178 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X863; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X644 5; test vector shifts converted to proper SSE2 vector shifts when the shift6; amounts are the same when using a shuffle splat.7 8define void @shift1a(<2 x i64> %val, ptr %dst, <2 x i64> %sh) nounwind {9; X86-LABEL: shift1a:10; X86: # %bb.0: # %entry11; X86-NEXT: movl {{[0-9]+}}(%esp), %eax12; X86-NEXT: psllq %xmm1, %xmm013; X86-NEXT: movdqa %xmm0, (%eax)14; X86-NEXT: retl15;16; X64-LABEL: shift1a:17; X64: # %bb.0: # %entry18; X64-NEXT: psllq %xmm1, %xmm019; X64-NEXT: movdqa %xmm0, (%rdi)20; X64-NEXT: retq21entry:22 %shamt = shufflevector <2 x i64> %sh, <2 x i64> undef, <2 x i32> <i32 0, i32 0>23 %shl = shl <2 x i64> %val, %shamt24 store <2 x i64> %shl, ptr %dst25 ret void26}27 28; shift1b can't use a packed shift but can shift lanes separately and shuffle back together29define void @shift1b(<2 x i64> %val, ptr %dst, <2 x i64> %sh) nounwind {30; X86-LABEL: shift1b:31; X86: # %bb.0: # %entry32; X86-NEXT: movl {{[0-9]+}}(%esp), %eax33; X86-NEXT: movdqa %xmm0, %xmm234; X86-NEXT: psllq %xmm1, %xmm235; X86-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]36; X86-NEXT: psllq %xmm1, %xmm037; X86-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1]38; X86-NEXT: movapd %xmm0, (%eax)39; X86-NEXT: retl40;41; X64-LABEL: shift1b:42; X64: # %bb.0: # %entry43; X64-NEXT: movdqa %xmm0, %xmm244; X64-NEXT: psllq %xmm1, %xmm245; X64-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]46; X64-NEXT: psllq %xmm1, %xmm047; X64-NEXT: movsd {{.*#+}} xmm0 = xmm2[0],xmm0[1]48; X64-NEXT: movapd %xmm0, (%rdi)49; X64-NEXT: retq50entry:51 %shamt = shufflevector <2 x i64> %sh, <2 x i64> undef, <2 x i32> <i32 0, i32 1>52 %shl = shl <2 x i64> %val, %shamt53 store <2 x i64> %shl, ptr %dst54 ret void55}56 57define void @shift2a(<4 x i32> %val, ptr %dst, <2 x i32> %amt) nounwind {58; X86-LABEL: shift2a:59; X86: # %bb.0: # %entry60; X86-NEXT: movl {{[0-9]+}}(%esp), %eax61; X86-NEXT: psrlq $32, %xmm162; X86-NEXT: pslld %xmm1, %xmm063; X86-NEXT: movdqa %xmm0, (%eax)64; X86-NEXT: retl65;66; X64-LABEL: shift2a:67; X64: # %bb.0: # %entry68; X64-NEXT: psrlq $32, %xmm169; X64-NEXT: pslld %xmm1, %xmm070; X64-NEXT: movdqa %xmm0, (%rdi)71; X64-NEXT: retq72entry:73 %shamt = shufflevector <2 x i32> %amt, <2 x i32> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 1>74 %shl = shl <4 x i32> %val, %shamt75 store <4 x i32> %shl, ptr %dst76 ret void77}78 79define void @shift2b(<4 x i32> %val, ptr %dst, <2 x i32> %amt) nounwind {80; X86-LABEL: shift2b:81; X86: # %bb.0: # %entry82; X86-NEXT: movl {{[0-9]+}}(%esp), %eax83; X86-NEXT: psrlq $32, %xmm184; X86-NEXT: pslld %xmm1, %xmm085; X86-NEXT: movdqa %xmm0, (%eax)86; X86-NEXT: retl87;88; X64-LABEL: shift2b:89; X64: # %bb.0: # %entry90; X64-NEXT: psrlq $32, %xmm191; X64-NEXT: pslld %xmm1, %xmm092; X64-NEXT: movdqa %xmm0, (%rdi)93; X64-NEXT: retq94entry:95 %shamt = shufflevector <2 x i32> %amt, <2 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 1, i32 1>96 %shl = shl <4 x i32> %val, %shamt97 store <4 x i32> %shl, ptr %dst98 ret void99}100 101define void @shift2c(<4 x i32> %val, ptr %dst, <2 x i32> %amt) nounwind {102; X86-LABEL: shift2c:103; X86: # %bb.0: # %entry104; X86-NEXT: movl {{[0-9]+}}(%esp), %eax105; X86-NEXT: psrlq $32, %xmm1106; X86-NEXT: pslld %xmm1, %xmm0107; X86-NEXT: movdqa %xmm0, (%eax)108; X86-NEXT: retl109;110; X64-LABEL: shift2c:111; X64: # %bb.0: # %entry112; X64-NEXT: psrlq $32, %xmm1113; X64-NEXT: pslld %xmm1, %xmm0114; X64-NEXT: movdqa %xmm0, (%rdi)115; X64-NEXT: retq116entry:117 %shamt = shufflevector <2 x i32> %amt, <2 x i32> undef, <4 x i32> <i32 1, i32 1, i32 1, i32 1>118 %shl = shl <4 x i32> %val, %shamt119 store <4 x i32> %shl, ptr %dst120 ret void121}122 123define void @shift3a(<8 x i16> %val, ptr %dst, <8 x i16> %amt) nounwind {124; X86-LABEL: shift3a:125; X86: # %bb.0: # %entry126; X86-NEXT: movl {{[0-9]+}}(%esp), %eax127; X86-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,6,6,6,6]128; X86-NEXT: psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero129; X86-NEXT: psllw %xmm1, %xmm0130; X86-NEXT: movdqa %xmm0, (%eax)131; X86-NEXT: retl132;133; X64-LABEL: shift3a:134; X64: # %bb.0: # %entry135; X64-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,6,6,6,6]136; X64-NEXT: psrldq {{.*#+}} xmm1 = xmm1[14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero137; X64-NEXT: psllw %xmm1, %xmm0138; X64-NEXT: movdqa %xmm0, (%rdi)139; X64-NEXT: retq140entry:141 %shamt = shufflevector <8 x i16> %amt, <8 x i16> undef, <8 x i32> <i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6, i32 6>142 %shl = shl <8 x i16> %val, %shamt143 store <8 x i16> %shl, ptr %dst144 ret void145}146 147define void @shift3b(<8 x i16> %val, ptr %dst, i16 %amt) nounwind {148; X86-LABEL: shift3b:149; X86: # %bb.0: # %entry150; X86-NEXT: movl {{[0-9]+}}(%esp), %eax151; X86-NEXT: movzwl {{[0-9]+}}(%esp), %ecx152; X86-NEXT: movd %ecx, %xmm1153; X86-NEXT: psllw %xmm1, %xmm0154; X86-NEXT: movdqa %xmm0, (%eax)155; X86-NEXT: retl156;157; X64-LABEL: shift3b:158; X64: # %bb.0: # %entry159; X64-NEXT: movzwl %si, %eax160; X64-NEXT: movd %eax, %xmm1161; X64-NEXT: psllw %xmm1, %xmm0162; X64-NEXT: movdqa %xmm0, (%rdi)163; X64-NEXT: retq164entry:165 %0 = insertelement <8 x i16> undef, i16 %amt, i32 0166 %1 = insertelement <8 x i16> %0, i16 %amt, i32 1167 %2 = insertelement <8 x i16> %1, i16 %amt, i32 2168 %3 = insertelement <8 x i16> %2, i16 %amt, i32 3169 %4 = insertelement <8 x i16> %3, i16 %amt, i32 4170 %5 = insertelement <8 x i16> %4, i16 %amt, i32 5171 %6 = insertelement <8 x i16> %5, i16 %amt, i32 6172 %7 = insertelement <8 x i16> %6, i16 %amt, i32 7173 %shl = shl <8 x i16> %val, %7174 store <8 x i16> %shl, ptr %dst175 ret void176}177 178