Disable loop re-rotation for -Oz

[aturetsk]

Summary

After changes in rL231820 loop re-rotation is performed even in -Oz mode. Since loop rotation is disabled for -Oz, it seems loop re-rotation should be disabled too.

[mzolotukhin]

Hi Andrey,

The change looks good to me, but I think the test might be slightly improved. You could commit the patch with this addressed.

Thanks,
Michael

Transforms/LoopRotate/oz-disable.ll

5–8

This check could be easily broken. A lot of stuff can happen to this loop when all passes are done, a simple example - the loop can be unrolled by 2, and then this check will fail.

I'd suggest we either check more specific regexps here, or just check debug output (add -debug -debug-only=loop-rotate and REQUIRES: asserts).

[hfinkel]

lib/Transforms/IPO/PassManagerBuilder.cpp

322

This comment is misleading. The rotation is not disabled at -Oz, it is header duplication that is disabled. The comment above the first invocation says:

// Rotate Loop - disable header duplication at -Oz

we should be specific here too about what is being disabled.

[aturetsk]

Hi Michael and Hal,

Thanks for the review.
I've fixed the test and the comment.

[hfinkel]

LGTM.

Transforms/LoopRotate/oz-disable.ll

+; RUN: opt -S -Os < %s | FileCheck %s -check-prefix=OS
+; RUN: opt -S -Oz < %s | FileCheck %s -check-prefix=OZ
+
+; Loop should be rotated for -Os but not for -Oz.
+; OS: icmp
+; OS: icmp
+; OZ: icmp
+; OZ-NOT: icmp
+
+@e = global i32 10
+
+declare void @use(i32)
+
+define void @test() {
+entry:
+  %end = load i32, i32* @e
+  br label %loop
+
+loop:
+  %n.phi = phi i32 [ %n, %loop.fin ], [ 0, %entry ]
+  %cond = icmp eq i32 %n.phi, %end
+  br i1 %cond, label %exit, label %loop.fin
+
+loop.fin:
+  %n = add i32 %n.phi, 1
+  call void @use(i32 %n)
+  br label %loop
+
+exit:
+  ret void
+}

lib/Transforms/IPO/PassManagerBuilder.cpp

void PassManagerBuilder::populateModulePassManager(
   // we must insert a no-op module pass to reset the pass manager.
   MPM.add(createBarrierNoopPass());
 
   if (RunFloat2Int)
     MPM.add(createFloat2IntPass());
 
   // Re-rotate loops in all our loop nests. These may have fallout out of
   // rotated form due to GVN or other transformations, and the vectorizer relies
-  // on the rotated form.
-  MPM.add(createLoopRotatePass());
+  // on the rotated form. Disabled at -Oz.
+  MPM.add(createLoopRotatePass(SizeLevel == 2 ? 0 : -1));
 
   // Distribute loops to allow partial vectorization.  I.e. isolate dependences
   // into separate loop that would otherwise inhibit vectorization.
   if (EnableLoopDistribute)
     MPM.add(createLoopDistributePass());
 
   MPM.add(createLoopVectorizePass(DisableUnrollLoops, LoopVectorize));
   // FIXME: Because of #pragma vectorize enable, the passes below are always