1; This file checks support for comparing vector values with the fcmp
2; instruction.
3
4; RUN: %p2i -i %s --filetype=obj --disassemble -a -O2 | FileCheck %s
5; RUN: %p2i -i %s --filetype=obj --disassemble -a -Om1 | FileCheck %s
6
7; RUN: %if --need=target_MIPS32 --need=allow_dump \
8; RUN:   --command %p2i --filetype=asm --assemble --disassemble --target mips32\
9; RUN:   -i %s --args -O2 \
10; RUN:   | %if --need=target_MIPS32 --need=allow_dump \
11; RUN:   --command FileCheck --check-prefix MIPS32 %s
12
13; Check that sext elimination occurs when the result of the comparison
14; instruction is alrady sign extended.  Sign extension to 4 x i32 uses
15; the pslld instruction.
16define internal <4 x i32> @sextElimination(<4 x float> %a, <4 x float> %b) {
17entry:
18  %res.trunc = fcmp oeq <4 x float> %a, %b
19  %res = sext <4 x i1> %res.trunc to <4 x i32>
20  ret <4 x i32> %res
21; CHECK-LABEL: sextElimination
22; CHECK: cmpeqps
23; CHECK-NOT: pslld
24}
25; MIPS32-LABEL: sextElimination
26; MIPS32: c.eq.s
27; MIPS32: li [[R:.*]],1
28; MIPS32: movf [[R]],zero,$fcc0
29; MIPS32: c.eq.s
30; MIPS32: li [[R:.*]],1
31; MIPS32: movf [[R]],zero,$fcc0
32; MIPS32: c.eq.s
33; MIPS32: li [[R:.*]],1
34; MIPS32: movf [[R]],zero,$fcc0
35; MIPS32: c.eq.s
36; MIPS32: li [[R:.*]],1
37; MIPS32: movf [[R]],zero,$fcc0
38
39define internal <4 x i32> @fcmpFalseVector(<4 x float> %a, <4 x float> %b) {
40entry:
41  %res.trunc = fcmp false <4 x float> %a, %b
42  %res = sext <4 x i1> %res.trunc to <4 x i32>
43  ret <4 x i32> %res
44; CHECK-LABEL: fcmpFalseVector
45; CHECK: pxor
46}
47; MIPS32-LABEL: fcmpFalseVector
48; MIPS32: li v0,0
49; MIPS32: li v1,0
50; MIPS32: li a0,0
51; MIPS32: li a1,0
52
53define internal <4 x i32> @fcmpOeqVector(<4 x float> %a, <4 x float> %b) {
54entry:
55  %res.trunc = fcmp oeq <4 x float> %a, %b
56  %res = sext <4 x i1> %res.trunc to <4 x i32>
57  ret <4 x i32> %res
58; CHECK-LABEL: fcmpOeqVector
59; CHECK: cmpeqps
60}
61; MIPS32-LABEL: fcmpOeqVector
62; MIPS32: c.eq.s
63; MIPS32: li [[R:.*]],1
64; MIPS32: movf [[R]],zero,$fcc0
65; MIPS32: c.eq.s
66; MIPS32: li [[R:.*]],1
67; MIPS32: movf [[R]],zero,$fcc0
68; MIPS32: c.eq.s
69; MIPS32: li [[R:.*]],1
70; MIPS32: movf [[R]],zero,$fcc0
71; MIPS32: c.eq.s
72; MIPS32: li [[R:.*]],1
73; MIPS32: movf [[R]],zero,$fcc0
74
75define internal <4 x i32> @fcmpOgeVector(<4 x float> %a, <4 x float> %b) {
76entry:
77  %res.trunc = fcmp oge <4 x float> %a, %b
78  %res = sext <4 x i1> %res.trunc to <4 x i32>
79  ret <4 x i32> %res
80; CHECK-LABEL: fcmpOgeVector
81; CHECK: cmpleps
82}
83; MIPS32-LABEL: fcmpOgeVector
84; MIPS32: c.ult.s
85; MIPS32: li [[R:.*]],1
86; MIPS32: movt [[R]],zero,$fcc0
87; MIPS32: c.ult.s
88; MIPS32: li [[R:.*]],1
89; MIPS32: movt [[R]],zero,$fcc0
