3cfd0c0d36
This change adds support for correctly lowering the `__scoped` Clang builtins, and corresponding scoped LLVM instructions. These were previously unconditionally lowered to Device scope, which is possibly incorrect. Furthermore, the default / implicit scope is changed from Device (an OpenCL assumption) to AllSvmDevices (aka System), since the SPIR-V BE is not OpenCL specific / can ingest IR coming from other language front-ends. OpenCL defaulting to Device scope is now reflected in the front-end handling of atomic ops, which seems preferable.
55 lines
2.1 KiB
LLVM
55 lines
2.1 KiB
LLVM
; RUN: llc -O0 -mtriple=spirv64-unknown-unknown %s -o - | FileCheck %s
|
|
; RUN: %if spirv-tools %{ llc -O0 -mtriple=spirv64-unknown-unknown %s -o - -filetype=obj | spirv-val %}
|
|
|
|
; CHECK-DAG: OpName %[[#GetScope:]] "_Z8getScopev"
|
|
; CHECK-DAG: %[[#Long:]] = OpTypeInt 32 0
|
|
; CHECK-DAG: %[[#WrkGrpConst2:]] = OpConstant %[[#Long]] 2
|
|
; CHECK-DAG: %[[#ScopeAllSvmDevices:]] = OpConstantNull %[[#Long]]
|
|
; CHECK-DAG: %[[#InvocationConst4:]] = OpConstant %[[#Long]] 4
|
|
; CHECK-DAG: %[[#Const8:]] = OpConstant %[[#Long]] 8
|
|
; CHECK-DAG: %[[#Const16:]] = OpConstant %[[#Long]] 16
|
|
; CHECK-DAG: %[[#Const3:]] = OpConstant %[[#Long]] 3
|
|
; CHECK-DAG: %[[#Const912:]] = OpConstant %[[#Long]] 912
|
|
; CHECK: OpMemoryBarrier %[[#ScopeAllSvmDevices]] %[[#WrkGrpConst2]]
|
|
; CHECK: OpMemoryBarrier %[[#ScopeAllSvmDevices]] %[[#InvocationConst4]]
|
|
; CHECK: OpMemoryBarrier %[[#ScopeAllSvmDevices]] %[[#Const8]]
|
|
; CHECK: OpMemoryBarrier %[[#InvocationConst4]] %[[#Const16]]
|
|
; CHECK: OpMemoryBarrier %[[#WrkGrpConst2]] %[[#InvocationConst4]]
|
|
; CHECK: OpFunctionEnd
|
|
; CHECK: %[[#ScopeId:]] = OpFunctionCall %[[#Long]] %[[#GetScope]]
|
|
; CHECK: OpControlBarrier %[[#Const3]] %[[#ScopeId:]] %[[#Const912]]
|
|
|
|
define spir_kernel void @fence_test_kernel1(ptr addrspace(1) noalias %s.ascast) {
|
|
fence acquire
|
|
ret void
|
|
}
|
|
|
|
define spir_kernel void @fence_test_kernel2(ptr addrspace(1) noalias %s.ascast) {
|
|
fence release
|
|
ret void
|
|
}
|
|
|
|
define spir_kernel void @fence_test_kernel3(ptr addrspace(1) noalias %s.ascast) {
|
|
fence acq_rel
|
|
ret void
|
|
}
|
|
|
|
define spir_kernel void @fence_test_kernel4(ptr addrspace(1) noalias %s.ascast) {
|
|
fence syncscope("singlethread") seq_cst
|
|
ret void
|
|
}
|
|
|
|
define spir_kernel void @fence_test_kernel5(ptr addrspace(1) noalias %s.ascast) {
|
|
fence syncscope("workgroup") release
|
|
ret void
|
|
}
|
|
|
|
define spir_func void @barrier_test1() {
|
|
%scope = call noundef i32 @_Z8getScopev()
|
|
call void @_Z22__spirv_ControlBarrieriii(i32 noundef 3, i32 noundef %scope, i32 noundef 912)
|
|
ret void
|
|
}
|
|
|
|
declare spir_func void @_Z22__spirv_ControlBarrieriii(i32 noundef, i32 noundef, i32 noundef)
|
|
declare spir_func i32 @_Z8getScopev()
|