| ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-attributes --check-globals all --version 4 | |
| ; RUN: opt -S -mtriple=amdgpu9.0a-unknown-amdhsa -passes=amdgpu-attributor %s | FileCheck %s | |
| ; Shrink result attribute list by preventing use of most attributes. | |
| define internal void @use_most() { | |
| ; CHECK-LABEL: define internal void @use_most( | |
| ; CHECK-SAME: ) #[[ATTR0:[0-9]+]] { | |
| ; CHECK-NEXT: [[ALLOCA:%.*]] = alloca [256 x i8], align 1, addrspace(5) | |
| ; CHECK-NEXT: [[ALLOCA_CAST:%.*]] = addrspacecast ptr addrspace(5) [[ALLOCA]] to ptr | |
| ; CHECK-NEXT: [[TMP1:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() | |
| ; CHECK-NEXT: [[TMP2:%.*]] = call i32 @llvm.amdgcn.workitem.id.y() | |
| ; CHECK-NEXT: [[TMP3:%.*]] = call i32 @llvm.amdgcn.workitem.id.z() | |
| ; CHECK-NEXT: [[TMP4:%.*]] = call i32 @llvm.amdgcn.workgroup.id.x() | |
| ; CHECK-NEXT: [[TMP5:%.*]] = call i32 @llvm.amdgcn.workgroup.id.y() | |
| ; CHECK-NEXT: [[TMP6:%.*]] = call i32 @llvm.amdgcn.workgroup.id.z() | |
| ; CHECK-NEXT: [[TMP10:%.*]] = call i32 @llvm.amdgcn.cluster.id.x() | |
| ; CHECK-NEXT: [[TMP11:%.*]] = call i32 @llvm.amdgcn.cluster.id.y() | |
| ; CHECK-NEXT: [[TMP12:%.*]] = call i32 @llvm.amdgcn.cluster.id.z() | |
| ; CHECK-NEXT: [[TMP7:%.*]] = call ptr addrspace(4) @llvm.amdgcn.dispatch.ptr() | |
| ; CHECK-NEXT: [[TMP8:%.*]] = call ptr addrspace(4) @llvm.amdgcn.queue.ptr() | |
| ; CHECK-NEXT: [[TMP9:%.*]] = call i64 @llvm.amdgcn.dispatch.id() | |
| ; CHECK-NEXT: [[TMP13:%.*]] = call i32 @llvm.amdgcn.lds.kernel.id() | |
| ; CHECK-NEXT: [[IMPLICIT_ARG_PTR:%.*]] = call ptr addrspace(4) @llvm.amdgcn.implicitarg.ptr() | |
| ; CHECK-NEXT: call void @llvm.memcpy.p0.p4.i64(ptr [[ALLOCA_CAST]], ptr addrspace(4) [[IMPLICIT_ARG_PTR]], i64 256, i1 false) | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %alloca = alloca [256 x i8], addrspace(5) | |
| %alloca.cast = addrspacecast ptr addrspace(5) %alloca to ptr | |
| call i32 @llvm.amdgcn.workitem.id.x() | |
| call i32 @llvm.amdgcn.workitem.id.y() | |
| call i32 @llvm.amdgcn.workitem.id.z() | |
| call i32 @llvm.amdgcn.workgroup.id.x() | |
| call i32 @llvm.amdgcn.workgroup.id.y() | |
| call i32 @llvm.amdgcn.workgroup.id.z() | |
| call i32 @llvm.amdgcn.cluster.id.x() | |
| call i32 @llvm.amdgcn.cluster.id.y() | |
| call i32 @llvm.amdgcn.cluster.id.z() | |
| call ptr addrspace(4) @llvm.amdgcn.dispatch.ptr() | |
| call ptr addrspace(4) @llvm.amdgcn.queue.ptr() | |
| call i64 @llvm.amdgcn.dispatch.id() | |
| call i32 @llvm.amdgcn.lds.kernel.id() | |
| %implicit.arg.ptr = call ptr addrspace(4) @llvm.amdgcn.implicitarg.ptr() | |
| call void @llvm.memcpy.p0.p4(ptr %alloca.cast, ptr addrspace(4) %implicit.arg.ptr, i64 256, i1 false) | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_virtreg() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg( | |
| ; CHECK-SAME: ) #[[ATTR1:[0-9]+]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "a"(i32 poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_virtreg_def() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def( | |
| ; CHECK-SAME: ) #[[ATTR1]] { | |
| ; CHECK-NEXT: [[DEF:%.*]] = call i32 asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %def = call i32 asm sideeffect "; def $0", "=a"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_physreg_def_tuple() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_def_tuple( | |
| ; CHECK-SAME: ) #[[ATTR2:[0-9]+]] { | |
| ; CHECK-NEXT: [[DEF:%.*]] = call i64 asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %def = call i64 asm sideeffect "; def $0", "={a[0:1]}"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_virtreg_second_arg() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_second_arg( | |
| ; CHECK-SAME: ) #[[ATTR1]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "v,a"(i32 poison, i32 poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_non_agpr_asm() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_non_agpr_asm( | |
| ; CHECK-SAME: ) #[[ATTR3:[0-9]+]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "v"(i32 poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_physreg() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg( | |
| ; CHECK-SAME: ) #[[ATTR1]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "{a0}"(i32 poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_physreg_tuple() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_tuple( | |
| ; CHECK-SAME: ) #[[ATTR2]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "{a[0:1]}"(i64 poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define void @func_uses_asm_virtreg_agpr() { | |
| ; CHECK-LABEL: define void @func_uses_asm_virtreg_agpr( | |
