1 ; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -enable-var-scope -check-prefix=GCN %s
3 ; GCN-LABEL: {{^}}select_and1:
4 ; GCN: v_cndmask_b32_e32 [[SEL:v[0-9]+]], 0, v{{[0-9]+}},
6 ; GCN: store_dword v[{{[0-9:]+}}], [[SEL]],
7 define amdgpu_kernel void @select_and1(i32 addrspace(1)* %p, i32 %x, i32 %y) {
8 %c = icmp slt i32 %x, 11
9 %s = select i1 %c, i32 0, i32 -1
11 store i32 %a, i32 addrspace(1)* %p, align 4
15 ; GCN-LABEL: {{^}}select_and2:
16 ; GCN: v_cndmask_b32_e32 [[SEL:v[0-9]+]], 0, v{{[0-9]+}},
18 ; GCN: store_dword v[{{[0-9:]+}}], [[SEL]],
19 define amdgpu_kernel void @select_and2(i32 addrspace(1)* %p, i32 %x, i32 %y) {
20 %c = icmp slt i32 %x, 11
21 %s = select i1 %c, i32 0, i32 -1
23 store i32 %a, i32 addrspace(1)* %p, align 4
27 ; GCN-LABEL: {{^}}select_and3:
28 ; GCN: v_cndmask_b32_e32 [[SEL:v[0-9]+]], 0, v{{[0-9]+}},
30 ; GCN: store_dword v[{{[0-9:]+}}], [[SEL]],
31 define amdgpu_kernel void @select_and3(i32 addrspace(1)* %p, i32 %x, i32 %y) {
32 %c = icmp slt i32 %x, 11
33 %s = select i1 %c, i32 -1, i32 0
35 store i32 %a, i32 addrspace(1)* %p, align 4
39 ; GCN-LABEL: {{^}}select_and_v4:
40 ; GCN: v_cndmask_b32_e32 [[SEL:v[0-9]+]], 0, v{{[0-9]+}},
41 ; GCN: v_cndmask_b32_e32 [[SEL:v[0-9]+]], 0, v{{[0-9]+}},
42 ; GCN: v_cndmask_b32_e32 [[SEL:v[0-9]+]], 0, v{{[0-9]+}},
43 ; GCN: v_cndmask_b32_e32 [[SEL:v[0-9]+]], 0, v{{[0-9]+}},
46 define amdgpu_kernel void @select_and_v4(<4 x i32> addrspace(1)* %p, i32 %x, <4 x i32> %y) {
47 %c = icmp slt i32 %x, 11
48 %s = select i1 %c, <4 x i32> zeroinitializer, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>
49 %a = and <4 x i32> %s, %y
50 store <4 x i32> %a, <4 x i32> addrspace(1)* %p, align 32
54 ; GCN-LABEL: {{^}}select_or1:
55 ; GCN: v_cndmask_b32_e32 [[SEL:v[0-9]+]], -1, v{{[0-9]+}},
57 ; GCN: store_dword v[{{[0-9:]+}}], [[SEL]],
58 define amdgpu_kernel void @select_or1(i32 addrspace(1)* %p, i32 %x, i32 %y) {
59 %c = icmp slt i32 %x, 11
60 %s = select i1 %c, i32 0, i32 -1
62 store i32 %a, i32 addrspace(1)* %p, align 4
66 ; GCN-LABEL: {{^}}select_or2:
67 ; GCN: v_cndmask_b32_e32 [[SEL:v[0-9]+]], -1, v{{[0-9]+}},
69 ; GCN: store_dword v[{{[0-9:]+}}], [[SEL]],
70 define amdgpu_kernel void @select_or2(i32 addrspace(1)* %p, i32 %x, i32 %y) {
71 %c = icmp slt i32 %x, 11
72 %s = select i1 %c, i32 0, i32 -1
74 store i32 %a, i32 addrspace(1)* %p, align 4
78 ; GCN-LABEL: {{^}}select_or3:
79 ; GCN: v_cndmask_b32_e32 [[SEL:v[0-9]+]], -1, v{{[0-9]+}},
81 ; GCN: store_dword v[{{[0-9:]+}}], [[SEL]],
82 define amdgpu_kernel void @select_or3(i32 addrspace(1)* %p, i32 %x, i32 %y) {
83 %c = icmp slt i32 %x, 11
84 %s = select i1 %c, i32 -1, i32 0
86 store i32 %a, i32 addrspace(1)* %p, align 4
90 ; GCN-LABEL: {{^}}select_or_v4:
