1 ; RUN: llc -march=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
2 ; RUN: llc -march=amdgcn -mcpu=tonga < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
4 declare double @llvm.minnum.f64(double, double) #0
5 declare <2 x double> @llvm.minnum.v2f64(<2 x double>, <2 x double>) #0
6 declare <4 x double> @llvm.minnum.v4f64(<4 x double>, <4 x double>) #0
7 declare <8 x double> @llvm.minnum.v8f64(<8 x double>, <8 x double>) #0
8 declare <16 x double> @llvm.minnum.v16f64(<16 x double>, <16 x double>) #0
10 ; FUNC-LABEL: {{^}}test_fmin_f64_ieee:
11 ; SI: s_load_dwordx2 [[A:s\[[0-9]+:[0-9]+\]]]
12 ; SI: s_load_dwordx2 [[B:s\[[0-9]+:[0-9]+\]]]
13 ; SI-DAG: v_max_f64 [[QUIETA:v\[[0-9]+:[0-9]+\]]], [[A]], [[A]]
14 ; SI-DAG: v_max_f64 [[QUIETB:v\[[0-9]+:[0-9]+\]]], [[B]], [[B]]
15 ; SI: v_min_f64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[QUIETA]], [[QUIETB]]
16 define amdgpu_kernel void @test_fmin_f64_ieee([8 x i32], double %a, [8 x i32], double %b) nounwind {
17 %val = call double @llvm.minnum.f64(double %a, double %b) #0
18 store double %val, double addrspace(1)* undef, align 8
22 ; FUNC-LABEL: {{^}}test_fmin_f64_no_ieee:
23 ; SI: ds_read_b64 [[VAL0:v\[[0-9]+:[0-9]+\]]]
24 ; SI: ds_read_b64 [[VAL1:v\[[0-9]+:[0-9]+\]]]
27 ; SI: v_min_f64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[VAL0]], [[VAL1]]
29 ; SI: ds_write_b64 v{{[0-9]+}}, [[RESULT]]
30 define amdgpu_ps void @test_fmin_f64_no_ieee() nounwind {
31 %a = load volatile double, double addrspace(3)* undef
32 %b = load volatile double, double addrspace(3)* undef
33 %val = call double @llvm.minnum.f64(double %a, double %b) #0
34 store volatile double %val, double addrspace(3)* undef
38 ; FUNC-LABEL: {{^}}test_fmin_v2f64:
41 define amdgpu_kernel void @test_fmin_v2f64(<2 x double> addrspace(1)* %out, <2 x double> %a, <2 x double> %b) nounwind {
42 %val = call <2 x double> @llvm.minnum.v2f64(<2 x double> %a, <2 x double> %b) #0
43 store <2 x double> %val, <2 x double> addrspace(1)* %out, align 16
47 ; FUNC-LABEL: {{^}}test_fmin_v4f64:
52 define amdgpu_kernel void @test_fmin_v4f64(<4 x double> addrspace(1)* %out, <4 x double> %a, <4 x double> %b) nounwind {
53 %val = call <4 x double> @llvm.minnum.v4f64(<4 x double> %a, <4 x double> %b) #0
54 store <4 x double> %val, <4 x double> addrspace(1)* %out, align 32
58 ; FUNC-LABEL: {{^}}test_fmin_v8f64:
67 define amdgpu_kernel void @test_fmin_v8f64(<8 x double> addrspace(1)* %out, <8 x double> %a, <8 x double> %b) nounwind {
68 %val = call <8 x double> @llvm.minnum.v8f64(<8 x double> %a, <8 x double> %b) #0
69 store <8 x double> %val, <8 x double> addrspace(1)* %out, align 64
73 ; FUNC-LABEL: {{^}}test_fmin_v16f64:
90 define amdgpu_kernel void @test_fmin_v16f64(<16 x double> addrspace(1)* %out, <16 x double> %a, <16 x double> %b) nounwind {
91 %val = call <16 x double> @llvm.minnum.v16f64(<16 x double> %a, <16 x double> %b) #0
92 store <16 x double> %val, <16 x double> addrspace(1)* %out, align 128
96 attributes #0 = { nounwind readnone }