Closed GoogleCodeExporter closed 8 years ago
If i do remember correctly,
i tested that already.
I've been surprised to note that, by removing the "A32(op)=A32(ref);" which is
indeed useless in 32 bits mode, the decoder became (slightly) slower, or at
best was unaffected by the change (depending on compiler and version).
So i let it this way, intentionnally.
Back then, i was suspecting this strange result was a side-effect of compiler
optimisations.
I can nonetheless re-test this.
Original comment by yann.col...@gmail.com
on 6 Dec 2012 at 1:01
I checked the removing on instruction A32(op)=A32(ref);
From an assembler perspective, it removes 2 instructions :
00113 8b 10 mov edx, DWORD PTR [eax]
00115 89 11 mov DWORD PTR [ecx], edx
In spite of this, the version without these instructions is actually slower
(although the difference is within 0.5%, but it tends to be repeatable).
It might be due to some pre-fetching side-effects.
Original comment by yann.col...@gmail.com
on 8 Dec 2012 at 12:16
Doesn't bring benefits to speed
Original comment by yann.col...@gmail.com
on 11 Dec 2012 at 1:42
Original issue reported on code.google.com by
strig...@gmail.com
on 5 Dec 2012 at 8:24