From 7377cff9e7641c75678fd5c80472942fd7ef869a Mon Sep 17 00:00:00 2001 From: Benjamin Kramer Date: Wed, 23 Oct 2013 19:19:04 +0000 Subject: X86: Custom lower zext v16i8 to v16i16. On sandy bridge (PR17654) we now get vpxor %xmm1, %xmm1, %xmm1 vpunpckhbw %xmm1, %xmm0, %xmm2 vpunpcklbw %xmm1, %xmm0, %xmm0 vinsertf128 $1, %xmm2, %ymm0, %ymm0 On haswell it's a simple vpmovzxbw %xmm0, %ymm0 There is a maze of duplicated and dead transforms and patterns in this area. Remove the dead custom lowering of zext v8i16 to v8i32, that's already handled by LowerAVXExtend. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193262 91177308-0d34-0410-b5e6-96231b3b80d8 --- test/CodeGen/X86/avx-zext.ll | 12 ++++++++++++ test/CodeGen/X86/avx2-conversions.ll | 9 +++++++++ 2 files changed, 21 insertions(+) (limited to 'test') diff --git a/test/CodeGen/X86/avx-zext.ll b/test/CodeGen/X86/avx-zext.ll index e2b6c552da..75117463bc 100644 --- a/test/CodeGen/X86/avx-zext.ll +++ b/test/CodeGen/X86/avx-zext.ll @@ -27,3 +27,15 @@ define <8 x i32> @zext_8i8_to_8i32(<8 x i8> %z) { %t = zext <8 x i8> %z to <8 x i32> ret <8 x i32> %t } + +; PR17654 +define <16 x i16> @zext_16i8_to_16i16(<16 x i8> %z) { +; CHECK-LABEL: zext_16i8_to_16i16: +; CHECK: vpxor +; CHECK: vpunpckhbw +; CHECK: vpunpcklbw +; CHECK: vinsertf128 +; CHECK: ret + %t = zext <16 x i8> %z to <16 x i16> + ret <16 x i16> %t +} diff --git a/test/CodeGen/X86/avx2-conversions.ll b/test/CodeGen/X86/avx2-conversions.ll index 3ce08dcc73..0143f18fe2 100644 --- a/test/CodeGen/X86/avx2-conversions.ll +++ b/test/CodeGen/X86/avx2-conversions.ll @@ -63,6 +63,15 @@ define <8 x i32> @zext_8i8_8i32(<8 x i8> %A) nounwind { ret <8 x i32>%B } +; CHECK-LABEL: zext_16i8_16i16: +; CHECK: vpmovzxbw +; CHECK-NOT: vinsert +; CHECK: ret +define <16 x i16> @zext_16i8_16i16(<16 x i8> %z) { + %t = zext <16 x i8> %z to <16 x i16> + ret <16 x i16> %t +} + ; CHECK: load_sext_test1 ; CHECK: vpmovsxdq (%r{{[^,]*}}), %ymm{{.*}} ; CHECK: ret -- cgit v1.2.3