90; MIPS32: c.ult.s
91; MIPS32: li [[R:.*]],1
92; MIPS32: movt [[R]],zero,$fcc0
93; MIPS32: c.ult.s
94; MIPS32: li [[R:.*]],1
95; MIPS32: movt [[R]],zero,$fcc0
96
97define internal <4 x i32> @fcmpOgtVector(<4 x float> %a, <4 x float> %b) {
98entry:
99  %res.trunc = fcmp ogt <4 x float> %a, %b
100  %res = sext <4 x i1> %res.trunc to <4 x i32>
101  ret <4 x i32> %res
102; CHECK-LABEL: fcmpOgtVector
103; CHECK: cmpltps
104}
105; MIPS32-LABEL: fcmpOgtVector
106; MIPS32: c.ule.s
107; MIPS32: li [[R:.*]],1
108; MIPS32: movt [[R]],zero,$fcc0
109; MIPS32: c.ule.s
110; MIPS32: li [[R:.*]],1
111; MIPS32: movt [[R]],zero,$fcc0
112; MIPS32: c.ule.s
113; MIPS32: li [[R:.*]],1
114; MIPS32: movt [[R]],zero,$fcc0
115; MIPS32: c.ule.s
116; MIPS32: li [[R:.*]],1
117; MIPS32: movt [[R]],zero,$fcc0
118
119define internal <4 x i32> @fcmpOleVector(<4 x float> %a, <4 x float> %b) {
120entry:
121  %res.trunc = fcmp ole <4 x float> %a, %b
122  %res = sext <4 x i1> %res.trunc to <4 x i32>
123  ret <4 x i32> %res
124; CHECK-LABEL: fcmpOleVector
125; CHECK: cmpleps
126}
127; MIPS32-LABEL: fcmpOleVector
128; MIPS32: c.ole.s
129; MIPS32: li [[R:.*]],1
130; MIPS32: movf [[R]],zero,$fcc0
131; MIPS32: c.ole.s
132; MIPS32: li [[R:.*]],1
133; MIPS32: movf [[R]],zero,$fcc0
134; MIPS32: c.ole.s
135; MIPS32: li [[R:.*]],1
136; MIPS32: movf [[R]],zero,$fcc0
137; MIPS32: c.ole.s
138; MIPS32: li [[R:.*]],1
139; MIPS32: movf [[R]],zero,$fcc0
140
141define internal <4 x i32> @fcmpOltVector(<4 x float> %a, <4 x float> %b) {
142entry:
143  %res.trunc = fcmp olt <4 x float> %a, %b
144  %res = sext <4 x i1> %res.trunc to <4 x i32>
145  ret <4 x i32> %res
146; CHECK-LABEL: fcmpOltVector
147; CHECK: cmpltps
148}
149; MIPS32-LABEL: fcmpOltVector
150; MIPS32: c.olt.s
151; MIPS32: li [[R:.*]],1
152; MIPS32: movf [[R]],zero,$fcc0
153; MIPS32: c.olt.s
154; MIPS32: li [[R:.*]],1
155; MIPS32: movf [[R]],zero,$fcc0
156; MIPS32: c.olt.s
157; MIPS32: li [[R:.*]],1
158; MIPS32: movf [[R]],zero,$fcc0
159; MIPS32: c.olt.s
160; MIPS32: li [[R:.*]],1
161; MIPS32: movf [[R]],zero,$fcc0
162
163define internal <4 x i32> @fcmpOneVector(<4 x float> %a, <4 x float> %b) {
164entry:
165  %res.trunc = fcmp one <4 x float> %a, %b
166  %res = sext <4 x i1> %res.trunc to <4 x i32>
167  ret <4 x i32> %res
168; CHECK-LABEL: fcmpOneVector
169; CHECK: cmpneqps
170; CHECK: cmpordps
171; CHECK: pand
172}
173; MIPS32-LABEL: fcmpOneVector
174; MIPS32: c.ueq.s
175; MIPS32: li [[R:.*]],1
176; MIPS32: movt [[R]],zero,$fcc0
177; MIPS32: c.ueq.s
178; MIPS32: li [[R:.*]],1
179; MIPS32: movt [[R]],zero,$fcc0
180; MIPS32: c.ueq.s
181; MIPS32: li [[R:.*]],1
182; MIPS32: movt [[R]],zero,$fcc0
183; MIPS32: c.ueq.s
184; MIPS32: li [[R:.*]],1
185; MIPS32: movt [[R]],zero,$fcc0
186
