[llvm] [SPIRV] Handle ptrcast between array and vector types (PR #166418)
Steven Perron via llvm-commits
llvm-commits at lists.llvm.org
Fri Nov 7 17:04:11 PST 2025
https://github.com/s-perron updated https://github.com/llvm/llvm-project/pull/166418
>From 3abe11ea28636b7f25b064725c53f9a7a474a2f0 Mon Sep 17 00:00:00 2001
From: Steven Perron <stevenperron at google.com>
Date: Thu, 30 Oct 2025 13:01:14 -0400
Subject: [PATCH 1/2] [SPIRV] Handle ptrcast between array and vector types
This commit adds support for legalizing pointer casts between array and vector types within the SPIRV backend.
This is necessary to handle cases where a vector is loaded from or stored to an array, which can occur with HLSL matrix types.
The following changes are included:
- Added to load a vector from an array.
- Added to store a vector to an array.
- Added the test case to verify the functionality.
---
.../Target/SPIRV/SPIRVLegalizePointerCast.cpp | 84 +++++++++++++++++++
.../pointers/load-store-vec-from-array.ll | 54 ++++++++++++
2 files changed, 138 insertions(+)
create mode 100644 llvm/test/CodeGen/SPIRV/pointers/load-store-vec-from-array.ll
diff --git a/llvm/lib/Target/SPIRV/SPIRVLegalizePointerCast.cpp b/llvm/lib/Target/SPIRV/SPIRVLegalizePointerCast.cpp
index 65dffc7908b78..87c0c8c5a7437 100644
--- a/llvm/lib/Target/SPIRV/SPIRVLegalizePointerCast.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVLegalizePointerCast.cpp
@@ -116,6 +116,85 @@ class SPIRVLegalizePointerCast : public FunctionPass {
return LI;
}
+ // Loads elements from an array and constructs a vector.
+ Value *loadVectorFromArray(IRBuilder<> &B, FixedVectorType *TargetType,
+ ArrayType *SourceType, Value *Source) {
+ // Ensure the element types of the array and vector are the same.
+ assert(TargetType->getElementType() == SourceType->getElementType() &&
+ "Element types of array and vector must be the same.");
+
+ // Load each element of the array.
+ SmallVector<Value *, 4> LoadedElements;
+ for (unsigned i = 0; i < TargetType->getNumElements(); ++i) {
+ // Create a GEP to access the i-th element of the array.
+ SmallVector<Type *, 2> Types = {Source->getType(), Source->getType()};
+ SmallVector<Value *, 4> Args;
+ Args.push_back(B.getInt1(true));
+ Args.push_back(Source);
+ Args.push_back(B.getInt32(0));
+ Args.push_back(ConstantInt::get(B.getInt32Ty(), i));
+ auto *ElementPtr = B.CreateIntrinsic(Intrinsic::spv_gep, {Types}, {Args});
+ GR->buildAssignPtr(B, TargetType->getElementType(), ElementPtr);
+
+ // Load the value from the element pointer.
+ Value *Load = B.CreateLoad(TargetType->getElementType(), ElementPtr);
+ buildAssignType(B, TargetType->getElementType(), Load);
+ LoadedElements.push_back(Load);
+ }
+
+ // Build the vector from the loaded elements.
+ Value *NewVector = UndefValue::get(TargetType);
+ buildAssignType(B, TargetType, NewVector);
+
+ for (unsigned i = 0; i < TargetType->getNumElements(); ++i) {
+ Value *Index = B.getInt32(i);
+ SmallVector<Type *, 4> Types = {TargetType, TargetType,
+ TargetType->getElementType(),
+ Index->getType()};
+ SmallVector<Value *> Args = {NewVector, LoadedElements[i], Index};
+ NewVector = B.CreateIntrinsic(Intrinsic::spv_insertelt, {Types}, {Args});
+ buildAssignType(B, TargetType, NewVector);
+ }
+ return NewVector;
+ }
+
+ // Stores elements from a vector into an array.
+ void storeArrayFromVector(IRBuilder<> &B, Value *SrcVector,
+ Value *DstArrayPtr, ArrayType *ArrTy,
+ Align Alignment) {
+ auto *VecTy = cast<FixedVectorType>(SrcVector->getType());
+
+ // Ensure the element types of the array and vector are the same.
+ assert(VecTy->getElementType() == ArrTy->getElementType() &&
+ "Element types of array and vector must be the same.");
+
+ for (unsigned i = 0; i < VecTy->getNumElements(); ++i) {
+ // Create a GEP to access the i-th element of the array.
