[AMDGPU] Generate canonical additions in AMDGPUPromoteAlloca #157810
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This stack of pull requests is managed by Graphite. Learn more about stacking. |
|
@llvm/pr-subscribers-backend-amdgpu Author: Fabian Ritter (ritter-x2a) ChangesWhen we know that one operand of an addition is a constant, we might was Full diff: https://github.com/llvm/llvm-project/pull/157810.diff 4 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp b/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
index bb77cdff778c0..7dbe1235a98b5 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
@@ -478,7 +478,7 @@ static Value *GEPToVectorIndex(GetElementPtrInst *GEP, AllocaInst *Alloca,
ConstantInt *ConstIndex =
ConstantInt::get(OffsetType, IndexQuot.getSExtValue());
- Value *IndexAdd = Builder.CreateAdd(ConstIndex, Offset);
+ Value *IndexAdd = Builder.CreateAdd(Offset, ConstIndex);
if (Instruction *NewInst = dyn_cast<Instruction>(IndexAdd))
NewInsts.push_back(NewInst);
return IndexAdd;
diff --git a/llvm/test/CodeGen/AMDGPU/promote-alloca-multidim.ll b/llvm/test/CodeGen/AMDGPU/promote-alloca-multidim.ll
index d72f158763c61..63622e67e7d0b 100644
--- a/llvm/test/CodeGen/AMDGPU/promote-alloca-multidim.ll
+++ b/llvm/test/CodeGen/AMDGPU/promote-alloca-multidim.ll
@@ -312,7 +312,7 @@ define amdgpu_kernel void @i64_2d_load_store_subvec_3_i64_offset_index(ptr %out)
; CHECK-NEXT: [[TMP15:%.*]] = insertelement <6 x i64> [[TMP14]], i64 4, i32 4
; CHECK-NEXT: [[TMP16:%.*]] = insertelement <6 x i64> [[TMP15]], i64 5, i32 5
; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[SEL3]], 3
-; CHECK-NEXT: [[TMP2:%.*]] = add i64 6, [[TMP1]]
+; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[TMP1]], 6
; CHECK-NEXT: [[TMP3:%.*]] = extractelement <6 x i64> [[TMP16]], i64 [[TMP2]]
; CHECK-NEXT: [[TMP4:%.*]] = insertelement <3 x i64> poison, i64 [[TMP3]], i64 0
; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[TMP2]], 1
@@ -464,7 +464,7 @@ define amdgpu_kernel void @i16_2d_load_store(ptr %out, i32 %sel) {
; CHECK-NEXT: [[TMP4:%.*]] = insertelement <6 x i16> [[TMP3]], i16 3, i32 3
; CHECK-NEXT: [[TMP5:%.*]] = insertelement <6 x i16> [[TMP4]], i16 4, i32 4
; CHECK-NEXT: [[TMP6:%.*]] = insertelement <6 x i16> [[TMP5]], i16 5, i32 5
-; CHECK-NEXT: [[TMP1:%.*]] = add i32 3, [[SEL]]
+; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[SEL]], 3
; CHECK-NEXT: [[TMP2:%.*]] = extractelement <6 x i16> [[TMP6]], i32 [[TMP1]]
; CHECK-NEXT: store i16 [[TMP2]], ptr [[OUT]], align 2
; CHECK-NEXT: ret void
@@ -498,7 +498,7 @@ define amdgpu_kernel void @float_2d_load_store(ptr %out, i32 %sel) {
