shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[Reland] Adjust bit cast instruction filter for DXIL Prepare pass (#143783)

Browse Source 
Relands https://github.com/llvm/llvm-project/pull/142678, with a new
change to remove an unnecessary gep argument, after a revert was needed
due to unforeseen bugs.
Fixes https://github.com/llvm/llvm-project/issues/139013
This commit is contained in:
Joshua Batista 2025-06-17 11:46:12 -07:00 committed by GitHub
parent 1f10c6a277
commit b59d4cf054
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 121 additions and 40 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

44
llvm/lib/Target/DirectX/DXILPrepare.cpp
View File

@ -148,9 +148,49 @@ class DXILPrepareModule : public ModulePass {
Type *Ty) { Type *Ty) {
// Omit bitcasts if the incoming value matches the instruction type. // Omit bitcasts if the incoming value matches the instruction type.
auto It = PointerTypes.find(Operand); auto It = PointerTypes.find(Operand);
if (It != PointerTypes.end()) if (It != PointerTypes.end()) {
if (cast<TypedPointerType>(It->second)->getElementType() == Ty) auto *OpTy = cast<TypedPointerType>(It->second)->getElementType();
if (OpTy == Ty)
return nullptr; return nullptr;
}
Type *ValTy = Operand->getType();
// Also omit the bitcast for matching global array types
if (auto *GlobalVar = dyn_cast<GlobalVariable>(Operand))
ValTy = GlobalVar->getValueType();
if (auto *AI = dyn_cast<AllocaInst>(Operand))
ValTy = AI->getAllocatedType();
if (auto *ArrTy = dyn_cast<ArrayType>(ValTy)) {
Type *ElTy = ArrTy->getElementType();
if (ElTy == Ty)
return nullptr;
}
// finally, drill down GEP instructions until we get the array
// that is being accessed, and compare element types
if (ConstantExpr *GEPInstr = dyn_cast<ConstantExpr>(Operand)) {
while (GEPInstr->getOpcode() == Instruction::GetElementPtr) {
Value *OpArg = GEPInstr->getOperand(0);
if (ConstantExpr *NewGEPInstr = dyn_cast<ConstantExpr>(OpArg)) {
GEPInstr = NewGEPInstr;
continue;
}
if (auto *GlobalVar = dyn_cast<GlobalVariable>(OpArg))
ValTy = GlobalVar->getValueType();
if (auto *AI = dyn_cast<AllocaInst>(Operand))
ValTy = AI->getAllocatedType();
if (auto *ArrTy = dyn_cast<ArrayType>(ValTy)) {
Type *ElTy = ArrTy->getElementType();
if (ElTy == Ty)
return nullptr;
}
break;
}
}
// Insert bitcasts where we are removing the instruction. // Insert bitcasts where we are removing the instruction.
Builder.SetInsertPoint(&Inst); Builder.SetInsertPoint(&Inst);
// This code only gets hit in opaque-pointer mode, so the type of the // This code only gets hit in opaque-pointer mode, so the type of the

