summaryrefslogtreecommitdiff
path: root/test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll
diff options
context:
space:
mode:
Diffstat (limited to 'test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll')
-rw-r--r--test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll100
1 files changed, 50 insertions, 50 deletions
diff --git a/test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll b/test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll
index 084a6933da267..e2620ce353c60 100644
--- a/test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll
+++ b/test/CodeGen/AMDGPU/annotate-kernel-features-hsa.ll
@@ -11,22 +11,22 @@ declare i32 @llvm.amdgcn.workitem.id.z() #0
declare i8 addrspace(2)* @llvm.amdgcn.dispatch.ptr() #0
declare i8 addrspace(2)* @llvm.amdgcn.queue.ptr() #0
-; HSA: define void @use_tgid_x(i32 addrspace(1)* %ptr) #1 {
-define void @use_tgid_x(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_tgid_x(i32 addrspace(1)* %ptr) #1 {
+define amdgpu_kernel void @use_tgid_x(i32 addrspace(1)* %ptr) #1 {
%val = call i32 @llvm.amdgcn.workgroup.id.x()
store i32 %val, i32 addrspace(1)* %ptr
ret void
}
-; HSA: define void @use_tgid_y(i32 addrspace(1)* %ptr) #2 {
-define void @use_tgid_y(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_tgid_y(i32 addrspace(1)* %ptr) #2 {
+define amdgpu_kernel void @use_tgid_y(i32 addrspace(1)* %ptr) #1 {
%val = call i32 @llvm.amdgcn.workgroup.id.y()
store i32 %val, i32 addrspace(1)* %ptr
ret void
}
-; HSA: define void @multi_use_tgid_y(i32 addrspace(1)* %ptr) #2 {
-define void @multi_use_tgid_y(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @multi_use_tgid_y(i32 addrspace(1)* %ptr) #2 {
+define amdgpu_kernel void @multi_use_tgid_y(i32 addrspace(1)* %ptr) #1 {
%val0 = call i32 @llvm.amdgcn.workgroup.id.y()
store volatile i32 %val0, i32 addrspace(1)* %ptr
%val1 = call i32 @llvm.amdgcn.workgroup.id.y()
@@ -34,8 +34,8 @@ define void @multi_use_tgid_y(i32 addrspace(1)* %ptr) #1 {
ret void
}
-; HSA: define void @use_tgid_x_y(i32 addrspace(1)* %ptr) #2 {
-define void @use_tgid_x_y(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_tgid_x_y(i32 addrspace(1)* %ptr) #2 {
+define amdgpu_kernel void @use_tgid_x_y(i32 addrspace(1)* %ptr) #1 {
%val0 = call i32 @llvm.amdgcn.workgroup.id.x()
%val1 = call i32 @llvm.amdgcn.workgroup.id.y()
store volatile i32 %val0, i32 addrspace(1)* %ptr
@@ -43,15 +43,15 @@ define void @use_tgid_x_y(i32 addrspace(1)* %ptr) #1 {
ret void
}
-; HSA: define void @use_tgid_z(i32 addrspace(1)* %ptr) #3 {
-define void @use_tgid_z(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_tgid_z(i32 addrspace(1)* %ptr) #3 {
+define amdgpu_kernel void @use_tgid_z(i32 addrspace(1)* %ptr) #1 {
%val = call i32 @llvm.amdgcn.workgroup.id.z()
store i32 %val, i32 addrspace(1)* %ptr
ret void
}
-; HSA: define void @use_tgid_x_z(i32 addrspace(1)* %ptr) #3 {
-define void @use_tgid_x_z(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_tgid_x_z(i32 addrspace(1)* %ptr) #3 {
+define amdgpu_kernel void @use_tgid_x_z(i32 addrspace(1)* %ptr) #1 {
