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: v_cmp_lg_f32_e32 vcc
101 ; SI-NEXT: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1, vcc
102 define void @f32_one(i32 addrspace(1)* %out, float %a, float %b) {
104 %0 = fcmp one float %a, %b
105 %1 = sext i1 %0 to i32
106 store i32 %1, i32 addrspace(1)* %out
110 ; FUNC-LABEL: {{^}}f32_ord:
111 ; R600-DAG: SETE_DX10
112 ; R600-DAG: SETE_DX10
114 ; R600-DAG: SETNE_INT
116 define void @f32_ord(i32 addrspace(1)* %out, float %a, float %b) {
118 %0 = fcmp ord float %a, %b
119 %1 = sext i1 %0 to i32
120 store i32 %1, i32 addrspace(1)* %out
124 ; FUNC-LABEL: {{^}}f32_ueq:
125 ; R600-DAG: SETNE_DX10
126 ; R600-DAG: SETNE_DX10
128 ; R600-DAG: SETE_DX10
130 ; R600-DAG: SETNE_INT
132 ; SI-DAG: v_cmp_u_f32_e32 vcc
133 ; SI-DAG: v_cmp_eq_f32_e64 [[CMP1:s\[[0-9]+:[0-9]+\]]]
134 ; SI: s_or_b64 [[OR:s\[[0-9]+:[0-9]+\]]], [[CMP1]], vcc
135 ; SI: v_cndmask_b32_e64 [[VRESULT:v[0-9]+]], 0, -1, [[OR]]
136 ; SI: buffer_store_dword [[VRESULT]]
137 define void @f32_ueq(i32 addrspace(1)* %out, float %a, float %b) {
139 %0 = fcmp ueq float %a, %b
140 %1 = sext i1 %0 to i32
141 store i32 %1, i32 addrspace(1)* %out
145 ; FUNC-LABEL: {{^}}f32_ugt:
148 ; SI: v_cmp_nle_f32_e32 vcc
149 ; SI-NEXT: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1, vcc
150 define void @f32_ugt(i32 addrspace(1)* %out, float %a, float %b) {
152 %0 = fcmp ugt float %a, %b
153 %1 = sext i1 %0 to i32
154 store i32 %1, i32 addrspace(1)* %out
158 ; FUNC-LABEL: {{^}}f32_uge:
162 ; SI: v_cmp_nlt_f32_e32 vcc
163 ; SI-NEXT: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1, vcc
164 define void @f32_uge(i32 addrspace(1)* %out, float %a, float %b) {
166 %0 = fcmp uge float %a, %b
167 %1 = sext i1 %0 to i32
168 store i32 %1, i32 addrspace(1)* %out
172 ; FUNC-LABEL: {{^}}f32_ult:
176 ; SI: v_cmp_nge_f32_e32 vcc
177 ; SI-NEXT: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1, vcc
178 define void @f32_ult(i32 addrspace(1)* %out, float %a, float %b) {
180 %0 = fcmp ult float %a, %b
181 %1 = sext i1 %0 to i32
182 store i32 %1, i32 addrspace(1)* %out
186 ; FUNC-LABEL: {{^}}f32_ule:
190 ; SI: v_cmp_ngt_f32_e32 vcc
191 ; SI-NEXT: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1, vcc
192 define void @f32_ule(i32 addrspace(1)* %out, float %a, float %b) {
194 %0 = fcmp ule float %a, %b
195 %1 = sext i1 %0 to i32
196 store i32 %1, i32 addrspace(1)* %out
200 ; FUNC-LABEL: {{^}}f32_une:
203 define void @f32_une(i32 addrspace(1)* %out, float %a, float %b) {
205 %0 = fcmp une float %a, %b
206 %1 = sext i1 %0 to i32
207 store i32 %1, i32 addrspace(1)* %out
211 ; FUNC-LABEL: {{^}}f32_uno:
217 define void @f32_uno(i32 addrspace(1)* %out, float %a, float %b) {
219 %0 = fcmp uno float %a, %b
220 %1 = sext i1 %0 to i32
221 store i32 %1, i32 addrspace(1)* %out
225 ;;;==========================================================================;;;
226 ;; 32-bit integer comparisons
227 ;;;==========================================================================;;;
229 ; FUNC-LABEL: {{^}}i32_eq:
232 define void @i32_eq(i32 addrspace(1)* %out, i32 %a, i32 %b) {
234 %0 = icmp eq i32 %a, %b
235 %1 = sext i1 %0 to i32
236 store i32 %1, i32 addrspace(1)* %out
240 ; FUNC-LABEL: {{^}}i32_ne:
243 define void @i32_ne(i32 addrspace(1)* %out, i32 %a, i32 %b) {
245 %0 = icmp ne i32 %a, %b
246 %1 = sext i1 %0 to i32
247 store i32 %1, i32 addrspace(1)* %out
