1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse4.2 | FileCheck %s --check-prefix=X86 3; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse4.2 | FileCheck %s --check-prefix=X64 4 5; truncate v2i64 to v2i32 6 7define void @convert_v2i64_to_v2i32(<2 x i32>* %dst.addr, <2 x i64> %src) nounwind { 8; X86-LABEL: convert_v2i64_to_v2i32: 9; X86: # %bb.0: # %entry 10; X86-NEXT: movl {{[0-9]+}}(%esp), %eax 11; X86-NEXT: pcmpeqd %xmm1, %xmm1 12; X86-NEXT: psubd %xmm1, %xmm0 13; X86-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] 14; X86-NEXT: movq %xmm0, (%eax) 15; X86-NEXT: retl 16; 17; X64-LABEL: convert_v2i64_to_v2i32: 18; X64: # %bb.0: # %entry 19; X64-NEXT: paddd {{.*}}(%rip), %xmm0 20; X64-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] 21; X64-NEXT: movq %xmm0, (%rdi) 22; X64-NEXT: retq 23entry: 24 %val = trunc <2 x i64> %src to <2 x i32> 25 %add = add <2 x i32> %val, < i32 1, i32 1 > 26 store <2 x i32> %add, <2 x i32>* %dst.addr 27 ret void 28} 29 30; truncate v3i32 to v3i8 31 32define void @convert_v3i32_to_v3i8(<3 x i8>* %dst.addr, <3 x i32>* %src.addr) nounwind { 33; X86-LABEL: convert_v3i32_to_v3i8: 34; X86: # %bb.0: # %entry 35; X86-NEXT: pushl %eax 36; X86-NEXT: movl {{[0-9]+}}(%esp), %eax 37; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx 38; X86-NEXT: movdqa (%ecx), %xmm0 39; X86-NEXT: pcmpeqd %xmm1, %xmm1 40; X86-NEXT: psubd %xmm1, %xmm0 41; X86-NEXT: pextrb $8, %xmm0, 2(%eax) 42; X86-NEXT: pshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u] 43; X86-NEXT: pextrw $0, %xmm0, (%eax) 44; X86-NEXT: popl %eax 45; X86-NEXT: retl 46; 47; X64-LABEL: convert_v3i32_to_v3i8: 48; X64: # %bb.0: # %entry 49; X64-NEXT: movdqa (%rsi), %xmm0 50; X64-NEXT: pcmpeqd %xmm1, %xmm1 51; X64-NEXT: psubd %xmm1, %xmm0 52; X64-NEXT: pextrb $8, %xmm0, 2(%rdi) 53; X64-NEXT: pshufb {{.*#+}} xmm0 = xmm0[0,4,8,12,u,u,u,u,u,u,u,u,u,u,u,u] 54; X64-NEXT: pextrw $0, %xmm0, (%rdi) 55; X64-NEXT: retq 56entry: 57 %load = load <3 x i32>, <3 x i32>* %src.addr 58 %val = trunc <3 x i32> %load to <3 x i8> 59 %add = add <3 x i8> %val, < i8 1, i8 1, i8 1 > 60 store <3 x i8> %add, <3 x i8>* %dst.addr 61 ret void 62} 63 64; truncate v5i16 to v5i8 65 66define void @convert_v5i16_to_v5i8(<5 x i8>* %dst.addr, <5 x i16>* %src.addr) nounwind { 67; X86-LABEL: convert_v5i16_to_v5i8: 68; X86: # %bb.0: # %entry 69; X86-NEXT: pushl %ebp 70; X86-NEXT: movl %esp, %ebp 71; X86-NEXT: andl $-8, %esp 72; X86-NEXT: subl $8, %esp 73; X86-NEXT: movl 8(%ebp), %eax 74; X86-NEXT: movl 12(%ebp), %ecx 75; X86-NEXT: movdqa (%ecx), %xmm0 76; X86-NEXT: pcmpeqd %xmm1, %xmm1 77; X86-NEXT: psubw %xmm1, %xmm0 78; X86-NEXT: pextrb $8, %xmm0, 4(%eax) 79; X86-NEXT: pshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u] 80; X86-NEXT: movd %xmm0, (%eax) 81; X86-NEXT: movl %ebp, %esp 82; X86-NEXT: popl %ebp 83; X86-NEXT: retl 84; 85; X64-LABEL: convert_v5i16_to_v5i8: 86; X64: # %bb.0: # %entry 87; X64-NEXT: movdqa (%rsi), %xmm0 88; X64-NEXT: pcmpeqd %xmm1, %xmm1 89; X64-NEXT: psubw %xmm1, %xmm0 90; X64-NEXT: pextrb $8, %xmm0, 4(%rdi) 91; X64-NEXT: pshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u] 92; X64-NEXT: movd %xmm0, (%rdi) 93; X64-NEXT: retq 94entry: 95 %load = load <5 x i16>, <5 x i16>* %src.addr 96 %val = trunc <5 x i16> %load to <5 x i8> 97 %add = add <5 x i8> %val, < i8 1, i8 1, i8 1, i8 1, i8 1 > 98 store <5 x i8> %add, <5 x i8>* %dst.addr 99 ret void 100} 101