%val0 = call i32 @llvm.amdgcn.workgroup.id.x()
%val1 = call i32 @llvm.amdgcn.workgroup.id.z()
store volatile i32 %val0, i32 addrspace(1)* %ptr
@@ -59,8 +59,8 @@ define void @use_tgid_x_z(i32 addrspace(1)* %ptr) #1 {
ret void
}
-; HSA: define void @use_tgid_y_z(i32 addrspace(1)* %ptr) #4 {
-define void @use_tgid_y_z(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_tgid_y_z(i32 addrspace(1)* %ptr) #4 {
+define amdgpu_kernel void @use_tgid_y_z(i32 addrspace(1)* %ptr) #1 {
%val0 = call i32 @llvm.amdgcn.workgroup.id.y()
%val1 = call i32 @llvm.amdgcn.workgroup.id.z()
store volatile i32 %val0, i32 addrspace(1)* %ptr
@@ -68,8 +68,8 @@ define void @use_tgid_y_z(i32 addrspace(1)* %ptr) #1 {
ret void
}
-; HSA: define void @use_tgid_x_y_z(i32 addrspace(1)* %ptr) #4 {
-define void @use_tgid_x_y_z(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_tgid_x_y_z(i32 addrspace(1)* %ptr) #4 {
+define amdgpu_kernel void @use_tgid_x_y_z(i32 addrspace(1)* %ptr) #1 {
%val0 = call i32 @llvm.amdgcn.workgroup.id.x()
%val1 = call i32 @llvm.amdgcn.workgroup.id.y()
%val2 = call i32 @llvm.amdgcn.workgroup.id.z()
@@ -79,29 +79,29 @@ define void @use_tgid_x_y_z(i32 addrspace(1)* %ptr) #1 {
ret void
}
-; HSA: define void @use_tidig_x(i32 addrspace(1)* %ptr) #1 {
-define void @use_tidig_x(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_tidig_x(i32 addrspace(1)* %ptr) #1 {
+define amdgpu_kernel void @use_tidig_x(i32 addrspace(1)* %ptr) #1 {
%val = call i32 @llvm.amdgcn.workitem.id.x()
store i32 %val, i32 addrspace(1)* %ptr
ret void
}
-; HSA: define void @use_tidig_y(i32 addrspace(1)* %ptr) #5 {
-define void @use_tidig_y(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_tidig_y(i32 addrspace(1)* %ptr) #5 {
+define amdgpu_kernel void @use_tidig_y(i32 addrspace(1)* %ptr) #1 {
%val = call i32 @llvm.amdgcn.workitem.id.y()
store i32 %val, i32 addrspace(1)* %ptr
ret void
}
-; HSA: define void @use_tidig_z(i32 addrspace(1)* %ptr) #6 {
-define void @use_tidig_z(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_tidig_z(i32 addrspace(1)* %ptr) #6 {
+define amdgpu_kernel void @use_tidig_z(i32 addrspace(1)* %ptr) #1 {
%val = call i32 @llvm.amdgcn.workitem.id.z()
store i32 %val, i32 addrspace(1)* %ptr
ret void
}
-; HSA: define void @use_tidig_x_tgid_x(i32 addrspace(1)* %ptr) #1 {
-define void @use_tidig_x_tgid_x(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_tidig_x_tgid_x(i32 addrspace(1)* %ptr) #1 {
+define amdgpu_kernel void @use_tidig_x_tgid_x(i32 addrspace(1)* %ptr) #1 {
%val0 = call i32 @llvm.amdgcn.workitem.id.x()
%val1 = call i32 @llvm.amdgcn.workgroup.id.x()
store volatile i32 %val0, i32 addrspace(1)* %ptr
@@ -109,8 +109,8 @@ define void @use_tidig_x_tgid_x(i32 addrspace(1)* %ptr) #1 {
ret void
}
-; HSA: define void @use_tidig_y_tgid_y(i32 addrspace(1)* %ptr) #7 {
-define void @use_tidig_y_tgid_y(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_tidig_y_tgid_y(i32 addrspace(1)* %ptr) #7 {
+define amdgpu_kernel void @use_tidig_y_tgid_y(i32 addrspace(1)* %ptr) #1 {
