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 4 5 ; FIXME: We shouldn't require both a movd and an insert in the wide version. 6 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 40 ; NARROW-NEXT: retl 41 ; 42 ; WIDE-LABEL: update: 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: movd {{.*#+}} xmm3 = mem[0],zero,zero,zero 61 ; WIDE-NEXT: pinsrd $1, 4(%ecx,%eax,8), %xmm3 62 ; WIDE-NEXT: psubb %xmm0, %xmm3 63 ; WIDE-NEXT: psrlw $2, %xmm3 64 ; WIDE-NEXT: pand %xmm1, %xmm3 65 ; WIDE-NEXT: pxor %xmm2, %xmm3 66 ; WIDE-NEXT: psubb %xmm2, %xmm3 67 ; WIDE-NEXT: pextrd $1, %xmm3, 4(%edx,%eax,8) 68 ; WIDE-NEXT: movd %xmm3, (%edx,%eax,8) 69 ; WIDE-NEXT: incl (%esp) 70 ; WIDE-NEXT: .LBB0_1: # %forcond 71 ; WIDE-NEXT: # =>This Inner Loop Header: Depth=1 72 ; WIDE-NEXT: movl (%esp), %eax 73 ; WIDE-NEXT: cmpl {{[0-9]+}}(%esp), %eax 74 ; WIDE-NEXT: jl .LBB0_2 75 ; WIDE-NEXT: # %bb.3: # %afterfor 76 ; WIDE-NEXT: addl $12, %esp 77 ; WIDE-NEXT: retl 78 entry: 79 %dst_i.addr = alloca i64* 80 %src_i.addr = alloca i64* 81 %n.addr = alloca i32 82 %i = alloca i32, align 4 83 %dst = alloca <8 x i8>*, align 4 84 %src = alloca <8 x i8>*, align 4 85 store i64* %dst_i, i64** %dst_i.addr 86 store i64* %src_i, i64** %src_i.addr 87 store i32 %n, i32* %n.addr 88 store i32 0, i32* %i 89 br label %forcond 90 91 forcond: 92 %tmp = load i32, i32* %i 93 %tmp1 = load i32, i32* %n.addr 94 %cmp = icmp slt i32 %tmp, %tmp1 95 br i1 %cmp, label %forbody, label %afterfor 96 97 forbody: 98 %tmp2 = load i32, i32* %i 99 %tmp3 = load i64*, i64** %dst_i.addr 100 %arrayidx = getelementptr i64, i64* %tmp3, i32 %tmp2 101 %conv = bitcast i64* %arrayidx to <8 x i8>* 102 store <8 x i8>* %conv, <8 x i8>** %dst 103 %tmp4 = load i32, i32* %i 104 %tmp5 = load i64*, i64** %src_i.addr 105 %arrayidx6 = getelementptr i64, i64* %tmp5, i32 %tmp4 106 %conv7 = bitcast i64* %arrayidx6 to <8 x i8>* 107 store <8 x i8>* %conv7, <8 x i8>** %src 108 %tmp8 = load i32, i32* %i 109 %tmp9 = load <8 x i8>*, <8 x i8>** %dst 110 %arrayidx10 = getelementptr <8 x i8>, <8 x i8>* %tmp9, i32 %tmp8 111 %tmp11 = load i32, i32* %i 112 %tmp12 = load <8 x i8>*, <8 x i8>** %src 113 %arrayidx13 = getelementptr <8 x i8>, <8 x i8>* %tmp12, i32 %tmp11 114 %tmp14 = load <8 x i8>, <8 x i8>* %arrayidx13 115 %add = add <8 x i8> %tmp14, < i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1 > 116 %shr = ashr <8 x i8> %add, < i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2 > 117 store <8 x i8> %shr, <8 x i8>* %arrayidx10 118 br label %forinc 119 120 forinc: 121 %tmp15 = load i32, i32* %i 122 %inc = add i32 %tmp15, 1 123 store i32 %inc, i32* %i 124 br label %forcond 125 126 afterfor: 127 ret void 128 } 129 130
