blob: 53d7235a3d136f8adb8e296226002f0cbc3850c0 [file] [log] [blame]
Chandler Carruthc1bb0e82014-10-02 06:52:19 +00001; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+sse2 | FileCheck %s --check-prefix=SSE --check-prefix=SSE2
2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+ssse3 | FileCheck %s --check-prefix=SSE --check-prefix=SSSE3
3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+sse4.1 | FileCheck %s --check-prefix=SSE --check-prefix=SSE41
4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+avx | FileCheck %s --check-prefix=AVX --check-prefix=AVX1
5; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2
Elena Demikhovsky0b0c5d82012-01-24 13:54:13 +00006
7define <8 x i32> @zext_8i16_to_8i32(<8 x i16> %A) nounwind uwtable readnone ssp {
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +00008; SSE2-LABEL: zext_8i16_to_8i32:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +00009; SSE2: # BB#0: # %entry
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +000010; SSE2-NEXT: movdqa %xmm0, %xmm2
11; SSE2-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
12; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [65535,65535,65535,65535]
13; SSE2-NEXT: pand %xmm1, %xmm2
14; SSE2-NEXT: punpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]
15; SSE2-NEXT: pand %xmm0, %xmm1
16; SSE2-NEXT: movdqa %xmm2, %xmm0
17; SSE2-NEXT: retq
18;
19; SSSE3-LABEL: zext_8i16_to_8i32:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +000020; SSSE3: # BB#0: # %entry
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +000021; SSSE3-NEXT: movdqa %xmm0, %xmm2
22; SSSE3-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
23; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [65535,65535,65535,65535]
24; SSSE3-NEXT: pand %xmm1, %xmm2
25; SSSE3-NEXT: punpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]
26; SSSE3-NEXT: pand %xmm0, %xmm1
27; SSSE3-NEXT: movdqa %xmm2, %xmm0
28; SSSE3-NEXT: retq
29;
30; SSE41-LABEL: zext_8i16_to_8i32:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +000031; SSE41: # BB#0: # %entry
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +000032; SSE41-NEXT: pmovzxwd %xmm0, %xmm2
33; SSE41-NEXT: movdqa {{.*#+}} xmm1 = [65535,65535,65535,65535]
34; SSE41-NEXT: pand %xmm1, %xmm2
35; SSE41-NEXT: punpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]
36; SSE41-NEXT: pand %xmm0, %xmm1
37; SSE41-NEXT: movdqa %xmm2, %xmm0
38; SSE41-NEXT: retq
39;
40; AVX1-LABEL: zext_8i16_to_8i32:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +000041; AVX1: # BB#0: # %entry
Chandler Carruth99627bf2014-10-04 03:52:55 +000042; AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
43; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
44; AVX1-NEXT: vpunpcklwd {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1],xmm1[2],xmm2[2],xmm1[3],xmm2[3]
45; AVX1-NEXT: vpmovzxwd %xmm0, %xmm0
46; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +000047; AVX1-NEXT: retq
48;
49; AVX2-LABEL: zext_8i16_to_8i32:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +000050; AVX2: # BB#0: # %entry
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +000051; AVX2-NEXT: vpmovzxwd %xmm0, %ymm0
52; AVX2-NEXT: retq
Chandler Carruth011088f2014-10-01 20:27:16 +000053entry:
Elena Demikhovsky0b0c5d82012-01-24 13:54:13 +000054 %B = zext <8 x i16> %A to <8 x i32>
55 ret <8 x i32>%B
56}
57
58define <4 x i64> @zext_4i32_to_4i64(<4 x i32> %A) nounwind uwtable readnone ssp {
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +000059; SSE2-LABEL: zext_4i32_to_4i64:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +000060; SSE2: # BB#0: # %entry
Chandler Carruth99627bf2014-10-04 03:52:55 +000061; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[0,1,1,3]
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +000062; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [4294967295,4294967295]
63; SSE2-NEXT: pand %xmm3, %xmm2
Chandler Carruth99627bf2014-10-04 03:52:55 +000064; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,2,3,3]
