• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl  | FileCheck %s --check-prefix=ALL --check-prefix=KNL
3; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx  | FileCheck %s --check-prefix=ALL --check-prefix=SKX
4
5define <16 x float> @sitof32(<16 x i32> %a) nounwind {
6; ALL-LABEL: sitof32:
7; ALL:       ## BB#0:
8; ALL-NEXT:    vcvtdq2ps %zmm0, %zmm0
9; ALL-NEXT:    retq
10  %b = sitofp <16 x i32> %a to <16 x float>
11  ret <16 x float> %b
12}
13
14define <8 x double> @sltof864(<8 x i64> %a) {
15; KNL-LABEL: sltof864:
16; KNL:       ## BB#0:
17; KNL-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
18; KNL-NEXT:    vpextrq $1, %xmm1, %rax
19; KNL-NEXT:    vcvtsi2sdq %rax, %xmm0, %xmm2
20; KNL-NEXT:    vmovq %xmm1, %rax
21; KNL-NEXT:    vcvtsi2sdq %rax, %xmm0, %xmm1
22; KNL-NEXT:    vunpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
23; KNL-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
24; KNL-NEXT:    vpextrq $1, %xmm2, %rax
25; KNL-NEXT:    vcvtsi2sdq %rax, %xmm0, %xmm3
26; KNL-NEXT:    vmovq %xmm2, %rax
27; KNL-NEXT:    vcvtsi2sdq %rax, %xmm0, %xmm2
28; KNL-NEXT:    vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
29; KNL-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
30; KNL-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
31; KNL-NEXT:    vpextrq $1, %xmm2, %rax
32; KNL-NEXT:    vcvtsi2sdq %rax, %xmm0, %xmm3
33; KNL-NEXT:    vmovq %xmm2, %rax
34; KNL-NEXT:    vcvtsi2sdq %rax, %xmm0, %xmm2
35; KNL-NEXT:    vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
36; KNL-NEXT:    vpextrq $1, %xmm0, %rax
37; KNL-NEXT:    vcvtsi2sdq %rax, %xmm0, %xmm3
38; KNL-NEXT:    vmovq %xmm0, %rax
39; KNL-NEXT:    vcvtsi2sdq %rax, %xmm0, %xmm0
40; KNL-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm3[0]
41; KNL-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
42; KNL-NEXT:    vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
43; KNL-NEXT:    retq
44;
45; SKX-LABEL: sltof864:
46; SKX:       ## BB#0:
47; SKX-NEXT:    vcvtqq2pd %zmm0, %zmm0
48; SKX-NEXT:    retq
49  %b = sitofp <8 x i64> %a to <8 x double>
50  ret <8 x double> %b
51}
52
53define <4 x double> @sltof464(<4 x i64> %a) {
54; KNL-LABEL: sltof464:
55; KNL:       ## BB#0:
56; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm1
57; KNL-NEXT:    vpextrq $1, %xmm1, %rax
58; KNL-NEXT:    vcvtsi2sdq %rax, %xmm0, %xmm2
59; KNL-NEXT:    vmovq %xmm1, %rax
60; KNL-NEXT:    vcvtsi2sdq %rax, %xmm0, %xmm1
61; KNL-NEXT:    vunpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
62; KNL-NEXT:    vpextrq $1, %xmm0, %rax
63; KNL-NEXT:    vcvtsi2sdq %rax, %xmm0, %xmm2
64; KNL-NEXT:    vmovq %xmm0, %rax
65; KNL-NEXT:    vcvtsi2sdq %rax, %xmm0, %xmm0
66; KNL-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm2[0]
67; KNL-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
68; KNL-NEXT:    retq
69;
70; SKX-LABEL: sltof464:
71; SKX:       ## BB#0:
72; SKX-NEXT:    vcvtqq2pd %ymm0, %ymm0
73; SKX-NEXT:    retq
74  %b = sitofp <4 x i64> %a to <4 x double>
75  ret <4 x double> %b
76}
77
78define <2 x float> @sltof2f32(<2 x i64> %a) {
79; KNL-LABEL: sltof2f32:
80; KNL:       ## BB#0:
81; KNL-NEXT:    vpextrq $1, %xmm0, %rax
82; KNL-NEXT:    vcvtsi2ssq %rax, %xmm0, %xmm1
83; KNL-NEXT:    vmovq %xmm0, %rax
84; KNL-NEXT:    vcvtsi2ssq %rax, %xmm0, %xmm0
