1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+sse4.2 | FileCheck %s --check-prefix=NARROW
3 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+sse4.2 -x86-experimental-vector-widening-legalization | FileCheck %s --check-prefix=WIDE
5 ; FIXME: We shouldn't require both a movd and an insert in the wide version.
7 define void @update(i64* %dst_i, i64* %src_i, i32 %n) nounwind {
8 ; NARROW-LABEL: update:
9 ; NARROW: # %bb.0: # %entry
10 ; NARROW-NEXT: subl $12, %esp
11 ; NARROW-NEXT: movl $0, (%esp)
12 ; NARROW-NEXT: pcmpeqd %xmm0, %xmm0
13 ; NARROW-NEXT: movdqa {{.*#+}} xmm1 = <0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u>
14 ; NARROW-NEXT: jmp .LBB0_1
15 ; NARROW-NEXT: .p2align 4, 0x90
16 ; NARROW-NEXT: .LBB0_2: # %forbody
17 ; NARROW-NEXT: # in Loop: Header=BB0_1 Depth=1
18 ; NARROW-NEXT: movl (%esp), %eax
19 ; NARROW-NEXT: leal (,%eax,8), %ecx
20 ; NARROW-NEXT: movl {{[0-9]+}}(%esp), %edx
21 ; NARROW-NEXT: addl %ecx, %edx
22 ; NARROW-NEXT: movl %edx, {{[0-9]+}}(%esp)
23 ; NARROW-NEXT: addl {{[0-9]+}}(%esp), %ecx
24 ; NARROW-NEXT: movl %ecx, {{[0-9]+}}(%esp)
25 ; NARROW-NEXT: pmovzxbw {{.*#+}} xmm2 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero,mem[4],zero,mem[5],zero,mem[6],zero,mem[7],zero
26 ; NARROW-NEXT: psubw %xmm0, %xmm2
27 ; NARROW-NEXT: psllw $8, %xmm2
28 ; NARROW-NEXT: psraw $8, %xmm2
29 ; NARROW-NEXT: psrlw $2, %xmm2
30 ; NARROW-NEXT: pshufb %xmm1, %xmm2
31 ; NARROW-NEXT: movq %xmm2, (%edx,%eax,8)
32 ; NARROW-NEXT: incl (%esp)
33 ; NARROW-NEXT: .LBB0_1: # %forcond
34 ; NARROW-NEXT: # =>This Inner Loop Header: Depth=1
35 ; NARROW-NEXT: movl (%esp), %eax
36 ; NARROW-NEXT: cmpl {{[0-9]+}}(%esp), %eax
37 ; NARROW-NEXT: jl .LBB0_2
38 ; NARROW-NEXT: # %bb.3: # %afterfor
39 ; NARROW-NEXT: addl $12, %esp
43 ; WIDE: # %bb.0: # %entry
44 ; WIDE-NEXT: subl $12, %esp
45 ; WIDE-NEXT: movl $0, (%esp)
46 ; WIDE-NEXT: pcmpeqd %xmm0, %xmm0
47 ; WIDE-NEXT: movdqa {{.*#+}} xmm1 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
48 ; WIDE-NEXT: movdqa {{.*#+}} xmm2 = [32,32,32,32,32,32,32,32,32,32,32,32,32,32,32,32]
49 ; WIDE-NEXT: jmp .LBB0_1
50 ; WIDE-NEXT: .p2align 4, 0x90
51 ; WIDE-NEXT: .LBB0_2: # %forbody
52 ; WIDE-NEXT: # in Loop: Header=BB0_1 Depth=1
53 ; WIDE-NEXT: movl (%esp), %eax
54 ; WIDE-NEXT: leal (,%eax,8), %ecx
55 ; WIDE-NEXT: movl {{[0-9]+}}(%esp), %edx
56 ; WIDE-NEXT: addl %ecx, %edx
57 ; WIDE-NEXT: movl %edx, {{[0-9]+}}(%esp)
58 ; WIDE-NEXT: addl {{[0-9]+}}(%esp), %ecx
59 ; WIDE-NEXT: movl %ecx, {{[0-9]+}}(%esp)
60 ; WIDE-NEXT: movq {{.*#+}} xmm3 = mem[0],zero
61 ; WIDE-NEXT: psubb %xmm0, %xmm3
62 ; WIDE-NEXT: psrlw $2, %xmm3
63 ; WIDE-NEXT: pand %xmm1, %xmm3
64 ; WIDE-NEXT: pxor %xmm2, %xmm3
65 ; WIDE-NEXT: psubb %xmm2, %xmm3
66 ; WIDE-NEXT: movq %xmm3, (%edx,%eax,8)
67 ; WIDE-NEXT: incl (%esp)
68 ; WIDE-NEXT: .LBB0_1: # %forcond
69 ; WIDE-NEXT: # =>This Inner Loop Header: Depth=1
70 ; WIDE-NEXT: movl (%esp), %eax
71 ; WIDE-NEXT: cmpl {{[0-9]+}}(%esp), %eax
72 ; WIDE-NEXT: jl .LBB0_2
73 ; WIDE-NEXT: # %bb.3: # %afterfor
74 ; WIDE-NEXT: addl $12, %esp
77 %dst_i.addr = alloca i64*
78 %src_i.addr = alloca i64*
80 %i = alloca i32, align 4
81 %dst = alloca <8 x i8>*, align 4
82 %src = alloca <8 x i8>*, align 4
83 store i64* %dst_i, i64** %dst_i.addr
84 store i64* %src_i, i64** %src_i.addr
85 store i32 %n, i32* %n.addr
90 %tmp = load i32, i32* %i
91 %tmp1 = load i32, i32* %n.addr
92 %cmp = icmp slt i32 %tmp, %tmp1
93 br i1 %cmp, label %forbody, label %afterfor
96 %tmp2 = load i32, i32* %i
97 %tmp3 = load i64*, i64** %dst_i.addr
98 %arrayidx = getelementptr i64, i64* %tmp3, i32 %tmp2
99 %conv = bitcast i64* %arrayidx to <8 x i8>*
100 store <8 x i8>* %conv, <8 x i8>** %dst
101 %tmp4 = load i32, i32* %i
102 %tmp5 = load i64*, i64** %src_i.addr
103 %arrayidx6 = getelementptr i64, i64* %tmp5, i32 %tmp4
104 %conv7 = bitcast i64* %arrayidx6 to <8 x i8>*
105 store <8 x i8>* %conv7, <8 x i8>** %src
106 %tmp8 = load i32, i32* %i
107 %tmp9 = load <8 x i8>*, <8 x i8>** %dst
108 %arrayidx10 = getelementptr <8 x i8>, <8 x i8>* %tmp9, i32 %tmp8
109 %tmp11 = load i32, i32* %i
110 %tmp12 = load <8 x i8>*, <8 x i8>** %src
111 %arrayidx13 = getelementptr <8 x i8>, <8 x i8>* %tmp12, i32 %tmp11
112 %tmp14 = load <8 x i8>, <8 x i8>* %arrayidx13
113 %add = add <8 x i8> %tmp14, < i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1 >
114 %shr = ashr <8 x i8> %add, < i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2 >
115 store <8 x i8> %shr, <8 x i8>* %arrayidx10
119 %tmp15 = load i32, i32* %i
120 %inc = add i32 %tmp15, 1
121 store i32 %inc, i32* %i