[m-dev.] Proposing a new grade component.
Paul Bone
pbone at csse.unimelb.edu.au
Thu Nov 10 14:57:09 AEDT 2011
On Thu, Nov 10, 2011 at 12:41:25PM +1100, Paul Bone wrote:
> On Tue, Nov 08, 2011 at 02:16:59PM +1100, Julien Fischer wrote:
> >
> > Given that it may have changed then I suggest you measure it again.
> >
>
> EXTRA_MCFLAGS =
> GRADE = asm_fast.gc
> mercury_compile.01 average of 5 with ignore=1 14.10
>
> EXTRA_MCFLAGS =
> GRADE = asm_fast.gc.profdeep
> mercury_compile.03 average of 5 with ignore=1 75.55
>
> EXTRA_MCFLAGS = --coverage-profiling
> GRADE = asm_fast.gc.profdeep
> mercury_compile.05 average of 5 with ignore=1 82.69
>
> Note that none of these tests use --profile-optimized.
>
> Test | Time sec | Slowdown | Slowdown compared to dp
> ----------------+-----------=---------------+-------------------------
> asm_fast.gc | 14.10 | 1.00 | 5.36
> dp | 75.55 | 0.19 | 1.00
> dp + coverage | 82.69 | 0.17 | 0.91
>
> Coverage profiling is about 9% slower than normal deep profiling.
> Deep profiling is about 81% slower than normal execution,
> and coverage profiling is about 83% slower than normal execution.
>
> I'm going to do another test, I want to see what it's like when I compile the
> library with profile optimized and compile the application (in this case mmc)
> without profile optimized.
>
With profile-optimized:
dp: 50.7s
dp + coverage: 53.05s
These are both much faster than before, and the gap between them has also
narrowed.
The net result of the proposed changes will actually be a faster deep profiling
grade. To re-iterate, we propose to enable coverage profiling by default, and
to enable --profile-optimized for the standard library.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 490 bytes
Desc: Digital signature
URL: <http://lists.mercurylang.org/archives/developers/attachments/20111110/70581662/attachment.sig>
More information about the developers
mailing list