blob: eeff2536b23229fbb49bca5dcb7035412c1d3337 [file] [log] [blame]
Tom Stellard49f8bfd2015-01-06 18:00:21 +00001; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
Marek Olsakfa6607d2015-02-11 14:26:46 +00002; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=VI -check-prefix=FUNC %s
Matt Arsenault257d48d2014-06-24 22:13:39 +00003; 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
Tom Stellard79243d92014-10-01 17:15:17 +00008; FUNC-LABEL: {{^}}rsq_clamped_f32:
Tom Stellard326d6ec2014-11-05 14:50:53 +00009; SI: v_rsq_clamp_f32_e32
Marek Olsakfa6607d2015-02-11 14:26:46 +000010
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
Matt Arsenault257d48d2014-06-24 22:13:39 +000017; EG: RECIPSQRT_CLAMPED
Marek Olsakfa6607d2015-02-11 14:26:46 +000018
Matt Arsenault257d48d2014-06-24 22:13:39 +000019define 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}