xref: /llvm-project/llvm/test/Transforms/InstCombine/X86/x86-pmaddwd.ll (revision d893ed78718e25a982dcba9cdba2d78212b79353) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -passes=instcombine -mtriple=x86_64-unknown-unknown -S | FileCheck %s
3
4;
5; UNDEF Elts
6;
7
8define <4 x i32> @undef_pmaddwd_128(<8 x i16> %a0) {
9; CHECK-LABEL: @undef_pmaddwd_128(
10; CHECK-NEXT:    ret <4 x i32> zeroinitializer
11;
12  %1 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %a0, <8 x i16> undef)
13  ret <4 x i32> %1
14}
15
16define <4 x i32> @undef_pmaddwd_128_commute(<8 x i16> %a0) {
17; CHECK-LABEL: @undef_pmaddwd_128_commute(
18; CHECK-NEXT:    ret <4 x i32> zeroinitializer
19;
20  %1 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> undef, <8 x i16> %a0)
21  ret <4 x i32> %1
22}
23
24define <8 x i32> @undef_pmaddwd_256(<16 x i16> %a0) {
25; CHECK-LABEL: @undef_pmaddwd_256(
26; CHECK-NEXT:    ret <8 x i32> zeroinitializer
27;
28  %1 = call <8 x i32> @llvm.x86.avx2.pmadd.wd(<16 x i16> %a0, <16 x i16> undef)
29  ret <8 x i32> %1
30}
31
32define <8 x i32> @undef_pmaddwd_256_commute(<16 x i16> %a0) {
33; CHECK-LABEL: @undef_pmaddwd_256_commute(
34; CHECK-NEXT:    ret <8 x i32> zeroinitializer
35;
36  %1 = call <8 x i32> @llvm.x86.avx2.pmadd.wd(<16 x i16> undef, <16 x i16> %a0)
37  ret <8 x i32> %1
38}
39
40define <16 x i32> @undef_pmaddwd_512(<32 x i16> %a0) {
41; CHECK-LABEL: @undef_pmaddwd_512(
42; CHECK-NEXT:    ret <16 x i32> zeroinitializer
43;
44  %1 = call <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16> %a0, <32 x i16> undef)
45  ret <16 x i32> %1
46}
47
48define <16 x i32> @undef_pmaddwd_512_commute(<32 x i16> %a0) {
49; CHECK-LABEL: @undef_pmaddwd_512_commute(
50; CHECK-NEXT:    ret <16 x i32> zeroinitializer
51;
52  %1 = call <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16> undef, <32 x i16> %a0)
53  ret <16 x i32> %1
54}
55
56;
57; Zero Elts
58;
59
60define <4 x i32> @zero_pmaddwd_128(<8 x i16> %a0) {
61; CHECK-LABEL: @zero_pmaddwd_128(
62; CHECK-NEXT:    ret <4 x i32> zeroinitializer
63;
64  %1 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %a0, <8 x i16> zeroinitializer)
65  ret <4 x i32> %1
66}
67
68define <4 x i32> @zero_pmaddwd_128_commute(<8 x i16> %a0) {
69; CHECK-LABEL: @zero_pmaddwd_128_commute(
70; CHECK-NEXT:    ret <4 x i32> zeroinitializer
71;
72  %1 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> zeroinitializer, <8 x i16> %a0)
73  ret <4 x i32> %1
74}
75
76define <8 x i32> @zero_pmaddwd_256(<16 x i16> %a0) {
77; CHECK-LABEL: @zero_pmaddwd_256(
78; CHECK-NEXT:    ret <8 x i32> zeroinitializer
79;
80  %1 = call <8 x i32> @llvm.x86.avx2.pmadd.wd(<16 x i16> %a0, <16 x i16> zeroinitializer)
81  ret <8 x i32> %1
82}
83
84define <8 x i32> @zero_pmaddwd_256_commute(<16 x i16> %a0) {
85; CHECK-LABEL: @zero_pmaddwd_256_commute(
86; CHECK-NEXT:    ret <8 x i32> zeroinitializer
87;
88  %1 = call <8 x i32> @llvm.x86.avx2.pmadd.wd(<16 x i16> zeroinitializer, <16 x i16> %a0)
89  ret <8 x i32> %1
90}
91
92define <16 x i32> @zero_pmaddwd_512(<32 x i16> %a0) {
93; CHECK-LABEL: @zero_pmaddwd_512(
94; CHECK-NEXT:    ret <16 x i32> zeroinitializer
95;
96  %1 = call <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16> %a0, <32 x i16> zeroinitializer)
97  ret <16 x i32> %1
98}
99
100define <16 x i32> @zero_pmaddwd_512_commute(<32 x i16> %a0) {
101; CHECK-LABEL: @zero_pmaddwd_512_commute(
102; CHECK-NEXT:    ret <16 x i32> zeroinitializer
103;
104  %1 = call <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16> zeroinitializer, <32 x i16> %a0)
105  ret <16 x i32> %1
106}
107
108;
109; Constant Folding
110;
111
112define <4 x i32> @fold_pmaddwd_128() {
113; CHECK-LABEL: @fold_pmaddwd_128(
114; CHECK-NEXT:    ret <4 x i32> <i32 19, i32 -229364, i32 -21, i32 -491429>
115;
116  %1 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> <i16 -1, i16 2, i16 3, i16 -4, i16 -5, i16 6, i16 7, i16 -8>, <8 x i16> <i16 -5, i16 7, i16 -32768, i16 32765, i16 -9, i16 -11, i16 -32763, i16 32761>)
117  ret <4 x i32> %1
118}
119
120define <8 x i32> @fold_pmaddwd_256() {
121; CHECK-LABEL: @fold_pmaddwd_256(
