[llvm] Adjust bit cast instruction filter for DXIL Prepare pass (PR #142678)
Joshua Batista via llvm-commits
llvm-commits at lists.llvm.org
Wed Jun 4 12:52:08 PDT 2025
https://github.com/bob80905 updated https://github.com/llvm/llvm-project/pull/142678
>From 0aff10620880dd7f58bd453ef5836fbf963f35dd Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista at microsoft.com>
Date: Tue, 3 Jun 2025 15:03:02 -0700
Subject: [PATCH 1/5] first attempt
---
llvm/lib/Target/DirectX/DXILPrepare.cpp | 37 +++++++++++++++++-
.../DirectX/noop_bitcast_global_array_type.ll | 39 +++++++++++++++++++
2 files changed, 74 insertions(+), 2 deletions(-)
create mode 100644 llvm/test/CodeGen/DirectX/noop_bitcast_global_array_type.ll
diff --git a/llvm/lib/Target/DirectX/DXILPrepare.cpp b/llvm/lib/Target/DirectX/DXILPrepare.cpp
index e0068787f5e5a..54e4499830ff0 100644
--- a/llvm/lib/Target/DirectX/DXILPrepare.cpp
+++ b/llvm/lib/Target/DirectX/DXILPrepare.cpp
@@ -148,9 +148,42 @@ class DXILPrepareModule : public ModulePass {
Type *Ty) {
// Omit bitcasts if the incoming value matches the instruction type.
auto It = PointerTypes.find(Operand);
- if (It != PointerTypes.end())
- if (cast<TypedPointerType>(It->second)->getElementType() == Ty)
+ if (It != PointerTypes.end()) {
+ auto OpTy = cast<TypedPointerType>(It->second)->getElementType();
+ if (OpTy == Ty)
return nullptr;
+ }
+
+ // Also omit the bitcast for matching global array types
+ if (auto *GlobalVar = llvm::dyn_cast<llvm::GlobalVariable>(Operand)) {
+ llvm::Type *ValTy = GlobalVar->getValueType();
+
+ if (auto *ArrTy = llvm::dyn_cast<llvm::ArrayType>(ValTy)) {
+ llvm::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 (auto *GEPInstr = llvm::dyn_cast<llvm::ConstantExpr>(Operand)) {
+ while (GEPInstr->getOpcode() == llvm::Instruction::GetElementPtr) {
+ llvm::Value *OpArg = GEPInstr->getOperand(0);
+
+ if (auto *GlobalVar = llvm::dyn_cast<llvm::GlobalVariable>(OpArg)) {
+ llvm::Constant *initializer = GlobalVar->getInitializer();
+
+ llvm::Type *ValTy = GlobalVar->getValueType();
+ if (auto *ArrTy = llvm::dyn_cast<llvm::ArrayType>(ValTy)) {
+ llvm::Type *ElTy = ArrTy->getElementType();
+ if (ElTy == Ty)
+ return nullptr;
+ }
+ }
+ }
+ }
+
// Insert bitcasts where we are removing the instruction.
