1 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
2 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
4 ; GCN-LABEL: {{^}}test_fmed3:
5 ; GCN: v_med3_f32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
6 define amdgpu_kernel void @test_fmed3(float addrspace(1)* %out, float %src0, float %src1, float %src2) #1 {
7 %med3 = call float @llvm.amdgcn.fmed3.f32(float %src0, float %src1, float %src2)
8 store float %med3, float addrspace(1)* %out
12 ; GCN-LABEL: {{^}}test_fmed3_srcmods:
13 ; GCN: v_med3_f32 v{{[0-9]+}}, -s{{[0-9]+}}, |v{{[0-9]+}}|, -|v{{[0-9]+}}|
14 define amdgpu_kernel void @test_fmed3_srcmods(float addrspace(1)* %out, float %src0, float %src1, float %src2) #1 {
15 %src0.fneg = fsub float -0.0, %src0
16 %src1.fabs = call float @llvm.fabs.f32(float %src1)
17 %src2.fabs = call float @llvm.fabs.f32(float %src2)
18 %src2.fneg.fabs = fsub float -0.0, %src2.fabs
19 %med3 = call float @llvm.amdgcn.fmed3.f32(float %src0.fneg, float %src1.fabs, float %src2.fneg.fabs)
20 store float %med3, float addrspace(1)* %out
24 ; GCN-LABEL: {{^}}test_fneg_fmed3:
25 ; GCN: v_med3_f32 v{{[0-9]+}}, -s{{[0-9]+}}, -v{{[0-9]+}}, -v{{[0-9]+}}
26 define amdgpu_kernel void @test_fneg_fmed3(float addrspace(1)* %out, float %src0, float %src1, float %src2) #1 {
27 %med3 = call float @llvm.amdgcn.fmed3.f32(float %src0, float %src1, float %src2)
28 %neg.med3 = fsub float -0.0, %med3
29 store float %neg.med3, float addrspace(1)* %out
33 ; GCN-LABEL: {{^}}test_fneg_fmed3_multi_use:
34 ; GCN: v_med3_f32 [[MED3:v[0-9]+]], -s{{[0-9]+}}, -v{{[0-9]+}}, -v{{[0-9]+}}
35 ; GCN: v_mul_f32_e32 v{{[0-9]+}}, -4.0, [[MED3]]
36 define amdgpu_kernel void @test_fneg_fmed3_multi_use(float addrspace(1)* %out, float %src0, float %src1, float %src2) #1 {
37 %med3 = call float @llvm.amdgcn.fmed3.f32(float %src0, float %src1, float %src2)
38 %neg.med3 = fsub float -0.0, %med3
39 %med3.user = fmul float %med3, 4.0
40 store volatile float %med3.user, float addrspace(1)* %out
41 store volatile float %neg.med3, float addrspace(1)* %out
45 ; GCN-LABEL: {{^}}test_fabs_fmed3:
46 ; GCN: v_med3_f32 [[MED3:v[0-9]+]], s{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
47 ; GCN: v_and_b32_e32 v{{[0-9]+}}, 0x7fffffff, [[MED3]]
48 define amdgpu_kernel void @test_fabs_fmed3(float addrspace(1)* %out, float %src0, float %src1, float %src2) #1 {
49 %med3 = call float @llvm.amdgcn.fmed3.f32(float %src0, float %src1, float %src2)
50 %fabs.med3 = call float @llvm.fabs.f32(float %med3)
51 store float %fabs.med3, float addrspace(1)* %out
55 ; GCN-LABEL: {{^}}test_fneg_fmed3_rr_0:
56 ; GCN: v_bfrev_b32_e32 [[NEG0:v[0-9]+]], 1
57 ; GCN: v_med3_f32 v{{[0-9]+}}, -s{{[0-9]+}}, -v{{[0-9]+}}, [[NEG0]]
58 define amdgpu_kernel void @test_fneg_fmed3_rr_0(float addrspace(1)* %out, float %src0, float %src1) #1 {
59 %med3 = call float @llvm.amdgcn.fmed3.f32(float %src0, float %src1, float 0.0)
60 %neg.med3 = fsub float -0.0, %med3
61 store float %neg.med3, float addrspace(1)* %out
65 ; FIXME: Worse off from folding this
66 ; GCN-LABEL: {{^}}test_fneg_fmed3_rr_0_foldable_user:
67 ; GCN: v_bfrev_b32_e32 [[NEG0:v[0-9]+]], 1
68 ; GCN: v_med3_f32 [[MED3:v[0-9]+]], -s{{[0-9]+}}, -v{{[0-9]+}}, [[NEG0]]
69 ; GCN: v_mul_f32_e32 v{{[0-9]+}}, s{{[0-9]+}}, [[MED3]]
70 define amdgpu_kernel void @test_fneg_fmed3_rr_0_foldable_user(float addrspace(1)* %out, float %src0, float %src1, float %mul.arg) #1 {
71 %med3 = call float @llvm.amdgcn.fmed3.f32(float %src0, float %src1, float 0.0)
72 %neg.med3 = fsub float -0.0, %med3
73 %mul = fmul float %neg.med3, %mul.arg
74 store float %mul, float addrspace(1)* %out
78 ; GCN-LABEL: {{^}}test_fneg_fmed3_r_inv2pi_0:
79 ; GCN-DAG: v_bfrev_b32_e32 [[NEG0:v[0-9]+]], 1
80 ; GCN-DAG: v_mov_b32_e32 [[NEG_INV:v[0-9]+]], 0xbe22f983
81 ; GCN: v_med3_f32 v{{[0-9]+}}, -s{{[0-9]+}}, [[NEG_INV]], [[NEG0]]
82 define amdgpu_kernel void @test_fneg_fmed3_r_inv2pi_0(float addrspace(1)* %out, float %src0) #1 {
83 %med3 = call float @llvm.amdgcn.fmed3.f32(float %src0, float 0x3FC45F3060000000, float 0.0)
84 %neg.med3 = fsub float -0.0, %med3
85 store float %neg.med3, float addrspace(1)* %out
89 ; GCN-LABEL: {{^}}test_fneg_fmed3_r_inv2pi_0_foldable_user:
90 ; GCN-DAG: v_bfrev_b32_e32 [[NEG0:v[0-9]+]], 1
91 ; GCN-DAG: v_mov_b32_e32 [[NEG_INV:v[0-9]+]], 0xbe22f983
92 ; GCN: v_med3_f32 [[MED3:v[0-9]+]], -s{{[0-9]+}}, [[NEG_INV]], [[NEG0]]
93 ; GCN: v_mul_f32_e32 v{{[0-9]+}}, s{{[0-9]+}}, [[MED3]]
94 define amdgpu_kernel void @test_fneg_fmed3_r_inv2pi_0_foldable_user(float addrspace(1)* %out, float %src0, float %mul.arg) #1 {
95 %med3 = call float @llvm.amdgcn.fmed3.f32(float %src0, float 0x3FC45F3060000000, float 0.0)
96 %neg.med3 = fsub float -0.0, %med3
97 %mul = fmul float %neg.med3, %mul.arg
98 store float %mul, float addrspace(1)* %out
102 declare float @llvm.amdgcn.fmed3.f32(float, float, float) #0
103 declare float @llvm.fabs.f32(float) #0
105 attributes #0 = { nounwind readnone }
106 attributes #1 = { nounwind }