85; KNL-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[2,3]
86; KNL-NEXT:    vcvtsi2ssq %rax, %xmm0, %xmm1
87; KNL-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1],xmm1[0],xmm0[3]
88; KNL-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1,2],xmm1[0]
89; KNL-NEXT:    retq
90;
91; SKX-LABEL: sltof2f32:
92; SKX:       ## BB#0:
93; SKX-NEXT:    ## kill: %XMM0<def> %XMM0<kill> %YMM0<def>
94; SKX-NEXT:    vcvtqq2ps %ymm0, %xmm0
95; SKX-NEXT:    retq
96  %b = sitofp <2 x i64> %a to <2 x float>
97  ret <2 x float>%b
98}
99
100define <4 x float> @sltof4f32_mem(<4 x i64>* %a) {
101; KNL-LABEL: sltof4f32_mem:
102; KNL:       ## BB#0:
103; KNL-NEXT:    vmovdqu (%rdi), %ymm0
104; KNL-NEXT:    vpextrq $1, %xmm0, %rax
105; KNL-NEXT:    vcvtsi2ssq %rax, %xmm0, %xmm1
106; KNL-NEXT:    vmovq %xmm0, %rax
107; KNL-NEXT:    vcvtsi2ssq %rax, %xmm0, %xmm2
108; KNL-NEXT:    vinsertps {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[2,3]
109; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
110; KNL-NEXT:    vmovq %xmm0, %rax
111; KNL-NEXT:    vcvtsi2ssq %rax, %xmm0, %xmm2
112; KNL-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0],xmm1[3]
113; KNL-NEXT:    vpextrq $1, %xmm0, %rax
114; KNL-NEXT:    vcvtsi2ssq %rax, %xmm0, %xmm0
115; KNL-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
116; KNL-NEXT:    retq
117;
118; SKX-LABEL: sltof4f32_mem:
119; SKX:       ## BB#0:
120; SKX-NEXT:    vcvtqq2psy (%rdi), %xmm0
121; SKX-NEXT:    retq
122  %a1 = load <4 x i64>, <4 x i64>* %a, align 8
123  %b = sitofp <4 x i64> %a1 to <4 x float>
124  ret <4 x float>%b
125}
126
127define <4 x i64> @f64tosl(<4 x double> %a) {
128; KNL-LABEL: f64tosl:
129; KNL:       ## BB#0:
130; KNL-NEXT:    vextractf128 $1, %ymm0, %xmm1
131; KNL-NEXT:    vcvttsd2si %xmm1, %rax
132; KNL-NEXT:    vmovq %rax, %xmm2
133; KNL-NEXT:    vpermilpd {{.*#+}} xmm1 = xmm1[1,0]
134; KNL-NEXT:    vcvttsd2si %xmm1, %rax
135; KNL-NEXT:    vmovq %rax, %xmm1
136; KNL-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
137; KNL-NEXT:    vcvttsd2si %xmm0, %rax
138; KNL-NEXT:    vmovq %rax, %xmm2
139; KNL-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
140; KNL-NEXT:    vcvttsd2si %xmm0, %rax
141; KNL-NEXT:    vmovq %rax, %xmm0
142; KNL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm0[0]
143; KNL-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
144; KNL-NEXT:    retq
145;
146; SKX-LABEL: f64tosl:
147; SKX:       ## BB#0:
148; SKX-NEXT:    vcvttpd2qq %ymm0, %ymm0
149; SKX-NEXT:    retq
150  %b = fptosi <4 x double> %a to <4 x i64>
151  ret <4 x i64> %b
152}
153
154define <4 x i64> @f32tosl(<4 x float> %a) {
155; KNL-LABEL: f32tosl:
156; KNL:       ## BB#0:
157; KNL-NEXT:    vpermilps {{.*#+}} xmm1 = xmm0[3,1,2,3]
158; KNL-NEXT:    vcvttss2si %xmm1, %rax
159; KNL-NEXT:    vmovq %rax, %xmm1
160; KNL-NEXT:    vpermilpd {{.*#+}} xmm2 = xmm0[1,0]
161; KNL-NEXT:    vcvttss2si %xmm2, %rax
162; KNL-NEXT:    vmovq %rax, %xmm2
163; KNL-NEXT:    vpunpcklqdq {{.*#+}} xmm1 = xmm2[0],xmm1[0]
164; KNL-NEXT:    vcvttss2si %xmm0, %rax
165; KNL-NEXT:    vmovq %rax, %xmm2
166; KNL-NEXT:    vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
167; KNL-NEXT:    vcvttss2si %xmm0, %rax
168; KNL-NEXT:    vmovq %rax, %xmm0
169; KNL-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm2[0],xmm0[0]