%val0 = call i32 @llvm.amdgcn.workitem.id.y()
%val1 = call i32 @llvm.amdgcn.workgroup.id.y()
store volatile i32 %val0, i32 addrspace(1)* %ptr
@@ -118,8 +118,8 @@ define void @use_tidig_y_tgid_y(i32 addrspace(1)* %ptr) #1 {
ret void
}
-; HSA: define void @use_tidig_x_y_z(i32 addrspace(1)* %ptr) #8 {
-define void @use_tidig_x_y_z(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_tidig_x_y_z(i32 addrspace(1)* %ptr) #8 {
+define amdgpu_kernel void @use_tidig_x_y_z(i32 addrspace(1)* %ptr) #1 {
%val0 = call i32 @llvm.amdgcn.workitem.id.x()
%val1 = call i32 @llvm.amdgcn.workitem.id.y()
%val2 = call i32 @llvm.amdgcn.workitem.id.z()
@@ -129,8 +129,8 @@ define void @use_tidig_x_y_z(i32 addrspace(1)* %ptr) #1 {
ret void
}
-; HSA: define void @use_all_workitems(i32 addrspace(1)* %ptr) #9 {
-define void @use_all_workitems(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_all_workitems(i32 addrspace(1)* %ptr) #9 {
+define amdgpu_kernel void @use_all_workitems(i32 addrspace(1)* %ptr) #1 {
%val0 = call i32 @llvm.amdgcn.workitem.id.x()
%val1 = call i32 @llvm.amdgcn.workitem.id.y()
%val2 = call i32 @llvm.amdgcn.workitem.id.z()
@@ -146,8 +146,8 @@ define void @use_all_workitems(i32 addrspace(1)* %ptr) #1 {
ret void
}
-; HSA: define void @use_dispatch_ptr(i32 addrspace(1)* %ptr) #10 {
-define void @use_dispatch_ptr(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_dispatch_ptr(i32 addrspace(1)* %ptr) #10 {
+define amdgpu_kernel void @use_dispatch_ptr(i32 addrspace(1)* %ptr) #1 {
%dispatch.ptr = call i8 addrspace(2)* @llvm.amdgcn.dispatch.ptr()
%bc = bitcast i8 addrspace(2)* %dispatch.ptr to i32 addrspace(2)*
%val = load i32, i32 addrspace(2)* %bc
@@ -155,8 +155,8 @@ define void @use_dispatch_ptr(i32 addrspace(1)* %ptr) #1 {
ret void
}
-; HSA: define void @use_queue_ptr(i32 addrspace(1)* %ptr) #11 {
-define void @use_queue_ptr(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_queue_ptr(i32 addrspace(1)* %ptr) #11 {
+define amdgpu_kernel void @use_queue_ptr(i32 addrspace(1)* %ptr) #1 {
%dispatch.ptr = call i8 addrspace(2)* @llvm.amdgcn.queue.ptr()
%bc = bitcast i8 addrspace(2)* %dispatch.ptr to i32 addrspace(2)*
%val = load i32, i32 addrspace(2)* %bc
@@ -164,58 +164,58 @@ define void @use_queue_ptr(i32 addrspace(1)* %ptr) #1 {
ret void
}
-; HSA: define void @use_group_to_flat_addrspacecast(i32 addrspace(3)* %ptr) #11 {
-define void @use_group_to_flat_addrspacecast(i32 addrspace(3)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_group_to_flat_addrspacecast(i32 addrspace(3)* %ptr) #11 {
+define amdgpu_kernel void @use_group_to_flat_addrspacecast(i32 addrspace(3)* %ptr) #1 {
%stof = addrspacecast i32 addrspace(3)* %ptr to i32 addrspace(4)*
store volatile i32 0, i32 addrspace(4)* %stof
ret void
}
-; HSA: define void @use_private_to_flat_addrspacecast(i32* %ptr) #11 {
-define void @use_private_to_flat_addrspacecast(i32* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_private_to_flat_addrspacecast(i32* %ptr) #11 {
+define amdgpu_kernel void @use_private_to_flat_addrspacecast(i32* %ptr) #1 {
