11-12 18:45:18 +08:00
return DAG.getNode(ISD::AND, DL, ResTy, Op->getOperand(1), BitMask);
}
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
SDValue MipsSETargetLowering::lowerINTRINSIC_WO_CHAIN(SDValue Op,
SelectionDAG &DAG) const {
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
SDLoc DL(Op);
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
unsigned Intrinsic = cast<ConstantSDNode>(Op->getOperand(0))->getZExtValue();
switch (Intrinsic) {
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
default:
return SDValue();
case Intrinsic::mips_shilo:
return lowerDSPIntr(Op, DAG, MipsISD::SHILO);
case Intrinsic::mips_dpau_h_qbl:
return lowerDSPIntr(Op, DAG, MipsISD::DPAU_H_QBL);
case Intrinsic::mips_dpau_h_qbr:
return lowerDSPIntr(Op, DAG, MipsISD::DPAU_H_QBR);
case Intrinsic::mips_dpsu_h_qbl:
return lowerDSPIntr(Op, DAG, MipsISD::DPSU_H_QBL);
case Intrinsic::mips_dpsu_h_qbr:
return lowerDSPIntr(Op, DAG, MipsISD::DPSU_H_QBR);
case Intrinsic::mips_dpa_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPA_W_PH);
case Intrinsic::mips_dps_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPS_W_PH);
case Intrinsic::mips_dpax_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPAX_W_PH);
case Intrinsic::mips_dpsx_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPSX_W_PH);
case Intrinsic::mips_mulsa_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::MULSA_W_PH);
case Intrinsic::mips_mult:
return lowerDSPIntr(Op, DAG, MipsISD::Mult);
case Intrinsic::mips_multu:
return lowerDSPIntr(Op, DAG, MipsISD::Multu);
case Intrinsic::mips_madd:
return lowerDSPIntr(Op, DAG, MipsISD::MAdd);
case Intrinsic::mips_maddu:
return lowerDSPIntr(Op, DAG, MipsISD::MAddu);
case Intrinsic::mips_msub:
return lowerDSPIntr(Op, DAG, MipsISD::MSub);
case Intrinsic::mips_msubu:
return lowerDSPIntr(Op, DAG, MipsISD::MSubu);
[mips][msa] Added support for matching addv from normal IR (i.e. not intrinsics) The corresponding intrinsic is now lowered into equivalent IR (ISD::ADD) before instruction selection. llvm-svn: 190507
2013-09-11 18:28:16 +08:00
case Intrinsic::mips_addv_b:
case Intrinsic::mips_addv_h:
case Intrinsic::mips_addv_w:
case Intrinsic::mips_addv_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::ADD, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
[mips][msa] Added support for matching addvi, and subvi from normal IR (i.e. not intrinsics) llvm-svn: 191203
2013-09-23 22:29:55 +08:00
case Intrinsic::mips_addvi_b:
case Intrinsic::mips_addvi_h:
case Intrinsic::mips_addvi_w:
case Intrinsic::mips_addvi_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::ADD, DL, Op->getValueType(0), Op->getOperand(1),
lowerMSASplatImm(Op, 2, DAG));
[mips][msa] Added support for matching and, or, and xor from normal IR (i.e. not intrinsics) llvm-svn: 191194
2013-09-23 20:57:42 +08:00
case Intrinsic::mips_and_v:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::AND, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
[mips][msa] Added support for matching andi, ori, nori, and xori from normal IR (i.e. not intrinsics) llvm-svn: 191293
2013-09-24 20:32:47 +08:00
case Intrinsic::mips_andi_b:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::AND, DL, Op->getValueType(0), Op->getOperand(1),
lowerMSASplatImm(Op, 2, DAG));
[mips][msa] Added support for matching bclr, and bclri from normal IR (i.e. not intrinsics) llvm-svn: 194471
2013-11-12 18:45:18 +08:00
case Intrinsic::mips_bclr_b:
case Intrinsic::mips_bclr_h:
case Intrinsic::mips_bclr_w:
case Intrinsic::mips_bclr_d:
return lowerMSABitClear(Op, DAG);
case Intrinsic::mips_bclri_b:
case Intrinsic::mips_bclri_h:
case Intrinsic::mips_bclri_w:
case Intrinsic::mips_bclri_d:
return lowerMSABitClearImm(Op, DAG);
[mips][msa] Added support for matching bins[lr]i.[bhwd] from normal IR (i.e. not intrinsics) This required correcting the definition of the bins[lr]i intrinsics because the result is also the first operand. It also required removing the (arbitrary) check for 32-bit immediates in MipsSEDAGToDAGISel::selectVSplat(). Currently using binsli.d with 2 bits set in the mask doesn't select binsli.d because the constant is legalized into a ConstantPool. Similar things can happen with binsri.d with more than 10 bits set in the mask. The resulting code when this happens is correct but not optimal. llvm-svn: 193687
2013-10-30 22:45:14 +08:00
case Intrinsic::mips_binsli_b:
case Intrinsic::mips_binsli_h:
case Intrinsic::mips_binsli_w:
case Intrinsic::mips_binsli_d: {
[mips] BSEL's and BINS[RL] operands are reversed compared to the vselect node used in the pattern. Summary: Correct the match patterns and the lowerings that made the CodeGen tests pass despite the mistakes. The original testcase that discovered the problem was SingleSource/UnitTests/SignlessType/factor.c in test-suite. During review, we also found that some of the existing CodeGen tests were incorrect and fixed them: * bitwise.ll: In bsel_v16i8 the IfSet/IfClear were reversed because bsel and bmnz have different operand orders and the test didn't correctly account for this. bmnz goes 'IfClear, IfSet, CondMask', while bsel goes 'CondMask, IfClear, IfSet'. * vec.ll: In the cases where a bsel is emitted as a bmnz (they are the same operation with a different input tied to the result) the operands were in the wrong order. * compare.ll and compare_float.ll: The bsel operand order was correct for a greater-than comparison, but a greater-than comparison instruction doesn't exist. Lowering this operation inverts the condition so the IfSet/IfClear need to be swapped to match. The differences between BSEL, BMNZ, and BMZ and how they map to/from vselect are rather confusing. I've therefore added a note to MSA.txt to explain this in a single place in addition to the comments that explain each case. Reviewers: matheusalmeida, jacksprat Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D3028 llvm-svn: 203657
2014-03-12 19:54:00 +08:00
// binsli_x(IfClear, IfSet, nbits) -> (vselect LBitsMask, IfSet, IfClear)
[mips][msa] Added support for matching bins[lr]i.[bhwd] from normal IR (i.e. not intrinsics) This required correcting the definition of the bins[lr]i intrinsics because the result is also the first operand. It also required removing the (arbitrary) check for 32-bit immediates in MipsSEDAGToDAGISel::selectVSplat(). Currently using binsli.d with 2 bits set in the mask doesn't select binsli.d because the constant is legalized into a ConstantPool. Similar things can happen with binsri.d with more than 10 bits set in the mask. The resulting code when this happens is correct but not optimal. llvm-svn: 193687
2013-10-30 22:45:14 +08:00
EVT VecTy = Op->getValueType(0);
EVT EltTy = VecTy.getVectorElementType();
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
if (Op->getConstantOperandVal(3) >= EltTy.getSizeInBits())
report_fatal_error("Immediate out of range");
[mips][msa] Added support for matching bins[lr]i.[bhwd] from normal IR (i.e. not intrinsics) This required correcting the definition of the bins[lr]i intrinsics because the result is also the first operand. It also required removing the (arbitrary) check for 32-bit immediates in MipsSEDAGToDAGISel::selectVSplat(). Currently using binsli.d with 2 bits set in the mask doesn't select binsli.d because the constant is legalized into a ConstantPool. Similar things can happen with binsri.d with more than 10 bits set in the mask. The resulting code when this happens is correct but not optimal. llvm-svn: 193687
2013-10-30 22:45:14 +08:00
APInt Mask = APInt::getHighBitsSet(EltTy.getSizeInBits(),
[mips][msa] Fix generation of bm(n)zi and bins[lr]i instructions We have two cases here, the first one being the following instruction selection from the builtin function: bm(n)zi builtin -> vselect node -> bins[lr]i machine instruction In case of bm(n)zi having an immediate which has either its high or low bits set, a bins[lr] instruction can be selected through the selectVSplatMask[LR] function. The function counts the number of bits set, and that value is being passed to the bins[lr]i instruction as its immediate, which in turn copies immediate modulo the size of the element in bits plus 1 as per specs, where we get the off-by-one-error. The other case is: bins[lr]i -> vselect node -> bsel.v In this case, a bsel.v instruction gets selected with a mask having one bit less set than required. Patch by Stefan Maksimovic. Differential Revision: https://reviews.llvm.org/D30579 llvm-svn: 299768
2017-04-07 21:31:36 +08:00
Op->getConstantOperandVal(3) + 1);
[mips][msa] Added support for matching bins[lr]i.[bhwd] from normal IR (i.e. not intrinsics) This required correcting the definition of the bins[lr]i intrinsics because the result is also the first operand. It also required removing the (arbitrary) check for 32-bit immediates in MipsSEDAGToDAGISel::selectVSplat(). Currently using binsli.d with 2 bits set in the mask doesn't select binsli.d because the constant is legalized into a ConstantPool. Similar things can happen with binsri.d with more than 10 bits set in the mask. The resulting code when this happens is correct but not optimal. llvm-svn: 193687
2013-10-30 22:45:14 +08:00
return DAG.getNode(ISD::VSELECT, DL, VecTy,
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
DAG.getConstant(Mask, DL, VecTy, true),
Op->getOperand(2), Op->getOperand(1));
[mips][msa] Added support for matching bins[lr]i.[bhwd] from normal IR (i.e. not intrinsics) This required correcting the definition of the bins[lr]i intrinsics because the result is also the first operand. It also required removing the (arbitrary) check for 32-bit immediates in MipsSEDAGToDAGISel::selectVSplat(). Currently using binsli.d with 2 bits set in the mask doesn't select binsli.d because the constant is legalized into a ConstantPool. Similar things can happen with binsri.d with more than 10 bits set in the mask. The resulting code when this happens is correct but not optimal. llvm-svn: 193687
2013-10-30 22:45:14 +08:00
}
case Intrinsic::mips_binsri_b:
case Intrinsic::mips_binsri_h:
case Intrinsic::mips_binsri_w:
case Intrinsic::mips_binsri_d: {
[mips] BSEL's and BINS[RL] operands are reversed compared to the vselect node used in the pattern. Summary: Correct the match patterns and the lowerings that made the CodeGen tests pass despite the mistakes. The original testcase that discovered the problem was SingleSource/UnitTests/SignlessType/factor.c in test-suite. During review, we also found that some of the existing CodeGen tests were incorrect and fixed them: * bitwise.ll: In bsel_v16i8 the IfSet/IfClear were reversed because bsel and bmnz have different operand orders and the test didn't correctly account for this. bmnz goes 'IfClear, IfSet, CondMask', while bsel goes 'CondMask, IfClear, IfSet'. * vec.ll: In the cases where a bsel is emitted as a bmnz (they are the same operation with a different input tied to the result) the operands were in the wrong order. * compare.ll and compare_float.ll: The bsel operand order was correct for a greater-than comparison, but a greater-than comparison instruction doesn't exist. Lowering this operation inverts the condition so the IfSet/IfClear need to be swapped to match. The differences between BSEL, BMNZ, and BMZ and how they map to/from vselect are rather confusing. I've therefore added a note to MSA.txt to explain this in a single place in addition to the comments that explain each case. Reviewers: matheusalmeida, jacksprat Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D3028 llvm-svn: 203657
2014-03-12 19:54:00 +08:00
// binsri_x(IfClear, IfSet, nbits) -> (vselect RBitsMask, IfSet, IfClear)
[mips][msa] Added support for matching bins[lr]i.[bhwd] from normal IR (i.e. not intrinsics) This required correcting the definition of the bins[lr]i intrinsics because the result is also the first operand. It also required removing the (arbitrary) check for 32-bit immediates in MipsSEDAGToDAGISel::selectVSplat(). Currently using binsli.d with 2 bits set in the mask doesn't select binsli.d because the constant is legalized into a ConstantPool. Similar things can happen with binsri.d with more than 10 bits set in the mask. The resulting code when this happens is correct but not optimal. llvm-svn: 193687
2013-10-30 22:45:14 +08:00
EVT VecTy = Op->getValueType(0);
EVT EltTy = VecTy.getVectorElementType();
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
if (Op->getConstantOperandVal(3) >= EltTy.getSizeInBits())
report_fatal_error("Immediate out of range");
[mips][msa] Added support for matching bins[lr]i.[bhwd] from normal IR (i.e. not intrinsics) This required correcting the definition of the bins[lr]i intrinsics because the result is also the first operand. It also required removing the (arbitrary) check for 32-bit immediates in MipsSEDAGToDAGISel::selectVSplat(). Currently using binsli.d with 2 bits set in the mask doesn't select binsli.d because the constant is legalized into a ConstantPool. Similar things can happen with binsri.d with more than 10 bits set in the mask. The resulting code when this happens is correct but not optimal. llvm-svn: 193687
2013-10-30 22:45:14 +08:00
APInt Mask = APInt::getLowBitsSet(EltTy.getSizeInBits(),
[mips][msa] Fix generation of bm(n)zi and bins[lr]i instructions We have two cases here, the first one being the following instruction selection from the builtin function: bm(n)zi builtin -> vselect node -> bins[lr]i machine instruction In case of bm(n)zi having an immediate which has either its high or low bits set, a bins[lr] instruction can be selected through the selectVSplatMask[LR] function. The function counts the number of bits set, and that value is being passed to the bins[lr]i instruction as its immediate, which in turn copies immediate modulo the size of the element in bits plus 1 as per specs, where we get the off-by-one-error. The other case is: bins[lr]i -> vselect node -> bsel.v In this case, a bsel.v instruction gets selected with a mask having one bit less set than required. Patch by Stefan Maksimovic. Differential Revision: https://reviews.llvm.org/D30579 llvm-svn: 299768
2017-04-07 21:31:36 +08:00
Op->getConstantOperandVal(3) + 1);
[mips][msa] Added support for matching bins[lr]i.[bhwd] from normal IR (i.e. not intrinsics) This required correcting the definition of the bins[lr]i intrinsics because the result is also the first operand. It also required removing the (arbitrary) check for 32-bit immediates in MipsSEDAGToDAGISel::selectVSplat(). Currently using binsli.d with 2 bits set in the mask doesn't select binsli.d because the constant is legalized into a ConstantPool. Similar things can happen with binsri.d with more than 10 bits set in the mask. The resulting code when this happens is correct but not optimal. llvm-svn: 193687
2013-10-30 22:45:14 +08:00
return DAG.getNode(ISD::VSELECT, DL, VecTy,
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
DAG.getConstant(Mask, DL, VecTy, true),
Op->getOperand(2), Op->getOperand(1));
[mips][msa] Added support for matching bins[lr]i.[bhwd] from normal IR (i.e. not intrinsics) This required correcting the definition of the bins[lr]i intrinsics because the result is also the first operand. It also required removing the (arbitrary) check for 32-bit immediates in MipsSEDAGToDAGISel::selectVSplat(). Currently using binsli.d with 2 bits set in the mask doesn't select binsli.d because the constant is legalized into a ConstantPool. Similar things can happen with binsri.d with more than 10 bits set in the mask. The resulting code when this happens is correct but not optimal. llvm-svn: 193687
2013-10-30 22:45:14 +08:00
}
[mips][msa] Added support for matching bmnz, bmnzi, bmz, and bmzi from normal IR (i.e. not intrinsics) Also corrected the definition of the intrinsics for these instructions (the result register is also the first operand), and added intrinsics for bsel and bseli to clang (they already existed in the backend). These four operations are mostly equivalent to bsel, and bseli (the difference is which operand is tied to the result). As a result some of the tests changed as described below. bitwise.ll: - bsel.v test adapted so that the mask is unknown at compile-time. This stops it emitting bmnzi.b instead of the intended bsel.v. - The bseli.b test now tests the right thing. Namely the case when one of the values is an uimm8, rather than when the condition is a uimm8 (which is covered by bmnzi.b) compare.ll: - bsel.v tests now (correctly) emits bmnz.v instead of bsel.v because this is the same operation (see MSA.txt). i8.ll - CHECK-DAG-ized test. - bmzi.b test now (correctly) emits equivalent bmnzi.b with swapped operands because this is the same operation (see MSA.txt). - bseli.b still emits bseli.b though because the immediate makes it distinguishable from bmnzi.b. vec.ll: - CHECK-DAG-ized test. - bmz.v tests now (correctly) emits bmnz.v with swapped operands (see MSA.txt). - bsel.v tests now (correctly) emits bmnz.v with swapped operands (see MSA.txt). llvm-svn: 193693
2013-10-30 23:20:38 +08:00
case Intrinsic::mips_bmnz_v:
return DAG.getNode(ISD::VSELECT, DL, Op->getValueType(0), Op->getOperand(3),
Op->getOperand(2), Op->getOperand(1));
case Intrinsic::mips_bmnzi_b:
return DAG.getNode(ISD::VSELECT, DL, Op->getValueType(0),
lowerMSASplatImm(Op, 3, DAG), Op->getOperand(2),
Op->getOperand(1));
case Intrinsic::mips_bmz_v:
return DAG.getNode(ISD::VSELECT, DL, Op->getValueType(0), Op->getOperand(3),
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_bmzi_b:
return DAG.getNode(ISD::VSELECT, DL, Op->getValueType(0),
lowerMSASplatImm(Op, 3, DAG), Op->getOperand(1),
Op->getOperand(2));
[mips][msa] Added support for matching bset, bseti, bneg, and bnegi from normal IR (i.e. not intrinsics) llvm-svn: 194469
2013-11-12 18:31:49 +08:00
case Intrinsic::mips_bneg_b:
case Intrinsic::mips_bneg_h:
case Intrinsic::mips_bneg_w:
case Intrinsic::mips_bneg_d: {
EVT VecTy = Op->getValueType(0);
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
SDValue One = DAG.getConstant(1, DL, VecTy);
[mips][msa] Added support for matching bset, bseti, bneg, and bnegi from normal IR (i.e. not intrinsics) llvm-svn: 194469
2013-11-12 18:31:49 +08:00
return DAG.getNode(ISD::XOR, DL, VecTy, Op->getOperand(1),
DAG.getNode(ISD::SHL, DL, VecTy, One,
[mips][msa] Mask vectors holding shift amounts Masked vectors which hold shift amounts when creating the following nodes: ISD::SHL, ISD::SRL or ISD::SRA. Instructions that use said nodes, which have had their arguments altered are sll, srl, sra, bneg, bclr and bset. For said instructions, the shift amount or the bit position that is specified in the corresponding vector elements will be interpreted as the shift amount/bit position modulo the size of the element in bits. The problem lies in compiling with -O2 enabled, where the instructions for formats .w and .d are not generated, but are instead optimized away. In this case, having shift amounts that are either negative or greater than the element bit size results in generation of incorrect results when constant folding. We remedy this by masking the operands for the nodes mentioned above before actually creating them, so that the final result is correct before placed into the constant pool. Patch by Stefan Maksimovic. Differential Revision: https://reviews.llvm.org/D31331 llvm-svn: 300839
2017-04-20 21:26:46 +08:00
truncateVecElts(Op, DAG)));
[mips][msa] Added support for matching bset, bseti, bneg, and bnegi from normal IR (i.e. not intrinsics) llvm-svn: 194469
2013-11-12 18:31:49 +08:00
}
case Intrinsic::mips_bnegi_b:
case Intrinsic::mips_bnegi_h:
case Intrinsic::mips_bnegi_w:
case Intrinsic::mips_bnegi_d:
return lowerMSABinaryBitImmIntr(Op, DAG, ISD::XOR, Op->getOperand(2),
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
!Subtarget.isLittle());
[mips][msa] Added bnz.df, bnz.v, bz.df, and bz.v These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes, are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as a branch/mov sequence to evaluate to 0 or 1. Note: The resulting code is sub-optimal since it doesnt seem to be possible to feed the result of an intrinsic directly into a brcond. At the moment it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily evaluates the boolean twice. llvm-svn: 189478
2013-08-28 20:14:50 +08:00
case Intrinsic::mips_bnz_b:
case Intrinsic::mips_bnz_h:
case Intrinsic::mips_bnz_w:
case Intrinsic::mips_bnz_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::VALL_NONZERO, DL, Op->getValueType(0),
Op->getOperand(1));
[mips][msa] Added bnz.df, bnz.v, bz.df, and bz.v These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes, are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as a branch/mov sequence to evaluate to 0 or 1. Note: The resulting code is sub-optimal since it doesnt seem to be possible to feed the result of an intrinsic directly into a brcond. At the moment it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily evaluates the boolean twice. llvm-svn: 189478
2013-08-28 20:14:50 +08:00
case Intrinsic::mips_bnz_v:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::VANY_NONZERO, DL, Op->getValueType(0),
Op->getOperand(1));
[mips][msa] Added support for matching bsel and bseli from normal IR (i.e. not intrinsics) This required correcting the definition of the bsel and bseli intrinsics. llvm-svn: 191290
2013-09-24 20:04:44 +08:00
case Intrinsic::mips_bsel_v:
[mips] BSEL's and BINS[RL] operands are reversed compared to the vselect node used in the pattern. Summary: Correct the match patterns and the lowerings that made the CodeGen tests pass despite the mistakes. The original testcase that discovered the problem was SingleSource/UnitTests/SignlessType/factor.c in test-suite. During review, we also found that some of the existing CodeGen tests were incorrect and fixed them: * bitwise.ll: In bsel_v16i8 the IfSet/IfClear were reversed because bsel and bmnz have different operand orders and the test didn't correctly account for this. bmnz goes 'IfClear, IfSet, CondMask', while bsel goes 'CondMask, IfClear, IfSet'. * vec.ll: In the cases where a bsel is emitted as a bmnz (they are the same operation with a different input tied to the result) the operands were in the wrong order. * compare.ll and compare_float.ll: The bsel operand order was correct for a greater-than comparison, but a greater-than comparison instruction doesn't exist. Lowering this operation inverts the condition so the IfSet/IfClear need to be swapped to match. The differences between BSEL, BMNZ, and BMZ and how they map to/from vselect are rather confusing. I've therefore added a note to MSA.txt to explain this in a single place in addition to the comments that explain each case. Reviewers: matheusalmeida, jacksprat Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D3028 llvm-svn: 203657
2014-03-12 19:54:00 +08:00
// bsel_v(Mask, IfClear, IfSet) -> (vselect Mask, IfSet, IfClear)
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::VSELECT, DL, Op->getValueType(0),
[mips] BSEL's and BINS[RL] operands are reversed compared to the vselect node used in the pattern. Summary: Correct the match patterns and the lowerings that made the CodeGen tests pass despite the mistakes. The original testcase that discovered the problem was SingleSource/UnitTests/SignlessType/factor.c in test-suite. During review, we also found that some of the existing CodeGen tests were incorrect and fixed them: * bitwise.ll: In bsel_v16i8 the IfSet/IfClear were reversed because bsel and bmnz have different operand orders and the test didn't correctly account for this. bmnz goes 'IfClear, IfSet, CondMask', while bsel goes 'CondMask, IfClear, IfSet'. * vec.ll: In the cases where a bsel is emitted as a bmnz (they are the same operation with a different input tied to the result) the operands were in the wrong order. * compare.ll and compare_float.ll: The bsel operand order was correct for a greater-than comparison, but a greater-than comparison instruction doesn't exist. Lowering this operation inverts the condition so the IfSet/IfClear need to be swapped to match. The differences between BSEL, BMNZ, and BMZ and how they map to/from vselect are rather confusing. I've therefore added a note to MSA.txt to explain this in a single place in addition to the comments that explain each case. Reviewers: matheusalmeida, jacksprat Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D3028 llvm-svn: 203657
