summaryrefslogtreecommitdiff
path: root/test/CodeGen/AMDGPU/large-work-group-promote-alloca.ll
diff options
context:
space:
mode:
Diffstat (limited to 'test/CodeGen/AMDGPU/large-work-group-promote-alloca.ll')
-rw-r--r--test/CodeGen/AMDGPU/large-work-group-promote-alloca.ll72
1 files changed, 41 insertions, 31 deletions
diff --git a/test/CodeGen/AMDGPU/large-work-group-promote-alloca.ll b/test/CodeGen/AMDGPU/large-work-group-promote-alloca.ll
index 906a688febd24..13dd7058c50a5 100644
--- a/test/CodeGen/AMDGPU/large-work-group-promote-alloca.ll
+++ b/test/CodeGen/AMDGPU/large-work-group-promote-alloca.ll
@@ -1,8 +1,10 @@
-; RUN: opt -S -mtriple=amdgcn-unknown-unknown -amdgpu-promote-alloca < %s | FileCheck %s
+; RUN: opt -S -mtriple=amdgcn-unknown-unknown -amdgpu-promote-alloca < %s | FileCheck --check-prefix=SI --check-prefix=ALL %s
+; RUN: opt -S -mcpu=tonga -mtriple=amdgcn-unknown-unknown -amdgpu-promote-alloca < %s | FileCheck --check-prefix=CI --check-prefix=ALL %s
-; CHECK: @promote_alloca_size_63.stack = internal unnamed_addr addrspace(3) global [63 x [5 x i32]] undef, align 4
+; SI-NOT: @promote_alloca_size_63.stack = internal unnamed_addr addrspace(3) global [63 x [5 x i32]] undef, align 4
+; CI: @promote_alloca_size_63.stack = internal unnamed_addr addrspace(3) global [63 x [5 x i32]] undef, align 4
-define void @promote_alloca_size_63(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #0 {
+define amdgpu_kernel void @promote_alloca_size_63(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #0 {
entry:
%stack = alloca [5 x i32], align 4
%0 = load i32, i32 addrspace(1)* %in, align 4
@@ -22,9 +24,9 @@ entry:
ret void
}
-; CHECK: @promote_alloca_size_256.stack = internal unnamed_addr addrspace(3) global [256 x [5 x i32]] undef, align 4
+; ALL: @promote_alloca_size_256.stack = internal unnamed_addr addrspace(3) global [256 x [5 x i32]] undef, align 4
-define void @promote_alloca_size_256(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #1 {
+define amdgpu_kernel void @promote_alloca_size_256(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #1 {
entry:
%stack = alloca [5 x i32], align 4
%0 = load i32, i32 addrspace(1)* %in, align 4
@@ -44,9 +46,9 @@ entry:
ret void
}
-; CHECK: @promote_alloca_size_1600.stack = internal unnamed_addr addrspace(3) global [1600 x [5 x i32]] undef, align 4
+; ALL: @promote_alloca_size_1600.stack = internal unnamed_addr addrspace(3) global [1600 x [5 x i32]] undef, align 4
-define void @promote_alloca_size_1600(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #2 {
+define amdgpu_kernel void @promote_alloca_size_1600(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #2 {
entry:
%stack = alloca [5 x i32], align 4
%0 = load i32, i32 addrspace(1)* %in, align 4
@@ -66,9 +68,10 @@ entry:
ret void
}
-; CHECK-LABEL: @occupancy_0(
-; CHECK: alloca [5 x i32]
-define void @occupancy_0(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #3 {
+; ALL-LABEL: @occupancy_0(
+; CI-NOT: alloca [5 x i32]
+; SI: alloca [5 x i32]
+define amdgpu_kernel void @occupancy_0(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #3 {
entry:
%stack = alloca [5 x i32], align 4
%0 = load i32, i32 addrspace(1)* %in, align 4
@@ -88,9 +91,10 @@ entry:
ret void
}
-; CHECK-LABEL: @occupancy_max(