+ SmallVector<Type *, 2> Types = {DstArrayPtr->getType(),
+ DstArrayPtr->getType()};
+ SmallVector<Value *, 4> Args;
+ Args.push_back(B.getInt1(true));
+ Args.push_back(DstArrayPtr);
+ Args.push_back(B.getInt32(0));
+ Args.push_back(ConstantInt::get(B.getInt32Ty(), i));
+ auto *ElementPtr = B.CreateIntrinsic(Intrinsic::spv_gep, {Types}, {Args});
+ GR->buildAssignPtr(B, ArrTy->getElementType(), ElementPtr);
+
+ // Extract the element from the vector and store it.
+ Value *Index = B.getInt32(i);
+ SmallVector<Type *, 3> EltTypes = {VecTy->getElementType(), VecTy,
+ Index->getType()};
+ SmallVector<Value *, 2> EltArgs = {SrcVector, Index};
+ Value *Element =
+ B.CreateIntrinsic(Intrinsic::spv_extractelt, {EltTypes}, {EltArgs});
+ buildAssignType(B, VecTy->getElementType(), Element);
+
+ Types = {Element->getType(), ElementPtr->getType()};
+ Args = {Element, ElementPtr, B.getInt16(2), B.getInt8(Alignment.value())};
+ B.CreateIntrinsic(Intrinsic::spv_store, {Types}, {Args});
+ }
+ }
+
// Replaces the load instruction to get rid of the ptrcast used as source
// operand.
void transformLoad(IRBuilder<> &B, LoadInst *LI, Value *CastedOperand,
@@ -154,6 +233,8 @@ class SPIRVLegalizePointerCast : public FunctionPass {
// - float v = s.m;
else if (SST && SST->getTypeAtIndex(0u) == ToTy)
Output = loadFirstValueFromAggregate(B, ToTy, OriginalOperand, LI);
+ else if (SAT && DVT && SAT->getElementType() == DVT->getElementType())
+ Output = loadVectorFromArray(B, DVT, SAT, OriginalOperand);
else
llvm_unreachable("Unimplemented implicit down-cast from load.");
@@ -288,6 +369,7 @@ class SPIRVLegalizePointerCast : public FunctionPass {
auto *S_VT = dyn_cast<FixedVectorType>(FromTy);
auto *D_ST = dyn_cast<StructType>(ToTy);
auto *D_VT = dyn_cast<FixedVectorType>(ToTy);
+ auto *D_AT = dyn_cast<ArrayType>(ToTy);
B.SetInsertPoint(BadStore);
if (D_ST && isTypeFirstElementAggregate(FromTy, D_ST))
@@ -296,6 +378,8 @@ class SPIRVLegalizePointerCast : public FunctionPass {
storeVectorFromVector(B, Src, Dst, Alignment);
else if (D_VT && !S_VT && FromTy == D_VT->getElementType())
storeToFirstValueAggregate(B, Src, Dst, D_VT, Alignment);
+ else if (D_AT && S_VT && S_VT->getElementType() == D_AT->getElementType())
+ storeArrayFromVector(B, Src, Dst, D_AT, Alignment);
else
llvm_unreachable("Unsupported ptrcast use in store. Please fix.");
diff --git a/llvm/test/CodeGen/SPIRV/pointers/load-store-vec-from-array.ll b/llvm/test/CodeGen/SPIRV/pointers/load-store-vec-from-array.ll
new file mode 100644
index 0000000000000..2a6d799c5bff6
--- /dev/null
+++ b/llvm/test/CodeGen/SPIRV/pointers/load-store-vec-from-array.ll
@@ -0,0 +1,54 @@
+; RUN: llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-vulkan %s -o - | FileCheck %s
+; RUN: %if spirv-tools %{ llc -O0 -verify-machineinstrs -mtriple=spirv-unknown-vulkan %s -o - -filetype=obj | spirv-val %}
+
+; CHECK-DAG: [[FLOAT:%[0-9]+]] = OpTypeFloat 32
+; CHECK-DAG: [[VEC4FLOAT:%[0-9]+]] = OpTypeVector [[FLOAT]] 4
+; CHECK-DAG: [[UINT_TYPE:%[0-9]+]] = OpTypeInt 32 0
+; CHECK-DAG: [[UINT4:%[0-9]+]] = OpConstant [[UINT_TYPE]] 4