2014-03-12 19:54:00 +08:00
Op->getOperand(1), Op->getOperand(3),
Op->getOperand(2));
[mips][msa] Added support for matching bsel and bseli from normal IR (i.e. not intrinsics) This required correcting the definition of the bsel and bseli intrinsics. llvm-svn: 191290
2013-09-24 20:04:44 +08:00
case Intrinsic::mips_bseli_b:
[mips] BSEL's and BINS[RL] operands are reversed compared to the vselect node used in the pattern. Summary: Correct the match patterns and the lowerings that made the CodeGen tests pass despite the mistakes. The original testcase that discovered the problem was SingleSource/UnitTests/SignlessType/factor.c in test-suite. During review, we also found that some of the existing CodeGen tests were incorrect and fixed them: * bitwise.ll: In bsel_v16i8 the IfSet/IfClear were reversed because bsel and bmnz have different operand orders and the test didn't correctly account for this. bmnz goes 'IfClear, IfSet, CondMask', while bsel goes 'CondMask, IfClear, IfSet'. * vec.ll: In the cases where a bsel is emitted as a bmnz (they are the same operation with a different input tied to the result) the operands were in the wrong order. * compare.ll and compare_float.ll: The bsel operand order was correct for a greater-than comparison, but a greater-than comparison instruction doesn't exist. Lowering this operation inverts the condition so the IfSet/IfClear need to be swapped to match. The differences between BSEL, BMNZ, and BMZ and how they map to/from vselect are rather confusing. I've therefore added a note to MSA.txt to explain this in a single place in addition to the comments that explain each case. Reviewers: matheusalmeida, jacksprat Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D3028 llvm-svn: 203657
2014-03-12 19:54:00 +08:00
// bseli_v(Mask, IfClear, IfSet) -> (vselect Mask, IfSet, IfClear)
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::VSELECT, DL, Op->getValueType(0),
[mips] BSEL's and BINS[RL] operands are reversed compared to the vselect node used in the pattern. Summary: Correct the match patterns and the lowerings that made the CodeGen tests pass despite the mistakes. The original testcase that discovered the problem was SingleSource/UnitTests/SignlessType/factor.c in test-suite. During review, we also found that some of the existing CodeGen tests were incorrect and fixed them: * bitwise.ll: In bsel_v16i8 the IfSet/IfClear were reversed because bsel and bmnz have different operand orders and the test didn't correctly account for this. bmnz goes 'IfClear, IfSet, CondMask', while bsel goes 'CondMask, IfClear, IfSet'. * vec.ll: In the cases where a bsel is emitted as a bmnz (they are the same operation with a different input tied to the result) the operands were in the wrong order. * compare.ll and compare_float.ll: The bsel operand order was correct for a greater-than comparison, but a greater-than comparison instruction doesn't exist. Lowering this operation inverts the condition so the IfSet/IfClear need to be swapped to match. The differences between BSEL, BMNZ, and BMZ and how they map to/from vselect are rather confusing. I've therefore added a note to MSA.txt to explain this in a single place in addition to the comments that explain each case. Reviewers: matheusalmeida, jacksprat Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D3028 llvm-svn: 203657
2014-03-12 19:54:00 +08:00
Op->getOperand(1), lowerMSASplatImm(Op, 3, DAG),
Op->getOperand(2));
[mips][msa] Added support for matching bset, bseti, bneg, and bnegi from normal IR (i.e. not intrinsics) llvm-svn: 194469
2013-11-12 18:31:49 +08:00
case Intrinsic::mips_bset_b:
case Intrinsic::mips_bset_h:
case Intrinsic::mips_bset_w:
case Intrinsic::mips_bset_d: {
EVT VecTy = Op->getValueType(0);
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
SDValue One = DAG.getConstant(1, DL, VecTy);
[mips][msa] Added support for matching bset, bseti, bneg, and bnegi from normal IR (i.e. not intrinsics) llvm-svn: 194469
2013-11-12 18:31:49 +08:00
return DAG.getNode(ISD::OR, DL, VecTy, Op->getOperand(1),
DAG.getNode(ISD::SHL, DL, VecTy, One,
[mips][msa] Mask vectors holding shift amounts Masked vectors which hold shift amounts when creating the following nodes: ISD::SHL, ISD::SRL or ISD::SRA. Instructions that use said nodes, which have had their arguments altered are sll, srl, sra, bneg, bclr and bset. For said instructions, the shift amount or the bit position that is specified in the corresponding vector elements will be interpreted as the shift amount/bit position modulo the size of the element in bits. The problem lies in compiling with -O2 enabled, where the instructions for formats .w and .d are not generated, but are instead optimized away. In this case, having shift amounts that are either negative or greater than the element bit size results in generation of incorrect results when constant folding. We remedy this by masking the operands for the nodes mentioned above before actually creating them, so that the final result is correct before placed into the constant pool. Patch by Stefan Maksimovic. Differential Revision: https://reviews.llvm.org/D31331 llvm-svn: 300839
2017-04-20 21:26:46 +08:00
truncateVecElts(Op, DAG)));
[mips][msa] Added support for matching bset, bseti, bneg, and bnegi from normal IR (i.e. not intrinsics) llvm-svn: 194469
2013-11-12 18:31:49 +08:00
}
case Intrinsic::mips_bseti_b:
case Intrinsic::mips_bseti_h:
case Intrinsic::mips_bseti_w:
case Intrinsic::mips_bseti_d:
return lowerMSABinaryBitImmIntr(Op, DAG, ISD::OR, Op->getOperand(2),
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
!Subtarget.isLittle());
[mips][msa] Added bnz.df, bnz.v, bz.df, and bz.v These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes, are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as a branch/mov sequence to evaluate to 0 or 1. Note: The resulting code is sub-optimal since it doesnt seem to be possible to feed the result of an intrinsic directly into a brcond. At the moment it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily evaluates the boolean twice. llvm-svn: 189478
2013-08-28 20:14:50 +08:00
case Intrinsic::mips_bz_b:
case Intrinsic::mips_bz_h:
case Intrinsic::mips_bz_w:
case Intrinsic::mips_bz_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::VALL_ZERO, DL, Op->getValueType(0),
Op->getOperand(1));
[mips][msa] Added bnz.df, bnz.v, bz.df, and bz.v These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes, are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as a branch/mov sequence to evaluate to 0 or 1. Note: The resulting code is sub-optimal since it doesnt seem to be possible to feed the result of an intrinsic directly into a brcond. At the moment it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily evaluates the boolean twice. llvm-svn: 189478
2013-08-28 20:14:50 +08:00
case Intrinsic::mips_bz_v:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::VANY_ZERO, DL, Op->getValueType(0),
Op->getOperand(1));
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
case Intrinsic::mips_ceq_b:
case Intrinsic::mips_ceq_h:
case Intrinsic::mips_ceq_w:
case Intrinsic::mips_ceq_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
Op->getOperand(2), ISD::SETEQ);
case Intrinsic::mips_ceqi_b:
case Intrinsic::mips_ceqi_h:
case Intrinsic::mips_ceqi_w:
case Intrinsic::mips_ceqi_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
lowerMSASplatImm(Op, 2, DAG, true), ISD::SETEQ);
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
case Intrinsic::mips_cle_s_b:
case Intrinsic::mips_cle_s_h:
case Intrinsic::mips_cle_s_w:
case Intrinsic::mips_cle_s_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
Op->getOperand(2), ISD::SETLE);
case Intrinsic::mips_clei_s_b:
case Intrinsic::mips_clei_s_h:
case Intrinsic::mips_clei_s_w:
case Intrinsic::mips_clei_s_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
lowerMSASplatImm(Op, 2, DAG, true), ISD::SETLE);
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
case Intrinsic::mips_cle_u_b:
case Intrinsic::mips_cle_u_h:
case Intrinsic::mips_cle_u_w:
case Intrinsic::mips_cle_u_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
Op->getOperand(2), ISD::SETULE);
case Intrinsic::mips_clei_u_b:
case Intrinsic::mips_clei_u_h:
case Intrinsic::mips_clei_u_w:
case Intrinsic::mips_clei_u_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
lowerMSASplatImm(Op, 2, DAG), ISD::SETULE);
case Intrinsic::mips_clt_s_b:
case Intrinsic::mips_clt_s_h:
case Intrinsic::mips_clt_s_w:
case Intrinsic::mips_clt_s_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
Op->getOperand(2), ISD::SETLT);
case Intrinsic::mips_clti_s_b:
case Intrinsic::mips_clti_s_h:
case Intrinsic::mips_clti_s_w:
case Intrinsic::mips_clti_s_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
lowerMSASplatImm(Op, 2, DAG, true), ISD::SETLT);
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
case Intrinsic::mips_clt_u_b:
case Intrinsic::mips_clt_u_h:
case Intrinsic::mips_clt_u_w:
case Intrinsic::mips_clt_u_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
Op->getOperand(2), ISD::SETULT);
case Intrinsic::mips_clti_u_b:
case Intrinsic::mips_clti_u_h:
case Intrinsic::mips_clti_u_w:
case Intrinsic::mips_clti_u_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
lowerMSASplatImm(Op, 2, DAG), ISD::SETULT);
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
case Intrinsic::mips_copy_s_b:
case Intrinsic::mips_copy_s_h:
case Intrinsic::mips_copy_s_w:
return lowerMSACopyIntr(Op, DAG, MipsISD::VEXTRACT_SEXT_ELT);
[mips][msa] Implemented copy_[us].d intrinsic. This intrinsic is lowered into equivalent copy_s.w instructions during legalization. llvm-svn: 191518
2013-09-27 21:04:21 +08:00
case Intrinsic::mips_copy_s_d:
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (Subtarget.hasMips64())
[mips][msa] Add copy_{u,s}.d. These instructions are only available on Mips64 cores that implement the MSA ASE. llvm-svn: 200398
2014-01-29 22:05:28 +08:00
// Lower directly into VEXTRACT_SEXT_ELT since i64 is legal on Mips64.
return lowerMSACopyIntr(Op, DAG, MipsISD::VEXTRACT_SEXT_ELT);
else {
// Lower into the generic EXTRACT_VECTOR_ELT node and let the type
// legalizer and EXTRACT_VECTOR_ELT lowering sort it out.
return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SDLoc(Op),
Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
}
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
case Intrinsic::mips_copy_u_b:
case Intrinsic::mips_copy_u_h:
case Intrinsic::mips_copy_u_w:
return lowerMSACopyIntr(Op, DAG, MipsISD::VEXTRACT_ZEXT_ELT);
[mips][msa] Implemented copy_[us].d intrinsic. This intrinsic is lowered into equivalent copy_s.w instructions during legalization. llvm-svn: 191518
2013-09-27 21:04:21 +08:00
case Intrinsic::mips_copy_u_d:
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (Subtarget.hasMips64())
[mips][msa] Add copy_{u,s}.d. These instructions are only available on Mips64 cores that implement the MSA ASE. llvm-svn: 200398
2014-01-29 22:05:28 +08:00
// Lower directly into VEXTRACT_ZEXT_ELT since i64 is legal on Mips64.
return lowerMSACopyIntr(Op, DAG, MipsISD::VEXTRACT_ZEXT_ELT);
else {
// Lower into the generic EXTRACT_VECTOR_ELT node and let the type
// legalizer and EXTRACT_VECTOR_ELT lowering sort it out.
// Note: When i64 is illegal, this results in copy_s.w instructions
// instead of copy_u.w instructions. This makes no difference to the
// behaviour since i64 is only illegal when the register file is 32-bit.
return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, SDLoc(Op),
Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
}
[mips][msa] Added support for matching div_[su] from normal IR (i.e. not intrinsics) llvm-svn: 190509
2013-09-11 18:38:58 +08:00
case Intrinsic::mips_div_s_b:
case Intrinsic::mips_div_s_h:
case Intrinsic::mips_div_s_w:
case Intrinsic::mips_div_s_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::SDIV, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
[mips][msa] Added support for matching div_[su] from normal IR (i.e. not intrinsics) llvm-svn: 190509
2013-09-11 18:38:58 +08:00
case Intrinsic::mips_div_u_b:
case Intrinsic::mips_div_u_h:
case Intrinsic::mips_div_u_w:
case Intrinsic::mips_div_u_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::UDIV, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
[mips][msa] Added support for matching fadd, fdiv, flog2, fmul, frint, fsqrt, and fsub from normal IR (i.e. not intrinsics) llvm-svn: 190512
2013-09-11 18:51:30 +08:00
case Intrinsic::mips_fadd_w:
[Mips] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC). llvm-svn: 309993
2017-08-04 06:12:30 +08:00
case Intrinsic::mips_fadd_d:
propagate fast-math-flags on DAG nodes After D10403, we had FMF in the DAG but disabled by default. Nick reported no crashing errors after some stress testing, so I enabled them at r243687. However, Escha soon notified us of a bug not covered by any in-tree regression tests: if we don't propagate the flags, we may fail to CSE DAG nodes because differing FMF causes them to not match. There is one test case in this patch to prove that point. This patch hopes to fix or leave a 'TODO' for all of the in-tree places where we create nodes that are FMF-capable. I did this by putting an assert in SelectionDAG.getNode() to find any FMF-capable node that was being created without FMF ( D11807 ). I then ran all regression tests and test-suite and confirmed that everything passes. This patch exposes remaining work to get DAG FMF to be fully functional: (1) add the flags to non-binary nodes such as FCMP, FMA and FNEG; (2) add the flags to intrinsics; (3) use the flags as conditions for transforms rather than the current global settings. Differential Revision: http://reviews.llvm.org/D12095 llvm-svn: 247815
2015-09-17 00:31:21 +08:00
// TODO: If intrinsics have fast-math-flags, propagate them.
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::FADD, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
// Don't lower mips_fcaf_[wd] since LLVM folds SETFALSE condcodes away
case Intrinsic::mips_fceq_w:
case Intrinsic::mips_fceq_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
Op->getOperand(2), ISD::SETOEQ);
case Intrinsic::mips_fcle_w:
case Intrinsic::mips_fcle_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
Op->getOperand(2), ISD::SETOLE);
case Intrinsic::mips_fclt_w:
case Intrinsic::mips_fclt_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
Op->getOperand(2), ISD::SETOLT);
case Intrinsic::mips_fcne_w:
case Intrinsic::mips_fcne_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
Op->getOperand(2), ISD::SETONE);
case Intrinsic::mips_fcor_w:
case Intrinsic::mips_fcor_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
Op->getOperand(2), ISD::SETO);
case Intrinsic::mips_fcueq_w:
case Intrinsic::mips_fcueq_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
Op->getOperand(2), ISD::SETUEQ);
case Intrinsic::mips_fcule_w:
case Intrinsic::mips_fcule_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
Op->getOperand(2), ISD::SETULE);
case Intrinsic::mips_fcult_w:
case Intrinsic::mips_fcult_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
Op->getOperand(2), ISD::SETULT);
case Intrinsic::mips_fcun_w:
case Intrinsic::mips_fcun_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
Op->getOperand(2), ISD::SETUO);
case Intrinsic::mips_fcune_w:
case Intrinsic::mips_fcune_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getSetCC(DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Added support for matching comparisons from normal IR (i.e. not intrinsics) MIPS SelectionDAG changes: * Added VCEQ, VCL[ET]_[SU] nodes to represent vector comparisons that produce a bitmask. llvm-svn: 191286
2013-09-24 18:46:19 +08:00
Op->getOperand(2), ISD::SETUNE);
[mips][msa] Added support for matching fadd, fdiv, flog2, fmul, frint, fsqrt, and fsub from normal IR (i.e. not intrinsics) llvm-svn: 190512
2013-09-11 18:51:30 +08:00
case Intrinsic::mips_fdiv_w:
[Mips] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC). llvm-svn: 309993
2017-08-04 06:12:30 +08:00
case Intrinsic::mips_fdiv_d:
propagate fast-math-flags on DAG nodes After D10403, we had FMF in the DAG but disabled by default. Nick reported no crashing errors after some stress testing, so I enabled them at r243687. However, Escha soon notified us of a bug not covered by any in-tree regression tests: if we don't propagate the flags, we may fail to CSE DAG nodes because differing FMF causes them to not match. There is one test case in this patch to prove that point. This patch hopes to fix or leave a 'TODO' for all of the in-tree places where we create nodes that are FMF-capable. I did this by putting an assert in SelectionDAG.getNode() to find any FMF-capable node that was being created without FMF ( D11807 ). I then ran all regression tests and test-suite and confirmed that everything passes. This patch exposes remaining work to get DAG FMF to be fully functional: (1) add the flags to non-binary nodes such as FCMP, FMA and FNEG; (2) add the flags to intrinsics; (3) use the flags as conditions for transforms rather than the current global settings. Differential Revision: http://reviews.llvm.org/D12095 llvm-svn: 247815
2015-09-17 00:31:21 +08:00
// TODO: If intrinsics have fast-math-flags, propagate them.
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::FDIV, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
[mips][msa] Added support for matching ffint_[us].[wd], and ftrunc_[us].[wd] from normal IR (i.e. not intrinsics) llvm-svn: 192429
2013-10-11 18:00:06 +08:00
case Intrinsic::mips_ffint_u_w:
case Intrinsic::mips_ffint_u_d:
return DAG.getNode(ISD::UINT_TO_FP, DL, Op->getValueType(0),
Op->getOperand(1));
case Intrinsic::mips_ffint_s_w:
case Intrinsic::mips_ffint_s_d:
return DAG.getNode(ISD::SINT_TO_FP, DL, Op->getValueType(0),
Op->getOperand(1));
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
case Intrinsic::mips_fill_b:
case Intrinsic::mips_fill_h:
[mips][msa] Implemented fill.d intrinsic. This intrinsic is lowered into an equivalent BUILD_VECTOR which is further lowered into a sequence of insert.w's on MIPS32. llvm-svn: 191519
2013-09-27 21:20:41 +08:00
case Intrinsic::mips_fill_w:
case Intrinsic::mips_fill_d: {
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
EVT ResTy = Op->getValueType(0);
Prefer SmallVector::append/insert over push_back loops. Same functionality, but hoists the vector growth out of the loop. llvm-svn: 229500
2015-02-17 23:29:18 +08:00
SmallVector<SDValue, 16> Ops(ResTy.getVectorNumElements(),
Op->getOperand(1));
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
[mips][msa] Implemented fill.d intrinsic. This intrinsic is lowered into an equivalent BUILD_VECTOR which is further lowered into a sequence of insert.w's on MIPS32. llvm-svn: 191519
2013-09-27 21:20:41 +08:00
// If ResTy is v2i64 then the type legalizer will break this node down into
// an equivalent v4i32.
[CodeGen] Add getBuildVector and getSplatBuildVector helpers. NFCI. Differential Revision: http://reviews.llvm.org/D17176 llvm-svn: 267606
2016-04-27 05:15:30 +08:00
return DAG.getBuildVector(ResTy, DL, Ops);
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
}
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
case Intrinsic::mips_fexp2_w:
case Intrinsic::mips_fexp2_d: {
propagate fast-math-flags on DAG nodes After D10403, we had FMF in the DAG but disabled by default. Nick reported no crashing errors after some stress testing, so I enabled them at r243687. However, Escha soon notified us of a bug not covered by any in-tree regression tests: if we don't propagate the flags, we may fail to CSE DAG nodes because differing FMF causes them to not match. There is one test case in this patch to prove that point. This patch hopes to fix or leave a 'TODO' for all of the in-tree places where we create nodes that are FMF-capable. I did this by putting an assert in SelectionDAG.getNode() to find any FMF-capable node that was being created without FMF ( D11807 ). I then ran all regression tests and test-suite and confirmed that everything passes. This patch exposes remaining work to get DAG FMF to be fully functional: (1) add the flags to non-binary nodes such as FCMP, FMA and FNEG; (2) add the flags to intrinsics; (3) use the flags as conditions for transforms rather than the current global settings. Differential Revision: http://reviews.llvm.org/D12095 llvm-svn: 247815
2015-09-17 00:31:21 +08:00
// TODO: If intrinsics have fast-math-flags, propagate them.
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
EVT ResTy = Op->getValueType(0);
return DAG.getNode(
ISD::FMUL, SDLoc(Op), ResTy, Op->getOperand(1),
DAG.getNode(ISD::FEXP2, SDLoc(Op), ResTy, Op->getOperand(2)));
}
[mips][msa] Added support for matching fadd, fdiv, flog2, fmul, frint, fsqrt, and fsub from normal IR (i.e. not intrinsics) llvm-svn: 190512
2013-09-11 18:51:30 +08:00
case Intrinsic::mips_flog2_w:
case Intrinsic::mips_flog2_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::FLOG2, DL, Op->getValueType(0), Op->getOperand(1));
[mips][msa] Added support for matching fmadd.[wd] from normal IR (i.e. not intrinsics) llvm-svn: 192430
2013-10-11 18:14:25 +08:00
case Intrinsic::mips_fmadd_w:
case Intrinsic::mips_fmadd_d:
return DAG.getNode(ISD::FMA, SDLoc(Op), Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2), Op->getOperand(3));
[mips][msa] Added support for matching fadd, fdiv, flog2, fmul, frint, fsqrt, and fsub from normal IR (i.e. not intrinsics) llvm-svn: 190512
2013-09-11 18:51:30 +08:00
case Intrinsic::mips_fmul_w:
[Mips] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC). llvm-svn: 309993
2017-08-04 06:12:30 +08:00
case Intrinsic::mips_fmul_d:
propagate fast-math-flags on DAG nodes After D10403, we had FMF in the DAG but disabled by default. Nick reported no crashing errors after some stress testing, so I enabled them at r243687. However, Escha soon notified us of a bug not covered by any in-tree regression tests: if we don't propagate the flags, we may fail to CSE DAG nodes because differing FMF causes them to not match. There is one test case in this patch to prove that point. This patch hopes to fix or leave a 'TODO' for all of the in-tree places where we create nodes that are FMF-capable. I did this by putting an assert in SelectionDAG.getNode() to find any FMF-capable node that was being created without FMF ( D11807 ). I then ran all regression tests and test-suite and confirmed that everything passes. This patch exposes remaining work to get DAG FMF to be fully functional: (1) add the flags to non-binary nodes such as FCMP, FMA and FNEG; (2) add the flags to intrinsics; (3) use the flags as conditions for transforms rather than the current global settings. Differential Revision: http://reviews.llvm.org/D12095 llvm-svn: 247815
2015-09-17 00:31:21 +08:00
// TODO: If intrinsics have fast-math-flags, propagate them.
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::FMUL, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
[mips][msa] Added support for matching fmsub.[wd] from normal IR (i.e. not intrinsics) llvm-svn: 192435
2013-10-11 18:27:32 +08:00
case Intrinsic::mips_fmsub_w:
case Intrinsic::mips_fmsub_d: {
propagate fast-math-flags on DAG nodes After D10403, we had FMF in the DAG but disabled by default. Nick reported no crashing errors after some stress testing, so I enabled them at r243687. However, Escha soon notified us of a bug not covered by any in-tree regression tests: if we don't propagate the flags, we may fail to CSE DAG nodes because differing FMF causes them to not match. There is one test case in this patch to prove that point. This patch hopes to fix or leave a 'TODO' for all of the in-tree places where we create nodes that are FMF-capable. I did this by putting an assert in SelectionDAG.getNode() to find any FMF-capable node that was being created without FMF ( D11807 ). I then ran all regression tests and test-suite and confirmed that everything passes. This patch exposes remaining work to get DAG FMF to be fully functional: (1) add the flags to non-binary nodes such as FCMP, FMA and FNEG; (2) add the flags to intrinsics; (3) use the flags as conditions for transforms rather than the current global settings. Differential Revision: http://reviews.llvm.org/D12095 llvm-svn: 247815
2015-09-17 00:31:21 +08:00
// TODO: If intrinsics have fast-math-flags, propagate them.
