Skip to content

Commit 62ed9b7

Browse files
fhahnfhahn
committed
[AArch64] Add tests for shuffle (tbl2, tbl2) -> tbl4 fold.
Add extra tests where shuffle (tbl2, tbl2) can be folded to tbl4. Regenerate check lines automatically as well. (cherry-picked from 7d4ee32)
1 parent f086a07 commit 62ed9b7

File tree

1 file changed

+205
-32
lines changed

1 file changed

+205
-32
lines changed

‎llvm/test/CodeGen/AArch64/arm64-tbl.ll

Lines changed: 205 additions & 32 deletions
Original file line numberDiff line numberDiff line change
@@ -1,61 +1,200 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
12
; RUN: llc < %s -mtriple=arm64-eabi -aarch64-neon-syntax=apple | FileCheck %s
23

34
define <8 x i8> @tbl1_8b(<16 x i8> %A, <8 x i8> %B) nounwind {
4-
; CHECK: tbl1_8b
5-
; CHECK: tbl.8b
5+
; CHECK-LABEL: tbl1_8b:
6+
; CHECK: // %bb.0:
7+
; CHECK-NEXT: tbl.8b v0, { v0 }, v1
8+
; CHECK-NEXT: ret
69
%tmp3 = call <8 x i8> @llvm.aarch64.neon.tbl1.v8i8(<16 x i8> %A, <8 x i8> %B)
710
ret <8 x i8> %tmp3
811
}
912

1013
define <16 x i8> @tbl1_16b(<16 x i8> %A, <16 x i8> %B) nounwind {
11-
; CHECK: tbl1_16b
12-
; CHECK: tbl.16b
14+
; CHECK-LABEL: tbl1_16b:
15+
; CHECK: // %bb.0:
16+
; CHECK-NEXT: tbl.16b v0, { v0 }, v1
17+
; CHECK-NEXT: ret
1318
%tmp3 = call <16 x i8> @llvm.aarch64.neon.tbl1.v16i8(<16 x i8> %A, <16 x i8> %B)
1419
ret <16 x i8> %tmp3
1520
}
1621

1722
define <8 x i8> @tbl2_8b(<16 x i8> %A, <16 x i8> %B, <8 x i8> %C) {
18-
; CHECK: tbl2_8b
19-
; CHECK: tbl.8b
23+
; CHECK-LABEL: tbl2_8b:
24+
; CHECK: // %bb.0:
25+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q0_q1 def $q0_q1
26+
; CHECK-NEXT: // kill: def $q0 killed $q0 killed $q0_q1 def $q0_q1
27+
; CHECK-NEXT: tbl.8b v0, { v0, v1 }, v2
28+
; CHECK-NEXT: ret
2029
%tmp3 = call <8 x i8> @llvm.aarch64.neon.tbl2.v8i8(<16 x i8> %A, <16 x i8> %B, <8 x i8> %C)
2130
ret <8 x i8> %tmp3
2231
}
2332

2433
define <16 x i8> @tbl2_16b(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C) {
25-
; CHECK: tbl2_16b
26-
; CHECK: tbl.16b
34+
; CHECK-LABEL: tbl2_16b:
35+
; CHECK: // %bb.0:
36+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q0_q1 def $q0_q1
37+
; CHECK-NEXT: // kill: def $q0 killed $q0 killed $q0_q1 def $q0_q1
38+
; CHECK-NEXT: tbl.16b v0, { v0, v1 }, v2
39+
; CHECK-NEXT: ret
2740
%tmp3 = call <16 x i8> @llvm.aarch64.neon.tbl2.v16i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C)
2841
ret <16 x i8> %tmp3
2942
}
3043

3144
define <8 x i8> @tbl3_8b(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <8 x i8> %D) {
32-
; CHECK: tbl3_8b
33-
; CHECK: tbl.8b
45+
; CHECK-LABEL: tbl3_8b:
46+
; CHECK: // %bb.0:
47+
; CHECK-NEXT: // kill: def $q2 killed $q2 killed $q0_q1_q2 def $q0_q1_q2
48+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q0_q1_q2 def $q0_q1_q2
49+
; CHECK-NEXT: // kill: def $q0 killed $q0 killed $q0_q1_q2 def $q0_q1_q2
50+
; CHECK-NEXT: tbl.8b v0, { v0, v1, v2 }, v3
51+
; CHECK-NEXT: ret
3452
%tmp3 = call <8 x i8> @llvm.aarch64.neon.tbl3.v8i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <8 x i8> %D)
3553
ret <8 x i8> %tmp3
3654
}
3755

