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/PowerPC/PPCMIPeephole.cpp

1191 lines
46 KiB
C++
Raw Normal View History

[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
//===-------------- PPCMIPeephole.cpp - MI Peephole Cleanups -------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===---------------------------------------------------------------------===//
//
// This pass performs peephole optimizations to clean up ugly code
// sequences at the MachineInstruction layer. It runs at the end of
// the SSA phases, following VSX swap removal. A pass of dead code
// elimination follows this one for quick clean-up of any dead
// instructions introduced here. Although we could do this as callbacks
// from the generic peephole pass, this would have a couple of bad
// effects: it might remove optimization opportunities for VSX swap
// removal, and it would miss cleanups made possible following VSX
// swap removal.
//
//===---------------------------------------------------------------------===//
#include "PPC.h"
#include "PPCInstrBuilder.h"
Sort the remaining #include lines in include/... and lib/.... I did this a long time ago with a janky python script, but now clang-format has built-in support for this. I fed clang-format every line with a #include and let it re-sort things according to the precise LLVM rules for include ordering baked into clang-format these days. I've reverted a number of files where the results of sorting includes isn't healthy. Either places where we have legacy code relying on particular include ordering (where possible, I'll fix these separately) or where we have particular formatting around #include lines that I didn't want to disturb in this patch. This patch is *entirely* mechanical. If you get merge conflicts or anything, just ignore the changes in this patch and run clang-format over your #include lines in the files. Sorry for any noise here, but it is important to keep these things stable. I was seeing an increasing number of patches with irrelevant re-ordering of #include lines because clang-format was used. This patch at least isolates that churn, makes it easy to skip when resolving conflicts, and gets us to a clean baseline (again). llvm-svn: 304787
2017-06-06 19:49:48 +08:00
#include "PPCInstrInfo.h"
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
#include "PPCTargetMachine.h"
[PowerPC Peephole] Constants into a join add, use ADDI over LI/ADD. Two blocks prior to the join each perform an li and the the join block has an add using the initialized register. Optimize each predecessor block to instead use addi and delete the li's and add. Differential Revision: https://reviews.llvm.org/D36734 llvm-svn: 313639
2017-09-20 00:14:37 +08:00
#include "llvm/ADT/Statistic.h"
#include "llvm/CodeGen/MachineDominators.h"
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
#include "llvm/CodeGen/MachineFunctionPass.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/Support/Debug.h"
[PowerPC] Eliminate sign- and zero-extensions if already sign- or zero-extended This patch enables redundant sign- and zero-extension elimination in PowerPC MI Peephole pass. If the input value of a sign- or zero-extension is known to be already sign- or zero-extended, the operation is redundant and can be eliminated. One common case is sign-extensions for a method parameter or for a method return value; they must be sign- or zero-extended as defined in PPC ELF ABI. For example of the following simple code, two extsw instructions are generated before the invocation of int_func and before the return. With this patch, both extsw are eliminated. void int_func(int); void ii_test(int a) { if (a & 1) return int_func(a); } Such redundant sign- or zero-extensions are quite common in many programs; e.g. I observed about 60,000 occurrences of the elimination while compiling the LLVM+CLANG. Differential Revision: https://reviews.llvm.org/D31319 llvm-svn: 315888
2017-10-16 12:12:57 +08:00
#include "llvm/ADT/Statistic.h"
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
#include "MCTargetDesc/PPCPredicates.h"
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
using namespace llvm;
#define DEBUG_TYPE "ppc-mi-peepholes"
[PowerPC] Eliminate sign- and zero-extensions if already sign- or zero-extended This patch enables redundant sign- and zero-extension elimination in PowerPC MI Peephole pass. If the input value of a sign- or zero-extension is known to be already sign- or zero-extended, the operation is redundant and can be eliminated. One common case is sign-extensions for a method parameter or for a method return value; they must be sign- or zero-extended as defined in PPC ELF ABI. For example of the following simple code, two extsw instructions are generated before the invocation of int_func and before the return. With this patch, both extsw are eliminated. void int_func(int); void ii_test(int a) { if (a & 1) return int_func(a); } Such redundant sign- or zero-extensions are quite common in many programs; e.g. I observed about 60,000 occurrences of the elimination while compiling the LLVM+CLANG. Differential Revision: https://reviews.llvm.org/D31319 llvm-svn: 315888
2017-10-16 12:12:57 +08:00
STATISTIC(NumEliminatedSExt, "Number of eliminated sign-extensions");
STATISTIC(NumEliminatedZExt, "Number of eliminated zero-extensions");
[PowerPC Peephole] Constants into a join add, use ADDI over LI/ADD. Two blocks prior to the join each perform an li and the the join block has an add using the initialized register. Optimize each predecessor block to instead use addi and delete the li's and add. Differential Revision: https://reviews.llvm.org/D36734 llvm-svn: 313639
2017-09-20 00:14:37 +08:00
STATISTIC(NumOptADDLIs, "Number of optimized ADD instruction fed by LI");
[PowerPC] Eliminate sign- and zero-extensions if already sign- or zero-extended This patch enables redundant sign- and zero-extension elimination in PowerPC MI Peephole pass. If the input value of a sign- or zero-extension is known to be already sign- or zero-extended, the operation is redundant and can be eliminated. One common case is sign-extensions for a method parameter or for a method return value; they must be sign- or zero-extended as defined in PPC ELF ABI. For example of the following simple code, two extsw instructions are generated before the invocation of int_func and before the return. With this patch, both extsw are eliminated. void int_func(int); void ii_test(int a) { if (a & 1) return int_func(a); } Such redundant sign- or zero-extensions are quite common in many programs; e.g. I observed about 60,000 occurrences of the elimination while compiling the LLVM+CLANG. Differential Revision: https://reviews.llvm.org/D31319 llvm-svn: 315888
2017-10-16 12:12:57 +08:00
static cl::opt<bool>
EnableSExtElimination("ppc-eliminate-signext",
cl::desc("enable elimination of sign-extensions"),
cl::init(true), cl::Hidden);
static cl::opt<bool>
EnableZExtElimination("ppc-eliminate-zeroext",
cl::desc("enable elimination of zero-extensions"),
cl::init(true), cl::Hidden);
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
namespace llvm {
void initializePPCMIPeepholePass(PassRegistry&);
}
namespace {
struct PPCMIPeephole : public MachineFunctionPass {
static char ID;
const PPCInstrInfo *TII;
MachineFunction *MF;
MachineRegisterInfo *MRI;
PPCMIPeephole() : MachineFunctionPass(ID) {
initializePPCMIPeepholePass(*PassRegistry::getPassRegistry());
}
private:
[PowerPC Peephole] Constants into a join add, use ADDI over LI/ADD. Two blocks prior to the join each perform an li and the the join block has an add using the initialized register. Optimize each predecessor block to instead use addi and delete the li's and add. Differential Revision: https://reviews.llvm.org/D36734 llvm-svn: 313639
2017-09-20 00:14:37 +08:00
MachineDominatorTree *MDT;
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
// Initialize class variables.
void initialize(MachineFunction &MFParm);
// Perform peepholes.
bool simplifyCode(void);
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
// Perform peepholes.
bool eliminateRedundantCompare(void);
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
// Find the "true" register represented by SrcReg (following chains
// of copies and subreg_to_reg operations).
unsigned lookThruCopyLike(unsigned SrcReg);
public:
[PowerPC Peephole] Constants into a join add, use ADDI over LI/ADD. Two blocks prior to the join each perform an li and the the join block has an add using the initialized register. Optimize each predecessor block to instead use addi and delete the li's and add. Differential Revision: https://reviews.llvm.org/D36734 llvm-svn: 313639
2017-09-20 00:14:37 +08:00
void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.addRequired<MachineDominatorTree>();
AU.addPreserved<MachineDominatorTree>();
MachineFunctionPass::getAnalysisUsage(AU);
}
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
// Main entry point for this pass.
bool runOnMachineFunction(MachineFunction &MF) override {
Add optimization bisect opt-in calls for PowerPC passes Differential Revision: http://reviews.llvm.org/D19554 llvm-svn: 267769
2016-04-28 03:39:32 +08:00
if (skipFunction(*MF.getFunction()))
return false;
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
initialize(MF);
return simplifyCode();
}
};
// Initialize class variables.
void PPCMIPeephole::initialize(MachineFunction &MFParm) {
MF = &MFParm;
MRI = &MF->getRegInfo();
[PowerPC Peephole] Constants into a join add, use ADDI over LI/ADD. Two blocks prior to the join each perform an li and the the join block has an add using the initialized register. Optimize each predecessor block to instead use addi and delete the li's and add. Differential Revision: https://reviews.llvm.org/D36734 llvm-svn: 313639
2017-09-20 00:14:37 +08:00
MDT = &getAnalysis<MachineDominatorTree>();
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
TII = MF->getSubtarget<PPCSubtarget>().getInstrInfo();
DEBUG(dbgs() << "*** PowerPC MI peephole pass ***\n\n");
DEBUG(MF->dump());
}
[PowerPC Peephole] Constants into a join add, use ADDI over LI/ADD. Two blocks prior to the join each perform an li and the the join block has an add using the initialized register. Optimize each predecessor block to instead use addi and delete the li's and add. Differential Revision: https://reviews.llvm.org/D36734 llvm-svn: 313639
2017-09-20 00:14:37 +08:00
static MachineInstr *getVRegDefOrNull(MachineOperand *Op,
MachineRegisterInfo *MRI) {
assert(Op && "Invalid Operand!");
if (!Op->isReg())
return nullptr;
unsigned Reg = Op->getReg();
if (!TargetRegisterInfo::isVirtualRegister(Reg))
return nullptr;
return MRI->getVRegDef(Reg);
}
[PowerPC] Eliminate sign- and zero-extensions if already sign- or zero-extended This patch enables redundant sign- and zero-extension elimination in PowerPC MI Peephole pass. If the input value of a sign- or zero-extension is known to be already sign- or zero-extended, the operation is redundant and can be eliminated. One common case is sign-extensions for a method parameter or for a method return value; they must be sign- or zero-extended as defined in PPC ELF ABI. For example of the following simple code, two extsw instructions are generated before the invocation of int_func and before the return. With this patch, both extsw are eliminated. void int_func(int); void ii_test(int a) { if (a & 1) return int_func(a); } Such redundant sign- or zero-extensions are quite common in many programs; e.g. I observed about 60,000 occurrences of the elimination while compiling the LLVM+CLANG. Differential Revision: https://reviews.llvm.org/D31319 llvm-svn: 315888
2017-10-16 12:12:57 +08:00
// This function returns number of known zero bits in output of MI
// starting from the most significant bit.
