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=CHECK --check-prefix=X32 3; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=CHECK --check-prefix=X64 4 5; test vector shifts converted to proper SSE2 vector shifts when the shift 6; amounts are the same. 7 8define void @shift1a(<2 x i64> %val, <2 x i64>* %dst) nounwind { 9; X32-LABEL: shift1a: 10; X32: # %bb.0: # %entry 11; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 12; X32-NEXT: psrlq $32, %xmm0 13; X32-NEXT: movdqa %xmm0, (%eax) 14; X32-NEXT: retl 15; 16; X64-LABEL: shift1a: 17; X64: # %bb.0: # %entry 18; X64-NEXT: psrlq $32, %xmm0 19; X64-NEXT: movdqa %xmm0, (%rdi) 20; X64-NEXT: retq 21entry: 22 %lshr = lshr <2 x i64> %val, < i64 32, i64 32 > 23 store <2 x i64> %lshr, <2 x i64>* %dst 24 ret void 25} 26 27define void @shift1b(<2 x i64> %val, <2 x i64>* %dst, i64 %amt) nounwind { 28; X32-LABEL: shift1b: 29; X32: # %bb.0: # %entry 30; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 31; X32-NEXT: movq {{.*#+}} xmm1 = mem[0],zero 32; X32-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,1,0,1] 33; X32-NEXT: psrlq %xmm1, %xmm0 34; X32-NEXT: movdqa %xmm0, (%eax) 35; X32-NEXT: retl 36; 37; X64-LABEL: shift1b: 38; X64: # %bb.0: # %entry 39; X64-NEXT: movq %rsi, %xmm1 40; X64-NEXT: psrlq %xmm1, %xmm0 41; X64-NEXT: movdqa %xmm0, (%rdi) 42; X64-NEXT: retq 43entry: 44 %0 = insertelement <2 x i64> undef, i64 %amt, i32 0 45 %1 = insertelement <2 x i64> %0, i64 %amt, i32 1 46 %lshr = lshr <2 x i64> %val, %1 47 store <2 x i64> %lshr, <2 x i64>* %dst 48 ret void 49} 50 51define void @shift2a(<4 x i32> %val, <4 x i32>* %dst) nounwind { 52; X32-LABEL: shift2a: 53; X32: # %bb.0: # %entry 54; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 55; X32-NEXT: psrld $17, %xmm0 56; X32-NEXT: movdqa %xmm0, (%eax) 57; X32-NEXT: retl 58; 59; X64-LABEL: shift2a: 60; X64: # %bb.0: # %entry 61; X64-NEXT: psrld $17, %xmm0 62; X64-NEXT: movdqa %xmm0, (%rdi) 63; X64-NEXT: retq 64entry: 65 %lshr = lshr <4 x i32> %val, < i32 17, i32 17, i32 17, i32 17 > 66 store <4 x i32> %lshr, <4 x i32>* %dst 67 ret void 68} 69 70define void @shift2b(<4 x i32> %val, <4 x i32>* %dst, i32 %amt) nounwind { 71; X32-LABEL: shift2b: 72; X32: # %bb.0: # %entry 73; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 74; X32-NEXT: movd {{.*#+}} xmm1 = mem[0],zero,zero,zero 75; X32-NEXT: psrld %xmm1, %xmm0 76; X32-NEXT: movdqa %xmm0, (%eax) 77; X32-NEXT: retl 78; 79; X64-LABEL: shift2b: 80; X64: # %bb.0: # %entry 81; X64-NEXT: movd %esi, %xmm1 82; X64-NEXT: psrld %xmm1, %xmm0 83; X64-NEXT: movdqa %xmm0, (%rdi) 84; X64-NEXT: retq 85entry: 86 %0 = insertelement <4 x i32> undef, i32 %amt, i32 0 87 %1 = insertelement <4 x i32> %0, i32 %amt, i32 1 88 %2 = insertelement <4 x i32> %1, i32 %amt, i32 2 89 %3 = insertelement <4 x i32> %2, i32 %amt, i32 3 90 %lshr = lshr <4 x i32> %val, %3 91 store <4 x i32> %lshr, <4 x i32>* %dst 92 ret void 93} 94 95 96define void @shift3a(<8 x i16> %val, <8 x i16>* %dst) nounwind { 97; X32-LABEL: shift3a: 98; X32: # %bb.0: # %entry 99; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 100; X32-NEXT: psrlw $5, %xmm0 101; X32-NEXT: movdqa %xmm0, (%eax) 102; X32-NEXT: retl 103; 104; X64-LABEL: shift3a: 105; X64: # %bb.0: # %entry 106; X64-NEXT: psrlw $5, %xmm0 107; X64-NEXT: movdqa %xmm0, (%rdi) 108; X64-NEXT: retq 109entry: 110 %lshr = lshr <8 x i16> %val, < i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5 > 111 store <8 x i16> %lshr, <8 x i16>* %dst 112 ret void 113} 114 115; properly zero extend the shift amount 116define void @shift3b(<8 x i16> %val, <8 x i16>* %dst, i16 %amt) nounwind { 117; X32-LABEL: shift3b: 118; X32: # %bb.0: # %entry 119; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 120; X32-NEXT: movzwl {{[0-9]+}}(%esp), %ecx 121; X32-NEXT: movd %ecx, %xmm1 122; X32-NEXT: psrlw %xmm1, %xmm0 123; X32-NEXT: movdqa %xmm0, (%eax) 124; X32-NEXT: retl 125; 126; X64-LABEL: shift3b: 127; X64: # %bb.0: # %entry 128; X64-NEXT: movzwl %si, %eax 129; X64-NEXT: movd %eax, %xmm1 130; X64-NEXT: psrlw %xmm1, %xmm0 131; X64-NEXT: movdqa %xmm0, (%rdi) 132; X64-NEXT: retq 133entry: 134 %0 = insertelement <8 x i16> undef, i16 %amt, i32 0 135 %1 = insertelement <8 x i16> %0, i16 %amt, i32 1 136 %2 = insertelement <8 x i16> %1, i16 %amt, i32 2 137 %3 = insertelement <8 x i16> %2, i16 %amt, i32 3 138 %4 = insertelement <8 x i16> %3, i16 %amt, i32 4 139 %5 = insertelement <8 x i16> %4, i16 %amt, i32 5 140 %6 = insertelement <8 x i16> %5, i16 %amt, i32 6 141 %7 = insertelement <8 x i16> %6, i16 %amt, i32 7 142 %lshr = lshr <8 x i16> %val, %7 143 store <8 x i16> %lshr, <8 x i16>* %dst 144 ret void 145} 146