1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 03:02:36 +01:00

[InstCombine] Add sext(ashr(shl(trunc(x),c),c)) folding support for vectors

Replacing m_ConstantInt with m_Constant permits folding of vectors as well as scalars.

Differential Revision: https://reviews.llvm.org/D83058
This commit is contained in:
Simon Pilgrim 2020-07-02 19:24:38 +01:00
parent fdbe62d99b
commit 8257854bda
3 changed files with 20 additions and 28 deletions

View File

@ -1460,16 +1460,17 @@ Instruction *InstCombiner::visitSExt(SExtInst &CI) {
// %d = ashr i32 %a, 30 // %d = ashr i32 %a, 30
Value *A = nullptr; Value *A = nullptr;
// TODO: Eventually this could be subsumed by EvaluateInDifferentType. // TODO: Eventually this could be subsumed by EvaluateInDifferentType.
ConstantInt *BA = nullptr, *CA = nullptr; Constant *BA = nullptr, *CA = nullptr;
if (match(Src, m_AShr(m_Shl(m_Trunc(m_Value(A)), m_ConstantInt(BA)), if (match(Src, m_AShr(m_Shl(m_Trunc(m_Value(A)), m_Constant(BA)),
m_ConstantInt(CA))) && m_Constant(CA))) &&
BA == CA && A->getType() == CI.getType()) { BA == CA && A->getType() == CI.getType()) {
unsigned MidSize = Src->getType()->getScalarSizeInBits(); unsigned MidSize = Src->getType()->getScalarSizeInBits();
unsigned SrcDstSize = CI.getType()->getScalarSizeInBits(); unsigned SrcDstSize = CI.getType()->getScalarSizeInBits();
unsigned ShAmt = CA->getZExtValue()+SrcDstSize-MidSize; Constant *SizeDiff = ConstantInt::get(CA->getType(), SrcDstSize - MidSize);
Constant *ShAmtV = ConstantInt::get(CI.getType(), ShAmt); Constant *ShAmt = ConstantExpr::getAdd(CA, SizeDiff);
A = Builder.CreateShl(A, ShAmtV, CI.getName()); Constant *ShAmtExt = ConstantExpr::getSExt(ShAmt, CI.getType());
return BinaryOperator::CreateAShr(A, ShAmtV); A = Builder.CreateShl(A, ShAmtExt, CI.getName());
return BinaryOperator::CreateAShr(A, ShAmtExt);
} }
return nullptr; return nullptr;

View File

@ -140,10 +140,8 @@ define i32 @test10(i32 %i) {
define <2 x i32> @test10_vec(<2 x i32> %i) { define <2 x i32> @test10_vec(<2 x i32> %i) {
; CHECK-LABEL: @test10_vec( ; CHECK-LABEL: @test10_vec(
; CHECK-NEXT: [[A:%.*]] = trunc <2 x i32> [[I:%.*]] to <2 x i8> ; CHECK-NEXT: [[D1:%.*]] = shl <2 x i32> [[I:%.*]], <i32 30, i32 30>
; CHECK-NEXT: [[B:%.*]] = shl <2 x i8> [[A]], <i8 6, i8 6> ; CHECK-NEXT: [[D:%.*]] = ashr exact <2 x i32> [[D1]], <i32 30, i32 30>
; CHECK-NEXT: [[C:%.*]] = ashr exact <2 x i8> [[B]], <i8 6, i8 6>
; CHECK-NEXT: [[D:%.*]] = sext <2 x i8> [[C]] to <2 x i32>
; CHECK-NEXT: ret <2 x i32> [[D]] ; CHECK-NEXT: ret <2 x i32> [[D]]
; ;
%A = trunc <2 x i32> %i to <2 x i8> %A = trunc <2 x i32> %i to <2 x i8>
@ -155,10 +153,8 @@ define <2 x i32> @test10_vec(<2 x i32> %i) {
define <2 x i32> @test10_vec_nonuniform(<2 x i32> %i) { define <2 x i32> @test10_vec_nonuniform(<2 x i32> %i) {
; CHECK-LABEL: @test10_vec_nonuniform( ; CHECK-LABEL: @test10_vec_nonuniform(
; CHECK-NEXT: [[A:%.*]] = trunc <2 x i32> [[I:%.*]] to <2 x i8> ; CHECK-NEXT: [[D1:%.*]] = shl <2 x i32> [[I:%.*]], <i32 30, i32 27>
; CHECK-NEXT: [[B:%.*]] = shl <2 x i8> [[A]], <i8 6, i8 3> ; CHECK-NEXT: [[D:%.*]] = ashr <2 x i32> [[D1]], <i32 30, i32 27>
; CHECK-NEXT: [[C:%.*]] = ashr <2 x i8> [[B]], <i8 6, i8 3>
; CHECK-NEXT: [[D:%.*]] = sext <2 x i8> [[C]] to <2 x i32>
; CHECK-NEXT: ret <2 x i32> [[D]] ; CHECK-NEXT: ret <2 x i32> [[D]]
; ;
%A = trunc <2 x i32> %i to <2 x i8> %A = trunc <2 x i32> %i to <2 x i8>
@ -170,10 +166,8 @@ define <2 x i32> @test10_vec_nonuniform(<2 x i32> %i) {
define <2 x i32> @test10_vec_undef(<2 x i32> %i) { define <2 x i32> @test10_vec_undef(<2 x i32> %i) {
; CHECK-LABEL: @test10_vec_undef( ; CHECK-LABEL: @test10_vec_undef(
; CHECK-NEXT: [[A:%.*]] = trunc <2 x i32> [[I:%.*]] to <2 x i8> ; CHECK-NEXT: [[D1:%.*]] = shl <2 x i32> [[I:%.*]], <i32 30, i32 0>
; CHECK-NEXT: [[B:%.*]] = shl <2 x i8> [[A]], <i8 6, i8 undef> ; CHECK-NEXT: [[D:%.*]] = ashr <2 x i32> [[D1]], <i32 30, i32 0>
; CHECK-NEXT: [[C:%.*]] = ashr <2 x i8> [[B]], <i8 6, i8 undef>
; CHECK-NEXT: [[D:%.*]] = sext <2 x i8> [[C]] to <2 x i32>
; CHECK-NEXT: ret <2 x i32> [[D]] ; CHECK-NEXT: ret <2 x i32> [[D]]
; ;
%A = trunc <2 x i32> %i to <2 x i8> %A = trunc <2 x i32> %i to <2 x i8>

