• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
2
3; These tests check that floating point comparisons which are used by select
4; to store integer true (-1) and false (0) values are lowered to one of the
5; SET*DX10 instructions.
6
7; CHECK: {{^}}fcmp_une_select_fptosi:
8; CHECK: LSHR
9; CHECK-NEXT: SETNE_DX10 * {{\** *}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z, literal.y,
10; CHECK-NEXT: 1084227584(5.000000e+00)
11define amdgpu_kernel void @fcmp_une_select_fptosi(i32 addrspace(1)* %out, float %in) {
12entry:
13  %0 = fcmp une float %in, 5.0
14  %1 = select i1 %0, float 1.000000e+00, float 0.000000e+00
15  %2 = fsub float -0.000000e+00, %1
16  %3 = fptosi float %2 to i32
17  store i32 %3, i32 addrspace(1)* %out
18  ret void
19}
20
21; CHECK: {{^}}fcmp_une_select_i32:
22; CHECK: LSHR
23; CHECK-NEXT: SETNE_DX10 * {{\** *}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z, literal.y,
24; CHECK-NEXT: 1084227584(5.000000e+00)
25define amdgpu_kernel void @fcmp_une_select_i32(i32 addrspace(1)* %out, float %in) {
26entry:
27  %0 = fcmp une float %in, 5.0
28  %1 = select i1 %0, i32 -1, i32 0
29  store i32 %1, i32 addrspace(1)* %out
30  ret void
31}
32
33; CHECK: {{^}}fcmp_oeq_select_fptosi:
34; CHECK: LSHR
35; CHECK-NEXT: SETE_DX10 * {{\** *}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z, literal.y,
36; CHECK-NEXT: 1084227584(5.000000e+00)
37define amdgpu_kernel void @fcmp_oeq_select_fptosi(i32 addrspace(1)* %out, float %in) {
38entry:
39  %0 = fcmp oeq float %in, 5.0
40  %1 = select i1 %0, float 1.000000e+00, float 0.000000e+00
41  %2 = fsub float -0.000000e+00, %1
42  %3 = fptosi float %2 to i32
43  store i32 %3, i32 addrspace(1)* %out
44  ret void
45}
46
47; CHECK: {{^}}fcmp_oeq_select_i32:
48; CHECK: LSHR
49; CHECK-NEXT: SETE_DX10 * {{\** *}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z, literal.y,
50; CHECK-NEXT: 1084227584(5.000000e+00)
51define amdgpu_kernel void @fcmp_oeq_select_i32(i32 addrspace(1)* %out, float %in) {
52entry:
53  %0 = fcmp oeq float %in, 5.0
54  %1 = select i1 %0, i32 -1, i32 0
55  store i32 %1, i32 addrspace(1)* %out
56  ret void
57}
58
59; CHECK: {{^}}fcmp_ogt_select_fptosi:
60; CHECK: LSHR
61; CHECK-NEXT: SETGT_DX10 * {{\** *}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z, literal.y,
62; CHECK-NEXT: 1084227584(5.000000e+00)
63define amdgpu_kernel void @fcmp_ogt_select_fptosi(i32 addrspace(1)* %out, float %in) {
64entry:
65  %0 = fcmp ogt float %in, 5.0
66  %1 = select i1 %0, float 1.000000e+00, float 0.000000e+00
67  %2 = fsub float -0.000000e+00, %1
68  %3 = fptosi float %2 to i32
69  store i32 %3, i32 addrspace(1)* %out
70  ret void
71}
72
73; CHECK: {{^}}fcmp_ogt_select_i32:
74; CHECK: LSHR
75; CHECK-NEXT: SETGT_DX10 * {{\** *}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z, literal.y,
76; CHECK-NEXT: 1084227584(5.000000e+00)
77define amdgpu_kernel void @fcmp_ogt_select_i32(i32 addrspace(1)* %out, float %in) {
