35 lines · plain
1; RUN: llc -mtriple=amdgcn -mcpu=tahiti -stop-after=amdgpu-isel -o - %s | FileCheck -check-prefixes=GCN,GFX8 %s2; RUN: llc -mtriple=amdgcn -mcpu=tonga -stop-after=amdgpu-isel -o - %s | FileCheck -check-prefixes=GCN,GFX9 %s3 4; RUN: not llc -mtriple=amdgcn -mcpu=tahiti < %s 2>&1 | FileCheck %s5; RUN: not llc -mtriple=amdgcn -mcpu=tonga < %s 2>&1 | FileCheck %s6 7; CHECK: error: lds: unsupported initializer for address space8 9@lds = addrspace(3) global [256 x i32] zeroinitializer10 11define amdgpu_kernel void @load_zeroinit_lds_global(ptr addrspace(1) %out, i1 %p) {12 ; GCN-LABEL: name: load_zeroinit_lds_global13 ; GCN: bb.0 (%ir-block.0):14 ; GCN: liveins: $sgpr4_sgpr515 ; GCN: [[COPY:%[0-9]+]]:sgpr_64(p4) = COPY $sgpr4_sgpr516 ; GFX8: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[COPY]](p4), 9, 017 ; GFX9: [[S_LOAD_DWORDX2_IMM:%[0-9]+]]:sreg_64_xexec = S_LOAD_DWORDX2_IMM [[COPY]](p4), 36, 018 ; GFX8: [[COPY1:%[0-9]+]]:sreg_32 = COPY [[S_LOAD_DWORDX2_IMM]].sub119 ; GFX8: [[COPY2:%[0-9]+]]:sreg_32 = COPY [[S_LOAD_DWORDX2_IMM]].sub020 ; GFX8: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 6144021 ; GFX8: [[S_MOV_B32_1:%[0-9]+]]:sreg_32 = S_MOV_B32 -122 ; GFX8: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE killed [[COPY2]], %subreg.sub0, killed [[COPY1]], %subreg.sub1, killed [[S_MOV_B32_1]], %subreg.sub2, killed [[S_MOV_B32_]], %subreg.sub323 ; GCN: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 target-flags(amdgpu-abs32-lo) @lds, implicit $exec24 ; GCN: SI_INIT_M0 -1, implicit-def $m025 ; GCN: [[DS_READ_B32_:%[0-9]+]]:vgpr_32 = DS_READ_B32 killed [[V_MOV_B32_e32_]], 40, 0, implicit $m0, implicit $exec26 ; GFX9: [[COPY1:%[0-9]+]]:vreg_64 = COPY [[S_LOAD_DWORDX2_IMM]]27 ; GFX8: BUFFER_STORE_DWORD_OFFSET killed [[DS_READ_B32_]], killed [[REG_SEQUENCE]], 0, 0, 0, 0, implicit $exec28 ; GFX9: FLAT_STORE_DWORD killed [[COPY1]], killed [[DS_READ_B32_]], 0, 0, implicit $exec, implicit $flat_scr29 ; GCN: S_ENDPGM 030 %gep = getelementptr [256 x i32], ptr addrspace(3) @lds, i32 0, i32 1031 %ld = load i32, ptr addrspace(3) %gep32 store i32 %ld, ptr addrspace(1) %out33 ret void34}35