91 ; GCN: v_cndmask_b32_e32 [[SEL:v[0-9]+]], -1, v{{[0-9]+}},
92 ; GCN: v_cndmask_b32_e32 [[SEL:v[0-9]+]], -1, v{{[0-9]+}},
93 ; GCN: v_cndmask_b32_e32 [[SEL:v[0-9]+]], -1, v{{[0-9]+}},
94 ; GCN: v_cndmask_b32_e32 [[SEL:v[0-9]+]], -1, v{{[0-9]+}},
97 define amdgpu_kernel void @select_or_v4(<4 x i32> addrspace(1)* %p, i32 %x, <4 x i32> %y) {
98 %c = icmp slt i32 %x, 11
99 %s = select i1 %c, <4 x i32> zeroinitializer, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 -1>
100 %a = or <4 x i32> %s, %y
101 store <4 x i32> %a, <4 x i32> addrspace(1)* %p, align 32
105 ; GCN-LABEL: {{^}}sel_constants_sub_constant_sel_constants:
106 ; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 2, 9,
107 define amdgpu_kernel void @sel_constants_sub_constant_sel_constants(i32 addrspace(1)* %p, i1 %cond) {
108 %sel = select i1 %cond, i32 -4, i32 3
109 %bo = sub i32 5, %sel
110 store i32 %bo, i32 addrspace(1)* %p, align 4
114 ; GCN-LABEL: {{^}}sel_constants_sub_constant_sel_constants_i16:
115 ; TODO: shrink i16 constant. This is correct but suboptimal.
116 ; GCN: v_mov_b32_e32 [[T:v[0-9]+]], 0xffff0009
117 ; GCN: v_cndmask_b32_e32 v{{[0-9]+}}, 2, [[T]],
118 define amdgpu_kernel void @sel_constants_sub_constant_sel_constants_i16(i16 addrspace(1)* %p, i1 %cond) {
119 %sel = select i1 %cond, i16 -4, i16 3
120 %bo = sub i16 5, %sel
121 store i16 %bo, i16 addrspace(1)* %p, align 2
125 ; GCN-LABEL: {{^}}sel_constants_sub_constant_sel_constants_i16_neg:
126 ; GCN: v_mov_b32_e32 [[F:v[0-9]+]], 0xfffff449
127 ; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, [[F]], -3,
128 define amdgpu_kernel void @sel_constants_sub_constant_sel_constants_i16_neg(i16 addrspace(1)* %p, i1 %cond) {
129 %sel = select i1 %cond, i16 4, i16 3000
130 %bo = sub i16 1, %sel
131 store i16 %bo, i16 addrspace(1)* %p, align 2
135 ; GCN-LABEL: {{^}}sel_constants_sub_constant_sel_constants_v2i16:
136 ; GCN-DAG: v_mov_b32_e32 [[F:v[0-9]+]], 0x60002
137 ; GCN-DAG: v_mov_b32_e32 [[T:v[0-9]+]], 0x50009
138 ; GCN: v_cndmask_b32_e32 v{{[0-9]+}}, [[F]], [[T]],
139 define amdgpu_kernel void @sel_constants_sub_constant_sel_constants_v2i16(<2 x i16> addrspace(1)* %p, i1 %cond) {
140 %sel = select i1 %cond, <2 x i16> <i16 -4, i16 2>, <2 x i16> <i16 3, i16 1>
141 %bo = sub <2 x i16> <i16 5, i16 7>, %sel
142 store <2 x i16> %bo, <2 x i16> addrspace(1)* %p, align 4
146 ; GCN-LABEL: {{^}}sel_constants_sub_constant_sel_constants_v4i32:
147 ; GCN-DAG: v_cndmask_b32_e64 v{{[0-9]+}}, 2, 9,
148 ; GCN-DAG: v_cndmask_b32_e64 v{{[0-9]+}}, 6, 5,
149 ; GCN-DAG: v_cndmask_b32_e64 v{{[0-9]+}}, 10, 6,
150 ; GCN-DAG: v_cndmask_b32_e64 v{{[0-9]+}}, 14, 7,
151 define amdgpu_kernel void @sel_constants_sub_constant_sel_constants_v4i32(<4 x i32> addrspace(1)* %p, i1 %cond) {
