Fix a bug in the cost calculation of vector casts. Detect situations where bitcasts cost zero.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167170 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 89 insertions, 26 deletions

diff --git a/lib/Target/TargetTransformImpl.cpp b/lib/Target/TargetTransformImpl.cpp
index 38c704f2b1..dbbf37a7dd 100644
--- a/lib/Target/TargetTransformImpl.cpp
+++ b/lib/Target/TargetTransformImpl.cpp
@@ -211,40 +211,55 @@ unsigned VectorTargetTransformImpl::getCastInstrCost(unsigned Opcode, Type *Dst,
   std::pair<unsigned, EVT> DstLT =
   getTypeLegalizationCost(Dst->getContext(), TLI->getValueType(Dst));
 
-  // If the cast is between same-sized registers, then the check is simple.
-  if (SrcLT.first == DstLT.first &&
-      SrcLT.second.getSizeInBits() == DstLT.second.getSizeInBits()) {
-    // Just check the op cost:
-    if (!TLI->isOperationExpand(ISD, DstLT.second)) {
-      // The operation is legal. Assume it costs 1. Multiply
-      // by the type-legalization overhead.
-      return SrcLT.first * 1;
-    }
+  // Handle scalar conversions.
+  if (!Src->isVectorTy() && !Dst->isVectorTy()) {
+    // Just check the op cost. If the operation is legal then assume it costs 1.
+    if (!TLI->isOperationExpand(ISD, DstLT.second))
+      return  1;
+
+    // Assume that illegal scalar instruction are expensive.
+    return 4;
   }
 
-  unsigned ScalarizationCost = 1;
+  // Check vector-to-vector casts.
+  if (Dst->isVectorTy() && Src->isVectorTy()) {
 
-  // Otherwise, assume that the cast is scalarized.
-  if (Dst->isVectorTy()) {
-    unsigned Num = Dst->getVectorNumElements();
-    unsigned Cost = getCastInstrCost(Opcode, Src->getScalarType(),
-                                     Dst->getScalarType());
-    // return the cost of multiple scalar invocation plus the cost of inserting
-    // and extracting the values.
-    ScalarizationCost *= getScalarizationOverhead(Dst, true, true) + Num * Cost;
-  }
+    // If the cast is between same-sized registers, then the check is simple.
+    if (SrcLT.first == DstLT.first &&
+        SrcLT.second.getSizeInBits() == DstLT.second.getSizeInBits()) {
 
-  if (Src->isVectorTy()) {
-    unsigned Num = Src->getVectorNumElements();
+      // Bitcast between types that are legalized to the same type are free.
+      if (Opcode == Instruction::BitCast)
+        return 0;
+
+      // Just check the op cost. If the operation is legal then assume it costs
+      // 1 and multiply by the type-legalization overhead.
+      if (!TLI->isOperationExpand(ISD, DstLT.second))
+        return SrcLT.first * 1;
+    }
+
+    // If we are converting vectors and the operation is illegal, or
+    // if the vectors are legalized to different types, estimate the
+    // scalarization costs.
+    unsigned Num = Dst->getVectorNumElements();
     unsigned Cost = getCastInstrCost(Opcode, Dst->getScalarType(),
                                      Src->getScalarType());
-    // return the cost of multiple scalar invocation plus the cost of inserting
-    // and extracting the values.
-    ScalarizationCost *= getScalarizationOverhead(Src, true, true) + Num * Cost;
+
+    // Return the cost of multiple scalar invocation plus the cost of
+    // inserting and extracting the values.
+    return getScalarizationOverhead(Dst, true, true) + Num * Cost;
   }
 
-  return ScalarizationCost;
-}
+  // We already handled vector-to-vector and scalar-to-scalar conversions. This 
+  // is where we handle bitcast between vectors and scalars. We need to assume
+  //  that the conversion is scalarized in one way or another.
+  if (Opcode == Instruction::BitCast)
+    // Illegal bitcasts are done by storing and loading from a stack slot.
+    return (Src->isVectorTy()? getScalarizationOverhead(Src, false, true):0) +
+           (Dst->isVectorTy()? getScalarizationOverhead(Dst, true, false):0);
+
+  llvm_unreachable("Unhandled cast");
+ }
 
 unsigned VectorTargetTransformImpl::getCFInstrCost(unsigned Opcode) const {
   return 1;
diff --git a/test/Transforms/LoopVectorize/X86/conversion-cost.ll b/test/Transforms/LoopVectorize/X86/conversion-cost.ll
new file mode 100644
index 0000000000..8582613617
--- /dev/null
+++ b/test/Transforms/LoopVectorize/X86/conversion-cost.ll
@@ -0,0 +1,48 @@
+; RUN: opt < %s  -loop-vectorize -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx -S | FileCheck %s
+
+target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.8.0"
+
+;CHECK: @conversion_cost1
+;CHECK: store <8 x i8>
+;CHECK: ret
+define i32 @conversion_cost1(i32 %n, i8* nocapture %A, float* nocapture %B) nounwind uwtable ssp {
+  %1 = icmp sgt i32 %n, 3
+  br i1 %1, label %.lr.ph, label %._crit_edge
+
+.lr.ph:                                           ; preds = %0, %.lr.ph
+  %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 3, %0 ]
+  %2 = trunc i64 %indvars.iv to i8
+  %3 = getelementptr inbounds i8* %A, i64 %indvars.iv
+  store i8 %2, i8* %3, align 1
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, %n
+  br i1 %exitcond, label %._crit_edge, label %.lr.ph
+
+._crit_edge:                                      ; preds = %.lr.ph, %0
+  ret i32 undef
+}
+
+;CHECK: @conversion_cost2
+;CHECK: store <8 x float>
+;CHECK: ret
+define i32 @conversion_cost2(i32 %n, i8* nocapture %A, float* nocapture %B) nounwind uwtable ssp {
+  %1 = icmp sgt i32 %n, 9
+  br i1 %1, label %.lr.ph, label %._crit_edge
+
+.lr.ph:                                           ; preds = %0, %.lr.ph
+  %indvars.iv = phi i64 [ %indvars.iv.next, %.lr.ph ], [ 9, %0 ]
+  %2 = add nsw i64 %indvars.iv, 3
+  %3 = trunc i64 %2 to i32
+  %4 = sitofp i32 %3 to float
+  %5 = getelementptr inbounds float* %B, i64 %indvars.iv
+  store float %4, float* %5, align 4
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, %n
+  br i1 %exitcond, label %._crit_edge, label %.lr.ph
+
+._crit_edge:                                      ; preds = %.lr.ph, %0
+  ret i32 undef
+}