170; KNL-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm0
171; KNL-NEXT:    retq
172;
173; SKX-LABEL: f32tosl:
174; SKX:       ## BB#0:
175; SKX-NEXT:    vcvttps2qq %xmm0, %ymm0
176; SKX-NEXT:    retq
177  %b = fptosi <4 x float> %a to <4 x i64>
178  ret <4 x i64> %b
179}
180
181define <4 x float> @sltof432(<4 x i64> %a) {
182; KNL-LABEL: sltof432:
183; KNL:       ## BB#0:
184; KNL-NEXT:    vpextrq $1, %xmm0, %rax
185; KNL-NEXT:    vcvtsi2ssq %rax, %xmm0, %xmm1
186; KNL-NEXT:    vmovq %xmm0, %rax
187; KNL-NEXT:    vcvtsi2ssq %rax, %xmm0, %xmm2
188; KNL-NEXT:    vinsertps {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[2,3]
189; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
190; KNL-NEXT:    vmovq %xmm0, %rax
191; KNL-NEXT:    vcvtsi2ssq %rax, %xmm0, %xmm2
192; KNL-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0],xmm1[3]
193; KNL-NEXT:    vpextrq $1, %xmm0, %rax
194; KNL-NEXT:    vcvtsi2ssq %rax, %xmm0, %xmm0
195; KNL-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
196; KNL-NEXT:    retq
197;
198; SKX-LABEL: sltof432:
199; SKX:       ## BB#0:
200; SKX-NEXT:    vcvtqq2ps %ymm0, %xmm0
201; SKX-NEXT:    retq
202  %b = sitofp <4 x i64> %a to <4 x float>
203  ret <4 x float> %b
204}
205
206define <4 x float> @ultof432(<4 x i64> %a) {
207; KNL-LABEL: ultof432:
208; KNL:       ## BB#0:
209; KNL-NEXT:    vpextrq $1, %xmm0, %rax
210; KNL-NEXT:    vcvtusi2ssq %rax, %xmm0, %xmm1
211; KNL-NEXT:    vmovq %xmm0, %rax
212; KNL-NEXT:    vcvtusi2ssq %rax, %xmm0, %xmm2
213; KNL-NEXT:    vinsertps {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[2,3]
214; KNL-NEXT:    vextracti128 $1, %ymm0, %xmm0
215; KNL-NEXT:    vmovq %xmm0, %rax
216; KNL-NEXT:    vcvtusi2ssq %rax, %xmm0, %xmm2
217; KNL-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,1],xmm2[0],xmm1[3]
218; KNL-NEXT:    vpextrq $1, %xmm0, %rax
219; KNL-NEXT:    vcvtusi2ssq %rax, %xmm0, %xmm0
220; KNL-NEXT:    vinsertps {{.*#+}} xmm0 = xmm1[0,1,2],xmm0[0]
221; KNL-NEXT:    retq
222;
223; SKX-LABEL: ultof432:
224; SKX:       ## BB#0:
225; SKX-NEXT:    vcvtuqq2ps %ymm0, %xmm0
226; SKX-NEXT:    retq
227  %b = uitofp <4 x i64> %a to <4 x float>
228  ret <4 x float> %b
229}
230
231define <8 x double> @ultof64(<8 x i64> %a) {
232; KNL-LABEL: ultof64:
233; KNL:       ## BB#0:
234; KNL-NEXT:    vextracti32x4 $3, %zmm0, %xmm1
235; KNL-NEXT:    vpextrq $1, %xmm1, %rax
236; KNL-NEXT:    vcvtusi2sdq %rax, %xmm0, %xmm2
237; KNL-NEXT:    vmovq %xmm1, %rax
238; KNL-NEXT:    vcvtusi2sdq %rax, %xmm0, %xmm1
239; KNL-NEXT:    vunpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
240; KNL-NEXT:    vextracti32x4 $2, %zmm0, %xmm2
241; KNL-NEXT:    vpextrq $1, %xmm2, %rax
242; KNL-NEXT:    vcvtusi2sdq %rax, %xmm0, %xmm3
243; KNL-NEXT:    vmovq %xmm2, %rax
244; KNL-NEXT:    vcvtusi2sdq %rax, %xmm0, %xmm2
245; KNL-NEXT:    vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
246; KNL-NEXT:    vinsertf128 $1, %xmm1, %ymm2, %ymm1
247; KNL-NEXT:    vextracti32x4 $1, %zmm0, %xmm2
248; KNL-NEXT:    vpextrq $1, %xmm2, %rax
249; KNL-NEXT:    vcvtusi2sdq %rax, %xmm0, %xmm3
250; KNL-NEXT:    vmovq %xmm2, %rax
251; KNL-NEXT:    vcvtusi2sdq %rax, %xmm0, %xmm2
252; KNL-NEXT:    vunpcklpd {{.*#+}} xmm2 = xmm2[0],xmm3[0]
253; KNL-NEXT:    vpextrq $1, %xmm0, %rax
254; KNL-NEXT:    vcvtusi2sdq %rax, %xmm0, %xmm3
