[llvm] r301990 - [PowerPC, DAGCombiner] Fold a << (b % (sizeof(a) * 8)) back to a single instruction
Tim Shen via llvm-commits
llvm-commits at lists.llvm.org
Tue May 2 17:07:02 PDT 2017
Author: timshen
Date: Tue May 2 19:07:02 2017
New Revision: 301990
URL: http://llvm.org/viewvc/llvm-project?rev=301990&view=rev
Log:
[PowerPC, DAGCombiner] Fold a << (b % (sizeof(a) * 8)) back to a single instruction
Summary:
This is the corresponding llvm change to D28037 to ensure no performance
regression.
Reviewers: bogner, kbarton, hfinkel, iteratee, echristo
Subscribers: nemanjai, llvm-commits
Differential Revision: https://reviews.llvm.org/D28329
Modified:
llvm/trunk/include/llvm/Target/TargetLowering.h
llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
llvm/trunk/lib/Target/PowerPC/PPCISelLowering.h
llvm/trunk/test/CodeGen/PowerPC/shift_mask.ll
Modified: llvm/trunk/include/llvm/Target/TargetLowering.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/Target/TargetLowering.h?rev=301990&r1=301989&r2=301990&view=diff
==============================================================================
--- llvm/trunk/include/llvm/Target/TargetLowering.h (original)
+++ llvm/trunk/include/llvm/Target/TargetLowering.h Tue May 2 19:07:02 2017
@@ -2061,6 +2061,14 @@ public:
return false;
}
+ // Return true if the instruction that performs a << b actually performs
+ // a << (b % (sizeof(a) * 8)).
+ virtual bool supportsModuloShift(ISD::NodeType Inst, EVT ReturnType) const {
+ assert((Inst == ISD::SHL || Inst == ISD::SRA || Inst == ISD::SRL) &&
+ "Expect a shift instruction");
+ return false;
+ }
+
//===--------------------------------------------------------------------===//
// Runtime Library hooks
//
Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=301990&r1=301989&r2=301990&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Tue May 2 19:07:02 2017
@@ -5294,6 +5294,17 @@ SDValue DAGCombiner::visitSHL(SDNode *N)
}
}
+ // If the target supports masking y in (shl, y),
+ // fold (shl x, (and y, ((1 << numbits(x)) - 1))) -> (shl x, y)
+ if (TLI.isOperationLegal(ISD::SHL, VT) &&
+ TLI.supportsModuloShift(ISD::SHL, VT) && N1->getOpcode() == ISD::AND) {
+ if (ConstantSDNode *Mask = isConstOrConstSplat(N1->getOperand(1))) {
+ if (Mask->getZExtValue() == OpSizeInBits - 1) {
+ return DAG.getNode(ISD::SHL, SDLoc(N), VT, N0, N1->getOperand(0));
+ }
+ }
+ }
+
ConstantSDNode *N1C = isConstOrConstSplat(N1);
// fold (shl c1, c2) -> c1<<c2
@@ -5492,6 +5503,17 @@ SDValue DAGCombiner::visitSRA(SDNode *N)
EVT VT = N0.getValueType();
unsigned OpSizeInBits = VT.getScalarSizeInBits();
+ // If the target supports masking y in (sra, y),
+ // fold (sra x, (and y, ((1 << numbits(x)) - 1))) -> (sra x, y)
+ if (TLI.isOperationLegal(ISD::SRA, VT) &&
+ TLI.supportsModuloShift(ISD::SRA, VT) && N1->getOpcode() == ISD::AND) {
+ if (ConstantSDNode *Mask = isConstOrConstSplat(N1->getOperand(1))) {
+ if (Mask->getZExtValue() == OpSizeInBits - 1) {
+ return DAG.getNode(ISD::SRA, SDLoc(N), VT, N0, N1->getOperand(0));
+ }
+ }
+ }
+
// Arithmetic shifting an all-sign-bit value is a no-op.
