• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X86
3; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X64
4
5; This test makes sure that the compiler does not crash with an
6; assertion failure when trying to fold a vector shift left
7; by immediate count if the type of the input vector is different
8; to the result type.
9;
10; This happens for example when lowering a shift left of a MVT::v16i8 vector.
11; This is custom lowered into the following sequence:
12;     count << 5
13;     A =  VSHLI(MVT::v8i16, r & (char16)15, 4)
14;     B = BITCAST MVT::v16i8, A
15;     VSELECT(r, B, count);
16;     count += count
17;     C = VSHLI(MVT::v8i16, r & (char16)63, 2)
18;     D = BITCAST MVT::v16i8, C
19;     r = VSELECT(r, C, count);
20;     count += count
21;     VSELECT(r, r+r, count);
22;     count = count << 5;
23;
24; Where 'r' is a vector of type MVT::v16i8, and
25; 'count' is the vector shift count.
26
27define <16 x i8> @do_not_crash(i8*, i32*, i64*, i32, i64, i8) {
28; X86-LABEL: do_not_crash:
29; X86:       # %bb.0: # %entry
30; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
31; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
32; X86-NEXT:    movb %al, (%ecx)
33; X86-NEXT:    movd %eax, %xmm0
34; X86-NEXT:    psllq $56, %xmm0
35; X86-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,255]
36; X86-NEXT:    movdqa %xmm2, %xmm1
37; X86-NEXT:    pandn %xmm0, %xmm1
38; X86-NEXT:    por %xmm2, %xmm1
39; X86-NEXT:    pcmpeqd %xmm2, %xmm2
40; X86-NEXT:    psllw $5, %xmm1
41; X86-NEXT:    pxor %xmm3, %xmm3
42; X86-NEXT:    pxor %xmm0, %xmm0
43; X86-NEXT:    pcmpgtb %xmm1, %xmm0
44; X86-NEXT:    pxor %xmm0, %xmm2
45; X86-NEXT:    pand {{\.LCPI.*}}, %xmm0
46; X86-NEXT:    por %xmm2, %xmm0
47; X86-NEXT:    paddb %xmm1, %xmm1
48; X86-NEXT:    pxor %xmm2, %xmm2
49; X86-NEXT:    pcmpgtb %xmm1, %xmm2
50; X86-NEXT:    movdqa %xmm2, %xmm4
51; X86-NEXT:    pandn %xmm0, %xmm4
52; X86-NEXT:    psllw $2, %xmm0
53; X86-NEXT:    pand %xmm2, %xmm0
54; X86-NEXT:    pand {{\.LCPI.*}}, %xmm0
55; X86-NEXT:    por %xmm4, %xmm0
56; X86-NEXT:    paddb %xmm1, %xmm1
57; X86-NEXT:    pcmpgtb %xmm1, %xmm3
58; X86-NEXT:    movdqa %xmm3, %xmm1
59; X86-NEXT:    pandn %xmm0, %xmm1
60; X86-NEXT:    paddb %xmm0, %xmm0
61; X86-NEXT:    pand %xmm3, %xmm0
62; X86-NEXT:    por %xmm1, %xmm0
63; X86-NEXT:    retl
64;
65; X64-LABEL: do_not_crash:
66; X64:       # %bb.0: # %entry
67; X64-NEXT:    movb %r9b, (%rdi)
68; X64-NEXT:    movd %r9d, %xmm0
69; X64-NEXT:    psllq $56, %xmm0
70; X64-NEXT:    movdqa {{.*#+}} xmm2 = [255,255,255,255,255,255,255,0,255,255,255,255,255,255,255,255]
71; X64-NEXT:    movdqa %xmm2, %xmm1
72; X64-NEXT:    pandn %xmm0, %xmm1
73; X64-NEXT:    por %xmm2, %xmm1
74; X64-NEXT:    pcmpeqd %xmm2, %xmm2
75; X64-NEXT:    psllw $5, %xmm1
76; X64-NEXT:    pxor %xmm3, %xmm3
77; X64-NEXT:    pxor %xmm0, %xmm0
78; X64-NEXT:    pcmpgtb %xmm1, %xmm0
79; X64-NEXT:    pxor %xmm0, %xmm2
80; X64-NEXT:    pand {{.*}}(%rip), %xmm0
81; X64-NEXT:    por %xmm2, %xmm0
82; X64-NEXT:    paddb %xmm1, %xmm1
83; X64-NEXT:    pxor %xmm2, %xmm2
84; X64-NEXT:    pcmpgtb %xmm1, %xmm2
85; X64-NEXT:    movdqa %xmm2, %xmm4
86; X64-NEXT:    pandn %xmm0, %xmm4
87; X64-NEXT:    psllw $2, %xmm0
88; X64-NEXT:    pand %xmm2, %xmm0
89; X64-NEXT:    pand {{.*}}(%rip), %xmm0
90; X64-NEXT:    por %xmm4, %xmm0
91; X64-NEXT:    paddb %xmm1, %xmm1
92; X64-NEXT:    pcmpgtb %xmm1, %xmm3
93; X64-NEXT:    movdqa %xmm3, %xmm1
94; X64-NEXT:    pandn %xmm0, %xmm1
95; X64-NEXT:    paddb %xmm0, %xmm0
96; X64-NEXT:    pand %xmm3, %xmm0
97; X64-NEXT:    por %xmm1, %xmm0
98; X64-NEXT:    retq
99entry:
100  store i8 %5, i8* %0
101  %L5 = load i8, i8* %0
102  %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
103  %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
104  ret <16 x i8> %B51
105}
106