[PATCH] Make SLPVectorizer cost model take credit for instructions that will be erased

[Arch D. Robison]

Hi #llvm,

The patch refines the cost model in SLPVectorizer to take credit for extractelement instructions that will be erased.  The refinement was found to be essential to some cases arising from compilation of Julia.

The patch also adds a test that requires the patch to pass.  The new test adds a new RUN: line because it needs to run with a different (and perhaps more realistic) -slp-threshold than some of the existing tests.  

I'm new to the LLVM testing infrastructure and have a question about the test.  Will the "target = ..." information from the earlier tests be used by the test that I added?

http://llvm-reviews.chandlerc.com/D3160

Files:
  test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll
  lib/Transforms/Vectorize/SLPVectorizer.cpp

Index: test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll
===================================================================
--- test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll
+++ test/Transforms/SLPVectorizer/X86/insert-element-build-vector.ll
@@ -195,3 +195,29 @@
 }
 
 attributes #0 = { nounwind ssp uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf"="true" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
+
+; RUN: opt -S -slp-vectorizer -slp-threshold=0 < %s | FileCheck %s
+
+; Check that cost model for vectorization takes credit for 
+; instructions that are erased.
+define <4 x float> @take_credit(<4 x float> %a, <4 x float> %b) {
+; CHECK-LABEL: @take_credit(
+; CHECK: %1 = fadd <4 x float> %a, %b
+  %a0 = extractelement <4 x float> %a, i32 0
+  %b0 = extractelement <4 x float> %b, i32 0
+  %c0 = fadd float %a0, %b0
+  %a1 = extractelement <4 x float> %a, i32 1
+  %b1 = extractelement <4 x float> %b, i32 1
+  %c1 = fadd float %a1, %b1
+  %a2 = extractelement <4 x float> %a, i32 2
+  %b2 = extractelement <4 x float> %b, i32 2
+  %c2 = fadd float %a2, %b2
+  %a3 = extractelement <4 x float> %a, i32 3
+  %b3 = extractelement <4 x float> %b, i32 3
+  %c3 = fadd float %a3, %b3
+  %v0 = insertelement <4 x float> undef, float %c0, i32 0
+  %v1 = insertelement <4 x float> %v0, float %c1, i32 1
+  %v2 = insertelement <4 x float> %v1, float %c2, i32 2
+  %v3 = insertelement <4 x float> %v2, float %c3, i32 3
+  ret <4 x float> %v3
+}
Index: lib/Transforms/Vectorize/SLPVectorizer.cpp
===================================================================
--- lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -1012,8 +1012,16 @@
       return 0;
     }
     case Instruction::ExtractElement: {
-      if (CanReuseExtract(VL))
-        return 0;
+      if (CanReuseExtract(VL)) {
+        int DeadCost = 0;
+        for (unsigned i = 0, e = VL.size(); i < e; ++i) {
+          ExtractElementInst *E = cast<ExtractElementInst>(VL[i]);
+          if (E->hasOneUse())
+            // Take credit for instruction that will become dead.
+            DeadCost += TTI->getVectorInstrCost(Instruction::ExtractElement, VecTy, i);
+        }
+        return -DeadCost;
+      }
       return getGatherCost(VecTy);
     }
     case Instruction::ZExt:
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D3160.1.patch
Type: text/x-patch
Size: 2469 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140324/6d0a8fdb/attachment.bin>