[llvm-commits] [llvm-gcc-4.2] r63023 - /llvm-gcc-4.2/trunk/gcc/llvm-backend.cpp

Chris Lattner clattner at apple.com
Tue Jan 27 00:03:16 PST 2009


On Jan 27, 2009, at 12:00 AM, Evan Cheng wrote:
> Looks like there are confusion about the semantics of the value being
> passed to createLoopUnswitchPass. My understanding is if it's true
> then it does loop switching but not aggressively. That is, it reduces
> the aggressiveness so it doesn't bloat code size. From your statement,
> that doesn't seem like the case?
>
> At -O2, we want to do this optimization, but we want to limit code
> size increase.

Evan's right, we want to run it at both -Os and -O2, but not in the  
"aggressive" mode.

In fact, since you can't have -Os -O3, should the "aggressive" mode  
just be enabled when optimize >= 3 and drop the optimize_size condition?

-Chris



More information about the llvm-commits mailing list