Builder.SetInsertPoint(&Inst);
// This code only gets hit in opaque-pointer mode, so the type of the
diff --git a/llvm/test/CodeGen/DirectX/noop_bitcast_global_array_type.ll b/llvm/test/CodeGen/DirectX/noop_bitcast_global_array_type.ll
new file mode 100644
index 0000000000000..4cca76efcf461
--- /dev/null
+++ b/llvm/test/CodeGen/DirectX/noop_bitcast_global_array_type.ll
@@ -0,0 +1,39 @@
+; 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"
+
+ at inputTile.1dim = local_unnamed_addr addrspace(3) global [3 x float] zeroinitializer, align 2
+
+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: load float, ptr addrspace(3) @inputTile.1dim, align 2
+ %v = load float, ptr addrspace(3) @inputTile.1dim, align 2
+
+ ret float %v
+}
+
+define void @teststore() local_unnamed_addr {
+ ; CHECK: 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
+}
+
+define float @testGEPConst() local_unnamed_addr {
+ ; CHECK: 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
+}
+
+define float @testGEPNonConst(i32 %i) local_unnamed_addr {
+ ; CHECK: 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
+}
>From 9c6cad47e2c21d79a118c2e949ac29da032f9404 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista at microsoft.com>
Date: Tue, 3 Jun 2025 16:24:48 -0700
Subject: [PATCH 2/5] fix while loop
---
llvm/lib/Target/DirectX/DXILPrepare.cpp | 11 ++++++++---
1 file changed, 8 insertions(+), 3 deletions(-)
diff --git a/llvm/lib/Target/DirectX/DXILPrepare.cpp b/llvm/lib/Target/DirectX/DXILPrepare.cpp
index 54e4499830ff0..8bd5e40100de0 100644
--- a/llvm/lib/Target/DirectX/DXILPrepare.cpp
+++ b/llvm/lib/Target/DirectX/DXILPrepare.cpp
@@ -167,13 +167,17 @@ class DXILPrepareModule : public ModulePass {
// finally, drill down GEP instructions until we get the array
// that is being accessed, and compare element types
- if (auto *GEPInstr = llvm::dyn_cast<llvm::ConstantExpr>(Operand)) {
+ if (llvm::ConstantExpr *GEPInstr =
+ llvm::dyn_cast<llvm::ConstantExpr>(Operand)) {
while (GEPInstr->getOpcode() == llvm::Instruction::GetElementPtr) {
llvm::Value *OpArg = GEPInstr->getOperand(0);
+ if (llvm::ConstantExpr *NewGEPInstr =
+ llvm::dyn_cast<llvm::ConstantExpr>(OpArg)) {
+ GEPInstr = NewGEPInstr;
+ continue;
+ }
if (auto *GlobalVar = llvm::dyn_cast<llvm::GlobalVariable>(OpArg)) {
- llvm::Constant *initializer = GlobalVar->getInitializer();
-
llvm::Type *ValTy = GlobalVar->getValueType();
if (auto *ArrTy = llvm::dyn_cast<llvm::ArrayType>(ValTy)) {
llvm::Type *ElTy = ArrTy->getElementType();
@@ -181,6 +185,7 @@ class DXILPrepareModule : public ModulePass {
return nullptr;
}
}
+ break;
}
}
>From 40ec16ff4bebd1e8e38912c959cc1ffee66d6101 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista at microsoft.com>
Date: Tue, 3 Jun 2025 17:05:46 -0700
Subject: [PATCH 3/5] add check label and check next
---
.../DirectX/noop_bitcast_global_array_type.ll | 12 ++++++++----
1 file changed, 8 insertions(+), 4 deletions(-)
diff --git a/llvm/test/CodeGen/DirectX/noop_bitcast_global_array_type.ll b/llvm/test/CodeGen/DirectX/noop_bitcast_global_array_type.ll
index 4cca76efcf461..396643b06609a 100644
--- a/llvm/test/CodeGen/DirectX/noop_bitcast_global_array_type.ll
+++ b/llvm/test/CodeGen/DirectX/noop_bitcast_global_array_type.ll
@@ -7,31 +7,35 @@ 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: load float, ptr addrspace(3) @inputTile.1dim, align 2
+ ; 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: store float 2.000000e+00, ptr addrspace(3) @inputTile.1dim, align 2
+ ; 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: load float, ptr addrspace(3) getelementptr (float, ptr addrspace(3) @inputTile.1dim, i32 1), align 4