; CHECK-NEXT: [[TMP4:%.*]] = insertelement <6 x float> [[TMP3]], float 3.000000e+00, i32 3
; CHECK-NEXT: [[TMP5:%.*]] = insertelement <6 x float> [[TMP4]], float 4.000000e+00, i32 4
; CHECK-NEXT: [[TMP6:%.*]] = insertelement <6 x float> [[TMP5]], float 5.000000e+00, i32 5
-; CHECK-NEXT: [[TMP1:%.*]] = add i32 3, [[SEL]]
+; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[SEL]], 3
; CHECK-NEXT: [[TMP2:%.*]] = extractelement <6 x float> [[TMP6]], i32 [[TMP1]]
; CHECK-NEXT: store float [[TMP2]], ptr [[OUT]], align 4
; CHECK-NEXT: ret void
@@ -538,7 +538,7 @@ define amdgpu_kernel void @ptr_2d_load_store(ptr %out, i32 %sel) {
; CHECK-NEXT: [[TMP4:%.*]] = insertelement <6 x ptr> [[TMP3]], ptr [[PTR_3]], i32 3
; CHECK-NEXT: [[TMP5:%.*]] = insertelement <6 x ptr> [[TMP4]], ptr [[PTR_4]], i32 4
; CHECK-NEXT: [[TMP6:%.*]] = insertelement <6 x ptr> [[TMP5]], ptr [[PTR_5]], i32 5
-; CHECK-NEXT: [[TMP7:%.*]] = add i32 3, [[SEL]]
+; CHECK-NEXT: [[TMP7:%.*]] = add i32 [[SEL]], 3
; CHECK-NEXT: [[TMP8:%.*]] = extractelement <6 x ptr> [[TMP6]], i32 [[TMP7]]
; CHECK-NEXT: store ptr [[TMP8]], ptr [[OUT]], align 8
; CHECK-NEXT: ret void
diff --git a/llvm/test/CodeGen/AMDGPU/promote-alloca-negative-index.ll b/llvm/test/CodeGen/AMDGPU/promote-alloca-negative-index.ll
index 1b6ac0bd93c19..a865bf5058d6a 100644
--- a/llvm/test/CodeGen/AMDGPU/promote-alloca-negative-index.ll
+++ b/llvm/test/CodeGen/AMDGPU/promote-alloca-negative-index.ll
@@ -11,7 +11,7 @@ define amdgpu_kernel void @negative_index_byte(ptr %out, i64 %offset) {
; CHECK-NEXT: [[TMP2:%.*]] = insertelement <4 x i8> [[TMP1]], i8 1, i32 1
; CHECK-NEXT: [[TMP3:%.*]] = insertelement <4 x i8> [[TMP2]], i8 2, i32 2
; CHECK-NEXT: [[TMP4:%.*]] = insertelement <4 x i8> [[TMP3]], i8 3, i32 3
-; CHECK-NEXT: [[TMP5:%.*]] = add i64 -1, [[OFFSET:%.*]]
+; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET:%.*]], -1
; CHECK-NEXT: [[TMP6:%.*]] = extractelement <4 x i8> [[TMP4]], i64 [[TMP5]]
; CHECK-NEXT: store i8 [[TMP6]], ptr [[OUT:%.*]], align 1
; CHECK-NEXT: ret void
@@ -39,7 +39,7 @@ define amdgpu_kernel void @negative_index_word(ptr %out, i64 %offset) {
; CHECK-NEXT: [[TMP2:%.*]] = insertelement <4 x i32> [[TMP1]], i32 1, i32 1
; CHECK-NEXT: [[TMP3:%.*]] = insertelement <4 x i32> [[TMP2]], i32 2, i32 2
; CHECK-NEXT: [[TMP4:%.*]] = insertelement <4 x i32> [[TMP3]], i32 3, i32 3
-; CHECK-NEXT: [[TMP5:%.*]] = add i64 -1, [[OFFSET:%.*]]
+; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET:%.*]], -1
; CHECK-NEXT: [[TMP6:%.*]] = extractelement <4 x i32> [[TMP4]], i64 [[TMP5]]
; CHECK-NEXT: store i32 [[TMP6]], ptr [[OUT:%.*]], align 4
; CHECK-NEXT: ret void