3856
define <16 x i8> @tbl3_16b(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D) {
39-
; CHECK: tbl3_16b
40-
; CHECK: tbl.16b
57+
; CHECK-LABEL: tbl3_16b:
58+
; CHECK: // %bb.0:
59+
; CHECK-NEXT: // kill: def $q2 killed $q2 killed $q0_q1_q2 def $q0_q1_q2
60+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q0_q1_q2 def $q0_q1_q2
61+
; CHECK-NEXT: // kill: def $q0 killed $q0 killed $q0_q1_q2 def $q0_q1_q2
62+
; CHECK-NEXT: tbl.16b v0, { v0, v1, v2 }, v3
63+
; CHECK-NEXT: ret
4164
%tmp3 = call <16 x i8> @llvm.aarch64.neon.tbl3.v16i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D)
4265
ret <16 x i8> %tmp3
4366
}
4467

4568
define <8 x i8> @tbl4_8b(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, <8 x i8> %E) {
46-
; CHECK: tbl4_8b
47-
; CHECK: tbl.8b
69+
; CHECK-LABEL: tbl4_8b:
70+
; CHECK: // %bb.0:
71+
; CHECK-NEXT: // kill: def $q3 killed $q3 killed $q0_q1_q2_q3 def $q0_q1_q2_q3
72+
; CHECK-NEXT: // kill: def $q2 killed $q2 killed $q0_q1_q2_q3 def $q0_q1_q2_q3
73+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q0_q1_q2_q3 def $q0_q1_q2_q3
74+
; CHECK-NEXT: // kill: def $q0 killed $q0 killed $q0_q1_q2_q3 def $q0_q1_q2_q3
75+
; CHECK-NEXT: tbl.8b v0, { v0, v1, v2, v3 }, v4
76+
; CHECK-NEXT: ret
4877
%tmp3 = call <8 x i8> @llvm.aarch64.neon.tbl4.v8i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, <8 x i8> %E)
4978
ret <8 x i8> %tmp3
5079
}
5180

5281
define <16 x i8> @tbl4_16b(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, <16 x i8> %E) {
53-
; CHECK: tbl4_16b
54-
; CHECK: tbl.16b
82+
; CHECK-LABEL: tbl4_16b:
83+
; CHECK: // %bb.0:
84+
; CHECK-NEXT: // kill: def $q3 killed $q3 killed $q0_q1_q2_q3 def $q0_q1_q2_q3
85+
; CHECK-NEXT: // kill: def $q2 killed $q2 killed $q0_q1_q2_q3 def $q0_q1_q2_q3
86+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q0_q1_q2_q3 def $q0_q1_q2_q3
87+
; CHECK-NEXT: // kill: def $q0 killed $q0 killed $q0_q1_q2_q3 def $q0_q1_q2_q3
88+
; CHECK-NEXT: tbl.16b v0, { v0, v1, v2, v3 }, v4
89+
; CHECK-NEXT: ret
5590
%tmp3 = call <16 x i8> @llvm.aarch64.neon.tbl4.v16i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, <16 x i8> %E)
5691
ret <16 x i8> %tmp3
5792
}
5893