static unsigned
getKnownLeadingZeroCount(MachineInstr *MI, const PPCInstrInfo *TII) {
unsigned Opcode = MI->getOpcode();
if (Opcode == PPC::RLDICL || Opcode == PPC::RLDICLo ||
Opcode == PPC::RLDCL || Opcode == PPC::RLDCLo)
return MI->getOperand(3).getImm();
if ((Opcode == PPC::RLDIC || Opcode == PPC::RLDICo) &&
MI->getOperand(3).getImm() <= 63 - MI->getOperand(2).getImm())
return MI->getOperand(3).getImm();
if ((Opcode == PPC::RLWINM || Opcode == PPC::RLWINMo ||
Opcode == PPC::RLWNM || Opcode == PPC::RLWNMo ||
Opcode == PPC::RLWINM8 || Opcode == PPC::RLWNM8) &&
MI->getOperand(3).getImm() <= MI->getOperand(4).getImm())
return 32 + MI->getOperand(3).getImm();
if (Opcode == PPC::ANDIo) {
uint16_t Imm = MI->getOperand(2).getImm();
return 48 + countLeadingZeros(Imm);
}
if (Opcode == PPC::CNTLZW || Opcode == PPC::CNTLZWo ||
Opcode == PPC::CNTTZW || Opcode == PPC::CNTTZWo ||
Opcode == PPC::CNTLZW8 || Opcode == PPC::CNTTZW8)
// The result ranges from 0 to 32.
return 58;
if (Opcode == PPC::CNTLZD || Opcode == PPC::CNTLZDo ||
Opcode == PPC::CNTTZD || Opcode == PPC::CNTTZDo)
// The result ranges from 0 to 64.
return 57;
if (Opcode == PPC::LHZ || Opcode == PPC::LHZX ||
Opcode == PPC::LHZ8 || Opcode == PPC::LHZX8 ||
Opcode == PPC::LHZU || Opcode == PPC::LHZUX ||
Opcode == PPC::LHZU8 || Opcode == PPC::LHZUX8)
return 48;
if (Opcode == PPC::LBZ || Opcode == PPC::LBZX ||
Opcode == PPC::LBZ8 || Opcode == PPC::LBZX8 ||
Opcode == PPC::LBZU || Opcode == PPC::LBZUX ||
Opcode == PPC::LBZU8 || Opcode == PPC::LBZUX8)
return 56;
if (TII->isZeroExtended(*MI))
return 32;
return 0;
}
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
// Perform peephole optimizations.
bool PPCMIPeephole::simplifyCode(void) {
bool Simplified = false;
MachineInstr* ToErase = nullptr;
for (MachineBasicBlock &MBB : *MF) {
for (MachineInstr &MI : MBB) {
// If the previous instruction was marked for elimination,
// remove it now.
if (ToErase) {
ToErase->eraseFromParent();
ToErase = nullptr;
}
// Ignore debug instructions.
if (MI.isDebugValue())
continue;
// Per-opcode peepholes.
switch (MI.getOpcode()) {
default:
break;
case PPC::XXPERMDI: {
// Perform simplifications of 2x64 vector swaps and splats.
// A swap is identified by an immediate value of 2, and a splat
// is identified by an immediate value of 0 or 3.
int Immed = MI.getOperand(3).getImm();
if (Immed != 1) {
// For each of these simplifications, we need the two source
// regs to match. Unfortunately, MachineCSE ignores COPY and
// SUBREG_TO_REG, so for example we can see
// XXPERMDI t, SUBREG_TO_REG(s), SUBREG_TO_REG(s), immed.
// We have to look through chains of COPY and SUBREG_TO_REG
// to find the real source values for comparison.
unsigned TrueReg1 = lookThruCopyLike(MI.getOperand(1).getReg());
unsigned TrueReg2 = lookThruCopyLike(MI.getOperand(2).getReg());
if (TrueReg1 == TrueReg2
&& TargetRegisterInfo::isVirtualRegister(TrueReg1)) {
MachineInstr *DefMI = MRI->getVRegDef(TrueReg1);
[PowerPC] Improvements for BUILD_VECTOR Vol. 4 This is the final patch in the series of patches that improves BUILD_VECTOR handling on PowerPC. This adds a few peephole optimizations to remove redundant instructions. It also adds a large test case which encompasses a large set of code patterns that build vectors - this test case was the motivator for this series of patches. Differential Revision: https://reviews.llvm.org/D26066 llvm-svn: 288800
2016-12-06 19:47:14 +08:00
unsigned DefOpc = DefMI ? DefMI->getOpcode() : 0;
// If this is a splat fed by a splatting load, the splat is
// redundant. Replace with a copy. This doesn't happen directly due
// to code in PPCDAGToDAGISel.cpp, but it can happen when converting
// a load of a double to a vector of 64-bit integers.
auto isConversionOfLoadAndSplat = [=]() -> bool {
if (DefOpc != PPC::XVCVDPSXDS && DefOpc != PPC::XVCVDPUXDS)
return false;
unsigned DefReg = lookThruCopyLike(DefMI->getOperand(1).getReg());
if (TargetRegisterInfo::isVirtualRegister(DefReg)) {
MachineInstr *LoadMI = MRI->getVRegDef(DefReg);
if (LoadMI && LoadMI->getOpcode() == PPC::LXVDSX)
return true;
}
return false;
};
if (DefMI && (Immed == 0 || Immed == 3)) {
if (DefOpc == PPC::LXVDSX || isConversionOfLoadAndSplat()) {
DEBUG(dbgs()
<< "Optimizing load-and-splat/splat "
"to load-and-splat/copy: ");
DEBUG(MI.dump());
[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
BuildMI(MBB, &MI, MI.getDebugLoc(), TII->get(PPC::COPY),
MI.getOperand(0).getReg())
.add(MI.getOperand(1));
[PowerPC] Improvements for BUILD_VECTOR Vol. 4 This is the final patch in the series of patches that improves BUILD_VECTOR handling on PowerPC. This adds a few peephole optimizations to remove redundant instructions. It also adds a large test case which encompasses a large set of code patterns that build vectors - this test case was the motivator for this series of patches. Differential Revision: https://reviews.llvm.org/D26066 llvm-svn: 288800
2016-12-06 19:47:14 +08:00
ToErase = &MI;
Simplified = true;
}
}
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
// If this is a splat or a swap fed by another splat, we
// can replace it with a copy.
[PowerPC] Improvements for BUILD_VECTOR Vol. 4 This is the final patch in the series of patches that improves BUILD_VECTOR handling on PowerPC. This adds a few peephole optimizations to remove redundant instructions. It also adds a large test case which encompasses a large set of code patterns that build vectors - this test case was the motivator for this series of patches. Differential Revision: https://reviews.llvm.org/D26066 llvm-svn: 288800
2016-12-06 19:47:14 +08:00
if (DefOpc == PPC::XXPERMDI) {
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
unsigned FeedImmed = DefMI->getOperand(3).getImm();
unsigned FeedReg1
= lookThruCopyLike(DefMI->getOperand(1).getReg());
unsigned FeedReg2
= lookThruCopyLike(DefMI->getOperand(2).getReg());
if ((FeedImmed == 0 || FeedImmed == 3) && FeedReg1 == FeedReg2) {
DEBUG(dbgs()
<< "Optimizing splat/swap or splat/splat "
"to splat/copy: ");
DEBUG(MI.dump());
[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
BuildMI(MBB, &MI, MI.getDebugLoc(), TII->get(PPC::COPY),
MI.getOperand(0).getReg())
.add(MI.getOperand(1));
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
ToErase = &MI;
Simplified = true;
}
// If this is a splat fed by a swap, we can simplify modify
// the splat to splat the other value from the swap's input
// parameter.
else if ((Immed == 0 || Immed == 3)
&& FeedImmed == 2 && FeedReg1 == FeedReg2) {
DEBUG(dbgs() << "Optimizing swap/splat => splat: ");
DEBUG(MI.dump());
MI.getOperand(1).setReg(DefMI->getOperand(1).getReg());
MI.getOperand(2).setReg(DefMI->getOperand(2).getReg());
MI.getOperand(3).setImm(3 - Immed);
Simplified = true;
}
// If this is a swap fed by a swap, we can replace it
// with a copy from the first swap's input.
