1; RUN: llc -march=amdgcn -mcpu=hawaii -verify-machineinstrs < %s | FileCheck -check-prefix=CI -check-prefix=COMMON %s 2; RUN: llc -march=amdgcn -mcpu=tahiti -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=COMMON %s 3; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=CI -check-prefix=COMMON %s 4 5 6; COMMON-LABEL: {{^}}fdiv_f64: 7; COMMON-DAG: buffer_load_dwordx2 [[NUM:v\[[0-9]+:[0-9]+\]]], off, {{s\[[0-9]+:[0-9]+\]}}, 0 8; COMMON-DAG: buffer_load_dwordx2 [[DEN:v\[[0-9]+:[0-9]+\]]], off, {{s\[[0-9]+:[0-9]+\]}}, 0 offset:8 9; CI-DAG: v_div_scale_f64 [[SCALE0:v\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, [[DEN]], [[DEN]], [[NUM]] 10; CI-DAG: v_div_scale_f64 [[SCALE1:v\[[0-9]+:[0-9]+\]]], vcc, [[NUM]], [[DEN]], [[NUM]] 11 12; Check for div_scale bug workaround on SI 13; SI-DAG: v_div_scale_f64 [[SCALE0:v\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, [[DEN]], [[DEN]], [[NUM]] 14; SI-DAG: v_div_scale_f64 [[SCALE1:v\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, [[NUM]], [[DEN]], [[NUM]] 15 16; COMMON-DAG: v_rcp_f64_e32 [[RCP_SCALE0:v\[[0-9]+:[0-9]+\]]], [[SCALE0]] 17 18; SI-DAG: v_cmp_eq_i32_e32 vcc, {{v[0-9]+}}, {{v[0-9]+}} 19; SI-DAG: v_cmp_eq_i32_e64 [[CMP0:s\[[0-9]+:[0-9]+\]]], {{v[0-9]+}}, {{v[0-9]+}} 20; SI-DAG: s_xor_b64 vcc, [[CMP0]], vcc 21 22; COMMON-DAG: v_fma_f64 [[FMA0:v\[[0-9]+:[0-9]+\]]], -[[SCALE0]], [[RCP_SCALE0]], 1.0 23; COMMON-DAG: v_fma_f64 [[FMA1:v\[[0-9]+:[0-9]+\]]], [[RCP_SCALE0]], [[FMA0]], [[RCP_SCALE0]] 24; COMMON-DAG: v_fma_f64 [[FMA2:v\[[0-9]+:[0-9]+\]]], -[[SCALE0]], [[FMA1]], 1.0 25; COMMON-DAG: v_fma_f64 [[FMA3:v\[[0-9]+:[0-9]+\]]], [[FMA1]], [[FMA2]], [[FMA1]] 26; COMMON-DAG: v_mul_f64 [[MUL:v\[[0-9]+:[0-9]+\]]], [[SCALE1]], [[FMA3]] 27; COMMON-DAG: v_fma_f64 [[FMA4:v\[[0-9]+:[0-9]+\]]], -[[SCALE0]], [[MUL]], [[SCALE1]] 28; COMMON: v_div_fmas_f64 [[FMAS:v\[[0-9]+:[0-9]+\]]], [[FMA4]], [[FMA3]], [[MUL]] 29; COMMON: v_div_fixup_f64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[FMAS]], [[DEN]], [[NUM]] 30; COMMON: buffer_store_dwordx2 [[RESULT]] 31; COMMON: s_endpgm 32define void @fdiv_f64(double addrspace(1)* %out, double addrspace(1)* %in) nounwind { 33 %gep.1 = getelementptr double, double addrspace(1)* %in, i32 1 34 %num = load volatile double, double addrspace(1)* %in 35 %den = load volatile double, double addrspace(1)* %gep.1 36 %result = fdiv double %num, %den 37 store double %result, double addrspace(1)* %out 38 ret void 39} 40 41; COMMON-LABEL: {{^}}fdiv_f64_s_v: 42define void @fdiv_f64_s_v(double addrspace(1)* %out, double addrspace(1)* %in, double %num) nounwind { 43 %den = load double, double addrspace(1)* %in 44 %result = fdiv double %num, %den 45 store double %result, double addrspace(1)* %out 46 ret void 47} 48 49; COMMON-LABEL: {{^}}fdiv_f64_v_s: 50define void @fdiv_f64_v_s(double addrspace(1)* %out, double addrspace(1)* %in, double %den) nounwind { 51 %num = load double, double addrspace(1)* %in 52 %result = fdiv double %num, %den 53 store double %result, double addrspace(1)* %out 54 ret void 55} 56 57; COMMON-LABEL: {{^}}fdiv_f64_s_s: 58define void @fdiv_f64_s_s(double addrspace(1)* %out, double %num, double %den) nounwind { 59 %result = fdiv double %num, %den 60 store double %result, double addrspace(1)* %out 61 ret void 62} 63 64; COMMON-LABEL: {{^}}v_fdiv_v2f64: 65define void @v_fdiv_v2f64(<2 x double> addrspace(1)* %out, <2 x double> addrspace(1)* %in) nounwind { 66 %gep.1 = getelementptr <2 x double>, <2 x double> addrspace(1)* %in, i32 1 67 %num = load <2 x double>, <2 x double> addrspace(1)* %in 68 %den = load <2 x double>, <2 x double> addrspace(1)* %gep.1 69 %result = fdiv <2 x double> %num, %den 70 store <2 x double> %result, <2 x double> addrspace(1)* %out 71 ret void 72} 73 74; COMMON-LABEL: {{^}}s_fdiv_v2f64: 75define void @s_fdiv_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %num, <2 x double> %den) { 76 %result = fdiv <2 x double> %num, %den 77 store <2 x double> %result, <2 x double> addrspace(1)* %out 78 ret void 79} 80 81; COMMON-LABEL: {{^}}v_fdiv_v4f64: 82define void @v_fdiv_v4f64(<4 x double> addrspace(1)* %out, <4 x double> addrspace(1)* %in) nounwind { 83 %gep.1 = getelementptr <4 x double>, <4 x double> addrspace(1)* %in, i32 1 84 %num = load <4 x double>, <4 x double> addrspace(1)* %in 85 %den = load <4 x double>, <4 x double> addrspace(1)* %gep.1 86 %result = fdiv <4 x double> %num, %den 87 store <4 x double> %result, <4 x double> addrspace(1)* %out 88 ret void 89} 90 91; COMMON-LABEL: {{^}}s_fdiv_v4f64: 92define void @s_fdiv_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %num, <4 x double> %den) { 93 %result = fdiv <4 x double> %num, %den 94 store <4 x double> %result, <4 x double> addrspace(1)* %out 95 ret void 96} 97