1 ; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=R600 -check-prefix=FUNC %s
2 ; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
4 declare i32 @llvm.r600.read.tidig.x() nounwind readnone
6 ; FUNC-LABEL: {{^}}setcc_v2i32:
7 ; R600-DAG: SETE_INT * T{{[0-9]+\.[XYZW]}}, KC0[3].X, KC0[3].Z
8 ; R600-DAG: SETE_INT * T{{[0-9]+\.[XYZW]}}, KC0[2].W, KC0[3].Y
10 define void @setcc_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> %a, <2 x i32> %b) {
11 %result = icmp eq <2 x i32> %a, %b
12 %sext = sext <2 x i1> %result to <2 x i32>
13 store <2 x i32> %sext, <2 x i32> addrspace(1)* %out
17 ; FUNC-LABEL: {{^}}setcc_v4i32:
18 ; R600-DAG: SETE_INT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
19 ; R600-DAG: SETE_INT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
20 ; R600-DAG: SETE_INT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
21 ; R600-DAG: SETE_INT * T{{[0-9]+\.[XYZW], T[0-9]+\.[XYZW], T[0-9]+\.[XYZW]}}
23 define void @setcc_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
24 %b_ptr = getelementptr <4 x i32> addrspace(1)* %in, i32 1
25 %a = load <4 x i32> addrspace(1) * %in
26 %b = load <4 x i32> addrspace(1) * %b_ptr
27 %result = icmp eq <4 x i32> %a, %b
28 %sext = sext <4 x i1> %result to <4 x i32>
29 store <4 x i32> %sext, <4 x i32> addrspace(1)* %out
33 ;;;==========================================================================;;;
35 ;;;==========================================================================;;;
37 ; FUNC-LABEL: {{^}}f32_oeq:
40 define void @f32_oeq(i32 addrspace(1)* %out, float %a, float %b) {
42 %0 = fcmp oeq float %a, %b
43 %1 = sext i1 %0 to i32
44 store i32 %1, i32 addrspace(1)* %out
48 ; FUNC-LABEL: {{^}}f32_ogt:
51 define void @f32_ogt(i32 addrspace(1)* %out, float %a, float %b) {
53 %0 = fcmp ogt float %a, %b
54 %1 = sext i1 %0 to i32
55 store i32 %1, i32 addrspace(1)* %out
59 ; FUNC-LABEL: {{^}}f32_oge:
62 define void @f32_oge(i32 addrspace(1)* %out, float %a, float %b) {
64 %0 = fcmp oge float %a, %b
65 %1 = sext i1 %0 to i32
66 store i32 %1, i32 addrspace(1)* %out
70 ; FUNC-LABEL: {{^}}f32_olt:
73 define void @f32_olt(i32 addrspace(1)* %out, float %a, float %b) {
75 %0 = fcmp olt float %a, %b
76 %1 = sext i1 %0 to i32
77 store i32 %1, i32 addrspace(1)* %out
81 ; FUNC-LABEL: {{^}}f32_ole:
84 define void @f32_ole(i32 addrspace(1)* %out, float %a, float %b) {
86 %0 = fcmp ole float %a, %b
87 %1 = sext i1 %0 to i32
88 store i32 %1, i32 addrspace(1)* %out
92 ; FUNC-LABEL: {{^}}f32_one:
96 ; R600-DAG: SETNE_DX10
100 ; SI-DAG: v_cmp_o_f32_e32 vcc
101 ; SI-DAG: v_cmp_neq_f32_e64 [[CMP1:s\[[0-9]+:[0-9]+\]]]
102 ; SI: s_and_b64 [[AND:s\[[0-9]+:[0-9]+\]]], [[CMP1]], vcc
103 ; SI: v_cndmask_b32_e64 [[VRESULT:v[0-9]+]], 0, -1, [[AND]]
104 ; SI: buffer_store_dword [[VRESULT]]
105 define void @f32_one(i32 addrspace(1)* %out, float %a, float %b) {
107 %0 = fcmp one float %a, %b
108 %1 = sext i1 %0 to i32
109 store i32 %1, i32 addrspace(1)* %out
113 ; FUNC-LABEL: {{^}}f32_ord:
114 ; R600-DAG: SETE_DX10