| ; CHECK-SAME: ) #[[ATTR1]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "a"(i32 poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define void @func_uses_asm_physreg_agpr() { | |
| ; CHECK-LABEL: define void @func_uses_asm_physreg_agpr( | |
| ; CHECK-SAME: ) #[[ATTR1]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "{a0}"(i32 poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define void @func_uses_asm_physreg_agpr_tuple() { | |
| ; CHECK-LABEL: define void @func_uses_asm_physreg_agpr_tuple( | |
| ; CHECK-SAME: ) #[[ATTR2]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "{a[0:1]}"(i64 poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| declare void @unknown() | |
| define amdgpu_kernel void @kernel_calls_extern() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_extern() { | |
| ; CHECK-NEXT: call void @unknown() | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @unknown() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_calls_extern_marked_callsite() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_extern_marked_callsite() { | |
| ; CHECK-NEXT: call void @unknown() #[[ATTR37:[0-9]+]] | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @unknown() #0 | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_calls_indirect(ptr %indirect) { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_indirect( | |
| ; CHECK-SAME: ptr [[INDIRECT:%.*]]) { | |
| ; CHECK-NEXT: call void [[INDIRECT]]() | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void %indirect() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_calls_indirect_marked_callsite(ptr %indirect) { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_indirect_marked_callsite( | |
| ; CHECK-SAME: ptr [[INDIRECT:%.*]]) { | |
| ; CHECK-NEXT: call void [[INDIRECT]]() #[[ATTR37]] | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void %indirect() #0 | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_transitively_uses_agpr_asm() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_transitively_uses_agpr_asm( | |
| ; CHECK-SAME: ) #[[ATTR1]] { | |
| ; CHECK-NEXT: call void @func_uses_asm_physreg_agpr() | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @func_uses_asm_physreg_agpr() | |
| call void @use_most() | |
| ret void | |
| } | |
| define void @empty() { | |
| ; CHECK-LABEL: define void @empty( | |
| ; CHECK-SAME: ) #[[ATTR0]] { | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @use_most() | |
| ret void | |
| } | |
| define void @also_empty() { | |
| ; CHECK-LABEL: define void @also_empty( | |
| ; CHECK-SAME: ) #[[ATTR0]] { | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_calls_empty() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_empty( | |
| ; CHECK-SAME: ) #[[ATTR0]] { | |
| ; CHECK-NEXT: call void @empty() | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @empty() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_calls_non_agpr_and_agpr() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_non_agpr_and_agpr( | |
| ; CHECK-SAME: ) #[[ATTR1]] { | |
| ; CHECK-NEXT: call void @empty() | |
| ; CHECK-NEXT: call void @func_uses_asm_physreg_agpr() | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @empty() | |
| call void @func_uses_asm_physreg_agpr() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_calls_generic_intrinsic(ptr %ptr0, ptr %ptr1, i64 %size) { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_generic_intrinsic( | |
| ; CHECK-SAME: ptr [[PTR0:%.*]], ptr [[PTR1:%.*]], i64 [[SIZE:%.*]]) #[[ATTR0]] { | |
| ; CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr [[PTR0]], ptr [[PTR1]], i64 [[SIZE]], i1 false) | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @llvm.memcpy.p0.p0.i64(ptr %ptr0, ptr %ptr1, i64 %size, i1 false) | |
| call void @use_most() | |
| ret void | |
| } | |
| declare <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float, float, <32 x float>, i32 immarg, i32 immarg, i32 immarg) | |
| define amdgpu_kernel void @kernel_calls_mfma.f32.32x32x1f32(ptr addrspace(1) %out, float %a, float %b, <32 x float> %c) { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_mfma.f32.32x32x1f32( | |
| ; CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], float [[A:%.*]], float [[B:%.*]], <32 x float> [[C:%.*]]) #[[ATTR0]] { | |
| ; CHECK-NEXT: [[RESULT:%.*]] = call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float [[A]], float [[B]], <32 x float> [[C]], i32 0, i32 0, i32 0) | |
| ; CHECK-NEXT: store <32 x float> [[RESULT]], ptr addrspace(1) [[OUT]], align 128 | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %result = call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float %a, float %b, <32 x float> %c, i32 0, i32 0, i32 0) | |
