1 ; RUN: llc -mtriple=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefixes=GCN,GFX678 %s
2 ; RUN: llc -mtriple=amdgcn -mcpu=tonga < %s | FileCheck -check-prefixes=GCN,GFX678 %s
3 ; RUN: llc -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX9 %s
5 declare double @llvm.minnum.f64(double, double) #0
6 declare <2 x double> @llvm.minnum.v2f64(<2 x double>, <2 x double>) #0
7 declare <4 x double> @llvm.minnum.v4f64(<4 x double>, <4 x double>) #0
8 declare <8 x double> @llvm.minnum.v8f64(<8 x double>, <8 x double>) #0
9 declare <16 x double> @llvm.minnum.v16f64(<16 x double>, <16 x double>) #0
11 ; GCN-LABEL: {{^}}test_fmin_f64_ieee_noflush:
12 ; GCN: s_load_dwordx2 [[A:s\[[0-9]+:[0-9]+\]]]
13 ; GCN: s_load_dwordx2 [[B:s\[[0-9]+:[0-9]+\]]]
15 ; GCN-DAG: v_max_f64 [[QUIETA:v\[[0-9]+:[0-9]+\]]], [[A]], [[A]]
16 ; GCN-DAG: v_max_f64 [[QUIETB:v\[[0-9]+:[0-9]+\]]], [[B]], [[B]]
18 ; GCN: v_min_f64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[QUIETB]], [[QUIETA]]
19 define amdgpu_kernel void @test_fmin_f64_ieee_noflush([8 x i32], double %a, [8 x i32], double %b) #1 {
20 %val = call double @llvm.minnum.f64(double %a, double %b) #0
21 store double %val, ptr addrspace(1) undef, align 8
25 ; GCN-LABEL: {{^}}test_fmin_f64_ieee_flush:
26 ; GCN: s_load_dwordx2 [[A:s\[[0-9]+:[0-9]+\]]]
27 ; GCN: s_load_dwordx2 [[B:s\[[0-9]+:[0-9]+\]]]
28 ; GFX678-DAG: v_mul_f64 [[QUIETA:v\[[0-9]+:[0-9]+\]]], 1.0, [[A]]
29 ; GFX678-DAG: v_mul_f64 [[QUIETB:v\[[0-9]+:[0-9]+\]]], 1.0, [[B]]
31 ; GFX9-DAG: v_max_f64 [[QUIETA:v\[[0-9]+:[0-9]+\]]], [[A]], [[A]]
32 ; GFX9-DAG: v_max_f64 [[QUIETB:v\[[0-9]+:[0-9]+\]]], [[B]], [[B]]
34 ; GCN: v_min_f64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[QUIETB]], [[QUIETA]]
35 define amdgpu_kernel void @test_fmin_f64_ieee_flush([8 x i32], double %a, [8 x i32], double %b) #2 {
36 %val = call double @llvm.minnum.f64(double %a, double %b) #0
37 store double %val, ptr addrspace(1) undef, align 8
41 ; GCN-LABEL: {{^}}test_fmin_f64_no_ieee:
42 ; GCN: ds_read_b64 [[VAL0:v\[[0-9]+:[0-9]+\]]]
43 ; GCN: ds_read_b64 [[VAL1:v\[[0-9]+:[0-9]+\]]]
46 ; GCN: v_min_f64 [[RESULT:v\[[0-9]+:[0-9]+\]]], [[VAL0]], [[VAL1]]
48 ; GCN: ds_write_b64 v{{[0-9]+}}, [[RESULT]]
49 define amdgpu_ps void @test_fmin_f64_no_ieee() nounwind {
50 %a = load volatile double, ptr addrspace(3) undef
51 %b = load volatile double, ptr addrspace(3) undef
52 %val = call double @llvm.minnum.f64(double %a, double %b) #0
53 store volatile double %val, ptr addrspace(3) undef
57 ; GCN-LABEL: {{^}}test_fmin_v2f64:
60 define amdgpu_kernel void @test_fmin_v2f64(ptr addrspace(1) %out, <2 x double> %a, <2 x double> %b) nounwind {
61 %val = call <2 x double> @llvm.minnum.v2f64(<2 x double> %a, <2 x double> %b) #0
62 store <2 x double> %val, ptr addrspace(1) %out, align 16
66 ; GCN-LABEL: {{^}}test_fmin_v4f64:
71 define amdgpu_kernel void @test_fmin_v4f64(ptr addrspace(1) %out, <4 x double> %a, <4 x double> %b) nounwind {
72 %val = call <4 x double> @llvm.minnum.v4f64(<4 x double> %a, <4 x double> %b) #0
73 store <4 x double> %val, ptr addrspace(1) %out, align 32
77 ; GCN-LABEL: {{^}}test_fmin_v8f64:
86 define amdgpu_kernel void @test_fmin_v8f64(ptr addrspace(1) %out, <8 x double> %a, <8 x double> %b) nounwind {
87 %val = call <8 x double> @llvm.minnum.v8f64(<8 x double> %a, <8 x double> %b) #0
88 store <8 x double> %val, ptr addrspace(1) %out, align 64
92 ; GCN-LABEL: {{^}}test_fmin_v16f64:
109 define amdgpu_kernel void @test_fmin_v16f64(ptr addrspace(1) %out, <16 x double> %a, <16 x double> %b) nounwind {
110 %val = call <16 x double> @llvm.minnum.v16f64(<16 x double> %a, <16 x double> %b) #0
111 store <16 x double> %val, ptr addrspace(1) %out, align 128
115 attributes #0 = { nounwind readnone }
116 attributes #1 = { nounwind "denormal-fp-math"="ieee,ieee" }
117 attributes #2 = { nounwind "denormal-fp-math"="preserve-sign,preserve-sign" }