• Home
  • Raw
  • Download

Lines Matching refs:SSE2

2 …c < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X86 --check-prefix=X86-SSE2
4 …< %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X64 --check-prefix=X64-SSE2
10 ; X86-SSE2-LABEL: convert_v2i16_to_v2f32:
11 ; X86-SSE2: # BB#0: # %entry
12 ; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %eax
13 ; X86-SSE2-NEXT: psllq $48, %xmm0
14 ; X86-SSE2-NEXT: psrad $16, %xmm0
15 ; X86-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
16 ; X86-SSE2-NEXT: cvtdq2ps %xmm0, %xmm0
17 ; X86-SSE2-NEXT: movss %xmm0, (%eax)
18 ; X86-SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,2,3]
19 ; X86-SSE2-NEXT: movss %xmm0, 4(%eax)
20 ; X86-SSE2-NEXT: retl
50 ; X86-SSE2-LABEL: convert_v3i8_to_v3f32:
51 ; X86-SSE2: # BB#0: # %entry
52 ; X86-SSE2-NEXT: pushl %ebp
53 ; X86-SSE2-NEXT: movl %esp, %ebp
54 ; X86-SSE2-NEXT: pushl %esi
55 ; X86-SSE2-NEXT: andl $-16, %esp
56 ; X86-SSE2-NEXT: subl $32, %esp
57 ; X86-SSE2-NEXT: movl 8(%ebp), %eax
58 ; X86-SSE2-NEXT: movl 12(%ebp), %ecx
59 ; X86-SSE2-NEXT: movzwl (%ecx), %edx
60 ; X86-SSE2-NEXT: movd %edx, %xmm0
61 ; X86-SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
62 ; X86-SSE2-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
63 ; X86-SSE2-NEXT: movdqa %xmm0, (%esp)
64 ; X86-SSE2-NEXT: movl {{[0-9]+}}(%esp), %edx
65 ; X86-SSE2-NEXT: shll $8, %edx
66 ; X86-SSE2-NEXT: movzbl (%esp), %esi
67 ; X86-SSE2-NEXT: orl %edx, %esi
68 ; X86-SSE2-NEXT: pinsrw $0, %esi, %xmm0
69 ; X86-SSE2-NEXT: movzbl 2(%ecx), %ecx
70 ; X86-SSE2-NEXT: pinsrw $1, %ecx, %xmm0
71 ; X86-SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
72 ; X86-SSE2-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
73 ; X86-SSE2-NEXT: psrad $24, %xmm0
74 ; X86-SSE2-NEXT: cvtdq2ps %xmm0, %xmm0
75 ; X86-SSE2-NEXT: movss %xmm0, (%eax)
76 ; X86-SSE2-NEXT: movaps %xmm0, %xmm1
77 ; X86-SSE2-NEXT: shufpd {{.*#+}} xmm1 = xmm1[1,0]
78 ; X86-SSE2-NEXT: movss %xmm1, 8(%eax)
79 ; X86-SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,2,3]
80 ; X86-SSE2-NEXT: movss %xmm0, 4(%eax)
81 ; X86-SSE2-NEXT: leal -4(%ebp), %esp
82 ; X86-SSE2-NEXT: popl %esi
83 ; X86-SSE2-NEXT: popl %ebp
84 ; X86-SSE2-NEXT: retl
105 ; X64-SSE2-LABEL: convert_v3i8_to_v3f32:
106 ; X64-SSE2: # BB#0: # %entry
107 ; X64-SSE2-NEXT: movzwl (%rsi), %eax
108 ; X64-SSE2-NEXT: movd %rax, %xmm0
109 ; X64-SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
110 ; X64-SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7]
111 ; X64-SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
112 ; X64-SSE2-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
113 ; X64-SSE2-NEXT: movdqa %xmm0, -{{[0-9]+}}(%rsp)
114 ; X64-SSE2-NEXT: movl -{{[0-9]+}}(%rsp), %eax
115 ; X64-SSE2-NEXT: shll $8, %eax
116 ; X64-SSE2-NEXT: movzbl -{{[0-9]+}}(%rsp), %ecx
117 ; X64-SSE2-NEXT: orl %eax, %ecx
118 ; X64-SSE2-NEXT: pinsrw $0, %ecx, %xmm0
119 ; X64-SSE2-NEXT: movzbl 2(%rsi), %eax
120 ; X64-SSE2-NEXT: pinsrw $1, %eax, %xmm0
121 ; X64-SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
122 ; X64-SSE2-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
123 ; X64-SSE2-NEXT: psrad $24, %xmm0
124 ; X64-SSE2-NEXT: cvtdq2ps %xmm0, %xmm0
125 ; X64-SSE2-NEXT: movlps %xmm0, (%rdi)
126 ; X64-SSE2-NEXT: shufpd {{.*#+}} xmm0 = xmm0[1,0]
127 ; X64-SSE2-NEXT: movss %xmm0, 8(%rdi)
128 ; X64-SSE2-NEXT: retq