+; CHECK-DAG: [[ARRAY4FLOAT:%[0-9]+]] = OpTypeArray [[FLOAT]] [[UINT4]]
+; CHECK-DAG: [[PTR_ARRAY4FLOAT:%[0-9]+]] = OpTypePointer Private [[ARRAY4FLOAT]]
+; CHECK-DAG: [[G_IN:%[0-9]+]] = OpVariable [[PTR_ARRAY4FLOAT]] Private
+; CHECK-DAG: [[G_OUT:%[0-9]+]] = OpVariable [[PTR_ARRAY4FLOAT]] Private
+; CHECK-DAG: [[UINT0:%[0-9]+]] = OpConstant [[UINT_TYPE]] 0
+; CHECK-DAG: [[UINT1:%[0-9]+]] = OpConstant [[UINT_TYPE]] 1
+; CHECK-DAG: [[UINT2:%[0-9]+]] = OpConstant [[UINT_TYPE]] 2
+; CHECK-DAG: [[UINT3:%[0-9]+]] = OpConstant [[UINT_TYPE]] 3
+; CHECK-DAG: [[PTR_FLOAT:%[0-9]+]] = OpTypePointer Private [[FLOAT]]
+; CHECK-DAG: [[UNDEF_VEC:%[0-9]+]] = OpUndef [[VEC4FLOAT]]
+
+ at G_in = internal addrspace(10) global [4 x float] zeroinitializer
+ at G_out = internal addrspace(10) global [4 x float] zeroinitializer
+
+define spir_func void @main() {
+entry:
+; CHECK: [[GEP0:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_IN]] [[UINT0]]
+; CHECK-NEXT: [[LOAD0:%[0-9]+]] = OpLoad [[FLOAT]] [[GEP0]]
+; CHECK-NEXT: [[GEP1:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_IN]] [[UINT1]]
+; CHECK-NEXT: [[LOAD1:%[0-9]+]] = OpLoad [[FLOAT]] [[GEP1]]
+; CHECK-NEXT: [[GEP2:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_IN]] [[UINT2]]
+; CHECK-NEXT: [[LOAD2:%[0-9]+]] = OpLoad [[FLOAT]] [[GEP2]]
+; CHECK-NEXT: [[GEP3:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_IN]] [[UINT3]]
+; CHECK-NEXT: [[LOAD3:%[0-9]+]] = OpLoad [[FLOAT]] [[GEP3]]
+; CHECK-NEXT: [[VEC_INSERT0:%[0-9]+]] = OpCompositeInsert [[VEC4FLOAT]] [[LOAD0]] [[UNDEF_VEC]] 0
+; CHECK-NEXT: [[VEC_INSERT1:%[0-9]+]] = OpCompositeInsert [[VEC4FLOAT]] [[LOAD1]] [[VEC_INSERT0]] 1
+; CHECK-NEXT: [[VEC_INSERT2:%[0-9]+]] = OpCompositeInsert [[VEC4FLOAT]] [[LOAD2]] [[VEC_INSERT1]] 2
+; CHECK-NEXT: [[VEC:%[0-9]+]] = OpCompositeInsert [[VEC4FLOAT]] [[LOAD3]] [[VEC_INSERT2]] 3
+ %0 = load <4 x float>, ptr addrspace(10) @G_in, align 64
+
+; CHECK-NEXT: [[GEP_OUT0:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_OUT]] [[UINT0]]
+; CHECK-NEXT: [[VEC_EXTRACT0:%[0-9]+]] = OpCompositeExtract [[FLOAT]] [[VEC]] 0
+; CHECK-NEXT: OpStore [[GEP_OUT0]] [[VEC_EXTRACT0]]
+; CHECK-NEXT: [[GEP_OUT1:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_OUT]] [[UINT1]]
+; CHECK-NEXT: [[VEC_EXTRACT1:%[0-9]+]] = OpCompositeExtract [[FLOAT]] [[VEC]] 1
+; CHECK-NEXT: OpStore [[GEP_OUT1]] [[VEC_EXTRACT1]]
+; CHECK-NEXT: [[GEP_OUT2:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_OUT]] [[UINT2]]
+; CHECK-NEXT: [[VEC_EXTRACT2:%[0-9]+]] = OpCompositeExtract [[FLOAT]] [[VEC]] 2
+; CHECK-NEXT: OpStore [[GEP_OUT2]] [[VEC_EXTRACT2]]
+; CHECK-NEXT: [[GEP_OUT3:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_OUT]] [[UINT3]]
+; CHECK-NEXT: [[VEC_EXTRACT3:%[0-9]+]] = OpCompositeExtract [[FLOAT]] [[VEC]] 3
+; CHECK-NEXT: OpStore [[GEP_OUT3]] [[VEC_EXTRACT3]]
+ store <4 x float> %0, ptr addrspace(10) @G_out, align 64
+
+; CHECK-NEXT: OpReturn
+ ret void
+}
>From 3c0e19709efbd4f8a68a67e73ac9e0d6e01dadc8 Mon Sep 17 00:00:00 2001
From: Steven Perron <stevenperron at google.com>
Date: Fri, 7 Nov 2025 20:03:57 -0500
Subject: [PATCH 2/2] Changes from code review.
