1 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s --check-prefix=ALL --check-prefix=SSE2
2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=sse4.1 | FileCheck %s --check-prefix=ALL --check-prefix=SSE41
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=core-avx2 | FileCheck %s --check-prefix=AVX2
5 define <16 x i8> @mul8c(<16 x i8> %i) nounwind {
7 ; SSE2: # BB#0: # %entry
8 ; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [117,117,117,117,117,117,117,117,117,117,117,117,117,117,117,117]
9 ; SSE2-NEXT: psraw $8, %xmm1
10 ; SSE2-NEXT: movdqa %xmm0, %xmm2
11 ; SSE2-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
12 ; SSE2-NEXT: psraw $8, %xmm2
13 ; SSE2-NEXT: pmullw %xmm1, %xmm2
14 ; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
15 ; SSE2-NEXT: pand %xmm3, %xmm2
16 ; SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
17 ; SSE2-NEXT: psraw $8, %xmm0
18 ; SSE2-NEXT: pmullw %xmm1, %xmm0
19 ; SSE2-NEXT: pand %xmm3, %xmm0
20 ; SSE2-NEXT: packuswb %xmm2, %xmm0
24 ; SSE41: # BB#0: # %entry
25 ; SSE41-NEXT: pmovsxbw %xmm0, %xmm1
26 ; SSE41-NEXT: pmovsxbw {{.*}}(%rip), %xmm2
27 ; SSE41-NEXT: pmullw %xmm2, %xmm1
28 ; SSE41-NEXT: movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
29 ; SSE41-NEXT: pand %xmm3, %xmm1
30 ; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
31 ; SSE41-NEXT: pmovsxbw %xmm0, %xmm0
32 ; SSE41-NEXT: pmullw %xmm2, %xmm0
33 ; SSE41-NEXT: pand %xmm3, %xmm0
34 ; SSE41-NEXT: packuswb %xmm0, %xmm1
35 ; SSE41-NEXT: movdqa %xmm1, %xmm0
38 %A = mul <16 x i8> %i, < i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117, i8 117 >
42 define <8 x i16> @mul16c(<8 x i16> %i) nounwind {
44 ; ALL: # BB#0: # %entry
45 ; ALL-NEXT: pmullw {{.*}}(%rip), %xmm0
48 %A = mul <8 x i16> %i, < i16 117, i16 117, i16 117, i16 117, i16 117, i16 117, i16 117, i16 117 >
52 define <4 x i32> @a(<4 x i32> %i) nounwind {
54 ; SSE2: # BB#0: # %entry
55 ; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [117,117,117,117]
56 ; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
57 ; SSE2-NEXT: pmuludq %xmm1, %xmm0
58 ; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
59 ; SSE2-NEXT: pmuludq %xmm1, %xmm2
60 ; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
61 ; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
65 ; SSE41: # BB#0: # %entry
66 ; SSE41-NEXT: pmulld {{.*}}(%rip), %xmm0
69 %A = mul <4 x i32> %i, < i32 117, i32 117, i32 117, i32 117 >
73 define <2 x i64> @b(<2 x i64> %i) nounwind {
75 ; ALL: # BB#0: # %entry
76 ; ALL-NEXT: movdqa {{.*#+}} xmm1 = [117,117]
77 ; ALL-NEXT: movdqa %xmm0, %xmm2
78 ; ALL-NEXT: pmuludq %xmm1, %xmm2
79 ; ALL-NEXT: psrlq $32, %xmm0
80 ; ALL-NEXT: pmuludq %xmm1, %xmm0
81 ; ALL-NEXT: psllq $32, %xmm0
82 ; ALL-NEXT: paddq %xmm2, %xmm0
