xref: /llvm-project/llvm/test/CodeGen/X86/stack-folding-int-avxvnni.ll (revision 756f5978410809530150f5e1cd425e85ad94d1cd) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc -O3 -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avxvnni < %s | FileCheck %s
3
4target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
5target triple = "x86_64-unknown-unknown"
6
7declare <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32>, <4 x i32>, <4 x i32>)
8declare <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32>, <8 x i32>, <8 x i32>)
9declare <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32>, <4 x i32>, <4 x i32>)
10declare <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32>, <8 x i32>, <8 x i32>)
11declare <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32>, <4 x i32>, <4 x i32>)
12declare <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32>, <8 x i32>, <8 x i32>)
13declare <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32>, <4 x i32>, <4 x i32>)
14declare <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32>, <8 x i32>, <8 x i32>)
15
16define <4 x i32> @stack_fold_vpdpwssd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) {
17; CHECK-LABEL: stack_fold_vpdpwssd:
18; CHECK:       # %bb.0:
19; CHECK-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
20; CHECK-NEXT:    #APP
21; CHECK-NEXT:    nop
22; CHECK-NEXT:    #NO_APP
23; CHECK-NEXT:    {vex} vpdpwssd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload
24; CHECK-NEXT:    retq
25  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
26  %2 = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2)
27  ret <4 x i32> %2
28}
29
30define <4 x i32> @stack_fold_vpdpwssd_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) {
31; CHECK-LABEL: stack_fold_vpdpwssd_commuted:
32; CHECK:       # %bb.0:
33; CHECK-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
34; CHECK-NEXT:    #APP
35; CHECK-NEXT:    nop
36; CHECK-NEXT:    #NO_APP
37; CHECK-NEXT:    {vex} vpdpwssd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload
38; CHECK-NEXT:    retq
39  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
40  %2 = call <4 x i32> @llvm.x86.avx512.vpdpwssd.128(<4 x i32> %a0, <4 x i32> %a2, <4 x i32> %a1)
41  ret <4 x i32> %2
42}
43
44define <8 x i32> @stack_fold_vpdpwssd_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) {
45; CHECK-LABEL: stack_fold_vpdpwssd_256:
46; CHECK:       # %bb.0:
47; CHECK-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
48; CHECK-NEXT:    #APP
49; CHECK-NEXT:    nop
50; CHECK-NEXT:    #NO_APP
51; CHECK-NEXT:    {vex} vpdpwssd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
52; CHECK-NEXT:    retq
53  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
54  %2 = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2)
55  ret <8 x i32> %2
56}
57
58define <8 x i32> @stack_fold_vpdpwssd_256_commuted(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) {
59; CHECK-LABEL: stack_fold_vpdpwssd_256_commuted:
60; CHECK:       # %bb.0:
61; CHECK-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
62; CHECK-NEXT:    #APP
63; CHECK-NEXT:    nop
64; CHECK-NEXT:    #NO_APP
65; CHECK-NEXT:    {vex} vpdpwssd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
66; CHECK-NEXT:    retq
67  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
68  %2 = call <8 x i32> @llvm.x86.avx512.vpdpwssd.256(<8 x i32> %a0, <8 x i32> %a2, <8 x i32> %a1)
69  ret <8 x i32> %2
70}
71
72define <4 x i32> @stack_fold_vpdpwssds(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) {
73; CHECK-LABEL: stack_fold_vpdpwssds:
74; CHECK:       # %bb.0:
75; CHECK-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
76; CHECK-NEXT:    #APP
77; CHECK-NEXT:    nop
78; CHECK-NEXT:    #NO_APP
79; CHECK-NEXT:    {vex} vpdpwssds {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload
80; CHECK-NEXT:    retq
81  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
82  %2 = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2)
83  ret <4 x i32> %2
84}
85
86define <4 x i32> @stack_fold_vpdpwssds_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) {
87; CHECK-LABEL: stack_fold_vpdpwssds_commuted:
88; CHECK:       # %bb.0:
89; CHECK-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
90; CHECK-NEXT:    #APP
91; CHECK-NEXT:    nop
92; CHECK-NEXT:    #NO_APP
93; CHECK-NEXT:    {vex} vpdpwssds {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload
94; CHECK-NEXT:    retq
95  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
96  %2 = call <4 x i32> @llvm.x86.avx512.vpdpwssds.128(<4 x i32> %a0, <4 x i32> %a2, <4 x i32> %a1)
97  ret <4 x i32> %2
98}
99
100define <8 x i32> @stack_fold_vpdpwssds_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) {
101; CHECK-LABEL: stack_fold_vpdpwssds_256:
102; CHECK:       # %bb.0:
103; CHECK-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
104; CHECK-NEXT:    #APP
105; CHECK-NEXT:    nop
106; CHECK-NEXT:    #NO_APP
107; CHECK-NEXT:    {vex} vpdpwssds {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
108; CHECK-NEXT:    retq
109  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
110  %2 = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2)
111  ret <8 x i32> %2
112}
113
114define <8 x i32> @stack_fold_vpdpwssds_256_commuted(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) {
115; CHECK-LABEL: stack_fold_vpdpwssds_256_commuted:
116; CHECK:       # %bb.0:
117; CHECK-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