152 %sel = select i1 %cond, <4 x i32> <i32 -4, i32 2, i32 3, i32 4>, <4 x i32> <i32 3, i32 1, i32 -1, i32 -3>
153 %bo = sub <4 x i32> <i32 5, i32 7, i32 9, i32 11>, %sel
154 store <4 x i32> %bo, <4 x i32> addrspace(1)* %p, align 32
158 ; GCN-LABEL: {{^}}sdiv_constant_sel_constants:
159 ; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 5, 0,
160 define amdgpu_kernel void @sdiv_constant_sel_constants(i64 addrspace(1)* %p, i1 %cond) {
161 %sel = select i1 %cond, i64 121, i64 23
162 %bo = sdiv i64 120, %sel
163 store i64 %bo, i64 addrspace(1)* %p, align 8
167 ; GCN-LABEL: {{^}}udiv_constant_sel_constants:
168 ; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 5, 0,
169 define amdgpu_kernel void @udiv_constant_sel_constants(i64 addrspace(1)* %p, i1 %cond) {
170 %sel = select i1 %cond, i64 -4, i64 23
171 %bo = udiv i64 120, %sel
172 store i64 %bo, i64 addrspace(1)* %p, align 8
176 ; GCN-LABEL: {{^}}srem_constant_sel_constants:
177 ; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 3, 33,
178 define amdgpu_kernel void @srem_constant_sel_constants(i64 addrspace(1)* %p, i1 %cond) {
179 %sel = select i1 %cond, i64 34, i64 15
180 %bo = srem i64 33, %sel
181 store i64 %bo, i64 addrspace(1)* %p, align 8
185 ; GCN-LABEL: {{^}}urem_constant_sel_constants:
186 ; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 3, 33,
187 define amdgpu_kernel void @urem_constant_sel_constants(i64 addrspace(1)* %p, i1 %cond) {
188 %sel = select i1 %cond, i64 34, i64 15
189 %bo = urem i64 33, %sel
190 store i64 %bo, i64 addrspace(1)* %p, align 8
194 ; GCN-LABEL: {{^}}shl_constant_sel_constants:
195 ; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 8, 4,
196 define amdgpu_kernel void @shl_constant_sel_constants(i32 addrspace(1)* %p, i1 %cond) {
197 %sel = select i1 %cond, i32 2, i32 3
198 %bo = shl i32 1, %sel
199 store i32 %bo, i32 addrspace(1)* %p, align 4
203 ; GCN-LABEL: {{^}}lshr_constant_sel_constants:
204 ; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 8, 16,
205 define amdgpu_kernel void @lshr_constant_sel_constants(i32 addrspace(1)* %p, i1 %cond) {
206 %sel = select i1 %cond, i32 2, i32 3
207 %bo = lshr i32 64, %sel
208 store i32 %bo, i32 addrspace(1)* %p, align 4
212 ; GCN-LABEL: {{^}}ashr_constant_sel_constants:
213 ; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 16, 32,
214 define amdgpu_kernel void @ashr_constant_sel_constants(i32 addrspace(1)* %p, i1 %cond) {
215 %sel = select i1 %cond, i32 2, i32 3
216 %bo = ashr i32 128, %sel
217 store i32 %bo, i32 addrspace(1)* %p, align 4
221 ; GCN-LABEL: {{^}}fsub_constant_sel_constants:
222 ; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, -4.0, 1.0,
223 define amdgpu_kernel void @fsub_constant_sel_constants(float addrspace(1)* %p, i1 %cond) {