| store <32 x float> %result, ptr addrspace(1) %out | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_calls_workitem_id_x(ptr addrspace(1) %out) { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_workitem_id_x( | |
| ; CHECK-SAME: ptr addrspace(1) [[OUT:%.*]]) #[[ATTR0]] { | |
| ; CHECK-NEXT: [[RESULT:%.*]] = call i32 @llvm.amdgcn.workitem.id.x() | |
| ; CHECK-NEXT: store i32 [[RESULT]], ptr addrspace(1) [[OUT]], align 4 | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %result = call i32 @llvm.amdgcn.workitem.id.x() | |
| store i32 %result, ptr addrspace(1) %out | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @indirect_calls_none_agpr(i1 %cond) { | |
| ; CHECK-LABEL: define amdgpu_kernel void @indirect_calls_none_agpr( | |
| ; CHECK-SAME: i1 [[COND:%.*]]) #[[ATTR0]] { | |
| ; CHECK-NEXT: [[FPTR:%.*]] = select i1 [[COND]], ptr @empty, ptr @also_empty | |
| ; CHECK-NEXT: [[TMP1:%.*]] = icmp eq ptr [[FPTR]], @also_empty | |
| ; CHECK-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP3:%.*]] | |
| ; CHECK: 2: | |
| ; CHECK-NEXT: call void @also_empty() | |
| ; CHECK-NEXT: br label [[TMP6:%.*]] | |
| ; CHECK: 3: | |
| ; CHECK-NEXT: br i1 true, label [[TMP4:%.*]], label [[TMP5:%.*]] | |
| ; CHECK: 4: | |
| ; CHECK-NEXT: call void @empty() | |
| ; CHECK-NEXT: br label [[TMP6]] | |
| ; CHECK: 5: | |
| ; CHECK-NEXT: unreachable | |
| ; CHECK: 6: | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %fptr = select i1 %cond, ptr @empty, ptr @also_empty | |
| call void %fptr() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_virtreg_def_struct_0() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def_struct_0( | |
| ; CHECK-SAME: ) #[[ATTR2]] { | |
| ; CHECK-NEXT: [[DEF:%.*]] = call { i32, i32 } asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %def = call {i32, i32} asm sideeffect "; def $0", "=a,=a"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_virtreg_use_struct_1() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_use_struct_1( | |
| ; CHECK-SAME: ) #[[ATTR5:[0-9]+]] { | |
| ; CHECK-NEXT: [[DEF:%.*]] = call { i32, <2 x i32> } asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %def = call {i32, <2 x i32>} asm sideeffect "; def $0", "=a,=a"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_virtreg_use_struct_2() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_use_struct_2( | |
| ; CHECK-SAME: ) #[[ATTR1]] { | |
| ; CHECK-NEXT: [[DEF:%.*]] = call { i32, <2 x i32> } asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %def = call {i32, <2 x i32>} asm sideeffect "; def $0", "=a,=v"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_virtreg_ptr_ty() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_ptr_ty( | |
| ; CHECK-SAME: ) #[[ATTR2]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "a"(ptr poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_virtreg_def_ptr_ty() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def_ptr_ty( | |
| ; CHECK-SAME: ) #[[ATTR2]] { | |
| ; CHECK-NEXT: [[DEF:%.*]] = call ptr asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %def = call ptr asm sideeffect "; def $0", "=a"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_virtreg_def_vector_ptr_ty() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def_vector_ptr_ty( | |
| ; CHECK-SAME: ) #[[ATTR5]] { | |
| ; CHECK-NEXT: [[DEF:%.*]] = call <2 x ptr> asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %def = call <2 x ptr> asm sideeffect "; def $0", "=a"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_physreg_def_struct_0() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_def_struct_0( | |
| ; CHECK-SAME: ) #[[ATTR6:[0-9]+]] { | |
| ; CHECK-NEXT: [[DEF:%.*]] = call { i32, i32 } asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %def = call {i32, i32} asm sideeffect "; def $0", "={a0},={a[4:5]}"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_clobber() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber( | |
| ; CHECK-SAME: ) #[[ATTR7:[0-9]+]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; clobber $0", "~{a4}"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_clobber_tuple() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber_tuple( | |
| ; CHECK-SAME: ) #[[ATTR8:[0-9]+]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; clobber $0", "~{a[10:13]}"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_clobber_oob() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber_oob( | |
| ; CHECK-SAME: ) #[[ATTR9:[0-9]+]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; clobber $0", "~{a256}"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_clobber_max() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_clobber_max( | |