else if (Immed == 2 && FeedImmed == 2 && FeedReg1 == FeedReg2) {
DEBUG(dbgs() << "Optimizing swap/swap => copy: ");
DEBUG(MI.dump());
[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
BuildMI(MBB, &MI, MI.getDebugLoc(), TII->get(PPC::COPY),
MI.getOperand(0).getReg())
.add(DefMI->getOperand(1));
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
ToErase = &MI;
Simplified = true;
}
[PowerPC] Improvements for BUILD_VECTOR Vol. 4 This is the final patch in the series of patches that improves BUILD_VECTOR handling on PowerPC. This adds a few peephole optimizations to remove redundant instructions. It also adds a large test case which encompasses a large set of code patterns that build vectors - this test case was the motivator for this series of patches. Differential Revision: https://reviews.llvm.org/D26066 llvm-svn: 288800
2016-12-06 19:47:14 +08:00
} else if ((Immed == 0 || Immed == 3) && DefOpc == PPC::XXPERMDIs &&
(DefMI->getOperand(2).getImm() == 0 ||
DefMI->getOperand(2).getImm() == 3)) {
[Power9] Part-word VSX integer scalar loads/stores and sign extend instructions This patch corresponds to review: https://reviews.llvm.org/D23155 This patch removes the VSHRC register class (based on D20310) and adds exploitation of the Power9 sub-word integer loads into VSX registers as well as vector sign extensions. The new instructions are useful for a few purposes: Int to Fp conversions of 1 or 2-byte values loaded from memory Building vectors of 1 or 2-byte integers with values loaded from memory Storing individual 1 or 2-byte elements from integer vectors This patch implements all of those uses. llvm-svn: 283190
2016-10-04 14:59:23 +08:00
// Splat fed by another splat - switch the output of the first
// and remove the second.
DefMI->getOperand(0).setReg(MI.getOperand(0).getReg());
ToErase = &MI;
Simplified = true;
DEBUG(dbgs() << "Removing redundant splat: ");
DEBUG(MI.dump());
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
}
}
}
break;
}
[Power9] Part-word VSX integer scalar loads/stores and sign extend instructions This patch corresponds to review: https://reviews.llvm.org/D23155 This patch removes the VSHRC register class (based on D20310) and adds exploitation of the Power9 sub-word integer loads into VSX registers as well as vector sign extensions. The new instructions are useful for a few purposes: Int to Fp conversions of 1 or 2-byte values loaded from memory Building vectors of 1 or 2-byte integers with values loaded from memory Storing individual 1 or 2-byte elements from integer vectors This patch implements all of those uses. llvm-svn: 283190
2016-10-04 14:59:23 +08:00
case PPC::VSPLTB:
case PPC::VSPLTH:
case PPC::XXSPLTW: {
unsigned MyOpcode = MI.getOpcode();
unsigned OpNo = MyOpcode == PPC::XXSPLTW ? 1 : 2;
unsigned TrueReg = lookThruCopyLike(MI.getOperand(OpNo).getReg());
[PowerPC] Improvements for BUILD_VECTOR Vol. 4 This is the final patch in the series of patches that improves BUILD_VECTOR handling on PowerPC. This adds a few peephole optimizations to remove redundant instructions. It also adds a large test case which encompasses a large set of code patterns that build vectors - this test case was the motivator for this series of patches. Differential Revision: https://reviews.llvm.org/D26066 llvm-svn: 288800
2016-12-06 19:47:14 +08:00
if (!TargetRegisterInfo::isVirtualRegister(TrueReg))
break;
[Power9] Part-word VSX integer scalar loads/stores and sign extend instructions This patch corresponds to review: https://reviews.llvm.org/D23155 This patch removes the VSHRC register class (based on D20310) and adds exploitation of the Power9 sub-word integer loads into VSX registers as well as vector sign extensions. The new instructions are useful for a few purposes: Int to Fp conversions of 1 or 2-byte values loaded from memory Building vectors of 1 or 2-byte integers with values loaded from memory Storing individual 1 or 2-byte elements from integer vectors This patch implements all of those uses. llvm-svn: 283190
2016-10-04 14:59:23 +08:00
MachineInstr *DefMI = MRI->getVRegDef(TrueReg);
if (!DefMI)
break;
unsigned DefOpcode = DefMI->getOpcode();
[PowerPC] Improvements for BUILD_VECTOR Vol. 4 This is the final patch in the series of patches that improves BUILD_VECTOR handling on PowerPC. This adds a few peephole optimizations to remove redundant instructions. It also adds a large test case which encompasses a large set of code patterns that build vectors - this test case was the motivator for this series of patches. Differential Revision: https://reviews.llvm.org/D26066 llvm-svn: 288800
2016-12-06 19:47:14 +08:00
auto isConvertOfSplat = [=]() -> bool {
if (DefOpcode != PPC::XVCVSPSXWS && DefOpcode != PPC::XVCVSPUXWS)
return false;
unsigned ConvReg = DefMI->getOperand(1).getReg();
if (!TargetRegisterInfo::isVirtualRegister(ConvReg))
return false;
MachineInstr *Splt = MRI->getVRegDef(ConvReg);
return Splt && (Splt->getOpcode() == PPC::LXVWSX ||
Splt->getOpcode() == PPC::XXSPLTW);
};
bool AlreadySplat = (MyOpcode == DefOpcode) ||
[Power9] Part-word VSX integer scalar loads/stores and sign extend instructions This patch corresponds to review: https://reviews.llvm.org/D23155 This patch removes the VSHRC register class (based on D20310) and adds exploitation of the Power9 sub-word integer loads into VSX registers as well as vector sign extensions. The new instructions are useful for a few purposes: Int to Fp conversions of 1 or 2-byte values loaded from memory Building vectors of 1 or 2-byte integers with values loaded from memory Storing individual 1 or 2-byte elements from integer vectors This patch implements all of those uses. llvm-svn: 283190
2016-10-04 14:59:23 +08:00
(MyOpcode == PPC::VSPLTB && DefOpcode == PPC::VSPLTBs) ||
(MyOpcode == PPC::VSPLTH && DefOpcode == PPC::VSPLTHs) ||
[PowerPC] Improvements for BUILD_VECTOR Vol. 4 This is the final patch in the series of patches that improves BUILD_VECTOR handling on PowerPC. This adds a few peephole optimizations to remove redundant instructions. It also adds a large test case which encompasses a large set of code patterns that build vectors - this test case was the motivator for this series of patches. Differential Revision: https://reviews.llvm.org/D26066 llvm-svn: 288800
2016-12-06 19:47:14 +08:00
(MyOpcode == PPC::XXSPLTW && DefOpcode == PPC::XXSPLTWs) ||
(MyOpcode == PPC::XXSPLTW && DefOpcode == PPC::LXVWSX) ||
(MyOpcode == PPC::XXSPLTW && DefOpcode == PPC::MTVSRWS)||
(MyOpcode == PPC::XXSPLTW && isConvertOfSplat());
// If the instruction[s] that feed this splat have already splat
// the value, this splat is redundant.
if (AlreadySplat) {
[PPCMIPeephole] Fix splat elimination Summary: In PPCMIPeephole, when we see two splat instructions, we can't simply do the following transformation: B = Splat A C = Splat B => C = Splat A because B may still be used between these two instructions. Instead, we should make the second Splat a PPC::COPY and let later passes decide whether to remove it or not: B = Splat A C = Splat B => B = Splat A C = COPY B Fixes PR30663. Reviewers: echristo, iteratee, kbarton, nemanjai Subscribers: mehdi_amini, llvm-commits Differential Revision: https://reviews.llvm.org/D25493 llvm-svn: 283961
2016-10-12 08:48:25 +08:00
DEBUG(dbgs() << "Changing redundant splat to a copy: ");
DEBUG(MI.dump());
BuildMI(MBB, &MI, MI.getDebugLoc(), TII->get(PPC::COPY),
MI.getOperand(0).getReg())
[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
.add(MI.getOperand(OpNo));
[Power9] Part-word VSX integer scalar loads/stores and sign extend instructions This patch corresponds to review: https://reviews.llvm.org/D23155 This patch removes the VSHRC register class (based on D20310) and adds exploitation of the Power9 sub-word integer loads into VSX registers as well as vector sign extensions. The new instructions are useful for a few purposes: Int to Fp conversions of 1 or 2-byte values loaded from memory Building vectors of 1 or 2-byte integers with values loaded from memory Storing individual 1 or 2-byte elements from integer vectors This patch implements all of those uses. llvm-svn: 283190
2016-10-04 14:59:23 +08:00
ToErase = &MI;
Simplified = true;
}
// Splat fed by a shift. Usually when we align value to splat into
// vector element zero.
if (DefOpcode == PPC::XXSLDWI) {
unsigned ShiftRes = DefMI->getOperand(0).getReg();
unsigned ShiftOp1 = DefMI->getOperand(1).getReg();
unsigned ShiftOp2 = DefMI->getOperand(2).getReg();
unsigned ShiftImm = DefMI->getOperand(3).getImm();
unsigned SplatImm = MI.getOperand(2).getImm();
if (ShiftOp1 == ShiftOp2) {
unsigned NewElem = (SplatImm + ShiftImm) & 0x3;
if (MRI->hasOneNonDBGUse(ShiftRes)) {
DEBUG(dbgs() << "Removing redundant shift: ");
DEBUG(DefMI->dump());
ToErase = DefMI;
}
Simplified = true;
DEBUG(dbgs() << "Changing splat immediate from " << SplatImm <<
" to " << NewElem << " in instruction: ");
DEBUG(MI.dump());
MI.getOperand(1).setReg(ShiftOp1);
MI.getOperand(2).setImm(NewElem);
}
}
break;
}
[PowerPC] Improvements for BUILD_VECTOR Vol. 4 This is the final patch in the series of patches that improves BUILD_VECTOR handling on PowerPC. This adds a few peephole optimizations to remove redundant instructions. It also adds a large test case which encompasses a large set of code patterns that build vectors - this test case was the motivator for this series of patches. Differential Revision: https://reviews.llvm.org/D26066 llvm-svn: 288800
2016-12-06 19:47:14 +08:00
case PPC::XVCVDPSP: {
// If this is a DP->SP conversion fed by an FRSP, the FRSP is redundant.