94+
; CHECK-LABEL: .LCPI8_0:
95+
; CHECK-NEXT: .byte 0 // 0x0
96+
; CHECK-NEXT: .byte 4 // 0x4
97+
; CHECK-NEXT: .byte 8 // 0x8
98+
; CHECK-NEXT: .byte 12 // 0xc
99+
; CHECK-NEXT: .byte 16 // 0x10
100+
; CHECK-NEXT: .byte 20 // 0x14
101+
; CHECK-NEXT: .byte 24 // 0x18
102+
; CHECK-NEXT: .byte 28 // 0x1c
103+
; CHECK-NEXT: .byte 255 // 0xff
104+
; CHECK-NEXT: .byte 255 // 0xff
105+
; CHECK-NEXT: .byte 255 // 0xff
106+
; CHECK-NEXT: .byte 255 // 0xff
107+
; CHECK-NEXT: .byte 255 // 0xff
108+
; CHECK-NEXT: .byte 255 // 0xff
109+
; CHECK-NEXT: .byte 255 // 0xff
110+
; CHECK-NEXT: .byte 255 // 0xff
111+
112+
define <16 x i8> @shuffled_tbl2_to_tbl4(<16 x i8> %a, <16 x i8> %b, <16 x i8> %c, <16 x i8> %d) {
113+
; CHECK-LABEL: shuffled_tbl2_to_tbl4:
114+
; CHECK: // %bb.0:
115+
; CHECK-NEXT: adrp x8, .LCPI8_0
116+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q0_q1 def $q0_q1
117+
; CHECK-NEXT: // kill: def $q3 killed $q3 killed $q2_q3 def $q2_q3
118+
; CHECK-NEXT: // kill: def $q0 killed $q0 killed $q0_q1 def $q0_q1
119+
; CHECK-NEXT: // kill: def $q2 killed $q2 killed $q2_q3 def $q2_q3
120+
; CHECK-NEXT: ldr q4, [x8, :lo12:.LCPI8_0]
121+
; CHECK-NEXT: tbl.16b v0, { v0, v1 }, v4
122+
; CHECK-NEXT: tbl.16b v1, { v2, v3 }, v4
123+
; CHECK-NEXT: mov.d v0[1], v1[0]
124+
; CHECK-NEXT: ret
125+
%t1 = call <16 x i8> @llvm.aarch64.neon.tbl2.v16i8(<16 x i8> %a, <16 x i8> %b, <16 x i8> <i8 0, i8 4, i8 8, i8 12, i8 16, i8 20, i8 24, i8 28, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>)
126+
%t2 = call <16 x i8> @llvm.aarch64.neon.tbl2.v16i8(<16 x i8> %c, <16 x i8> %d, <16 x i8> <i8 0, i8 4, i8 8, i8 12, i8 16, i8 20, i8 24, i8 28, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>)
127+
%s = shufflevector <16 x i8> %t1, <16 x i8> %t2, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
128+
ret <16 x i8> %s
129+
}
130+
131+
define <16 x i8> @shuffled_tbl2_to_tbl4_incompatible_shuffle(<16 x i8> %a, <16 x i8> %b, <16 x i8> %c, <16 x i8> %d) {
132+
; CHECK-LABEL: shuffled_tbl2_to_tbl4_incompatible_shuffle:
133+
; CHECK: // %bb.0:
134+
; CHECK-NEXT: adrp x8, .LCPI9_0
135+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q0_q1 def $q0_q1
136+
; CHECK-NEXT: // kill: def $q3 killed $q3 killed $q2_q3 def $q2_q3
137+
; CHECK-NEXT: // kill: def $q0 killed $q0 killed $q0_q1 def $q0_q1
138+
; CHECK-NEXT: // kill: def $q2 killed $q2 killed $q2_q3 def $q2_q3
139+
; CHECK-NEXT: ldr q4, [x8, :lo12:.LCPI9_0]
140+
; CHECK-NEXT: adrp x8, .LCPI9_1
141+
; CHECK-NEXT: tbl.16b v0, { v0, v1 }, v4
142+
; CHECK-NEXT: tbl.16b v1, { v2, v3 }, v4
143+
; CHECK-NEXT: ldr q2, [x8, :lo12:.LCPI9_1]
144+
; CHECK-NEXT: tbl.16b v0, { v0, v1 }, v2
145+
; CHECK-NEXT: ret
146+
%t1 = call <16 x i8> @llvm.aarch64.neon.tbl2.v16i8(<16 x i8> %a, <16 x i8> %b, <16 x i8> <i8 0, i8 4, i8 8, i8 12, i8 16, i8 20, i8 24, i8 28, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>)
147+
%t2 = call <16 x i8> @llvm.aarch64.neon.tbl2.v16i8(<16 x i8> %c, <16 x i8> %d, <16 x i8> <i8 0, i8 4, i8 8, i8 12, i8 16, i8 20, i8 24, i8 28, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>)
148+
%s = shufflevector <16 x i8> %t1, <16 x i8> %t2, <16 x i32> <i32 0, i32 1, i32 21, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
149+
ret <16 x i8> %s
150+
}
151+
152+
define <16 x i8> @shuffled_tbl2_to_tbl4_incompatible_tbl2_mask1(<16 x i8> %a, <16 x i8> %b, <16 x i8> %c, <16 x i8> %d) {
153+
