mixed arithmetic is not accelerated in 6.1 64 bits images

Hi all,
if I inquire about
    (Smalltalk vmParameterAt: 75).
    (Smalltalk vmParameterAt: 77).
I can see that they are both answering false.
This results in such performances:

[2.0+2] bench.
 '69,200,000 per second. 14.4 nanoseconds per run. 0 % GC time.'
[2.0>2] bench.
 '78,300,000 per second. 12.8 nanoseconds per run. 0 % GC time.'

But if I do this:
  (Smalltalk vmParameterAt: 75 put: true).
  (Smalltalk vmParameterAt: 77 put: true).

then performance logically increases:

[2.0+2] bench.
 '341,000,000 per second. 2.94 nanoseconds per run. 0 % GC time.'
[2.0>2] bench.
 '387,000,000 per second. 2.58 nanoseconds per run. 0 % GC time.'

Why wouldn't we want a x4 in such micro benchmarks?
Maybe I hallucinated, but I wonder if I did not see some source code
changes dedicated to accelerate a bit such mixed arithmetic at image side,
a pity, given that the problem is already solved at VM side...

Nicolas

Squeak-dev mailing list -- [email protected]
To unsubscribe send an email to [email protected]