64
llvm/test/CodeGen/DirectX/llc-vector-load-scalarize.ll
View File

@ -60,19 +60,15 @@ define <4 x i32> @load_array_vec_test() #0 {
define <4 x i32> @load_vec_test() #0 { define <4 x i32> @load_vec_test() #0 {
; CHECK-LABEL: define <4 x i32> @load_vec_test( ; CHECK-LABEL: define <4 x i32> @load_vec_test(
; CHECK-SAME: ) #[[ATTR0]] { ; CHECK-SAME: ) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = bitcast ptr addrspace(3) @vecData.scalarized to ptr addrspace(3) ; CHECK-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(3) @vecData.scalarized, align 4
; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr addrspace(3) [[TMP1]], align 4 ; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @vecData.scalarized, i32 1), align 4
; CHECK-NEXT: [[TMP3:%.*]] = bitcast ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @vecData.scalarized, i32 1) to ptr addrspace(3) ; CHECK-NEXT: [[TMP3:%.*]] = load i32, ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @vecData.scalarized, i32 2), align 4
; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr addrspace(3) [[TMP3]], align 4 ; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @vecData.scalarized, i32 3), align 4
; CHECK-NEXT: [[TMP5:%.*]] = bitcast ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @vecData.scalarized, i32 2) to ptr addrspace(3) ; CHECK-NEXT: [[DOTUPTO0:%.*]] = insertelement <4 x i32> poison, i32 [[TMP1]], i32 0
; CHECK-NEXT: [[TMP6:%.*]] = load i32, ptr addrspace(3) [[TMP5]], align 4 ; CHECK-NEXT: [[DOTUPTO1:%.*]] = insertelement <4 x i32> [[DOTUPTO0]], i32 [[TMP2]], i32 1
; CHECK-NEXT: [[TMP7:%.*]] = bitcast ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @vecData.scalarized, i32 3) to ptr addrspace(3) ; CHECK-NEXT: [[DOTUPTO2:%.*]] = insertelement <4 x i32> [[DOTUPTO1]], i32 [[TMP3]], i32 2
; CHECK-NEXT: [[TMP8:%.*]] = load i32, ptr addrspace(3) [[TMP7]], align 4 ; CHECK-NEXT: [[TMP5:%.*]] = insertelement <4 x i32> [[DOTUPTO2]], i32 [[TMP4]], i32 3
; CHECK-NEXT: [[DOTUPTO0:%.*]] = insertelement <4 x i32> poison, i32 [[TMP2]], i32 0 ; CHECK-NEXT: ret <4 x i32> [[TMP5]]
; CHECK-NEXT: [[DOTUPTO1:%.*]] = insertelement <4 x i32> [[DOTUPTO0]], i32 [[TMP4]], i32 1
; CHECK-NEXT: [[DOTUPTO2:%.*]] = insertelement <4 x i32> [[DOTUPTO1]], i32 [[TMP6]], i32 2
; CHECK-NEXT: [[TMP9:%.*]] = insertelement <4 x i32> [[DOTUPTO2]], i32 [[TMP8]], i32 3
; CHECK-NEXT: ret <4 x i32> [[TMP9]]
; ;
%1 = load <4 x i32>, <4 x i32> addrspace(3)* @"vecData", align 4 %1 = load <4 x i32>, <4 x i32> addrspace(3)* @"vecData", align 4
ret <4 x i32> %1 ret <4 x i32> %1
@ -103,31 +99,23 @@ define <4 x i32> @load_static_array_of_vec_test(i32 %index) #0 {
define <4 x i32> @multid_load_test() #0 { define <4 x i32> @multid_load_test() #0 {
; CHECK-LABEL: define <4 x i32> @multid_load_test( ; CHECK-LABEL: define <4 x i32> @multid_load_test(
; CHECK-SAME: ) #[[ATTR0]] { ; CHECK-SAME: ) #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = bitcast ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim to ptr addrspace(3) ; CHECK-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, align 4
; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr addrspace(3) [[TMP1]], align 4 ; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 1), align 4
; CHECK-NEXT: [[TMP3:%.*]] = bitcast ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 1) to ptr addrspace(3) ; CHECK-NEXT: [[TMP3:%.*]] = load i32, ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 2), align 4
; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr addrspace(3) [[TMP3]], align 4 ; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 3), align 4
; CHECK-NEXT: [[TMP5:%.*]] = bitcast ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 2) to ptr addrspace(3) ; CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr addrspace(3) getelementptr inbounds ([36 x i32], ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 0, i32 4), align 4
