maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[AArch64][SVE] Implement ptrue intrinsic 

Reviewers: sdesmalen, eli.friedman, dancgr, mgudim, cameron.mcinally,
huntergr, efriedma

Reviewed By: sdesmalen

Subscribers: tschuett, kristof.beyls, hiraditya, rkruppe, psnobl,
llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D71457
This commit is contained in:
Cullen Rhodes 2019-12-12 15:24:30 +00:00
parent ac3f9e4842
commit 49199465a3
5 changed files with 73 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
llvm/include/llvm/IR/IntrinsicsAArch64.td
View File

@ -923,6 +923,11 @@ let TargetPrefix = "aarch64" in { // All intrinsics start with "llvm.aarch64.".
LLVMVectorElementType<0>],
[IntrNoMem]>;
class AdvSIMD_SVE_PTRUE_Intrinsic
: Intrinsic<[llvm_anyvector_ty],
[llvm_i32_ty],
[IntrNoMem, ImmArg<0>]>;
class AdvSIMD_SVE_PUNPKHI_Intrinsic
: Intrinsic<[LLVMHalfElementsVectorType<0>],
[llvm_anyvector_ty],
@ -951,7 +956,7 @@ let TargetPrefix = "aarch64" in { // All intrinsics start with "llvm.aarch64.".
class AdvSIMD_SVE_CNTB_Intrinsic
: Intrinsic<[llvm_i64_ty],
[llvm_i32_ty],
[IntrNoMem]>;
[IntrNoMem, ImmArg<0>]>;
class AdvSIMD_SVE_CNTP_Intrinsic
: Intrinsic<[llvm_i64_ty],
@ -1426,6 +1431,12 @@ def int_aarch64_sve_ucvtf_f16i64 : Builtin_SVCVT<"svcvt_f16_u64_m", llvm_nxv8
def int_aarch64_sve_ucvtf_f32i64 : Builtin_SVCVT<"svcvt_f32_u64_m", llvm_nxv4f32_ty, llvm_nxv2i64_ty>;
def int_aarch64_sve_ucvtf_f64i32 : Builtin_SVCVT<"svcvt_f64_u32_m", llvm_nxv2f64_ty, llvm_nxv4i32_ty>;
//
// Predicate creation
//
def int_aarch64_sve_ptrue : AdvSIMD_SVE_PTRUE_Intrinsic;
//
// Predicate operations
//

4
llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
View File

@ -1347,6 +1347,7 @@ const char *AArch64TargetLowering::getTargetNodeName(unsigned Opcode) const {
case AArch64ISD::UUNPKHI: return "AArch64ISD::UUNPKHI";
case AArch64ISD::UUNPKLO: return "AArch64ISD::UUNPKLO";
case AArch64ISD::INSR: return "AArch64ISD::INSR";
case AArch64ISD::PTRUE: return "AArch64ISD::PTRUE";
case AArch64ISD::GLD1: return "AArch64ISD::GLD1";
case AArch64ISD::GLD1_SCALED: return "AArch64ISD::GLD1_SCALED";
case AArch64ISD::GLD1_SXTW: return "AArch64ISD::GLD1_SXTW";
@ -2921,6 +2922,9 @@ SDValue AArch64TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
case Intrinsic::aarch64_sve_uunpklo:
return DAG.getNode(AArch64ISD::UUNPKLO, dl, Op.getValueType(),
Op.getOperand(1));
case Intrinsic::aarch64_sve_ptrue:
return DAG.getNode(AArch64ISD::PTRUE, dl, Op.getValueType(),
Op.getOperand(1));
case Intrinsic::aarch64_sve_insr: {
SDValue Scalar = Op.getOperand(2);

1
llvm/lib/Target/AArch64/AArch64ISelLowering.h
View File

@ -205,6 +205,7 @@ enum NodeType : unsigned {
UUNPKLO,
INSR,
PTRUE,
// Unsigned gather loads.
GLD1,