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +000065; SSE2-NEXT: pand %xmm3, %xmm1
66; SSE2-NEXT: movdqa %xmm2, %xmm0
67; SSE2-NEXT: retq
68;
69; SSSE3-LABEL: zext_4i32_to_4i64:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +000070; SSSE3: # BB#0: # %entry
Chandler Carruth99627bf2014-10-04 03:52:55 +000071; SSSE3-NEXT: pshufd {{.*#+}} xmm2 = xmm0[0,1,1,3]
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +000072; SSSE3-NEXT: movdqa {{.*#+}} xmm3 = [4294967295,4294967295]
73; SSSE3-NEXT: pand %xmm3, %xmm2
Chandler Carruth99627bf2014-10-04 03:52:55 +000074; SSSE3-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,2,3,3]
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +000075; SSSE3-NEXT: pand %xmm3, %xmm1
76; SSSE3-NEXT: movdqa %xmm2, %xmm0
77; SSSE3-NEXT: retq
78;
79; SSE41-LABEL: zext_4i32_to_4i64:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +000080; SSE41: # BB#0: # %entry
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +000081; SSE41-NEXT: pmovzxdq %xmm0, %xmm2
82; SSE41-NEXT: movdqa {{.*#+}} xmm3 = [4294967295,4294967295]
83; SSE41-NEXT: pand %xmm3, %xmm2
Chandler Carruth99627bf2014-10-04 03:52:55 +000084; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm0[2,2,3,3]
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +000085; SSE41-NEXT: pand %xmm3, %xmm1
86; SSE41-NEXT: movdqa %xmm2, %xmm0
87; SSE41-NEXT: retq
88;
89; AVX1-LABEL: zext_4i32_to_4i64:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +000090; AVX1: # BB#0: # %entry
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +000091; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
Chandler Carruth99627bf2014-10-04 03:52:55 +000092; AVX1-NEXT: vpunpckhdq {{.*#+}} xmm1 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
93; AVX1-NEXT: vpmovzxdq %xmm0, %xmm0
94; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +000095; AVX1-NEXT: retq
96;
97; AVX2-LABEL: zext_4i32_to_4i64:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +000098; AVX2: # BB#0: # %entry
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +000099; AVX2-NEXT: vpmovzxdq %xmm0, %ymm0
100; AVX2-NEXT: retq
Chandler Carruth011088f2014-10-01 20:27:16 +0000101entry:
Elena Demikhovsky0b0c5d82012-01-24 13:54:13 +0000102 %B = zext <4 x i32> %A to <4 x i64>
103 ret <4 x i64>%B
104}
Nadav Rotem34ca89a2012-02-12 15:05:31 +0000105
Nadav Rotem34ca89a2012-02-12 15:05:31 +0000106define <8 x i32> @zext_8i8_to_8i32(<8 x i8> %z) {
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +0000107; SSE2-LABEL: zext_8i8_to_8i32:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +0000108; SSE2: # BB#0: # %entry
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +0000109; SSE2-NEXT: movdqa %xmm0, %xmm2
110; SSE2-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
111; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [255,255,255,255]
112; SSE2-NEXT: pand %xmm1, %xmm2
113; SSE2-NEXT: punpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]
114; SSE2-NEXT: pand %xmm0, %xmm1
115; SSE2-NEXT: movdqa %xmm2, %xmm0
116; SSE2-NEXT: retq
117;
118; SSSE3-LABEL: zext_8i8_to_8i32:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +0000119; SSSE3: # BB#0: # %entry
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +0000120; SSSE3-NEXT: movdqa %xmm0, %xmm2
121; SSSE3-NEXT: punpcklwd {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3]
122; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [255,255,255,255]
123; SSSE3-NEXT: pand %xmm1, %xmm2
124; SSSE3-NEXT: punpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]
125; SSSE3-NEXT: pand %xmm0, %xmm1
126; SSSE3-NEXT: movdqa %xmm2, %xmm0
127; SSSE3-NEXT: retq
128;
129; SSE41-LABEL: zext_8i8_to_8i32:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +0000130; SSE41: # BB#0: # %entry
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +0000131; SSE41-NEXT: pmovzxwd %xmm0, %xmm2
132; SSE41-NEXT: movdqa {{.*#+}} xmm1 = [255,255,255,255]
133; SSE41-NEXT: pand %xmm1, %xmm2
134; SSE41-NEXT: punpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]