187define internal <4 x i32> @fcmpOrdVector(<4 x float> %a, <4 x float> %b) {
188entry:
189  %res.trunc = fcmp ord <4 x float> %a, %b
190  %res = sext <4 x i1> %res.trunc to <4 x i32>
191  ret <4 x i32> %res
192; CHECK-LABEL: fcmpOrdVector
193; CHECK: cmpordps
194}
195; MIPS32-LABEL: fcmpOrdVector
196; MIPS32: c.un.s
197; MIPS32: li [[R:.*]],1
198; MIPS32: movt [[R]],zero,$fcc0
199; MIPS32: c.un.s
200; MIPS32: li [[R:.*]],1
201; MIPS32: movt [[R]],zero,$fcc0
202; MIPS32: c.un.s
203; MIPS32: li [[R:.*]],1
204; MIPS32: movt [[R]],zero,$fcc0
205; MIPS32: c.un.s
206; MIPS32: li [[R:.*]],1
207; MIPS32: movt [[R]],zero,$fcc0
208
209define internal <4 x i32> @fcmpTrueVector(<4 x float> %a, <4 x float> %b) {
210entry:
211  %res.trunc = fcmp true <4 x float> %a, %b
212  %res = sext <4 x i1> %res.trunc to <4 x i32>
213  ret <4 x i32> %res
214; CHECK-LABEL: fcmpTrueVector
215; CHECK: pcmpeqd
216}
217; MIPS32-LABEL: fcmpTrueVector
218; MIPS32: li v0,1
219; MIPS32: li v1,1
220; MIPS32: li a0,1
221; MIPS32: li a1,1
222
223define internal <4 x i32> @fcmpUeqVector(<4 x float> %a, <4 x float> %b) {
224entry:
225  %res.trunc = fcmp ueq <4 x float> %a, %b
226  %res = sext <4 x i1> %res.trunc to <4 x i32>
227  ret <4 x i32> %res
228; CHECK-LABEL: fcmpUeqVector
229; CHECK: cmpeqps
230; CHECK: cmpunordps
231; CHECK: por
232}
233; MIPS32-LABEL: fcmpUeqVector
234; MIPS32: c.ueq.s
235; MIPS32: li [[R:.*]],1
236; MIPS32: movf [[R]],zero,$fcc0
237; MIPS32: c.ueq.s
238; MIPS32: li [[R:.*]],1
239; MIPS32: movf [[R]],zero,$fcc0
240; MIPS32: c.ueq.s
241; MIPS32: li [[R:.*]],1
242; MIPS32: movf [[R]],zero,$fcc0
243; MIPS32: c.ueq.s
244; MIPS32: li [[R:.*]],1
245; MIPS32: movf [[R]],zero,$fcc0
246
247define internal <4 x i32> @fcmpUgeVector(<4 x float> %a, <4 x float> %b) {
248entry:
249  %res.trunc = fcmp uge <4 x float> %a, %b
250  %res = sext <4 x i1> %res.trunc to <4 x i32>
251  ret <4 x i32> %res
252; CHECK-LABEL: fcmpUgeVector
253; CHECK: cmpnltps
254}
255; MIPS32-LABEL: fcmpUgeVector
256; MIPS32: c.olt.s
257; MIPS32: li [[R:.*]],1
258; MIPS32: movt [[R]],zero,$fcc0
259; MIPS32: c.olt.s
260; MIPS32: li [[R:.*]],1
261; MIPS32: movt [[R]],zero,$fcc0
262; MIPS32: c.olt.s
263; MIPS32: li [[R:.*]],1
264; MIPS32: movt [[R]],zero,$fcc0
265; MIPS32: c.olt.s
266; MIPS32: li [[R:.*]],1
267; MIPS32: movt [[R]],zero,$fcc0
268
269define internal <4 x i32> @fcmpUgtVector(<4 x float> %a, <4 x float> %b) {
270entry:
271  %res.trunc = fcmp ugt <4 x float> %a, %b
272  %res = sext <4 x i1> %res.trunc to <4 x i32>
273  ret <4 x i32> %res
274; CHECK-LABEL: fcmpUgtVector
275; CHECK: cmpnleps
276}
277; MIPS32-LABEL: fcmpUgtVector
278; MIPS32: c.ole.s
279; MIPS32: li [[R:.*]],1
280; MIPS32: movt [[R]],zero,$fcc0
281; MIPS32: c.ole.s
282; MIPS32: li [[R:.*]],1
