[PATCH] Splitting of virtual definitions (aka. V_MUST_DEF)

[Brian Booth]

On Tue, 2004-06-08 at 08:09, Diego Novillo wrote:
> Could you test the patch against POOMA
> (http://www.tat.physik.uni-tuebingen.de/~rguenth/gcc/monitor-summary.html) 

Here is the average of three runs:

flags: -O2 -funroll-loops -ffast-math

	WITH PATCH	WITHOUT PATCH
real	83m54.488s	79m35.982s
usr	3m38.307s	3m37.057s
sys	0m27.477s	0m24.467s

assembly
code 	8,643,479	8,649,104
size
(bytes)

Ah, thrashing. Anyway, this shows a compile time (usr time) increase of
roughly 0.58% which was fairly consistent over all three runs and a code
size improvement of 0.07%.

> and PR8361? 

Here is the average of three runs:

flags: -O3

	WITH PATCH	WITHOUT PATCH
real	0m21.108s	0m20.944s
usr	0m17.293s	0m17.347s
sys	0m1.470s	0m1.583s

assembly
code 	4139		4139
size
(bytes)

No code size improvement and a slight decrease in compile time (roughly
0.31%). However, since one was not consistently faster than the other
over the three runs, I'm assuming that this is just noise.

Brian
---