; CHECK-NEXT: [[TMP6:%.*]] = load i32, ptr addrspace(3) [[TMP5]], align 4 ; CHECK-NEXT: [[DOTI13:%.*]] = load i32, ptr addrspace(3) getelementptr (i32, ptr addrspace(3) getelementptr inbounds ([36 x i32], ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 0, i32 4), i32 1), align 4
; CHECK-NEXT: [[TMP7:%.*]] = bitcast ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 3) to ptr addrspace(3) ; CHECK-NEXT: [[DOTI25:%.*]] = load i32, ptr addrspace(3) getelementptr (i32, ptr addrspace(3) getelementptr inbounds ([36 x i32], ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 0, i32 4), i32 2), align 4
; CHECK-NEXT: [[TMP8:%.*]] = load i32, ptr addrspace(3) [[TMP7]], align 4 ; CHECK-NEXT: [[DOTI37:%.*]] = load i32, ptr addrspace(3) getelementptr (i32, ptr addrspace(3) getelementptr inbounds ([36 x i32], ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 0, i32 4), i32 3), align 4
; CHECK-NEXT: [[TMP9:%.*]] = bitcast ptr addrspace(3) getelementptr inbounds ([36 x i32], ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 0, i32 4) to ptr addrspace(3) ; CHECK-NEXT: [[DOTI08:%.*]] = add i32 [[TMP1]], [[TMP5]]
; CHECK-NEXT: [[TMP10:%.*]] = load i32, ptr addrspace(3) [[TMP9]], align 4 ; CHECK-NEXT: [[DOTI19:%.*]] = add i32 [[TMP2]], [[DOTI13]]
; CHECK-NEXT: [[TMP11:%.*]] = bitcast ptr addrspace(3) getelementptr (i32, ptr addrspace(3) getelementptr inbounds ([36 x i32], ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 0, i32 4), i32 1) to ptr addrspace(3) ; CHECK-NEXT: [[DOTI210:%.*]] = add i32 [[TMP3]], [[DOTI25]]
; CHECK-NEXT: [[TMP12:%.*]] = load i32, ptr addrspace(3) [[TMP11]], align 4 ; CHECK-NEXT: [[DOTI311:%.*]] = add i32 [[TMP4]], [[DOTI37]]
; CHECK-NEXT: [[TMP13:%.*]] = bitcast ptr addrspace(3) getelementptr (i32, ptr addrspace(3) getelementptr inbounds ([36 x i32], ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 0, i32 4), i32 2) to ptr addrspace(3) ; CHECK-NEXT: [[DOTUPTO015:%.*]] = insertelement <4 x i32> poison, i32 [[DOTI08]], i32 0
; CHECK-NEXT: [[TMP14:%.*]] = load i32, ptr addrspace(3) [[TMP13]], align 4 ; CHECK-NEXT: [[DOTUPTO116:%.*]] = insertelement <4 x i32> [[DOTUPTO015]], i32 [[DOTI19]], i32 1
; CHECK-NEXT: [[TMP15:%.*]] = bitcast ptr addrspace(3) getelementptr (i32, ptr addrspace(3) getelementptr inbounds ([36 x i32], ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 0, i32 4), i32 3) to ptr addrspace(3) ; CHECK-NEXT: [[DOTUPTO217:%.*]] = insertelement <4 x i32> [[DOTUPTO116]], i32 [[DOTI210]], i32 2
; CHECK-NEXT: [[TMP16:%.*]] = load i32, ptr addrspace(3) [[TMP15]], align 4 ; CHECK-NEXT: [[TMP6:%.*]] = insertelement <4 x i32> [[DOTUPTO217]], i32 [[DOTI311]], i32 3
; CHECK-NEXT: [[DOTI05:%.*]] = add i32 [[TMP2]], [[TMP10]] ; CHECK-NEXT: ret <4 x i32> [[TMP6]]
; CHECK-NEXT: [[DOTI16:%.*]] = add i32 [[TMP4]], [[TMP12]]
; CHECK-NEXT: [[DOTI27:%.*]] = add i32 [[TMP6]], [[TMP14]]
; CHECK-NEXT: [[DOTI38:%.*]] = add i32 [[TMP8]], [[TMP16]]
; CHECK-NEXT: [[DOTUPTO01215:%.*]] = insertelement <4 x i32> poison, i32 [[DOTI05]], i32 0
; CHECK-NEXT: [[DOTUPTO11316:%.*]] = insertelement <4 x i32> [[DOTUPTO01215]], i32 [[DOTI16]], i32 1
; CHECK-NEXT: [[DOTUPTO21417:%.*]] = insertelement <4 x i32> [[DOTUPTO11316]], i32 [[DOTI27]], i32 2
; CHECK-NEXT: [[TMP17:%.*]] = insertelement <4 x i32> [[DOTUPTO21417]], i32 [[DOTI38]], i32 3
; CHECK-NEXT: ret <4 x i32> [[TMP17]]
; ;
%1 = load <4 x i32>, <4 x i32> addrspace(3)* getelementptr inbounds ([3 x [3 x <4 x i32>]], [3 x [3 x <4 x i32>]] addrspace(3)* @"groushared2dArrayofVectors", i32 0, i32 0, i32 0), align 4 %1 = load <4 x i32>, <4 x i32> addrspace(3)* getelementptr inbounds ([3 x [3 x <4 x i32>]], [3 x [3 x <4 x i32>]] addrspace(3)* @"groushared2dArrayofVectors", i32 0, i32 0, i32 0), align 4
%2 = load <4 x i32>, <4 x i32> addrspace(3)* getelementptr inbounds ([3 x [3 x <4 x i32>]], [3 x [3 x <4 x i32>]] addrspace(3)* @"groushared2dArrayofVectors", i32 0, i32 1, i32 1), align 4 %2 = load <4 x i32>, <4 x i32> addrspace(3)* getelementptr inbounds ([3 x [3 x <4 x i32>]], [3 x [3 x <4 x i32>]] addrspace(3)* @"groushared2dArrayofVectors", i32 0, i32 1, i32 1), align 4