+ ; 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: getelementptr float, ptr addrspace(3) @inputTile.1dim, i32 %i
+ ; 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
>From 5df995655979bd36f235f4d5c8acc955021e99e3 Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista at microsoft.com>
Date: Wed, 4 Jun 2025 11:31:28 -0700
Subject: [PATCH 4/5] update test
---
.../DirectX/llc-vector-load-scalarize.ll | 58 ++++++++-----------
1 file changed, 23 insertions(+), 35 deletions(-)
diff --git a/llvm/test/CodeGen/DirectX/llc-vector-load-scalarize.ll b/llvm/test/CodeGen/DirectX/llc-vector-load-scalarize.ll
index 7e5a92e1311f8..e5aa4ff1e1db9 100644
--- a/llvm/test/CodeGen/DirectX/llc-vector-load-scalarize.ll
+++ b/llvm/test/CodeGen/DirectX/llc-vector-load-scalarize.ll
@@ -60,19 +60,15 @@ define <4 x i32> @load_array_vec_test() #0 {
define <4 x i32> @load_vec_test() #0 {
; CHECK-LABEL: define <4 x i32> @load_vec_test(
; CHECK-SAME: ) #[[ATTR0]] {
-; CHECK-NEXT: [[TMP1:%.*]] = bitcast ptr addrspace(3) @vecData.scalarized to ptr addrspace(3)
-; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr addrspace(3) [[TMP1]], align 4
-; CHECK-NEXT: [[TMP3:%.*]] = bitcast ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @vecData.scalarized, i32 1) to ptr addrspace(3)
-; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr addrspace(3) [[TMP3]], align 4
-; CHECK-NEXT: [[TMP5:%.*]] = bitcast ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @vecData.scalarized, i32 2) to ptr addrspace(3)
-; CHECK-NEXT: [[TMP6:%.*]] = load i32, ptr addrspace(3) [[TMP5]], align 4
-; CHECK-NEXT: [[TMP7:%.*]] = bitcast ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @vecData.scalarized, i32 3) to ptr addrspace(3)
-; CHECK-NEXT: [[TMP8:%.*]] = load i32, ptr addrspace(3) [[TMP7]], align 4
-; CHECK-NEXT: [[DOTUPTO0:%.*]] = insertelement <4 x i32> poison, i32 [[TMP2]], i32 0
-; 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]]
+; CHECK-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(3) @vecData.scalarized, align 4
+; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr addrspace(3) getelementptr (i32, ptr addrspace(3) @vecData.scalarized, i32 1), align 4
+; 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) getelementptr (i32, ptr addrspace(3) @vecData.scalarized, i32 3), align 4
+; CHECK-NEXT: [[DOTUPTO0:%.*]] = insertelement <4 x i32> poison, i32 [[TMP1]], i32 0
+; CHECK-NEXT: [[DOTUPTO1:%.*]] = insertelement <4 x i32> [[DOTUPTO0]], i32 [[TMP2]], i32 1
+; CHECK-NEXT: [[DOTUPTO2:%.*]] = insertelement <4 x i32> [[DOTUPTO1]], i32 [[TMP3]], i32 2
+; CHECK-NEXT: [[TMP5:%.*]] = insertelement <4 x i32> [[DOTUPTO2]], i32 [[TMP4]], i32 3
+; CHECK-NEXT: ret <4 x i32> [[TMP5]]
;
%1 = load <4 x i32>, <4 x i32> addrspace(3)* @"vecData", align 4
ret <4 x i32> %1
@@ -107,31 +103,23 @@ define <4 x i32> @load_static_array_of_vec_test(i32 %index) #0 {
define <4 x i32> @multid_load_test() #0 {
; CHECK-LABEL: define <4 x i32> @multid_load_test(
; CHECK-SAME: ) #[[ATTR0]] {
-; CHECK-NEXT: [[TMP1:%.*]] = bitcast ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim to ptr addrspace(3)
-; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr addrspace(3) [[TMP1]], 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: [[TMP4:%.*]] = load i32, ptr addrspace(3) [[TMP3]], 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: [[TMP6:%.*]] = load i32, ptr addrspace(3) [[TMP5]], 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: [[TMP8:%.*]] = load i32, ptr addrspace(3) [[TMP7]], align 4
-; CHECK-NEXT: [[TMP11:%.*]] = bitcast ptr addrspace(3) getelementptr inbounds ([36 x i32], ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 1) to ptr addrspace(3)