unsigned TrueReg = lookThruCopyLike(MI.getOperand(1).getReg());
if (!TargetRegisterInfo::isVirtualRegister(TrueReg))
break;
MachineInstr *DefMI = MRI->getVRegDef(TrueReg);
// This can occur when building a vector of single precision or integer
// values.
if (DefMI && DefMI->getOpcode() == PPC::XXPERMDI) {
unsigned DefsReg1 = lookThruCopyLike(DefMI->getOperand(1).getReg());
unsigned DefsReg2 = lookThruCopyLike(DefMI->getOperand(2).getReg());
if (!TargetRegisterInfo::isVirtualRegister(DefsReg1) ||
!TargetRegisterInfo::isVirtualRegister(DefsReg2))
break;
MachineInstr *P1 = MRI->getVRegDef(DefsReg1);
MachineInstr *P2 = MRI->getVRegDef(DefsReg2);
if (!P1 || !P2)
break;
// Remove the passed FRSP instruction if it only feeds this MI and
// set any uses of that FRSP (in this MI) to the source of the FRSP.
auto removeFRSPIfPossible = [&](MachineInstr *RoundInstr) {
if (RoundInstr->getOpcode() == PPC::FRSP &&
MRI->hasOneNonDBGUse(RoundInstr->getOperand(0).getReg())) {
Simplified = true;
unsigned ConvReg1 = RoundInstr->getOperand(1).getReg();
unsigned FRSPDefines = RoundInstr->getOperand(0).getReg();
MachineInstr &Use = *(MRI->use_instr_begin(FRSPDefines));
for (int i = 0, e = Use.getNumOperands(); i < e; ++i)
if (Use.getOperand(i).isReg() &&
Use.getOperand(i).getReg() == FRSPDefines)
Use.getOperand(i).setReg(ConvReg1);
DEBUG(dbgs() << "Removing redundant FRSP:\n");
DEBUG(RoundInstr->dump());
DEBUG(dbgs() << "As it feeds instruction:\n");
DEBUG(MI.dump());
DEBUG(dbgs() << "Through instruction:\n");
DEBUG(DefMI->dump());
RoundInstr->eraseFromParent();
}
};
// If the input to XVCVDPSP is a vector that was built (even
// partially) out of FRSP's, the FRSP(s) can safely be removed
// since this instruction performs the same operation.
if (P1 != P2) {
removeFRSPIfPossible(P1);
removeFRSPIfPossible(P2);
break;
}
removeFRSPIfPossible(P1);
}
break;
}
[PowerPC] Eliminate sign- and zero-extensions if already sign- or zero-extended This patch enables redundant sign- and zero-extension elimination in PowerPC MI Peephole pass. If the input value of a sign- or zero-extension is known to be already sign- or zero-extended, the operation is redundant and can be eliminated. One common case is sign-extensions for a method parameter or for a method return value; they must be sign- or zero-extended as defined in PPC ELF ABI. For example of the following simple code, two extsw instructions are generated before the invocation of int_func and before the return. With this patch, both extsw are eliminated. void int_func(int); void ii_test(int a) { if (a & 1) return int_func(a); } Such redundant sign- or zero-extensions are quite common in many programs; e.g. I observed about 60,000 occurrences of the elimination while compiling the LLVM+CLANG. Differential Revision: https://reviews.llvm.org/D31319 llvm-svn: 315888
2017-10-16 12:12:57 +08:00
case PPC::EXTSH:
case PPC::EXTSH8:
case PPC::EXTSH8_32_64: {
if (!EnableSExtElimination) break;
unsigned NarrowReg = MI.getOperand(1).getReg();
if (!TargetRegisterInfo::isVirtualRegister(NarrowReg))
break;
MachineInstr *SrcMI = MRI->getVRegDef(NarrowReg);
// If we've used a zero-extending load that we will sign-extend,
// just do a sign-extending load.
if (SrcMI->getOpcode() == PPC::LHZ ||
SrcMI->getOpcode() == PPC::LHZX) {
if (!MRI->hasOneNonDBGUse(SrcMI->getOperand(0).getReg()))
break;
auto is64Bit = [] (unsigned Opcode) {
return Opcode == PPC::EXTSH8;
};
auto isXForm = [] (unsigned Opcode) {
return Opcode == PPC::LHZX;
};
auto getSextLoadOp = [] (bool is64Bit, bool isXForm) {
if (is64Bit)
if (isXForm) return PPC::LHAX8;
else return PPC::LHA8;
else
if (isXForm) return PPC::LHAX;
else return PPC::LHA;
};
unsigned Opc = getSextLoadOp(is64Bit(MI.getOpcode()),
isXForm(SrcMI->getOpcode()));
DEBUG(dbgs() << "Zero-extending load\n");
DEBUG(SrcMI->dump());
DEBUG(dbgs() << "and sign-extension\n");
DEBUG(MI.dump());
DEBUG(dbgs() << "are merged into sign-extending load\n");
SrcMI->setDesc(TII->get(Opc));
SrcMI->getOperand(0).setReg(MI.getOperand(0).getReg());
ToErase = &MI;
Simplified = true;
NumEliminatedSExt++;
}
break;
}
case PPC::EXTSW:
case PPC::EXTSW_32:
case PPC::EXTSW_32_64: {
if (!EnableSExtElimination) break;
unsigned NarrowReg = MI.getOperand(1).getReg();
if (!TargetRegisterInfo::isVirtualRegister(NarrowReg))
break;
MachineInstr *SrcMI = MRI->getVRegDef(NarrowReg);
// If we've used a zero-extending load that we will sign-extend,
// just do a sign-extending load.
if (SrcMI->getOpcode() == PPC::LWZ ||
SrcMI->getOpcode() == PPC::LWZX) {
if (!MRI->hasOneNonDBGUse(SrcMI->getOperand(0).getReg()))
break;
auto is64Bit = [] (unsigned Opcode) {
return Opcode == PPC::EXTSW || Opcode == PPC::EXTSW_32_64;
};
auto isXForm = [] (unsigned Opcode) {
return Opcode == PPC::LWZX;
};
auto getSextLoadOp = [] (bool is64Bit, bool isXForm) {
if (is64Bit)
if (isXForm) return PPC::LWAX;
else return PPC::LWA;
else
if (isXForm) return PPC::LWAX_32;
else return PPC::LWA_32;
};
unsigned Opc = getSextLoadOp(is64Bit(MI.getOpcode()),
isXForm(SrcMI->getOpcode()));
DEBUG(dbgs() << "Zero-extending load\n");
DEBUG(SrcMI->dump());
DEBUG(dbgs() << "and sign-extension\n");
DEBUG(MI.dump());
DEBUG(dbgs() << "are merged into sign-extending load\n");
SrcMI->setDesc(TII->get(Opc));
SrcMI->getOperand(0).setReg(MI.getOperand(0).getReg());
ToErase = &MI;
Simplified = true;
NumEliminatedSExt++;
} else if (MI.getOpcode() == PPC::EXTSW_32_64 &&
TII->isSignExtended(*SrcMI)) {
// We can eliminate EXTSW if the input is known to be already
// sign-extended.
DEBUG(dbgs() << "Removing redundant sign-extension\n");
unsigned TmpReg =
MF->getRegInfo().createVirtualRegister(&PPC::G8RCRegClass);
BuildMI(MBB, &MI, MI.getDebugLoc(), TII->get(PPC::IMPLICIT_DEF),
TmpReg);
BuildMI(MBB, &MI, MI.getDebugLoc(), TII->get(PPC::INSERT_SUBREG),
MI.getOperand(0).getReg())
.addReg(TmpReg)
.addReg(NarrowReg)
.addImm(PPC::sub_32);
ToErase = &MI;
Simplified = true;
NumEliminatedSExt++;
}
break;
}
case PPC::RLDICL: {
// We can eliminate RLDICL (e.g. for zero-extension)
// if all bits to clear are already zero in the input.
// This code assume following code sequence for zero-extension.
