175 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-globals2; RUN: opt -mtriple=amdgcn-unknown-amdhsa -S -passes=amdgpu-attributor < %s | FileCheck -check-prefix=HSA %s3 4declare void @llvm.memcpy.p1.p4.i32(ptr addrspace(1) nocapture, ptr addrspace(4) nocapture, i32, i1) #05 6@lds.i32 = unnamed_addr addrspace(3) global i32 poison, align 47@lds.arr = unnamed_addr addrspace(3) global [256 x i32] poison, align 48 9@global.i32 = unnamed_addr addrspace(1) global i32 poison, align 410@global.arr = unnamed_addr addrspace(1) global [256 x i32] poison, align 411 12;.13; HSA: @lds.i32 = unnamed_addr addrspace(3) global i32 poison, align 414; HSA: @lds.arr = unnamed_addr addrspace(3) global [256 x i32] poison, align 415; HSA: @global.i32 = unnamed_addr addrspace(1) global i32 poison, align 416; HSA: @global.arr = unnamed_addr addrspace(1) global [256 x i32] poison, align 417;.18define amdgpu_kernel void @store_cast_0_flat_to_group_addrspacecast() #1 {19; HSA-LABEL: define {{[^@]+}}@store_cast_0_flat_to_group_addrspacecast20; HSA-SAME: () #[[ATTR1:[0-9]+]] {21; HSA-NEXT: store i32 7, ptr addrspace(3) addrspacecast (ptr addrspace(4) null to ptr addrspace(3)), align 422; HSA-NEXT: ret void23;24 store i32 7, ptr addrspace(3) addrspacecast (ptr addrspace(4) null to ptr addrspace(3))25 ret void26}27 28define amdgpu_kernel void @store_cast_0_group_to_flat_addrspacecast() #1 {29; HSA-LABEL: define {{[^@]+}}@store_cast_0_group_to_flat_addrspacecast30; HSA-SAME: () #[[ATTR2:[0-9]+]] {31; HSA-NEXT: store i32 7, ptr addrspace(4) addrspacecast (ptr addrspace(3) null to ptr addrspace(4)), align 432; HSA-NEXT: ret void33;34 store i32 7, ptr addrspace(4) addrspacecast (ptr addrspace(3) null to ptr addrspace(4))35 ret void36}37 38define amdgpu_kernel void @store_constant_cast_group_gv_to_flat() #1 {39; HSA-LABEL: define {{[^@]+}}@store_constant_cast_group_gv_to_flat40; HSA-SAME: () #[[ATTR2]] {41; HSA-NEXT: store i32 7, ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.i32 to ptr addrspace(4)), align 442; HSA-NEXT: ret void43;44 store i32 7, ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.i32 to ptr addrspace(4))45 ret void46}47 48define amdgpu_kernel void @store_constant_cast_group_gv_gep_to_flat() #1 {49; HSA-LABEL: define {{[^@]+}}@store_constant_cast_group_gv_gep_to_flat50; HSA-SAME: () #[[ATTR2]] {51; HSA-NEXT: store i32 7, ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8), align 452; HSA-NEXT: ret void53;54 store i32 7, ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8)55 ret void56}57 58define amdgpu_kernel void @store_constant_cast_global_gv_to_flat() #1 {59; HSA-LABEL: define {{[^@]+}}@store_constant_cast_global_gv_to_flat60; HSA-SAME: () #[[ATTR1]] {61; HSA-NEXT: store i32 7, ptr addrspace(4) addrspacecast (ptr addrspace(1) @global.i32 to ptr addrspace(4)), align 462; HSA-NEXT: ret void63;64 store i32 7, ptr addrspace(4) addrspacecast (ptr addrspace(1) @global.i32 to ptr addrspace(4))65 ret void66}67 68define amdgpu_kernel void @store_constant_cast_global_gv_gep_to_flat() #1 {69; HSA-LABEL: define {{[^@]+}}@store_constant_cast_global_gv_gep_to_flat70; HSA-SAME: () #[[ATTR1]] {71; HSA-NEXT: store i32 7, ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(1) @global.arr to ptr addrspace(4)), i64 0, i64 8), align 472; HSA-NEXT: ret void73;74 store i32 7, ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(1) @global.arr to ptr addrspace(4)), i64 0, i64 8)75 ret void76}77 78define amdgpu_kernel void @load_constant_cast_group_gv_gep_to_flat(ptr addrspace(1) %out) #1 {79; HSA-LABEL: define {{[^@]+}}@load_constant_cast_group_gv_gep_to_flat80; HSA-SAME: (ptr addrspace(1) [[OUT:%.*]]) #[[ATTR2]] {81; HSA-NEXT: [[VAL:%.*]] = load i32, ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8), align 482; HSA-NEXT: store i32 [[VAL]], ptr addrspace(1) [[OUT]], align 483; HSA-NEXT: ret void84;85 %val = load i32, ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8)86 store i32 %val, ptr addrspace(1) %out87 ret void88}89 90define amdgpu_kernel void @atomicrmw_constant_cast_group_gv_gep_to_flat(ptr addrspace(1) %out) #1 {91; HSA-LABEL: define {{[^@]+}}@atomicrmw_constant_cast_group_gv_gep_to_flat92; HSA-SAME: (ptr addrspace(1) [[OUT:%.*]]) #[[ATTR2]] {93; HSA-NEXT: [[VAL:%.