[m-rev.] for review: fix mantis bug #559, step 2

[Julien Fischer]

On Fri, 6 May 2022, Zoltan Somogyi wrote:

> 2022-05-06 15:12 GMT+10:00 "Julien Fischer" <jfischer at opturion.com>:
>> The rest looks fine -- thanks for looking into this so promptly.
>
> You are welcome. I followed all your suggestions.
>
> I would like to ask you to benchmark the effect of this change as well,
> maybe with fewer optimination levels (say just 2 and 5), but with more
> repetitions, say 12 or 16 instead of 6.

Results attached.  Baseline is commit 2cb46b2, while modified is commit
b96a90f. (I meant to benchmark both asm_fast.gc and hlc.gc at -O2 and
-O5, but didn't realise I had set things up incorrectly until this
morning; let me know if you you want to run the other combinations).

Julien.
-------------- next part --------------
benchmark 1:
EXTRA_MCFLAGS = -O2
GRADE = asm_fast.gc
base mercury_compile.01 average of 16 with ignore=1     22.60
mod  mercury_compile.01 average of 16 with ignore=1     22.55 speedup  0.22%

benchmark 2:
EXTRA_MCFLAGS = -O2 --intermod-opt
GRADE = asm_fast.gc
base mercury_compile.02 average of 16 with ignore=1     22.00
mod  mercury_compile.02 average of 16 with ignore=1     21.85 speedup  0.68%

benchmark 3:
EXTRA_MCFLAGS = -O5
GRADE = hlc.gc
base mercury_compile.03 average of 16 with ignore=1     22.39
mod  mercury_compile.03 average of 16 with ignore=1     22.46 speedup -0.31%

benchmark 4:
EXTRA_MCFLAGS = -O5 --intermod-opt
GRADE = hlc.gc
base mercury_compile.04 average of 16 with ignore=1     19.85
mod  mercury_compile.04 average of 16 with ignore=1     19.73 speedup  0.60%

total baseline time  86.84
total modified time  86.59 speedup  0.29%
baseline wins   1
modified wins   3