| ; CHECK-SAME: ) #[[ATTR9]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; clobber $0", "~{a255}"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_physreg_oob() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_physreg_oob( | |
| ; CHECK-SAME: ) #[[ATTR9]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "{a256}"(i32 poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_virtreg_def_max_ty() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_def_max_ty( | |
| ; CHECK-SAME: ) #[[ATTR10:[0-9]+]] { | |
| ; CHECK-NEXT: [[DEF:%.*]] = call <32 x i32> asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %def = call <32 x i32> asm sideeffect "; def $0", "=a"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_virtreg_use_max_ty() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_use_max_ty( | |
| ; CHECK-SAME: ) #[[ATTR10]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "a"(<32 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_asm_virtreg_use_def_max_ty() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_asm_virtreg_use_def_max_ty( | |
| ; CHECK-SAME: ) #[[ATTR10]] { | |
| ; CHECK-NEXT: [[DEF:%.*]] = call <32 x i32> asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %def = call <32 x i32> asm sideeffect "; use $0", "=a,a"(<32 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @vreg_use_exceeds_register_file() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @vreg_use_exceeds_register_file( | |
| ; CHECK-SAME: ) #[[ATTR9]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "a"(<257 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @vreg_def_exceeds_register_file() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @vreg_def_exceeds_register_file( | |
| ; CHECK-SAME: ) #[[ATTR9]] { | |
| ; CHECK-NEXT: [[DEF:%.*]] = call <257 x i32> asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %def = call <257 x i32> asm sideeffect "; def $0", "=a"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @multiple() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @multiple( | |
| ; CHECK-SAME: ) #[[ATTR10]] { | |
| ; CHECK-NEXT: [[DEF:%.*]] = call { <16 x i32>, <8 x i32>, <8 x i32> } asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %def = call {<16 x i32>, <8 x i32>, <8 x i32>} asm sideeffect "; def $0", "=a,=a,=a,a,a,a"(<4 x i32> splat (i32 0), <8 x i32> splat (i32 1), i64 999) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @earlyclobber_0() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @earlyclobber_0( | |
| ; CHECK-SAME: ) #[[ATTR11:[0-9]+]] { | |
| ; CHECK-NEXT: [[DEF:%.*]] = call <8 x i32> asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %def = call <8 x i32> asm sideeffect "; def $0", "=&a,a"(i32 0) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @earlyclobber_1() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @earlyclobber_1( | |
| ; CHECK-SAME: ) #[[ATTR12:[0-9]+]] { | |
| ; CHECK-NEXT: [[DEF:%.*]] = call { <8 x i32>, <16 x i32> } asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %def = call { <8 x i32>, <16 x i32 > } asm sideeffect "; def $0, $1", "=&a,=&a,a,a"(i32 0, <16 x i32> splat (i32 1)) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @physreg_a32__vreg_a256__vreg_a512() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @physreg_a32__vreg_a256__vreg_a512( | |
| ; CHECK-SAME: ) #[[ATTR13:[0-9]+]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0, $1, $2", "{a16},a,a"(i32 poison, <8 x i32> poison, <16 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @physreg_def_a32__def_vreg_a256__def_vreg_a512() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @physreg_def_a32__def_vreg_a256__def_vreg_a512( | |
| ; CHECK-SAME: ) #[[ATTR13]] { | |
| ; CHECK-NEXT: [[TMP1:%.*]] = call { i32, <8 x i32>, <16 x i32> } asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call {i32, <8 x i32>, <16 x i32>} asm sideeffect "; def $0, $1, $2", "={a16},=a,=a"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @physreg_def_a32___def_vreg_a512_use_vreg_a256() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @physreg_def_a32___def_vreg_a512_use_vreg_a256( | |
| ; CHECK-SAME: ) #[[ATTR14:[0-9]+]] { | |
| ; CHECK-NEXT: [[TMP1:%.*]] = call { i32, <16 x i32> } asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call {i32, <16 x i32>} asm sideeffect "; def $0, $1, $2", "={a16},=a,a"(<8 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @mixed_physreg_vreg_tuples_0() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @mixed_physreg_vreg_tuples_0( | |
