61 lines · plain
1; RUN: llc -mtriple=amdgcn < %s | FileCheck -check-prefix=SI -check-prefix=GCN -check-prefix=FUNC %s2; RUN: llc -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck -check-prefix=VI -check-prefix=GCN -check-prefix=FUNC %s3 4; FUNC-LABEL: {{^}}fneg_f64:5; GCN: s_xor_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x800000006define amdgpu_kernel void @fneg_f64(ptr addrspace(1) %out, double %in) {7 %fneg = fsub double -0.000000e+00, %in8 store double %fneg, ptr addrspace(1) %out9 ret void10}11 12; FUNC-LABEL: {{^}}fneg_v2f64:13; GCN: s_xor_b3214; GCN: s_xor_b3215define amdgpu_kernel void @fneg_v2f64(ptr addrspace(1) nocapture %out, <2 x double> %in) {16 %fneg = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %in17 store <2 x double> %fneg, ptr addrspace(1) %out18 ret void19}20 21; FUNC-LABEL: {{^}}fneg_v4f64:22; R600: -PV23; R600: -T24; R600: -PV25; R600: -PV26 27; GCN: s_xor_b3228; GCN: s_xor_b3229; GCN: s_xor_b3230; GCN: s_xor_b3231define amdgpu_kernel void @fneg_v4f64(ptr addrspace(1) nocapture %out, <4 x double> %in) {32 %fneg = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %in33 store <4 x double> %fneg, ptr addrspace(1) %out34 ret void35}36 37; DAGCombiner will transform:38; (fneg (f64 bitcast (i64 a))) => (f64 bitcast (xor (i64 a), 0x80000000))39; unless the target returns true for isNegFree()40 41; FUNC-LABEL: {{^}}fneg_free_f64:42; GCN: v_add_f64 {{v\[[0-9]+:[0-9]+\]}}, -{{s\[[0-9]+:[0-9]+\]}}, 0{{$}}43define amdgpu_kernel void @fneg_free_f64(ptr addrspace(1) %out, i64 %in) {44 %bc = bitcast i64 %in to double45 %fsub = fsub double 0.0, %bc46 store double %fsub, ptr addrspace(1) %out47 ret void48}49 50; GCN-LABEL: {{^}}fneg_fold_f64:51; SI: s_load_dwordx2 [[NEG_VALUE:s\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, 0x1352; VI: s_load_dwordx2 [[NEG_VALUE:s\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, 0x4c53; GCN-NOT: xor54; GCN: v_mul_f64 {{v\[[0-9]+:[0-9]+\]}}, -[[NEG_VALUE]], [[NEG_VALUE]]55define amdgpu_kernel void @fneg_fold_f64(ptr addrspace(1) %out, [8 x i32], double %in) {56 %fsub = fsub double -0.0, %in57 %fmul = fmul double %fsub, %in58 store double %fmul, ptr addrspace(1) %out59 ret void60}61