78entry:
79  %0 = fcmp ogt float %in, 5.0
80  %1 = select i1 %0, i32 -1, i32 0
81  store i32 %1, i32 addrspace(1)* %out
82  ret void
83}
84
85; CHECK: {{^}}fcmp_oge_select_fptosi:
86; CHECK: LSHR
87; CHECK-NEXT: SETGE_DX10 * {{\** *}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z, literal.y,
88; CHECK-NEXT: 1084227584(5.000000e+00)
89define amdgpu_kernel void @fcmp_oge_select_fptosi(i32 addrspace(1)* %out, float %in) {
90entry:
91  %0 = fcmp oge float %in, 5.0
92  %1 = select i1 %0, float 1.000000e+00, float 0.000000e+00
93  %2 = fsub float -0.000000e+00, %1
94  %3 = fptosi float %2 to i32
95  store i32 %3, i32 addrspace(1)* %out
96  ret void
97}
98
99; CHECK: {{^}}fcmp_oge_select_i32:
100; CHECK: LSHR
101; CHECK-NEXT: SETGE_DX10 * {{\** *}}T{{[0-9]+\.[XYZW]}}, KC0[2].Z, literal.y,
102; CHECK-NEXT: 1084227584(5.000000e+00)
103define amdgpu_kernel void @fcmp_oge_select_i32(i32 addrspace(1)* %out, float %in) {
104entry:
105  %0 = fcmp oge float %in, 5.0
106  %1 = select i1 %0, i32 -1, i32 0
107  store i32 %1, i32 addrspace(1)* %out
108  ret void
109}
110
111; CHECK: {{^}}fcmp_ole_select_fptosi:
112; CHECK: LSHR
113; CHECK-NEXT: SETGE_DX10 * {{\** *}}T{{[0-9]+\.[XYZW]}}, literal.y, KC0[2].Z,
114; CHECK-NEXT: 1084227584(5.000000e+00)
115define amdgpu_kernel void @fcmp_ole_select_fptosi(i32 addrspace(1)* %out, float %in) {
116entry:
117  %0 = fcmp ole float %in, 5.0
118  %1 = select i1 %0, float 1.000000e+00, float 0.000000e+00
119  %2 = fsub float -0.000000e+00, %1
120  %3 = fptosi float %2 to i32
121  store i32 %3, i32 addrspace(1)* %out
122  ret void
123}
124
125; CHECK: {{^}}fcmp_ole_select_i32:
126; CHECK: LSHR
127; CHECK-NEXT: SETGE_DX10 * {{\** *}}T{{[0-9]+\.[XYZW]}}, literal.y, KC0[2].Z,
128; CHECK-NEXT: 1084227584(5.000000e+00)
129define amdgpu_kernel void @fcmp_ole_select_i32(i32 addrspace(1)* %out, float %in) {
130entry:
131  %0 = fcmp ole float %in, 5.0
132  %1 = select i1 %0, i32 -1, i32 0
133  store i32 %1, i32 addrspace(1)* %out
134  ret void
135}
136
137; CHECK: {{^}}fcmp_olt_select_fptosi:
138; CHECK: LSHR
139; CHECK-NEXT: SETGT_DX10 * {{\** *}}T{{[0-9]+\.[XYZW]}}, literal.y, KC0[2].Z,
140; CHECK-NEXT: 1084227584(5.000000e+00)
141define amdgpu_kernel void @fcmp_olt_select_fptosi(i32 addrspace(1)* %out, float %in) {
142entry:
143  %0 = fcmp olt float %in, 5.0
144  %1 = select i1 %0, float 1.000000e+00, float 0.000000e+00
145  %2 = fsub float -0.000000e+00, %1
146  %3 = fptosi float %2 to i32
147  store i32 %3, i32 addrspace(1)* %out
148  ret void
149}
150
151; CHECK: {{^}}fcmp_olt_select_i32:
152; CHECK: LSHR
153; CHECK-NEXT: SETGT_DX10 * {{\** *}}T{{[0-9]+\.[XYZW]}}, literal.y, KC0[2].Z,
154; CHECK-NEXT: 1084227584(5.000000e+00)
155define amdgpu_kernel void @fcmp_olt_select_i32(i32 addrspace(1)* %out, float %in) {
156entry:
157  %0 = fcmp olt float %in, 5.0
158  %1 = select i1 %0, i32 -1, i32 0
159  store i32 %1, i32 addrspace(1)* %out
160  ret void
161}
162