135; SSE41-NEXT: pand %xmm0, %xmm1
136; SSE41-NEXT: movdqa %xmm2, %xmm0
137; SSE41-NEXT: retq
138;
139; AVX1-LABEL: zext_8i8_to_8i32:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +0000140; AVX1: # BB#0: # %entry
Chandler Carruth99627bf2014-10-04 03:52:55 +0000141; AVX1-NEXT: vpmovzxwd %xmm0, %xmm1
142; AVX1-NEXT: vpunpckhwd {{.*#+}} xmm0 = xmm0[4,4,5,5,6,6,7,7]
143; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
144; AVX1-NEXT: vandps {{.*}}, %ymm0, %ymm0
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +0000145; AVX1-NEXT: retq
146;
147; AVX2-LABEL: zext_8i8_to_8i32:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +0000148; AVX2: # BB#0: # %entry
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +0000149; AVX2-NEXT: vpmovzxwd %xmm0, %ymm0
Chandler Carruth99627bf2014-10-04 03:52:55 +0000150; AVX2-NEXT: vpbroadcastd {{.*}}, %ymm1
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +0000151; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm0
152; AVX2-NEXT: retq
Chandler Carruth011088f2014-10-01 20:27:16 +0000153entry:
Nadav Rotem34ca89a2012-02-12 15:05:31 +0000154 %t = zext <8 x i8> %z to <8 x i32>
155 ret <8 x i32> %t
156}
Benjamin Kramerda8446b2013-10-23 19:19:04 +0000157
158; PR17654
159define <16 x i16> @zext_16i8_to_16i16(<16 x i8> %z) {
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +0000160; SSE2-LABEL: zext_16i8_to_16i16:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +0000161; SSE2: # BB#0: # %entry
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +0000162; SSE2-NEXT: movdqa %xmm0, %xmm2
163; SSE2-NEXT: punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
164; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
165; SSE2-NEXT: pand %xmm1, %xmm2
166; SSE2-NEXT: punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
167; SSE2-NEXT: pand %xmm0, %xmm1
168; SSE2-NEXT: movdqa %xmm2, %xmm0
169; SSE2-NEXT: retq
170;
171; SSSE3-LABEL: zext_16i8_to_16i16:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +0000172; SSSE3: # BB#0: # %entry
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +0000173; SSSE3-NEXT: movdqa %xmm0, %xmm2
174; SSSE3-NEXT: punpcklbw {{.*#+}} xmm2 = xmm2[0],xmm0[0],xmm2[1],xmm0[1],xmm2[2],xmm0[2],xmm2[3],xmm0[3],xmm2[4],xmm0[4],xmm2[5],xmm0[5],xmm2[6],xmm0[6],xmm2[7],xmm0[7]
175; SSSE3-NEXT: movdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
176; SSSE3-NEXT: pand %xmm1, %xmm2
177; SSSE3-NEXT: punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
178; SSSE3-NEXT: pand %xmm0, %xmm1
179; SSSE3-NEXT: movdqa %xmm2, %xmm0
180; SSSE3-NEXT: retq
181;
182; SSE41-LABEL: zext_16i8_to_16i16:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +0000183; SSE41: # BB#0: # %entry
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +0000184; SSE41-NEXT: pmovzxbw %xmm0, %xmm2
185; SSE41-NEXT: movdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
186; SSE41-NEXT: pand %xmm1, %xmm2
187; SSE41-NEXT: punpckhbw {{.*#+}} xmm0 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
188; SSE41-NEXT: pand %xmm0, %xmm1
189; SSE41-NEXT: movdqa %xmm2, %xmm0
190; SSE41-NEXT: retq
191;
192; AVX1-LABEL: zext_16i8_to_16i16:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +0000193; AVX1: # BB#0: # %entry
Chandler Carruthacecdc02014-10-05 12:07:34 +0000194; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
195; AVX1-NEXT: vpunpckhbw {{.*#+}} xmm1 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15]
Chandler Carruth99627bf2014-10-04 03:52:55 +0000196; AVX1-NEXT: vpmovzxbw %xmm0, %xmm0
197; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +0000198; AVX1-NEXT: retq
199;
200; AVX2-LABEL: zext_16i8_to_16i16:
Chandler Carruthc1bb0e82014-10-02 06:52:19 +0000201; AVX2: # BB#0: # %entry
Chandler Carruthbbbdb9f2014-10-01 20:41:36 +0000202; AVX2-NEXT: vpmovzxbw %xmm0, %ymm0
203; AVX2-NEXT: retq
Chandler Carruth011088f2014-10-01 20:27:16 +0000204entry:
Benjamin Kramerda8446b2013-10-23 19:19:04 +0000205 %t = zext <16 x i8> %z to <16 x i16>
206 ret <16 x i16> %t
207}