// %vreg6<def> = COPY %vreg5:sub_32; (optional)
// %vreg8<def> = IMPLICIT_DEF;
// %vreg7<def,tied1> = INSERT_SUBREG %vreg8<tied0>, %vreg6, sub_32;
if (!EnableZExtElimination) break;
if (MI.getOperand(2).getImm() != 0)
break;
unsigned SrcReg = MI.getOperand(1).getReg();
if (!TargetRegisterInfo::isVirtualRegister(SrcReg))
break;
MachineInstr *SrcMI = MRI->getVRegDef(SrcReg);
if (!(SrcMI && SrcMI->getOpcode() == PPC::INSERT_SUBREG &&
SrcMI->getOperand(0).isReg() && SrcMI->getOperand(1).isReg()))
break;
MachineInstr *ImpDefMI, *SubRegMI;
ImpDefMI = MRI->getVRegDef(SrcMI->getOperand(1).getReg());
SubRegMI = MRI->getVRegDef(SrcMI->getOperand(2).getReg());
if (ImpDefMI->getOpcode() != PPC::IMPLICIT_DEF) break;
SrcMI = SubRegMI;
if (SubRegMI->getOpcode() == PPC::COPY) {
unsigned CopyReg = SubRegMI->getOperand(1).getReg();
if (TargetRegisterInfo::isVirtualRegister(CopyReg))
SrcMI = MRI->getVRegDef(CopyReg);
}
unsigned KnownZeroCount = getKnownLeadingZeroCount(SrcMI, TII);
if (MI.getOperand(3).getImm() <= KnownZeroCount) {
DEBUG(dbgs() << "Removing redundant zero-extension\n");
BuildMI(MBB, &MI, MI.getDebugLoc(), TII->get(PPC::COPY),
MI.getOperand(0).getReg())
.addReg(SrcReg);
ToErase = &MI;
Simplified = true;
NumEliminatedZExt++;
}
break;
}
[PowerPC Peephole] Constants into a join add, use ADDI over LI/ADD. Two blocks prior to the join each perform an li and the the join block has an add using the initialized register. Optimize each predecessor block to instead use addi and delete the li's and add. Differential Revision: https://reviews.llvm.org/D36734 llvm-svn: 313639
2017-09-20 00:14:37 +08:00
// TODO: Any instruction that has an immediate form fed only by a PHI
// whose operands are all load immediate can be folded away. We currently
// do this for ADD instructions, but should expand it to arithmetic and
// binary instructions with immediate forms in the future.
case PPC::ADD4:
case PPC::ADD8: {
auto isSingleUsePHI = [&](MachineOperand *PhiOp) {
assert(PhiOp && "Invalid Operand!");
MachineInstr *DefPhiMI = getVRegDefOrNull(PhiOp, MRI);
return DefPhiMI && (DefPhiMI->getOpcode() == PPC::PHI) &&
MRI->hasOneNonDBGUse(DefPhiMI->getOperand(0).getReg());
};
auto dominatesAllSingleUseLIs = [&](MachineOperand *DominatorOp,
MachineOperand *PhiOp) {
assert(PhiOp && "Invalid Operand!");
assert(DominatorOp && "Invalid Operand!");
MachineInstr *DefPhiMI = getVRegDefOrNull(PhiOp, MRI);
MachineInstr *DefDomMI = getVRegDefOrNull(DominatorOp, MRI);
// Note: the vregs only show up at odd indices position of PHI Node,
// the even indices position save the BB info.
for (unsigned i = 1; i < DefPhiMI->getNumOperands(); i += 2) {
MachineInstr *LiMI =
getVRegDefOrNull(&DefPhiMI->getOperand(i), MRI);
Fix for PR34888. The issue is that we assume operand zero of the input to the add instruction is a register. In this case, the input comes from inline assembly and operand zero is not a register thereby causing a crash. The code will bail anyway if the input instruction doesn't have the right opcode. So do that check first and let short-circuiting prevent the crash. llvm-svn: 315285
2017-10-10 16:46:10 +08:00
if (!LiMI ||
(LiMI->getOpcode() != PPC::LI && LiMI->getOpcode() != PPC::LI8)
|| !MRI->hasOneNonDBGUse(LiMI->getOperand(0).getReg()) ||
!MDT->dominates(DefDomMI, LiMI))
[PowerPC Peephole] Constants into a join add, use ADDI over LI/ADD. Two blocks prior to the join each perform an li and the the join block has an add using the initialized register. Optimize each predecessor block to instead use addi and delete the li's and add. Differential Revision: https://reviews.llvm.org/D36734 llvm-svn: 313639
2017-09-20 00:14:37 +08:00
return false;
}
return true;
};
MachineOperand Op1 = MI.getOperand(1);
MachineOperand Op2 = MI.getOperand(2);
if (isSingleUsePHI(&Op2) && dominatesAllSingleUseLIs(&Op1, &Op2))
std::swap(Op1, Op2);
else if (!isSingleUsePHI(&Op1) || !dominatesAllSingleUseLIs(&Op2, &Op1))
break; // We don't have an ADD fed by LI's that can be transformed
// Now we know that Op1 is the PHI node and Op2 is the dominator
unsigned DominatorReg = Op2.getReg();
const TargetRegisterClass *TRC = MI.getOpcode() == PPC::ADD8
? &PPC::G8RC_and_G8RC_NOX0RegClass
: &PPC::GPRC_and_GPRC_NOR0RegClass;
MRI->setRegClass(DominatorReg, TRC);
// replace LIs with ADDIs
MachineInstr *DefPhiMI = getVRegDefOrNull(&Op1, MRI);
for (unsigned i = 1; i < DefPhiMI->getNumOperands(); i += 2) {
MachineInstr *LiMI = getVRegDefOrNull(&DefPhiMI->getOperand(i), MRI);
DEBUG(dbgs() << "Optimizing LI to ADDI: ");
DEBUG(LiMI->dump());
// There could be repeated registers in the PHI, e.g: %vreg1<def> =
// PHI %vreg6, <BB#2>, %vreg8, <BB#3>, %vreg8, <BB#6>; So if we've
// already replaced the def instruction, skip.
if (LiMI->getOpcode() == PPC::ADDI || LiMI->getOpcode() == PPC::ADDI8)
continue;
assert((LiMI->getOpcode() == PPC::LI ||
LiMI->getOpcode() == PPC::LI8) &&
"Invalid Opcode!");
auto LiImm = LiMI->getOperand(1).getImm(); // save the imm of LI
LiMI->RemoveOperand(1); // remove the imm of LI
LiMI->setDesc(TII->get(LiMI->getOpcode() == PPC::LI ? PPC::ADDI
: PPC::ADDI8));
MachineInstrBuilder(*LiMI->getParent()->getParent(), *LiMI)
.addReg(DominatorReg)
.addImm(LiImm); // restore the imm of LI
DEBUG(LiMI->dump());
}
// Replace ADD with COPY
DEBUG(dbgs() << "Optimizing ADD to COPY: ");
DEBUG(MI.dump());
BuildMI(MBB, &MI, MI.getDebugLoc(), TII->get(PPC::COPY),
MI.getOperand(0).getReg())
.add(Op1);
ToErase = &MI;
Simplified = true;
NumOptADDLIs++;
break;
}
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
}
}
[PowerPC Peephole] Constants into a join add, use ADDI over LI/ADD. Two blocks prior to the join each perform an li and the the join block has an add using the initialized register. Optimize each predecessor block to instead use addi and delete the li's and add. Differential Revision: https://reviews.llvm.org/D36734 llvm-svn: 313639
2017-09-20 00:14:37 +08:00
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
// If the last instruction was marked for elimination,
// remove it now.
if (ToErase) {
ToErase->eraseFromParent();
ToErase = nullptr;
}
}
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
// We try to eliminate redundant compare instruction.
Simplified |= eliminateRedundantCompare();
return Simplified;
}
// helper functions for eliminateRedundantCompare
static bool isEqOrNe(MachineInstr *BI) {
PPC::Predicate Pred = (PPC::Predicate)BI->getOperand(0).getImm();
unsigned PredCond = PPC::getPredicateCondition(Pred);
return (PredCond == PPC::PRED_EQ || PredCond == PPC::PRED_NE);
}
static bool isSupportedCmpOp(unsigned opCode) {
return (opCode == PPC::CMPLD || opCode == PPC::CMPD ||
opCode == PPC::CMPLW || opCode == PPC::CMPW ||
opCode == PPC::CMPLDI || opCode == PPC::CMPDI ||
opCode == PPC::CMPLWI || opCode == PPC::CMPWI);
}
static bool is64bitCmpOp(unsigned opCode) {
return (opCode == PPC::CMPLD || opCode == PPC::CMPD ||
opCode == PPC::CMPLDI || opCode == PPC::CMPDI);
}
static bool isSignedCmpOp(unsigned opCode) {
return (opCode == PPC::CMPD || opCode == PPC::CMPW ||
opCode == PPC::CMPDI || opCode == PPC::CMPWI);
}
static unsigned getSignedCmpOpCode(unsigned opCode) {
if (opCode == PPC::CMPLD) return PPC::CMPD;
if (opCode == PPC::CMPLW) return PPC::CMPW;
if (opCode == PPC::CMPLDI) return PPC::CMPDI;
if (opCode == PPC::CMPLWI) return PPC::CMPWI;
return opCode;
}
// We can decrement immediate x in (GE x) by changing it to (GT x-1) or
// (LT x) to (LE x-1)
static unsigned getPredicateToDecImm(MachineInstr *BI, MachineInstr *CMPI) {
uint64_t Imm = CMPI->getOperand(2).getImm();
bool SignedCmp = isSignedCmpOp(CMPI->getOpcode());
if ((!SignedCmp && Imm == 0) || (SignedCmp && Imm == 0x8000))
return 0;
PPC::Predicate Pred = (PPC::Predicate)BI->getOperand(0).getImm();
unsigned PredCond = PPC::getPredicateCondition(Pred);
unsigned PredHint = PPC::getPredicateHint(Pred);
if (PredCond == PPC::PRED_GE)
return PPC::getPredicate(PPC::PRED_GT, PredHint);
if (PredCond == PPC::PRED_LT)
return PPC::getPredicate(PPC::PRED_LE, PredHint);
return 0;
}
// We can increment immediate x in (GT x) by changing it to (GE x+1) or
// (LE x) to (LT x+1)
static unsigned getPredicateToIncImm(MachineInstr *BI, MachineInstr *CMPI) {
uint64_t Imm = CMPI->getOperand(2).getImm();
bool SignedCmp = isSignedCmpOp(CMPI->getOpcode());
if ((!SignedCmp && Imm == 0xFFFF) || (SignedCmp && Imm == 0x7FFF))
return 0;
PPC::Predicate Pred = (PPC::Predicate)BI->getOperand(0).getImm();
unsigned PredCond = PPC::getPredicateCondition(Pred);
unsigned PredHint = PPC::getPredicateHint(Pred);
if (PredCond == PPC::PRED_GT)
return PPC::getPredicate(PPC::PRED_GE, PredHint);
if (PredCond == PPC::PRED_LE)
return PPC::getPredicate(PPC::PRED_LT, PredHint);
return 0;
}
[PowerPC] eliminate partially redundant compare instruction This is a follow-on of D37211. D37211 eliminates a compare instruction if two conditional branches can be made based on the one compare instruction, e.g. if (a == 0) { ... } else if (a < 0) { ... } This patch extends this optimization to support partially redundant cases, which often happen in while loops. For example, one compare instruction is moved from the loop body into the preheader by this optimization in the following example. do { if (a == 0) dummy1(); a = func(a); } while (a > 0); Differential Revision: https://reviews.llvm.org/D38236 llvm-svn: 314390
2017-09-28 16:38:19 +08:00
// This takes a Phi node and returns a register value for the spefied BB.
