1; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s 2; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=bonaire < %s | FileCheck -check-prefix=CI -check-prefix=FUNC %s 3; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=CI -check-prefix=FUNC %s 4 5declare double @llvm.trunc.f64(double) nounwind readnone 6declare <2 x double> @llvm.trunc.v2f64(<2 x double>) nounwind readnone 7declare <3 x double> @llvm.trunc.v3f64(<3 x double>) nounwind readnone 8declare <4 x double> @llvm.trunc.v4f64(<4 x double>) nounwind readnone 9declare <8 x double> @llvm.trunc.v8f64(<8 x double>) nounwind readnone 10declare <16 x double> @llvm.trunc.v16f64(<16 x double>) nounwind readnone 11 12; FUNC-LABEL: {{^}}v_ftrunc_f64: 13; CI: v_trunc_f64 14; SI: v_bfe_u32 {{v[0-9]+}}, {{v[0-9]+}}, 20, 11 15; SI: s_endpgm 16define amdgpu_kernel void @v_ftrunc_f64(double addrspace(1)* %out, double addrspace(1)* %in) { 17 %x = load double, double addrspace(1)* %in, align 8 18 %y = call double @llvm.trunc.f64(double %x) nounwind readnone 19 store double %y, double addrspace(1)* %out, align 8 20 ret void 21} 22 23; FUNC-LABEL: {{^}}ftrunc_f64: 24; CI: v_trunc_f64_e32 25 26; SI: s_bfe_u32 [[SEXP:s[0-9]+]], {{s[0-9]+}}, 0xb0014 27; SI-DAG: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x80000000 28; SI-DAG: s_add_i32 [[SEXP1:s[0-9]+]], [[SEXP]], 0xfffffc01 29; SI-DAG: s_lshr_b64 s[{{[0-9]+:[0-9]+}}], s[{{[0-9]+:[0-9]+}}], [[SEXP1]] 30; SI-DAG: s_andn2_b64 31; SI-DAG: cmp_gt_i32 32; SI-DAG: cndmask_b32 33; SI-DAG: cndmask_b32 34; SI-DAG: cmp_lt_i32 35; SI-DAG: cndmask_b32 36; SI-DAG: cndmask_b32 37; SI: s_endpgm 38define amdgpu_kernel void @ftrunc_f64(double addrspace(1)* %out, double %x) { 39 %y = call double @llvm.trunc.f64(double %x) nounwind readnone 40 store double %y, double addrspace(1)* %out 41 ret void 42} 43 44; FUNC-LABEL: {{^}}ftrunc_v2f64: 45; CI: v_trunc_f64_e32 46; CI: v_trunc_f64_e32 47define amdgpu_kernel void @ftrunc_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %x) { 48 %y = call <2 x double> @llvm.trunc.v2f64(<2 x double> %x) nounwind readnone 49 store <2 x double> %y, <2 x double> addrspace(1)* %out 50 ret void 51} 52 53; FIXME-FUNC-LABEL: {{^}}ftrunc_v3f64: 54; FIXME-CI: v_trunc_f64_e32 55; FIXME-CI: v_trunc_f64_e32 56; FIXME-CI: v_trunc_f64_e32 57; define amdgpu_kernel void @ftrunc_v3f64(<3 x double> addrspace(1)* %out, <3 x double> %x) { 58; %y = call <3 x double> @llvm.trunc.v3f64(<3 x double> %x) nounwind readnone 59; store <3 x double> %y, <3 x double> addrspace(1)* %out 60; ret void 61; } 62 63; FUNC-LABEL: {{^}}ftrunc_v4f64: 64; CI: v_trunc_f64_e32 65; CI: v_trunc_f64_e32 66; CI: v_trunc_f64_e32 67; CI: v_trunc_f64_e32 68define amdgpu_kernel void @ftrunc_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %x) { 69 %y = call <4 x double> @llvm.trunc.v4f64(<4 x double> %x) nounwind readnone 70 store <4 x double> %y, <4 x double> addrspace(1)* %out 71 ret void 72} 73 74; FUNC-LABEL: {{^}}ftrunc_v8f64: 75; CI: v_trunc_f64_e32 76; CI: v_trunc_f64_e32 77; CI: v_trunc_f64_e32 78; CI: v_trunc_f64_e32 79; CI: v_trunc_f64_e32 80; CI: v_trunc_f64_e32 81; CI: v_trunc_f64_e32 82; CI: v_trunc_f64_e32 83define amdgpu_kernel void @ftrunc_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %x) { 84 %y = call <8 x double> @llvm.trunc.v8f64(<8 x double> %x) nounwind readnone 85 store <8 x double> %y, <8 x double> addrspace(1)* %out 86 ret void 87} 88 89; FUNC-LABEL: {{^}}ftrunc_v16f64: 90; CI: v_trunc_f64_e32 91; CI: v_trunc_f64_e32 92; CI: v_trunc_f64_e32 93; CI: v_trunc_f64_e32 94; CI: v_trunc_f64_e32 95; CI: v_trunc_f64_e32 96; CI: v_trunc_f64_e32 97; CI: v_trunc_f64_e32 98; CI: v_trunc_f64_e32 99; CI: v_trunc_f64_e32 100; CI: v_trunc_f64_e32 101; CI: v_trunc_f64_e32 102; CI: v_trunc_f64_e32 103; CI: v_trunc_f64_e32 104; CI: v_trunc_f64_e32 105; CI: v_trunc_f64_e32 106define amdgpu_kernel void @ftrunc_v16f64(<16 x double> addrspace(1)* %out, <16 x double> %x) { 107 %y = call <16 x double> @llvm.trunc.v16f64(<16 x double> %x) nounwind readnone 108 store <16 x double> %y, <16 x double> addrspace(1)* %out 109 ret void 110} 111