[mips][msa] Added support for matching fmsub.[wd] from normal IR (i.e. not intrinsics) llvm-svn: 192435
2013-10-11 18:27:32 +08:00
EVT ResTy = Op->getValueType(0);
return DAG.getNode(ISD::FSUB, SDLoc(Op), ResTy, Op->getOperand(1),
DAG.getNode(ISD::FMUL, SDLoc(Op), ResTy,
Op->getOperand(2), Op->getOperand(3)));
}
[mips][msa] Added support for matching fadd, fdiv, flog2, fmul, frint, fsqrt, and fsub from normal IR (i.e. not intrinsics) llvm-svn: 190512
2013-09-11 18:51:30 +08:00
case Intrinsic::mips_frint_w:
case Intrinsic::mips_frint_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::FRINT, DL, Op->getValueType(0), Op->getOperand(1));
[mips][msa] Added support for matching fadd, fdiv, flog2, fmul, frint, fsqrt, and fsub from normal IR (i.e. not intrinsics) llvm-svn: 190512
2013-09-11 18:51:30 +08:00
case Intrinsic::mips_fsqrt_w:
case Intrinsic::mips_fsqrt_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::FSQRT, DL, Op->getValueType(0), Op->getOperand(1));
[mips][msa] Added support for matching fadd, fdiv, flog2, fmul, frint, fsqrt, and fsub from normal IR (i.e. not intrinsics) llvm-svn: 190512
2013-09-11 18:51:30 +08:00
case Intrinsic::mips_fsub_w:
[Mips] Fix some Clang-tidy modernize-use-using and Include What You Use warnings; other minor fixes (NFC). llvm-svn: 309993
2017-08-04 06:12:30 +08:00
case Intrinsic::mips_fsub_d:
propagate fast-math-flags on DAG nodes After D10403, we had FMF in the DAG but disabled by default. Nick reported no crashing errors after some stress testing, so I enabled them at r243687. However, Escha soon notified us of a bug not covered by any in-tree regression tests: if we don't propagate the flags, we may fail to CSE DAG nodes because differing FMF causes them to not match. There is one test case in this patch to prove that point. This patch hopes to fix or leave a 'TODO' for all of the in-tree places where we create nodes that are FMF-capable. I did this by putting an assert in SelectionDAG.getNode() to find any FMF-capable node that was being created without FMF ( D11807 ). I then ran all regression tests and test-suite and confirmed that everything passes. This patch exposes remaining work to get DAG FMF to be fully functional: (1) add the flags to non-binary nodes such as FCMP, FMA and FNEG; (2) add the flags to intrinsics; (3) use the flags as conditions for transforms rather than the current global settings. Differential Revision: http://reviews.llvm.org/D12095 llvm-svn: 247815
2015-09-17 00:31:21 +08:00
// TODO: If intrinsics have fast-math-flags, propagate them.
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::FSUB, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
[mips][msa] Added support for matching ffint_[us].[wd], and ftrunc_[us].[wd] from normal IR (i.e. not intrinsics) llvm-svn: 192429
2013-10-11 18:00:06 +08:00
case Intrinsic::mips_ftrunc_u_w:
case Intrinsic::mips_ftrunc_u_d:
return DAG.getNode(ISD::FP_TO_UINT, DL, Op->getValueType(0),
Op->getOperand(1));
case Intrinsic::mips_ftrunc_s_w:
case Intrinsic::mips_ftrunc_s_d:
return DAG.getNode(ISD::FP_TO_SINT, DL, Op->getValueType(0),
Op->getOperand(1));
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
case Intrinsic::mips_ilvev_b:
case Intrinsic::mips_ilvev_h:
case Intrinsic::mips_ilvev_w:
case Intrinsic::mips_ilvev_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::ILVEV, DL, Op->getValueType(0),
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_ilvl_b:
case Intrinsic::mips_ilvl_h:
case Intrinsic::mips_ilvl_w:
case Intrinsic::mips_ilvl_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::ILVL, DL, Op->getValueType(0),
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_ilvod_b:
case Intrinsic::mips_ilvod_h:
case Intrinsic::mips_ilvod_w:
case Intrinsic::mips_ilvod_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::ILVOD, DL, Op->getValueType(0),
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_ilvr_b:
case Intrinsic::mips_ilvr_h:
case Intrinsic::mips_ilvr_w:
case Intrinsic::mips_ilvr_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::ILVR, DL, Op->getValueType(0),
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
Op->getOperand(1), Op->getOperand(2));
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
case Intrinsic::mips_insert_b:
case Intrinsic::mips_insert_h:
case Intrinsic::mips_insert_w:
[mips][msa] Implemented insert.d intrinsic. This intrinsic is lowered into an equivalent INSERT_VECTOR_ELT which is further lowered into a sequence of insert.w's on MIPS32. llvm-svn: 191521
2013-09-27 21:36:54 +08:00
case Intrinsic::mips_insert_d:
return DAG.getNode(ISD::INSERT_VECTOR_ELT, SDLoc(Op), Op->getValueType(0),
Op->getOperand(1), Op->getOperand(3), Op->getOperand(2));
[mips] Rewrite MipsAsmParser and MipsOperand. Summary: Highlights: - Registers are resolved much later (by the render method). Prior to that point, GPR32's/GPR64's are GPR's regardless of register size. Similarly FGR32's/FGR64's/AFGR64's are FGR's regardless of register size or FR mode. Numeric registers can be anything. - All registers are parsed the same way everywhere (even when handling symbol aliasing) - One consequence is that all registers can be specified numerically almost anywhere (e.g. $fccX, $wX). The exception is symbol aliasing but that can be easily resolved. - Removes the need for the hasConsumedDollar hack - Parenthesis and Bracket suffixes are handled generically - Micromips instructions are parsed directly instead of going through the standard encodings first. - rdhwr accepts all 32 registers, and the following instructions that previously xfailed now work: ddiv, ddivu, div, divu, cvt.l.[ds], se[bh], wsbh, floor.w.[ds], c.ngl.d, c.sf.s, dsbh, dshd, madd.s, msub.s, nmadd.s, nmsub.s, swxc1 - Diagnostics involving registers point at the correct character (the $) - There's only one kind of immediate in MipsOperand. LSA immediates are handled by the predicate and renderer. Lowlights: - Hardcoded '$zero' in the div patterns is handled with a hack. MipsOperand::isReg() will return true for a k_RegisterIndex token with Index == 0 and getReg() will return ZERO for this case. Note that it doesn't return ZERO_64 on isGP64() targets. - I haven't cleaned up all of the now-unused functions. Some more of the generic parser could be removed too (integers and relocs for example). - insve.df needed a custom decoder to handle the implicit fourth operand that was needed to make it parse correctly. The difficulty was that the matcher expected a Token<'0'> but gets an Imm<0>. Adding an implicit zero solved this. Reviewers: matheusalmeida, vmedic Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D3222 llvm-svn: 205292
2014-04-01 18:35:28 +08:00
case Intrinsic::mips_insve_b:
case Intrinsic::mips_insve_h:
case Intrinsic::mips_insve_w:
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
case Intrinsic::mips_insve_d: {
// Report an error for out of range values.
int64_t Max;
switch (Intrinsic) {
case Intrinsic::mips_insve_b: Max = 15; break;
case Intrinsic::mips_insve_h: Max = 7; break;
case Intrinsic::mips_insve_w: Max = 3; break;
case Intrinsic::mips_insve_d: Max = 1; break;
default: llvm_unreachable("Unmatched intrinsic");
}
int64_t Value = cast<ConstantSDNode>(Op->getOperand(2))->getSExtValue();
if (Value < 0 || Value > Max)
report_fatal_error("Immediate out of range");
[mips] Rewrite MipsAsmParser and MipsOperand. Summary: Highlights: - Registers are resolved much later (by the render method). Prior to that point, GPR32's/GPR64's are GPR's regardless of register size. Similarly FGR32's/FGR64's/AFGR64's are FGR's regardless of register size or FR mode. Numeric registers can be anything. - All registers are parsed the same way everywhere (even when handling symbol aliasing) - One consequence is that all registers can be specified numerically almost anywhere (e.g. $fccX, $wX). The exception is symbol aliasing but that can be easily resolved. - Removes the need for the hasConsumedDollar hack - Parenthesis and Bracket suffixes are handled generically - Micromips instructions are parsed directly instead of going through the standard encodings first. - rdhwr accepts all 32 registers, and the following instructions that previously xfailed now work: ddiv, ddivu, div, divu, cvt.l.[ds], se[bh], wsbh, floor.w.[ds], c.ngl.d, c.sf.s, dsbh, dshd, madd.s, msub.s, nmadd.s, nmsub.s, swxc1 - Diagnostics involving registers point at the correct character (the $) - There's only one kind of immediate in MipsOperand. LSA immediates are handled by the predicate and renderer. Lowlights: - Hardcoded '$zero' in the div patterns is handled with a hack. MipsOperand::isReg() will return true for a k_RegisterIndex token with Index == 0 and getReg() will return ZERO for this case. Note that it doesn't return ZERO_64 on isGP64() targets. - I haven't cleaned up all of the now-unused functions. Some more of the generic parser could be removed too (integers and relocs for example). - insve.df needed a custom decoder to handle the implicit fourth operand that was needed to make it parse correctly. The difficulty was that the matcher expected a Token<'0'> but gets an Imm<0>. Adding an implicit zero solved this. Reviewers: matheusalmeida, vmedic Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D3222 llvm-svn: 205292
2014-04-01 18:35:28 +08:00
return DAG.getNode(MipsISD::INSVE, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2), Op->getOperand(3),
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
DAG.getConstant(0, DL, MVT::i32));
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
}
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
case Intrinsic::mips_ldi_b:
case Intrinsic::mips_ldi_h:
case Intrinsic::mips_ldi_w:
case Intrinsic::mips_ldi_d:
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
return lowerMSASplatImm(Op, 1, DAG, true);
[mips][msa] Add DLSA instruction. llvm-svn: 201081
2014-02-10 20:05:17 +08:00
case Intrinsic::mips_lsa:
case Intrinsic::mips_dlsa: {
[mips][msa] Added lsa instruction llvm-svn: 192895
2013-10-17 21:38:20 +08:00
EVT ResTy = Op->getValueType(0);
return DAG.getNode(ISD::ADD, SDLoc(Op), ResTy, Op->getOperand(1),
DAG.getNode(ISD::SHL, SDLoc(Op), ResTy,
Op->getOperand(2), Op->getOperand(3)));
}
[mips][msa] Added support for matching maddv.[bhwd], and msubv.[bhwd] from normal IR (i.e. not intrinsics) llvm-svn: 192438
2013-10-11 18:50:42 +08:00
case Intrinsic::mips_maddv_b:
case Intrinsic::mips_maddv_h:
case Intrinsic::mips_maddv_w:
case Intrinsic::mips_maddv_d: {
EVT ResTy = Op->getValueType(0);
return DAG.getNode(ISD::ADD, SDLoc(Op), ResTy, Op->getOperand(1),
DAG.getNode(ISD::MUL, SDLoc(Op), ResTy,
Op->getOperand(2), Op->getOperand(3)));
}
[mips][msa] Added support for matching max, maxi, min, mini from normal IR (i.e. not intrinsics) llvm-svn: 191291
2013-09-24 20:18:31 +08:00
case Intrinsic::mips_max_s_b:
case Intrinsic::mips_max_s_h:
case Intrinsic::mips_max_s_w:
case Intrinsic::mips_max_s_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::VSMAX, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
[mips][msa] Added support for matching max, maxi, min, mini from normal IR (i.e. not intrinsics) llvm-svn: 191291
2013-09-24 20:18:31 +08:00
case Intrinsic::mips_max_u_b:
case Intrinsic::mips_max_u_h:
case Intrinsic::mips_max_u_w:
case Intrinsic::mips_max_u_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::VUMAX, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
[mips][msa] Added support for matching max, maxi, min, mini from normal IR (i.e. not intrinsics) llvm-svn: 191291
2013-09-24 20:18:31 +08:00
case Intrinsic::mips_maxi_s_b:
case Intrinsic::mips_maxi_s_h:
case Intrinsic::mips_maxi_s_w:
case Intrinsic::mips_maxi_s_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::VSMAX, DL, Op->getValueType(0),
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG, true));
[mips][msa] Added support for matching max, maxi, min, mini from normal IR (i.e. not intrinsics) llvm-svn: 191291
2013-09-24 20:18:31 +08:00
case Intrinsic::mips_maxi_u_b:
case Intrinsic::mips_maxi_u_h:
case Intrinsic::mips_maxi_u_w:
case Intrinsic::mips_maxi_u_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::VUMAX, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
[mips][msa] Added support for matching max, maxi, min, mini from normal IR (i.e. not intrinsics) llvm-svn: 191291
2013-09-24 20:18:31 +08:00
case Intrinsic::mips_min_s_b:
case Intrinsic::mips_min_s_h:
case Intrinsic::mips_min_s_w:
case Intrinsic::mips_min_s_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::VSMIN, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
[mips][msa] Added support for matching max, maxi, min, mini from normal IR (i.e. not intrinsics) llvm-svn: 191291
2013-09-24 20:18:31 +08:00
case Intrinsic::mips_min_u_b:
case Intrinsic::mips_min_u_h:
case Intrinsic::mips_min_u_w:
case Intrinsic::mips_min_u_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::VUMIN, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
[mips][msa] Added support for matching max, maxi, min, mini from normal IR (i.e. not intrinsics) llvm-svn: 191291
2013-09-24 20:18:31 +08:00
case Intrinsic::mips_mini_s_b:
case Intrinsic::mips_mini_s_h:
case Intrinsic::mips_mini_s_w:
case Intrinsic::mips_mini_s_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::VSMIN, DL, Op->getValueType(0),
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG, true));
[mips][msa] Added support for matching max, maxi, min, mini from normal IR (i.e. not intrinsics) llvm-svn: 191291
2013-09-24 20:18:31 +08:00
case Intrinsic::mips_mini_u_b:
case Intrinsic::mips_mini_u_h:
case Intrinsic::mips_mini_u_w:
case Intrinsic::mips_mini_u_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::VUMIN, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
[mips][msa] Added support for matching mod_[us] from normal IR (i.e. not intrinsics) llvm-svn: 191737
2013-10-01 18:22:35 +08:00
case Intrinsic::mips_mod_s_b:
case Intrinsic::mips_mod_s_h:
case Intrinsic::mips_mod_s_w:
case Intrinsic::mips_mod_s_d:
return DAG.getNode(ISD::SREM, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
case Intrinsic::mips_mod_u_b:
case Intrinsic::mips_mod_u_h:
case Intrinsic::mips_mod_u_w:
case Intrinsic::mips_mod_u_d:
return DAG.getNode(ISD::UREM, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
[mips][msa] Added support for matching mulv, nlzc, sll, sra, srl, and subv from normal IR (i.e. not intrinsics) llvm-svn: 190518
2013-09-11 19:58:30 +08:00
case Intrinsic::mips_mulv_b:
case Intrinsic::mips_mulv_h:
case Intrinsic::mips_mulv_w:
case Intrinsic::mips_mulv_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::MUL, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
[mips][msa] Added support for matching maddv.[bhwd], and msubv.[bhwd] from normal IR (i.e. not intrinsics) llvm-svn: 192438
2013-10-11 18:50:42 +08:00
case Intrinsic::mips_msubv_b:
case Intrinsic::mips_msubv_h:
case Intrinsic::mips_msubv_w:
case Intrinsic::mips_msubv_d: {
EVT ResTy = Op->getValueType(0);
return DAG.getNode(ISD::SUB, SDLoc(Op), ResTy, Op->getOperand(1),
DAG.getNode(ISD::MUL, SDLoc(Op), ResTy,
Op->getOperand(2), Op->getOperand(3)));
}
[mips][msa] Added support for matching mulv, nlzc, sll, sra, srl, and subv from normal IR (i.e. not intrinsics) llvm-svn: 190518
2013-09-11 19:58:30 +08:00
case Intrinsic::mips_nlzc_b:
case Intrinsic::mips_nlzc_h:
case Intrinsic::mips_nlzc_w:
case Intrinsic::mips_nlzc_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::CTLZ, DL, Op->getValueType(0), Op->getOperand(1));
[mips][msa] Added support for matching nor from normal IR (i.e. not intrinsics) llvm-svn: 191195
2013-09-23 21:22:24 +08:00
case Intrinsic::mips_nor_v: {
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
SDValue Res = DAG.getNode(ISD::OR, DL, Op->getValueType(0),
Op->getOperand(1), Op->getOperand(2));
return DAG.getNOT(DL, Res, Res->getValueType(0));
[mips][msa] Added support for matching nor from normal IR (i.e. not intrinsics) llvm-svn: 191195
2013-09-23 21:22:24 +08:00
}
[mips][msa] Added support for matching andi, ori, nori, and xori from normal IR (i.e. not intrinsics) llvm-svn: 191293
2013-09-24 20:32:47 +08:00
case Intrinsic::mips_nori_b: {
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
SDValue Res = DAG.getNode(ISD::OR, DL, Op->getValueType(0),
Op->getOperand(1),
lowerMSASplatImm(Op, 2, DAG));
return DAG.getNOT(DL, Res, Res->getValueType(0));
[mips][msa] Added support for matching andi, ori, nori, and xori from normal IR (i.e. not intrinsics) llvm-svn: 191293
2013-09-24 20:32:47 +08:00
}
[mips][msa] Added support for matching and, or, and xor from normal IR (i.e. not intrinsics) llvm-svn: 191194
2013-09-23 20:57:42 +08:00
case Intrinsic::mips_or_v:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::OR, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
[mips][msa] Added support for matching andi, ori, nori, and xori from normal IR (i.e. not intrinsics) llvm-svn: 191293
2013-09-24 20:32:47 +08:00
case Intrinsic::mips_ori_b:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::OR, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
case Intrinsic::mips_pckev_b:
case Intrinsic::mips_pckev_h:
case Intrinsic::mips_pckev_w:
case Intrinsic::mips_pckev_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::PCKEV, DL, Op->getValueType(0),
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
Op->getOperand(1), Op->getOperand(2));
case Intrinsic::mips_pckod_b:
case Intrinsic::mips_pckod_h:
case Intrinsic::mips_pckod_w:
case Intrinsic::mips_pckod_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::PCKOD, DL, Op->getValueType(0),
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
Op->getOperand(1), Op->getOperand(2));
[mips][msa] Added support for matching pcnt from normal IR (i.e. not intrinsics) llvm-svn: 191198
2013-09-23 21:40:21 +08:00
case Intrinsic::mips_pcnt_b:
case Intrinsic::mips_pcnt_h:
case Intrinsic::mips_pcnt_w:
case Intrinsic::mips_pcnt_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::CTPOP, DL, Op->getValueType(0), Op->getOperand(1));
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
case Intrinsic::mips_sat_s_b:
case Intrinsic::mips_sat_s_h:
case Intrinsic::mips_sat_s_w:
case Intrinsic::mips_sat_s_d:
case Intrinsic::mips_sat_u_b:
case Intrinsic::mips_sat_u_h:
case Intrinsic::mips_sat_u_w:
case Intrinsic::mips_sat_u_d: {
// Report an error for out of range values.
int64_t Max;
switch (Intrinsic) {
case Intrinsic::mips_sat_s_b:
case Intrinsic::mips_sat_u_b: Max = 7; break;
case Intrinsic::mips_sat_s_h:
case Intrinsic::mips_sat_u_h: Max = 15; break;
case Intrinsic::mips_sat_s_w:
case Intrinsic::mips_sat_u_w: Max = 31; break;
case Intrinsic::mips_sat_s_d:
case Intrinsic::mips_sat_u_d: Max = 63; break;
default: llvm_unreachable("Unmatched intrinsic");
}
int64_t Value = cast<ConstantSDNode>(Op->getOperand(2))->getSExtValue();
if (Value < 0 || Value > Max)
report_fatal_error("Immediate out of range");
return SDValue();
}
[mips][msa] Added support for matching shf from normal IR (i.e. not intrinsics) llvm-svn: 191302
2013-09-24 22:20:00 +08:00
case Intrinsic::mips_shf_b:
case Intrinsic::mips_shf_h:
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
case Intrinsic::mips_shf_w: {
int64_t Value = cast<ConstantSDNode>(Op->getOperand(2))->getSExtValue();
if (Value < 0 || Value > 255)
report_fatal_error("Immediate out of range");
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::SHF, DL, Op->getValueType(0),
[mips][msa] Added support for matching shf from normal IR (i.e. not intrinsics) llvm-svn: 191302
2013-09-24 22:20:00 +08:00
Op->getOperand(2), Op->getOperand(1));
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
}
case Intrinsic::mips_sldi_b:
case Intrinsic::mips_sldi_h:
case Intrinsic::mips_sldi_w:
case Intrinsic::mips_sldi_d: {
// Report an error for out of range values.
int64_t Max;
switch (Intrinsic) {
case Intrinsic::mips_sldi_b: Max = 15; break;
case Intrinsic::mips_sldi_h: Max = 7; break;
case Intrinsic::mips_sldi_w: Max = 3; break;
case Intrinsic::mips_sldi_d: Max = 1; break;
default: llvm_unreachable("Unmatched intrinsic");
}
int64_t Value = cast<ConstantSDNode>(Op->getOperand(3))->getSExtValue();
if (Value < 0 || Value > Max)
report_fatal_error("Immediate out of range");
return SDValue();
}
[mips][msa] Added support for matching mulv, nlzc, sll, sra, srl, and subv from normal IR (i.e. not intrinsics) llvm-svn: 190518
2013-09-11 19:58:30 +08:00
case Intrinsic::mips_sll_b:
case Intrinsic::mips_sll_h:
case Intrinsic::mips_sll_w:
case Intrinsic::mips_sll_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::SHL, DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Mask vectors holding shift amounts Masked vectors which hold shift amounts when creating the following nodes: ISD::SHL, ISD::SRL or ISD::SRA. Instructions that use said nodes, which have had their arguments altered are sll, srl, sra, bneg, bclr and bset. For said instructions, the shift amount or the bit position that is specified in the corresponding vector elements will be interpreted as the shift amount/bit position modulo the size of the element in bits. The problem lies in compiling with -O2 enabled, where the instructions for formats .w and .d are not generated, but are instead optimized away. In this case, having shift amounts that are either negative or greater than the element bit size results in generation of incorrect results when constant folding. We remedy this by masking the operands for the nodes mentioned above before actually creating them, so that the final result is correct before placed into the constant pool. Patch by Stefan Maksimovic. Differential Revision: https://reviews.llvm.org/D31331 llvm-svn: 300839
2017-04-20 21:26:46 +08:00
truncateVecElts(Op, DAG));
[mips][msa] Added support for matching slli, srai, and srli from normal IR (i.e. not intrinsics) llvm-svn: 191285
2013-09-24 18:28:18 +08:00
case Intrinsic::mips_slli_b:
case Intrinsic::mips_slli_h:
case Intrinsic::mips_slli_w:
case Intrinsic::mips_slli_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::SHL, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
[mips][msa] Added support for matching splat.[bhw] from normal IR (i.e. not intrinsics) splat.d is implemented but this subtest is currently disabled. This is because it is difficult to match the appropriate IR on MIPS32. There is a patch under review that should help with this so I hope to enable the subtest soon. llvm-svn: 193680
2013-10-30 21:07:44 +08:00
case Intrinsic::mips_splat_b:
case Intrinsic::mips_splat_h:
case Intrinsic::mips_splat_w:
case Intrinsic::mips_splat_d:
// We can't lower via VECTOR_SHUFFLE because it requires constant shuffle
// masks, nor can we lower via BUILD_VECTOR & EXTRACT_VECTOR_ELT because
// EXTRACT_VECTOR_ELT can't extract i64's on MIPS32.
