• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=GCN -check-prefix=FUNC %s
2; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=VI -check-prefix=GCN -check-prefix=FUNC %s
3; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=R600 -check-prefix=FUNC %s
4
5; FUNC-LABEL: {{^}}fneg_f32:
6; R600: -PV
7
8; GCN: v_xor_b32
9define void @fneg_f32(float addrspace(1)* %out, float %in) {
10  %fneg = fsub float -0.000000e+00, %in
11  store float %fneg, float addrspace(1)* %out
12  ret void
13}
14
15; FUNC-LABEL: {{^}}fneg_v2f32:
16; R600: -PV
17; R600: -PV
18
19; GCN: v_xor_b32
20; GCN: v_xor_b32
21define void @fneg_v2f32(<2 x float> addrspace(1)* nocapture %out, <2 x float> %in) {
22  %fneg = fsub <2 x float> <float -0.000000e+00, float -0.000000e+00>, %in
23  store <2 x float> %fneg, <2 x float> addrspace(1)* %out
24  ret void
25}
26
27; FUNC-LABEL: {{^}}fneg_v4f32:
28; R600: -PV
29; R600: -T
30; R600: -PV
31; R600: -PV
32
33; GCN: v_xor_b32
34; GCN: v_xor_b32
35; GCN: v_xor_b32
36; GCN: v_xor_b32
37define void @fneg_v4f32(<4 x float> addrspace(1)* nocapture %out, <4 x float> %in) {
38  %fneg = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %in
39  store <4 x float> %fneg, <4 x float> addrspace(1)* %out
40  ret void
41}
42
43; DAGCombiner will transform:
44; (fneg (f32 bitcast (i32 a))) => (f32 bitcast (xor (i32 a), 0x80000000))
45; unless the target returns true for isNegFree()
46
47; FUNC-LABEL: {{^}}fneg_free_f32:
48; R600-NOT: XOR
49; R600: -KC0[2].Z
50
51; XXX: We could use v_add_f32_e64 with the negate bit here instead.
52; GCN: v_sub_f32_e64 v{{[0-9]}}, 0, s{{[0-9]+$}}
53define void @fneg_free_f32(float addrspace(1)* %out, i32 %in) {
54  %bc = bitcast i32 %in to float
55  %fsub = fsub float 0.0, %bc
56  store float %fsub, float addrspace(1)* %out
57  ret void
58}
59
60; FUNC-LABEL: {{^}}fneg_fold_f32:
61; SI: s_load_dword [[NEG_VALUE:s[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0xb
62; VI: s_load_dword [[NEG_VALUE:s[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0x2c
63; GCN-NOT: xor
64; GCN: v_mul_f32_e64 v{{[0-9]+}}, -[[NEG_VALUE]], [[NEG_VALUE]]
65define void @fneg_fold_f32(float addrspace(1)* %out, float %in) {
66  %fsub = fsub float -0.0, %in
67  %fmul = fmul float %fsub, %in
68  store float %fmul, float addrspace(1)* %out
69  ret void
70}
71