| ; CHECK-SAME: ) #[[ATTR11]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0, $1", "{a[1:4]},a"(<4 x i32> poison, <4 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @mixed_physreg_vreg_tuples_1() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @mixed_physreg_vreg_tuples_1( | |
| ; CHECK-SAME: ) #[[ATTR15:[0-9]+]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0, $1", "a,{a[0:3]}"(<4 x i32> poison, <4 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @physreg_raises_limit() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @physreg_raises_limit( | |
| ; CHECK-SAME: ) #[[ATTR16:[0-9]+]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0, $1", "a,{a[5:8]}"(<4 x i32> poison, <4 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @physreg_tuple_alignment_raises_limit() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @physreg_tuple_alignment_raises_limit( | |
| ; CHECK-SAME: ) #[[ATTR11]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0, $1", "a,{a[1:4]}"(<4 x i32> poison, <4 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @align3_virtreg() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @align3_virtreg( | |
| ; CHECK-SAME: ) #[[ATTR6]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0, $1", "a,a"(<3 x i32> poison, <3 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @align3_align4_virtreg() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @align3_align4_virtreg( | |
| ; CHECK-SAME: ) #[[ATTR15]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0, $1", "a,a"(<3 x i32> poison, <4 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @align2_align4_virtreg() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @align2_align4_virtreg( | |
| ; CHECK-SAME: ) #[[ATTR15]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0, $1", "a,a"(<2 x i32> poison, <4 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_write_register_a55() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_write_register_a55( | |
| ; CHECK-SAME: ) #[[ATTR17:[0-9]+]] { | |
| ; CHECK-NEXT: call void @llvm.write_register.i32(metadata [[META0:![0-9]+]], i32 0) | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @llvm.write_register.i64(metadata !0, i32 0) | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_write_register_v55() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_write_register_v55( | |
| ; CHECK-SAME: ) #[[ATTR0]] { | |
| ; CHECK-NEXT: call void @llvm.write_register.i32(metadata [[META1:![0-9]+]], i32 0) | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @llvm.write_register.i64(metadata !1, i32 0) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_write_register_a55_57() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_write_register_a55_57( | |
| ; CHECK-SAME: ) #[[ATTR18:[0-9]+]] { | |
| ; CHECK-NEXT: call void @llvm.write_register.i96(metadata [[META2:![0-9]+]], i96 0) | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @llvm.write_register.i64(metadata !2, i96 0) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_read_register_a55(ptr addrspace(1) %ptr) { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_read_register_a55( | |
| ; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR19:[0-9]+]] { | |
| ; CHECK-NEXT: [[REG:%.*]] = call i32 @llvm.read_register.i32(metadata [[META0]]) | |
| ; CHECK-NEXT: store i32 [[REG]], ptr addrspace(1) [[PTR]], align 4 | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %reg = call i32 @llvm.read_register.i64(metadata !0) | |
| store i32 %reg, ptr addrspace(1) %ptr | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_read_volatile_register_a55(ptr addrspace(1) %ptr) { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_read_volatile_register_a55( | |
| ; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR20:[0-9]+]] { | |
| ; CHECK-NEXT: [[REG:%.*]] = call i32 @llvm.read_volatile_register.i32(metadata [[META0]]) | |
| ; CHECK-NEXT: store i32 [[REG]], ptr addrspace(1) [[PTR]], align 4 | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %reg = call i32 @llvm.read_volatile_register.i64(metadata !0) | |
| store i32 %reg, ptr addrspace(1) %ptr | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_read_register_a56_59(ptr addrspace(1) %ptr) { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_read_register_a56_59( | |
| ; CHECK-SAME: ptr addrspace(1) [[PTR:%.*]]) #[[ATTR21:[0-9]+]] { | |
| ; CHECK-NEXT: [[REG:%.*]] = call i128 @llvm.read_register.i128(metadata [[META3:![0-9]+]]) | |
| ; CHECK-NEXT: store i128 [[REG]], ptr addrspace(1) [[PTR]], align 8 | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %reg = call i128 @llvm.read_register.i64(metadata !3) | |
| store i128 %reg, ptr addrspace(1) %ptr | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_write_register_out_of_bounds_a256() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_write_register_out_of_bounds_a256( | |
