AMDGPU: Add regression test for multiple frame index lowering #140784
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
Member
|
@llvm/pr-subscribers-backend-amdgpu Author: Matt Arsenault (arsenm) ChangesFull diff: https://github.com/llvm/llvm-project/pull/140784.diff 1 Files Affected:
diff --git a/llvm/test/CodeGen/AMDGPU/frame-index-elimination.ll b/llvm/test/CodeGen/AMDGPU/frame-index-elimination.ll
index ee62359cffc63..aea6329d56885 100644
--- a/llvm/test/CodeGen/AMDGPU/frame-index-elimination.ll
+++ b/llvm/test/CodeGen/AMDGPU/frame-index-elimination.ll
@@ -424,4 +424,115 @@ vector.body.i.i.i.i: ; preds = %.shuffle.then.i.i.i
ret void
}
+; Check that we do not produce a verifier error after prolog
+; epilog. alloca1 and alloca2 will lower to literals.
+
+; GCN-LABEL: {{^}}s_multiple_frame_indexes_literal_offsets:
+; GCN: s_load_dword [[ARG0:s[0-9]+]]
+; GCN: s_movk_i32 [[ALLOCA1:s[0-9]+]], 0x44
+; GCN: s_cmp_eq_u32 [[ARG0]], 0
+; GCN: s_cselect_b32 [[SELECT:s[0-9]+]], [[ALLOCA1]], 0x48
+; GCN: s_mov_b32 [[ALLOCA0:s[0-9]+]], 0
+; GCN: ; use [[SELECT]], [[ALLOCA0]]
+define amdgpu_kernel void @s_multiple_frame_indexes_literal_offsets(i32 inreg %arg0) #0 {
+ %alloca0 = alloca [17 x i32], align 8, addrspace(5)
+ %alloca1 = alloca i32, align 4, addrspace(5)
+ %alloca2 = alloca i32, align 4, addrspace(5)
+ %cmp = icmp eq i32 %arg0, 0
+ %select = select i1 %cmp, ptr addrspace(5) %alloca1, ptr addrspace(5) %alloca2
+ call void asm sideeffect "; use $0, $1","s,s"(ptr addrspace(5) %select, ptr addrspace(5) %alloca0)
+ ret void
+}
+
+; %alloca1 or alloca2 will lower to an inline constant, and one will
+; %be a literal, so we could fold both indexes into the instruction.
+
+; GCN-LABEL: {{^}}s_multiple_frame_indexes_one_imm_one_literal_offset:
+; GCN: s_load_dword [[ARG0:s[0-9]+]]
+; GCN: s_mov_b32 [[ALLOCA1:s[0-9]+]], 64
+; GCN: s_cmp_eq_u32 [[ARG0]], 0
+; GCN: s_cselect_b32 [[SELECT:s[0-9]+]], [[ALLOCA1]], 0x44
+; GCN: s_mov_b32 [[ALLOCA0:s[0-9]+]], 0
+; GCN: ; use [[SELECT]], [[ALLOCA0]]
+define amdgpu_kernel void @s_multiple_frame_indexes_one_imm_one_literal_offset(i32 inreg %arg0) #0 {
+ %alloca0 = alloca [16 x i32], align 8, addrspace(5)
+ %alloca1 = alloca i32, align 4, addrspace(5)
+ %alloca2 = alloca i32, align 4, addrspace(5)
+ %cmp = icmp eq i32 %arg0, 0
+ %select = select i1 %cmp, ptr addrspace(5) %alloca1, ptr addrspace(5) %alloca2
+ call void asm sideeffect "; use $0, $1","s,s"(ptr addrspace(5) %select, ptr addrspace(5) %alloca0)
+ ret void
+}
+
+; GCN-LABEL: {{^}}s_multiple_frame_indexes_imm_offsets:
+; GCN: s_load_dword [[ARG0:s[0-9]+]]
+; GCN: s_mov_b32 [[ALLOCA1:s[0-9]+]], 16
+; GCN: s_cmp_eq_u32 [[ARG0]], 0
+; GCN: s_cselect_b32 [[SELECT:s[0-9]+]], [[ALLOCA1]], 20
+; GCN: s_mov_b32 [[ALLOCA0:s[0-9]+]], 0
+; GCN: ; use [[SELECT]], [[ALLOCA0]]
+define amdgpu_kernel void @s_multiple_frame_indexes_imm_offsets(i32 inreg %arg0) #0 {
+ %alloca0 = alloca [4 x i32], align 8, addrspace(5)
+ %alloca1 = alloca i32, align 4, addrspace(5)
+ %alloca2 = alloca i32, align 4, addrspace(5)
+ %cmp = icmp eq i32 %arg0, 0
+ %select = select i1 %cmp, ptr addrspace(5) %alloca1, ptr addrspace(5) %alloca2