// Instead we lower to MipsISD::VSHF and match from there.
return DAG.getNode(MipsISD::VSHF, DL, Op->getValueType(0),
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
lowerMSASplatZExt(Op, 2, DAG), Op->getOperand(1),
[mips][msa] Added support for matching splat.[bhw] from normal IR (i.e. not intrinsics) splat.d is implemented but this subtest is currently disabled. This is because it is difficult to match the appropriate IR on MIPS32. There is a patch under review that should help with this so I hope to enable the subtest soon. llvm-svn: 193680
2013-10-30 21:07:44 +08:00
Op->getOperand(1));
[mips][msa] Added support for matching splati from normal IR (i.e. not intrinsics) Updated some of the vshf since they (correctly) emit splati's now llvm-svn: 191511
2013-09-27 19:48:57 +08:00
case Intrinsic::mips_splati_b:
case Intrinsic::mips_splati_h:
case Intrinsic::mips_splati_w:
case Intrinsic::mips_splati_d:
return DAG.getNode(MipsISD::VSHF, DL, Op->getValueType(0),
lowerMSASplatImm(Op, 2, DAG), Op->getOperand(1),
Op->getOperand(1));
[mips][msa] Added support for matching mulv, nlzc, sll, sra, srl, and subv from normal IR (i.e. not intrinsics) llvm-svn: 190518
2013-09-11 19:58:30 +08:00
case Intrinsic::mips_sra_b:
case Intrinsic::mips_sra_h:
case Intrinsic::mips_sra_w:
case Intrinsic::mips_sra_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::SRA, DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Mask vectors holding shift amounts Masked vectors which hold shift amounts when creating the following nodes: ISD::SHL, ISD::SRL or ISD::SRA. Instructions that use said nodes, which have had their arguments altered are sll, srl, sra, bneg, bclr and bset. For said instructions, the shift amount or the bit position that is specified in the corresponding vector elements will be interpreted as the shift amount/bit position modulo the size of the element in bits. The problem lies in compiling with -O2 enabled, where the instructions for formats .w and .d are not generated, but are instead optimized away. In this case, having shift amounts that are either negative or greater than the element bit size results in generation of incorrect results when constant folding. We remedy this by masking the operands for the nodes mentioned above before actually creating them, so that the final result is correct before placed into the constant pool. Patch by Stefan Maksimovic. Differential Revision: https://reviews.llvm.org/D31331 llvm-svn: 300839
2017-04-20 21:26:46 +08:00
truncateVecElts(Op, DAG));
[mips][msa] Added support for matching slli, srai, and srli from normal IR (i.e. not intrinsics) llvm-svn: 191285
2013-09-24 18:28:18 +08:00
case Intrinsic::mips_srai_b:
case Intrinsic::mips_srai_h:
case Intrinsic::mips_srai_w:
case Intrinsic::mips_srai_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::SRA, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
case Intrinsic::mips_srari_b:
case Intrinsic::mips_srari_h:
case Intrinsic::mips_srari_w:
case Intrinsic::mips_srari_d: {
// Report an error for out of range values.
int64_t Max;
switch (Intrinsic) {
case Intrinsic::mips_srari_b: Max = 7; break;
case Intrinsic::mips_srari_h: Max = 15; break;
case Intrinsic::mips_srari_w: Max = 31; break;
case Intrinsic::mips_srari_d: Max = 63; break;
default: llvm_unreachable("Unmatched intrinsic");
}
int64_t Value = cast<ConstantSDNode>(Op->getOperand(2))->getSExtValue();
if (Value < 0 || Value > Max)
report_fatal_error("Immediate out of range");
return SDValue();
}
[mips][msa] Added support for matching mulv, nlzc, sll, sra, srl, and subv from normal IR (i.e. not intrinsics) llvm-svn: 190518
2013-09-11 19:58:30 +08:00
case Intrinsic::mips_srl_b:
case Intrinsic::mips_srl_h:
case Intrinsic::mips_srl_w:
case Intrinsic::mips_srl_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::SRL, DL, Op->getValueType(0), Op->getOperand(1),
[mips][msa] Mask vectors holding shift amounts Masked vectors which hold shift amounts when creating the following nodes: ISD::SHL, ISD::SRL or ISD::SRA. Instructions that use said nodes, which have had their arguments altered are sll, srl, sra, bneg, bclr and bset. For said instructions, the shift amount or the bit position that is specified in the corresponding vector elements will be interpreted as the shift amount/bit position modulo the size of the element in bits. The problem lies in compiling with -O2 enabled, where the instructions for formats .w and .d are not generated, but are instead optimized away. In this case, having shift amounts that are either negative or greater than the element bit size results in generation of incorrect results when constant folding. We remedy this by masking the operands for the nodes mentioned above before actually creating them, so that the final result is correct before placed into the constant pool. Patch by Stefan Maksimovic. Differential Revision: https://reviews.llvm.org/D31331 llvm-svn: 300839
2017-04-20 21:26:46 +08:00
truncateVecElts(Op, DAG));
[mips][msa] Added support for matching slli, srai, and srli from normal IR (i.e. not intrinsics) llvm-svn: 191285
2013-09-24 18:28:18 +08:00
case Intrinsic::mips_srli_b:
case Intrinsic::mips_srli_h:
case Intrinsic::mips_srli_w:
case Intrinsic::mips_srli_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::SRL, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
case Intrinsic::mips_srlri_b:
case Intrinsic::mips_srlri_h:
case Intrinsic::mips_srlri_w:
case Intrinsic::mips_srlri_d: {
// Report an error for out of range values.
int64_t Max;
switch (Intrinsic) {
case Intrinsic::mips_srlri_b: Max = 7; break;
case Intrinsic::mips_srlri_h: Max = 15; break;
case Intrinsic::mips_srlri_w: Max = 31; break;
case Intrinsic::mips_srlri_d: Max = 63; break;
default: llvm_unreachable("Unmatched intrinsic");
}
int64_t Value = cast<ConstantSDNode>(Op->getOperand(2))->getSExtValue();
if (Value < 0 || Value > Max)
report_fatal_error("Immediate out of range");
return SDValue();
}
[mips][msa] Added support for matching mulv, nlzc, sll, sra, srl, and subv from normal IR (i.e. not intrinsics) llvm-svn: 190518
2013-09-11 19:58:30 +08:00
case Intrinsic::mips_subv_b:
case Intrinsic::mips_subv_h:
case Intrinsic::mips_subv_w:
case Intrinsic::mips_subv_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::SUB, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
[mips][msa] Added support for matching addvi, and subvi from normal IR (i.e. not intrinsics) llvm-svn: 191203
2013-09-23 22:29:55 +08:00
case Intrinsic::mips_subvi_b:
case Intrinsic::mips_subvi_h:
case Intrinsic::mips_subvi_w:
case Intrinsic::mips_subvi_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::SUB, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
[mips][msa] Added support for matching vshf from normal IR (i.e. not intrinsics) llvm-svn: 191301
2013-09-24 22:02:15 +08:00
case Intrinsic::mips_vshf_b:
case Intrinsic::mips_vshf_h:
case Intrinsic::mips_vshf_w:
case Intrinsic::mips_vshf_d:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(MipsISD::VSHF, DL, Op->getValueType(0),
[mips][msa] Added support for matching vshf from normal IR (i.e. not intrinsics) llvm-svn: 191301
2013-09-24 22:02:15 +08:00
Op->getOperand(1), Op->getOperand(2), Op->getOperand(3));
[mips][msa] Added support for matching and, or, and xor from normal IR (i.e. not intrinsics) llvm-svn: 191194
2013-09-23 20:57:42 +08:00
case Intrinsic::mips_xor_v:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::XOR, DL, Op->getValueType(0), Op->getOperand(1),
Op->getOperand(2));
[mips][msa] Added support for matching andi, ori, nori, and xori from normal IR (i.e. not intrinsics) llvm-svn: 191293
2013-09-24 20:32:47 +08:00
case Intrinsic::mips_xori_b:
[mips][msa] Tidy up lowerMSABinaryIntr, lowerMSABinaryImmIntr, lowerMSABranchIntr, and lowerMSAUnaryIntr were trivially small functions. Inlined them into their callers. lowerMSASplat now takes its callers SDLoc instead of making a new one. No functional change. llvm-svn: 191503
2013-09-27 18:25:41 +08:00
return DAG.getNode(ISD::XOR, DL, Op->getValueType(0),
Op->getOperand(1), lowerMSASplatImm(Op, 2, DAG));
[Mips] Add support for llvm.thread.pointer intrinsic. This will be used to implement __builtin_thread_pointer in clang. Differential Revision: http://reviews.llvm.org/D19569 llvm-svn: 267743
2016-04-28 01:21:49 +08:00
case Intrinsic::thread_pointer: {
EVT PtrVT = getPointerTy(DAG.getDataLayout());
return DAG.getNode(MipsISD::ThreadPointer, DL, PtrVT);
}
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
}
}
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
static SDValue lowerMSALoadIntr(SDValue Op, SelectionDAG &DAG, unsigned Intr,
const MipsSubtarget &Subtarget) {
[mips][msa] Added load/store intrinsics. llvm-svn: 189476
2013-08-28 20:04:29 +08:00
SDLoc DL(Op);
SDValue ChainIn = Op->getOperand(0);
SDValue Address = Op->getOperand(2);
SDValue Offset = Op->getOperand(3);
EVT ResTy = Op->getValueType(0);
EVT PtrTy = Address->getValueType(0);
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
// For N64 addresses have the underlying type MVT::i64. This intrinsic
// however takes an i32 signed constant offset. The actual type of the
// intrinsic is a scaled signed i10.
if (Subtarget.isABI_N64())
Offset = DAG.getNode(ISD::SIGN_EXTEND, DL, PtrTy, Offset);
[mips][msa] Added load/store intrinsics. llvm-svn: 189476
2013-08-28 20:04:29 +08:00
Address = DAG.getNode(ISD::ADD, DL, PtrTy, Address, Offset);
[SelectionDAG] Get rid of bool parameters in SelectionDAG::getLoad, getStore, and friends. Summary: Instead, we take a single flags arg (a bitset). Also add a default 0 alignment, and change the order of arguments so the alignment comes before the flags. This greatly simplifies many callsites, and fixes a bug in AMDGPUISelLowering, wherein the order of the args to getLoad was inverted. It also greatly simplifies the process of adding another flag to getLoad. Reviewers: chandlerc, tstellarAMD Subscribers: jholewinski, arsenm, jyknight, dsanders, nemanjai, llvm-commits Differential Revision: http://reviews.llvm.org/D22249 llvm-svn: 275592
2016-07-16 02:27:10 +08:00
return DAG.getLoad(ResTy, DL, ChainIn, Address, MachinePointerInfo(),
/* Alignment = */ 16);
[mips][msa] Added load/store intrinsics. llvm-svn: 189476
2013-08-28 20:04:29 +08:00
}
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
SDValue MipsSETargetLowering::lowerINTRINSIC_W_CHAIN(SDValue Op,
SelectionDAG &DAG) const {
[mips][msa] Added load/store intrinsics. llvm-svn: 189476
2013-08-28 20:04:29 +08:00
unsigned Intr = cast<ConstantSDNode>(Op->getOperand(1))->getZExtValue();
switch (Intr) {
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
default:
return SDValue();
case Intrinsic::mips_extp:
return lowerDSPIntr(Op, DAG, MipsISD::EXTP);
case Intrinsic::mips_extpdp:
return lowerDSPIntr(Op, DAG, MipsISD::EXTPDP);
case Intrinsic::mips_extr_w:
return lowerDSPIntr(Op, DAG, MipsISD::EXTR_W);
case Intrinsic::mips_extr_r_w:
return lowerDSPIntr(Op, DAG, MipsISD::EXTR_R_W);
case Intrinsic::mips_extr_rs_w:
return lowerDSPIntr(Op, DAG, MipsISD::EXTR_RS_W);
case Intrinsic::mips_extr_s_h:
return lowerDSPIntr(Op, DAG, MipsISD::EXTR_S_H);
case Intrinsic::mips_mthlip:
return lowerDSPIntr(Op, DAG, MipsISD::MTHLIP);
case Intrinsic::mips_mulsaq_s_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::MULSAQ_S_W_PH);
case Intrinsic::mips_maq_s_w_phl:
return lowerDSPIntr(Op, DAG, MipsISD::MAQ_S_W_PHL);
case Intrinsic::mips_maq_s_w_phr:
return lowerDSPIntr(Op, DAG, MipsISD::MAQ_S_W_PHR);
case Intrinsic::mips_maq_sa_w_phl:
return lowerDSPIntr(Op, DAG, MipsISD::MAQ_SA_W_PHL);
case Intrinsic::mips_maq_sa_w_phr:
return lowerDSPIntr(Op, DAG, MipsISD::MAQ_SA_W_PHR);
case Intrinsic::mips_dpaq_s_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPAQ_S_W_PH);
case Intrinsic::mips_dpsq_s_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPSQ_S_W_PH);
case Intrinsic::mips_dpaq_sa_l_w:
return lowerDSPIntr(Op, DAG, MipsISD::DPAQ_SA_L_W);
case Intrinsic::mips_dpsq_sa_l_w:
return lowerDSPIntr(Op, DAG, MipsISD::DPSQ_SA_L_W);
case Intrinsic::mips_dpaqx_s_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPAQX_S_W_PH);
case Intrinsic::mips_dpaqx_sa_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPAQX_SA_W_PH);
case Intrinsic::mips_dpsqx_s_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPSQX_S_W_PH);
case Intrinsic::mips_dpsqx_sa_w_ph:
return lowerDSPIntr(Op, DAG, MipsISD::DPSQX_SA_W_PH);
[mips][msa] Added load/store intrinsics. llvm-svn: 189476
2013-08-28 20:04:29 +08:00
case Intrinsic::mips_ld_b:
case Intrinsic::mips_ld_h:
case Intrinsic::mips_ld_w:
case Intrinsic::mips_ld_d:
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
return lowerMSALoadIntr(Op, DAG, Intr, Subtarget);
[mips][msa] Added load/store intrinsics. llvm-svn: 189476
2013-08-28 20:04:29 +08:00
}
}
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
static SDValue lowerMSAStoreIntr(SDValue Op, SelectionDAG &DAG, unsigned Intr,
const MipsSubtarget &Subtarget) {
[mips][msa] Added load/store intrinsics. llvm-svn: 189476
2013-08-28 20:04:29 +08:00
SDLoc DL(Op);
SDValue ChainIn = Op->getOperand(0);
SDValue Value = Op->getOperand(2);
SDValue Address = Op->getOperand(3);
SDValue Offset = Op->getOperand(4);
EVT PtrTy = Address->getValueType(0);
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
// For N64 addresses have the underlying type MVT::i64. This intrinsic
// however takes an i32 signed constant offset. The actual type of the
// intrinsic is a scaled signed i10.
if (Subtarget.isABI_N64())
Offset = DAG.getNode(ISD::SIGN_EXTEND, DL, PtrTy, Offset);
[mips][msa] Added load/store intrinsics. llvm-svn: 189476
2013-08-28 20:04:29 +08:00
Address = DAG.getNode(ISD::ADD, DL, PtrTy, Address, Offset);
[SelectionDAG] Get rid of bool parameters in SelectionDAG::getLoad, getStore, and friends. Summary: Instead, we take a single flags arg (a bitset). Also add a default 0 alignment, and change the order of arguments so the alignment comes before the flags. This greatly simplifies many callsites, and fixes a bug in AMDGPUISelLowering, wherein the order of the args to getLoad was inverted. It also greatly simplifies the process of adding another flag to getLoad. Reviewers: chandlerc, tstellarAMD Subscribers: jholewinski, arsenm, jyknight, dsanders, nemanjai, llvm-commits Differential Revision: http://reviews.llvm.org/D22249 llvm-svn: 275592
2016-07-16 02:27:10 +08:00
return DAG.getStore(ChainIn, DL, Value, Address, MachinePointerInfo(),
/* Alignment = */ 16);
[mips][msa] Added load/store intrinsics. llvm-svn: 189476
2013-08-28 20:04:29 +08:00
}
SDValue MipsSETargetLowering::lowerINTRINSIC_VOID(SDValue Op,
SelectionDAG &DAG) const {
unsigned Intr = cast<ConstantSDNode>(Op->getOperand(1))->getZExtValue();
switch (Intr) {
default:
return SDValue();
case Intrinsic::mips_st_b:
case Intrinsic::mips_st_h:
case Intrinsic::mips_st_w:
case Intrinsic::mips_st_d:
[mips] Fix Mips MSA instrinsics The usage of some MIPS MSA instrinsics that took immediates could crash LLVM during lowering. This patch addresses that behaviour. Crucially this patch also makes the use of intrinsics with out of range immediates as producing an internal error. The ld,st instrinsics would trigger an assertion failure for MIPS64 as their lowering would attempt to add an i32 offset to a i64 pointer. Reviewers: vkalintiris, slthakur Differential Revision: https://reviews.llvm.org/D25438 llvm-svn: 291571
2017-01-11 00:40:57 +08:00
return lowerMSAStoreIntr(Op, DAG, Intr, Subtarget);
[mips] Move MipsTargetLowering::lowerINTRINSIC_W_CHAIN and lowerINTRINSIC_WO_CHAIN into MipsSETargetLowering. No functionality changes. llvm-svn: 179444
2013-04-13 10:13:30 +08:00
}
}
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
/// \brief Check if the given BuildVectorSDNode is a splat.
/// This method currently relies on DAG nodes being reused when equivalent,
/// so it's possible for this to return false even when isConstantSplat returns
/// true.
static bool isSplatVector(const BuildVectorSDNode *N) {
unsigned int nOps = N->getNumOperands();
[mips][msa] Added support for matching bmnz, bmnzi, bmz, and bmzi from normal IR (i.e. not intrinsics) Also corrected the definition of the intrinsics for these instructions (the result register is also the first operand), and added intrinsics for bsel and bseli to clang (they already existed in the backend). These four operations are mostly equivalent to bsel, and bseli (the difference is which operand is tied to the result). As a result some of the tests changed as described below. bitwise.ll: - bsel.v test adapted so that the mask is unknown at compile-time. This stops it emitting bmnzi.b instead of the intended bsel.v. - The bseli.b test now tests the right thing. Namely the case when one of the values is an uimm8, rather than when the condition is a uimm8 (which is covered by bmnzi.b) compare.ll: - bsel.v tests now (correctly) emits bmnz.v instead of bsel.v because this is the same operation (see MSA.txt). i8.ll - CHECK-DAG-ized test. - bmzi.b test now (correctly) emits equivalent bmnzi.b with swapped operands because this is the same operation (see MSA.txt). - bseli.b still emits bseli.b though because the immediate makes it distinguishable from bmnzi.b. vec.ll: - CHECK-DAG-ized test. - bmz.v tests now (correctly) emits bmnz.v with swapped operands (see MSA.txt). - bsel.v tests now (correctly) emits bmnz.v with swapped operands (see MSA.txt). llvm-svn: 193693
2013-10-30 23:20:38 +08:00
assert(nOps > 1 && "isSplatVector has 0 or 1 sized build vector");
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
SDValue Operand0 = N->getOperand(0);
for (unsigned int i = 1; i < nOps; ++i) {
if (N->getOperand(i) != Operand0)
return false;
}
return true;
}
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
// Lower ISD::EXTRACT_VECTOR_ELT into MipsISD::VEXTRACT_SEXT_ELT.
//
// The non-value bits resulting from ISD::EXTRACT_VECTOR_ELT are undefined. We
// choose to sign-extend but we could have equally chosen zero-extend. The
// DAGCombiner will fold any sign/zero extension of the ISD::EXTRACT_VECTOR_ELT
// result into this node later (possibly changing it to a zero-extend in the
// process).
SDValue MipsSETargetLowering::
lowerEXTRACT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const {
SDLoc DL(Op);
EVT ResTy = Op->getValueType(0);
SDValue Op0 = Op->getOperand(0);
[mips][msa] Implemented extract_vector_elt for v4f32 or v2f64 For v4f32 and v2f64, EXTRACT_VECTOR_ELT is matched by a pseudo-insn which may be expanded to subregister copies and/or instructions as appropriate. llvm-svn: 191514
2013-09-27 20:17:32 +08:00
EVT VecTy = Op0->getValueType(0);
if (!VecTy.is128BitVector())
return SDValue();
if (ResTy.isInteger()) {
SDValue Op1 = Op->getOperand(1);
EVT EltTy = VecTy.getVectorElementType();
return DAG.getNode(MipsISD::VEXTRACT_SEXT_ELT, DL, ResTy, Op0, Op1,
DAG.getValueType(EltTy));
}
return Op;
[mips][msa] Added support for matching insert and copy from normal IR (i.e. not intrinsics) Changes to MIPS SelectionDAG: * Added nodes VEXTRACT_[SZ]EXT_ELT to represent extract and extend in a single operation and implemented the DAG combines necessary to fold sign/zero extends into the extract. llvm-svn: 191199
2013-09-23 22:03:12 +08:00
}
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
static bool isConstantOrUndef(const SDValue Op) {
[DAG] use isUndef() ; NFCI llvm-svn: 263448
2016-03-15 01:28:46 +08:00
if (Op->isUndef())
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
return true;
[mips] Use isa<> instead of dyn_cast<> with unused value. NFC. Reviewers: dsanders Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D7615 llvm-svn: 229138
2015-02-14 03:12:16 +08:00
if (isa<ConstantSDNode>(Op))
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
return true;
[mips] Use isa<> instead of dyn_cast<> with unused value. NFC. Reviewers: dsanders Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D7615 llvm-svn: 229138
2015-02-14 03:12:16 +08:00
if (isa<ConstantFPSDNode>(Op))
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
return true;
return false;
}
static bool isConstantOrUndefBUILD_VECTOR(const BuildVectorSDNode *Op) {
for (unsigned i = 0; i < Op->getNumOperands(); ++i)
if (isConstantOrUndef(Op->getOperand(i)))
return true;
return false;
}
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
// Lowers ISD::BUILD_VECTOR into appropriate SelectionDAG nodes for the
// backend.
//
// Lowers according to the following rules:
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
// - Constant splats are legal as-is as long as the SplatBitSize is a power of
// 2 less than or equal to 64 and the value fits into a signed 10-bit
// immediate
// - Constant splats are lowered to bitconverted BUILD_VECTORs if SplatBitSize
// is a power of 2 less than or equal to 64 and the value does not fit into a
// signed 10-bit immediate
// - Non-constant splats are legal as-is.
// - Non-constant non-splats are lowered to sequences of INSERT_VECTOR_ELT.
// - All others are illegal and must be expanded.
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
SDValue MipsSETargetLowering::lowerBUILD_VECTOR(SDValue Op,
SelectionDAG &DAG) const {
BuildVectorSDNode *Node = cast<BuildVectorSDNode>(Op);
EVT ResTy = Op->getValueType(0);
SDLoc DL(Op);
APInt SplatValue, SplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
if (!Subtarget.hasMSA() || !ResTy.is128BitVector())
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
return SDValue();
if (Node->isConstantSplat(SplatValue, SplatUndef, SplatBitSize,
HasAnyUndefs, 8,
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
!Subtarget.isLittle()) && SplatBitSize <= 64) {
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
// We can only cope with 8, 16, 32, or 64-bit elements
if (SplatBitSize != 8 && SplatBitSize != 16 && SplatBitSize != 32 &&
SplatBitSize != 64)
return SDValue();
[mips][msa] Accept more values for constant splats This patches teaches the MIPS backend to accept more values for constant splats. Previously, only 10 bit signed immediates or values that could be loaded using an ldi.[bhwd] instruction would be acceptted. This patch relaxes that constraint so that any constant value that be splatted is accepted. As a result, the constant pool is used less for vector operations, and the suite of bit manipulation instructions b(clr|set|neg)i can now be used with the full range of their immediate operand. Reviewers: slthakur Differential Revision: https://reviews.llvm.org/D30640 llvm-svn: 297457
2017-03-10 21:27:14 +08:00
// If the value isn't an integer type we will have to bitcast
// from an integer type first. Also, if there are any undefs, we must
// lower them to defined values first.
if (ResTy.isInteger() && !HasAnyUndefs)
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
return Op;
EVT ViaVecTy;
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
switch (SplatBitSize) {
default:
return SDValue();
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
case 8:
ViaVecTy = MVT::v16i8;
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
break;
case 16:
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
ViaVecTy = MVT::v8i16;
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
break;
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
case 32:
ViaVecTy = MVT::v4i32;
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
break;
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
case 64:
// There's no fill.d to fall back on for 64-bit values
return SDValue();
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
}
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
// SelectionDAG::getConstant will promote SplatValue appropriately.