---
.../Target/SPIRV/SPIRVLegalizePointerCast.cpp | 14 +++++---------
.../SPIRV/pointers/load-store-vec-from-array.ll | 16 ++++++++--------
2 files changed, 13 insertions(+), 17 deletions(-)
diff --git a/llvm/lib/Target/SPIRV/SPIRVLegalizePointerCast.cpp b/llvm/lib/Target/SPIRV/SPIRVLegalizePointerCast.cpp
index 87c0c8c5a7437..4ce871b6f5e5d 100644
--- a/llvm/lib/Target/SPIRV/SPIRVLegalizePointerCast.cpp
+++ b/llvm/lib/Target/SPIRV/SPIRVLegalizePointerCast.cpp
@@ -118,18 +118,14 @@ class SPIRVLegalizePointerCast : public FunctionPass {
// Loads elements from an array and constructs a vector.
Value *loadVectorFromArray(IRBuilder<> &B, FixedVectorType *TargetType,
- ArrayType *SourceType, Value *Source) {
- // Ensure the element types of the array and vector are the same.
- assert(TargetType->getElementType() == SourceType->getElementType() &&
- "Element types of array and vector must be the same.");
-
+ Value *Source) {
// Load each element of the array.
SmallVector<Value *, 4> LoadedElements;
for (unsigned i = 0; i < TargetType->getNumElements(); ++i) {
// Create a GEP to access the i-th element of the array.
SmallVector<Type *, 2> Types = {Source->getType(), Source->getType()};
SmallVector<Value *, 4> Args;
- Args.push_back(B.getInt1(true));
+ Args.push_back(B.getInt1(false));
Args.push_back(Source);
Args.push_back(B.getInt32(0));
Args.push_back(ConstantInt::get(B.getInt32Ty(), i));
@@ -143,7 +139,7 @@ class SPIRVLegalizePointerCast : public FunctionPass {
}
// Build the vector from the loaded elements.
- Value *NewVector = UndefValue::get(TargetType);
+ Value *NewVector = PoisonValue::get(TargetType);
buildAssignType(B, TargetType, NewVector);
for (unsigned i = 0; i < TargetType->getNumElements(); ++i) {
@@ -173,7 +169,7 @@ class SPIRVLegalizePointerCast : public FunctionPass {
SmallVector<Type *, 2> Types = {DstArrayPtr->getType(),
DstArrayPtr->getType()};
SmallVector<Value *, 4> Args;
- Args.push_back(B.getInt1(true));
+ Args.push_back(B.getInt1(false));
Args.push_back(DstArrayPtr);
Args.push_back(B.getInt32(0));
Args.push_back(ConstantInt::get(B.getInt32Ty(), i));
@@ -234,7 +230,7 @@ class SPIRVLegalizePointerCast : public FunctionPass {
else if (SST && SST->getTypeAtIndex(0u) == ToTy)
Output = loadFirstValueFromAggregate(B, ToTy, OriginalOperand, LI);
else if (SAT && DVT && SAT->getElementType() == DVT->getElementType())
- Output = loadVectorFromArray(B, DVT, SAT, OriginalOperand);
+ Output = loadVectorFromArray(B, DVT, OriginalOperand);
else
llvm_unreachable("Unimplemented implicit down-cast from load.");
diff --git a/llvm/test/CodeGen/SPIRV/pointers/load-store-vec-from-array.ll b/llvm/test/CodeGen/SPIRV/pointers/load-store-vec-from-array.ll
index 2a6d799c5bff6..917bb27afad00 100644
--- a/llvm/test/CodeGen/SPIRV/pointers/load-store-vec-from-array.ll