85 %A = mul <2 x i64> %i, < i64 117, i64 117 >
89 define <16 x i8> @mul8(<16 x i8> %i, <16 x i8> %j) nounwind {
91 ; SSE2: # BB#0: # %entry
92 ; SSE2-NEXT: movdqa %xmm1, %xmm2
93 ; SSE2-NEXT: punpckhbw {{.*#+}} xmm2 = xmm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
94 ; SSE2-NEXT: psraw $8, %xmm2
95 ; SSE2-NEXT: movdqa %xmm0, %xmm3
96 ; SSE2-NEXT: punpckhbw {{.*#+}} xmm3 = xmm3[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
97 ; SSE2-NEXT: psraw $8, %xmm3
98 ; SSE2-NEXT: pmullw %xmm2, %xmm3
99 ; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,255]
100 ; SSE2-NEXT: pand %xmm2, %xmm3
101 ; SSE2-NEXT: punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
102 ; SSE2-NEXT: psraw $8, %xmm1
103 ; SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
104 ; SSE2-NEXT: psraw $8, %xmm0
105 ; SSE2-NEXT: pmullw %xmm1, %xmm0
106 ; SSE2-NEXT: pand %xmm2, %xmm0
107 ; SSE2-NEXT: packuswb %xmm3, %xmm0
111 ; SSE41: # BB#0: # %entry
112 ; SSE41-NEXT: pmovsxbw %xmm1, %xmm3
113 ; SSE41-NEXT: pmovsxbw %xmm0, %xmm2
114 ; SSE41-NEXT: pmullw %xmm3, %xmm2
115 ; SSE41-NEXT: movdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255]
116 ; SSE41-NEXT: pand %xmm3, %xmm2
117 ; SSE41-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,0,1]
118 ; SSE41-NEXT: pmovsxbw %xmm1, %xmm1
119 ; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
120 ; SSE41-NEXT: pmovsxbw %xmm0, %xmm0
121 ; SSE41-NEXT: pmullw %xmm1, %xmm0
122 ; SSE41-NEXT: pand %xmm3, %xmm0
123 ; SSE41-NEXT: packuswb %xmm0, %xmm2
124 ; SSE41-NEXT: movdqa %xmm2, %xmm0
127 %A = mul <16 x i8> %i, %j
131 define <8 x i16> @mul16(<8 x i16> %i, <8 x i16> %j) nounwind {
133 ; ALL: # BB#0: # %entry
134 ; ALL-NEXT: pmullw %xmm1, %xmm0
137 %A = mul <8 x i16> %i, %j
141 define <4 x i32> @c(<4 x i32> %i, <4 x i32> %j) nounwind {
143 ; SSE2: # BB#0: # %entry
144 ; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
145 ; SSE2-NEXT: pmuludq %xmm1, %xmm0
146 ; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
147 ; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[1,1,3,3]
148 ; SSE2-NEXT: pmuludq %xmm2, %xmm1
149 ; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3]
150 ; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
154 ; SSE41: # BB#0: # %entry
155 ; SSE41-NEXT: pmulld %xmm1, %xmm0
158 %A = mul <4 x i32> %i, %j
162 define <2 x i64> @d(<2 x i64> %i, <2 x i64> %j) nounwind {
164 ; ALL: # BB#0: # %entry
165 ; ALL-NEXT: movdqa %xmm0, %xmm2
166 ; ALL-NEXT: pmuludq %xmm1, %xmm2
167 ; ALL-NEXT: movdqa %xmm1, %xmm3
168 ; ALL-NEXT: psrlq $32, %xmm3
169 ; ALL-NEXT: pmuludq %xmm0, %xmm3
170 ; ALL-NEXT: psllq $32, %xmm3
171 ; ALL-NEXT: paddq %xmm3, %xmm2
172 ; ALL-NEXT: psrlq $32, %xmm0
173 ; ALL-NEXT: pmuludq %xmm1, %xmm0