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
SDValue Result = DAG.getConstant(SplatValue, DL, ViaVecTy);
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
Fix illegal DAG produced by SelectionDAG::getConstant() for v2i64 type Summary: When getConstant() is called for an expanded vector type, it is split into multiple scalar constants which are then combined using appropriate build_vector and bitcast operations. In addition to the usual big/little endian differences, the case where the element-order of the vector does not have the same endianness as the elements themselves is also accounted for. For example, for v4i32 on big-endian MIPS, the byte-order of the vector is <3210,7654,BA98,FEDC>. For little-endian, it is <0123,4567,89AB,CDEF>. Handling this case turns out to be a nop since getConstant() returns a splatted vector (so reversing the element order doesn't change the value) This fixes a number of cases in MIPS MSA where calling getConstant() during operation legalization introduces illegal types (e.g. to legalize v2i64 UNDEF into a v2i64 BUILD_VECTOR of illegal i64 zeros). It should also handle bigger differences between illegal and legal types such as legalizing v2i64 into v8i16. lowerMSASplatImm() in the MIPS backend no longer needs to avoid calling getConstant() so this function has been updated in the same patch. For the sake of transparency, the steps I've taken since the review are: * Added 'virtual' to isVectorEltOrderLittleEndian() as requested. This revealed that the MIPS tests were falsely passing because a polymorphic function was not actually polymorphic in the reviewed patch. * Fixed the tests that were now failing. This involved deleting the code to handle the MIPS MSA element-order (which was previously doing an byte-order swap instead of an element-order swap). This left isVectorEltOrderLittleEndian() unused and it was deleted. * Fixed build failures caused by rebasing beyond r194467-r194472. These build failures involved the bset, bneg, and bclr instructions added in these commits using lowerMSASplatImm() in a way that was no longer valid after this patch. Some of these were fixed by calling SelectionDAG::getConstant() instead, others were fixed by a new function getBuildVectorSplat() that provided the removed functionality of lowerMSASplatImm() in a more sensible way. Reviewers: bkramer Reviewed By: bkramer CC: llvm-commits Differential Revision: http://llvm-reviews.chandlerc.com/D1973 llvm-svn: 194811
2013-11-15 20:56:49 +08:00
// Bitcast to the type we originally wanted
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
if (ViaVecTy != ResTy)
Result = DAG.getNode(ISD::BITCAST, SDLoc(Node), ResTy, Result);
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
return Result;
[mips][msa] Remove the VSPLAT and VSPLATD nodes in favour of matching BUILD_VECTOR. Most constant BUILD_VECTOR's are matched using ComplexPatterns which cover bitcasted as well as normal vectors. However, it doesn't seem to be possible to match ldi.[bhwd] in a type-agnostic manner (e.g. to support the widest range of immediates, it should be possible to use ldi.b to load v2i64) using TableGen so ldi.[bhwd] is matched using custom code in MipsSEISelDAGToDAG.cpp This made the majority of the constant splat BUILD_VECTOR lowering redundant. The only transformation remaining for constant splats is when an (up-to) 32-bit constant splat is possible but the value does not fit into a 10-bit signed integer. In this case, the BUILD_VECTOR is transformed into a bitcasted BUILD_VECTOR so that fill.[bhw] can be used to splat the vector from a GPR32 register (which is initialized using the usual lui/addui sequence). There are no additional tests since this is a re-implementation of previous functionality. The change is intended to make it easier to implement some of the upcoming instruction selection patches since they can rely on existing support for BUILD_VECTOR's in the DAGCombiner. compare_float.ll changed slightly because a BITCAST is no longer introduced during legalization. llvm-svn: 191299
2013-09-24 21:33:07 +08:00
} else if (isSplatVector(Node))
return Op;
else if (!isConstantOrUndefBUILD_VECTOR(Node)) {
[mips][msa] Non-constant BUILD_VECTOR's should be expanded to INSERT_VECTOR_ELT instead of memory operations. The resulting code is the same length, but doesnt cause memory traffic or latency. llvm-svn: 191297
2013-09-24 21:16:15 +08:00
// Use INSERT_VECTOR_ELT operations rather than expand to stores.
// The resulting code is the same length as the expansion, but it doesn't
// use memory operations
EVT ResTy = Node->getValueType(0);
assert(ResTy.isVector());
unsigned NumElts = ResTy.getVectorNumElements();
SDValue Vector = DAG.getUNDEF(ResTy);
for (unsigned i = 0; i < NumElts; ++i) {
Vector = DAG.getNode(ISD::INSERT_VECTOR_ELT, DL, ResTy, Vector,
Node->getOperand(i),
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
DAG.getConstant(i, DL, MVT::i32));
[mips][msa] Non-constant BUILD_VECTOR's should be expanded to INSERT_VECTOR_ELT instead of memory operations. The resulting code is the same length, but doesnt cause memory traffic or latency. llvm-svn: 191297
2013-09-24 21:16:15 +08:00
}
return Vector;
}
[mips][msa] Implemented build_vector using ldi, fill, and custom SelectionDAG nodes (VSPLAT and VSPLATD) Note: There's a later patch on my branch that re-implements this to select build_vector without the custom SelectionDAG nodes. The future patch avoids the constant-folding problems stemming from the custom node (i.e. it doesn't need to re-implement all the DAG combines related to BUILD_VECTOR). Changes to MIPS specific SelectionDAG nodes: * Added VSPLAT This is a special case of BUILD_VECTOR that covers the case the BUILD_VECTOR is a splat operation. * Added VSPLATD This is a special case of VSPLAT that handles the cases when v2i64 is legal llvm-svn: 191191
2013-09-23 20:02:46 +08:00
return SDValue();
}
[mips][msa] Added support for matching shf from normal IR (i.e. not intrinsics) llvm-svn: 191302
2013-09-24 22:20:00 +08:00
// Lower VECTOR_SHUFFLE into SHF (if possible).
//
// SHF splits the vector into blocks of four elements, then shuffles these
// elements according to a <4 x i2> constant (encoded as an integer immediate).
//
// It is therefore possible to lower into SHF when the mask takes the form:
// <a, b, c, d, a+4, b+4, c+4, d+4, a+8, b+8, c+8, d+8, ...>
// When undef's appear they are treated as if they were whatever value is
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// necessary in order to fit the above forms.
[mips][msa] Added support for matching shf from normal IR (i.e. not intrinsics) llvm-svn: 191302
2013-09-24 22:20:00 +08:00
//
// For example:
// %2 = shufflevector <8 x i16> %0, <8 x i16> undef,
// <8 x i32> <i32 3, i32 2, i32 1, i32 0,
// i32 7, i32 6, i32 5, i32 4>
// is lowered to:
// (SHF_H $w0, $w1, 27)
// where the 27 comes from:
// 3 + (2 << 2) + (1 << 4) + (0 << 6)
static SDValue lowerVECTOR_SHUFFLE_SHF(SDValue Op, EVT ResTy,
SmallVector<int, 16> Indices,
SelectionDAG &DAG) {
int SHFIndices[4] = { -1, -1, -1, -1 };
if (Indices.size() < 4)
return SDValue();
for (unsigned i = 0; i < 4; ++i) {
for (unsigned j = i; j < Indices.size(); j += 4) {
int Idx = Indices[j];
// Convert from vector index to 4-element subvector index
// If an index refers to an element outside of the subvector then give up
if (Idx != -1) {
Idx -= 4 * (j / 4);
if (Idx < 0 || Idx >= 4)
return SDValue();
}
// If the mask has an undef, replace it with the current index.
// Note that it might still be undef if the current index is also undef
if (SHFIndices[i] == -1)
SHFIndices[i] = Idx;
// Check that non-undef values are the same as in the mask. If they
// aren't then give up
if (!(Idx == -1 || Idx == SHFIndices[i]))
return SDValue();
}
}
// Calculate the immediate. Replace any remaining undefs with zero
APInt Imm(32, 0);
for (int i = 3; i >= 0; --i) {
int Idx = SHFIndices[i];
if (Idx == -1)
Idx = 0;
Imm <<= 2;
Imm |= Idx & 0x3;
}
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
SDLoc DL(Op);
return DAG.getNode(MipsISD::SHF, DL, ResTy,
DAG.getConstant(Imm, DL, MVT::i32), Op->getOperand(0));
[mips][msa] Added support for matching shf from normal IR (i.e. not intrinsics) llvm-svn: 191302
2013-09-24 22:20:00 +08:00
}
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
/// Determine whether a range fits a regular pattern of values.
/// This function accounts for the possibility of jumping over the End iterator.
template <typename ValType>
static bool
fitsRegularPattern(typename SmallVectorImpl<ValType>::const_iterator Begin,
unsigned CheckStride,
typename SmallVectorImpl<ValType>::const_iterator End,
ValType ExpectedIndex, unsigned ExpectedIndexStride) {
auto &I = Begin;
while (I != End) {
if (*I != -1 && *I != ExpectedIndex)
return false;
ExpectedIndex += ExpectedIndexStride;
// Incrementing past End is undefined behaviour so we must increment one
// step at a time and check for End at each step.
for (unsigned n = 0; n < CheckStride && I != End; ++n, ++I)
; // Empty loop body.
}
return true;
}
// Determine whether VECTOR_SHUFFLE is a SPLATI.
//
// It is a SPLATI when the mask is:
// <x, x, x, ...>
// where x is any valid index.
//
// When undef's appear in the mask they are treated as if they were whatever
// value is necessary in order to fit the above form.
static bool isVECTOR_SHUFFLE_SPLATI(SDValue Op, EVT ResTy,
SmallVector<int, 16> Indices,
SelectionDAG &DAG) {
assert((Indices.size() % 2) == 0);
int SplatIndex = -1;
for (const auto &V : Indices) {
if (V != -1) {
SplatIndex = V;
break;
}
}
return fitsRegularPattern<int>(Indices.begin(), 1, Indices.end(), SplatIndex,
0);
}
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
// Lower VECTOR_SHUFFLE into ILVEV (if possible).
//
// ILVEV interleaves the even elements from each vector.
//
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// It is possible to lower into ILVEV when the mask consists of two of the
// following forms interleaved:
// <0, 2, 4, ...>
// <n, n+2, n+4, ...>
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
// where n is the number of elements in the vector.
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// For example:
// <0, 0, 2, 2, 4, 4, ...>
// <0, n, 2, n+2, 4, n+4, ...>
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
//
// When undef's appear in the mask they are treated as if they were whatever
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// value is necessary in order to fit the above forms.
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
static SDValue lowerVECTOR_SHUFFLE_ILVEV(SDValue Op, EVT ResTy,
SmallVector<int, 16> Indices,
SelectionDAG &DAG) {
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
assert((Indices.size() % 2) == 0);
SDValue Wt;
SDValue Ws;
const auto &Begin = Indices.begin();
const auto &End = Indices.end();
// Check even elements are taken from the even elements of one half or the
// other and pick an operand accordingly.
if (fitsRegularPattern<int>(Begin, 2, End, 0, 2))
Wt = Op->getOperand(0);
else if (fitsRegularPattern<int>(Begin, 2, End, Indices.size(), 2))
Wt = Op->getOperand(1);
else
return SDValue();
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// Check odd elements are taken from the even elements of one half or the
// other and pick an operand accordingly.
if (fitsRegularPattern<int>(Begin + 1, 2, End, 0, 2))
Ws = Op->getOperand(0);
else if (fitsRegularPattern<int>(Begin + 1, 2, End, Indices.size(), 2))
Ws = Op->getOperand(1);
else
return SDValue();
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
return DAG.getNode(MipsISD::ILVEV, SDLoc(Op), ResTy, Ws, Wt);
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
}
// Lower VECTOR_SHUFFLE into ILVOD (if possible).
//
// ILVOD interleaves the odd elements from each vector.
//
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// It is possible to lower into ILVOD when the mask consists of two of the
// following forms interleaved:
// <1, 3, 5, ...>
// <n+1, n+3, n+5, ...>
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
// where n is the number of elements in the vector.
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// For example:
// <1, 1, 3, 3, 5, 5, ...>
// <1, n+1, 3, n+3, 5, n+5, ...>
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
//
// When undef's appear in the mask they are treated as if they were whatever
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// value is necessary in order to fit the above forms.
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
static SDValue lowerVECTOR_SHUFFLE_ILVOD(SDValue Op, EVT ResTy,
SmallVector<int, 16> Indices,
SelectionDAG &DAG) {
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
assert((Indices.size() % 2) == 0);
SDValue Wt;
SDValue Ws;
const auto &Begin = Indices.begin();
const auto &End = Indices.end();
// Check even elements are taken from the odd elements of one half or the
// other and pick an operand accordingly.
if (fitsRegularPattern<int>(Begin, 2, End, 1, 2))
Wt = Op->getOperand(0);
else if (fitsRegularPattern<int>(Begin, 2, End, Indices.size() + 1, 2))
Wt = Op->getOperand(1);
else
return SDValue();
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// Check odd elements are taken from the odd elements of one half or the
// other and pick an operand accordingly.
if (fitsRegularPattern<int>(Begin + 1, 2, End, 1, 2))
Ws = Op->getOperand(0);
else if (fitsRegularPattern<int>(Begin + 1, 2, End, Indices.size() + 1, 2))
Ws = Op->getOperand(1);
else
return SDValue();
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
return DAG.getNode(MipsISD::ILVOD, SDLoc(Op), ResTy, Wt, Ws);
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
}
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// Lower VECTOR_SHUFFLE into ILVR (if possible).
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
//
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// ILVR interleaves consecutive elements from the right (lowest-indexed) half of
// each vector.
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
//
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// It is possible to lower into ILVR when the mask consists of two of the
// following forms interleaved:
// <0, 1, 2, ...>
// <n, n+1, n+2, ...>
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
// where n is the number of elements in the vector.
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// For example:
// <0, 0, 1, 1, 2, 2, ...>
// <0, n, 1, n+1, 2, n+2, ...>
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
//
// When undef's appear in the mask they are treated as if they were whatever
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// value is necessary in order to fit the above forms.
static SDValue lowerVECTOR_SHUFFLE_ILVR(SDValue Op, EVT ResTy,
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
SmallVector<int, 16> Indices,
SelectionDAG &DAG) {
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
assert((Indices.size() % 2) == 0);
SDValue Wt;
SDValue Ws;
const auto &Begin = Indices.begin();
const auto &End = Indices.end();
// Check even elements are taken from the right (lowest-indexed) elements of
// one half or the other and pick an operand accordingly.
if (fitsRegularPattern<int>(Begin, 2, End, 0, 1))
Wt = Op->getOperand(0);
else if (fitsRegularPattern<int>(Begin, 2, End, Indices.size(), 1))
Wt = Op->getOperand(1);
else
return SDValue();
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// Check odd elements are taken from the right (lowest-indexed) elements of
// one half or the other and pick an operand accordingly.
if (fitsRegularPattern<int>(Begin + 1, 2, End, 0, 1))
Ws = Op->getOperand(0);
else if (fitsRegularPattern<int>(Begin + 1, 2, End, Indices.size(), 1))
Ws = Op->getOperand(1);
else
return SDValue();
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
return DAG.getNode(MipsISD::ILVR, SDLoc(Op), ResTy, Ws, Wt);
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
}
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// Lower VECTOR_SHUFFLE into ILVL (if possible).
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
//
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// ILVL interleaves consecutive elements from the left (highest-indexed) half
// of each vector.
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
//
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// It is possible to lower into ILVL when the mask consists of two of the
// following forms interleaved:
// <x, x+1, x+2, ...>
// <n+x, n+x+1, n+x+2, ...>
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
// where n is the number of elements in the vector and x is half n.
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// For example:
// <x, x, x+1, x+1, x+2, x+2, ...>
// <x, n+x, x+1, n+x+1, x+2, n+x+2, ...>
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
//
// When undef's appear in the mask they are treated as if they were whatever
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// value is necessary in order to fit the above forms.
static SDValue lowerVECTOR_SHUFFLE_ILVL(SDValue Op, EVT ResTy,
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
SmallVector<int, 16> Indices,
SelectionDAG &DAG) {
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
assert((Indices.size() % 2) == 0);
unsigned HalfSize = Indices.size() / 2;
SDValue Wt;
SDValue Ws;
const auto &Begin = Indices.begin();
const auto &End = Indices.end();
// Check even elements are taken from the left (highest-indexed) elements of
// one half or the other and pick an operand accordingly.
if (fitsRegularPattern<int>(Begin, 2, End, HalfSize, 1))
Wt = Op->getOperand(0);
else if (fitsRegularPattern<int>(Begin, 2, End, Indices.size() + HalfSize, 1))
Wt = Op->getOperand(1);
else
return SDValue();
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// Check odd elements are taken from the left (highest-indexed) elements of
// one half or the other and pick an operand accordingly.
if (fitsRegularPattern<int>(Begin + 1, 2, End, HalfSize, 1))
Ws = Op->getOperand(0);
else if (fitsRegularPattern<int>(Begin + 1, 2, End, Indices.size() + HalfSize,
1))
Ws = Op->getOperand(1);
else
return SDValue();
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
return DAG.getNode(MipsISD::ILVL, SDLoc(Op), ResTy, Ws, Wt);
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
}
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
// Lower VECTOR_SHUFFLE into PCKEV (if possible).
//
// PCKEV copies the even elements of each vector into the result vector.
//
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// It is possible to lower into PCKEV when the mask consists of two of the
// following forms concatenated:
// <0, 2, 4, ...>
// <n, n+2, n+4, ...>
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
// where n is the number of elements in the vector.
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// For example:
// <0, 2, 4, ..., 0, 2, 4, ...>
// <0, 2, 4, ..., n, n+2, n+4, ...>
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
//
// When undef's appear in the mask they are treated as if they were whatever
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// value is necessary in order to fit the above forms.
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
static SDValue lowerVECTOR_SHUFFLE_PCKEV(SDValue Op, EVT ResTy,
SmallVector<int, 16> Indices,
SelectionDAG &DAG) {
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
assert((Indices.size() % 2) == 0);
SDValue Wt;
SDValue Ws;
const auto &Begin = Indices.begin();
const auto &Mid = Indices.begin() + Indices.size() / 2;
const auto &End = Indices.end();
if (fitsRegularPattern<int>(Begin, 1, Mid, 0, 2))
Wt = Op->getOperand(0);
else if (fitsRegularPattern<int>(Begin, 1, Mid, Indices.size(), 2))
Wt = Op->getOperand(1);
else
return SDValue();
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
if (fitsRegularPattern<int>(Mid, 1, End, 0, 2))
Ws = Op->getOperand(0);
else if (fitsRegularPattern<int>(Mid, 1, End, Indices.size(), 2))
Ws = Op->getOperand(1);
else
return SDValue();
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
return DAG.getNode(MipsISD::PCKEV, SDLoc(Op), ResTy, Ws, Wt);
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
}
// Lower VECTOR_SHUFFLE into PCKOD (if possible).
//
// PCKOD copies the odd elements of each vector into the result vector.
//
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// It is possible to lower into PCKOD when the mask consists of two of the
// following forms concatenated:
// <1, 3, 5, ...>
// <n+1, n+3, n+5, ...>
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
// where n is the number of elements in the vector.
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// For example:
// <1, 3, 5, ..., 1, 3, 5, ...>
// <1, 3, 5, ..., n+1, n+3, n+5, ...>
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
//
// When undef's appear in the mask they are treated as if they were whatever
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// value is necessary in order to fit the above forms.
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
static SDValue lowerVECTOR_SHUFFLE_PCKOD(SDValue Op, EVT ResTy,
SmallVector<int, 16> Indices,
SelectionDAG &DAG) {
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
assert((Indices.size() % 2) == 0);
SDValue Wt;
SDValue Ws;
const auto &Begin = Indices.begin();
const auto &Mid = Indices.begin() + Indices.size() / 2;
const auto &End = Indices.end();
if (fitsRegularPattern<int>(Begin, 1, Mid, 1, 2))
Wt = Op->getOperand(0);
else if (fitsRegularPattern<int>(Begin, 1, Mid, Indices.size() + 1, 2))
Wt = Op->getOperand(1);
else
return SDValue();
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
if (fitsRegularPattern<int>(Mid, 1, End, 1, 2))
Ws = Op->getOperand(0);
else if (fitsRegularPattern<int>(Mid, 1, End, Indices.size() + 1, 2))
Ws = Op->getOperand(1);
else
return SDValue();
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
return DAG.getNode(MipsISD::PCKOD, SDLoc(Op), ResTy, Ws, Wt);
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
}
[mips][msa] Added support for matching vshf from normal IR (i.e. not intrinsics) llvm-svn: 191301
2013-09-24 22:02:15 +08:00
// Lower VECTOR_SHUFFLE into VSHF.
//
// This mostly consists of converting the shuffle indices in Indices into a
// BUILD_VECTOR and adding it as an operand to the resulting VSHF. There is
// also code to eliminate unused operands of the VECTOR_SHUFFLE. For example,
// if the type is v8i16 and all the indices are less than 8 then the second
// operand is unused and can be replaced with anything. We choose to replace it
// with the used operand since this reduces the number of instructions overall.
static SDValue lowerVECTOR_SHUFFLE_VSHF(SDValue Op, EVT ResTy,
SmallVector<int, 16> Indices,
SelectionDAG &DAG) {
SmallVector<SDValue, 16> Ops;
SDValue Op0;
SDValue Op1;
EVT MaskVecTy = ResTy.changeVectorElementTypeToInteger();
EVT MaskEltTy = MaskVecTy.getVectorElementType();
bool Using1stVec = false;
bool Using2ndVec = false;
SDLoc DL(Op);
int ResTyNumElts = ResTy.getVectorNumElements();
for (int i = 0; i < ResTyNumElts; ++i) {
// Idx == -1 means UNDEF
int Idx = Indices[i];
if (0 <= Idx && Idx < ResTyNumElts)
Using1stVec = true;
if (ResTyNumElts <= Idx && Idx < ResTyNumElts * 2)
Using2ndVec = true;
}
for (SmallVector<int, 16>::iterator I = Indices.begin(); I != Indices.end();
++I)
Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" [DebugInfo] Add debug locations to constant SD nodes This adds debug location to constant nodes of Selection DAG and updates all places that create constants to pass debug locations (see PR13269). Can't guarantee that all locations are correct, but in a lot of cases choice is obvious, so most of them should be. At least all tests pass. Tests for these changes do not cover everything, instead just check it for SDNodes, ARM and AArch64 where it's easy to get incorrect locations on constants. This is not complete fix as FastISel contains workaround for wrong debug locations, which drops locations from instructions on processing constants, but there isn't currently a way to use debug locations from constants there as llvm::Constant doesn't cache it (yet). Although this is a bit different issue, not directly related to these changes. Differential Revision: http://reviews.llvm.org/D9084 llvm-svn: 235989
2015-04-28 22:05:47 +08:00
Ops.push_back(DAG.getTargetConstant(*I, DL, MaskEltTy));
[mips][msa] Added support for matching vshf from normal IR (i.e. not intrinsics) llvm-svn: 191301
2013-09-24 22:02:15 +08:00
[CodeGen] Add getBuildVector and getSplatBuildVector helpers. NFCI. Differential Revision: http://reviews.llvm.org/D17176 llvm-svn: 267606
2016-04-27 05:15:30 +08:00
SDValue MaskVec = DAG.getBuildVector(MaskVecTy, DL, Ops);
[mips][msa] Added support for matching vshf from normal IR (i.e. not intrinsics) llvm-svn: 191301
2013-09-24 22:02:15 +08:00
if (Using1stVec && Using2ndVec) {
Op0 = Op->getOperand(0);
Op1 = Op->getOperand(1);
} else if (Using1stVec)
Op0 = Op1 = Op->getOperand(0);
else if (Using2ndVec)
Op0 = Op1 = Op->getOperand(1);
else
llvm_unreachable("shuffle vector mask references neither vector operand?");
[mips] Correct lowering of VECTOR_SHUFFLE to VSHF. Summary: VECTOR_SHUFFLE concatenates the vectors in an vectorwise fashion. <0b00, 0b01> + <0b10, 0b11> -> <0b00, 0b01, 0b10, 0b11> VSHF concatenates the vectors in a bitwise fashion: <0b00, 0b01> + <0b10, 0b11> -> 0b0100 + 0b1110 -> 0b01001110 <0b10, 0b11, 0b00, 0b01> We must therefore swap the operands to get the correct result. The test case that discovered the issue was MultiSource/Benchmarks/nbench. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D3142 llvm-svn: 204480
2014-03-22 00:56:51 +08:00
// VECTOR_SHUFFLE concatenates the vectors in an vectorwise fashion.