255; KNL-NEXT:    vmovq %xmm0, %rax
256; KNL-NEXT:    vcvtusi2sdq %rax, %xmm0, %xmm0
257; KNL-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],xmm3[0]
258; KNL-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
259; KNL-NEXT:    vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
260; KNL-NEXT:    retq
261;
262; SKX-LABEL: ultof64:
263; SKX:       ## BB#0:
264; SKX-NEXT:    vcvtuqq2pd %zmm0, %zmm0
265; SKX-NEXT:    retq
266  %b = uitofp <8 x i64> %a to <8 x double>
267  ret <8 x double> %b
268}
269
270define <16 x i32> @fptosi00(<16 x float> %a) nounwind {
271; ALL-LABEL: fptosi00:
272; ALL:       ## BB#0:
273; ALL-NEXT:    vcvttps2dq %zmm0, %zmm0
274; ALL-NEXT:    retq
275  %b = fptosi <16 x float> %a to <16 x i32>
276  ret <16 x i32> %b
277}
278
279define <16 x i32> @fptoui00(<16 x float> %a) nounwind {
280; ALL-LABEL: fptoui00:
281; ALL:       ## BB#0:
282; ALL-NEXT:    vcvttps2udq %zmm0, %zmm0
283; ALL-NEXT:    retq
284  %b = fptoui <16 x float> %a to <16 x i32>
285  ret <16 x i32> %b
286}
287
288define <8 x i32> @fptoui_256(<8 x float> %a) nounwind {
289; KNL-LABEL: fptoui_256:
290; KNL:       ## BB#0:
291; KNL-NEXT:    ## kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
292; KNL-NEXT:    vcvttps2udq %zmm0, %zmm0
293; KNL-NEXT:    ## kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
294; KNL-NEXT:    retq
295;
296; SKX-LABEL: fptoui_256:
297; SKX:       ## BB#0:
298; SKX-NEXT:    vcvttps2udq %ymm0, %ymm0
299; SKX-NEXT:    retq
300  %b = fptoui <8 x float> %a to <8 x i32>
301  ret <8 x i32> %b
302}
303
304define <4 x i32> @fptoui_128(<4 x float> %a) nounwind {
305; KNL-LABEL: fptoui_128:
306; KNL:       ## BB#0:
307; KNL-NEXT:    ## kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
308; KNL-NEXT:    vcvttps2udq %zmm0, %zmm0
309; KNL-NEXT:    ## kill: %XMM0<def> %XMM0<kill> %ZMM0<kill>
310; KNL-NEXT:    retq
311;
312; SKX-LABEL: fptoui_128:
313; SKX:       ## BB#0:
314; SKX-NEXT:    vcvttps2udq %xmm0, %xmm0
315; SKX-NEXT:    retq
316  %b = fptoui <4 x float> %a to <4 x i32>
317  ret <4 x i32> %b
318}
319
320define <8 x i32> @fptoui01(<8 x double> %a) nounwind {
321; ALL-LABEL: fptoui01:
322; ALL:       ## BB#0:
323; ALL-NEXT:    vcvttpd2udq %zmm0, %ymm0
324; ALL-NEXT:    retq
325  %b = fptoui <8 x double> %a to <8 x i32>
326  ret <8 x i32> %b
327}
328
329define <4 x i32> @fptoui_256d(<4 x double> %a) nounwind {
330; KNL-LABEL: fptoui_256d:
331; KNL:       ## BB#0:
332; KNL-NEXT:    ## kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
333; KNL-NEXT:    vcvttpd2udq %zmm0, %ymm0
334; KNL-NEXT:    ## kill: %XMM0<def> %XMM0<kill> %YMM0<kill>
335; KNL-NEXT:    retq
336;
337; SKX-LABEL: fptoui_256d:
338; SKX:       ## BB#0:
339; SKX-NEXT:    vcvttpd2udq %ymm0, %xmm0
340; SKX-NEXT:    retq
341  %b = fptoui <4 x double> %a to <4 x i32>
342  ret <4 x i32> %b
343}
344
345define <8 x double> @sitof64(<8 x i32> %a) {
346; ALL-LABEL: sitof64:
347; ALL:       ## BB#0:
348; ALL-NEXT:    vcvtdq2pd %ymm0, %zmm0
349; ALL-NEXT:    retq
350  %b = sitofp <8 x i32> %a to <8 x double>
351  ret <8 x double> %b
352}
353
354define <8 x i32> @fptosi01(<8 x double> %a) {
355; ALL-LABEL: fptosi01:
356; ALL:       ## BB#0:
357; ALL-NEXT:    vcvttpd2dq %zmm0, %ymm0
358; ALL-NEXT:    retq
359  %b = fptosi <8 x double> %a to <8 x i32>