24
llvm/lib/Target/AArch64/SVEInstrFormats.td
View File

@ -18,7 +18,7 @@ def SVEPatternOperand : AsmOperandClass {
let DiagnosticType = "InvalidSVEPattern";
}
def sve_pred_enum : Operand<i32>, ImmLeaf<i32, [{
def sve_pred_enum : Operand<i32>, TImmLeaf<i32, [{
return (((uint32_t)Imm) < 32);
}]> {
@ -249,11 +249,12 @@ def sve_incdec_imm : Operand<i32>, ImmLeaf<i32, [{
// it's important we define them first.
//===----------------------------------------------------------------------===//
class sve_int_ptrue<bits<2> sz8_64, bits<3> opc, string asm, PPRRegOp pprty>
class sve_int_ptrue<bits<2> sz8_64, bits<3> opc, string asm, PPRRegOp pprty,
ValueType vt, SDPatternOperator op>
: I<(outs pprty:$Pd), (ins sve_pred_enum:$pattern),
asm, "\t$Pd, $pattern",
"",
[]>, Sched<[]> {
[(set (vt pprty:$Pd), (op sve_pred_enum:$pattern))]>, Sched<[]> {
bits<4> Pd;
bits<5> pattern;
let Inst{31-24} = 0b00100101;
@ -269,11 +270,11 @@ class sve_int_ptrue<bits<2> sz8_64, bits<3> opc, string asm, PPRRegOp pprty>
let Defs = !if(!eq (opc{0}, 1), [NZCV], []);
}
multiclass sve_int_ptrue<bits<3> opc, string asm> {
def _B : sve_int_ptrue<0b00, opc, asm, PPR8>;
def _H : sve_int_ptrue<0b01, opc, asm, PPR16>;
def _S : sve_int_ptrue<0b10, opc, asm, PPR32>;
def _D : sve_int_ptrue<0b11, opc, asm, PPR64>;
multiclass sve_int_ptrue<bits<3> opc, string asm, SDPatternOperator op> {
def _B : sve_int_ptrue<0b00, opc, asm, PPR8, nxv16i1, op>;
def _H : sve_int_ptrue<0b01, opc, asm, PPR16, nxv8i1, op>;
def _S : sve_int_ptrue<0b10, opc, asm, PPR32, nxv4i1, op>;
def _D : sve_int_ptrue<0b11, opc, asm, PPR64, nxv2i1, op>;
def : InstAlias<asm # "\t$Pd",
(!cast<Instruction>(NAME # _B) PPR8:$Pd, 0b11111), 1>;
@ -285,9 +286,12 @@ multiclass sve_int_ptrue<bits<3> opc, string asm> {
(!cast<Instruction>(NAME # _D) PPR64:$Pd, 0b11111), 1>;
}
def SDT_AArch64PTrue : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
def AArch64ptrue : SDNode<"AArch64ISD::PTRUE", SDT_AArch64PTrue>;
let Predicates = [HasSVE] in {
defm PTRUE : sve_int_ptrue<0b000, "ptrue">;
defm PTRUES : sve_int_ptrue<0b001, "ptrues">;
defm PTRUE : sve_int_ptrue<0b000, "ptrue", AArch64ptrue>;
defm PTRUES : sve_int_ptrue<0b001, "ptrues", null_frag>;
}
//===----------------------------------------------------------------------===//

42
llvm/test/CodeGen/AArch64/sve-intrinsics-pred-creation.ll Normal file
View File

@ -0,0 +1,42 @@
; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s
;
; PTRUE
;
define <vscale x 16 x i1> @ptrue_b8() {
; CHECK-LABEL: ptrue_b8:
; CHECK: ptrue p0.b, pow2
; CHECK-NEXT: ret
%out = call <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 0)
ret <vscale x 16 x i1> %out
}
define <vscale x 8 x i1> @ptrue_b16() {
; CHECK-LABEL: ptrue_b16:
; CHECK: ptrue p0.h, vl1
; CHECK-NEXT: ret
%out = call <vscale x 8 x i1> @llvm.aarch64.sve.ptrue.nxv8i1(i32 1)
ret <vscale x 8 x i1> %out
}
define <vscale x 4 x i1> @ptrue_b32() {
; CHECK-LABEL: ptrue_b32:
; CHECK: ptrue p0.s, mul3
; CHECK-NEXT: ret
%out = call <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 30)
ret <vscale x 4 x i1> %out
}
define <vscale x 2 x i1> @ptrue_b64() {
; CHECK-LABEL: ptrue_b64:
; CHECK: ptrue p0.d
; CHECK-NEXT: ret
%out = call <vscale x 2 x i1> @llvm.aarch64.sve.ptrue.nxv2i1(i32 31)
ret <vscale x 2 x i1> %out
}
declare <vscale x 16 x i1> @llvm.aarch64.sve.ptrue.nxv16i1(i32 %pattern)
declare <vscale x 8 x i1> @llvm.aarch64.sve.ptrue.nxv8i1(i32 %pattern)
declare <vscale x 4 x i1> @llvm.aarch64.sve.ptrue.nxv4i1(i32 %pattern)
declare <vscale x 2 x i1> @llvm.aarch64.sve.ptrue.nxv2i1(i32 %pattern)