[llvm-commits] [llvm] r167412 - in /llvm/trunk: lib/Target/TargetTransformImpl.cpp test/Analysis/CostModel/X86/cast.ll test/Transforms/LoopVectorize/X86/conversion-cost.ll

[Nadav Rotem]

Author: nadav
Date: Mon Nov  5 16:20:53 2012
New Revision: 167412

URL: http://llvm.org/viewvc/llvm-project?rev=167412&view=rev
Log:
Code Model: Improve the accuracy of the zext/sext/trunc vector cost estimation.

Added:
    llvm/trunk/test/Analysis/CostModel/X86/cast.ll
Modified:
    llvm/trunk/lib/Target/TargetTransformImpl.cpp
    llvm/trunk/test/Transforms/LoopVectorize/X86/conversion-cost.ll

Modified: llvm/trunk/lib/Target/TargetTransformImpl.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/TargetTransformImpl.cpp?rev=167412&r1=167411&r2=167412&view=diff
==============================================================================
--- llvm/trunk/lib/Target/TargetTransformImpl.cpp (original)
+++ llvm/trunk/lib/Target/TargetTransformImpl.cpp Mon Nov  5 16:20:53 2012
@@ -101,7 +101,7 @@
   case AtomicRMW:      return 0;
   case Trunc:          return ISD::TRUNCATE;
   case ZExt:           return ISD::ZERO_EXTEND;
-  case SExt:           return ISD::SEXTLOAD;
+  case SExt:           return ISD::SIGN_EXTEND;
   case FPToUI:         return ISD::FP_TO_UINT;
   case FPToSI:         return ISD::FP_TO_SINT;
   case UIToFP:         return ISD::UINT_TO_FP;
@@ -235,9 +235,17 @@
         SrcLT.second.getSizeInBits() == DstLT.second.getSizeInBits()) {
 
       // Bitcast between types that are legalized to the same type are free.
-      if (Opcode == Instruction::BitCast)
+      if (Opcode == Instruction::BitCast || Opcode == Instruction::Trunc)
         return 0;
 
+      // Assume that Zext is done using AND.
+      if (Opcode == Instruction::ZExt)
+        return 1;
+
+      // Assume that sext is done using SHL and SRA.
+      if (Opcode == Instruction::SExt)
+        return 2;
+
       // 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))
@@ -310,7 +318,6 @@
   return 1;
 }
 
-/// Returns the expected cost of Vector Insert and Extract.
 unsigned VectorTargetTransformImpl::getVectorInstrCost(unsigned Opcode,
                                                        Type *Val,
                                                        unsigned Index) const {

Added: llvm/trunk/test/Analysis/CostModel/X86/cast.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Analysis/CostModel/X86/cast.ll?rev=167412&view=auto
==============================================================================
--- llvm/trunk/test/Analysis/CostModel/X86/cast.ll (added)
+++ llvm/trunk/test/Analysis/CostModel/X86/cast.ll Mon Nov  5 16:20:53 2012
@@ -0,0 +1,34 @@
+; RUN: opt < %s  -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | 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"
+
+define i32 @add(i32 %arg) {
+
+  ; -- Same size registeres --
+  ;CHECK: cost of 1 {{.*}} zext
+  %A = zext <4 x i1> undef to <4 x i32>
+  ;CHECK: cost of 2 {{.*}} sext
+  %B = sext <4 x i1> undef to <4 x i32>
+  ;CHECK: cost of 0 {{.*}} trunc
+  %C = trunc <4 x i32> undef to <4 x i1>
+
+  ; -- Different size registers --
+  ;CHECK-NOT: cost of 1 {{.*}} zext
+  %D = zext <8 x i1> undef to <8 x i32>
+  ;CHECK-NOT: cost of 2 {{.*}} sext
+  %E = sext <8 x i1> undef to <8 x i32>
+  ;CHECK-NOT: cost of 2 {{.*}} trunc
+  %F = trunc <8 x i32> undef to <8 x i1>
+
+  ; -- scalars --
+
+  ;CHECK: cost of 1 {{.*}} zext
+  %G = zext i1 undef to i32
+  ;CHECK: cost of 0 {{.*}} trunc
+  %H = trunc i32 undef to i1
+
+  ;CHECK: cost of 1 {{.*}} ret
+  ret i32 undef
+}
+

Modified: llvm/trunk/test/Transforms/LoopVectorize/X86/conversion-cost.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/X86/conversion-cost.ll?rev=167412&r1=167411&r2=167412&view=diff
==============================================================================
--- llvm/trunk/test/Transforms/LoopVectorize/X86/conversion-cost.ll (original)
+++ llvm/trunk/test/Transforms/LoopVectorize/X86/conversion-cost.ll Mon Nov  5 16:20:53 2012
@@ -4,7 +4,7 @@
 target triple = "x86_64-apple-macosx10.8.0"
 
 ;CHECK: @conversion_cost1
-;CHECK: store <8 x i8>
+;CHECK: store <2 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
@@ -25,7 +25,7 @@
 }
 
 ;CHECK: @conversion_cost2
-;CHECK-NOT: <8 x float>
+;CHECK: <2 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