; CHECK-LABEL: shuffled_tbl2_to_tbl4_incompatible_tbl2_mask1:
154+
; CHECK: // %bb.0:
155+
; CHECK-NEXT: adrp x8, .LCPI10_0
156+
; CHECK-NEXT: adrp x9, .LCPI10_1
157+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q0_q1 def $q0_q1
158+
; CHECK-NEXT: // kill: def $q3 killed $q3 killed $q2_q3 def $q2_q3
159+
; CHECK-NEXT: // kill: def $q0 killed $q0 killed $q0_q1 def $q0_q1
160+
; CHECK-NEXT: // kill: def $q2 killed $q2 killed $q2_q3 def $q2_q3
161+
; CHECK-NEXT: ldr q4, [x8, :lo12:.LCPI10_0]
162+
; CHECK-NEXT: adrp x8, .LCPI10_2
163+
; CHECK-NEXT: ldr q5, [x9, :lo12:.LCPI10_1]
164+
; CHECK-NEXT: tbl.16b v0, { v0, v1 }, v4
165+
; CHECK-NEXT: tbl.16b v1, { v2, v3 }, v5
166+
; CHECK-NEXT: ldr q2, [x8, :lo12:.LCPI10_2]
167+
; CHECK-NEXT: tbl.16b v0, { v0, v1 }, v2
168+
; CHECK-NEXT: ret
169+
%t1 = call <16 x i8> @llvm.aarch64.neon.tbl2.v16i8(<16 x i8> %a, <16 x i8> %b, <16 x i8> <i8 0, i8 4, i8 8, i8 12, i8 16, i8 20, i8 24, i8 28, i8 0, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>)
170+
%t2 = call <16 x i8> @llvm.aarch64.neon.tbl2.v16i8(<16 x i8> %c, <16 x i8> %d, <16 x i8> <i8 0, i8 4, i8 8, i8 12, i8 16, i8 20, i8 24, i8 28, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>)
171+
%s = shufflevector <16 x i8> %t1, <16 x i8> %t2, <16 x i32> <i32 0, i32 1, i32 21, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
172+
ret <16 x i8> %s
173+
}
174+
175+
define <16 x i8> @shuffled_tbl2_to_tbl4_incompatible_tbl2_mask2(<16 x i8> %a, <16 x i8> %b, <16 x i8> %c, <16 x i8> %d) {
176+
; CHECK-LABEL: shuffled_tbl2_to_tbl4_incompatible_tbl2_mask2:
177+
; CHECK: // %bb.0:
178+
; CHECK-NEXT: adrp x8, .LCPI11_0
179+
; CHECK-NEXT: adrp x9, .LCPI11_1
180+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q0_q1 def $q0_q1
181+
; CHECK-NEXT: // kill: def $q3 killed $q3 killed $q2_q3 def $q2_q3
182+
; CHECK-NEXT: // kill: def $q0 killed $q0 killed $q0_q1 def $q0_q1
183+
; CHECK-NEXT: // kill: def $q2 killed $q2 killed $q2_q3 def $q2_q3
184+
; CHECK-NEXT: ldr q4, [x8, :lo12:.LCPI11_0]
185+
; CHECK-NEXT: adrp x8, .LCPI11_2
186+
; CHECK-NEXT: ldr q5, [x9, :lo12:.LCPI11_1]
187+
; CHECK-NEXT: tbl.16b v0, { v0, v1 }, v4
188+
; CHECK-NEXT: tbl.16b v1, { v2, v3 }, v5
189+
; CHECK-NEXT: ldr q2, [x8, :lo12:.LCPI11_2]
190+
; CHECK-NEXT: tbl.16b v0, { v0, v1 }, v2
191+
; CHECK-NEXT: ret
192+
%t1 = call <16 x i8> @llvm.aarch64.neon.tbl2.v16i8(<16 x i8> %a, <16 x i8> %b, <16 x i8> <i8 0, i8 4, i8 8, i8 12, i8 16, i8 20, i8 24, i8 28, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>)
193+
%t2 = call <16 x i8> @llvm.aarch64.neon.tbl2.v16i8(<16 x i8> %c, <16 x i8> %d, <16 x i8> <i8 0, i8 4, i8 8, i8 12, i8 16, i8 20, i8 24, i8 28, i8 0, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>)
194+
%s = shufflevector <16 x i8> %t1, <16 x i8> %t2, <16 x i32> <i32 0, i32 1, i32 21, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
195+
ret <16 x i8> %s
196+
}
197+
59198
declare <8 x i8> @llvm.aarch64.neon.tbl1.v8i8(<16 x i8>, <8 x i8>) nounwind readnone
60199
declare <16 x i8> @llvm.aarch64.neon.tbl1.v16i8(<16 x i8>, <16 x i8>) nounwind readnone
61200
declare <8 x i8> @llvm.aarch64.neon.tbl2.v8i8(<16 x i8>, <16 x i8>, <8 x i8>) nounwind readnone
@@ -66,57 +205,91 @@ declare <8 x i8> @llvm.aarch64.neon.tbl4.v8i8(<16 x i8>, <16 x i8>, <16 x i8>, <
66205
declare <16 x i8> @llvm.aarch64.neon.tbl4.v16i8(<16 x i8>, <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8>) nounwind readnone
67206