| ; CHECK-SAME: ) #[[ATTR22:[0-9]+]] { | |
| ; CHECK-NEXT: call void @llvm.write_register.i32(metadata [[META4:![0-9]+]], i32 0) | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @llvm.write_register.i64(metadata !4, i32 0) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_multiple_uses() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_multiple_uses( | |
| ; CHECK-SAME: ) #[[ATTR5]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "a"(i64 poison) | |
| call void asm sideeffect "; use $0", "a"(i32 poison) | |
| call void asm sideeffect "; use $0", "a"(i128 poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_multiple_defs() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_multiple_defs( | |
| ; CHECK-SAME: ) #[[ATTR5]] { | |
| ; CHECK-NEXT: [[TMP1:%.*]] = call i64 asm sideeffect " | |
| ; CHECK-NEXT: [[TMP2:%.*]] = call i32 asm sideeffect " | |
| ; CHECK-NEXT: [[TMP3:%.*]] = call i128 asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call i64 asm sideeffect "; def $0", "=a"() | |
| call i32 asm sideeffect "; def $0", "=a"() | |
| call i128 asm sideeffect "; def $0", "=a"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_multiple_use_defs() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_multiple_use_defs( | |
| ; CHECK-SAME: ) #[[ATTR5]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: [[TMP1:%.*]] = call i128 asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "a"(i32 poison) | |
| call i128 asm sideeffect "; def $0", "=a"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define void @callgraph_b() { | |
| ; CHECK-LABEL: define void @callgraph_b( | |
| ; CHECK-SAME: ) #[[ATTR15]] { | |
| ; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i32> asm sideeffect " | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call <4 x i32> asm sideeffect "; def $0", "=a"() | |
| call void asm sideeffect "; use $0", "a"(<8 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define void @callgraph_c() { | |
| ; CHECK-LABEL: define void @callgraph_c( | |
| ; CHECK-SAME: ) #[[ATTR2]] { | |
| ; CHECK-NEXT: [[TMP1:%.*]] = call i32 asm sideeffect " | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call i32 asm sideeffect "; def $0", "=a"() | |
| call void asm sideeffect "; use $0", "a"(<2 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define void @callgraph_a(i1 %cond) { | |
| ; CHECK-LABEL: define void @callgraph_a( | |
| ; CHECK-SAME: i1 [[COND:%.*]]) #[[ATTR15]] { | |
| ; CHECK-NEXT: br i1 [[COND]], label [[A:%.*]], label [[B:%.*]] | |
| ; CHECK: a: | |
| ; CHECK-NEXT: call void @callgraph_b() | |
| ; CHECK-NEXT: ret void | |
| ; CHECK: b: | |
| ; CHECK-NEXT: call void @callgraph_c() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| br i1 %cond, label %a, label %b | |
| a: | |
| call void @callgraph_b() | |
| ret void | |
| b: | |
| call void @callgraph_c() | |
| ret void | |
| } | |
| define void @kernel_max_callgraph(i1 %cond) { | |
| ; CHECK-LABEL: define void @kernel_max_callgraph( | |
| ; CHECK-SAME: i1 [[COND:%.*]]) #[[ATTR15]] { | |
| ; CHECK-NEXT: call void @callgraph_a(i1 [[COND]]) | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @callgraph_a(i1 %cond) | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_all_virtregs() #1 { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_all_virtregs( | |
| ; CHECK-SAME: ) #[[ATTR23:[0-9]+]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "a,a,a,a,a,a,a,a"(<32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_uses_all_virtregs_plus_1() #1 { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_uses_all_virtregs_plus_1( | |
| ; CHECK-SAME: ) #[[ATTR23]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "a,a,a,a,a,a,a,a,a"(<32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, <32 x i32> poison, i32 poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define void @recursive() { | |
| ; CHECK-LABEL: define void @recursive( | |
| ; CHECK-SAME: ) #[[ATTR24:[0-9]+]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: call void @recursive() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "a"(<7 x i32> poison) | |
| call void @use_most() | |
| call void @recursive() | |
| ret void | |
| } | |
| define void @indirect_0() { | |
| ; CHECK-LABEL: define void @indirect_0( | |
| ; CHECK-SAME: ) #[[ATTR24]] { | |
| ; CHECK-NEXT: call void asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void asm sideeffect "; use $0", "a"(<7 x i32> poison) | |
| call void @use_most() | |
| ret void | |
| } | |
| define void @indirect_1() { | |
| ; CHECK-LABEL: define void @indirect_1( | |
| ; CHECK-SAME: ) #[[ATTR25:[0-9]+]] { | |
