1 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=sse2 | FileCheck %s --check-prefix=X64 2 ; RUN: llc < %s -mtriple=i386-unknown-unknown -mattr=sse2 | FileCheck %s --check-prefix=X32 3 ; RUN: llc < %s -mtriple=i386-unknown-unknown -mattr=avx2 | FileCheck %s --check-prefix=X32AVX 4 5 ; Use movq or movsd to load / store i64 values if sse2 is available. 6 ; rdar://6659858 7 8 define void @foo(i64* %x, i64* %y) { 9 ; X64-LABEL: foo: 10 ; X64: # BB#0: 11 ; X64-NEXT: movq (%rsi), %rax 12 ; X64-NEXT: movq %rax, (%rdi) 13 ; X64-NEXT: retq 14 ; 15 ; X32-LABEL: foo: 16 ; X32: # BB#0: 17 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 18 ; X32-NEXT: movl {{[0-9]+}}(%esp), %ecx 19 ; X32-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero 20 ; X32-NEXT: movsd %xmm0, (%eax) 21 ; X32-NEXT: retl 22 %tmp1 = load i64, i64* %y, align 8 23 store i64 %tmp1, i64* %x, align 8 24 ret void 25 } 26 27 ; Verify that a 64-bit chunk extracted from a vector is stored with a movq 28 ; regardless of whether the system is 64-bit. 29 30 define void @store_i64_from_vector(<8 x i16> %x, <8 x i16> %y, i64* %i) { 31 ; X64-LABEL: store_i64_from_vector: 32 ; X64: # BB#0: 33 ; X64-NEXT: paddw %xmm1, %xmm0 34 ; X64-NEXT: movq %xmm0, (%rdi) 35 ; X64-NEXT: retq 36 ; 37 ; X32-LABEL: store_i64_from_vector: 38 ; X32: # BB#0: 39 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax 40 ; X32-NEXT: paddw %xmm1, %xmm0 41 ; X32-NEXT: movq %xmm0, (%eax) 42 ; X32-NEXT: retl 43 %z = add <8 x i16> %x, %y ; force execution domain 44 %bc = bitcast <8 x i16> %z to <2 x i64> 45 %vecext = extractelement <2 x i64> %bc, i32 0 46 store i64 %vecext, i64* %i, align 8 47 ret void 48 } 49 50 define void @store_i64_from_vector256(<16 x i16> %x, <16 x i16> %y, i64* %i) { 51 ; X32AVX-LABEL: store_i64_from_vector256: 52 ; X32AVX: # BB#0: 53 ; X32AVX-NEXT: movl {{[0-9]+}}(%esp), %eax 54 ; X32AVX-NEXT: vpaddw %ymm1, %ymm0, %ymm0 55 ; X32AVX-NEXT: vextracti128 $1, %ymm0, %xmm0 56 ; X32AVX-NEXT: vmovq %xmm0, (%eax) 57 ; X32AVX-NEXT: vzeroupper 58 ; X32AVX-NEXT: retl 59 %z = add <16 x i16> %x, %y ; force execution domain 60 %bc = bitcast <16 x i16> %z to <4 x i64> 61 %vecext = extractelement <4 x i64> %bc, i32 2 62 store i64 %vecext, i64* %i, align 8 63 ret void 64 } 65 66 ; PR23476 67 ; Handle extraction from a non-simple / pre-legalization type. 68 69 define void @PR23476(<5 x i64> %in, i64* %out, i32 %index) { 70 ; X32-LABEL: PR23476: 71 ; X32: movsd {{.*#+}} xmm0 = mem[0],zero 72 ; X32-NEXT: movsd %xmm0, (%eax) 73 %ext = extractelement <5 x i64> %in, i32 %index 74 store i64 %ext, i64* %out, align 8 75 ret void 76 } 77