68207
define <8 x i8> @tbx1_8b(<8 x i8> %A, <16 x i8> %B, <8 x i8> %C) nounwind {
69-
; CHECK: tbx1_8b
70-
; CHECK: tbx.8b
208+
; CHECK-LABEL: tbx1_8b:
209+
; CHECK: // %bb.0:
210+
; CHECK-NEXT: tbx.8b v0, { v1 }, v2
211+
; CHECK-NEXT: ret
71212
%tmp3 = call <8 x i8> @llvm.aarch64.neon.tbx1.v8i8(<8 x i8> %A, <16 x i8> %B, <8 x i8> %C)
72213
ret <8 x i8> %tmp3
73214
}
74215

75216
define <16 x i8> @tbx1_16b(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C) nounwind {
76-
; CHECK: tbx1_16b
77-
; CHECK: tbx.16b
217+
; CHECK-LABEL: tbx1_16b:
218+
; CHECK: // %bb.0:
219+
; CHECK-NEXT: tbx.16b v0, { v1 }, v2
220+
; CHECK-NEXT: ret
78221
%tmp3 = call <16 x i8> @llvm.aarch64.neon.tbx1.v16i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C)
79222
ret <16 x i8> %tmp3
80223
}
81224

82225
define <8 x i8> @tbx2_8b(<8 x i8> %A, <16 x i8> %B, <16 x i8> %C, <8 x i8> %D) {
83-
; CHECK: tbx2_8b
84-
; CHECK: tbx.8b
226+
; CHECK-LABEL: tbx2_8b:
227+
; CHECK: // %bb.0:
228+
; CHECK-NEXT: // kill: def $q2 killed $q2 killed $q1_q2 def $q1_q2
229+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q1_q2 def $q1_q2
230+
; CHECK-NEXT: tbx.8b v0, { v1, v2 }, v3
231+
; CHECK-NEXT: ret
85232
%tmp3 = call <8 x i8> @llvm.aarch64.neon.tbx2.v8i8(<8 x i8> %A, <16 x i8> %B, <16 x i8> %C, <8 x i8> %D)
86233
ret <8 x i8> %tmp3
87234
}
88235

89236
define <16 x i8> @tbx2_16b(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D) {
90-
; CHECK: tbx2_16b
91-
; CHECK: tbx.16b
237+
; CHECK-LABEL: tbx2_16b:
238+
; CHECK: // %bb.0:
239+
; CHECK-NEXT: // kill: def $q2 killed $q2 killed $q1_q2 def $q1_q2
240+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q1_q2 def $q1_q2
241+
; CHECK-NEXT: tbx.16b v0, { v1, v2 }, v3
242+
; CHECK-NEXT: ret
92243
%tmp3 = call <16 x i8> @llvm.aarch64.neon.tbx2.v16i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D)
93244
ret <16 x i8> %tmp3
94245
}
95246