-; CHECK: alloca [5 x i32]
-define void @occupancy_max(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #4 {
+; ALL-LABEL: @occupancy_max(
+; CI-NOT: alloca [5 x i32]
+; SI: alloca [5 x i32]
+define amdgpu_kernel void @occupancy_max(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #4 {
entry:
%stack = alloca [5 x i32], align 4
%0 = load i32, i32 addrspace(1)* %in, align 4
@@ -110,9 +114,11 @@ entry:
ret void
}
-; CHECK-LABEL: @occupancy_6(
-; CHECK-NOT: alloca
-define void @occupancy_6(i8 addrspace(1)* nocapture %out, i8 addrspace(1)* nocapture %in) #5 {
+; SI-LABEL: @occupancy_6(
+; CI-LABEL: @occupancy_6(
+; SI: alloca
+; CI-NOT: alloca
+define amdgpu_kernel void @occupancy_6(i8 addrspace(1)* nocapture %out, i8 addrspace(1)* nocapture %in) #5 {
entry:
%stack = alloca [42 x i8], align 4
%tmp = load i8, i8 addrspace(1)* %in, align 1
@@ -134,9 +140,9 @@ entry:
ret void
}
-; CHECK-LABEL: @occupancy_6_over(
-; CHECK: alloca [43 x i8]
-define void @occupancy_6_over(i8 addrspace(1)* nocapture %out, i8 addrspace(1)* nocapture %in) #5 {
+; ALL-LABEL: @occupancy_6_over(
+; ALL: alloca [43 x i8]
+define amdgpu_kernel void @occupancy_6_over(i8 addrspace(1)* nocapture %out, i8 addrspace(1)* nocapture %in) #5 {
entry:
%stack = alloca [43 x i8], align 4
%tmp = load i8, i8 addrspace(1)* %in, align 1
@@ -158,9 +164,11 @@ entry:
ret void
}
-; CHECK-LABEL: @occupancy_8(
-; CHECK-NOT: alloca
-define void @occupancy_8(i8 addrspace(1)* nocapture %out, i8 addrspace(1)* nocapture %in) #6 {
+; SI-LABEL: @occupancy_8(
+; CI-LABEL: @occupancy_8(
+; SI: alloca
+; CI-NOT: alloca
+define amdgpu_kernel void @occupancy_8(i8 addrspace(1)* nocapture %out, i8 addrspace(1)* nocapture %in) #6 {
entry:
%stack = alloca [32 x i8], align 4
%tmp = load i8, i8 addrspace(1)* %in, align 1
@@ -182,9 +190,9 @@ entry:
ret void
}
-; CHECK-LABEL: @occupancy_8_over(
-; CHECK: alloca [33 x i8]
-define void @occupancy_8_over(i8 addrspace(1)* nocapture %out, i8 addrspace(1)* nocapture %in) #6 {
+; ALL-LABEL: @occupancy_8_over(
+; ALL: alloca [33 x i8]
+define amdgpu_kernel void @occupancy_8_over(i8 addrspace(1)* nocapture %out, i8 addrspace(1)* nocapture %in) #6 {
entry:
%stack = alloca [33 x i8], align 4
%tmp = load i8, i8 addrspace(1)* %in, align 1
@@ -206,9 +214,11 @@ entry:
ret void
}
-; CHECK-LABEL: @occupancy_9(
-; CHECK-NOT: alloca
-define void @occupancy_9(i8 addrspace(1)* nocapture %out, i8 addrspace(1)* nocapture %in) #7 {
+; SI-LABEL: @occupancy_9(
+; CI-LABEL: @occupancy_9(
+; SI: alloca
+; CI-NOT: alloca
+define amdgpu_kernel void @occupancy_9(i8 addrspace(1)* nocapture %out, i8 addrspace(1)* nocapture %in) #7 {
entry:
%stack = alloca [28 x i8], align 4
%tmp = load i8, i8 addrspace(1)* %in, align 1
@@ -230,9 +240,9 @@ entry:
ret void
}
-; CHECK-LABEL: @occupancy_9_over(
-; CHECK: alloca [29 x i8]
-define void @occupancy_9_over(i8 addrspace(1)* nocapture %out, i8 addrspace(1)* nocapture %in) #7 {
+; ALL-LABEL: @occupancy_9_over(
+; ALL: alloca [29 x i8]
+define amdgpu_kernel void @occupancy_9_over(i8 addrspace(1)* nocapture %out, i8 addrspace(1)* nocapture %in) #7 {
entry:
%stack = alloca [29 x i8], align 4
%tmp = load i8, i8 addrspace(1)* %in, align 1
generated by cgit v1.2.3 (git 2.25.1) at 2025-08-03 04:09:58 +0000