1 ; RUN: llc -march=amdgcn -verify-machineinstrs -enable-misched=0 < %s | FileCheck -check-prefix=GCN -check-prefix=SI -check-prefix=FUNC %s
2 ; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global,-xnack -enable-misched=0 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI -check-prefix=FUNC %s
3 ; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=R600 -check-prefix=FUNC %s
6 ; DAGCombiner will transform:
7 ; (fabsf (f32 bitcast (i32 a))) => (f32 bitcast (and (i32 a), 0x7FFFFFFF))
8 ; unless isFabsFree returns true
10 ; FUNC-LABEL: {{^}}s_fabsf_fn_free:
12 ; R600: |PV.{{[XYZW]}}|
14 ; GCN: s_bitset0_b32 s{{[0-9]+}}, 31
15 define amdgpu_kernel void @s_fabsf_fn_free(ptr addrspace(1) %out, i32 %in) {
16 %bc= bitcast i32 %in to float
17 %fabs = call float @fabsf(float %bc)
18 store float %fabs, ptr addrspace(1) %out
22 ; FUNC-LABEL: {{^}}s_fabsf_free:
24 ; R600: |PV.{{[XYZW]}}|
26 ; GCN: s_bitset0_b32 s{{[0-9]+}}, 31
27 define amdgpu_kernel void @s_fabsf_free(ptr addrspace(1) %out, i32 %in) {
28 %bc= bitcast i32 %in to float
29 %fabs = call float @llvm.fabs.f32(float %bc)
30 store float %fabs, ptr addrspace(1) %out
34 ; FUNC-LABEL: {{^}}s_fabsf_f32:
35 ; R600: |{{(PV|T[0-9])\.[XYZW]}}|
37 ; GCN: s_bitset0_b32 s{{[0-9]+}}, 31
38 define amdgpu_kernel void @s_fabsf_f32(ptr addrspace(1) %out, float %in) {
39 %fabs = call float @llvm.fabs.f32(float %in)
40 store float %fabs, ptr addrspace(1) %out
44 ; FUNC-LABEL: {{^}}fabs_v2f32:
45 ; R600: |{{(PV|T[0-9])\.[XYZW]}}|
46 ; R600: |{{(PV|T[0-9])\.[XYZW]}}|
48 ; GCN: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x7fffffff
49 ; GCN: s_and_b32 s{{[0-9]+}}, s{{[0-9]+}}, 0x7fffffff
50 define amdgpu_kernel void @fabs_v2f32(ptr addrspace(1) %out, <2 x float> %in) {
51 %fabs = call <2 x float> @llvm.fabs.v2f32(<2 x float> %in)
52 store <2 x float> %fabs, ptr addrspace(1) %out
56 ; FUNC-LABEL: {{^}}fabsf_v4f32:
57 ; R600: |{{(PV|T[0-9])\.[XYZW]}}|
58 ; R600: |{{(PV|T[0-9])\.[XYZW]}}|
59 ; R600: |{{(PV|T[0-9])\.[XYZW]}}|
60 ; R600: |{{(PV|T[0-9])\.[XYZW]}}|
66 define amdgpu_kernel void @fabsf_v4f32(ptr addrspace(1) %out, <4 x float> %in) {
67 %fabs = call <4 x float> @llvm.fabs.v4f32(<4 x float> %in)
68 store <4 x float> %fabs, ptr addrspace(1) %out
72 ; GCN-LABEL: {{^}}fabsf_fn_fold:
73 ; SI: s_load_dwordx4 s[[[#LOAD:]]:[[#END:]]], s[{{[0-9]+:[0-9]+}}], 0x9
74 ; VI: s_load_dwordx4 s[[[#LOAD:]]:[[#END:]]], s[{{[0-9]+:[0-9]+}}], 0x24
76 ; GCN: v_mov_b32_e32 [[V_MUL_VI:v[0-9]+]], s[[#LOAD + 3]]
77 ; GCN: v_mul_f32_e64 v{{[0-9]+}}, |s[[#LOAD + 2]]|, [[V_MUL_VI]]
78 define amdgpu_kernel void @fabsf_fn_fold(ptr addrspace(1) %out, float %in0, float %in1) {
79 %fabs = call float @fabsf(float %in0)
80 %fmul = fmul float %fabs, %in1
81 store float %fmul, ptr addrspace(1) %out
85 ; FUNC-LABEL: {{^}}fabs_fold:
86 ; SI: s_load_dwordx4 s[[[#LOAD:]]:[[#END:]]], s[{{[0-9]+:[0-9]+}}], 0x9
87 ; VI: s_load_dwordx4 s[[[#LOAD:]]:[[#END:]]], s[{{[0-9]+:[0-9]+}}], 0x24
89 ; GCN: v_mov_b32_e32 [[V_MUL_VI:v[0-9]+]], s[[#LOAD + 3]]
90 ; GCN: v_mul_f32_e64 v{{[0-9]+}}, |s[[#LOAD + 2]]|, [[V_MUL_VI]]
91 define amdgpu_kernel void @fabs_fold(ptr addrspace(1) %out, float %in0, float %in1) {
92 %fabs = call float @llvm.fabs.f32(float %in0)
93 %fmul = fmul float %fabs, %in1
94 store float %fmul, ptr addrspace(1) %out
98 ; Make sure we turn some integer operations back into fabsf
99 ; FUNC-LABEL: {{^}}bitpreserve_fabsf_f32:
100 ; GCN: v_add_f32_e64 v{{[0-9]+}}, |s{{[0-9]+}}|, 1.0
101 define amdgpu_kernel void @bitpreserve_fabsf_f32(ptr addrspace(1) %out, float %in) {
102 %in.bc = bitcast float %in to i32
103 %int.abs = and i32 %in.bc, 2147483647
104 %bc = bitcast i32 %int.abs to float
105 %fadd = fadd float %bc, 1.0
106 store float %fadd, ptr addrspace(1) %out
110 declare float @fabsf(float) readnone
111 declare float @llvm.fabs.f32(float) readnone
112 declare <2 x float> @llvm.fabs.v2f32(<2 x float>) readnone
113 declare <4 x float> @llvm.fabs.v4f32(<4 x float>) readnone