53
llvm/test/CodeGen/DirectX/noop_bitcast_global_array_type.ll Normal file
View File

@ -0,0 +1,53 @@
; RUN: opt -S --dxil-prepare %s | FileCheck %s
; Test that global arrays do not get a bitcast instruction
; after the dxil-prepare pass.
target triple = "dxilv1.2-unknown-shadermodel6.2-compute"
@inputTile.1dim = local_unnamed_addr addrspace(3) global [3 x float] zeroinitializer, align 2
; CHECK-LABEL: testload
define float @testload() local_unnamed_addr {
; NOTE: this would be "bitcast ptr addrspace(3)..." before the change that introduced this test,
; after the dxil-prepare pass is run
; CHECK-NEXT: load float, ptr addrspace(3) @inputTile.1dim, align 2
%v = load float, ptr addrspace(3) @inputTile.1dim, align 2
ret float %v
}
; CHECK-LABEL: teststore
define void @teststore() local_unnamed_addr {
; CHECK-next: store float 2.000000e+00, ptr addrspace(3) @inputTile.1dim, align 2
store float 2.000000e+00, ptr addrspace(3) @inputTile.1dim, align 2
ret void
}
; CHECK-LABEL: testGEPConst
define float @testGEPConst() local_unnamed_addr {
; CHECK-NEXT: load float, ptr addrspace(3) getelementptr (float, ptr addrspace(3) @inputTile.1dim, i32 1), align 4
%v = load float, ptr addrspace(3) getelementptr (float, ptr addrspace(3) @inputTile.1dim, i32 1), align 4
ret float %v
}
; CHECK-LABEL: testGEPNonConst
define float @testGEPNonConst(i32 %i) local_unnamed_addr {
; CHECK-NEXT: getelementptr float, ptr addrspace(3) @inputTile.1dim, i32 %i
%gep = getelementptr float, ptr addrspace(3) @inputTile.1dim, i32 %i
%v = load float, ptr addrspace(3) %gep
ret float %v
}
; CHECK-LABEL: testAlloca
define float @testAlloca(i32 %i) local_unnamed_addr {
; CHECK-NEXT: alloca [3 x float], align 4
%arr = alloca [3 x float], align 4
; CHECK-NEXT: getelementptr [3 x float], ptr %arr, i32 1
%gep = getelementptr [3 x float], ptr %arr, i32 1
%v = load float, ptr %gep
ret float %v
}