360  ret <8 x i32> %b
361}
362
363define <4 x i32> @fptosi03(<4 x double> %a) {
364; KNL-LABEL: fptosi03:
365; KNL:       ## BB#0:
366; KNL-NEXT:    vcvttpd2dqy %ymm0, %xmm0
367; KNL-NEXT:    retq
368;
369; SKX-LABEL: fptosi03:
370; SKX:       ## BB#0:
371; SKX-NEXT:    vcvttpd2dq %ymm0, %xmm0
372; SKX-NEXT:    retq
373  %b = fptosi <4 x double> %a to <4 x i32>
374  ret <4 x i32> %b
375}
376
377define <16 x float> @fptrunc00(<16 x double> %b) nounwind {
378; KNL-LABEL: fptrunc00:
379; KNL:       ## BB#0:
380; KNL-NEXT:    vcvtpd2ps %zmm0, %ymm0
381; KNL-NEXT:    vcvtpd2ps %zmm1, %ymm1
382; KNL-NEXT:    vinsertf64x4 $1, %ymm1, %zmm0, %zmm0
383; KNL-NEXT:    retq
384;
385; SKX-LABEL: fptrunc00:
386; SKX:       ## BB#0:
387; SKX-NEXT:    vcvtpd2ps %zmm0, %ymm0
388; SKX-NEXT:    vcvtpd2ps %zmm1, %ymm1
389; SKX-NEXT:    vinsertf32x8 $1, %ymm1, %zmm0, %zmm0
390; SKX-NEXT:    retq
391  %a = fptrunc <16 x double> %b to <16 x float>
392  ret <16 x float> %a
393}
394
395define <4 x float> @fptrunc01(<4 x double> %b) {
396; KNL-LABEL: fptrunc01:
397; KNL:       ## BB#0:
398; KNL-NEXT:    vcvtpd2psy %ymm0, %xmm0
399; KNL-NEXT:    retq
400;
401; SKX-LABEL: fptrunc01:
402; SKX:       ## BB#0:
403; SKX-NEXT:    vcvtpd2ps %ymm0, %xmm0
404; SKX-NEXT:    retq
405  %a = fptrunc <4 x double> %b to <4 x float>
406  ret <4 x float> %a
407}
408
409define <4 x float> @fptrunc02(<4 x double> %b, <4 x i1> %mask) {
410; KNL-LABEL: fptrunc02:
411; KNL:       ## BB#0:
412; KNL-NEXT:    vpslld $31, %xmm1, %xmm1
413; KNL-NEXT:    vcvtpd2psy %ymm0, %xmm0
414; KNL-NEXT:    vxorps %xmm2, %xmm2, %xmm2
415; KNL-NEXT:    vblendvps %xmm1, %xmm0, %xmm2, %xmm0
416; KNL-NEXT:    retq
417;
418; SKX-LABEL: fptrunc02:
419; SKX:       ## BB#0:
420; SKX-NEXT:    vpslld $31, %xmm1, %xmm1
421; SKX-NEXT:    vptestmd %xmm1, %xmm1, %k1
422; SKX-NEXT:    vcvtpd2ps %ymm0, %xmm0 {%k1} {z}
423; SKX-NEXT:    retq
424  %a = fptrunc <4 x double> %b to <4 x float>
425  %c = select <4 x i1>%mask, <4 x float>%a, <4 x float> zeroinitializer
426  ret <4 x float> %c
427}
428
429define <8 x double> @fpext00(<8 x float> %b) nounwind {
430; ALL-LABEL: fpext00:
431; ALL:       ## BB#0:
432; ALL-NEXT:    vcvtps2pd %ymm0, %zmm0
433; ALL-NEXT:    retq
434  %a = fpext <8 x float> %b to <8 x double>
435  ret <8 x double> %a
436}
437
438define <4 x double> @fpext01(<4 x float> %b, <4 x double>%b1, <4 x double>%a1) {
439; KNL-LABEL: fpext01:
440; KNL:       ## BB#0:
441; KNL-NEXT:    vcvtps2pd %xmm0, %ymm0
442; KNL-NEXT:    vcmpltpd %ymm2, %ymm1, %ymm1
443; KNL-NEXT:    vxorpd %ymm2, %ymm2, %ymm2
444; KNL-NEXT:    vblendvpd %ymm1, %ymm0, %ymm2, %ymm0
445; KNL-NEXT:    retq
446;
447; SKX-LABEL: fpext01:
448; SKX:       ## BB#0:
449; SKX-NEXT:    vcmpltpd %ymm2, %ymm1, %k1
450; SKX-NEXT:    vcvtps2pd %xmm0, %ymm0 {%k1} {z}
451; SKX-NEXT:    retq
452  %a = fpext <4 x float> %b to <4 x double>
453  %mask = fcmp ogt <4 x double>%a1, %b1
454  %c = select <4 x i1>%mask,  <4 x double>%a, <4 x double>zeroinitializer
455  ret <4 x double> %c
456}
457
458define double @funcA(i64* nocapture %e) {
459; ALL-LABEL: funcA:
460; ALL:       ## BB#0: ## %entry
461; ALL-NEXT:    vcvtsi2sdq (%rdi), %xmm0, %xmm0
462; ALL-NEXT:    retq
463entry:
464  %tmp1 = load i64, i64* %e, align 8
465  %conv = sitofp i64 %tmp1 to double