static unsigned getIncomingRegForBlock(MachineInstr *Phi,
MachineBasicBlock *MBB) {
for (unsigned I = 2, E = Phi->getNumOperands() + 1; I != E; I += 2) {
MachineOperand &MO = Phi->getOperand(I);
if (MO.getMBB() == MBB)
return Phi->getOperand(I-1).getReg();
}
llvm_unreachable("invalid src basic block for this Phi node\n");
return 0;
}
// This function tracks the source of the register through register copy.
// If BB1 and BB2 are non-NULL, we also track PHI instruction in BB2
// assuming that the control comes from BB1 into BB2.
static unsigned getSrcVReg(unsigned Reg, MachineBasicBlock *BB1,
MachineBasicBlock *BB2, MachineRegisterInfo *MRI) {
unsigned SrcReg = Reg;
while (1) {
unsigned NextReg = SrcReg;
MachineInstr *Inst = MRI->getVRegDef(SrcReg);
if (BB1 && Inst->getOpcode() == PPC::PHI && Inst->getParent() == BB2) {
NextReg = getIncomingRegForBlock(Inst, BB1);
// We track through PHI only once to avoid infinite loop.
BB1 = nullptr;
}
else if (Inst->isFullCopy())
NextReg = Inst->getOperand(1).getReg();
if (NextReg == SrcReg || !TargetRegisterInfo::isVirtualRegister(NextReg))
break;
SrcReg = NextReg;
}
return SrcReg;
}
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
static bool eligibleForCompareElimination(MachineBasicBlock &MBB,
[PowerPC] eliminate partially redundant compare instruction This is a follow-on of D37211. D37211 eliminates a compare instruction if two conditional branches can be made based on the one compare instruction, e.g. if (a == 0) { ... } else if (a < 0) { ... } This patch extends this optimization to support partially redundant cases, which often happen in while loops. For example, one compare instruction is moved from the loop body into the preheader by this optimization in the following example. do { if (a == 0) dummy1(); a = func(a); } while (a > 0); Differential Revision: https://reviews.llvm.org/D38236 llvm-svn: 314390
2017-09-28 16:38:19 +08:00
MachineBasicBlock *&PredMBB,
MachineBasicBlock *&MBBtoMoveCmp,
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
MachineRegisterInfo *MRI) {
auto isEligibleBB = [&](MachineBasicBlock &BB) {
auto BII = BB.getFirstInstrTerminator();
// We optimize BBs ending with a conditional branch.
// We check only for BCC here, not BCCLR, because BCCLR
// will be formed only later in the pipeline.
if (BB.succ_size() == 2 &&
BII != BB.instr_end() &&
(*BII).getOpcode() == PPC::BCC &&
(*BII).getOperand(1).isReg()) {
// We optimize only if the condition code is used only by one BCC.
unsigned CndReg = (*BII).getOperand(1).getReg();
if (!TargetRegisterInfo::isVirtualRegister(CndReg) ||
!MRI->hasOneNonDBGUse(CndReg))
return false;
// We skip this BB if a physical register is used in comparison.
MachineInstr *CMPI = MRI->getVRegDef(CndReg);
for (MachineOperand &MO : CMPI->operands())
if (MO.isReg() && !TargetRegisterInfo::isVirtualRegister(MO.getReg()))
return false;
return true;
}
return false;
};
[PowerPC] eliminate partially redundant compare instruction This is a follow-on of D37211. D37211 eliminates a compare instruction if two conditional branches can be made based on the one compare instruction, e.g. if (a == 0) { ... } else if (a < 0) { ... } This patch extends this optimization to support partially redundant cases, which often happen in while loops. For example, one compare instruction is moved from the loop body into the preheader by this optimization in the following example. do { if (a == 0) dummy1(); a = func(a); } while (a > 0); Differential Revision: https://reviews.llvm.org/D38236 llvm-svn: 314390
2017-09-28 16:38:19 +08:00
// If this BB has more than one successor, we can create a new BB and
// move the compare instruction in the new BB.
// So far, we do not move compare instruction to a BB having multiple
// successors to avoid potentially increasing code size.
auto isEligibleForMoveCmp = [](MachineBasicBlock &BB) {
return BB.succ_size() == 1;
};
if (!isEligibleBB(MBB))
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
return false;
[PowerPC] eliminate partially redundant compare instruction This is a follow-on of D37211. D37211 eliminates a compare instruction if two conditional branches can be made based on the one compare instruction, e.g. if (a == 0) { ... } else if (a < 0) { ... } This patch extends this optimization to support partially redundant cases, which often happen in while loops. For example, one compare instruction is moved from the loop body into the preheader by this optimization in the following example. do { if (a == 0) dummy1(); a = func(a); } while (a > 0); Differential Revision: https://reviews.llvm.org/D38236 llvm-svn: 314390
2017-09-28 16:38:19 +08:00
unsigned NumPredBBs = MBB.pred_size();
if (NumPredBBs == 1) {
MachineBasicBlock *TmpMBB = *MBB.pred_begin();
if (isEligibleBB(*TmpMBB)) {
PredMBB = TmpMBB;
MBBtoMoveCmp = nullptr;
return true;
}
}
else if (NumPredBBs == 2) {
// We check for partially redundant case.
// So far, we support cases with only two predecessors
// to avoid increasing the number of instructions.
MachineBasicBlock::pred_iterator PI = MBB.pred_begin();
MachineBasicBlock *Pred1MBB = *PI;
MachineBasicBlock *Pred2MBB = *(PI+1);
if (isEligibleBB(*Pred1MBB) && isEligibleForMoveCmp(*Pred2MBB)) {
// We assume Pred1MBB is the BB containing the compare to be merged and
// Pred2MBB is the BB to which we will append a compare instruction.
// Hence we can proceed as is.
}
else if (isEligibleBB(*Pred2MBB) && isEligibleForMoveCmp(*Pred1MBB)) {
// We need to swap Pred1MBB and Pred2MBB to canonicalize.
std::swap(Pred1MBB, Pred2MBB);
}
else return false;
// Here, Pred2MBB is the BB to which we need to append a compare inst.
// We cannot move the compare instruction if operands are not available
// in Pred2MBB (i.e. defined in MBB by an instruction other than PHI).
MachineInstr *BI = &*MBB.getFirstInstrTerminator();
MachineInstr *CMPI = MRI->getVRegDef(BI->getOperand(1).getReg());
for (int I = 1; I <= 2; I++)
if (CMPI->getOperand(I).isReg()) {
MachineInstr *Inst = MRI->getVRegDef(CMPI->getOperand(I).getReg());
if (Inst->getParent() == &MBB && Inst->getOpcode() != PPC::PHI)
return false;
}
PredMBB = Pred1MBB;
MBBtoMoveCmp = Pred2MBB;
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
return true;
[PowerPC] eliminate partially redundant compare instruction This is a follow-on of D37211. D37211 eliminates a compare instruction if two conditional branches can be made based on the one compare instruction, e.g. if (a == 0) { ... } else if (a < 0) { ... } This patch extends this optimization to support partially redundant cases, which often happen in while loops. For example, one compare instruction is moved from the loop body into the preheader by this optimization in the following example. do { if (a == 0) dummy1(); a = func(a); } while (a > 0); Differential Revision: https://reviews.llvm.org/D38236 llvm-svn: 314390
2017-09-28 16:38:19 +08:00
}
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
return false;
}
// If multiple conditional branches are executed based on the (essentially)
// same comparison, we merge compare instructions into one and make multiple
// conditional branches on this comparison.
// For example,
// if (a == 0) { ... }
// else if (a < 0) { ... }
// can be executed by one compare and two conditional branches instead of
// two pairs of a compare and a conditional branch.
//
// This method merges two compare instructions in two MBBs and modifies the
// compare and conditional branch instructions if needed.