// <0b00, 0b01> + <0b10, 0b11> -> <0b00, 0b01, 0b10, 0b11>
// VSHF concatenates the vectors in a bitwise fashion:
// <0b00, 0b01> + <0b10, 0b11> ->
// 0b0100 + 0b1110 -> 0b01001110
// <0b10, 0b11, 0b00, 0b01>
// We must therefore swap the operands to get the correct result.
return DAG.getNode(MipsISD::VSHF, DL, ResTy, MaskVec, Op1, Op0);
[mips][msa] Added support for matching vshf from normal IR (i.e. not intrinsics) llvm-svn: 191301
2013-09-24 22:02:15 +08:00
}
// Lower VECTOR_SHUFFLE into one of a number of instructions depending on the
// indices in the shuffle.
SDValue MipsSETargetLowering::lowerVECTOR_SHUFFLE(SDValue Op,
SelectionDAG &DAG) const {
ShuffleVectorSDNode *Node = cast<ShuffleVectorSDNode>(Op);
EVT ResTy = Op->getValueType(0);
if (!ResTy.is128BitVector())
return SDValue();
int ResTyNumElts = ResTy.getVectorNumElements();
SmallVector<int, 16> Indices;
for (int i = 0; i < ResTyNumElts; ++i)
Indices.push_back(Node->getMaskElt(i));
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
// splati.[bhwd] is preferable to the others but is matched from
// MipsISD::VSHF.
if (isVECTOR_SHUFFLE_SPLATI(Op, ResTy, Indices, DAG))
return lowerVECTOR_SHUFFLE_VSHF(Op, ResTy, Indices, DAG);
[CodeGen] Prefer "if (SDValue R = ...)" to "if (R.getNode())". NFCI. llvm-svn: 260316
2016-02-10 06:54:12 +08:00
SDValue Result;
if ((Result = lowerVECTOR_SHUFFLE_ILVEV(Op, ResTy, Indices, DAG)))
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
return Result;
[CodeGen] Prefer "if (SDValue R = ...)" to "if (R.getNode())". NFCI. llvm-svn: 260316
2016-02-10 06:54:12 +08:00
if ((Result = lowerVECTOR_SHUFFLE_ILVOD(Op, ResTy, Indices, DAG)))
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
return Result;
[CodeGen] Prefer "if (SDValue R = ...)" to "if (R.getNode())". NFCI. llvm-svn: 260316
2016-02-10 06:54:12 +08:00
if ((Result = lowerVECTOR_SHUFFLE_ILVL(Op, ResTy, Indices, DAG)))
[mips][msa] Added support for matching ilv[lr], ilvod, and ilvev from normal IR (i.e. not intrinsics) llvm-svn: 191304
2013-09-24 22:36:12 +08:00
return Result;
[CodeGen] Prefer "if (SDValue R = ...)" to "if (R.getNode())". NFCI. llvm-svn: 260316
2016-02-10 06:54:12 +08:00
if ((Result = lowerVECTOR_SHUFFLE_ILVR(Op, ResTy, Indices, DAG)))
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
return Result;
[CodeGen] Prefer "if (SDValue R = ...)" to "if (R.getNode())". NFCI. llvm-svn: 260316
2016-02-10 06:54:12 +08:00
if ((Result = lowerVECTOR_SHUFFLE_PCKEV(Op, ResTy, Indices, DAG)))
[mips][msa] Added support for matching pckev, and pckod from normal IR (i.e. not intrinsics) llvm-svn: 191306
2013-09-24 22:53:25 +08:00
return Result;
[CodeGen] Prefer "if (SDValue R = ...)" to "if (R.getNode())". NFCI. llvm-svn: 260316
2016-02-10 06:54:12 +08:00
if ((Result = lowerVECTOR_SHUFFLE_PCKOD(Op, ResTy, Indices, DAG)))
[mips] Correct and improve special-case shuffle instructions. Summary: The documentation writes vectors highest-index first whereas LLVM-IR writes them lowest-index first. As a result, instructions defined in terms of left_half() and right_half() had the halves reversed. In addition to correcting them, they have been improved to allow shuffles that use the same operand twice or in reverse order. For example, ilvev used to accept masks of the form: <0, n, 2, n+2, 4, n+4, ...> but now accepts: <0, 0, 2, 2, 4, 4, ...> <n, n, n+2, n+2, n+4, n+4, ...> <0, n, 2, n+2, 4, n+4, ...> <n, 0, n+2, 2, n+4, 4, ...> One further improvement is that splati.[bhwd] is now the preferred instruction for splat-like operations. The other special shuffles are no longer used for splats. This lead to the discovery that <0, 0, ...> would not cause splati.[hwd] to be selected and this has also been fixed. This fixes the enc-3des test from the test-suite on Mips64r6 with MSA. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9660 llvm-svn: 237689
2015-05-19 20:24:52 +08:00
return Result;
[CodeGen] Prefer "if (SDValue R = ...)" to "if (R.getNode())". NFCI. llvm-svn: 260316
2016-02-10 06:54:12 +08:00
if ((Result = lowerVECTOR_SHUFFLE_SHF(Op, ResTy, Indices, DAG)))
[mips][msa] Added support for matching shf from normal IR (i.e. not intrinsics) llvm-svn: 191302
2013-09-24 22:20:00 +08:00
return Result;
[mips][msa] Added support for matching vshf from normal IR (i.e. not intrinsics) llvm-svn: 191301
2013-09-24 22:02:15 +08:00
return lowerVECTOR_SHUFFLE_VSHF(Op, ResTy, Indices, DAG);
}
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MachineBasicBlock *
MipsSETargetLowering::emitBPOSGE32(MachineInstr &MI,
MachineBasicBlock *BB) const {
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
// $bb:
// bposge32_pseudo $vr0
// =>
// $bb:
// bposge32 $tbb
// $fbb:
// li $vr2, 0
// b $sink
// $tbb:
// li $vr1, 1
// $sink:
// $vr0 = phi($vr2, $fbb, $vr1, $tbb)
MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
Remove most of the TargetMachine::getSubtarget/getSubtargetImpl calls that don't take a Function argument from Mips. Notable exceptions: the AsmPrinter and MipsTargetObjectFile. The latter needs to be fixed, and the former will be fixed when the general AsmPrinter changes happen. llvm-svn: 227512
2015-01-30 07:27:36 +08:00
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
[mips] Rename register classes CPURegs and CPU64Regs. llvm-svn: 187832
2013-08-07 07:08:38 +08:00
const TargetRegisterClass *RC = &Mips::GPR32RegClass;
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
DebugLoc DL = MI.getDebugLoc();
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
const BasicBlock *LLVM_BB = BB->getBasicBlock();
[C++11] Replace llvm::next and llvm::prior with std::next and std::prev. Remove the old functions. llvm-svn: 202636
2014-03-02 20:27:27 +08:00
MachineFunction::iterator It = std::next(MachineFunction::iterator(BB));
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
MachineFunction *F = BB->getParent();
MachineBasicBlock *FBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *TBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *Sink = F->CreateMachineBasicBlock(LLVM_BB);
F->insert(It, FBB);
F->insert(It, TBB);
F->insert(It, Sink);
// Transfer the remainder of BB and its successor edges to Sink.
[C++11] Replace llvm::next and llvm::prior with std::next and std::prev. Remove the old functions. llvm-svn: 202636
2014-03-02 20:27:27 +08:00
Sink->splice(Sink->begin(), BB, std::next(MachineBasicBlock::iterator(MI)),
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
BB->end());
Sink->transferSuccessorsAndUpdatePHIs(BB);
// Add successors.
BB->addSuccessor(FBB);
BB->addSuccessor(TBB);
FBB->addSuccessor(Sink);
TBB->addSuccessor(Sink);
// Insert the real bposge32 instruction to $BB.
BuildMI(BB, DL, TII->get(Mips::BPOSGE32)).addMBB(TBB);
[mips][microMIPS] Implement APPEND, BPOSGE32C, MODSUB, MULSA.W.PH and MULSAQ_S.W.PH instructions Differential Revision: http://reviews.llvm.org/D14117 llvm-svn: 269408
2016-05-13 19:32:53 +08:00
// Insert the real bposge32c instruction to $BB.
BuildMI(BB, DL, TII->get(Mips::BPOSGE32C_MMR3)).addMBB(TBB);
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
// Fill $FBB.
unsigned VR2 = RegInfo.createVirtualRegister(RC);
BuildMI(*FBB, FBB->end(), DL, TII->get(Mips::ADDiu), VR2)
.addReg(Mips::ZERO).addImm(0);
BuildMI(*FBB, FBB->end(), DL, TII->get(Mips::B)).addMBB(Sink);
// Fill $TBB.
unsigned VR1 = RegInfo.createVirtualRegister(RC);
BuildMI(*TBB, TBB->end(), DL, TII->get(Mips::ADDiu), VR1)
.addReg(Mips::ZERO).addImm(1);
// Insert phi function to $Sink.
BuildMI(*Sink, Sink->begin(), DL, TII->get(Mips::PHI),
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MI.getOperand(0).getReg())
.addReg(VR2)
.addMBB(FBB)
.addReg(VR1)
.addMBB(TBB);
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MI.eraseFromParent(); // The pseudo instruction is gone now.
[mips] Define two subclasses of MipsTargetLowering. Mips16TargetLowering is for mips16 and MipsSETargetLowering is for mips32/64. No functionality changes. llvm-svn: 176917
2013-03-13 08:54:29 +08:00
return Sink;
}
[mips][msa] Added bnz.df, bnz.v, bz.df, and bz.v These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes, are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as a branch/mov sequence to evaluate to 0 or 1. Note: The resulting code is sub-optimal since it doesnt seem to be possible to feed the result of an intrinsic directly into a brcond. At the moment it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily evaluates the boolean twice. llvm-svn: 189478
2013-08-28 20:14:50 +08:00
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MachineBasicBlock *MipsSETargetLowering::emitMSACBranchPseudo(
MachineInstr &MI, MachineBasicBlock *BB, unsigned BranchOp) const {
[mips][msa] Added bnz.df, bnz.v, bz.df, and bz.v These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes, are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as a branch/mov sequence to evaluate to 0 or 1. Note: The resulting code is sub-optimal since it doesnt seem to be possible to feed the result of an intrinsic directly into a brcond. At the moment it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily evaluates the boolean twice. llvm-svn: 189478
2013-08-28 20:14:50 +08:00
// $bb:
// vany_nonzero $rd, $ws
// =>
// $bb:
// bnz.b $ws, $tbb
// b $fbb
// $fbb:
// li $rd1, 0
// b $sink
// $tbb:
// li $rd2, 1
// $sink:
// $rd = phi($rd1, $fbb, $rd2, $tbb)
MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
Remove most of the TargetMachine::getSubtarget/getSubtargetImpl calls that don't take a Function argument from Mips. Notable exceptions: the AsmPrinter and MipsTargetObjectFile. The latter needs to be fixed, and the former will be fixed when the general AsmPrinter changes happen. llvm-svn: 227512
2015-01-30 07:27:36 +08:00
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
[mips][msa] Added bnz.df, bnz.v, bz.df, and bz.v These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes, are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as a branch/mov sequence to evaluate to 0 or 1. Note: The resulting code is sub-optimal since it doesnt seem to be possible to feed the result of an intrinsic directly into a brcond. At the moment it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily evaluates the boolean twice. llvm-svn: 189478
2013-08-28 20:14:50 +08:00
const TargetRegisterClass *RC = &Mips::GPR32RegClass;
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
DebugLoc DL = MI.getDebugLoc();
[mips][msa] Added bnz.df, bnz.v, bz.df, and bz.v These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes, are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as a branch/mov sequence to evaluate to 0 or 1. Note: The resulting code is sub-optimal since it doesnt seem to be possible to feed the result of an intrinsic directly into a brcond. At the moment it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily evaluates the boolean twice. llvm-svn: 189478
2013-08-28 20:14:50 +08:00
const BasicBlock *LLVM_BB = BB->getBasicBlock();
[C++11] Replace llvm::next and llvm::prior with std::next and std::prev. Remove the old functions. llvm-svn: 202636
2014-03-02 20:27:27 +08:00
MachineFunction::iterator It = std::next(MachineFunction::iterator(BB));
[mips][msa] Added bnz.df, bnz.v, bz.df, and bz.v These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes, are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as a branch/mov sequence to evaluate to 0 or 1. Note: The resulting code is sub-optimal since it doesnt seem to be possible to feed the result of an intrinsic directly into a brcond. At the moment it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily evaluates the boolean twice. llvm-svn: 189478
2013-08-28 20:14:50 +08:00
MachineFunction *F = BB->getParent();
MachineBasicBlock *FBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *TBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *Sink = F->CreateMachineBasicBlock(LLVM_BB);
F->insert(It, FBB);
F->insert(It, TBB);
F->insert(It, Sink);
// Transfer the remainder of BB and its successor edges to Sink.
[C++11] Replace llvm::next and llvm::prior with std::next and std::prev. Remove the old functions. llvm-svn: 202636
2014-03-02 20:27:27 +08:00
Sink->splice(Sink->begin(), BB, std::next(MachineBasicBlock::iterator(MI)),
[mips][msa] Added bnz.df, bnz.v, bz.df, and bz.v These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes, are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as a branch/mov sequence to evaluate to 0 or 1. Note: The resulting code is sub-optimal since it doesnt seem to be possible to feed the result of an intrinsic directly into a brcond. At the moment it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily evaluates the boolean twice. llvm-svn: 189478
2013-08-28 20:14:50 +08:00
BB->end());
Sink->transferSuccessorsAndUpdatePHIs(BB);
// Add successors.
BB->addSuccessor(FBB);
BB->addSuccessor(TBB);
FBB->addSuccessor(Sink);
TBB->addSuccessor(Sink);
// Insert the real bnz.b instruction to $BB.
BuildMI(BB, DL, TII->get(BranchOp))
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
.addReg(MI.getOperand(1).getReg())
.addMBB(TBB);
[mips][msa] Added bnz.df, bnz.v, bz.df, and bz.v These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes, are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as a branch/mov sequence to evaluate to 0 or 1. Note: The resulting code is sub-optimal since it doesnt seem to be possible to feed the result of an intrinsic directly into a brcond. At the moment it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily evaluates the boolean twice. llvm-svn: 189478
2013-08-28 20:14:50 +08:00
// Fill $FBB.
unsigned RD1 = RegInfo.createVirtualRegister(RC);
BuildMI(*FBB, FBB->end(), DL, TII->get(Mips::ADDiu), RD1)
.addReg(Mips::ZERO).addImm(0);
BuildMI(*FBB, FBB->end(), DL, TII->get(Mips::B)).addMBB(Sink);
// Fill $TBB.
unsigned RD2 = RegInfo.createVirtualRegister(RC);
BuildMI(*TBB, TBB->end(), DL, TII->get(Mips::ADDiu), RD2)
.addReg(Mips::ZERO).addImm(1);
// Insert phi function to $Sink.
BuildMI(*Sink, Sink->begin(), DL, TII->get(Mips::PHI),
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MI.getOperand(0).getReg())
.addReg(RD1)
.addMBB(FBB)
.addReg(RD2)
.addMBB(TBB);
[mips][msa] Added bnz.df, bnz.v, bz.df, and bz.v These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes, are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as a branch/mov sequence to evaluate to 0 or 1. Note: The resulting code is sub-optimal since it doesnt seem to be possible to feed the result of an intrinsic directly into a brcond. At the moment it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily evaluates the boolean twice. llvm-svn: 189478
2013-08-28 20:14:50 +08:00
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MI.eraseFromParent(); // The pseudo instruction is gone now.
[mips][msa] Added bnz.df, bnz.v, bz.df, and bz.v These intrinsics are legalized to V(ALL|ANY)_(NON)?ZERO nodes, are matched as SN?Z_[BHWDV]_PSEUDO pseudo's, and emitted as a branch/mov sequence to evaluate to 0 or 1. Note: The resulting code is sub-optimal since it doesnt seem to be possible to feed the result of an intrinsic directly into a brcond. At the moment it uses (SETCC (VALL_ZERO $ws), 0, SETEQ) and similar which unnecessarily evaluates the boolean twice. llvm-svn: 189478
2013-08-28 20:14:50 +08:00
return Sink;
}
[mips][msa] Implemented extract_vector_elt for v4f32 or v2f64 For v4f32 and v2f64, EXTRACT_VECTOR_ELT is matched by a pseudo-insn which may be expanded to subregister copies and/or instructions as appropriate. llvm-svn: 191514
2013-09-27 20:17:32 +08:00
// Emit the COPY_FW pseudo instruction.
//
// copy_fw_pseudo $fd, $ws, n
// =>
// copy_u_w $rt, $ws, $n
// mtc1 $rt, $fd
//
// When n is zero, the equivalent operation can be performed with (potentially)
// zero instructions due to register overlaps. This optimization is never valid
// for lane 1 because it would require FR=0 mode which isn't supported by MSA.
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MachineBasicBlock *
MipsSETargetLowering::emitCOPY_FW(MachineInstr &MI,
MachineBasicBlock *BB) const {
Remove most of the TargetMachine::getSubtarget/getSubtargetImpl calls that don't take a Function argument from Mips. Notable exceptions: the AsmPrinter and MipsTargetObjectFile. The latter needs to be fixed, and the former will be fixed when the general AsmPrinter changes happen. llvm-svn: 227512
2015-01-30 07:27:36 +08:00
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
[mips][msa] Implemented extract_vector_elt for v4f32 or v2f64 For v4f32 and v2f64, EXTRACT_VECTOR_ELT is matched by a pseudo-insn which may be expanded to subregister copies and/or instructions as appropriate. llvm-svn: 191514
2013-09-27 20:17:32 +08:00
MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
DebugLoc DL = MI.getDebugLoc();
unsigned Fd = MI.getOperand(0).getReg();
unsigned Ws = MI.getOperand(1).getReg();
unsigned Lane = MI.getOperand(2).getImm();
[mips][msa] Implemented extract_vector_elt for v4f32 or v2f64 For v4f32 and v2f64, EXTRACT_VECTOR_ELT is matched by a pseudo-insn which may be expanded to subregister copies and/or instructions as appropriate. llvm-svn: 191514
2013-09-27 20:17:32 +08:00
[mips] Honour -mno-odd-spreg for vector insert/extract when MSA is enabled. Summary: -mno-odd-spreg prohibits the use of odd-numbered single-precision floating point registers. However, vector insert/extract was still using them when manipulating the subregisters of an MSA register. Fixed this by ensuring that insertion/extraction is only performed on even-numbered vector registers when -mno-odd-spreg is given. Reviewers: vmedic, sstankovic Reviewed By: sstankovic Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D7672 llvm-svn: 230235
2015-02-24 01:22:16 +08:00
if (Lane == 0) {
unsigned Wt = Ws;
if (!Subtarget.useOddSPReg()) {
// We must copy to an even-numbered MSA register so that the
// single-precision sub-register is also guaranteed to be even-numbered.
Wt = RegInfo.createVirtualRegister(&Mips::MSA128WEvensRegClass);
BuildMI(*BB, MI, DL, TII->get(Mips::COPY), Wt).addReg(Ws);
}
BuildMI(*BB, MI, DL, TII->get(Mips::COPY), Fd).addReg(Wt, 0, Mips::sub_lo);
} else {
unsigned Wt = RegInfo.createVirtualRegister(
Subtarget.useOddSPReg() ? &Mips::MSA128WRegClass :
&Mips::MSA128WEvensRegClass);
[mips][msa] Implemented extract_vector_elt for v4f32 or v2f64 For v4f32 and v2f64, EXTRACT_VECTOR_ELT is matched by a pseudo-insn which may be expanded to subregister copies and/or instructions as appropriate. llvm-svn: 191514
2013-09-27 20:17:32 +08:00
[mips][msa] Correct the behaviour of the COPY_FW pseudo on lanes 2 and 3. Summary: Previously, attempting to extract lanes 2 and 3 would actually extract lane 1. The MSA CodeGen tests only covered lanes 0 and 1. Differential Revision: http://llvm-reviews.chandlerc.com/D2935 llvm-svn: 202848
2014-03-04 21:54:30 +08:00
BuildMI(*BB, MI, DL, TII->get(Mips::SPLATI_W), Wt).addReg(Ws).addImm(Lane);
[mips][msa] Implemented extract_vector_elt for v4f32 or v2f64 For v4f32 and v2f64, EXTRACT_VECTOR_ELT is matched by a pseudo-insn which may be expanded to subregister copies and/or instructions as appropriate. llvm-svn: 191514
2013-09-27 20:17:32 +08:00
BuildMI(*BB, MI, DL, TII->get(Mips::COPY), Fd).addReg(Wt, 0, Mips::sub_lo);
}
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MI.eraseFromParent(); // The pseudo instruction is gone now.
[mips][msa] Implemented extract_vector_elt for v4f32 or v2f64 For v4f32 and v2f64, EXTRACT_VECTOR_ELT is matched by a pseudo-insn which may be expanded to subregister copies and/or instructions as appropriate. llvm-svn: 191514
2013-09-27 20:17:32 +08:00
return BB;
}
// Emit the COPY_FD pseudo instruction.
//
// copy_fd_pseudo $fd, $ws, n
// =>
// splati.d $wt, $ws, $n
// copy $fd, $wt:sub_64
//
// When n is zero, the equivalent operation can be performed with (potentially)
// zero instructions due to register overlaps. This optimization is always
// valid because FR=1 mode which is the only supported mode in MSA.
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MachineBasicBlock *
MipsSETargetLowering::emitCOPY_FD(MachineInstr &MI,
MachineBasicBlock *BB) const {
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
assert(Subtarget.isFP64bit());
[mips][msa] Implemented extract_vector_elt for v4f32 or v2f64 For v4f32 and v2f64, EXTRACT_VECTOR_ELT is matched by a pseudo-insn which may be expanded to subregister copies and/or instructions as appropriate. llvm-svn: 191514
2013-09-27 20:17:32 +08:00
Remove most of the TargetMachine::getSubtarget/getSubtargetImpl calls that don't take a Function argument from Mips. Notable exceptions: the AsmPrinter and MipsTargetObjectFile. The latter needs to be fixed, and the former will be fixed when the general AsmPrinter changes happen. llvm-svn: 227512
2015-01-30 07:27:36 +08:00
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
[mips][msa] Implemented extract_vector_elt for v4f32 or v2f64 For v4f32 and v2f64, EXTRACT_VECTOR_ELT is matched by a pseudo-insn which may be expanded to subregister copies and/or instructions as appropriate. llvm-svn: 191514
2013-09-27 20:17:32 +08:00
MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
unsigned Fd = MI.getOperand(0).getReg();
unsigned Ws = MI.getOperand(1).getReg();
unsigned Lane = MI.getOperand(2).getImm() * 2;
DebugLoc DL = MI.getDebugLoc();
[mips][msa] Implemented extract_vector_elt for v4f32 or v2f64 For v4f32 and v2f64, EXTRACT_VECTOR_ELT is matched by a pseudo-insn which may be expanded to subregister copies and/or instructions as appropriate. llvm-svn: 191514
2013-09-27 20:17:32 +08:00
if (Lane == 0)
BuildMI(*BB, MI, DL, TII->get(Mips::COPY), Fd).addReg(Ws, 0, Mips::sub_64);
else {
unsigned Wt = RegInfo.createVirtualRegister(&Mips::MSA128DRegClass);
BuildMI(*BB, MI, DL, TII->get(Mips::SPLATI_D), Wt).addReg(Ws).addImm(1);
BuildMI(*BB, MI, DL, TII->get(Mips::COPY), Fd).addReg(Wt, 0, Mips::sub_64);
}
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MI.eraseFromParent(); // The pseudo instruction is gone now.
[mips][msa] Implemented extract_vector_elt for v4f32 or v2f64 For v4f32 and v2f64, EXTRACT_VECTOR_ELT is matched by a pseudo-insn which may be expanded to subregister copies and/or instructions as appropriate. llvm-svn: 191514
2013-09-27 20:17:32 +08:00
return BB;
}
[mips][msa] Implemented insert_vector_elt for v4f32 and v2f64. For v4f32 and v2f64, INSERT_VECTOR_ELT is matched by a pseudo-insn which is later expanded to appropriate insve.[wd] insns. llvm-svn: 191515
2013-09-27 20:31:32 +08:00
// Emit the INSERT_FW pseudo instruction.