| ; CHECK-NEXT: [[TMP1:%.*]] = call <3 x i32> asm sideeffect " | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call <3 x i32> asm sideeffect "; def $0", "=a"() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @knowable_indirect_call(i1 %cond) { | |
| ; CHECK-LABEL: define amdgpu_kernel void @knowable_indirect_call( | |
| ; CHECK-SAME: i1 [[COND:%.*]]) #[[ATTR24]] { | |
| ; CHECK-NEXT: [[FPTR:%.*]] = select i1 [[COND]], ptr @indirect_0, ptr @indirect_1 | |
| ; CHECK-NEXT: [[TMP1:%.*]] = icmp eq ptr [[FPTR]], @indirect_1 | |
| ; CHECK-NEXT: br i1 [[TMP1]], label [[TMP2:%.*]], label [[TMP3:%.*]] | |
| ; CHECK: 2: | |
| ; CHECK-NEXT: call void @indirect_1() | |
| ; CHECK-NEXT: br label [[TMP6:%.*]] | |
| ; CHECK: 3: | |
| ; CHECK-NEXT: br i1 true, label [[TMP4:%.*]], label [[TMP5:%.*]] | |
| ; CHECK: 4: | |
| ; CHECK-NEXT: call void @indirect_0() | |
| ; CHECK-NEXT: br label [[TMP6]] | |
| ; CHECK: 5: | |
| ; CHECK-NEXT: unreachable | |
| ; CHECK: 6: | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| %fptr = select i1 %cond, ptr @indirect_0, ptr @indirect_1 | |
| call void %fptr() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @calls_poison(i1 %cond) { | |
| ; CHECK-LABEL: define amdgpu_kernel void @calls_poison( | |
| ; CHECK-SAME: i1 [[COND:%.*]]) { | |
| ; CHECK-NEXT: call void poison() | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void poison() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @calls_null(i1 %cond) { | |
| ; CHECK-LABEL: define amdgpu_kernel void @calls_null( | |
| ; CHECK-SAME: i1 [[COND:%.*]]) { | |
| ; CHECK-NEXT: call void null() | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void null() | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @indirect_unknown(ptr %fptr) { | |
| ; CHECK-LABEL: define amdgpu_kernel void @indirect_unknown( | |
| ; CHECK-SAME: ptr [[FPTR:%.*]]) { | |
| ; CHECK-NEXT: call void [[FPTR]]() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void %fptr() | |
| ret void | |
| } | |
| ; Sanitizer tests: amdgpu-agpr-alloc=0 must not be inferred for functions with | |
| ; sanitizer attributes. Sanitizer runtime calls (e.g. __asan_malloc_impl) are | |
| ; introduced after the Attributor runs and may use AGPRs. | |
| define amdgpu_kernel void @kernel_sanitize_address() sanitize_address { | |
| ; CHECK: Function Attrs: sanitize_address | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_sanitize_address( | |
| ; CHECK-SAME: ) #[[ATTR26:[0-9]+]] { | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_sanitize_memory() sanitize_memory { | |
| ; CHECK: Function Attrs: sanitize_memory | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_sanitize_memory( | |
| ; CHECK-SAME: ) #[[ATTR27:[0-9]+]] { | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_sanitize_thread() sanitize_thread { | |
| ; CHECK: Function Attrs: sanitize_thread | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_sanitize_thread( | |
| ; CHECK-SAME: ) #[[ATTR28:[0-9]+]] { | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_sanitize_hwaddress() sanitize_hwaddress { | |
| ; CHECK: Function Attrs: sanitize_hwaddress | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_sanitize_hwaddress( | |
| ; CHECK-SAME: ) #[[ATTR29:[0-9]+]] { | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @use_most() | |
| ret void | |
| } | |
| ; A sanitized function with a user-defined amdgpu-agpr-alloc=0 (mimics an | |
| ; annotation from e.g. inline asm constraints). The early return after | |
| ; indicateOptimisticFixpoint() means the pre-existing attribute is trusted | |
| ; and the sanitizer check is skipped. | |
| define amdgpu_kernel void @kernel_sanitize_address_preannotated() #2 { | |
| ; CHECK: Function Attrs: sanitize_address | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_sanitize_address_preannotated( | |
| ; CHECK-SAME: ) #[[ATTR30:[0-9]+]] { | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @use_most() | |
| ret void | |
| } | |
| ; A non-sanitized kernel calling a sanitized callee. The callee's pessimistic | |
| ; fixpoint propagates up, so the caller must not get amdgpu-agpr-alloc=0 either. | |
| define void @sanitized_callee() sanitize_address { | |
| ; CHECK: Function Attrs: sanitize_address | |
| ; CHECK-LABEL: define void @sanitized_callee( | |
| ; CHECK-SAME: ) #[[ATTR26]] { | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @use_most() | |
| ret void | |
| } | |
| define amdgpu_kernel void @kernel_calls_sanitized_callee() { | |
| ; CHECK-LABEL: define amdgpu_kernel void @kernel_calls_sanitized_callee( | |
| ; CHECK-SAME: ) #[[ATTR31:[0-9]+]] { | |
| ; CHECK-NEXT: call void @sanitized_callee() | |
| ; CHECK-NEXT: call void @use_most() | |
| ; CHECK-NEXT: ret void | |