466  ret double %conv
467}
468
469define double @funcB(i32* %e) {
470; ALL-LABEL: funcB:
471; ALL:       ## BB#0: ## %entry
472; ALL-NEXT:    vcvtsi2sdl (%rdi), %xmm0, %xmm0
473; ALL-NEXT:    retq
474entry:
475  %tmp1 = load i32, i32* %e, align 4
476  %conv = sitofp i32 %tmp1 to double
477  ret double %conv
478}
479
480define float @funcC(i32* %e) {
481; ALL-LABEL: funcC:
482; ALL:       ## BB#0: ## %entry
483; ALL-NEXT:    vcvtsi2ssl (%rdi), %xmm0, %xmm0
484; ALL-NEXT:    retq
485entry:
486  %tmp1 = load i32, i32* %e, align 4
487  %conv = sitofp i32 %tmp1 to float
488  ret float %conv
489}
490
491define float @i64tof32(i64* %e) {
492; ALL-LABEL: i64tof32:
493; ALL:       ## BB#0: ## %entry
494; ALL-NEXT:    vcvtsi2ssq (%rdi), %xmm0, %xmm0
495; ALL-NEXT:    retq
496entry:
497  %tmp1 = load i64, i64* %e, align 8
498  %conv = sitofp i64 %tmp1 to float
499  ret float %conv
500}
501
502define void @fpext() {
503; ALL-LABEL: fpext:
504; ALL:       ## BB#0: ## %entry
505; ALL-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
506; ALL-NEXT:    vcvtss2sd %xmm0, %xmm0, %xmm0
507; ALL-NEXT:    vmovsd %xmm0, -{{[0-9]+}}(%rsp)
508; ALL-NEXT:    retq
509entry:
510  %f = alloca float, align 4
511  %d = alloca double, align 8
512  %tmp = load float, float* %f, align 4
513  %conv = fpext float %tmp to double
514  store double %conv, double* %d, align 8
515  ret void
516}
517
518define void @fpround_scalar() nounwind uwtable {
519; ALL-LABEL: fpround_scalar:
520; ALL:       ## BB#0: ## %entry
521; ALL-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
522; ALL-NEXT:    vcvtsd2ss %xmm0, %xmm0, %xmm0
523; ALL-NEXT:    vmovss %xmm0, -{{[0-9]+}}(%rsp)
524; ALL-NEXT:    retq
525entry:
526  %f = alloca float, align 4
527  %d = alloca double, align 8
528  %tmp = load double, double* %d, align 8
529  %conv = fptrunc double %tmp to float
530  store float %conv, float* %f, align 4
531  ret void
532}
533
534define double @long_to_double(i64 %x) {
535; ALL-LABEL: long_to_double:
536; ALL:       ## BB#0:
537; ALL-NEXT:    vmovq %rdi, %xmm0
538; ALL-NEXT:    retq
539   %res = bitcast i64 %x to double
540   ret double %res
541}
542
543define i64 @double_to_long(double %x) {
544; ALL-LABEL: double_to_long:
545; ALL:       ## BB#0:
546; ALL-NEXT:    vmovq %xmm0, %rax
547; ALL-NEXT:    retq
548   %res = bitcast double %x to i64
549   ret i64 %res
550}
551
552define float @int_to_float(i32 %x) {
553; ALL-LABEL: int_to_float:
554; ALL:       ## BB#0:
555; ALL-NEXT:    vmovd %edi, %xmm0
556; ALL-NEXT:    retq
557   %res = bitcast i32 %x to float
558   ret float %res
559}
560
561define i32 @float_to_int(float %x) {
562; ALL-LABEL: float_to_int:
563; ALL:       ## BB#0:
564; ALL-NEXT:    vmovd %xmm0, %eax
565; ALL-NEXT:    retq
566   %res = bitcast float %x to i32
567   ret i32 %res
568}
569
570define <16 x double> @uitof64(<16 x i32> %a) nounwind {
571; KNL-LABEL: uitof64:
572; KNL:       ## BB#0:
573; KNL-NEXT:    vcvtudq2pd %ymm0, %zmm2
574; KNL-NEXT:    vextracti64x4 $1, %zmm0, %ymm0
575; KNL-NEXT:    vcvtudq2pd %ymm0, %zmm1
576; KNL-NEXT:    vmovaps %zmm2, %zmm0
577; KNL-NEXT:    retq
578;
579; SKX-LABEL: uitof64:
580; SKX:       ## BB#0:
581; SKX-NEXT:    vcvtudq2pd %ymm0, %zmm2
582; SKX-NEXT:    vextracti32x8 $1, %zmm0, %ymm0
583; SKX-NEXT:    vcvtudq2pd %ymm0, %zmm1
