[llvm-commits] [dragonegg] r141945 - in /dragonegg/trunk: README src/Backend.cpp

[Duncan Sands]

Author: baldrick
Date: Fri Oct 14 01:10:09 2011
New Revision: 141945

URL: http://llvm.org/viewvc/llvm-project?rev=141945&view=rev
Log:
Polyhedron 2005 benchmark results show that when GCC optimizations
are enabled the level of LLVM optimization doesn't matter: -O1, -O2
and -O3 all give essentially the same code quality and more or less
the same compile time.  I'm not convinced: my not very scientific
testing shows LLVM -O2 as being the sweet spot when GCC optimizes
at -O3.

Modified:
    dragonegg/trunk/README
    dragonegg/trunk/src/Backend.cpp

Modified: dragonegg/trunk/README
URL: http://llvm.org/viewvc/llvm-project/dragonegg/trunk/README?rev=141945&r1=141944&r2=141945&view=diff
==============================================================================
--- dragonegg/trunk/README (original)
+++ dragonegg/trunk/README Fri Oct 14 01:10:09 2011
@@ -112,9 +112,9 @@
 
 -fplugin-arg-dragonegg-enable-gcc-optzns
   Run the GCC tree optimizers rather than the LLVM IR optimizers (normally all
-  GCC optimizations are disabled).  By default this severely reduces the amount
-  of LLVM IR optimization.  Use -O4 to -O6 to have LLVM optimize harder, or set
-  an explicit level using the -fplugin-arg-dragonegg-llvm-ir-optimize option.
+  GCC optimizations are disabled).  By default this reduces the amount of LLVM
+  IR optimization.  Use -O4 to have LLVM optimize harder, or explicitly set a
+  level using the -fplugin-arg-dragonegg-llvm-ir-optimize option.
 
 -fplugin-arg-dragonegg-save-gcc-output
   GCC assembler output is normally redirected to /dev/null so that it doesn't

Modified: dragonegg/trunk/src/Backend.cpp
URL: http://llvm.org/viewvc/llvm-project/dragonegg/trunk/src/Backend.cpp?rev=141945&r1=141944&r2=141945&view=diff
==============================================================================
--- dragonegg/trunk/src/Backend.cpp (original)
+++ dragonegg/trunk/src/Backend.cpp Fri Oct 14 01:10:09 2011
@@ -210,13 +210,6 @@
   // If the user supplied an LLVM optimization level then use it.
   if (LLVMIROptimizeArg >= 0)
     return LLVMIROptimizeArg;
-  // If the GCC optimizers were run then tone down the LLVM optimization level:
-  //   GCC | LLVM
-  //   ----------
-  //     0 |   0
-  //     1 |   1 (per-function maximum)
-  if (EnableGCCOptimizations)
-    return (optimize + 1) / 2;
   // Otherwise use the GCC optimization level.
   return optimize;
 }
@@ -233,12 +226,10 @@
   //     0 |   0
   //     1 |   0
   //     2 |   1
-  //     3 |   1
-  //     4 |   2
-  //     5 |   2
-  //     6 |   3 (per-module maximum)
+  //     3 |   2
+  //     4 |   3 (per-module maximum)
   if (EnableGCCOptimizations)
-    return optimize / 2;
+    return optimize > 0 ? optimize - 1 : 0;
   // Otherwise use the GCC optimization level.
   return optimize;
 }