1 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s 2 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=VI -check-prefix=FUNC %s 3 4 declare float @llvm.amdgcn.rsq.clamp.f32(float) #1 5 declare double @llvm.amdgcn.rsq.clamp.f64(double) #1 6 7 ; FUNC-LABEL: {{^}}rsq_clamp_f32: 8 ; SI: v_rsq_clamp_f32_e32 9 10 ; VI: s_load_dword [[SRC:s[0-9]+]] 11 ; VI-DAG: v_rsq_f32_e32 [[RSQ:v[0-9]+]], [[SRC]] 12 ; VI-DAG: v_min_f32_e32 [[MIN:v[0-9]+]], 0x7f7fffff, [[RSQ]] 13 ; TODO: this constant should be folded: 14 ; VI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0xff7fffff 15 ; VI: v_max_f32_e32 [[RESULT:v[0-9]+]], [[MIN]], [[K]] 16 ; VI: buffer_store_dword [[RESULT]] 17 define void @rsq_clamp_f32(float addrspace(1)* %out, float %src) #0 { 18 %rsq_clamp = call float @llvm.amdgcn.rsq.clamp.f32(float %src) 19 store float %rsq_clamp, float addrspace(1)* %out 20 ret void 21 } 22 23 24 ; FUNC-LABEL: {{^}}rsq_clamp_f64: 25 ; SI: v_rsq_clamp_f64_e32 26 27 ; TODO: this constant should be folded: 28 ; VI-DAG: s_mov_b32 s[[LOW1:[0-9+]]], -1 29 ; VI-DAG: s_mov_b32 s[[HIGH1:[0-9+]]], 0x7fefffff 30 ; VI-DAG: s_mov_b32 s[[HIGH2:[0-9+]]], 0xffefffff 31 ; VI-DAG: v_rsq_f64_e32 [[RSQ:v\[[0-9]+:[0-9]+\]]], s[{{[0-9]+:[0-9]+}} 32 ; VI-DAG: v_min_f64 v[0:1], [[RSQ]], s{{\[}}[[LOW1]]:[[HIGH1]]] 33 ; VI-DAG: v_max_f64 v[0:1], v[0:1], s{{\[}}[[LOW1]]:[[HIGH2]]] 34 define void @rsq_clamp_f64(double addrspace(1)* %out, double %src) #0 { 35 %rsq_clamp = call double @llvm.amdgcn.rsq.clamp.f64(double %src) 36 store double %rsq_clamp, double addrspace(1)* %out 37 ret void 38 } 39 40 ; FUNC-LABEL: {{^}}rsq_clamp_undef_f32: 41 ; SI-NOT: v_rsq_clamp_f32 42 define void @rsq_clamp_undef_f32(float addrspace(1)* %out) #0 { 43 %rsq_clamp = call float @llvm.amdgcn.rsq.clamp.f32(float undef) 44 store float %rsq_clamp, float addrspace(1)* %out 45 ret void 46 } 47 48 attributes #0 = { nounwind } 49 attributes #1 = { nounwind readnone } 50