584; SKX-NEXT:    vmovaps %zmm2, %zmm0
585; SKX-NEXT:    retq
586  %b = uitofp <16 x i32> %a to <16 x double>
587  ret <16 x double> %b
588}
589
590define <4 x double> @uitof64_256(<4 x i32> %a) nounwind {
591; KNL-LABEL: uitof64_256:
592; KNL:       ## BB#0:
593; KNL-NEXT:    ## kill: %XMM0<def> %XMM0<kill> %YMM0<def>
594; KNL-NEXT:    vcvtudq2pd %ymm0, %zmm0
595; KNL-NEXT:    ## kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
596; KNL-NEXT:    retq
597;
598; SKX-LABEL: uitof64_256:
599; SKX:       ## BB#0:
600; SKX-NEXT:    vcvtudq2pd %xmm0, %ymm0
601; SKX-NEXT:    retq
602  %b = uitofp <4 x i32> %a to <4 x double>
603  ret <4 x double> %b
604}
605
606define <16 x float> @uitof32(<16 x i32> %a) nounwind {
607; ALL-LABEL: uitof32:
608; ALL:       ## BB#0:
609; ALL-NEXT:    vcvtudq2ps %zmm0, %zmm0
610; ALL-NEXT:    retq
611  %b = uitofp <16 x i32> %a to <16 x float>
612  ret <16 x float> %b
613}
614
615define <8 x float> @uitof32_256(<8 x i32> %a) nounwind {
616; KNL-LABEL: uitof32_256:
617; KNL:       ## BB#0:
618; KNL-NEXT:    ## kill: %YMM0<def> %YMM0<kill> %ZMM0<def>
619; KNL-NEXT:    vcvtudq2ps %zmm0, %zmm0
620; KNL-NEXT:    ## kill: %YMM0<def> %YMM0<kill> %ZMM0<kill>
621; KNL-NEXT:    retq
622;
623; SKX-LABEL: uitof32_256:
624; SKX:       ## BB#0:
625; SKX-NEXT:    vcvtudq2ps %ymm0, %ymm0
626; SKX-NEXT:    retq
627  %b = uitofp <8 x i32> %a to <8 x float>
628  ret <8 x float> %b
629}
630
631define <4 x float> @uitof32_128(<4 x i32> %a) nounwind {
632; KNL-LABEL: uitof32_128:
633; KNL:       ## BB#0:
634; KNL-NEXT:    ## kill: %XMM0<def> %XMM0<kill> %ZMM0<def>
635; KNL-NEXT:    vcvtudq2ps %zmm0, %zmm0
636; KNL-NEXT:    ## kill: %XMM0<def> %XMM0<kill> %ZMM0<kill>
637; KNL-NEXT:    retq
638;
639; SKX-LABEL: uitof32_128:
640; SKX:       ## BB#0:
641; SKX-NEXT:    vcvtudq2ps %xmm0, %xmm0
642; SKX-NEXT:    retq
643  %b = uitofp <4 x i32> %a to <4 x float>
644  ret <4 x float> %b
645}
646
647define i32 @fptosi02(float %a) nounwind {
648; ALL-LABEL: fptosi02:
649; ALL:       ## BB#0:
650; ALL-NEXT:    vcvttss2si %xmm0, %eax
651; ALL-NEXT:    retq
652  %b = fptosi float %a to i32
653  ret i32 %b
654}
655
656define i32 @fptoui02(float %a) nounwind {
657; ALL-LABEL: fptoui02:
658; ALL:       ## BB#0:
659; ALL-NEXT:    vcvttss2usi %xmm0, %eax
660; ALL-NEXT:    retq
661  %b = fptoui float %a to i32
662  ret i32 %b
663}
664
665define float @uitofp02(i32 %a) nounwind {
666; ALL-LABEL: uitofp02:
667; ALL:       ## BB#0:
668; ALL-NEXT:    vcvtusi2ssl %edi, %xmm0, %xmm0
669; ALL-NEXT:    retq
670  %b = uitofp i32 %a to float
671  ret float %b
672}
673
674define double @uitofp03(i32 %a) nounwind {
675; ALL-LABEL: uitofp03:
676; ALL:       ## BB#0:
677; ALL-NEXT:    vcvtusi2sdl %edi, %xmm0, %xmm0
678; ALL-NEXT:    retq
679  %b = uitofp i32 %a to double
680  ret double %b
681}
682
683define <16 x float> @sitofp_16i1_float(<16 x i32> %a) {
684; KNL-LABEL: sitofp_16i1_float:
685; KNL:       ## BB#0:
686; KNL-NEXT:    vpxord %zmm1, %zmm1, %zmm1
687; KNL-NEXT:    vpcmpgtd %zmm0, %zmm1, %k1
688; KNL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0
689; KNL-NEXT:    vmovdqa32 %zmm0, %zmm0 {%k1} {z}
690; KNL-NEXT:    vcvtdq2ps %zmm0, %zmm0
691; KNL-NEXT:    retq
692;
693; SKX-LABEL: sitofp_16i1_float:
694; SKX:       ## BB#0:
695; SKX-NEXT:    vpxord %zmm1, %zmm1, %zmm1