//
// insert_fw_pseudo $wd, $wd_in, $n, $fs
// =>
// subreg_to_reg $wt:sub_lo, $fs
// insve_w $wd[$n], $wd_in, $wt[0]
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
MachineBasicBlock *
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MipsSETargetLowering::emitINSERT_FW(MachineInstr &MI,
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
MachineBasicBlock *BB) const {
Remove most of the TargetMachine::getSubtarget/getSubtargetImpl calls that don't take a Function argument from Mips. Notable exceptions: the AsmPrinter and MipsTargetObjectFile. The latter needs to be fixed, and the former will be fixed when the general AsmPrinter changes happen. llvm-svn: 227512
2015-01-30 07:27:36 +08:00
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
[mips][msa] Implemented insert_vector_elt for v4f32 and v2f64. For v4f32 and v2f64, INSERT_VECTOR_ELT is matched by a pseudo-insn which is later expanded to appropriate insve.[wd] insns. llvm-svn: 191515
2013-09-27 20:31:32 +08:00
MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
DebugLoc DL = MI.getDebugLoc();
unsigned Wd = MI.getOperand(0).getReg();
unsigned Wd_in = MI.getOperand(1).getReg();
unsigned Lane = MI.getOperand(2).getImm();
unsigned Fs = MI.getOperand(3).getReg();
[mips] Honour -mno-odd-spreg for vector insert/extract when MSA is enabled. Summary: -mno-odd-spreg prohibits the use of odd-numbered single-precision floating point registers. However, vector insert/extract was still using them when manipulating the subregisters of an MSA register. Fixed this by ensuring that insertion/extraction is only performed on even-numbered vector registers when -mno-odd-spreg is given. Reviewers: vmedic, sstankovic Reviewed By: sstankovic Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D7672 llvm-svn: 230235
2015-02-24 01:22:16 +08:00
unsigned Wt = RegInfo.createVirtualRegister(
Subtarget.useOddSPReg() ? &Mips::MSA128WRegClass :
&Mips::MSA128WEvensRegClass);
[mips][msa] Implemented insert_vector_elt for v4f32 and v2f64. For v4f32 and v2f64, INSERT_VECTOR_ELT is matched by a pseudo-insn which is later expanded to appropriate insve.[wd] insns. llvm-svn: 191515
2013-09-27 20:31:32 +08:00
BuildMI(*BB, MI, DL, TII->get(Mips::SUBREG_TO_REG), Wt)
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
.addImm(0)
.addReg(Fs)
.addImm(Mips::sub_lo);
[mips][msa] Implemented insert_vector_elt for v4f32 and v2f64. For v4f32 and v2f64, INSERT_VECTOR_ELT is matched by a pseudo-insn which is later expanded to appropriate insve.[wd] insns. llvm-svn: 191515
2013-09-27 20:31:32 +08:00
BuildMI(*BB, MI, DL, TII->get(Mips::INSVE_W), Wd)
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
.addReg(Wd_in)
.addImm(Lane)
[mips] Rewrite MipsAsmParser and MipsOperand. Summary: Highlights: - Registers are resolved much later (by the render method). Prior to that point, GPR32's/GPR64's are GPR's regardless of register size. Similarly FGR32's/FGR64's/AFGR64's are FGR's regardless of register size or FR mode. Numeric registers can be anything. - All registers are parsed the same way everywhere (even when handling symbol aliasing) - One consequence is that all registers can be specified numerically almost anywhere (e.g. $fccX, $wX). The exception is symbol aliasing but that can be easily resolved. - Removes the need for the hasConsumedDollar hack - Parenthesis and Bracket suffixes are handled generically - Micromips instructions are parsed directly instead of going through the standard encodings first. - rdhwr accepts all 32 registers, and the following instructions that previously xfailed now work: ddiv, ddivu, div, divu, cvt.l.[ds], se[bh], wsbh, floor.w.[ds], c.ngl.d, c.sf.s, dsbh, dshd, madd.s, msub.s, nmadd.s, nmsub.s, swxc1 - Diagnostics involving registers point at the correct character (the $) - There's only one kind of immediate in MipsOperand. LSA immediates are handled by the predicate and renderer. Lowlights: - Hardcoded '$zero' in the div patterns is handled with a hack. MipsOperand::isReg() will return true for a k_RegisterIndex token with Index == 0 and getReg() will return ZERO for this case. Note that it doesn't return ZERO_64 on isGP64() targets. - I haven't cleaned up all of the now-unused functions. Some more of the generic parser could be removed too (integers and relocs for example). - insve.df needed a custom decoder to handle the implicit fourth operand that was needed to make it parse correctly. The difficulty was that the matcher expected a Token<'0'> but gets an Imm<0>. Adding an implicit zero solved this. Reviewers: matheusalmeida, vmedic Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D3222 llvm-svn: 205292
2014-04-01 18:35:28 +08:00
.addReg(Wt)
.addImm(0);
[mips][msa] Implemented insert_vector_elt for v4f32 and v2f64. For v4f32 and v2f64, INSERT_VECTOR_ELT is matched by a pseudo-insn which is later expanded to appropriate insve.[wd] insns. llvm-svn: 191515
2013-09-27 20:31:32 +08:00
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MI.eraseFromParent(); // The pseudo instruction is gone now.
[mips][msa] Implemented insert_vector_elt for v4f32 and v2f64. For v4f32 and v2f64, INSERT_VECTOR_ELT is matched by a pseudo-insn which is later expanded to appropriate insve.[wd] insns. llvm-svn: 191515
2013-09-27 20:31:32 +08:00
return BB;
}
// Emit the INSERT_FD pseudo instruction.
//
// insert_fd_pseudo $wd, $fs, n
// =>
// subreg_to_reg $wt:sub_64, $fs
// insve_d $wd[$n], $wd_in, $wt[0]
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
MachineBasicBlock *
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MipsSETargetLowering::emitINSERT_FD(MachineInstr &MI,
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
MachineBasicBlock *BB) const {
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
assert(Subtarget.isFP64bit());
[mips][msa] Implemented insert_vector_elt for v4f32 and v2f64. For v4f32 and v2f64, INSERT_VECTOR_ELT is matched by a pseudo-insn which is later expanded to appropriate insve.[wd] insns. llvm-svn: 191515
2013-09-27 20:31:32 +08:00
Remove most of the TargetMachine::getSubtarget/getSubtargetImpl calls that don't take a Function argument from Mips. Notable exceptions: the AsmPrinter and MipsTargetObjectFile. The latter needs to be fixed, and the former will be fixed when the general AsmPrinter changes happen. llvm-svn: 227512
2015-01-30 07:27:36 +08:00
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
[mips][msa] Implemented insert_vector_elt for v4f32 and v2f64. For v4f32 and v2f64, INSERT_VECTOR_ELT is matched by a pseudo-insn which is later expanded to appropriate insve.[wd] insns. llvm-svn: 191515
2013-09-27 20:31:32 +08:00
MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
DebugLoc DL = MI.getDebugLoc();
unsigned Wd = MI.getOperand(0).getReg();
unsigned Wd_in = MI.getOperand(1).getReg();
unsigned Lane = MI.getOperand(2).getImm();
unsigned Fs = MI.getOperand(3).getReg();
[mips][msa] Implemented insert_vector_elt for v4f32 and v2f64. For v4f32 and v2f64, INSERT_VECTOR_ELT is matched by a pseudo-insn which is later expanded to appropriate insve.[wd] insns. llvm-svn: 191515
2013-09-27 20:31:32 +08:00
unsigned Wt = RegInfo.createVirtualRegister(&Mips::MSA128DRegClass);
BuildMI(*BB, MI, DL, TII->get(Mips::SUBREG_TO_REG), Wt)
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
.addImm(0)
.addReg(Fs)
.addImm(Mips::sub_64);
[mips][msa] Implemented insert_vector_elt for v4f32 and v2f64. For v4f32 and v2f64, INSERT_VECTOR_ELT is matched by a pseudo-insn which is later expanded to appropriate insve.[wd] insns. llvm-svn: 191515
2013-09-27 20:31:32 +08:00
BuildMI(*BB, MI, DL, TII->get(Mips::INSVE_D), Wd)
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
.addReg(Wd_in)
.addImm(Lane)
[mips] Rewrite MipsAsmParser and MipsOperand. Summary: Highlights: - Registers are resolved much later (by the render method). Prior to that point, GPR32's/GPR64's are GPR's regardless of register size. Similarly FGR32's/FGR64's/AFGR64's are FGR's regardless of register size or FR mode. Numeric registers can be anything. - All registers are parsed the same way everywhere (even when handling symbol aliasing) - One consequence is that all registers can be specified numerically almost anywhere (e.g. $fccX, $wX). The exception is symbol aliasing but that can be easily resolved. - Removes the need for the hasConsumedDollar hack - Parenthesis and Bracket suffixes are handled generically - Micromips instructions are parsed directly instead of going through the standard encodings first. - rdhwr accepts all 32 registers, and the following instructions that previously xfailed now work: ddiv, ddivu, div, divu, cvt.l.[ds], se[bh], wsbh, floor.w.[ds], c.ngl.d, c.sf.s, dsbh, dshd, madd.s, msub.s, nmadd.s, nmsub.s, swxc1 - Diagnostics involving registers point at the correct character (the $) - There's only one kind of immediate in MipsOperand. LSA immediates are handled by the predicate and renderer. Lowlights: - Hardcoded '$zero' in the div patterns is handled with a hack. MipsOperand::isReg() will return true for a k_RegisterIndex token with Index == 0 and getReg() will return ZERO for this case. Note that it doesn't return ZERO_64 on isGP64() targets. - I haven't cleaned up all of the now-unused functions. Some more of the generic parser could be removed too (integers and relocs for example). - insve.df needed a custom decoder to handle the implicit fourth operand that was needed to make it parse correctly. The difficulty was that the matcher expected a Token<'0'> but gets an Imm<0>. Adding an implicit zero solved this. Reviewers: matheusalmeida, vmedic Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D3222 llvm-svn: 205292
2014-04-01 18:35:28 +08:00
.addReg(Wt)
.addImm(0);
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MI.eraseFromParent(); // The pseudo instruction is gone now.
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
return BB;
}
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
// Emit the INSERT_([BHWD]|F[WD])_VIDX pseudo instruction.
//
// For integer:
// (INSERT_([BHWD]|F[WD])_PSEUDO $wd, $wd_in, $n, $rs)
// =>
// (SLL $lanetmp1, $lane, <log2size)
// (SLD_B $wdtmp1, $wd_in, $wd_in, $lanetmp1)
// (INSERT_[BHWD], $wdtmp2, $wdtmp1, 0, $rs)
// (NEG $lanetmp2, $lanetmp1)
// (SLD_B $wd, $wdtmp2, $wdtmp2, $lanetmp2)
//
// For floating point:
// (INSERT_([BHWD]|F[WD])_PSEUDO $wd, $wd_in, $n, $fs)
// =>
// (SUBREG_TO_REG $wt, $fs, <subreg>)
// (SLL $lanetmp1, $lane, <log2size)
// (SLD_B $wdtmp1, $wd_in, $wd_in, $lanetmp1)
// (INSVE_[WD], $wdtmp2, 0, $wdtmp1, 0)
// (NEG $lanetmp2, $lanetmp1)
// (SLD_B $wd, $wdtmp2, $wdtmp2, $lanetmp2)
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MachineBasicBlock *MipsSETargetLowering::emitINSERT_DF_VIDX(
MachineInstr &MI, MachineBasicBlock *BB, unsigned EltSizeInBytes,
bool IsFP) const {
Remove most of the TargetMachine::getSubtarget/getSubtargetImpl calls that don't take a Function argument from Mips. Notable exceptions: the AsmPrinter and MipsTargetObjectFile. The latter needs to be fixed, and the former will be fixed when the general AsmPrinter changes happen. llvm-svn: 227512
2015-01-30 07:27:36 +08:00
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
DebugLoc DL = MI.getDebugLoc();
unsigned Wd = MI.getOperand(0).getReg();
unsigned SrcVecReg = MI.getOperand(1).getReg();
unsigned LaneReg = MI.getOperand(2).getReg();
unsigned SrcValReg = MI.getOperand(3).getReg();
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
const TargetRegisterClass *VecRC = nullptr;
[mips][msa] Fix register/register-class mismatches in emitINSERT_DF_VIDX(). Reviewers: sdardis Subscribers: dsanders, sdardis, llvm-commits Differential Revision: http://reviews.llvm.org/D21068 llvm-svn: 272765
2016-06-15 16:43:23 +08:00
// FIXME: This should be true for N32 too.
So that we can include target lowering in the subtarget, remove include circular dependency with the subtarget by inlining accessor methods and outlining a routine. llvm-svn: 212234
2014-07-03 07:18:40 +08:00
const TargetRegisterClass *GPRRC =
[mips][msa] Test basic operations for the N32 ABI too. Summary: This required adding instruction aliases for dneg. N64 will be enabled shortly but requires additional bugfixes. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9341 llvm-svn: 236489
2015-05-05 16:48:35 +08:00
Subtarget.isABI_N64() ? &Mips::GPR64RegClass : &Mips::GPR32RegClass;
[mips][msa] Fix register/register-class mismatches in emitINSERT_DF_VIDX(). Reviewers: sdardis Subscribers: dsanders, sdardis, llvm-commits Differential Revision: http://reviews.llvm.org/D21068 llvm-svn: 272765
2016-06-15 16:43:23 +08:00
unsigned SubRegIdx = Subtarget.isABI_N64() ? Mips::sub_32 : 0;
unsigned ShiftOp = Subtarget.isABI_N64() ? Mips::DSLL : Mips::SLL;
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
unsigned EltLog2Size;
unsigned InsertOp = 0;
unsigned InsveOp = 0;
switch (EltSizeInBytes) {
default:
llvm_unreachable("Unexpected size");
case 1:
EltLog2Size = 0;
InsertOp = Mips::INSERT_B;
InsveOp = Mips::INSVE_B;
VecRC = &Mips::MSA128BRegClass;
break;
case 2:
EltLog2Size = 1;
InsertOp = Mips::INSERT_H;
InsveOp = Mips::INSVE_H;
VecRC = &Mips::MSA128HRegClass;
break;
case 4:
EltLog2Size = 2;
InsertOp = Mips::INSERT_W;
InsveOp = Mips::INSVE_W;
VecRC = &Mips::MSA128WRegClass;
break;
case 8:
EltLog2Size = 3;
InsertOp = Mips::INSERT_D;
InsveOp = Mips::INSVE_D;
VecRC = &Mips::MSA128DRegClass;
break;
}
if (IsFP) {
unsigned Wt = RegInfo.createVirtualRegister(VecRC);
BuildMI(*BB, MI, DL, TII->get(Mips::SUBREG_TO_REG), Wt)
.addImm(0)
.addReg(SrcValReg)
.addImm(EltSizeInBytes == 8 ? Mips::sub_64 : Mips::sub_lo);
SrcValReg = Wt;
}
// Convert the lane index into a byte index
if (EltSizeInBytes != 1) {
unsigned LaneTmp1 = RegInfo.createVirtualRegister(GPRRC);
[mips][msa] Fix register/register-class mismatches in emitINSERT_DF_VIDX(). Reviewers: sdardis Subscribers: dsanders, sdardis, llvm-commits Differential Revision: http://reviews.llvm.org/D21068 llvm-svn: 272765
2016-06-15 16:43:23 +08:00
BuildMI(*BB, MI, DL, TII->get(ShiftOp), LaneTmp1)
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
.addReg(LaneReg)
.addImm(EltLog2Size);
LaneReg = LaneTmp1;
}
// Rotate bytes around so that the desired lane is element zero
unsigned WdTmp1 = RegInfo.createVirtualRegister(VecRC);
BuildMI(*BB, MI, DL, TII->get(Mips::SLD_B), WdTmp1)
.addReg(SrcVecReg)
.addReg(SrcVecReg)
[mips][msa] Fix register/register-class mismatches in emitINSERT_DF_VIDX(). Reviewers: sdardis Subscribers: dsanders, sdardis, llvm-commits Differential Revision: http://reviews.llvm.org/D21068 llvm-svn: 272765
2016-06-15 16:43:23 +08:00
.addReg(LaneReg, 0, SubRegIdx);
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
unsigned WdTmp2 = RegInfo.createVirtualRegister(VecRC);
if (IsFP) {
// Use insve.df to insert to element zero
BuildMI(*BB, MI, DL, TII->get(InsveOp), WdTmp2)
.addReg(WdTmp1)
.addImm(0)
.addReg(SrcValReg)
.addImm(0);
} else {
// Use insert.df to insert to element zero
BuildMI(*BB, MI, DL, TII->get(InsertOp), WdTmp2)
.addReg(WdTmp1)
.addReg(SrcValReg)
.addImm(0);
}
// Rotate elements the rest of the way for a full rotation.
// sld.df inteprets $rt modulo the number of columns so we only need to negate
// the lane index to do this.
unsigned LaneTmp2 = RegInfo.createVirtualRegister(GPRRC);
[mips][msa] Test basic operations for the N32 ABI too. Summary: This required adding instruction aliases for dneg. N64 will be enabled shortly but requires additional bugfixes. Reviewers: vkalintiris Reviewed By: vkalintiris Subscribers: llvm-commits Differential Revision: http://reviews.llvm.org/D9341 llvm-svn: 236489
2015-05-05 16:48:35 +08:00
BuildMI(*BB, MI, DL, TII->get(Subtarget.isABI_N64() ? Mips::DSUB : Mips::SUB),
LaneTmp2)
.addReg(Subtarget.isABI_N64() ? Mips::ZERO_64 : Mips::ZERO)
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
.addReg(LaneReg);
BuildMI(*BB, MI, DL, TII->get(Mips::SLD_B), Wd)
.addReg(WdTmp2)
.addReg(WdTmp2)
[mips][msa] Fix register/register-class mismatches in emitINSERT_DF_VIDX(). Reviewers: sdardis Subscribers: dsanders, sdardis, llvm-commits Differential Revision: http://reviews.llvm.org/D21068 llvm-svn: 272765
2016-06-15 16:43:23 +08:00
.addReg(LaneTmp2, 0, SubRegIdx);
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MI.eraseFromParent(); // The pseudo instruction is gone now.
[mips][msa] Fix vector insertions where the index is variable Summary: This isn't supported directly so we rotate the vector by the desired number of elements, insert to element zero, then rotate back. The i64 case generates rather poor code on MIPS32. There is an obvious optimisation to be made in future (do both insert.w's inside a shared rotate/unrotate sequence) but for now it's sufficient to select valid code instead of aborting. Depends on D3536 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://reviews.llvm.org/D3537 llvm-svn: 207640
2014-04-30 20:09:32 +08:00
return BB;
}
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
// Emit the FILL_FW pseudo instruction.
//
// fill_fw_pseudo $wd, $fs
// =>
// implicit_def $wt1
// insert_subreg $wt2:subreg_lo, $wt1, $fs
// splati.w $wd, $wt2[0]
MachineBasicBlock *
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MipsSETargetLowering::emitFILL_FW(MachineInstr &MI,
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
MachineBasicBlock *BB) const {
Remove most of the TargetMachine::getSubtarget/getSubtargetImpl calls that don't take a Function argument from Mips. Notable exceptions: the AsmPrinter and MipsTargetObjectFile. The latter needs to be fixed, and the former will be fixed when the general AsmPrinter changes happen. llvm-svn: 227512
2015-01-30 07:27:36 +08:00
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
DebugLoc DL = MI.getDebugLoc();
unsigned Wd = MI.getOperand(0).getReg();
unsigned Fs = MI.getOperand(1).getReg();
[mips] Honour -mno-odd-spreg for vector splat (again) Previous the lowering of FILL_FW would use the MSA128W register class when performing a vector splat. Instead it should be honouring -mno-odd-spreg and only use the even registers when performing a splat from word to vector register. Logical follow-on from r230235. This fixes PR/31369. A previous commit was missing the test case and had another differential in it. Reviewers: slthakur Differential Revision: https://reviews.llvm.org/D28373 llvm-svn: 291566
2017-01-10 23:53:10 +08:00
unsigned Wt1 = RegInfo.createVirtualRegister(
Subtarget.useOddSPReg() ? &Mips::MSA128WRegClass
: &Mips::MSA128WEvensRegClass);
unsigned Wt2 = RegInfo.createVirtualRegister(
Subtarget.useOddSPReg() ? &Mips::MSA128WRegClass
: &Mips::MSA128WEvensRegClass);
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
BuildMI(*BB, MI, DL, TII->get(Mips::IMPLICIT_DEF), Wt1);
BuildMI(*BB, MI, DL, TII->get(Mips::INSERT_SUBREG), Wt2)
.addReg(Wt1)
.addReg(Fs)
.addImm(Mips::sub_lo);
BuildMI(*BB, MI, DL, TII->get(Mips::SPLATI_W), Wd).addReg(Wt2).addImm(0);
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MI.eraseFromParent(); // The pseudo instruction is gone now.
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
return BB;
}
// Emit the FILL_FD pseudo instruction.
//
// fill_fd_pseudo $wd, $fs
// =>
// implicit_def $wt1
// insert_subreg $wt2:subreg_64, $wt1, $fs
// splati.d $wd, $wt2[0]
MachineBasicBlock *
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MipsSETargetLowering::emitFILL_FD(MachineInstr &MI,
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
MachineBasicBlock *BB) const {
The subtarget in MipsTargetLowering isn't going to change and so doesn't need to be a pointer, but a reference. llvm-svn: 213422
2014-07-19 06:55:25 +08:00
assert(Subtarget.isFP64bit());
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
Remove most of the TargetMachine::getSubtarget/getSubtargetImpl calls that don't take a Function argument from Mips. Notable exceptions: the AsmPrinter and MipsTargetObjectFile. The latter needs to be fixed, and the former will be fixed when the general AsmPrinter changes happen. llvm-svn: 227512
2015-01-30 07:27:36 +08:00
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
DebugLoc DL = MI.getDebugLoc();
unsigned Wd = MI.getOperand(0).getReg();
unsigned Fs = MI.getOperand(1).getReg();
[mips][msa] Added support for build_vector for v4f32 and v2f64. llvm-svn: 192699
2013-10-15 21:14:41 +08:00
unsigned Wt1 = RegInfo.createVirtualRegister(&Mips::MSA128DRegClass);
unsigned Wt2 = RegInfo.createVirtualRegister(&Mips::MSA128DRegClass);
BuildMI(*BB, MI, DL, TII->get(Mips::IMPLICIT_DEF), Wt1);
BuildMI(*BB, MI, DL, TII->get(Mips::INSERT_SUBREG), Wt2)
.addReg(Wt1)
.addReg(Fs)
.addImm(Mips::sub_64);
BuildMI(*BB, MI, DL, TII->get(Mips::SPLATI_D), Wd).addReg(Wt2).addImm(0);
[mips][msa] Implemented insert_vector_elt for v4f32 and v2f64. For v4f32 and v2f64, INSERT_VECTOR_ELT is matched by a pseudo-insn which is later expanded to appropriate insve.[wd] insns. llvm-svn: 191515
2013-09-27 20:31:32 +08:00
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MI.eraseFromParent(); // The pseudo instruction is gone now.
[mips][msa] Implemented insert_vector_elt for v4f32 and v2f64. For v4f32 and v2f64, INSERT_VECTOR_ELT is matched by a pseudo-insn which is later expanded to appropriate insve.[wd] insns. llvm-svn: 191515
2013-09-27 20:31:32 +08:00
return BB;
}
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
[mips][msa] Implement f16 support The MIPS MSA ASE provides instructions to convert to and from half precision floating point. This patch teaches the MIPS backend to treat f16 as a legal type and how to promote such values to f32 for the usual set of operations. As a result of this, the fexup[lr].w intrinsics no longer crash LLVM during type legalization. Reviewers: zoran.jovanvoic, vkalintiris Differential Revision: https://reviews.llvm.org/D26398 llvm-svn: 287349
2016-11-19 00:17:44 +08:00
// Emit the ST_F16_PSEDUO instruction to store a f16 value from an MSA
// register.