// For the above example, the input for this pass looks like:
// cmplwi r3, 0
// beq 0, .LBB0_3
// cmpwi r3, -1
// bgt 0, .LBB0_4
// So, before merging two compares, we need to modify these instructions as
// cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq
// beq 0, .LBB0_3
// cmpwi r3, 0 ; greather than -1 means greater or equal to 0
// bge 0, .LBB0_4
bool PPCMIPeephole::eliminateRedundantCompare(void) {
bool Simplified = false;
for (MachineBasicBlock &MBB2 : *MF) {
[PowerPC] eliminate partially redundant compare instruction This is a follow-on of D37211. D37211 eliminates a compare instruction if two conditional branches can be made based on the one compare instruction, e.g. if (a == 0) { ... } else if (a < 0) { ... } This patch extends this optimization to support partially redundant cases, which often happen in while loops. For example, one compare instruction is moved from the loop body into the preheader by this optimization in the following example. do { if (a == 0) dummy1(); a = func(a); } while (a > 0); Differential Revision: https://reviews.llvm.org/D38236 llvm-svn: 314390
2017-09-28 16:38:19 +08:00
MachineBasicBlock *MBB1 = nullptr, *MBBtoMoveCmp = nullptr;
// For fully redundant case, we select two basic blocks MBB1 and MBB2
// as an optimization target if
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
// - both MBBs end with a conditional branch,
// - MBB1 is the only predecessor of MBB2, and
// - compare does not take a physical register as a operand in both MBBs.
[PowerPC] eliminate partially redundant compare instruction This is a follow-on of D37211. D37211 eliminates a compare instruction if two conditional branches can be made based on the one compare instruction, e.g. if (a == 0) { ... } else if (a < 0) { ... } This patch extends this optimization to support partially redundant cases, which often happen in while loops. For example, one compare instruction is moved from the loop body into the preheader by this optimization in the following example. do { if (a == 0) dummy1(); a = func(a); } while (a > 0); Differential Revision: https://reviews.llvm.org/D38236 llvm-svn: 314390
2017-09-28 16:38:19 +08:00
// In this case, eligibleForCompareElimination sets MBBtoMoveCmp nullptr.
//
// As partially redundant case, we additionally handle if MBB2 has one
// additional predecessor, which has only one successor (MBB2).
// In this case, we move the compre instruction originally in MBB2 into
// MBBtoMoveCmp. This partially redundant case is typically appear by
// compiling a while loop; here, MBBtoMoveCmp is the loop preheader.
//
// Overview of CFG of related basic blocks
// Fully redundant case Partially redundant case
// -------- ---------------- --------
// | MBB1 | (w/ 2 succ) | MBBtoMoveCmp | | MBB1 | (w/ 2 succ)
// -------- ---------------- --------
// | \ (w/ 1 succ) \ | \
// | \ \ | \
// | \ |
// -------- --------
// | MBB2 | (w/ 1 pred | MBB2 | (w/ 2 pred
// -------- and 2 succ) -------- and 2 succ)
// | \ | \
// | \ | \
//
if (!eligibleForCompareElimination(MBB2, MBB1, MBBtoMoveCmp, MRI))
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
continue;
MachineInstr *BI1 = &*MBB1->getFirstInstrTerminator();
MachineInstr *CMPI1 = MRI->getVRegDef(BI1->getOperand(1).getReg());
MachineInstr *BI2 = &*MBB2.getFirstInstrTerminator();
MachineInstr *CMPI2 = MRI->getVRegDef(BI2->getOperand(1).getReg());
[PowerPC] eliminate partially redundant compare instruction This is a follow-on of D37211. D37211 eliminates a compare instruction if two conditional branches can be made based on the one compare instruction, e.g. if (a == 0) { ... } else if (a < 0) { ... } This patch extends this optimization to support partially redundant cases, which often happen in while loops. For example, one compare instruction is moved from the loop body into the preheader by this optimization in the following example. do { if (a == 0) dummy1(); a = func(a); } while (a > 0); Differential Revision: https://reviews.llvm.org/D38236 llvm-svn: 314390
2017-09-28 16:38:19 +08:00
bool IsPartiallyRedundant = (MBBtoMoveCmp != nullptr);
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
// We cannot optimize an unsupported compare opcode or
// a mix of 32-bit and 64-bit comaprisons
if (!isSupportedCmpOp(CMPI1->getOpcode()) ||
!isSupportedCmpOp(CMPI2->getOpcode()) ||
is64bitCmpOp(CMPI1->getOpcode()) != is64bitCmpOp(CMPI2->getOpcode()))
continue;
unsigned NewOpCode = 0;
unsigned NewPredicate1 = 0, NewPredicate2 = 0;
int16_t Imm1 = 0, NewImm1 = 0, Imm2 = 0, NewImm2 = 0;
[PowerPC] eliminate partially redundant compare instruction This is a follow-on of D37211. D37211 eliminates a compare instruction if two conditional branches can be made based on the one compare instruction, e.g. if (a == 0) { ... } else if (a < 0) { ... } This patch extends this optimization to support partially redundant cases, which often happen in while loops. For example, one compare instruction is moved from the loop body into the preheader by this optimization in the following example. do { if (a == 0) dummy1(); a = func(a); } while (a > 0); Differential Revision: https://reviews.llvm.org/D38236 llvm-svn: 314390
2017-09-28 16:38:19 +08:00
bool SwapOperands = false;
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
if (CMPI1->getOpcode() != CMPI2->getOpcode()) {
// Typically, unsigned comparison is used for equality check, but
// we replace it with a signed comparison if the comparison
// to be merged is a signed comparison.
// In other cases of opcode mismatch, we cannot optimize this.
if (isEqOrNe(BI2) &&
CMPI1->getOpcode() == getSignedCmpOpCode(CMPI2->getOpcode()))
NewOpCode = CMPI1->getOpcode();
else if (isEqOrNe(BI1) &&
getSignedCmpOpCode(CMPI1->getOpcode()) == CMPI2->getOpcode())
NewOpCode = CMPI2->getOpcode();
else continue;
}
if (CMPI1->getOperand(2).isReg() && CMPI2->getOperand(2).isReg()) {
// In case of comparisons between two registers, these two registers
// must be same to merge two comparisons.
[PowerPC] eliminate partially redundant compare instruction This is a follow-on of D37211. D37211 eliminates a compare instruction if two conditional branches can be made based on the one compare instruction, e.g. if (a == 0) { ... } else if (a < 0) { ... } This patch extends this optimization to support partially redundant cases, which often happen in while loops. For example, one compare instruction is moved from the loop body into the preheader by this optimization in the following example. do { if (a == 0) dummy1(); a = func(a); } while (a > 0); Differential Revision: https://reviews.llvm.org/D38236 llvm-svn: 314390
2017-09-28 16:38:19 +08:00
unsigned Cmp1Operand1 = getSrcVReg(CMPI1->getOperand(1).getReg(),
nullptr, nullptr, MRI);
unsigned Cmp1Operand2 = getSrcVReg(CMPI1->getOperand(2).getReg(),
nullptr, nullptr, MRI);
unsigned Cmp2Operand1 = getSrcVReg(CMPI2->getOperand(1).getReg(),
MBB1, &MBB2, MRI);
unsigned Cmp2Operand2 = getSrcVReg(CMPI2->getOperand(2).getReg(),
MBB1, &MBB2, MRI);
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
if (Cmp1Operand1 == Cmp2Operand1 && Cmp1Operand2 == Cmp2Operand2) {
// Same pair of registers in the same order; ready to merge as is.
}
else if (Cmp1Operand1 == Cmp2Operand2 && Cmp1Operand2 == Cmp2Operand1) {
// Same pair of registers in different order.
// We reverse the predicate to merge compare instructions.
PPC::Predicate Pred = (PPC::Predicate)BI2->getOperand(0).getImm();
NewPredicate2 = (unsigned)PPC::getSwappedPredicate(Pred);
[PowerPC] eliminate partially redundant compare instruction This is a follow-on of D37211. D37211 eliminates a compare instruction if two conditional branches can be made based on the one compare instruction, e.g. if (a == 0) { ... } else if (a < 0) { ... } This patch extends this optimization to support partially redundant cases, which often happen in while loops. For example, one compare instruction is moved from the loop body into the preheader by this optimization in the following example. do { if (a == 0) dummy1(); a = func(a); } while (a > 0); Differential Revision: https://reviews.llvm.org/D38236 llvm-svn: 314390
2017-09-28 16:38:19 +08:00
// In case of partial redundancy, we need to swap operands
// in another compare instruction.
SwapOperands = true;
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
}
else continue;
}
[trivial] fix format, NFC llvm-svn: 314769
2017-10-03 15:28:58 +08:00
else if (CMPI1->getOperand(2).isImm() && CMPI2->getOperand(2).isImm()) {
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
// In case of comparisons between a register and an immediate,
// the operand register must be same for two compare instructions.
