[llvm-commits] [llvm] r158259 - in /llvm/trunk: lib/Target/PowerPC/PPCTargetMachine.cpp test/CodeGen/PowerPC/2007-05-22-tailmerge-3.ll

[Hal Finkel]

Author: hfinkel
Date: Fri Jun  8 22:14:50 2012
New Revision: 158259

URL: http://llvm.org/viewvc/llvm-project?rev=158259&view=rev
Log:
Enable tail merging on PPC.

Tail merging had been disabled on PPC because it would disturb bundling decisions
made during pre-RA scheduling on the 970 cores. Now, however, all bundling decisions
are made during post-RA scheduling, and tail merging is generally beneficial (the
average test-suite speedup is insignificantly positive).

Largest test-suite speedups:
MultiSource/Benchmarks/mediabench/gsm/toast/toast - 30%
MultiSource/Benchmarks/BitBench/uuencode/uuencode - 23%
SingleSource/Benchmarks/Shootout-C++/ary - 21%
SingleSource/Benchmarks/Stanford/Queens - 17%

Largest slowdowns:
MultiSource/Benchmarks/MiBench/security-sha/security-sha - 24%
MultiSource/Benchmarks/McCat/03-testtrie/testtrie - 22%
MultiSource/Applications/JM/ldecod/ldecod - 14%
MultiSource/Benchmarks/mediabench/g721/g721encode/encode - 9%

This is improved by using full (instead of just critical) anti-dependency breaking,
but doing so still causes miscompiles and so cannot yet be enabled by default.

Modified:
    llvm/trunk/lib/Target/PowerPC/PPCTargetMachine.cpp
    llvm/trunk/test/CodeGen/PowerPC/2007-05-22-tailmerge-3.ll

Modified: llvm/trunk/lib/Target/PowerPC/PPCTargetMachine.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/PowerPC/PPCTargetMachine.cpp?rev=158259&r1=158258&r2=158259&view=diff
==============================================================================
--- llvm/trunk/lib/Target/PowerPC/PPCTargetMachine.cpp (original)
+++ llvm/trunk/lib/Target/PowerPC/PPCTargetMachine.cpp Fri Jun  8 22:14:50 2012
@@ -93,13 +93,7 @@
 } // namespace
 
 TargetPassConfig *PPCTargetMachine::createPassConfig(PassManagerBase &PM) {
-  TargetPassConfig *PassConfig = new PPCPassConfig(this, PM);
-
-  // Override this for PowerPC.  Tail merging happily breaks up instruction issue
-  // groups, which typically degrades performance.
-  PassConfig->setEnableTailMerge(false);
-
-  return PassConfig;
+  return new PPCPassConfig(this, PM);
 }
 
 bool PPCPassConfig::addPreRegAlloc() {

Modified: llvm/trunk/test/CodeGen/PowerPC/2007-05-22-tailmerge-3.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/PowerPC/2007-05-22-tailmerge-3.ll?rev=158259&r1=158258&r2=158259&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/PowerPC/2007-05-22-tailmerge-3.ll (original)
+++ llvm/trunk/test/CodeGen/PowerPC/2007-05-22-tailmerge-3.ll Fri Jun  8 22:14:50 2012
@@ -1,8 +1,8 @@
-; RUN: llc < %s -march=ppc32 | grep bl.*baz | count 2
-; RUN: llc < %s -march=ppc32 | grep bl.*quux | count 2
-; RUN: llc < %s -march=ppc32 -enable-tail-merge | grep bl.*baz | count 1
-; RUN: llc < %s -march=ppc32 -enable-tail-merge=1 | grep bl.*quux | count 1
-; Check that tail merging is not the default on ppc, and that -enable-tail-merge works.
+; RUN: llc < %s -march=ppc32 -enable-tail-merge=0 | grep bl.*baz | count 2
+; RUN: llc < %s -march=ppc32 -enable-tail-merge=0 | grep bl.*quux | count 2
+; RUN: llc < %s -march=ppc32 | grep bl.*baz | count 1
+; RUN: llc < %s -march=ppc32 | grep bl.*quux | count 1
+; Check that tail merging is the default on ppc, and that -enable-tail-merge works.
 
 ; ModuleID = 'tail.c'
 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64"