251 ; FUNC-LABEL: {{^}}i32_ugt:
254 define void @i32_ugt(i32 addrspace(1)* %out, i32 %a, i32 %b) {
256 %0 = icmp ugt i32 %a, %b
257 %1 = sext i1 %0 to i32
258 store i32 %1, i32 addrspace(1)* %out
262 ; FUNC-LABEL: {{^}}i32_uge:
265 define void @i32_uge(i32 addrspace(1)* %out, i32 %a, i32 %b) {
267 %0 = icmp uge i32 %a, %b
268 %1 = sext i1 %0 to i32
269 store i32 %1, i32 addrspace(1)* %out
273 ; FUNC-LABEL: {{^}}i32_ult:
276 define void @i32_ult(i32 addrspace(1)* %out, i32 %a, i32 %b) {
278 %0 = icmp ult i32 %a, %b
279 %1 = sext i1 %0 to i32
280 store i32 %1, i32 addrspace(1)* %out
284 ; FUNC-LABEL: {{^}}i32_ule:
287 define void @i32_ule(i32 addrspace(1)* %out, i32 %a, i32 %b) {
289 %0 = icmp ule i32 %a, %b
290 %1 = sext i1 %0 to i32
291 store i32 %1, i32 addrspace(1)* %out
295 ; FUNC-LABEL: {{^}}i32_sgt:
298 define void @i32_sgt(i32 addrspace(1)* %out, i32 %a, i32 %b) {
300 %0 = icmp sgt i32 %a, %b
301 %1 = sext i1 %0 to i32
302 store i32 %1, i32 addrspace(1)* %out
306 ; FUNC-LABEL: {{^}}i32_sge:
309 define void @i32_sge(i32 addrspace(1)* %out, i32 %a, i32 %b) {
311 %0 = icmp sge i32 %a, %b
312 %1 = sext i1 %0 to i32
313 store i32 %1, i32 addrspace(1)* %out
317 ; FUNC-LABEL: {{^}}i32_slt:
320 define void @i32_slt(i32 addrspace(1)* %out, i32 %a, i32 %b) {
322 %0 = icmp slt i32 %a, %b
323 %1 = sext i1 %0 to i32
324 store i32 %1, i32 addrspace(1)* %out
328 ; FUNC-LABEL: {{^}}i32_sle:
331 define void @i32_sle(i32 addrspace(1)* %out, i32 %a, i32 %b) {
333 %0 = icmp sle i32 %a, %b
334 %1 = sext i1 %0 to i32
335 store i32 %1, i32 addrspace(1)* %out
339 ; FIXME: This does 4 compares
340 ; FUNC-LABEL: {{^}}v3i32_eq:
341 ; SI-DAG: v_cmp_eq_i32
342 ; SI-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
343 ; SI-DAG: v_cmp_eq_i32
344 ; SI-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
345 ; SI-DAG: v_cmp_eq_i32
346 ; SI-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
348 define void @v3i32_eq(<3 x i32> addrspace(1)* %out, <3 x i32> addrspace(1)* %ptra, <3 x i32> addrspace(1)* %ptrb) {
349 %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
350 %gep.a = getelementptr <3 x i32> addrspace(1)* %ptra, i32 %tid
351 %gep.b = getelementptr <3 x i32> addrspace(1)* %ptrb, i32 %tid
352 %gep.out = getelementptr <3 x i32> addrspace(1)* %out, i32 %tid
353 %a = load <3 x i32> addrspace(1)* %gep.a
354 %b = load <3 x i32> addrspace(1)* %gep.b
355 %cmp = icmp eq <3 x i32> %a, %b
356 %ext = sext <3 x i1> %cmp to <3 x i32>
357 store <3 x i32> %ext, <3 x i32> addrspace(1)* %gep.out
361 ; FUNC-LABEL: {{^}}v3i8_eq:
362 ; SI-DAG: v_cmp_eq_i32
363 ; SI-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
364 ; SI-DAG: v_cmp_eq_i32
365 ; SI-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
366 ; SI-DAG: v_cmp_eq_i32
367 ; SI-DAG: v_cndmask_b32_e64 {{v[0-9]+}}, 0, -1,
369 define void @v3i8_eq(<3 x i8> addrspace(1)* %out, <3 x i8> addrspace(1)* %ptra, <3 x i8> addrspace(1)* %ptrb) {
370 %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone
371 %gep.a = getelementptr <3 x i8> addrspace(1)* %ptra, i32 %tid
372 %gep.b = getelementptr <3 x i8> addrspace(1)* %ptrb, i32 %tid
373 %gep.out = getelementptr <3 x i8> addrspace(1)* %out, i32 %tid
374 %a = load <3 x i8> addrspace(1)* %gep.a
375 %b = load <3 x i8> addrspace(1)* %gep.b
376 %cmp = icmp eq <3 x i8> %a, %b
377 %ext = sext <3 x i1> %cmp to <3 x i8>
378 store <3 x i8> %ext, <3 x i8> addrspace(1)* %gep.out