122; CHECK-NEXT:    ret <8 x i32> <i32 -7, i32 32762, i32 91, i32 32750, i32 -239, i32 687938, i32 -451, i32 -32756>
123;
124  %1 = call <8 x i32> @llvm.x86.avx2.pmadd.wd(<16 x i16> <i16 0, i16 -1, i16 2, i16 3, i16 -4, i16 -5, i16 6, i16 7, i16 -8, i16 9, i16 -10, i16 11, i16 -12, i16 13, i16 -14, i16 -15>, <16 x i16> <i16 -5, i16 7, i16 -32768, i16 32766, i16 -9, i16 -11, i16 -32764, i16 32762, i16 13, i16 -15, i16 -32760, i16 32758, i16 17, i16 -19, i16 -32756, i16 32756>)
125  ret <8 x i32> %1
126}
127
128define <16 x i32> @fold_pmaddwd_512() {
129; CHECK-LABEL: @fold_pmaddwd_512(
130; CHECK-NEXT:    ret <16 x i32> <i32 -7, i32 32762, i32 91, i32 32750, i32 -239, i32 687938, i32 -451, i32 -32756, i32 -7, i32 32762, i32 91, i32 32750, i32 -239, i32 687938, i32 -451, i32 -32756>
131;
132  %1 = call <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16> <i16 0, i16 -1, i16 2, i16 3, i16 -4, i16 -5, i16 6, i16 7, i16 -8, i16 9, i16 -10, i16 11, i16 -12, i16 13, i16 -14, i16 -15, i16 -5, i16 7, i16 -32768, i16 32766, i16 -9, i16 -11, i16 -32764, i16 32762, i16 13, i16 -15, i16 -32760, i16 32758, i16 17, i16 -19, i16 -32756, i16 32756>, <32 x i16> <i16 -5, i16 7, i16 -32768, i16 32766, i16 -9, i16 -11, i16 -32764, i16 32762, i16 13, i16 -15, i16 -32760, i16 32758, i16 17, i16 -19, i16 -32756, i16 32756, i16 0, i16 -1, i16 2, i16 3, i16 -4, i16 -5, i16 6, i16 7, i16 -8, i16 9, i16 -10, i16 11, i16 -12, i16 13, i16 -14, i16 -15>)
133  ret <16 x i32> %1
134}
135
136;
137; Demanded Elts
138;
139
140define <4 x i32> @elts_pmaddwd_128(<8 x i16> %a0, <8 x i16> %a1) {
141; CHECK-LABEL: @elts_pmaddwd_128(
142; CHECK-NEXT:    [[TMP1:%.*]] = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> [[A0:%.*]], <8 x i16> [[A1:%.*]])
143; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <4 x i32> [[TMP1]], <4 x i32> poison, <4 x i32> zeroinitializer
144; CHECK-NEXT:    ret <4 x i32> [[TMP2]]
145;
146  %1 = shufflevector <8 x i16> %a0, <8 x i16> undef, <8 x i32> <i32 0, i32 1, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2>
147  %2 = shufflevector <8 x i16> %a1, <8 x i16> undef, <8 x i32> <i32 0, i32 1, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6>
148  %3 = call <4 x i32> @llvm.x86.sse2.pmadd.wd(<8 x i16> %1, <8 x i16> %2)
149  %4 = shufflevector <4 x i32> %3, <4 x i32> poison, <4 x i32> zeroinitializer
150  ret <4 x i32> %4
151}
152
153define <8 x i32> @elts_pmaddwd_256(<16 x i16> %a0, <16 x i16> %a1) {
154; CHECK-LABEL: @elts_pmaddwd_256(
155; CHECK-NEXT:    [[TMP1:%.*]] = call <8 x i32> @llvm.x86.avx2.pmadd.wd(<16 x i16> [[A0:%.*]], <16 x i16> [[A1:%.*]])
156; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <8 x i32> [[TMP1]], <8 x i32> poison, <8 x i32> zeroinitializer
157; CHECK-NEXT:    ret <8 x i32> [[TMP2]]
158;
159  %1 = shufflevector <16 x i16> %a0, <16 x i16> undef, <16 x i32> <i32 0, i32 1, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
160  %2 = shufflevector <16 x i16> %a1, <16 x i16> undef, <16 x i32> <i32 0, i32 1, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
161  %3 = call <8 x i32> @llvm.x86.avx2.pmadd.wd(<16 x i16> %1, <16 x i16> %2)
162  %4 = shufflevector <8 x i32> %3, <8 x i32> poison, <8 x i32> zeroinitializer
163  ret <8 x i32> %4
164}
165
166define <16 x i32> @elts_pmaddwd_512(<32 x i16> %a0, <32 x i16> %a1) {
167; CHECK-LABEL: @elts_pmaddwd_512(
168; CHECK-NEXT:    [[TMP1:%.*]] = call <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16> [[A0:%.*]], <32 x i16> [[A1:%.*]])
169; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <16 x i32> [[TMP1]], <16 x i32> poison, <16 x i32> zeroinitializer
170; CHECK-NEXT:    ret <16 x i32> [[TMP2]]
171;
172  %1 = shufflevector <32 x i16> %a0, <32 x i16> undef, <32 x i32> <i32 0, i32 1, i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
173  %2 = shufflevector <32 x i16> %a1, <32 x i16> undef, <32 x i32> <i32 0, i32 1, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
174  %3 = call <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16> %1, <32 x i16> %2)
175  %4 = shufflevector <16 x i32> %3, <16 x i32> poison, <16 x i32> zeroinitializer
176  ret <16 x i32> %4
177}
178