1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=X32
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=X64
; This test makes sure that the compiler does not crash with an
; assertion failure when trying to fold a vector shift left
; by immediate count if the type of the input vector is different
; to the result type.
;
; This happens for example when lowering a shift left of a MVT::v16i8 vector.
; This is custom lowered into the following sequence:
; count << 5
; A = VSHLI(MVT::v8i16, r & (char16)15, 4)
; B = BITCAST MVT::v16i8, A
; VSELECT(r, B, count);
; count += count
; C = VSHLI(MVT::v8i16, r & (char16)63, 2)
; D = BITCAST MVT::v16i8, C
; r = VSELECT(r, C, count);
; count += count
; VSELECT(r, r+r, count);
; count = count << 5;
;
; Where 'r' is a vector of type MVT::v16i8, and
; 'count' is the vector shift count.
define <16 x i8> @do_not_crash(i8*, i32*, i64*, i32, i64, i8) {
; X32-LABEL: do_not_crash:
; X32: # %bb.0: # %entry
; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx
; X32-NEXT: movb %al, (%ecx)
; X32-NEXT: movd %eax, %xmm0
; X32-NEXT: psllq $56, %xmm0
; X32-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,255]
; X32-NEXT: movdqa %xmm2, %xmm1
; X32-NEXT: pandn %xmm0, %xmm1
; X32-NEXT: por %xmm2, %xmm1
; X32-NEXT: pcmpeqd %xmm2, %xmm2
; X32-NEXT: psllw $5, %xmm1
; X32-NEXT: pxor %xmm3, %xmm3
; X32-NEXT: pxor %xmm0, %xmm0
; X32-NEXT: pcmpgtb %xmm1, %xmm0
; X32-NEXT: pxor %xmm0, %xmm2
; X32-NEXT: pand {{\.LCPI.*}}, %xmm0
; X32-NEXT: por %xmm2, %xmm0
; X32-NEXT: paddb %xmm1, %xmm1
; X32-NEXT: pxor %xmm2, %xmm2
; X32-NEXT: pcmpgtb %xmm1, %xmm2
; X32-NEXT: movdqa %xmm2, %xmm4
; X32-NEXT: pandn %xmm0, %xmm4
; X32-NEXT: psllw $2, %xmm0
; X32-NEXT: pand %xmm2, %xmm0
; X32-NEXT: pand {{\.LCPI.*}}, %xmm0
; X32-NEXT: por %xmm4, %xmm0
; X32-NEXT: paddb %xmm1, %xmm1
; X32-NEXT: pcmpgtb %xmm1, %xmm3
; X32-NEXT: movdqa %xmm3, %xmm1
; X32-NEXT: pandn %xmm0, %xmm1
; X32-NEXT: paddb %xmm0, %xmm0
; X32-NEXT: pand %xmm3, %xmm0
; X32-NEXT: por %xmm1, %xmm0
; X32-NEXT: retl
;
; X64-LABEL: do_not_crash:
; X64: # %bb.0: # %entry
; X64-NEXT: movb %r9b, (%rdi)
; X64-NEXT: movd %r9d, %xmm0
; X64-NEXT: psllq $56, %xmm0
; X64-NEXT: movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,255]
; X64-NEXT: movdqa %xmm2, %xmm1
; X64-NEXT: pandn %xmm0, %xmm1
; X64-NEXT: por %xmm2, %xmm1
; X64-NEXT: pcmpeqd %xmm2, %xmm2
; X64-NEXT: psllw $5, %xmm1
; X64-NEXT: pxor %xmm3, %xmm3
; X64-NEXT: pxor %xmm0, %xmm0
; X64-NEXT: pcmpgtb %xmm1, %xmm0
; X64-NEXT: pxor %xmm0, %xmm2
; X64-NEXT: pand {{.*}}(%rip), %xmm0
; X64-NEXT: por %xmm2, %xmm0
; X64-NEXT: paddb %xmm1, %xmm1
; X64-NEXT: pxor %xmm2, %xmm2
; X64-NEXT: pcmpgtb %xmm1, %xmm2
; X64-NEXT: movdqa %xmm2, %xmm4
; X64-NEXT: pandn %xmm0, %xmm4
; X64-NEXT: psllw $2, %xmm0
; X64-NEXT: pand %xmm2, %xmm0
; X64-NEXT: pand {{.*}}(%rip), %xmm0
; X64-NEXT: por %xmm4, %xmm0
; X64-NEXT: paddb %xmm1, %xmm1
; X64-NEXT: pcmpgtb %xmm1, %xmm3
; X64-NEXT: movdqa %xmm3, %xmm1
; X64-NEXT: pandn %xmm0, %xmm1
; X64-NEXT: paddb %xmm0, %xmm0
; X64-NEXT: pand %xmm3, %xmm0
; X64-NEXT: por %xmm1, %xmm0
; X64-NEXT: retq
entry:
store i8 %5, i8* %0
%L5 = load i8, i8* %0
%I8 = insertelement <16 x i8> <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>, i8 %L5, i32 7
%B51 = shl <16 x i8> <i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>, %I8
ret <16 x i8> %B51
}
|