[llvm] r179381 - LoopVectorizer: integer division is not a reduction operation

[Arnold Schwaighofer]

Author: arnolds
Date: Fri Apr 12 10:15:19 2013
New Revision: 179381

URL: http://llvm.org/viewvc/llvm-project?rev=179381&view=rev
Log:
LoopVectorizer: integer division is not a reduction operation

Don't classify idiv/udiv as a reduction operation. Integer division is lossy.
For example : (1 / 2) * 4 != 4/2.

Example:

int a[] = { 2, 5, 2, 2}
int x = 80;

for()
  x /= a[i];

Scalar:
  x /= 2 // = 40
  x /= 5 // = 8
  x /= 2 // = 4
  x /= 2 // = 2

Vectorized:

 <80, 1> / <2,5> //= <40,0>
 <40, 0> / <2,2> //= <20,0>

 20*0 = 0

radar://13640654

Added:
    llvm/trunk/test/Transforms/LoopVectorize/no_idiv_reduction.ll
Modified:
    llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp

Modified: llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp?rev=179381&r1=179380&r2=179381&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp (original)
+++ llvm/trunk/lib/Transforms/Vectorize/LoopVectorize.cpp Fri Apr 12 10:15:19 2013
@@ -2833,8 +2833,6 @@ LoopVectorizationLegality::isReductionIn
   case Instruction::Sub:
   case Instruction::Add:
     return Kind == RK_IntegerAdd;
-  case Instruction::SDiv:
-  case Instruction::UDiv:
   case Instruction::Mul:
     return Kind == RK_IntegerMult;
   case Instruction::And:

Added: llvm/trunk/test/Transforms/LoopVectorize/no_idiv_reduction.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/LoopVectorize/no_idiv_reduction.ll?rev=179381&view=auto
==============================================================================
--- llvm/trunk/test/Transforms/LoopVectorize/no_idiv_reduction.ll (added)
+++ llvm/trunk/test/Transforms/LoopVectorize/no_idiv_reduction.ll Fri Apr 12 10:15:19 2013
@@ -0,0 +1,24 @@
+; RUN: opt -loop-vectorize -force-vector-width=2 -force-vector-unroll=1 -S < %s | FileCheck %s
+ at a = common global [128 x i32] zeroinitializer, align 16
+
+;; Must not vectorize division reduction. Division is lossy.
+define i32 @g() {
+entry:
+  br label %for.body
+
+for.body:
+  ; CHECK: @g
+  ; CHECK-NOT: sdiv <2 x i32>
+  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
+  %r.05 = phi i32 [ 80, %entry ], [ %div, %for.body ]
+  %arrayidx = getelementptr inbounds [128 x i32]* @a, i64 0, i64 %indvars.iv
+  %0 = load i32* %arrayidx, align 4
+  %div = sdiv i32 %r.05, %0
+  %indvars.iv.next = add i64 %indvars.iv, 1
+  %lftr.wideiv = trunc i64 %indvars.iv.next to i32
+  %exitcond = icmp eq i32 %lftr.wideiv, 1024
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:
+  ret i32 %div
+}