X86 cost model: Vector shifts are expensive in most cases

The default logic does not correctly identify costs of casts because they are
marked as custom on x86.

For some cases, where the shift amount is a scalar we would be able to generate
better code. Unfortunately, when this is the case the value (the splat) will get
hoisted out of the loop, thereby making it invisible to ISel.

radar://13130673
radar://13537826

llvm-svn: 178703

1 files changed, 42 insertions, 0 deletions

diff --git a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
index 2336035bea0..a38b22bd789 100644
--- a/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/llvm/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -182,6 +182,16 @@ unsigned X86TTI::getArithmeticInstrCost(unsigned Opcode, Type *Ty) const {
     { ISD::SRL,     MVT::v2i64,    1 },
     { ISD::SHL,     MVT::v4i64,    1 },
     { ISD::SRL,     MVT::v4i64,    1 },
+
+    { ISD::SHL,  MVT::v32i8,  42 }, // cmpeqb sequence.
+    { ISD::SHL,  MVT::v16i16,  16*10 }, // Scalarized.
+
+    { ISD::SRL,  MVT::v32i8,  32*10 }, // Scalarized.
+    { ISD::SRL,  MVT::v16i16,  8*10 }, // Scalarized.
+
+    { ISD::SRA,  MVT::v32i8,  32*10 }, // Scalarized.
+    { ISD::SRA,  MVT::v16i16,  16*10 }, // Scalarized.
+    { ISD::SRA,  MVT::v4i64,  4*10 }, // Scalarized.
   };
 
   // Look for AVX2 lowering tricks.
@@ -192,6 +202,38 @@ unsigned X86TTI::getArithmeticInstrCost(unsigned Opcode, Type *Ty) const {
       return LT.first * AVX2CostTable[Idx].Cost;
   }
 
+  static const CostTblEntry<MVT> SSE2CostTable[] = {
+    // We don't correctly identify costs of casts because they are marked as
+    // custom.
+    // For some cases, where the shift amount is a scalar we would be able
+    // to generate better code. Unfortunately, when this is the case the value
+    // (the splat) will get hoisted out of the loop, thereby making it invisible
+    // to ISel. The cost model must return worst case assumptions because it is
+    // used for vectorization and we don't want to make vectorized code worse
+    // than scalar code.
+    { ISD::SHL,  MVT::v16i8,  30 }, // cmpeqb sequence.
+    { ISD::SHL,  MVT::v8i16,  8*10 }, // Scalarized.
+    { ISD::SHL,  MVT::v4i32,  2*5 }, // We optimized this using mul.
+    { ISD::SHL,  MVT::v2i64,  2*10 }, // Scalarized.
+
+    { ISD::SRL,  MVT::v16i8,  16*10 }, // Scalarized.
+    { ISD::SRL,  MVT::v8i16,  8*10 }, // Scalarized.
+    { ISD::SRL,  MVT::v4i32,  4*10 }, // Scalarized.
+    { ISD::SRL,  MVT::v2i64,  2*10 }, // Scalarized.
+
+    { ISD::SRA,  MVT::v16i8,  16*10 }, // Scalarized.
+    { ISD::SRA,  MVT::v8i16,  8*10 }, // Scalarized.
+    { ISD::SRA,  MVT::v4i32,  4*10 }, // Scalarized.
+    { ISD::SRA,  MVT::v2i64,  2*10 }, // Scalarized.
+  };
+
+  if (ST->hasSSE2()) {
+    int Idx = CostTableLookup<MVT>(SSE2CostTable, array_lengthof(SSE2CostTable),
+                                   ISD, LT.second);
+    if (Idx != -1)
+      return LT.first * SSE2CostTable[Idx].Cost;
+  }
+
   static const CostTblEntry<MVT> AVX1CostTable[] = {
     // We don't have to scalarize unsupported ops. We can issue two half-sized
     // operations and we only need to extract the upper YMM half.