//
// STF16 MSA128F16:$wd, mem_simm10:$addr
// =>
// copy_u.h $rtemp,$wd[0]
// sh $rtemp, $addr
//
// Safety: We can't use st.h & co as they would over write the memory after
// the destination. It would require half floats be allocated 16 bytes(!) of
// space.
MachineBasicBlock *
MipsSETargetLowering::emitST_F16_PSEUDO(MachineInstr &MI,
MachineBasicBlock *BB) const {
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
DebugLoc DL = MI.getDebugLoc();
unsigned Ws = MI.getOperand(0).getReg();
unsigned Rt = MI.getOperand(1).getReg();
const MachineMemOperand &MMO = **MI.memoperands_begin();
unsigned Imm = MMO.getOffset();
// Caution: A load via the GOT can expand to a GPR32 operand, a load via
// spill and reload can expand as a GPR64 operand. Examine the
// operand in detail and default to ABI.
const TargetRegisterClass *RC =
MI.getOperand(1).isReg() ? RegInfo.getRegClass(MI.getOperand(1).getReg())
: (Subtarget.isABI_O32() ? &Mips::GPR32RegClass
: &Mips::GPR64RegClass);
const bool UsingMips32 = RC == &Mips::GPR32RegClass;
[mips] Alter register classes for MSA pseudo f16 instructions This change introduces additional machine instructions in functions dealing with the expansion of msa pseudo f16 instructions due to register classes being inappropriate when checked with machine verifier. Differential Revision: https://reviews.llvm.org/D34276 llvm-svn: 308301
2017-07-18 20:05:35 +08:00
unsigned Rs = RegInfo.createVirtualRegister(&Mips::GPR32RegClass);
[mips][msa] Implement f16 support The MIPS MSA ASE provides instructions to convert to and from half precision floating point. This patch teaches the MIPS backend to treat f16 as a legal type and how to promote such values to f32 for the usual set of operations. As a result of this, the fexup[lr].w intrinsics no longer crash LLVM during type legalization. Reviewers: zoran.jovanvoic, vkalintiris Differential Revision: https://reviews.llvm.org/D26398 llvm-svn: 287349
2016-11-19 00:17:44 +08:00
BuildMI(*BB, MI, DL, TII->get(Mips::COPY_U_H), Rs).addReg(Ws).addImm(0);
[mips] Alter register classes for MSA pseudo f16 instructions This change introduces additional machine instructions in functions dealing with the expansion of msa pseudo f16 instructions due to register classes being inappropriate when checked with machine verifier. Differential Revision: https://reviews.llvm.org/D34276 llvm-svn: 308301
2017-07-18 20:05:35 +08:00
if(!UsingMips32) {
unsigned Tmp = RegInfo.createVirtualRegister(&Mips::GPR64RegClass);
BuildMI(*BB, MI, DL, TII->get(Mips::SUBREG_TO_REG), Tmp)
.addImm(0)
.addReg(Rs)
.addImm(Mips::sub_32);
Rs = Tmp;
}
[mips][msa] Implement f16 support The MIPS MSA ASE provides instructions to convert to and from half precision floating point. This patch teaches the MIPS backend to treat f16 as a legal type and how to promote such values to f32 for the usual set of operations. As a result of this, the fexup[lr].w intrinsics no longer crash LLVM during type legalization. Reviewers: zoran.jovanvoic, vkalintiris Differential Revision: https://reviews.llvm.org/D26398 llvm-svn: 287349
2016-11-19 00:17:44 +08:00
BuildMI(*BB, MI, DL, TII->get(UsingMips32 ? Mips::SH : Mips::SH64))
.addReg(Rs)
.addReg(Rt)
.addImm(Imm)
.addMemOperand(BB->getParent()->getMachineMemOperand(
&MMO, MMO.getOffset(), MMO.getSize()));
MI.eraseFromParent();
return BB;
}
// Emit the LD_F16_PSEDUO instruction to load a f16 value into an MSA register.
//
// LD_F16 MSA128F16:$wd, mem_simm10:$addr
// =>
// lh $rtemp, $addr
// fill.h $wd, $rtemp
//
// Safety: We can't use ld.h & co as they over-read from the source.
// Additionally, if the address is not modulo 16, 2 cases can occur:
// a) Segmentation fault as the load instruction reads from a memory page
// memory it's not supposed to.
// b) The load crosses an implementation specific boundary, requiring OS
// intervention.
MachineBasicBlock *
MipsSETargetLowering::emitLD_F16_PSEUDO(MachineInstr &MI,
MachineBasicBlock *BB) const {
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
DebugLoc DL = MI.getDebugLoc();
unsigned Wd = MI.getOperand(0).getReg();
// Caution: A load via the GOT can expand to a GPR32 operand, a load via
// spill and reload can expand as a GPR64 operand. Examine the
// operand in detail and default to ABI.
const TargetRegisterClass *RC =
MI.getOperand(1).isReg() ? RegInfo.getRegClass(MI.getOperand(1).getReg())
: (Subtarget.isABI_O32() ? &Mips::GPR32RegClass
: &Mips::GPR64RegClass);
const bool UsingMips32 = RC == &Mips::GPR32RegClass;
unsigned Rt = RegInfo.createVirtualRegister(RC);
MachineInstrBuilder MIB =
BuildMI(*BB, MI, DL, TII->get(UsingMips32 ? Mips::LH : Mips::LH64), Rt);
for (unsigned i = 1; i < MI.getNumOperands(); i++)
[CodeGen] Rename MachineInstrBuilder::addOperand. NFC Rename from addOperand to just add, to match the other method that has been added to MachineInstrBuilder for adding more than just 1 operand. See https://reviews.llvm.org/D28057 for the whole discussion. Differential Revision: https://reviews.llvm.org/D28556 llvm-svn: 291891
2017-01-13 17:58:52 +08:00
MIB.add(MI.getOperand(i));
[mips][msa] Implement f16 support The MIPS MSA ASE provides instructions to convert to and from half precision floating point. This patch teaches the MIPS backend to treat f16 as a legal type and how to promote such values to f32 for the usual set of operations. As a result of this, the fexup[lr].w intrinsics no longer crash LLVM during type legalization. Reviewers: zoran.jovanvoic, vkalintiris Differential Revision: https://reviews.llvm.org/D26398 llvm-svn: 287349
2016-11-19 00:17:44 +08:00
[mips] Alter register classes for MSA pseudo f16 instructions This change introduces additional machine instructions in functions dealing with the expansion of msa pseudo f16 instructions due to register classes being inappropriate when checked with machine verifier. Differential Revision: https://reviews.llvm.org/D34276 llvm-svn: 308301
2017-07-18 20:05:35 +08:00
if(!UsingMips32) {
unsigned Tmp = RegInfo.createVirtualRegister(&Mips::GPR32RegClass);
BuildMI(*BB, MI, DL, TII->get(Mips::COPY), Tmp).addReg(Rt, 0, Mips::sub_32);
Rt = Tmp;
}
[mips][msa] Implement f16 support The MIPS MSA ASE provides instructions to convert to and from half precision floating point. This patch teaches the MIPS backend to treat f16 as a legal type and how to promote such values to f32 for the usual set of operations. As a result of this, the fexup[lr].w intrinsics no longer crash LLVM during type legalization. Reviewers: zoran.jovanvoic, vkalintiris Differential Revision: https://reviews.llvm.org/D26398 llvm-svn: 287349
2016-11-19 00:17:44 +08:00
BuildMI(*BB, MI, DL, TII->get(Mips::FILL_H), Wd).addReg(Rt);
MI.eraseFromParent();
return BB;
}
// Emit the FPROUND_PSEUDO instruction.
//
// Round an FGR64Opnd, FGR32Opnd to an f16.
//
// Safety: Cycle the operand through the GPRs so the result always ends up
// the correct MSA register.
//
// FIXME: This copying is strictly unnecessary. If we could tie FGR32Opnd:$Fs
// / FGR64Opnd:$Fs and MSA128F16:$Wd to the same physical register
// (which they can be, as the MSA registers are defined to alias the
// FPU's 64 bit and 32 bit registers) the result can be accessed using
// the correct register class. That requires operands be tie-able across
// register classes which have a sub/super register class relationship.
//
// For FPG32Opnd:
//
// FPROUND MSA128F16:$wd, FGR32Opnd:$fs
// =>
// mfc1 $rtemp, $fs
// fill.w $rtemp, $wtemp
// fexdo.w $wd, $wtemp, $wtemp
//
// For FPG64Opnd on mips32r2+:
//
// FPROUND MSA128F16:$wd, FGR64Opnd:$fs
// =>
// mfc1 $rtemp, $fs
// fill.w $rtemp, $wtemp
// mfhc1 $rtemp2, $fs
// insert.w $wtemp[1], $rtemp2
// insert.w $wtemp[3], $rtemp2
// fexdo.w $wtemp2, $wtemp, $wtemp
// fexdo.h $wd, $temp2, $temp2
//
// For FGR64Opnd on mips64r2+:
//
// FPROUND MSA128F16:$wd, FGR64Opnd:$fs
// =>
// dmfc1 $rtemp, $fs
// fill.d $rtemp, $wtemp
// fexdo.w $wtemp2, $wtemp, $wtemp
// fexdo.h $wd, $wtemp2, $wtemp2
//
// Safety note: As $wtemp is UNDEF, we may provoke a spurious exception if the
// undef bits are "just right" and the exception enable bits are
// set. By using fill.w to replicate $fs into all elements over
// insert.w for one element, we avoid that potiential case. If
// fexdo.[hw] causes an exception in, the exception is valid and it
// occurs for all elements.
MachineBasicBlock *
MipsSETargetLowering::emitFPROUND_PSEUDO(MachineInstr &MI,
MachineBasicBlock *BB,
bool IsFGR64) const {
// Strictly speaking, we need MIPS32R5 to support MSA. We'll be generous
// here. It's technically doable to support MIPS32 here, but the ISA forbids
// it.
assert(Subtarget.hasMSA() && Subtarget.hasMips32r2());
bool IsFGR64onMips64 = Subtarget.hasMips64() && IsFGR64;
[mips] Alter register classes for MSA pseudo f16 instructions This change introduces additional machine instructions in functions dealing with the expansion of msa pseudo f16 instructions due to register classes being inappropriate when checked with machine verifier. Differential Revision: https://reviews.llvm.org/D34276 llvm-svn: 308301
2017-07-18 20:05:35 +08:00
bool IsFGR64onMips32 = !Subtarget.hasMips64() && IsFGR64;
[mips][msa] Implement f16 support The MIPS MSA ASE provides instructions to convert to and from half precision floating point. This patch teaches the MIPS backend to treat f16 as a legal type and how to promote such values to f32 for the usual set of operations. As a result of this, the fexup[lr].w intrinsics no longer crash LLVM during type legalization. Reviewers: zoran.jovanvoic, vkalintiris Differential Revision: https://reviews.llvm.org/D26398 llvm-svn: 287349
2016-11-19 00:17:44 +08:00
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
DebugLoc DL = MI.getDebugLoc();
unsigned Wd = MI.getOperand(0).getReg();
unsigned Fs = MI.getOperand(1).getReg();
MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
unsigned Wtemp = RegInfo.createVirtualRegister(&Mips::MSA128WRegClass);
const TargetRegisterClass *GPRRC =
IsFGR64onMips64 ? &Mips::GPR64RegClass : &Mips::GPR32RegClass;
[mips] Alter register classes for MSA pseudo f16 instructions This change introduces additional machine instructions in functions dealing with the expansion of msa pseudo f16 instructions due to register classes being inappropriate when checked with machine verifier. Differential Revision: https://reviews.llvm.org/D34276 llvm-svn: 308301
2017-07-18 20:05:35 +08:00
unsigned MFC1Opc = IsFGR64onMips64
? Mips::DMFC1
: (IsFGR64onMips32 ? Mips::MFC1_D64 : Mips::MFC1);
[mips][msa] Implement f16 support The MIPS MSA ASE provides instructions to convert to and from half precision floating point. This patch teaches the MIPS backend to treat f16 as a legal type and how to promote such values to f32 for the usual set of operations. As a result of this, the fexup[lr].w intrinsics no longer crash LLVM during type legalization. Reviewers: zoran.jovanvoic, vkalintiris Differential Revision: https://reviews.llvm.org/D26398 llvm-svn: 287349
2016-11-19 00:17:44 +08:00
unsigned FILLOpc = IsFGR64onMips64 ? Mips::FILL_D : Mips::FILL_W;
// Perform the register class copy as mentioned above.
unsigned Rtemp = RegInfo.createVirtualRegister(GPRRC);
BuildMI(*BB, MI, DL, TII->get(MFC1Opc), Rtemp).addReg(Fs);
BuildMI(*BB, MI, DL, TII->get(FILLOpc), Wtemp).addReg(Rtemp);
unsigned WPHI = Wtemp;
[mips] Alter register classes for MSA pseudo f16 instructions This change introduces additional machine instructions in functions dealing with the expansion of msa pseudo f16 instructions due to register classes being inappropriate when checked with machine verifier. Differential Revision: https://reviews.llvm.org/D34276 llvm-svn: 308301
2017-07-18 20:05:35 +08:00
if (IsFGR64onMips32) {
[mips][msa] Implement f16 support The MIPS MSA ASE provides instructions to convert to and from half precision floating point. This patch teaches the MIPS backend to treat f16 as a legal type and how to promote such values to f32 for the usual set of operations. As a result of this, the fexup[lr].w intrinsics no longer crash LLVM during type legalization. Reviewers: zoran.jovanvoic, vkalintiris Differential Revision: https://reviews.llvm.org/D26398 llvm-svn: 287349
2016-11-19 00:17:44 +08:00
unsigned Rtemp2 = RegInfo.createVirtualRegister(GPRRC);
BuildMI(*BB, MI, DL, TII->get(Mips::MFHC1_D64), Rtemp2).addReg(Fs);
unsigned Wtemp2 = RegInfo.createVirtualRegister(&Mips::MSA128WRegClass);
unsigned Wtemp3 = RegInfo.createVirtualRegister(&Mips::MSA128WRegClass);
BuildMI(*BB, MI, DL, TII->get(Mips::INSERT_W), Wtemp2)
.addReg(Wtemp)
.addReg(Rtemp2)
.addImm(1);
BuildMI(*BB, MI, DL, TII->get(Mips::INSERT_W), Wtemp3)
.addReg(Wtemp2)
.addReg(Rtemp2)
.addImm(3);
WPHI = Wtemp3;
}
if (IsFGR64) {
unsigned Wtemp2 = RegInfo.createVirtualRegister(&Mips::MSA128WRegClass);
BuildMI(*BB, MI, DL, TII->get(Mips::FEXDO_W), Wtemp2)
.addReg(WPHI)
.addReg(WPHI);
WPHI = Wtemp2;
}
BuildMI(*BB, MI, DL, TII->get(Mips::FEXDO_H), Wd).addReg(WPHI).addReg(WPHI);
MI.eraseFromParent();
return BB;
}
// Emit the FPEXTEND_PSEUDO instruction.
//
// Expand an f16 to either a FGR32Opnd or FGR64Opnd.
//
// Safety: Cycle the result through the GPRs so the result always ends up
// the correct floating point register.
//
// FIXME: This copying is strictly unnecessary. If we could tie FGR32Opnd:$Fd
// / FGR64Opnd:$Fd and MSA128F16:$Ws to the same physical register
// (which they can be, as the MSA registers are defined to alias the
// FPU's 64 bit and 32 bit registers) the result can be accessed using
// the correct register class. That requires operands be tie-able across
// register classes which have a sub/super register class relationship. I
// haven't checked.
//
// For FGR32Opnd:
//
// FPEXTEND FGR32Opnd:$fd, MSA128F16:$ws
// =>
// fexupr.w $wtemp, $ws
// copy_s.w $rtemp, $ws[0]
// mtc1 $rtemp, $fd
//
// For FGR64Opnd on Mips64:
//
// FPEXTEND FGR64Opnd:$fd, MSA128F16:$ws
// =>
// fexupr.w $wtemp, $ws
// fexupr.d $wtemp2, $wtemp
// copy_s.d $rtemp, $wtemp2s[0]
// dmtc1 $rtemp, $fd
//
// For FGR64Opnd on Mips32:
//
// FPEXTEND FGR64Opnd:$fd, MSA128F16:$ws
// =>
// fexupr.w $wtemp, $ws
// fexupr.d $wtemp2, $wtemp
// copy_s.w $rtemp, $wtemp2[0]
// mtc1 $rtemp, $ftemp
// copy_s.w $rtemp2, $wtemp2[1]
// $fd = mthc1 $rtemp2, $ftemp
MachineBasicBlock *
MipsSETargetLowering::emitFPEXTEND_PSEUDO(MachineInstr &MI,
MachineBasicBlock *BB,
bool IsFGR64) const {
// Strictly speaking, we need MIPS32R5 to support MSA. We'll be generous
// here. It's technically doable to support MIPS32 here, but the ISA forbids
// it.
assert(Subtarget.hasMSA() && Subtarget.hasMips32r2());
bool IsFGR64onMips64 = Subtarget.hasMips64() && IsFGR64;
bool IsFGR64onMips32 = !Subtarget.hasMips64() && IsFGR64;
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
DebugLoc DL = MI.getDebugLoc();
unsigned Fd = MI.getOperand(0).getReg();
unsigned Ws = MI.getOperand(1).getReg();
MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
const TargetRegisterClass *GPRRC =
IsFGR64onMips64 ? &Mips::GPR64RegClass : &Mips::GPR32RegClass;
[mips] Alter register classes for MSA pseudo f16 instructions This change introduces additional machine instructions in functions dealing with the expansion of msa pseudo f16 instructions due to register classes being inappropriate when checked with machine verifier. Differential Revision: https://reviews.llvm.org/D34276 llvm-svn: 308301
2017-07-18 20:05:35 +08:00
unsigned MTC1Opc = IsFGR64onMips64
? Mips::DMTC1
: (IsFGR64onMips32 ? Mips::MTC1_D64 : Mips::MTC1);
[mips][msa] Implement f16 support The MIPS MSA ASE provides instructions to convert to and from half precision floating point. This patch teaches the MIPS backend to treat f16 as a legal type and how to promote such values to f32 for the usual set of operations. As a result of this, the fexup[lr].w intrinsics no longer crash LLVM during type legalization. Reviewers: zoran.jovanvoic, vkalintiris Differential Revision: https://reviews.llvm.org/D26398 llvm-svn: 287349
2016-11-19 00:17:44 +08:00
unsigned COPYOpc = IsFGR64onMips64 ? Mips::COPY_S_D : Mips::COPY_S_W;
unsigned Wtemp = RegInfo.createVirtualRegister(&Mips::MSA128WRegClass);
unsigned WPHI = Wtemp;
BuildMI(*BB, MI, DL, TII->get(Mips::FEXUPR_W), Wtemp).addReg(Ws);
if (IsFGR64) {
WPHI = RegInfo.createVirtualRegister(&Mips::MSA128DRegClass);
BuildMI(*BB, MI, DL, TII->get(Mips::FEXUPR_D), WPHI).addReg(Wtemp);
}
// Perform the safety regclass copy mentioned above.
unsigned Rtemp = RegInfo.createVirtualRegister(GPRRC);
unsigned FPRPHI = IsFGR64onMips32
? RegInfo.createVirtualRegister(&Mips::FGR64RegClass)
: Fd;
BuildMI(*BB, MI, DL, TII->get(COPYOpc), Rtemp).addReg(WPHI).addImm(0);
BuildMI(*BB, MI, DL, TII->get(MTC1Opc), FPRPHI).addReg(Rtemp);
if (IsFGR64onMips32) {
unsigned Rtemp2 = RegInfo.createVirtualRegister(GPRRC);
BuildMI(*BB, MI, DL, TII->get(Mips::COPY_S_W), Rtemp2)
.addReg(WPHI)
.addImm(1);
BuildMI(*BB, MI, DL, TII->get(Mips::MTHC1_D64), Fd)
.addReg(FPRPHI)
.addReg(Rtemp2);
}
MI.eraseFromParent();
return BB;
}
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
// Emit the FEXP2_W_1 pseudo instructions.
//
// fexp2_w_1_pseudo $wd, $wt
// =>
// ldi.w $ws, 1
// fexp2.w $wd, $ws, $wt
MachineBasicBlock *
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MipsSETargetLowering::emitFEXP2_W_1(MachineInstr &MI,
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
MachineBasicBlock *BB) const {
Remove most of the TargetMachine::getSubtarget/getSubtargetImpl calls that don't take a Function argument from Mips. Notable exceptions: the AsmPrinter and MipsTargetObjectFile. The latter needs to be fixed, and the former will be fixed when the general AsmPrinter changes happen. llvm-svn: 227512
2015-01-30 07:27:36 +08:00
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
const TargetRegisterClass *RC = &Mips::MSA128WRegClass;
unsigned Ws1 = RegInfo.createVirtualRegister(RC);
unsigned Ws2 = RegInfo.createVirtualRegister(RC);
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
DebugLoc DL = MI.getDebugLoc();
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
// Splat 1.0 into a vector
BuildMI(*BB, MI, DL, TII->get(Mips::LDI_W), Ws1).addImm(1);
BuildMI(*BB, MI, DL, TII->get(Mips::FFINT_U_W), Ws2).addReg(Ws1);
// Emit 1.0 * fexp2(Wt)
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
BuildMI(*BB, MI, DL, TII->get(Mips::FEXP2_W), MI.getOperand(0).getReg())
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
.addReg(Ws2)
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
.addReg(MI.getOperand(1).getReg());
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MI.eraseFromParent(); // The pseudo instruction is gone now.
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
return BB;
}
// Emit the FEXP2_D_1 pseudo instructions.
//
// fexp2_d_1_pseudo $wd, $wt
// =>
// ldi.d $ws, 1
// fexp2.d $wd, $ws, $wt
MachineBasicBlock *
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MipsSETargetLowering::emitFEXP2_D_1(MachineInstr &MI,
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
MachineBasicBlock *BB) const {
Remove most of the TargetMachine::getSubtarget/getSubtargetImpl calls that don't take a Function argument from Mips. Notable exceptions: the AsmPrinter and MipsTargetObjectFile. The latter needs to be fixed, and the former will be fixed when the general AsmPrinter changes happen. llvm-svn: 227512
2015-01-30 07:27:36 +08:00
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
MachineRegisterInfo &RegInfo = BB->getParent()->getRegInfo();
const TargetRegisterClass *RC = &Mips::MSA128DRegClass;
unsigned Ws1 = RegInfo.createVirtualRegister(RC);
unsigned Ws2 = RegInfo.createVirtualRegister(RC);
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
DebugLoc DL = MI.getDebugLoc();
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
// Splat 1.0 into a vector
BuildMI(*BB, MI, DL, TII->get(Mips::LDI_D), Ws1).addImm(1);
BuildMI(*BB, MI, DL, TII->get(Mips::FFINT_U_D), Ws2).addReg(Ws1);
// Emit 1.0 * fexp2(Wt)
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
BuildMI(*BB, MI, DL, TII->get(Mips::FEXP2_D), MI.getOperand(0).getReg())
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
.addReg(Ws2)
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
.addReg(MI.getOperand(1).getReg());
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
CodeGen: Use MachineInstr& in TargetLowering, NFC This is a mechanical change to make TargetLowering API take MachineInstr& (instead of MachineInstr*), since the argument is expected to be a valid MachineInstr. In one case, changed a parameter from MachineInstr* to MachineBasicBlock::iterator, since it was used as an insertion point. As a side effect, this removes a bunch of MachineInstr* to MachineBasicBlock::iterator implicit conversions, a necessary step toward fixing PR26753. llvm-svn: 274287
2016-07-01 06:52:52 +08:00
MI.eraseFromParent(); // The pseudo instruction is gone now.
[mips][msa] Added support for matching fexp2 from normal IR (i.e. not intrinsics) llvm-svn: 193239
2013-10-23 18:36:52 +08:00
return BB;
}