commit 5bef9c627e44077e4ec5024be062faecc4eab3e5
author Simon Pilgrim <llvm-dev@redking.me.uk> Sun May 14 17:59:46 2017 +0000
committer Simon Pilgrim <llvm-dev@redking.me.uk> Sun May 14 17:59:46 2017 +0000
tree 68afdca220cfdad89eaf113ceda81b5fc34bf516
parent aa8dffb69bccbe4e41469e0023b7508dd44945f8

[X86][XOP] XOP's general v16i8 shifts will be used instead of v8i16 shift + mask.

Tweak cost model to match what lowering actually does.

llvm-svn: 303013

llvm/lib/Target/X86/X86TargetTransformInfo.cpp

diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
index ba6462a..2952925 100644
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -273,9 +273,12 @@
     if (ISD == ISD::SDIV && LT.second == MVT::v4i32 && ST->hasSSE41())
       return LT.first * 15;
 
-    if (const auto *Entry = CostTableLookup(SSE2UniformConstCostTable, ISD,
-                                            LT.second))
-      return LT.first * Entry->Cost;
+    // XOP has faster vXi8 shifts.
+    if ((ISD != ISD::SHL && ISD != ISD::SRL && ISD != ISD::SRA) ||
+        !ST->hasXOP())
+      if (const auto *Entry =
+              CostTableLookup(SSE2UniformConstCostTable, ISD, LT.second))
+        return LT.first * Entry->Cost;
   }
 
   static const CostTblEntry AVX2UniformCostTable[] = {