115 ; R600-DAG: SETE_DX10
117 ; R600-DAG: SETNE_INT
119 define void @f32_ord(i32 addrspace(1)* %out, float %a, float %b) {
121 %0 = fcmp ord float %a, %b
122 %1 = sext i1 %0 to i32
123 store i32 %1, i32 addrspace(1)* %out
127 ; FUNC-LABEL: {{^}}f32_ueq:
128 ; R600-DAG: SETNE_DX10
129 ; R600-DAG: SETNE_DX10
131 ; R600-DAG: SETE_DX10
133 ; R600-DAG: SETNE_INT
135 ; SI-DAG: v_cmp_u_f32_e32 vcc
136 ; SI-DAG: v_cmp_eq_f32_e64 [[CMP1:s\[[0-9]+:[0-9]+\]]]
137 ; SI: s_or_b64 [[OR:s\[[0-9]+:[0-9]+\]]], [[CMP1]], vcc
138 ; SI: v_cndmask_b32_e64 [[VRESULT:v[0-9]+]], 0, -1, [[OR]]
139 ; SI: buffer_store_dword [[VRESULT]]
140 define void @f32_ueq(i32 addrspace(1)* %out, float %a, float %b) {
142 %0 = fcmp ueq float %a, %b
143 %1 = sext i1 %0 to i32
144 store i32 %1, i32 addrspace(1)* %out
148 ; FUNC-LABEL: {{^}}f32_ugt:
155 define void @f32_ugt(i32 addrspace(1)* %out, float %a, float %b) {
157 %0 = fcmp ugt float %a, %b
158 %1 = sext i1 %0 to i32
159 store i32 %1, i32 addrspace(1)* %out
163 ; FUNC-LABEL: {{^}}f32_uge:
170 define void @f32_uge(i32 addrspace(1)* %out, float %a, float %b) {
172 %0 = fcmp uge float %a, %b
173 %1 = sext i1 %0 to i32
174 store i32 %1, i32 addrspace(1)* %out
178 ; FUNC-LABEL: {{^}}f32_ult:
185 define void @f32_ult(i32 addrspace(1)* %out, float %a, float %b) {
187 %0 = fcmp ult float %a, %b
188 %1 = sext i1 %0 to i32
189 store i32 %1, i32 addrspace(1)* %out
193 ; FUNC-LABEL: {{^}}f32_ule:
200 define void @f32_ule(i32 addrspace(1)* %out, float %a, float %b) {
202 %0 = fcmp ule float %a, %b
203 %1 = sext i1 %0 to i32
204 store i32 %1, i32 addrspace(1)* %out
208 ; FUNC-LABEL: {{^}}f32_une:
211 define void @f32_une(i32 addrspace(1)* %out, float %a, float %b) {
213 %0 = fcmp une float %a, %b
214 %1 = sext i1 %0 to i32
215 store i32 %1, i32 addrspace(1)* %out
219 ; FUNC-LABEL: {{^}}f32_uno:
225 define void @f32_uno(i32 addrspace(1)* %out, float %a, float %b) {
227 %0 = fcmp uno float %a, %b
228 %1 = sext i1 %0 to i32
229 store i32 %1, i32 addrspace(1)* %out
233 ;;;==========================================================================;;;
234 ;; 32-bit integer comparisons
235 ;;;==========================================================================;;;
237 ; FUNC-LABEL: {{^}}i32_eq:
240 define void @i32_eq(i32 addrspace(1)* %out, i32 %a, i32 %b) {
242 %0 = icmp eq i32 %a, %b
243 %1 = sext i1 %0 to i32
244 store i32 %1, i32 addrspace(1)* %out
248 ; FUNC-LABEL: {{^}}i32_ne:
251 define void @i32_ne(i32 addrspace(1)* %out, i32 %a, i32 %b) {
253 %0 = icmp ne i32 %a, %b
254 %1 = sext i1 %0 to i32
255 store i32 %1, i32 addrspace(1)* %out
259 ; FUNC-LABEL: {{^}}i32_ugt:
262 define void @i32_ugt(i32 addrspace(1)* %out, i32 %a, i32 %b) {
264 %0 = icmp ugt i32 %a, %b
265 %1 = sext i1 %0 to i32
266 store i32 %1, i32 addrspace(1)* %out
270 ; FUNC-LABEL: {{^}}i32_uge:
273 define void @i32_uge(i32 addrspace(1)* %out, i32 %a, i32 %b) {
275 %0 = icmp uge i32 %a, %b
276 %1 = sext i1 %0 to i32
277 store i32 %1, i32 addrspace(1)* %out
281 ; FUNC-LABEL: {{^}}i32_ult:
284 define void @i32_ult(i32 addrspace(1)* %out, i32 %a, i32 %b) {
286 %0 = icmp ult i32 %a, %b
287 %1 = sext i1 %0 to i32
288 store i32 %1, i32 addrspace(1)* %out
292 ; FUNC-LABEL: {{^}}i32_ule:
295 define void @i32_ule(i32 addrspace(1)* %out, i32 %a, i32 %b) {
297 %0 = icmp ule i32 %a, %b
298 %1 = sext i1 %0 to i32
299 store i32 %1, i32 addrspace(1)* %out
303 ; FUNC-LABEL: {{^}}i32_sgt:
306 define void @i32_sgt(i32 addrspace(1)* %out, i32 %a, i32 %b) {
308 %0 = icmp sgt i32 %a, %b
309 %1 = sext i1 %0 to i32
310 store i32 %1, i32 addrspace(1)* %out
314 ; FUNC-LABEL: {{^}}i32_sge:
317 define void @i32_sge(i32 addrspace(1)* %out, i32 %a, i32 %b) {
319 %0 = icmp sge i32 %a, %b
320 %1 = sext i1 %0 to i32
321 store i32 %1, i32 addrspace(1)* %out
325 ; FUNC-LABEL: {{^}}i32_slt:
328 define void @i32_slt(i32 addrspace(1)* %out, i32 %a, i32 %b) {
330 %0 = icmp slt i32 %a, %b
331 %1 = sext i1 %0 to i32
332 store i32 %1, i32 addrspace(1)* %out
336 ; FUNC-LABEL: {{^}}i32_sle:
339 define void @i32_sle(i32 addrspace(1)* %out, i32 %a, i32 %b) {
341 %0 = icmp sle i32 %a, %b
342 %1 = sext i1 %0 to i32
343 store i32 %1, i32 addrspace(1)* %out
347 ; FIXME: This does 4 compares
348 ; FUNC-LABEL: {{^}}v3i32_eq:
349 ; SI-DAG: v_cmp_eq_i32
350 ; SI-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
351 ; SI-DAG: v_cmp_eq_i32
352 ; SI-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
353 ; SI-DAG: v_cmp_eq_i32
354 ; SI-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
356 define void @v3i32_eq(<3 x i32> addrspace(1)* %out, <3 x i32> addrspace(1)* %ptra, <3 x i32> addrspace(1)* %ptrb) {
357 %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
358 %gep.a = getelementptr <3 x i32> addrspace(1)* %ptra, i32 %tid
359 %gep.b = getelementptr <3 x i32> addrspace(1)* %ptrb, i32 %tid
360 %gep.out = getelementptr <3 x i32> addrspace(1)* %out, i32 %tid
361 %a = load <3 x i32> addrspace(1)* %gep.a
362 %b = load <3 x i32> addrspace(1)* %gep.b
363 %cmp = icmp eq <3 x i32> %a, %b
364 %ext = sext <3 x i1> %cmp to <3 x i32>
365 store <3 x i32> %ext, <3 x i32> addrspace(1)* %gep.out
369 ; FUNC-LABEL: {{^}}v3i8_eq:
370 ; SI-DAG: v_cmp_eq_i32
371 ; SI-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
372 ; SI-DAG: v_cmp_eq_i32
373 ; SI-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
374 ; SI-DAG: v_cmp_eq_i32
375 ; SI-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
377 define void @v3i8_eq(<3 x i8> addrspace(1)* %out, <3 x i8> addrspace(1)* %ptra, <3 x i8> addrspace(1)* %ptrb) {
378 %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
379 %gep.a = getelementptr <3 x i8> addrspace(1)* %ptra, i32 %tid
380 %gep.b = getelementptr <3 x i8> addrspace(1)* %ptrb, i32 %tid
381 %gep.out = getelementptr <3 x i8> addrspace(1)* %out, i32 %tid
382 %a = load <3 x i8> addrspace(1)* %gep.a
383 %b = load <3 x i8> addrspace(1)* %gep.b
384 %cmp = icmp eq <3 x i8> %a, %b
385 %ext = sext <3 x i1> %cmp to <3 x i8>
386 store <3 x i8> %ext, <3 x i8> addrspace(1)* %gep.out