%stof = addrspacecast i32* %ptr to i32 addrspace(4)*
store volatile i32 0, i32 addrspace(4)* %stof
ret void
}
-; HSA: define void @use_flat_to_group_addrspacecast(i32 addrspace(4)* %ptr) #1 {
-define void @use_flat_to_group_addrspacecast(i32 addrspace(4)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_flat_to_group_addrspacecast(i32 addrspace(4)* %ptr) #1 {
+define amdgpu_kernel void @use_flat_to_group_addrspacecast(i32 addrspace(4)* %ptr) #1 {
%ftos = addrspacecast i32 addrspace(4)* %ptr to i32 addrspace(3)*
store volatile i32 0, i32 addrspace(3)* %ftos
ret void
}
-; HSA: define void @use_flat_to_private_addrspacecast(i32 addrspace(4)* %ptr) #1 {
-define void @use_flat_to_private_addrspacecast(i32 addrspace(4)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_flat_to_private_addrspacecast(i32 addrspace(4)* %ptr) #1 {
+define amdgpu_kernel void @use_flat_to_private_addrspacecast(i32 addrspace(4)* %ptr) #1 {
%ftos = addrspacecast i32 addrspace(4)* %ptr to i32*
store volatile i32 0, i32* %ftos
ret void
}
; No-op addrspacecast should not use queue ptr
-; HSA: define void @use_global_to_flat_addrspacecast(i32 addrspace(1)* %ptr) #1 {
-define void @use_global_to_flat_addrspacecast(i32 addrspace(1)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_global_to_flat_addrspacecast(i32 addrspace(1)* %ptr) #1 {
+define amdgpu_kernel void @use_global_to_flat_addrspacecast(i32 addrspace(1)* %ptr) #1 {
%stof = addrspacecast i32 addrspace(1)* %ptr to i32 addrspace(4)*
store volatile i32 0, i32 addrspace(4)* %stof
ret void
}
-; HSA: define void @use_constant_to_flat_addrspacecast(i32 addrspace(2)* %ptr) #1 {
-define void @use_constant_to_flat_addrspacecast(i32 addrspace(2)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_constant_to_flat_addrspacecast(i32 addrspace(2)* %ptr) #1 {
+define amdgpu_kernel void @use_constant_to_flat_addrspacecast(i32 addrspace(2)* %ptr) #1 {
%stof = addrspacecast i32 addrspace(2)* %ptr to i32 addrspace(4)*
%ld = load volatile i32, i32 addrspace(4)* %stof
ret void
}
-; HSA: define void @use_flat_to_global_addrspacecast(i32 addrspace(4)* %ptr) #1 {
-define void @use_flat_to_global_addrspacecast(i32 addrspace(4)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_flat_to_global_addrspacecast(i32 addrspace(4)* %ptr) #1 {
+define amdgpu_kernel void @use_flat_to_global_addrspacecast(i32 addrspace(4)* %ptr) #1 {
%ftos = addrspacecast i32 addrspace(4)* %ptr to i32 addrspace(1)*
store volatile i32 0, i32 addrspace(1)* %ftos
ret void
}
-; HSA: define void @use_flat_to_constant_addrspacecast(i32 addrspace(4)* %ptr) #1 {
-define void @use_flat_to_constant_addrspacecast(i32 addrspace(4)* %ptr) #1 {
+; HSA: define amdgpu_kernel void @use_flat_to_constant_addrspacecast(i32 addrspace(4)* %ptr) #1 {
+define amdgpu_kernel void @use_flat_to_constant_addrspacecast(i32 addrspace(4)* %ptr) #1 {
%ftos = addrspacecast i32 addrspace(4)* %ptr to i32 addrspace(2)*
%ld = load volatile i32, i32 addrspace(2)* %ftos
ret void
generated by cgit v1.2.3 (git 2.25.1) at 2025-06-15 03:39:09 +0000