if (DAG.ComputeNumSignBits(N0) == OpSizeInBits)
return N0;
@@ -5650,6 +5672,17 @@ SDValue DAGCombiner::visitSRL(SDNode *N)
EVT VT = N0.getValueType();
unsigned OpSizeInBits = VT.getScalarSizeInBits();
+ // If the target supports masking y in (srl, y),
+ // fold (srl x, (and y, ((1 << numbits(x)) - 1))) -> (srl x, y)
+ if (TLI.isOperationLegal(ISD::SRL, VT) &&
+ TLI.supportsModuloShift(ISD::SRL, VT) && N1->getOpcode() == ISD::AND) {
+ if (ConstantSDNode *Mask = isConstOrConstSplat(N1->getOperand(1))) {
+ if (Mask->getZExtValue() == OpSizeInBits - 1) {
+ return DAG.getNode(ISD::SRL, SDLoc(N), VT, N0, N1->getOperand(0));
+ }
+ }
+ }
+
// fold vector ops
if (VT.isVector())
if (SDValue FoldedVOp = SimplifyVBinOp(N))
Modified: llvm/trunk/lib/Target/PowerPC/PPCISelLowering.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPCISelLowering.h?rev=301990&r1=301989&r2=301990&view=diff
==============================================================================
--- llvm/trunk/lib/Target/PowerPC/PPCISelLowering.h (original)
+++ llvm/trunk/lib/Target/PowerPC/PPCISelLowering.h Tue May 2 19:07:02 2017
@@ -1017,6 +1017,14 @@ namespace llvm {
SDValue
combineElementTruncationToVectorTruncation(SDNode *N,
DAGCombinerInfo &DCI) const;
+
+ bool supportsModuloShift(ISD::NodeType Inst,
+ EVT ReturnType) const override {
+ assert((Inst == ISD::SHL || Inst == ISD::SRA || Inst == ISD::SRL) &&
+ "Expect a shift instruction");
+ assert(isOperationLegal(Inst, ReturnType));
+ return ReturnType.isVector();
+ }
};
namespace PPC {
Modified: llvm/trunk/test/CodeGen/PowerPC/shift_mask.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/PowerPC/shift_mask.ll?rev=301990&r1=301989&r2=301990&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/PowerPC/shift_mask.ll (original)
+++ llvm/trunk/test/CodeGen/PowerPC/shift_mask.ll Tue May 2 19:07:02 2017
@@ -49,8 +49,6 @@ define i64 @test003(i64 %a, i64 %b) {
define <16 x i8> @test010(<16 x i8> %a, <16 x i8> %b) {
; CHECK-LABEL: test010:
; CHECK: # BB#0:
-; CHECK-NEXT: vspltisb 4, 7
-; CHECK-NEXT: xxland 35, 35, 36
; CHECK-NEXT: vslb 2, 2, 3
; CHECK-NEXT: blr
%rem = and <16 x i8> %b, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
@@ -61,8 +59,6 @@ define <16 x i8> @test010(<16 x i8> %a,
define <8 x i16> @test011(<8 x i16> %a, <8 x i16> %b) {
; CHECK-LABEL: test011:
; CHECK: # BB#0:
-; CHECK-NEXT: vspltish 4, 15
-; CHECK-NEXT: xxland 35, 35, 36
; CHECK-NEXT: vslh 2, 2, 3
; CHECK-NEXT: blr
%rem = and <8 x i16> %b, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
@@ -73,10 +69,6 @@ define <8 x i16> @test011(<8 x i16> %a,
define <4 x i32> @test012(<4 x i32> %a, <4 x i32> %b) {
; CHECK-LABEL: test012:
; CHECK: # BB#0:
-; CHECK-NEXT: vspltisw 4, -16
-; CHECK-NEXT: vspltisw 5, 15
-; CHECK-NEXT: vsubuwm 4, 5, 4
-; CHECK-NEXT: xxland 35, 35, 36
; CHECK-NEXT: vslw 2, 2, 3
; CHECK-NEXT: blr
%rem = and <4 x i32> %b, <i32 31, i32 31, i32 31, i32 31>
@@ -87,11 +79,6 @@ define <4 x i32> @test012(<4 x i32> %a,
define <2 x i64> @test013(<2 x i64> %a, <2 x i64> %b) {
; CHECK-LABEL: test013:
; CHECK: # BB#0:
-; CHECK-NEXT: addis 3, 2, .LCPI7_0 at toc@ha
-; CHECK-NEXT: addi 3, 3, .LCPI7_0 at toc@l
-; CHECK-NEXT: lxvd2x 0, 0, 3
