1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=i686-unknown -mattr=+mmx,+sse2 | FileCheck %s --check-prefix=X86 3; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+mmx,+sse2 | FileCheck %s --check-prefix=X64 4 5define i32 @t0(i64 %x) nounwind { 6; X86-LABEL: t0: 7; X86: # %bb.0: # %entry 8; X86-NEXT: pshufw $238, {{[0-9]+}}(%esp), %mm0 # mm0 = mem[2,3,2,3] 9; X86-NEXT: movd %mm0, %eax 10; X86-NEXT: retl 11; 12; X64-LABEL: t0: 13; X64: # %bb.0: # %entry 14; X64-NEXT: movq %rdi, %mm0 15; X64-NEXT: pshufw $238, %mm0, %mm0 # mm0 = mm0[2,3,2,3] 16; X64-NEXT: movd %mm0, %eax 17; X64-NEXT: retq 18entry: 19 %0 = bitcast i64 %x to <4 x i16> 20 %1 = bitcast <4 x i16> %0 to x86_mmx 21 %2 = tail call x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx %1, i8 -18) 22 %3 = bitcast x86_mmx %2 to <4 x i16> 23 %4 = bitcast <4 x i16> %3 to <1 x i64> 24 %5 = extractelement <1 x i64> %4, i32 0 25 %6 = bitcast i64 %5 to <2 x i32> 26 %7 = extractelement <2 x i32> %6, i32 0 27 ret i32 %7 28} 29 30define i64 @t1(i64 %x, i32 %n) nounwind { 31; X86-LABEL: t1: 32; X86: # %bb.0: # %entry 33; X86-NEXT: pushl %ebp 34; X86-NEXT: movl %esp, %ebp 35; X86-NEXT: andl $-8, %esp 36; X86-NEXT: subl $8, %esp 37; X86-NEXT: movd 16(%ebp), %mm0 38; X86-NEXT: movq 8(%ebp), %mm1 39; X86-NEXT: psllq %mm0, %mm1 40; X86-NEXT: movq %mm1, (%esp) 41; X86-NEXT: movl (%esp), %eax 42; X86-NEXT: movl {{[0-9]+}}(%esp), %edx 43; X86-NEXT: movl %ebp, %esp 44; X86-NEXT: popl %ebp 45; X86-NEXT: retl 46; 47; X64-LABEL: t1: 48; X64: # %bb.0: # %entry 49; X64-NEXT: movd %esi, %mm0 50; X64-NEXT: movq %rdi, %mm1 51; X64-NEXT: psllq %mm0, %mm1 52; X64-NEXT: movq %mm1, %rax 53; X64-NEXT: retq 54entry: 55 %0 = bitcast i64 %x to x86_mmx 56 %1 = tail call x86_mmx @llvm.x86.mmx.pslli.q(x86_mmx %0, i32 %n) 57 %2 = bitcast x86_mmx %1 to i64 58 ret i64 %2 59} 60 61define i64 @t2(i64 %x, i32 %n, i32 %w) nounwind { 62; X86-LABEL: t2: 63; X86: # %bb.0: # %entry 64; X86-NEXT: pushl %ebp 65; X86-NEXT: movl %esp, %ebp 66; X86-NEXT: andl $-8, %esp 67; X86-NEXT: subl $8, %esp 68; X86-NEXT: movd 16(%ebp), %mm0 69; X86-NEXT: movd 20(%ebp), %mm1 70; X86-NEXT: psllq %mm0, %mm1 71; X86-NEXT: por 8(%ebp), %mm1 72; X86-NEXT: movq %mm1, (%esp) 73; X86-NEXT: movl (%esp), %eax 74; X86-NEXT: movl {{[0-9]+}}(%esp), %edx 75; X86-NEXT: movl %ebp, %esp 76; X86-NEXT: popl %ebp 77; X86-NEXT: retl 78; 79; X64-LABEL: t2: 80; X64: # %bb.0: # %entry 81; X64-NEXT: movd %esi, %mm0 82; X64-NEXT: movd %edx, %mm1 83; X64-NEXT: psllq %mm0, %mm1 84; X64-NEXT: movq %rdi, %mm0 85; X64-NEXT: por %mm1, %mm0 86; X64-NEXT: movq %mm0, %rax 87; X64-NEXT: retq 88entry: 89 %0 = insertelement <2 x i32> undef, i32 %w, i32 0 90 %1 = insertelement <2 x i32> %0, i32 0, i32 1 91 %2 = bitcast <2 x i32> %1 to x86_mmx 92 %3 = tail call x86_mmx @llvm.x86.mmx.pslli.q(x86_mmx %2, i32 %n) 93 %4 = bitcast i64 %x to x86_mmx 94 %5 = tail call x86_mmx @llvm.x86.mmx.por(x86_mmx %4, x86_mmx %3) 95 %6 = bitcast x86_mmx %5 to i64 96 ret i64 %6 97} 98 99define i64 @t3(<1 x i64>* %y, i32* %n) nounwind { 100; X86-LABEL: t3: 101; X86: # %bb.0: # %entry 102; X86-NEXT: pushl %ebp 103; X86-NEXT: movl %esp, %ebp 104; X86-NEXT: andl $-8, %esp 105; X86-NEXT: subl $8, %esp 106; X86-NEXT: movl 12(%ebp), %eax 107; X86-NEXT: movl 8(%ebp), %ecx 108; X86-NEXT: movq (%ecx), %mm0 109; X86-NEXT: movd (%eax), %mm1 110; X86-NEXT: psllq %mm1, %mm0 111; X86-NEXT: movq %mm0, (%esp) 112; X86-NEXT: movl (%esp), %eax 113; X86-NEXT: movl {{[0-9]+}}(%esp), %edx 114; X86-NEXT: movl %ebp, %esp 115; X86-NEXT: popl %ebp 116; X86-NEXT: retl 117; 118; X64-LABEL: t3: 119; X64: # %bb.0: # %entry 120; X64-NEXT: movq (%rdi), %mm0 121; X64-NEXT: movd (%rsi), %mm1 122; X64-NEXT: psllq %mm1, %mm0 123; X64-NEXT: movq %mm0, %rax 124; X64-NEXT: retq 125entry: 126 %0 = bitcast <1 x i64>* %y to x86_mmx* 127 %1 = load x86_mmx, x86_mmx* %0, align 8 128 %2 = load i32, i32* %n, align 4 129 %3 = tail call x86_mmx @llvm.x86.mmx.pslli.q(x86_mmx %1, i32 %2) 130 %4 = bitcast x86_mmx %3 to i64 131 ret i64 %4 132} 133 134declare x86_mmx @llvm.x86.sse.pshuf.w(x86_mmx, i8) 135declare x86_mmx @llvm.x86.mmx.pslli.q(x86_mmx, i32) 136declare x86_mmx @llvm.x86.mmx.por(x86_mmx, x86_mmx) 137 138