224 %sel = select i1 %cond, float -2.0, float 3.0
225 %bo = fsub float -1.0, %sel
226 store float %bo, float addrspace(1)* %p, align 4
230 ; GCN-LABEL: {{^}}fsub_constant_sel_constants_f16:
231 ; TODO: it shall be possible to fold constants with OpSel
232 ; GCN-DAG: v_mov_b32_e32 [[T:v[0-9]+]], 0x3c00
233 ; GCN-DAG: v_mov_b32_e32 [[F:v[0-9]+]], 0xc400
234 ; GCN: v_cndmask_b32_e32 v{{[0-9]+}}, [[F]], [[T]],
235 define amdgpu_kernel void @fsub_constant_sel_constants_f16(half addrspace(1)* %p, i1 %cond) {
236 %sel = select i1 %cond, half -2.0, half 3.0
237 %bo = fsub half -1.0, %sel
238 store half %bo, half addrspace(1)* %p, align 2
242 ; GCN-LABEL: {{^}}fsub_constant_sel_constants_v2f16:
243 ; GCN-DAG: v_mov_b32_e32 [[T:v[0-9]+]], 0x45003c00
244 ; GCN: v_cndmask_b32_e32 v{{[0-9]+}}, -2.0, [[T]],
245 define amdgpu_kernel void @fsub_constant_sel_constants_v2f16(<2 x half> addrspace(1)* %p, i1 %cond) {
246 %sel = select i1 %cond, <2 x half> <half -2.0, half -3.0>, <2 x half> <half -1.0, half 4.0>
247 %bo = fsub <2 x half> <half -1.0, half 2.0>, %sel
248 store <2 x half> %bo, <2 x half> addrspace(1)* %p, align 4
252 ; GCN-LABEL: {{^}}fsub_constant_sel_constants_v4f32:
253 ; GCN-DAG: v_mov_b32_e32 [[T2:v[0-9]+]], 0x40a00000
254 ; GCN-DAG: v_mov_b32_e32 [[T3:v[0-9]+]], 0x41100000
255 ; GCN-DAG: v_mov_b32_e32 [[T4:v[0-9]+]], 0x41500000
256 ; GCN-DAG: v_mov_b32_e32 [[F4:v[0-9]+]], 0x40c00000
257 ; GCN-DAG: v_cndmask_b32_e64 v{{[0-9]+}}, 0, 1.0,
258 ; GCN-DAG: v_cndmask_b32_e32 v{{[0-9]+}}, 2.0, [[T2]],
259 ; GCN-DAG: v_cndmask_b32_e32 v{{[0-9]+}}, 4.0, [[T3]],
260 ; GCN-DAG: v_cndmask_b32_e32 v{{[0-9]+}}, [[F4]], [[T4]],
261 define amdgpu_kernel void @fsub_constant_sel_constants_v4f32(<4 x float> addrspace(1)* %p, i1 %cond) {
262 %sel = select i1 %cond, <4 x float> <float -2.0, float -3.0, float -4.0, float -5.0>, <4 x float> <float -1.0, float 0.0, float 1.0, float 2.0>
263 %bo = fsub <4 x float> <float -1.0, float 2.0, float 5.0, float 8.0>, %sel
264 store <4 x float> %bo, <4 x float> addrspace(1)* %p, align 32
268 ; GCN-LABEL: {{^}}fdiv_constant_sel_constants:
269 ; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 4.0, -2.0,
270 define amdgpu_kernel void @fdiv_constant_sel_constants(float addrspace(1)* %p, i1 %cond) {
271 %sel = select i1 %cond, float -4.0, float 2.0
272 %bo = fdiv float 8.0, %sel
273 store float %bo, float addrspace(1)* %p, align 4
277 ; GCN-LABEL: {{^}}frem_constant_sel_constants:
278 ; GCN: v_cndmask_b32_e64 v{{[0-9]+}}, 2.0, 1.0,
279 define amdgpu_kernel void @frem_constant_sel_constants(float addrspace(1)* %p, i1 %cond) {
280 %sel = select i1 %cond, float -4.0, float 3.0
281 %bo = frem float 5.0, %sel
282 store float %bo, float addrspace(1)* %p, align 4