174 ; ALL-NEXT: psllq $32, %xmm0
175 ; ALL-NEXT: paddq %xmm2, %xmm0
178 %A = mul <2 x i64> %i, %j
184 define <4 x i32> @e(<4 x i32> %i, <4 x i32> %j) nounwind {
186 ; SSE2: # BB#0: # %entry
187 ; SSE2-NEXT: subq $40, %rsp
188 ; SSE2-NEXT: movaps %xmm1, {{[0-9]+}}(%rsp) # 16-byte Spill
189 ; SSE2-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
190 ; SSE2-NEXT: callq foo
191 ; SSE2-NEXT: movdqa (%rsp), %xmm0 # 16-byte Reload
192 ; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm0[1,1,3,3]
193 ; SSE2-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm2 # 16-byte Reload
194 ; SSE2-NEXT: pmuludq %xmm2, %xmm0
195 ; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
196 ; SSE2-NEXT: pshufd {{.*#+}} xmm2 = xmm2[1,1,3,3]
197 ; SSE2-NEXT: pmuludq %xmm1, %xmm2
198 ; SSE2-NEXT: pshufd {{.*#+}} xmm1 = xmm2[0,2,2,3]
199 ; SSE2-NEXT: punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
200 ; SSE2-NEXT: addq $40, %rsp
204 ; SSE41: # BB#0: # %entry
205 ; SSE41-NEXT: subq $40, %rsp
206 ; SSE41-NEXT: movaps %xmm1, {{[0-9]+}}(%rsp) # 16-byte Spill
207 ; SSE41-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
208 ; SSE41-NEXT: callq foo
209 ; SSE41-NEXT: movdqa (%rsp), %xmm0 # 16-byte Reload
210 ; SSE41-NEXT: pmulld {{[0-9]+}}(%rsp), %xmm0 # 16-byte Folded Reload
211 ; SSE41-NEXT: addq $40, %rsp
214 ; Use a call to force spills.
216 %A = mul <4 x i32> %i, %j
220 define <2 x i64> @f(<2 x i64> %i, <2 x i64> %j) nounwind {
222 ; ALL: # BB#0: # %entry
223 ; ALL-NEXT: subq $40, %rsp
224 ; ALL-NEXT: movaps %xmm1, {{[0-9]+}}(%rsp) # 16-byte Spill
225 ; ALL-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
226 ; ALL-NEXT: callq foo
227 ; ALL-NEXT: movdqa (%rsp), %xmm0 # 16-byte Reload
228 ; ALL-NEXT: movdqa %xmm0, %xmm2
229 ; ALL-NEXT: movdqa {{[0-9]+}}(%rsp), %xmm3 # 16-byte Reload
230 ; ALL-NEXT: pmuludq %xmm3, %xmm2
231 ; ALL-NEXT: movdqa %xmm3, %xmm1
232 ; ALL-NEXT: psrlq $32, %xmm1
233 ; ALL-NEXT: pmuludq %xmm0, %xmm1
234 ; ALL-NEXT: psllq $32, %xmm1
235 ; ALL-NEXT: paddq %xmm1, %xmm2
236 ; ALL-NEXT: psrlq $32, %xmm0
237 ; ALL-NEXT: pmuludq %xmm3, %xmm0
238 ; ALL-NEXT: psllq $32, %xmm0
239 ; ALL-NEXT: paddq %xmm2, %xmm0
240 ; ALL-NEXT: addq $40, %rsp
243 ; Use a call to force spills.
245 %A = mul <2 x i64> %i, %j
249 define <4 x i64> @b1(<4 x i64> %i) nounwind {
252 ; AVX2-NEXT: vpmuludq
253 ; AVX2-NEXT: vpsrlq $32
254 ; AVX2-NEXT: vpmuludq
255 ; AVX2-NEXT: vpsllq $32
259 %A = mul <4 x i64> %i, < i64 117, i64 117, i64 117, i64 117 >
263 define <4 x i64> @b2(<4 x i64> %i, <4 x i64> %j) nounwind {
266 ; AVX2-NEXT: vpsrlq $32
267 ; AVX2-NEXT: vpmuludq
268 ; AVX2-NEXT: vpsllq $32
270 ; AVX2-NEXT: vpsrlq $32
271 ; AVX2-NEXT: vpmuludq
272 ; AVX2-NEXT: vpsllq $32
276 %A = mul <4 x i64> %i, %j