-; CHECK-NEXT: xxswapd 36, 0
-; CHECK-NEXT: xxland 35, 35, 36
; CHECK-NEXT: vsld 2, 2, 3
; CHECK-NEXT: blr
%rem = and <2 x i64> %b, <i64 63, i64 63>
@@ -148,8 +135,6 @@ define i64 @test103(i64 %a, i64 %b) {
define <16 x i8> @test110(<16 x i8> %a, <16 x i8> %b) {
; CHECK-LABEL: test110:
; CHECK: # BB#0:
-; CHECK-NEXT: vspltisb 4, 7
-; CHECK-NEXT: xxland 35, 35, 36
; CHECK-NEXT: vsrb 2, 2, 3
; CHECK-NEXT: blr
%rem = and <16 x i8> %b, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
@@ -160,8 +145,6 @@ define <16 x i8> @test110(<16 x i8> %a,
define <8 x i16> @test111(<8 x i16> %a, <8 x i16> %b) {
; CHECK-LABEL: test111:
; CHECK: # BB#0:
-; CHECK-NEXT: vspltish 4, 15
-; CHECK-NEXT: xxland 35, 35, 36
; CHECK-NEXT: vsrh 2, 2, 3
; CHECK-NEXT: blr
%rem = and <8 x i16> %b, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
@@ -172,10 +155,6 @@ define <8 x i16> @test111(<8 x i16> %a,
define <4 x i32> @test112(<4 x i32> %a, <4 x i32> %b) {
; CHECK-LABEL: test112:
; CHECK: # BB#0:
-; CHECK-NEXT: vspltisw 4, -16
-; CHECK-NEXT: vspltisw 5, 15
-; CHECK-NEXT: vsubuwm 4, 5, 4
-; CHECK-NEXT: xxland 35, 35, 36
; CHECK-NEXT: vsrw 2, 2, 3
; CHECK-NEXT: blr
%rem = and <4 x i32> %b, <i32 31, i32 31, i32 31, i32 31>
@@ -186,11 +165,6 @@ define <4 x i32> @test112(<4 x i32> %a,
define <2 x i64> @test113(<2 x i64> %a, <2 x i64> %b) {
; CHECK-LABEL: test113:
; CHECK: # BB#0:
-; CHECK-NEXT: addis 3, 2, .LCPI15_0 at toc@ha
-; CHECK-NEXT: addi 3, 3, .LCPI15_0 at toc@l
-; CHECK-NEXT: lxvd2x 0, 0, 3
-; CHECK-NEXT: xxswapd 36, 0
-; CHECK-NEXT: xxland 35, 35, 36
; CHECK-NEXT: vsrd 2, 2, 3
; CHECK-NEXT: blr
%rem = and <2 x i64> %b, <i64 63, i64 63>
@@ -247,8 +221,6 @@ define i64 @test203(i64 %a, i64 %b) {
define <16 x i8> @test210(<16 x i8> %a, <16 x i8> %b) {
; CHECK-LABEL: test210:
; CHECK: # BB#0:
-; CHECK-NEXT: vspltisb 4, 7
-; CHECK-NEXT: xxland 35, 35, 36
; CHECK-NEXT: vsrab 2, 2, 3
; CHECK-NEXT: blr
%rem = and <16 x i8> %b, <i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7>
@@ -259,8 +231,6 @@ define <16 x i8> @test210(<16 x i8> %a,
define <8 x i16> @test211(<8 x i16> %a, <8 x i16> %b) {
; CHECK-LABEL: test211:
; CHECK: # BB#0:
-; CHECK-NEXT: vspltish 4, 15
-; CHECK-NEXT: xxland 35, 35, 36
; CHECK-NEXT: vsrah 2, 2, 3
; CHECK-NEXT: blr
%rem = and <8 x i16> %b, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
@@ -271,10 +241,6 @@ define <8 x i16> @test211(<8 x i16> %a,
define <4 x i32> @test212(<4 x i32> %a, <4 x i32> %b) {
; CHECK-LABEL: test212:
; CHECK: # BB#0:
-; CHECK-NEXT: vspltisw 4, -16
-; CHECK-NEXT: vspltisw 5, 15
-; CHECK-NEXT: vsubuwm 4, 5, 4
-; CHECK-NEXT: xxland 35, 35, 36
; CHECK-NEXT: vsraw 2, 2, 3
; CHECK-NEXT: blr
%rem = and <4 x i32> %b, <i32 31, i32 31, i32 31, i32 31>
@@ -285,11 +251,6 @@ define <4 x i32> @test212(<4 x i32> %a,
define <2 x i64> @test213(<2 x i64> %a, <2 x i64> %b) {
; CHECK-LABEL: test213:
; CHECK: # BB#0:
-; CHECK-NEXT: addis 3, 2, .LCPI23_0 at toc@ha
-; CHECK-NEXT: addi 3, 3, .LCPI23_0 at toc@l
-; CHECK-NEXT: lxvd2x 0, 0, 3
-; CHECK-NEXT: xxswapd 36, 0
-; CHECK-NEXT: xxland 35, 35, 36
; CHECK-NEXT: vsrad 2, 2, 3
; CHECK-NEXT: blr
%rem = and <2 x i64> %b, <i64 63, i64 63>
More information about the llvm-commits
mailing list