View File

@ -82,9 +82,8 @@ define i64 @test2(i64 %a) {
define <2 x i64> @test2_vec(<2 x i64> %a) { define <2 x i64> @test2_vec(<2 x i64> %a) {
; CHECK-LABEL: @test2_vec( ; CHECK-LABEL: @test2_vec(
; CHECK-NEXT: [[B:%.*]] = trunc <2 x i64> [[A:%.*]] to <2 x i32> ; CHECK-NEXT: [[B:%.*]] = trunc <2 x i64> [[A:%.*]] to <2 x i32>
; CHECK-NEXT: [[C:%.*]] = shl <2 x i32> [[B]], <i32 4, i32 4> ; CHECK-NEXT: [[D1:%.*]] = shl <2 x i64> [[A]], <i64 36, i64 36>
; CHECK-NEXT: [[Q:%.*]] = ashr exact <2 x i32> [[C]], <i32 4, i32 4> ; CHECK-NEXT: [[D:%.*]] = ashr exact <2 x i64> [[D1]], <i64 36, i64 36>
; CHECK-NEXT: [[D:%.*]] = sext <2 x i32> [[Q]] to <2 x i64>
; CHECK-NEXT: call void @use_vec(<2 x i32> [[B]]) ; CHECK-NEXT: call void @use_vec(<2 x i32> [[B]])
; CHECK-NEXT: ret <2 x i64> [[D]] ; CHECK-NEXT: ret <2 x i64> [[D]]
; ;
@ -99,9 +98,8 @@ define <2 x i64> @test2_vec(<2 x i64> %a) {
define <2 x i64> @test2_vec_nonuniform(<2 x i64> %a) { define <2 x i64> @test2_vec_nonuniform(<2 x i64> %a) {
; CHECK-LABEL: @test2_vec_nonuniform( ; CHECK-LABEL: @test2_vec_nonuniform(
; CHECK-NEXT: [[B:%.*]] = trunc <2 x i64> [[A:%.*]] to <2 x i32> ; CHECK-NEXT: [[B:%.*]] = trunc <2 x i64> [[A:%.*]] to <2 x i32>
; CHECK-NEXT: [[C:%.*]] = shl <2 x i32> [[B]], <i32 4, i32 5> ; CHECK-NEXT: [[D1:%.*]] = shl <2 x i64> [[A]], <i64 36, i64 37>
; CHECK-NEXT: [[Q:%.*]] = ashr <2 x i32> [[C]], <i32 4, i32 5> ; CHECK-NEXT: [[D:%.*]] = ashr <2 x i64> [[D1]], <i64 36, i64 37>
; CHECK-NEXT: [[D:%.*]] = sext <2 x i32> [[Q]] to <2 x i64>
; CHECK-NEXT: call void @use_vec(<2 x i32> [[B]]) ; CHECK-NEXT: call void @use_vec(<2 x i32> [[B]])
; CHECK-NEXT: ret <2 x i64> [[D]] ; CHECK-NEXT: ret <2 x i64> [[D]]
; ;
@ -116,9 +114,8 @@ define <2 x i64> @test2_vec_nonuniform(<2 x i64> %a) {
define <2 x i64> @test2_vec_undef(<2 x i64> %a) { define <2 x i64> @test2_vec_undef(<2 x i64> %a) {
; CHECK-LABEL: @test2_vec_undef( ; CHECK-LABEL: @test2_vec_undef(
; CHECK-NEXT: [[B:%.*]] = trunc <2 x i64> [[A:%.*]] to <2 x i32> ; CHECK-NEXT: [[B:%.*]] = trunc <2 x i64> [[A:%.*]] to <2 x i32>
; CHECK-NEXT: [[C:%.*]] = shl <2 x i32> [[B]], <i32 4, i32 undef> ; CHECK-NEXT: [[D1:%.*]] = shl <2 x i64> [[A]], <i64 36, i64 0>
; CHECK-NEXT: [[Q:%.*]] = ashr <2 x i32> [[C]], <i32 4, i32 undef> ; CHECK-NEXT: [[D:%.*]] = ashr <2 x i64> [[D1]], <i64 36, i64 0>
; CHECK-NEXT: [[D:%.*]] = sext <2 x i32> [[Q]] to <2 x i64>
; CHECK-NEXT: call void @use_vec(<2 x i32> [[B]]) ; CHECK-NEXT: call void @use_vec(<2 x i32> [[B]])
; CHECK-NEXT: ret <2 x i64> [[D]] ; CHECK-NEXT: ret <2 x i64> [[D]]
; ;