96247
define <8 x i8> @tbx3_8b(<8 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, <8 x i8> %E) {
97-
; CHECK: tbx3_8b
98-
; CHECK: tbx.8b
248+
; CHECK-LABEL: tbx3_8b:
249+
; CHECK: // %bb.0:
250+
; CHECK-NEXT: // kill: def $q3 killed $q3 killed $q1_q2_q3 def $q1_q2_q3
251+
; CHECK-NEXT: // kill: def $q2 killed $q2 killed $q1_q2_q3 def $q1_q2_q3
252+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q1_q2_q3 def $q1_q2_q3
253+
; CHECK-NEXT: tbx.8b v0, { v1, v2, v3 }, v4
254+
; CHECK-NEXT: ret
99255
%tmp3 = call <8 x i8> @llvm.aarch64.neon.tbx3.v8i8(< 8 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, <8 x i8> %E)
100256
ret <8 x i8> %tmp3
101257
}
102258

103259
define <16 x i8> @tbx3_16b(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, <16 x i8> %E) {
104-
; CHECK: tbx3_16b
105-
; CHECK: tbx.16b
260+
; CHECK-LABEL: tbx3_16b:
261+
; CHECK: // %bb.0:
262+
; CHECK-NEXT: // kill: def $q3 killed $q3 killed $q1_q2_q3 def $q1_q2_q3
263+
; CHECK-NEXT: // kill: def $q2 killed $q2 killed $q1_q2_q3 def $q1_q2_q3
264+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q1_q2_q3 def $q1_q2_q3
265+
; CHECK-NEXT: tbx.16b v0, { v1, v2, v3 }, v4
266+
; CHECK-NEXT: ret
106267
%tmp3 = call <16 x i8> @llvm.aarch64.neon.tbx3.v16i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, <16 x i8> %E)
107268
ret <16 x i8> %tmp3
108269
}
109270

110271
define <8 x i8> @tbx4_8b(<8 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, <16 x i8> %E, <8 x i8> %F) {
111-
; CHECK: tbx4_8b
112-
; CHECK: tbx.8b
272+
; CHECK-LABEL: tbx4_8b:
273+
; CHECK: // %bb.0:
274+
; CHECK-NEXT: // kill: def $q4 killed $q4 killed $q1_q2_q3_q4 def $q1_q2_q3_q4
275+
; CHECK-NEXT: // kill: def $q3 killed $q3 killed $q1_q2_q3_q4 def $q1_q2_q3_q4
276+
; CHECK-NEXT: // kill: def $q2 killed $q2 killed $q1_q2_q3_q4 def $q1_q2_q3_q4
277+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q1_q2_q3_q4 def $q1_q2_q3_q4
278+
; CHECK-NEXT: tbx.8b v0, { v1, v2, v3, v4 }, v5
279+
; CHECK-NEXT: ret
113280
%tmp3 = call <8 x i8> @llvm.aarch64.neon.tbx4.v8i8(<8 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, <16 x i8> %E, <8 x i8> %F)
114281
ret <8 x i8> %tmp3
115282
}
116283

117284
define <16 x i8> @tbx4_16b(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, <16 x i8> %E, <16 x i8> %F) {
118-
; CHECK: tbx4_16b
119-
; CHECK: tbx.16b
285+
; CHECK-LABEL: tbx4_16b:
286+
; CHECK: // %bb.0:
287+
; CHECK-NEXT: // kill: def $q4 killed $q4 killed $q1_q2_q3_q4 def $q1_q2_q3_q4
288+
; CHECK-NEXT: // kill: def $q3 killed $q3 killed $q1_q2_q3_q4 def $q1_q2_q3_q4
289+
; CHECK-NEXT: // kill: def $q2 killed $q2 killed $q1_q2_q3_q4 def $q1_q2_q3_q4
290+
; CHECK-NEXT: // kill: def $q1 killed $q1 killed $q1_q2_q3_q4 def $q1_q2_q3_q4
291+
; CHECK-NEXT: tbx.16b v0, { v1, v2, v3, v4 }, v5
292+
; CHECK-NEXT: ret
120293
%tmp3 = call <16 x i8> @llvm.aarch64.neon.tbx4.v16i8(<16 x i8> %A, <16 x i8> %B, <16 x i8> %C, <16 x i8> %D, <16 x i8> %E, <16 x i8> %F)
121294
ret <16 x i8> %tmp3
122295
}

0 commit comments

Comments
 (0)