+ call void asm sideeffect "; use $0, $1","s,s"(ptr addrspace(5) %select, ptr addrspace(5) %alloca0)
+ ret void
+}
+
+; GCN-LABEL: {{^}}v_multiple_frame_indexes_literal_offsets:
+; GCN: v_mov_b32_e32 [[ALLOCA1:v[0-9]+]], 0x48
+; GCN: v_mov_b32_e32 [[ALLOCA2:v[0-9]+]], 0x44
+; GCN: v_cmp_eq_u32_e32 vcc, 0, v0
+; GCN: v_cndmask_b32_e32 [[SELECT:v[0-9]+]], [[ALLOCA1]], [[ALLOCA2]], vcc
+; GCN: v_mov_b32_e32 [[ALLOCA0:v[0-9]+]], 0{{$}}
+; GCN: ; use [[SELECT]], [[ALLOCA0]]
+define amdgpu_kernel void @v_multiple_frame_indexes_literal_offsets() #0 {
+ %vgpr = call i32 @llvm.amdgcn.workitem.id.x()
+ %alloca0 = alloca [17 x i32], align 8, addrspace(5)
+ %alloca1 = alloca i32, align 4, addrspace(5)
+ %alloca2 = alloca i32, align 4, addrspace(5)
+ %cmp = icmp eq i32 %vgpr, 0
+ %select = select i1 %cmp, ptr addrspace(5) %alloca1, ptr addrspace(5) %alloca2
+ call void asm sideeffect "; use $0, $1","v,v"(ptr addrspace(5) %select, ptr addrspace(5) %alloca0)
+ ret void
+}
+
+; GCN-LABEL: {{^}}v_multiple_frame_indexes_one_imm_one_literal_offset:
+; GCN: v_mov_b32_e32 [[ALLOCA1:v[0-9]+]], 0x44
+; GCN: v_mov_b32_e32 [[ALLOCA2:v[0-9]+]], 64
+; GCN: v_cmp_eq_u32_e32 vcc, 0, v0
+; GCN: v_cndmask_b32_e32 [[SELECT:v[0-9]+]], [[ALLOCA1]], [[ALLOCA2]], vcc
+; GCN: v_mov_b32_e32 [[ALLOCA0:v[0-9]+]], 0{{$}}
+; GCN: ; use [[SELECT]], [[ALLOCA0]]
+define amdgpu_kernel void @v_multiple_frame_indexes_one_imm_one_literal_offset() #0 {
+ %vgpr = call i32 @llvm.amdgcn.workitem.id.x()
+ %alloca0 = alloca [16 x i32], align 8, addrspace(5)
+ %alloca1 = alloca i32, align 4, addrspace(5)
+ %alloca2 = alloca i32, align 4, addrspace(5)
+ %cmp = icmp eq i32 %vgpr, 0
+ %select = select i1 %cmp, ptr addrspace(5) %alloca1, ptr addrspace(5) %alloca2
+ call void asm sideeffect "; use $0, $1","v,v"(ptr addrspace(5) %select, ptr addrspace(5) %alloca0)
+ ret void
+}
+
+; GCN-LABEL: {{^}}v_multiple_frame_indexes_imm_offsets:
+; GCN: v_mov_b32_e32 [[ALLOCA1:v[0-9]+]], 12
+; GCN: v_mov_b32_e32 [[ALLOCA2:v[0-9]+]], 8
+; GCN: v_cmp_eq_u32_e32 vcc, 0, v0
+; GCN: v_cndmask_b32_e32 [[SELECT:v[0-9]+]], [[ALLOCA1]], [[ALLOCA2]], vcc
+; GCN: v_mov_b32_e32 [[ALLOCA0:v[0-9]+]], 0{{$}}
+; GCN: ; use [[SELECT]], [[ALLOCA0]]
+define amdgpu_kernel void @v_multiple_frame_indexes_imm_offsets() #0 {
+ %vgpr = call i32 @llvm.amdgcn.workitem.id.x()
+ %alloca0 = alloca [2 x i32], align 8, addrspace(5)
+ %alloca1 = alloca i32, align 4, addrspace(5)
+ %alloca2 = alloca i32, align 4, addrspace(5)
+ %cmp = icmp eq i32 %vgpr, 0
+ %select = select i1 %cmp, ptr addrspace(5) %alloca1, ptr addrspace(5) %alloca2
+ call void asm sideeffect "; use $0, $1","v,v"(ptr addrspace(5) %select, ptr addrspace(5) %alloca0)
+ ret void
+}
+
attributes #0 = { nounwind }
|
arsenm
requested review from
cdevadas,
dstutt,
jayfoad,
mbrkusanin,
rampitec and
shiltian
rampitec
approved these changes
May 20, 2025
arsenm
commented
May 20, 2025
Contributor
Author
arsenm
deleted the
users/arsenm/amdgpu/add-frame-index-elimination-lowering-test
branch
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Failures appeared after #140587 but this case wasn't covered