283; MIPS32: movt [[R]],zero,$fcc0
284; MIPS32: c.ole.s
285; MIPS32: li [[R:.*]],1
286; MIPS32: movt [[R]],zero,$fcc0
287; MIPS32: c.ole.s
288; MIPS32: li [[R:.*]],1
289; MIPS32: movt [[R]],zero,$fcc0
290
291define internal <4 x i32> @fcmpUleVector(<4 x float> %a, <4 x float> %b) {
292entry:
293  %res.trunc = fcmp ule <4 x float> %a, %b
294  %res = sext <4 x i1> %res.trunc to <4 x i32>
295  ret <4 x i32> %res
296; CHECK-LABEL: fcmpUleVector
297; CHECK: cmpnltps
298}
299; MIPS32-LABEL: fcmpUleVector
300; MIPS32: c.ule.s
301; MIPS32: li [[R:.*]],1
302; MIPS32: movf [[R]],zero,$fcc0
303; MIPS32: c.ule.s
304; MIPS32: li [[R:.*]],1
305; MIPS32: movf [[R]],zero,$fcc0
306; MIPS32: c.ule.s
307; MIPS32: li [[R:.*]],1
308; MIPS32: movf [[R]],zero,$fcc0
309; MIPS32: c.ule.s
310; MIPS32: li [[R:.*]],1
311; MIPS32: movf [[R]],zero,$fcc0
312
313define internal <4 x i32> @fcmpUltVector(<4 x float> %a, <4 x float> %b) {
314entry:
315  %res.trunc = fcmp ult <4 x float> %a, %b
316  %res = sext <4 x i1> %res.trunc to <4 x i32>
317  ret <4 x i32> %res
318; CHECK-LABEL: fcmpUltVector
319; CHECK: cmpnleps
320}
321; MIPS32-LABEL: fcmpUltVector
322; MIPS32: c.ult.s
323; MIPS32: li [[R:.*]],1
324; MIPS32: movf [[R]],zero,$fcc0
325; MIPS32: c.ult.s
326; MIPS32: li [[R:.*]],1
327; MIPS32: movf [[R]],zero,$fcc0
328; MIPS32: c.ult.s
329; MIPS32: li [[R:.*]],1
330; MIPS32: movf [[R]],zero,$fcc0
331; MIPS32: c.ult.s
332; MIPS32: li [[R:.*]],1
333; MIPS32: movf [[R]],zero,$fcc0
334
335define internal <4 x i32> @fcmpUneVector(<4 x float> %a, <4 x float> %b) {
336entry:
337  %res.trunc = fcmp une <4 x float> %a, %b
338  %res = sext <4 x i1> %res.trunc to <4 x i32>
339  ret <4 x i32> %res
340; CHECK-LABEL: fcmpUneVector
341; CHECK: cmpneqps
342}
343; MIPS32-LABEL: fcmpUneVector
344; MIPS32: c.eq.s
345; MIPS32: li [[R:.*]],1
346; MIPS32: movt [[R]],zero,$fcc0
347; MIPS32: c.eq.s
348; MIPS32: li [[R:.*]],1
349; MIPS32: movt [[R]],zero,$fcc0
350; MIPS32: c.eq.s
351; MIPS32: li [[R:.*]],1
352; MIPS32: movt [[R]],zero,$fcc0
353; MIPS32: c.eq.s
354; MIPS32: li [[R:.*]],1
355; MIPS32: movt [[R]],zero,$fcc0
356
357define internal <4 x i32> @fcmpUnoVector(<4 x float> %a, <4 x float> %b) {
358entry:
359  %res.trunc = fcmp uno <4 x float> %a, %b
360  %res = sext <4 x i1> %res.trunc to <4 x i32>
361  ret <4 x i32> %res
362; CHECK-LABEL: fcmpUnoVector
363; CHECK: cmpunordps
364}
365; MIPS32-LABEL: fcmpUnoVector
366; MIPS32: c.un.s
367; MIPS32: li [[R:.*]],1
368; MIPS32: movf [[R]],zero,$fcc0
369; MIPS32: c.un.s
370; MIPS32: li [[R:.*]],1
371; MIPS32: movf [[R]],zero,$fcc0
372; MIPS32: c.un.s
373; MIPS32: li [[R:.*]],1
374; MIPS32: movf [[R]],zero,$fcc0
375; MIPS32: c.un.s
376; MIPS32: li [[R:.*]],1
377; MIPS32: movf [[R]],zero,$fcc0
378