*]] = atomicrmw add ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8), i32 1 seq_cst, align 494; HSA-NEXT: store i32 [[VAL]], ptr addrspace(1) [[OUT]], align 495; HSA-NEXT: ret void96;97 %val = atomicrmw add ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8), i32 1 seq_cst98 store i32 %val, ptr addrspace(1) %out99 ret void100}101 102define amdgpu_kernel void @cmpxchg_constant_cast_group_gv_gep_to_flat(ptr addrspace(1) %out) #1 {103; HSA-LABEL: define {{[^@]+}}@cmpxchg_constant_cast_group_gv_gep_to_flat104; HSA-SAME: (ptr addrspace(1) [[OUT:%.*]]) #[[ATTR2]] {105; HSA-NEXT: [[VAL:%.*]] = cmpxchg ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8), i32 0, i32 1 seq_cst seq_cst, align 4106; HSA-NEXT: [[VAL0:%.*]] = extractvalue { i32, i1 } [[VAL]], 0107; HSA-NEXT: store i32 [[VAL0]], ptr addrspace(1) [[OUT]], align 4108; HSA-NEXT: ret void109;110 %val = cmpxchg ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8), i32 0, i32 1 seq_cst seq_cst111 %val0 = extractvalue { i32, i1 } %val, 0112 store i32 %val0, ptr addrspace(1) %out113 ret void114}115 116define amdgpu_kernel void @memcpy_constant_cast_group_gv_gep_to_flat(ptr addrspace(1) %out) #1 {117; HSA-LABEL: define {{[^@]+}}@memcpy_constant_cast_group_gv_gep_to_flat118; HSA-SAME: (ptr addrspace(1) [[OUT:%.*]]) #[[ATTR2]] {119; HSA-NEXT: call void @llvm.memcpy.p1.p4.i32(ptr addrspace(1) align 4 [[OUT]], ptr addrspace(4) align 4 getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8), i32 32, i1 false)120; HSA-NEXT: ret void121;122 call void @llvm.memcpy.p1.p4.i32(ptr addrspace(1) align 4 %out, ptr addrspace(4) align 4 getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8), i32 32, i1 false)123 ret void124}125 126; Can't just search the pointer value127define amdgpu_kernel void @store_value_constant_cast_lds_gv_gep_to_flat(ptr addrspace(1) %out) #1 {128; HSA-LABEL: define {{[^@]+}}@store_value_constant_cast_lds_gv_gep_to_flat129; HSA-SAME: (ptr addrspace(1) [[OUT:%.*]]) #[[ATTR2]] {130; HSA-NEXT: store ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8), ptr addrspace(1) [[OUT]], align 8131; HSA-NEXT: ret void132;133 store ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8), ptr addrspace(1) %out134 ret void135}136 137; Can't just search pointer types138define amdgpu_kernel void @store_ptrtoint_value_constant_cast_lds_gv_gep_to_flat(ptr addrspace(1) %out) #1 {139; HSA-LABEL: define {{[^@]+}}@store_ptrtoint_value_constant_cast_lds_gv_gep_to_flat140; HSA-SAME: (ptr addrspace(1) [[OUT:%.*]]) #[[ATTR2]] {141; HSA-NEXT: store i64 ptrtoint (ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8) to i64), ptr addrspace(1) [[OUT]], align 8142; HSA-NEXT: ret void143;144 store i64 ptrtoint (ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8) to i64), ptr addrspace(1) %out145 ret void146}147 148; Cast group to flat, do GEP, cast back to group149define amdgpu_kernel void @store_constant_cast_group_gv_gep_to_flat_to_group() #1 {150; HSA-LABEL: define {{[^@]+}}@store_constant_cast_group_gv_gep_to_flat_to_group151; HSA-SAME: () #[[ATTR2]] {152; HSA-NEXT: store i32 7, ptr addrspace(3) addrspacecast (ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8) to ptr addrspace(3)), align 4153; HSA-NEXT: ret void154;155 store i32 7, ptr addrspace(3) addrspacecast (ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8) to ptr addrspace(3))156 ret void157}158 159define ptr addrspace(3) @ret_constant_cast_group_gv_gep_to_flat_to_group() #1 {160; HSA-LABEL: define {{[^@]+}}@ret_constant_cast_group_gv_gep_to_flat_to_group161; HSA-SAME: () #[[ATTR2]] {162; HSA-NEXT: ret ptr addrspace(3) addrspacecast (ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8) to ptr addrspace(3))163;164 ret ptr addrspace(3) addrspacecast (ptr addrspace(4) getelementptr ([256 x i32], ptr addrspace(4) addrspacecast (ptr addrspace(3) @lds.arr to ptr addrspace(4)), i64 0, i64 8) to ptr addrspace(3))165}166 167attributes #0 = { argmemonly nounwind }168attributes #1 = { nounwind }169 170;.171; HSA: attributes #[[ATTR0:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }172; HSA: attributes #[[ATTR1]] = { nounwind "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-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" "uniform-work-group-size"="false" }173; HSA: attributes #[[ATTR2]] = { nounwind "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-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" "uniform-work-group-size"="false" }174;.175