696; SKX-NEXT:    vpcmpgtd %zmm0, %zmm1, %k0
697; SKX-NEXT:    vpmovm2d %k0, %zmm0
698; SKX-NEXT:    vcvtdq2ps %zmm0, %zmm0
699; SKX-NEXT:    retq
700  %mask = icmp slt <16 x i32> %a, zeroinitializer
701  %1 = sitofp <16 x i1> %mask to <16 x float>
702  ret <16 x float> %1
703}
704
705define <16 x float> @sitofp_16i8_float(<16 x i8> %a) {
706; ALL-LABEL: sitofp_16i8_float:
707; ALL:       ## BB#0:
708; ALL-NEXT:    vpmovsxbd %xmm0, %zmm0
709; ALL-NEXT:    vcvtdq2ps %zmm0, %zmm0
710; ALL-NEXT:    retq
711  %1 = sitofp <16 x i8> %a to <16 x float>
712  ret <16 x float> %1
713}
714
715define <16 x float> @sitofp_16i16_float(<16 x i16> %a) {
716; ALL-LABEL: sitofp_16i16_float:
717; ALL:       ## BB#0:
718; ALL-NEXT:    vpmovsxwd %ymm0, %zmm0
719; ALL-NEXT:    vcvtdq2ps %zmm0, %zmm0
720; ALL-NEXT:    retq
721  %1 = sitofp <16 x i16> %a to <16 x float>
722  ret <16 x float> %1
723}
724
725define <8 x double> @sitofp_8i16_double(<8 x i16> %a) {
726; ALL-LABEL: sitofp_8i16_double:
727; ALL:       ## BB#0:
728; ALL-NEXT:    vpmovsxwd %xmm0, %ymm0
729; ALL-NEXT:    vcvtdq2pd %ymm0, %zmm0
730; ALL-NEXT:    retq
731  %1 = sitofp <8 x i16> %a to <8 x double>
732  ret <8 x double> %1
733}
734
735define <8 x double> @sitofp_8i8_double(<8 x i8> %a) {
736; ALL-LABEL: sitofp_8i8_double:
737; ALL:       ## BB#0:
738; ALL-NEXT:    vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
739; ALL-NEXT:    vpslld $24, %ymm0, %ymm0
740; ALL-NEXT:    vpsrad $24, %ymm0, %ymm0
741; ALL-NEXT:    vcvtdq2pd %ymm0, %zmm0
742; ALL-NEXT:    retq
743  %1 = sitofp <8 x i8> %a to <8 x double>
744  ret <8 x double> %1
745}
746
747define <8 x double> @sitofp_8i1_double(<8 x double> %a) {
748; KNL-LABEL: sitofp_8i1_double:
749; KNL:       ## BB#0:
750; KNL-NEXT:    vpxord %zmm1, %zmm1, %zmm1
751; KNL-NEXT:    vcmpltpd %zmm0, %zmm1, %k1
752; KNL-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0
753; KNL-NEXT:    vmovdqa64 %zmm0, %zmm0 {%k1} {z}
754; KNL-NEXT:    vpmovqd %zmm0, %ymm0
755; KNL-NEXT:    vcvtdq2pd %ymm0, %zmm0
756; KNL-NEXT:    retq
757;
758; SKX-LABEL: sitofp_8i1_double:
759; SKX:       ## BB#0:
760; SKX-NEXT:    vpxord %zmm1, %zmm1, %zmm1
761; SKX-NEXT:    vcmpltpd %zmm0, %zmm1, %k0
762; SKX-NEXT:    vpmovm2d %k0, %ymm0
763; SKX-NEXT:    vcvtdq2pd %ymm0, %zmm0
764; SKX-NEXT:    retq
765  %cmpres = fcmp ogt <8 x double> %a, zeroinitializer
766  %1 = sitofp <8 x i1> %cmpres to <8 x double>
767  ret <8 x double> %1
768}
769
770define <16 x float> @uitofp_16i8(<16 x i8>%a) {
771; ALL-LABEL: uitofp_16i8:
772; ALL:       ## BB#0:
773; ALL-NEXT:    vpmovzxbd {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero,xmm0[8],zero,zero,zero,xmm0[9],zero,zero,zero,xmm0[10],zero,zero,zero,xmm0[11],zero,zero,zero,xmm0[12],zero,zero,zero,xmm0[13],zero,zero,zero,xmm0[14],zero,zero,zero,xmm0[15],zero,zero,zero
774; ALL-NEXT:    vcvtudq2ps %zmm0, %zmm0
775; ALL-NEXT:    retq
776  %b = uitofp <16 x i8> %a to <16 x float>
777  ret <16 x float>%b
778}
779
780define <16 x float> @uitofp_16i16(<16 x i16>%a) {
781; ALL-LABEL: uitofp_16i16:
782; ALL:       ## BB#0:
783; ALL-NEXT:    vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero
784; ALL-NEXT:    vcvtudq2ps %zmm0, %zmm0
785; ALL-NEXT:    retq
786  %b = uitofp <16 x i16> %a to <16 x float>
787  ret <16 x float>%b
788}
789
790