From 6ada70d1b55b9efb8b0603575a0b86e14f049553 Mon Sep 17 00:00:00 2001 From: Simon Pilgrim Date: Sat, 19 Oct 2019 11:53:48 +0000 Subject: [X86][SSE] LowerUINT_TO_FP_i64 - only use HADDPD for size/fast-hops We were always generating a single source HADDPD, but really we should only do this if shouldUseHorizontalOp says its a good idea. Differential Revision: https://reviews.llvm.org/D69175 llvm-svn: 375341 --- llvm/test/CodeGen/X86/scalar-int-to-fp.ll | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) (limited to 'llvm/test/CodeGen/X86/scalar-int-to-fp.ll') diff --git a/llvm/test/CodeGen/X86/scalar-int-to-fp.ll b/llvm/test/CodeGen/X86/scalar-int-to-fp.ll index 580da1535e3..31faa728ce2 100644 --- a/llvm/test/CodeGen/X86/scalar-int-to-fp.ll +++ b/llvm/test/CodeGen/X86/scalar-int-to-fp.ll @@ -610,8 +610,9 @@ define double @u64_to_d(i64 %a) nounwind { ; AVX512F_32-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero ; AVX512F_32-NEXT: vunpcklps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1] ; AVX512F_32-NEXT: vsubpd {{\.LCPI.*}}, %xmm0, %xmm0 -; AVX512F_32-NEXT: vhaddpd %xmm0, %xmm0, %xmm0 -; AVX512F_32-NEXT: vmovlpd %xmm0, (%esp) +; AVX512F_32-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0] +; AVX512F_32-NEXT: vaddsd %xmm0, %xmm1, %xmm0 +; AVX512F_32-NEXT: vmovsd %xmm0, (%esp) ; AVX512F_32-NEXT: fldl (%esp) ; AVX512F_32-NEXT: movl %ebp, %esp ; AVX512F_32-NEXT: popl %ebp -- cgit v1.2.3