diff --git a/llvm/test/CodeGen/AMDGPU/promote-alloca-vector-gep-of-gep.ll b/llvm/test/CodeGen/AMDGPU/promote-alloca-vector-gep-of-gep.ll
index a24f041a17857..f95a6a8ec9b45 100644
--- a/llvm/test/CodeGen/AMDGPU/promote-alloca-vector-gep-of-gep.ll
+++ b/llvm/test/CodeGen/AMDGPU/promote-alloca-vector-gep-of-gep.ll
@@ -10,7 +10,7 @@ define amdgpu_ps void @scalar_alloca_ptr_with_vector_gep_of_gep(i32 %idx, ptr ad
; CHECK-NEXT: [[TMP2:%.*]] = add i32 [[TMP0]], 1
; CHECK-NEXT: [[TMP3:%.*]] = insertelement <20 x i32> [[TMP1]], i32 2, i32 [[TMP2]]
; CHECK-NEXT: [[TMP4:%.*]] = mul i32 [[IDX]], 2
-; CHECK-NEXT: [[TMP5:%.*]] = add i32 1, [[TMP4]]
+; CHECK-NEXT: [[TMP5:%.*]] = add i32 [[TMP4]], 1
; CHECK-NEXT: [[TMP6:%.*]] = extractelement <20 x i32> [[TMP3]], i32 [[TMP5]]
; CHECK-NEXT: store i32 [[TMP6]], ptr addrspace(1) [[OUTPUT]], align 4
; CHECK-NEXT: ret void
@@ -31,12 +31,12 @@ define amdgpu_ps void @scalar_alloca_ptr_with_vector_gep_of_gep3(i32 %idx, ptr a
; CHECK-NEXT: [[ENTRY:.*:]]
; CHECK-NEXT: [[ALLOCA:%.*]] = freeze <16 x i32> poison
; CHECK-NEXT: [[TMP0:%.*]] = mul i32 [[IDX]], 2
-; CHECK-NEXT: [[TMP1:%.*]] = add i32 8, [[TMP0]]
+; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[TMP0]], 8
; CHECK-NEXT: [[TMP2:%.*]] = insertelement <16 x i32> [[ALLOCA]], i32 10, i32 [[TMP1]]
; CHECK-NEXT: [[TMP3:%.*]] = add i32 [[TMP1]], 1
; CHECK-NEXT: [[TMP4:%.*]] = insertelement <16 x i32> [[TMP2]], i32 20, i32 [[TMP3]]
; CHECK-NEXT: [[TMP5:%.*]] = mul i32 [[IDX]], 2
-; CHECK-NEXT: [[TMP6:%.*]] = add i32 9, [[TMP5]]
+; CHECK-NEXT: [[TMP6:%.*]] = add i32 [[TMP5]], 9
; CHECK-NEXT: [[TMP7:%.*]] = extractelement <16 x i32> [[TMP4]], i32 [[TMP6]]
; CHECK-NEXT: store i32 [[TMP7]], ptr addrspace(1) [[OUTPUT]], align 4
; CHECK-NEXT: ret void
|
nikic
approved these changes
Sep 10, 2025
arsenm
approved these changes
Sep 10, 2025
Base automatically changed from
users/ritter-x2a/09-09-_amdgpu_treat_gep_offsets_as_signed_in_amdgpupromotealloca
to
main
September 10, 2025 09:32
When we know that one operand of an addition is a constant, we might was well put it on the right-hand side and avoid the work to canonicalize it in a later pass.
ritter-x2a
force-pushed
the
users/ritter-x2a/09-10-_amdgpu_generate_canonical_additions_in_amdgpupromotealloca
branch
from
f6a8f01 to
43fc13b
Compare
|
Manual rebase, since Graphite seems to have messed the automatic rebase up. |
Merge activity
|
ritter-x2a
deleted the
users/ritter-x2a/09-10-_amdgpu_generate_canonical_additions_in_amdgpupromotealloca
branch
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When we know that one operand of an addition is a constant, we might was
well put it on the right-hand side and avoid the work to canonicalize it
in a later pass.