1 ; RUN: llc %s -mtriple=x86_64-unknown-unknown -mattr='-sse4.1' -o - | FileCheck %s -check-prefix=NO_SSE_41 2 ; RUN: llc %s -mtriple=x86_64-unknown-unknown -mattr='+sse4.1' -o - | FileCheck %s -check-prefix=SSE_41 3 4 ; PR20472 ( http://llvm.org/bugs/show_bug.cgi?id=20472 ) 5 ; When sexting a trunc'd vector value, we can't eliminate the zext. 6 ; If we don't have SSE4.1, use punpck. 7 ; If we have SSE4.1, use pmovzx because it combines the load op. 8 ; There may be a better way to do this using pshufb + pmovsx, 9 ; but that is beyond our current codegen capabilities. 10 11 define <4 x i32> @trunc_sext(<4 x i16>* %in) { 12 %load = load <4 x i16>, <4 x i16>* %in 13 %trunc = trunc <4 x i16> %load to <4 x i8> 14 %sext = sext <4 x i8> %trunc to <4 x i32> 15 ret <4 x i32> %sext 16 17 ; NO_SSE_41-LABEL: trunc_sext: 18 ; NO_SSE_41: movq (%rdi), %xmm0 19 ; NO_SSE_41-NEXT: punpcklwd %xmm0, %xmm0 20 ; NO_SSE_41-NEXT: pslld $24, %xmm0 21 ; NO_SSE_41-NEXT: psrad $24, %xmm0 22 ; NO_SSE_41-NEXT: retq 23 24 ; SSE_41-LABEL: trunc_sext: 25 ; SSE_41: pmovzxwd (%rdi), %xmm0 26 ; SSE_41-NEXT: pslld $24, %xmm0 27 ; SSE_41-NEXT: psrad $24, %xmm0 28 ; SSE_41-NEXT: retq 29 } 30 31
