50 lines · plain
1; RUN: llc -mtriple=amdgcn -mcpu=verde < %s | FileCheck -check-prefix=GCN -check-prefix=VCCZ-BUG %s2; RUN: llc -mtriple=amdgcn -mcpu=bonaire < %s | FileCheck -check-prefix=GCN -check-prefix=VCCZ-BUG %s3; RUN: llc -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck -check-prefix=GCN %s4 5; GCN-LABEL: {{^}}vccz_workaround:6; GCN: s_load_dword [[REG:s[0-9]+]], s[{{[0-9]+:[0-9]+}}],7; GCN: v_cmp_neq_f32_e64 {{[^,]*}}, [[REG]], 0{{$}}8; VCCZ-BUG: s_waitcnt lgkmcnt(0)9; VCCZ-BUG: s_mov_b64 vcc, vcc10; GCN-NOT: s_mov_b64 vcc, vcc11; GCN: s_cbranch_vccnz [[EXIT:.L[0-9A-Za-z_]+]]12; GCN: buffer_store_dword13; GCN: [[EXIT]]:14; GCN: s_endpgm15define amdgpu_kernel void @vccz_workaround(ptr addrspace(4) %in, ptr addrspace(1) %out, float %cond) {16entry:17 %cnd = fcmp oeq float 0.0, %cond18 %sgpr = load volatile i32, ptr addrspace(4) %in19 br i1 %cnd, label %if, label %endif20 21if:22 store i32 %sgpr, ptr addrspace(1) %out23 br label %endif24 25endif:26 ret void27}28 29; GCN-LABEL: {{^}}vccz_noworkaround:30; GCN: v_cmp_neq_f32_e32 vcc, 0, v{{[0-9]+}}31; GCN-NOT: s_waitcnt lgkmcnt(0)32; GCN-NOT: s_mov_b64 vcc, vcc33; GCN: s_cbranch_vccnz [[EXIT:.L[0-9A-Za-z_]+]]34; GCN: buffer_store_dword35; GCN: [[EXIT]]:36; GCN: s_endpgm37define amdgpu_kernel void @vccz_noworkaround(ptr addrspace(1) %in, ptr addrspace(1) %out) {38entry:39 %vgpr = load volatile float, ptr addrspace(1) %in40 %cnd = fcmp oeq float 0.0, %vgpr41 br i1 %cnd, label %if, label %endif42 43if:44 store float %vgpr, ptr addrspace(1) %out45 br label %endif46 47endif:48 ret void49}50