1; RUN: llc -march=amdgcn < %s | FileCheck -allow-deprecated-dag-overlap -check-prefix=GCN -check-prefix=SI -check-prefix=FUNC %s 2; RUN: llc -march=amdgcn -mcpu=tonga < %s | FileCheck -allow-deprecated-dag-overlap -check-prefix=GCN -check-prefix=GFX8 --check-prefix=FUNC %s 3; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -allow-deprecated-dag-overlap -check-prefix=EG -check-prefix=FUNC %s 4; RUN: llc -march=r600 -mcpu=cayman < %s | FileCheck -allow-deprecated-dag-overlap -check-prefix=CM -check-prefix=FUNC %s 5 6; FUNC-LABEL: {{^}}test: 7; EG: LOG_IEEE 8; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 9; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 10; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 11; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} 12; GCN: v_log_f32_e32 v{{[0-9]+}}, v{{[0-9]+}} 13; GCN: v_mul_f32_e32 v{{[0-9]+}}, 0x3f317218, v{{[0-9]+}} 14define void @test(float addrspace(1)* %out, float %in) { 15entry: 16 %res = call float @llvm.log.f32(float %in) 17 store float %res, float addrspace(1)* %out 18 ret void 19} 20 21; FUNC-LABEL: {{^}}testv2: 22; EG: LOG_IEEE 23; EG: LOG_IEEE 24; FIXME: We should be able to merge these packets together on Cayman so we 25; have a maximum of 4 instructions. 26; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 27; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 28; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 29; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 30; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 31; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 32; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} 33; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} 34; GCN: v_log_f32_e32 v{{[0-9]+}}, v{{[0-9]+}} 35; SI: v_mov_b32_e32 v[[R_F32_LOG_CONST:[0-9]+]], 0x3f317218 36; GCN: v_log_f32_e32 v{{[0-9]+}}, v{{[0-9]+}} 37; GFX8: v_mov_b32_e32 v[[R_F32_LOG_CONST:[0-9]+]], 0x3f317218 38; GCN: v_mul_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v[[R_F32_LOG_CONST]] 39; GCN: v_mul_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v[[R_F32_LOG_CONST]] 40define void @testv2(<2 x float> addrspace(1)* %out, <2 x float> %in) { 41entry: 42 %res = call <2 x float> @llvm.log.v2f32(<2 x float> %in) 43 store <2 x float> %res, <2 x float> addrspace(1)* %out 44 ret void 45} 46 47; FUNC-LABEL: {{^}}testv4: 48; EG: LOG_IEEE 49; EG: LOG_IEEE 50; EG: LOG_IEEE 51; EG: LOG_IEEE 52; FIXME: We should be able to merge these packets together on Cayman so we 53; have a maximum of 4 instructions. 54; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 55; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 56; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 57; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 58; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 59; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 60; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 61; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 62; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 63; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 64; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 65; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} (MASKED) 66; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} 67; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} 68; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} 69; CM-DAG: LOG_IEEE T{{[0-9]+\.[XYZW]}} 70; GCN: v_log_f32_e32 v{{[0-9]+}}, v{{[0-9]+}} 71; SI: v_mov_b32_e32 v[[R_F32_LOG_CONST:[0-9]+]], 0x3f317218 72; GCN: v_log_f32_e32 v{{[0-9]+}}, v{{[0-9]+}} 73; GCN: v_log_f32_e32 v{{[0-9]+}}, v{{[0-9]+}} 74; GCN: v_log_f32_e32 v{{[0-9]+}}, v{{[0-9]+}} 75; GFX8: v_mov_b32_e32 v[[R_F32_LOG_CONST:[0-9]+]], 0x3f317218 76; GCN: v_mul_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v[[R_F32_LOG_CONST]] 77; GCN: v_mul_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v[[R_F32_LOG_CONST]] 78; GCN: v_mul_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v[[R_F32_LOG_CONST]] 79; GCN: v_mul_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v[[R_F32_LOG_CONST]] 80define void @testv4(<4 x float> addrspace(1)* %out, <4 x float> %in) { 81entry: 82 %res = call <4 x float> @llvm.log.v4f32(<4 x float> %in) 83 store <4 x float> %res, <4 x float> addrspace(1)* %out 84 ret void 85} 86 87declare float @llvm.log.f32(float) readnone 88declare <2 x float> @llvm.log.v2f32(<2 x float>) readnone 89declare <4 x float> @llvm.log.v4f32(<4 x float>) readnone 90