118; CHECK-NEXT:    #APP
119; CHECK-NEXT:    nop
120; CHECK-NEXT:    #NO_APP
121; CHECK-NEXT:    {vex} vpdpwssds {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
122; CHECK-NEXT:    retq
123  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
124  %2 = call <8 x i32> @llvm.x86.avx512.vpdpwssds.256(<8 x i32> %a0, <8 x i32> %a2, <8 x i32> %a1)
125  ret <8 x i32> %2
126}
127
128define <4 x i32> @stack_fold_vpdpbusd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) {
129; CHECK-LABEL: stack_fold_vpdpbusd:
130; CHECK:       # %bb.0:
131; CHECK-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
132; CHECK-NEXT:    #APP
133; CHECK-NEXT:    nop
134; CHECK-NEXT:    #NO_APP
135; CHECK-NEXT:    {vex} vpdpbusd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload
136; CHECK-NEXT:    retq
137  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
138  %2 = call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2)
139  ret <4 x i32> %2
140}
141
142define <4 x i32> @stack_fold_vpdpbusd_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) {
143; CHECK-LABEL: stack_fold_vpdpbusd_commuted:
144; CHECK:       # %bb.0:
145; CHECK-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
146; CHECK-NEXT:    #APP
147; CHECK-NEXT:    nop
148; CHECK-NEXT:    #NO_APP
149; CHECK-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
150; CHECK-NEXT:    {vex} vpdpbusd %xmm1, %xmm2, %xmm0
151; CHECK-NEXT:    retq
152  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
153  %2 = call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> %a0, <4 x i32> %a2, <4 x i32> %a1)
154  ret <4 x i32> %2
155}
156
157define <8 x i32> @stack_fold_vpdpbusd_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) {
158; CHECK-LABEL: stack_fold_vpdpbusd_256:
159; CHECK:       # %bb.0:
160; CHECK-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
161; CHECK-NEXT:    #APP
162; CHECK-NEXT:    nop
163; CHECK-NEXT:    #NO_APP
164; CHECK-NEXT:    {vex} vpdpbusd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
165; CHECK-NEXT:    retq
166  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
167  %2 = call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2)
168  ret <8 x i32> %2
169}
170
171define <8 x i32> @stack_fold_vpdpbusd_256_commuted(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) {
172; CHECK-LABEL: stack_fold_vpdpbusd_256_commuted:
173; CHECK:       # %bb.0:
174; CHECK-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
175; CHECK-NEXT:    #APP
176; CHECK-NEXT:    nop
177; CHECK-NEXT:    #NO_APP
178; CHECK-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
179; CHECK-NEXT:    {vex} vpdpbusd %ymm1, %ymm2, %ymm0
180; CHECK-NEXT:    retq
181  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
182  %2 = call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %a0, <8 x i32> %a2, <8 x i32> %a1)
183  ret <8 x i32> %2
184}
185
186define <4 x i32> @stack_fold_vpdpbusds(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) {
187; CHECK-LABEL: stack_fold_vpdpbusds:
188; CHECK:       # %bb.0:
189; CHECK-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
190; CHECK-NEXT:    #APP
191; CHECK-NEXT:    nop
192; CHECK-NEXT:    #NO_APP
193; CHECK-NEXT:    {vex} vpdpbusds {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload
194; CHECK-NEXT:    retq
195  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
196  %2 = call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2)
197  ret <4 x i32> %2
198}
199
200define <4 x i32> @stack_fold_vpdpbusds_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) {
201; CHECK-LABEL: stack_fold_vpdpbusds_commuted:
202; CHECK:       # %bb.0:
203; CHECK-NEXT:    vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
204; CHECK-NEXT:    #APP
205; CHECK-NEXT:    nop
206; CHECK-NEXT:    #NO_APP
207; CHECK-NEXT:    vmovdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
208; CHECK-NEXT:    {vex} vpdpbusds %xmm1, %xmm2, %xmm0
209; CHECK-NEXT:    retq
210  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
211  %2 = call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> %a0, <4 x i32> %a2, <4 x i32> %a1)
212  ret <4 x i32> %2
213}
214
215define <8 x i32> @stack_fold_vpdpbusds_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) {
216; CHECK-LABEL: stack_fold_vpdpbusds_256:
217; CHECK:       # %bb.0:
218; CHECK-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
219; CHECK-NEXT:    #APP
220; CHECK-NEXT:    nop
221; CHECK-NEXT:    #NO_APP
222; CHECK-NEXT:    {vex} vpdpbusds {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload
223; CHECK-NEXT:    retq
224  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
225  %2 = call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2)
226  ret <8 x i32> %2
227}
228
229define <8 x i32> @stack_fold_vpdpbusds_256_commuted(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) {
230; CHECK-LABEL: stack_fold_vpdpbusds_256_commuted:
231; CHECK:       # %bb.0:
232; CHECK-NEXT:    vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
233; CHECK-NEXT:    #APP
234; CHECK-NEXT:    nop
235; CHECK-NEXT:    #NO_APP
236; CHECK-NEXT:    vmovdqu {{[-0-9]+}}(%r{{[sb]}}p), %ymm2 # 32-byte Reload
237; CHECK-NEXT:    {vex} vpdpbusds %ymm1, %ymm2, %ymm0
238; CHECK-NEXT:    retq
239  %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"()
240  %2 = call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %a0, <8 x i32> %a2, <8 x i32> %a1)
241  ret <8 x i32> %2
242}
243