1 ; RUN: llc -march=amdgcn -mcpu=SI -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 ; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s 4 5 ; FIXME: Uses of this should be moved to llvm.amdgcn.rsq.clamped, and 6 ; an r600 variant added. 7 8 declare float @llvm.AMDGPU.rsq.clamped.f32(float) nounwind readnone 9 10 ; FUNC-LABEL: {{^}}rsq_clamped_f32: 11 ; SI: v_rsq_clamp_f32_e32 12 13 ; VI-DAG: v_rsq_f32_e32 [[RSQ:v[0-9]+]], {{s[0-9]+}} 14 ; VI-DAG: v_min_f32_e32 [[MIN:v[0-9]+]], 0x7f7fffff, [[RSQ]] 15 ; TODO: this constant should be folded: 16 ; VI-DAG: v_mov_b32_e32 [[MINFLT:v[0-9]+]], 0xff7fffff 17 ; VI: v_max_f32_e32 {{v[0-9]+}}, [[MIN]], [[MINFLT]] 18 19 ; EG: RECIPSQRT_CLAMPED 20 21 define void @rsq_clamped_f32(float addrspace(1)* %out, float %src) nounwind { 22 %rsq_clamped = call float @llvm.AMDGPU.rsq.clamped.f32(float %src) nounwind readnone 23 store float %rsq_clamped, float addrspace(1)* %out, align 4 24 ret void 25 } 26
