80 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 52; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -O3 --amdgpu-lower-module-lds-strategy=module < %s | FileCheck -check-prefix=GCN %s3; RUN: opt -S -mtriple=amdgcn-- -amdgpu-lower-module-lds --amdgpu-lower-module-lds-strategy=module < %s | FileCheck %s4; RUN: opt -S -mtriple=amdgcn-- -passes=amdgpu-lower-module-lds --amdgpu-lower-module-lds-strategy=module < %s | FileCheck %s5 6%vec_type = type { %vec_base }7%vec_base = type { %union.anon }8%union.anon = type { %"vec_base<char, 3>::n_vec_" }9%"vec_base<char, 3>::n_vec_" = type { [3 x i8] }10 11$_f1 = comdat any12$_f2 = comdat any13@_f1 = linkonce_odr hidden local_unnamed_addr addrspace(3) global %vec_type poison, comdat, align 114@_f2 = linkonce_odr hidden local_unnamed_addr addrspace(3) global %vec_type poison, comdat, align 115 16;.17; CHECK: @[[LLVM_AMDGCN_KERNEL_TEST_LDS:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [[LLVM_AMDGCN_KERNEL_TEST_LDS_T:%.*]] poison, align 4, !absolute_symbol !018;.19define protected amdgpu_kernel void @test(ptr addrspace(1) nocapture %ptr.coerce) local_unnamed_addr #0 {20; GCN-LABEL: test:21; GCN: ; %bb.0: ; %entry22; GCN-NEXT: v_mov_b32_e32 v0, 223; GCN-NEXT: v_mov_b32_e32 v1, 024; GCN-NEXT: ds_write_b8 v1, v025; GCN-NEXT: ds_read_u8 v2, v1 offset:226; GCN-NEXT: ds_read_u16 v3, v127; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x2428; GCN-NEXT: s_waitcnt lgkmcnt(0)29; GCN-NEXT: ds_write_b8 v1, v2 offset:630; GCN-NEXT: ds_write_b16 v1, v3 offset:431; GCN-NEXT: v_cmp_eq_u32_sdwa s[2:3], v3, v0 src0_sel:BYTE_0 src1_sel:DWORD32; GCN-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[2:3]33; GCN-NEXT: global_store_byte v1, v0, s[0:1]34; GCN-NEXT: s_endpgm35; CHECK-LABEL: define protected amdgpu_kernel void @test(36; CHECK-SAME: ptr addrspace(1) captures(none) [[PTR_COERCE:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {37; CHECK-NEXT: entry:38; CHECK-NEXT: store i8 3, ptr addrspace(3) @llvm.amdgcn.kernel.test.lds, align 4, !alias.scope !1, !noalias !439; CHECK-NEXT: tail call void @llvm.memcpy.p3.p3.i64(ptr addrspace(3) noundef align 1 dereferenceable(3) getelementptr inbounds ([[LLVM_AMDGCN_KERNEL_TEST_LDS_T:%.*]], ptr addrspace(3) @llvm.amdgcn.kernel.test.lds, i32 0, i32 2), ptr addrspace(3) noundef align 1 dereferenceable(3) @llvm.amdgcn.kernel.test.lds, i64 3, i1 false), !alias.scope !6, !noalias !740; CHECK-NEXT: [[TMP0:%.*]] = load i8, ptr addrspace(3) getelementptr inbounds ([[LLVM_AMDGCN_KERNEL_TEST_LDS_T]], ptr addrspace(3) @llvm.amdgcn.kernel.test.lds, i32 0, i32 2), align 4, !alias.scope !4, !noalias !141; CHECK-NEXT: [[CMP_I_I:%.*]] = icmp eq i8 [[TMP0]], 342; CHECK-NEXT: store i8 2, ptr addrspace(3) @llvm.amdgcn.kernel.test.lds, align 4, !alias.scope !1, !noalias !443; CHECK-NEXT: tail call void @llvm.memcpy.p3.p3.i64(ptr addrspace(3) noundef align 1 dereferenceable(3) getelementptr inbounds ([[LLVM_AMDGCN_KERNEL_TEST_LDS_T]], ptr addrspace(3) @llvm.amdgcn.kernel.test.lds, i32 0, i32 2), ptr addrspace(3) noundef align 1 dereferenceable(3) @llvm.amdgcn.kernel.test.lds, i64 3, i1 false), !alias.scope !6, !noalias !744; CHECK-NEXT: [[TMP1:%.*]] = load i8, ptr addrspace(3) getelementptr inbounds ([[LLVM_AMDGCN_KERNEL_TEST_LDS_T]], ptr addrspace(3) @llvm.amdgcn.kernel.test.lds, i32 0, i32 2), align 4, !alias.scope !4, !noalias !145; CHECK-NEXT: [[CMP_I_I19:%.*]] = icmp eq i8 [[TMP1]], 246; CHECK-NEXT: [[TMP2:%.*]] = and i1 [[CMP_I_I19]], [[CMP_I_I]]47; CHECK-NEXT: [[FROMBOOL8:%.*]] = zext i1 [[TMP2]] to i848; CHECK-NEXT: store i8 [[FROMBOOL8]], ptr addrspace(1) [[PTR_COERCE]], align 149; CHECK-NEXT: ret void50entry:51 store i8 3, ptr addrspace(3) @_f1, align 152 tail call void @llvm.memcpy.p3.p3.i64(ptr addrspace(3) noundef align 1 dereferenceable(3) @_f2, ptr addrspace(3) noundef align 1 dereferenceable(3) @_f1, i64 3, i1 false)53 %0 = load i8, ptr addrspace(3) @_f2, align 154 %cmp.i.i = icmp eq i8 %0, 355 store i8 2, ptr addrspace(3) @_f1, align 156 tail call void @llvm.memcpy.p3.p3.i64(ptr addrspace(3) noundef align 1 dereferenceable(3) @_f2, ptr addrspace(3) noundef align 1 dereferenceable(3) @_f1, i64 3, i1 false)57 %1 = load i8, ptr addrspace(3) @_f2, align 158 %cmp.i.i19 = icmp eq i8 %1, 259 %2 = and i1 %cmp.i.i19, %cmp.i.i60 %frombool8 = zext i1 %2 to i861 store i8 %frombool8, ptr addrspace(1) %ptr.coerce, align 162 ret void63}64 65declare void @llvm.memcpy.p3.p3.i64(ptr addrspace(3) noalias nocapture writeonly, ptr addrspace(3) noalias nocapture readonly, i64, i1 immarg) #166 67;.68; CHECK: attributes #[[ATTR0]] = { "amdgpu-lds-size"="7" }69; CHECK: attributes #[[ATTR1:[0-9]+]] = { nocallback nofree nounwind willreturn memory(argmem: readwrite) }70;.71; CHECK: [[META0:![0-9]+]] = !{i32 0, i32 1}72; CHECK: [[META1:![0-9]+]] = !{!2}73; CHECK: [[META2:![0-9]+]] = distinct !{!2, !3}74; CHECK: [[META3:![0-9]+]] = distinct !{!3}75; CHECK: [[META4:![0-9]+]] = !{!5}76; CHECK: [[META5:![0-9]+]] = distinct !{!5, !3}77; CHECK: [[META6:![0-9]+]] = !{!5, !2}78; CHECK: [[META7:![0-9]+]] = !{}79;.80