[PowerPC] eliminate partially redundant compare instruction This is a follow-on of D37211. D37211 eliminates a compare instruction if two conditional branches can be made based on the one compare instruction, e.g. if (a == 0) { ... } else if (a < 0) { ... } This patch extends this optimization to support partially redundant cases, which often happen in while loops. For example, one compare instruction is moved from the loop body into the preheader by this optimization in the following example. do { if (a == 0) dummy1(); a = func(a); } while (a > 0); Differential Revision: https://reviews.llvm.org/D38236 llvm-svn: 314390
2017-09-28 16:38:19 +08:00
unsigned Cmp1Operand1 = getSrcVReg(CMPI1->getOperand(1).getReg(),
nullptr, nullptr, MRI);
unsigned Cmp2Operand1 = getSrcVReg(CMPI2->getOperand(1).getReg(),
MBB1, &MBB2, MRI);
if (Cmp1Operand1 != Cmp2Operand1)
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
continue;
NewImm1 = Imm1 = (int16_t)CMPI1->getOperand(2).getImm();
NewImm2 = Imm2 = (int16_t)CMPI2->getOperand(2).getImm();
// If immediate are not same, we try to adjust by changing predicate;
// e.g. GT imm means GE (imm+1).
if (Imm1 != Imm2 && (!isEqOrNe(BI2) || !isEqOrNe(BI1))) {
int Diff = Imm1 - Imm2;
if (Diff < -2 || Diff > 2)
continue;
unsigned PredToInc1 = getPredicateToIncImm(BI1, CMPI1);
unsigned PredToDec1 = getPredicateToDecImm(BI1, CMPI1);
unsigned PredToInc2 = getPredicateToIncImm(BI2, CMPI2);
unsigned PredToDec2 = getPredicateToDecImm(BI2, CMPI2);
if (Diff == 2) {
if (PredToInc2 && PredToDec1) {
NewPredicate2 = PredToInc2;
NewPredicate1 = PredToDec1;
NewImm2++;
NewImm1--;
}
}
else if (Diff == 1) {
if (PredToInc2) {
NewImm2++;
NewPredicate2 = PredToInc2;
}
else if (PredToDec1) {
NewImm1--;
NewPredicate1 = PredToDec1;
}
}
else if (Diff == -1) {
if (PredToDec2) {
NewImm2--;
NewPredicate2 = PredToDec2;
}
else if (PredToInc1) {
NewImm1++;
NewPredicate1 = PredToInc1;
}
}
else if (Diff == -2) {
if (PredToDec2 && PredToInc1) {
NewPredicate2 = PredToDec2;
NewPredicate1 = PredToInc1;
NewImm2--;
NewImm1++;
}
}
}
// We cannnot merge two compares if the immediates are not same.
if (NewImm2 != NewImm1)
continue;
}
DEBUG(dbgs() << "Optimize two pairs of compare and branch:\n");
DEBUG(CMPI1->dump());
DEBUG(BI1->dump());
DEBUG(CMPI2->dump());
DEBUG(BI2->dump());
// We adjust opcode, predicates and immediate as we determined above.
if (NewOpCode != 0 && NewOpCode != CMPI1->getOpcode()) {
CMPI1->setDesc(TII->get(NewOpCode));
}
if (NewPredicate1) {
BI1->getOperand(0).setImm(NewPredicate1);
}
if (NewPredicate2) {
BI2->getOperand(0).setImm(NewPredicate2);
}
if (NewImm1 != Imm1) {
CMPI1->getOperand(2).setImm(NewImm1);
}
[PowerPC] eliminate partially redundant compare instruction This is a follow-on of D37211. D37211 eliminates a compare instruction if two conditional branches can be made based on the one compare instruction, e.g. if (a == 0) { ... } else if (a < 0) { ... } This patch extends this optimization to support partially redundant cases, which often happen in while loops. For example, one compare instruction is moved from the loop body into the preheader by this optimization in the following example. do { if (a == 0) dummy1(); a = func(a); } while (a > 0); Differential Revision: https://reviews.llvm.org/D38236 llvm-svn: 314390
2017-09-28 16:38:19 +08:00
if (IsPartiallyRedundant) {
// We touch up the compare instruction in MBB2 and move it to
// a previous BB to handle partially redundant case.
if (SwapOperands) {
unsigned Op1 = CMPI2->getOperand(1).getReg();
unsigned Op2 = CMPI2->getOperand(2).getReg();
CMPI2->getOperand(1).setReg(Op2);
CMPI2->getOperand(2).setReg(Op1);
}
if (NewImm2 != Imm2)
CMPI2->getOperand(2).setImm(NewImm2);
for (int I = 1; I <= 2; I++) {
if (CMPI2->getOperand(I).isReg()) {
MachineInstr *Inst = MRI->getVRegDef(CMPI2->getOperand(I).getReg());
if (Inst->getParent() != &MBB2)
continue;
assert(Inst->getOpcode() == PPC::PHI &&
"We cannot support if an operand comes from this BB.");
unsigned SrcReg = getIncomingRegForBlock(Inst, MBBtoMoveCmp);
CMPI2->getOperand(I).setReg(SrcReg);
}
}
auto I = MachineBasicBlock::iterator(MBBtoMoveCmp->getFirstTerminator());
MBBtoMoveCmp->splice(I, &MBB2, MachineBasicBlock::iterator(CMPI2));
DebugLoc DL = CMPI2->getDebugLoc();
unsigned NewVReg = MRI->createVirtualRegister(&PPC::CRRCRegClass);
BuildMI(MBB2, MBB2.begin(), DL,
TII->get(PPC::PHI), NewVReg)
.addReg(BI1->getOperand(1).getReg()).addMBB(MBB1)
.addReg(BI2->getOperand(1).getReg()).addMBB(MBBtoMoveCmp);
BI2->getOperand(1).setReg(NewVReg);
}
else {
// We finally eliminate compare instruction in MBB2.
BI2->getOperand(1).setReg(BI1->getOperand(1).getReg());
CMPI2->eraseFromParent();
}
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
BI2->getOperand(1).setIsKill(true);
BI1->getOperand(1).setIsKill(false);
DEBUG(dbgs() << "into a compare and two branches:\n");
DEBUG(CMPI1->dump());
DEBUG(BI1->dump());
DEBUG(BI2->dump());
[PowerPC] eliminate partially redundant compare instruction This is a follow-on of D37211. D37211 eliminates a compare instruction if two conditional branches can be made based on the one compare instruction, e.g. if (a == 0) { ... } else if (a < 0) { ... } This patch extends this optimization to support partially redundant cases, which often happen in while loops. For example, one compare instruction is moved from the loop body into the preheader by this optimization in the following example. do { if (a == 0) dummy1(); a = func(a); } while (a > 0); Differential Revision: https://reviews.llvm.org/D38236 llvm-svn: 314390
2017-09-28 16:38:19 +08:00
if (IsPartiallyRedundant) {
DEBUG(dbgs() << "The following compare is moved into BB#" <<
MBBtoMoveCmp->getNumber() << " to handle partial redundancy.\n");
DEBUG(CMPI2->dump());
}
[PowerPC] eliminate redundant compare instruction If multiple conditional branches are executed based on the same comparison, we can execute multiple conditional branches based on the result of one comparison on PPC. For example, if (a == 0) { ... } else if (a < 0) { ... } can be executed by one compare and two conditional branches instead of two pairs of a compare and a conditional branch. This patch identifies a code sequence of the two pairs of a compare and a conditional branch and merge the compares if possible. To maximize the opportunity, we do canonicalization of code sequence before merging compares. For the above example, the input for this pass looks like: cmplwi r3, 0 beq 0, .LBB0_3 cmpwi r3, -1 bgt 0, .LBB0_4 So, before merging two compares, we canonicalize it as cmpwi r3, 0 ; cmplwi and cmpwi yield same result for beq beq 0, .LBB0_3 cmpwi r3, 0 ; greather than -1 means greater or equal to 0 bge 0, .LBB0_4 The generated code should be cmpwi r3, 0 beq 0, .LBB0_3 bge 0, .LBB0_4 Differential Revision: https://reviews.llvm.org/D37211 llvm-svn: 312514
2017-09-05 12:15:17 +08:00
Simplified = true;
}
[PowerPC] Add an MI SSA peephole pass. This patch adds a pass for doing PowerPC peephole optimizations at the MI level while the code is still in SSA form. This allows for easy modifications to the instructions while depending on a subsequent pass of DCE. Both passes are very fast due to the characteristics of SSA. At this time, the only peepholes added are for cleaning up various redundancies involving the XXPERMDI instruction. However, I would expect this will be a useful place to add more peepholes for inefficiencies generated during instruction selection. The pass is placed after VSX swap optimization, as it is best to let that pass remove unnecessary swaps before performing any remaining clean-ups. The utility of these clean-ups are demonstrated by changes to four existing test cases, all of which now have tighter expected code generation. I've also added Eric Schweiz's bugpoint-reduced test from PR25157, for which we now generate tight code. One other test started failing for me, and I've fixed it (test/Transforms/PlaceSafepoints/finite-loops.ll) as well; this is not related to my changes, and I'm not sure why it works before and not after. The problem is that the CHECK-NOT: of "statepoint" from test1 fails because of the "statepoint" in test2, and so forth. Adding a CHECK-LABEL in between keeps the different occurrences of that string properly scoped. llvm-svn: 252651
2015-11-11 05:38:26 +08:00
return Simplified;
}
// This is used to find the "true" source register for an
// XXPERMDI instruction, since MachineCSE does not handle the
// "copy-like" operations (Copy and SubregToReg). Returns
// the original SrcReg unless it is the target of a copy-like
// operation, in which case we chain backwards through all
// such operations to the ultimate source register. If a
// physical register is encountered, we stop the search.
unsigned PPCMIPeephole::lookThruCopyLike(unsigned SrcReg) {
while (true) {
MachineInstr *MI = MRI->getVRegDef(SrcReg);
if (!MI->isCopyLike())
return SrcReg;
unsigned CopySrcReg;
if (MI->isCopy())
CopySrcReg = MI->getOperand(1).getReg();
else {
assert(MI->isSubregToReg() && "bad opcode for lookThruCopyLike");
CopySrcReg = MI->getOperand(2).getReg();
}
if (!TargetRegisterInfo::isVirtualRegister(CopySrcReg))
return CopySrcReg;
SrcReg = CopySrcReg;
}
}
} // end default namespace
INITIALIZE_PASS_BEGIN(PPCMIPeephole, DEBUG_TYPE,
"PowerPC MI Peephole Optimization", false, false)
INITIALIZE_PASS_END(PPCMIPeephole, DEBUG_TYPE,
"PowerPC MI Peephole Optimization", false, false)
char PPCMIPeephole::ID = 0;
FunctionPass*
llvm::createPPCMIPeepholePass() { return new PPCMIPeephole(); }