+++ b/llvm/test/CodeGen/SPIRV/pointers/load-store-vec-from-array.ll
@@ -21,13 +21,13 @@
define spir_func void @main() {
entry:
-; CHECK: [[GEP0:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_IN]] [[UINT0]]
+; CHECK: [[GEP0:%[0-9]+]] = OpAccessChain [[PTR_FLOAT]] [[G_IN]] [[UINT0]]
; CHECK-NEXT: [[LOAD0:%[0-9]+]] = OpLoad [[FLOAT]] [[GEP0]]
-; CHECK-NEXT: [[GEP1:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_IN]] [[UINT1]]
+; CHECK-NEXT: [[GEP1:%[0-9]+]] = OpAccessChain [[PTR_FLOAT]] [[G_IN]] [[UINT1]]
; CHECK-NEXT: [[LOAD1:%[0-9]+]] = OpLoad [[FLOAT]] [[GEP1]]
-; CHECK-NEXT: [[GEP2:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_IN]] [[UINT2]]
+; CHECK-NEXT: [[GEP2:%[0-9]+]] = OpAccessChain [[PTR_FLOAT]] [[G_IN]] [[UINT2]]
; CHECK-NEXT: [[LOAD2:%[0-9]+]] = OpLoad [[FLOAT]] [[GEP2]]
-; CHECK-NEXT: [[GEP3:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_IN]] [[UINT3]]
+; CHECK-NEXT: [[GEP3:%[0-9]+]] = OpAccessChain [[PTR_FLOAT]] [[G_IN]] [[UINT3]]
; CHECK-NEXT: [[LOAD3:%[0-9]+]] = OpLoad [[FLOAT]] [[GEP3]]
; CHECK-NEXT: [[VEC_INSERT0:%[0-9]+]] = OpCompositeInsert [[VEC4FLOAT]] [[LOAD0]] [[UNDEF_VEC]] 0
; CHECK-NEXT: [[VEC_INSERT1:%[0-9]+]] = OpCompositeInsert [[VEC4FLOAT]] [[LOAD1]] [[VEC_INSERT0]] 1
@@ -35,16 +35,16 @@ entry:
; CHECK-NEXT: [[VEC:%[0-9]+]] = OpCompositeInsert [[VEC4FLOAT]] [[LOAD3]] [[VEC_INSERT2]] 3
%0 = load <4 x float>, ptr addrspace(10) @G_in, align 64
-; CHECK-NEXT: [[GEP_OUT0:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_OUT]] [[UINT0]]
+; CHECK-NEXT: [[GEP_OUT0:%[0-9]+]] = OpAccessChain [[PTR_FLOAT]] [[G_OUT]] [[UINT0]]
; CHECK-NEXT: [[VEC_EXTRACT0:%[0-9]+]] = OpCompositeExtract [[FLOAT]] [[VEC]] 0
; CHECK-NEXT: OpStore [[GEP_OUT0]] [[VEC_EXTRACT0]]
-; CHECK-NEXT: [[GEP_OUT1:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_OUT]] [[UINT1]]
+; CHECK-NEXT: [[GEP_OUT1:%[0-9]+]] = OpAccessChain [[PTR_FLOAT]] [[G_OUT]] [[UINT1]]
; CHECK-NEXT: [[VEC_EXTRACT1:%[0-9]+]] = OpCompositeExtract [[FLOAT]] [[VEC]] 1
; CHECK-NEXT: OpStore [[GEP_OUT1]] [[VEC_EXTRACT1]]
-; CHECK-NEXT: [[GEP_OUT2:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_OUT]] [[UINT2]]
+; CHECK-NEXT: [[GEP_OUT2:%[0-9]+]] = OpAccessChain [[PTR_FLOAT]] [[G_OUT]] [[UINT2]]
; CHECK-NEXT: [[VEC_EXTRACT2:%[0-9]+]] = OpCompositeExtract [[FLOAT]] [[VEC]] 2
; CHECK-NEXT: OpStore [[GEP_OUT2]] [[VEC_EXTRACT2]]
-; CHECK-NEXT: [[GEP_OUT3:%[0-9]+]] = OpInBoundsAccessChain [[PTR_FLOAT]] [[G_OUT]] [[UINT3]]
+; CHECK-NEXT: [[GEP_OUT3:%[0-9]+]] = OpAccessChain [[PTR_FLOAT]] [[G_OUT]] [[UINT3]]
; CHECK-NEXT: [[VEC_EXTRACT3:%[0-9]+]] = OpCompositeExtract [[FLOAT]] [[VEC]] 3
; CHECK-NEXT: OpStore [[GEP_OUT3]] [[VEC_EXTRACT3]]
store <4 x float> %0, ptr addrspace(10) @G_out, align 64
More information about the llvm-commits
mailing list