1 ; RUN: llc < %s -march=x86 -mattr=+sse2 | FileCheck %s 2 3 ; test vector shifts converted to proper SSE2 vector shifts when the shift 4 ; amounts are the same. 5 6 define void @shift1a(<2 x i64> %val, <2 x i64>* %dst) nounwind { 7 entry: 8 ; CHECK: shift1a: 9 ; CHECK: psrlq 10 %lshr = lshr <2 x i64> %val, < i64 32, i64 32 > 11 store <2 x i64> %lshr, <2 x i64>* %dst 12 ret void 13 } 14 15 define void @shift1b(<2 x i64> %val, <2 x i64>* %dst, i64 %amt) nounwind { 16 entry: 17 ; CHECK: shift1b: 18 ; CHECK: movd 19 ; CHECK-NEXT: psrlq 20 %0 = insertelement <2 x i64> undef, i64 %amt, i32 0 21 %1 = insertelement <2 x i64> %0, i64 %amt, i32 1 22 %lshr = lshr <2 x i64> %val, %1 23 store <2 x i64> %lshr, <2 x i64>* %dst 24 ret void 25 } 26 27 define void @shift2a(<4 x i32> %val, <4 x i32>* %dst) nounwind { 28 entry: 29 ; CHECK: shift2a: 30 ; CHECK: psrld 31 %lshr = lshr <4 x i32> %val, < i32 17, i32 17, i32 17, i32 17 > 32 store <4 x i32> %lshr, <4 x i32>* %dst 33 ret void 34 } 35 36 define void @shift2b(<4 x i32> %val, <4 x i32>* %dst, i32 %amt) nounwind { 37 entry: 38 ; CHECK: shift2b: 39 ; CHECK: movd 40 ; CHECK-NEXT: psrld 41 %0 = insertelement <4 x i32> undef, i32 %amt, i32 0 42 %1 = insertelement <4 x i32> %0, i32 %amt, i32 1 43 %2 = insertelement <4 x i32> %1, i32 %amt, i32 2 44 %3 = insertelement <4 x i32> %2, i32 %amt, i32 3 45 %lshr = lshr <4 x i32> %val, %3 46 store <4 x i32> %lshr, <4 x i32>* %dst 47 ret void 48 } 49 50 51 define void @shift3a(<8 x i16> %val, <8 x i16>* %dst) nounwind { 52 entry: 53 ; CHECK: shift3a: 54 ; CHECK: psrlw 55 %lshr = lshr <8 x i16> %val, < i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5 > 56 store <8 x i16> %lshr, <8 x i16>* %dst 57 ret void 58 } 59 60 ; properly zero extend the shift amount 61 define void @shift3b(<8 x i16> %val, <8 x i16>* %dst, i16 %amt) nounwind { 62 entry: 63 ; CHECK: shift3b: 64 ; CHECK: movzwl 65 ; CHECK: movd 66 ; CHECK-NEXT: psrlw 67 %0 = insertelement <8 x i16> undef, i16 %amt, i32 0 68 %1 = insertelement <8 x i16> %0, i16 %amt, i32 1 69 %2 = insertelement <8 x i16> %0, i16 %amt, i32 2 70 %3 = insertelement <8 x i16> %0, i16 %amt, i32 3 71 %4 = insertelement <8 x i16> %0, i16 %amt, i32 4 72 %5 = insertelement <8 x i16> %0, i16 %amt, i32 5 73 %6 = insertelement <8 x i16> %0, i16 %amt, i32 6 74 %7 = insertelement <8 x i16> %0, i16 %amt, i32 7 75 %lshr = lshr <8 x i16> %val, %7 76 store <8 x i16> %lshr, <8 x i16>* %dst 77 ret void 78 } 79