| ; | |
| call void @sanitized_callee() | |
| call void @use_most() | |
| ret void | |
| } | |
| attributes #0 = { "amdgpu-agpr-alloc"="0" } | |
| attributes #1 = { "amdgpu-waves-per-eu"="1,1" } | |
| attributes #2 = { sanitize_address "amdgpu-agpr-alloc"="0" } | |
| !0 = !{!"a55"} | |
| !1 = !{!"v55"} | |
| !2 = !{!"a[55:57]"} | |
| !3 = !{!"a[56:59]"} | |
| !4 = !{!"a256"} | |
| ;. | |
| ; CHECK: attributes #[[ATTR0]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-lds-dma" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR1]] = { "amdgpu-agpr-alloc"="1" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR2]] = { "amdgpu-agpr-alloc"="2" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR3]] = { "amdgpu-agpr-alloc"="0" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR4:[0-9]+]] = { convergent nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) } | |
| ; CHECK: attributes #[[ATTR5]] = { "amdgpu-agpr-alloc"="4" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR6]] = { "amdgpu-agpr-alloc"="6" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR7]] = { "amdgpu-agpr-alloc"="5" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR8]] = { "amdgpu-agpr-alloc"="14" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR9]] = { "amdgpu-agpr-alloc"="256" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR10]] = { "amdgpu-agpr-alloc"="32" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR11]] = { "amdgpu-agpr-alloc"="9" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR12]] = { "amdgpu-agpr-alloc"="64" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR13]] = { "amdgpu-agpr-alloc"="49" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR14]] = { "amdgpu-agpr-alloc"="33" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR15]] = { "amdgpu-agpr-alloc"="8" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR16]] = { "amdgpu-agpr-alloc"="13" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR17]] = { "amdgpu-agpr-alloc"="56" "amdgpu-no-cluster-id-x" "amdgpu-no-cluster-id-y" "amdgpu-no-cluster-id-z" "amdgpu-no-completion-action" "amdgpu-no-default-queue" "amdgpu-no-dispatch-id" "amdgpu-no-dispatch-ptr" "amdgpu-no-flat-scratch-init" "amdgpu-no-heap-ptr" "amdgpu-no-hostcall-ptr" "amdgpu-no-implicitarg-ptr" "amdgpu-no-lds-dma" "amdgpu-no-lds-kernel-id" "amdgpu-no-multigrid-sync-arg" "amdgpu-no-queue-ptr" "amdgpu-no-workgroup-id-x" "amdgpu-no-workgroup-id-y" "amdgpu-no-workgroup-id-z" "amdgpu-no-workitem-id-x" "amdgpu-no-workitem-id-y" "amdgpu-no-workitem-id-z" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR18]] = { "amdgpu-agpr-alloc"="58" "amdgpu-no-lds-dma" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR19]] = { "amdgpu-agpr-alloc"="56" "amdgpu-no-lds-dma" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR20]] = { "amdgpu-agpr-alloc"="56" } | |
| ; CHECK: attributes #[[ATTR21]] = { "amdgpu-agpr-alloc"="60" "amdgpu-no-lds-dma" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR22]] = { "amdgpu-agpr-alloc"="256" "amdgpu-no-lds-dma" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR23]] = { "amdgpu-agpr-alloc"="256" "amdgpu-no-wwm" "amdgpu-waves-per-eu"="1,1" } | |
| ; CHECK: attributes #[[ATTR24]] = { "amdgpu-agpr-alloc"="7" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR25]] = { "amdgpu-agpr-alloc"="3" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR26]] = { sanitize_address "amdgpu-no-lds-dma" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR27]] = { sanitize_memory "amdgpu-no-lds-dma" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR28]] = { sanitize_thread "amdgpu-no-lds-dma" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR29]] = { sanitize_hwaddress "amdgpu-no-lds-dma" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR30]] = { sanitize_address "amdgpu-agpr-alloc"="0" "amdgpu-no-lds-dma" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR31]] = { "amdgpu-no-lds-dma" "amdgpu-no-wwm" } | |
| ; CHECK: attributes #[[ATTR32:[0-9]+]] = { nocallback nofree nosync nounwind speculatable willreturn memory(none) } | |
| ; CHECK: attributes #[[ATTR33:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(argmem: readwrite) } | |
| ; CHECK: attributes #[[ATTR34:[0-9]+]] = { nocallback nofree nosync nounwind willreturn memory(read) } | |
| ; CHECK: attributes #[[ATTR35:[0-9]+]] = { nounwind } | |
| ; CHECK: attributes #[[ATTR36:[0-9]+]] = { nocallback nounwind } | |
| ; CHECK: attributes #[[ATTR37]] = { "amdgpu-agpr-alloc"="0" } | |
| ;. | |
| ; CHECK: [[META0]] = !{!"a55"} | |
| ; CHECK: [[META1]] = !{!"v55"} | |
| ; CHECK: [[META2]] = !{!"a[55:57]"} | |
| ; CHECK: [[META3]] = !{!"a[56:59]"} | |
| ; CHECK: [[META4]] = !{!"a256"} | |
| ;. |