• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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
6declare float @llvm.AMDGPU.rsq.clamped.f32(float) nounwind readnone
7
8; FUNC-LABEL: {{^}}rsq_clamped_f32:
9; SI: v_rsq_clamp_f32_e32
10
11; VI: v_rsq_f32_e32 [[RSQ:v[0-9]+]], {{s[0-9]+}}
12; VI: v_min_f32_e32 [[MIN:v[0-9]+]], 0x7f7fffff, [[RSQ]]
13; TODO: this constant should be folded:
14; VI: v_mov_b32_e32 [[MINFLT:v[0-9]+]], 0xff7fffff
15; VI: v_max_f32_e32 {{v[0-9]+}}, [[MIN]], [[MINFLT]]
16
17; EG: RECIPSQRT_CLAMPED
18
19define void @rsq_clamped_f32(float addrspace(1)* %out, float %src) nounwind {
20  %rsq_clamped = call float @llvm.AMDGPU.rsq.clamped.f32(float %src) nounwind readnone
21  store float %rsq_clamped, float addrspace(1)* %out, align 4
22  ret void
23}
24