-; CHECK-NEXT: [[TMP12:%.*]] = load i32, ptr addrspace(3) [[TMP11]], align 4
-; CHECK-NEXT: [[DOTI12:%.*]] = bitcast ptr addrspace(3) getelementptr (i32, ptr addrspace(3) getelementptr inbounds ([36 x i32], ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 1), i32 1) to ptr addrspace(3)
-; CHECK-NEXT: [[DOTI13:%.*]] = load i32, ptr addrspace(3) [[DOTI12]], align 4
-; CHECK-NEXT: [[DOTI24:%.*]] = bitcast ptr addrspace(3) getelementptr (i32, ptr addrspace(3) getelementptr inbounds ([36 x i32], ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 1), i32 2) to ptr addrspace(3)
-; CHECK-NEXT: [[DOTI25:%.*]] = load i32, ptr addrspace(3) [[DOTI24]], align 4
-; CHECK-NEXT: [[DOTI36:%.*]] = bitcast ptr addrspace(3) getelementptr (i32, ptr addrspace(3) getelementptr inbounds ([36 x i32], ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 1), i32 3) to ptr addrspace(3)
-; CHECK-NEXT: [[DOTI37:%.*]] = load i32, ptr addrspace(3) [[DOTI36]], align 4
-; CHECK-NEXT: [[DOTI08:%.*]] = add i32 [[TMP2]], [[TMP12]]
-; CHECK-NEXT: [[DOTI19:%.*]] = add i32 [[TMP4]], [[DOTI13]]
-; CHECK-NEXT: [[DOTI210:%.*]] = add i32 [[TMP6]], [[DOTI25]]
-; CHECK-NEXT: [[DOTI311:%.*]] = add i32 [[TMP8]], [[DOTI37]]
+; CHECK-NEXT: [[TMP1:%.*]] = load i32, ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, 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:%.*]] = 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) getelementptr (i32, ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 3), align 4
+; CHECK-NEXT: [[TMP5:%.*]] = load i32, ptr addrspace(3) getelementptr inbounds ([36 x i32], ptr addrspace(3) @groushared2dArrayofVectors.scalarized.1dim, i32 1), 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 1), i32 1), align 4
+; 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 1), i32 2), 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 1), i32 3), align 4
+; CHECK-NEXT: [[DOTI08:%.*]] = add i32 [[TMP1]], [[TMP5]]
+; CHECK-NEXT: [[DOTI19:%.*]] = add i32 [[TMP2]], [[DOTI13]]
+; CHECK-NEXT: [[DOTI210:%.*]] = add i32 [[TMP3]], [[DOTI25]]
+; CHECK-NEXT: [[DOTI311:%.*]] = add i32 [[TMP4]], [[DOTI37]]
; CHECK-NEXT: [[DOTUPTO015:%.*]] = insertelement <4 x i32> poison, i32 [[DOTI08]], i32 0
; CHECK-NEXT: [[DOTUPTO116:%.*]] = insertelement <4 x i32> [[DOTUPTO015]], i32 [[DOTI19]], i32 1
; CHECK-NEXT: [[DOTUPTO217:%.*]] = insertelement <4 x i32> [[DOTUPTO116]], i32 [[DOTI210]], i32 2
-; CHECK-NEXT: [[TMP16:%.*]] = insertelement <4 x i32> [[DOTUPTO217]], i32 [[DOTI311]], i32 3
-; CHECK-NEXT: ret <4 x i32> [[TMP16]]
+; CHECK-NEXT: [[TMP6:%.*]] = insertelement <4 x i32> [[DOTUPTO217]], i32 [[DOTI311]], i32 3
+; CHECK-NEXT: ret <4 x i32> [[TMP6]]
;
%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
>From b42ee72a0785ea189ae6f01501d8d725d5e1736f Mon Sep 17 00:00:00 2001
From: Joshua Batista <jbatista at microsoft.com>
Date: Wed, 4 Jun 2025 12:51:34 -0700
Subject: [PATCH 5/5] force pointer type deduction
---
llvm/lib/Target/DirectX/DXILPrepare.cpp | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/llvm/lib/Target/DirectX/DXILPrepare.cpp b/llvm/lib/Target/DirectX/DXILPrepare.cpp
index 8bd5e40100de0..ec50aa149ceb5 100644
--- a/llvm/lib/Target/DirectX/DXILPrepare.cpp
+++ b/llvm/lib/Target/DirectX/DXILPrepare.cpp
@@ -149,7 +149,7 @@ class DXILPrepareModule : public ModulePass {
// Omit bitcasts if the incoming value matches the instruction type.
auto It = PointerTypes.find(Operand);
if (It != PointerTypes.end()) {
- auto OpTy = cast<TypedPointerType>(It->second)->getElementType();
+ auto *OpTy = cast<TypedPointerType>(It->second)->getElementType();
if (OpTy == Ty)
return nullptr;
}
More information about the llvm-commits
mailing list