[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [ft-devel] ttinterp.c optimization
From: |
Alexei Podtelezhnikov |
Subject: |
Re: [ft-devel] ttinterp.c optimization |
Date: |
Tue, 26 Aug 2014 01:34:28 -0400 |
On Sun, Aug 24, 2014 at 10:24 PM, Alexei Podtelezhnikov
<address@hidden> wrote:
>>> I want to move CURRENT_Ppem outside of the point loops. It seems to
>>> work fine. Anything wrong with that?
>>
>> This is OK, since the ppem (measured along the projection vector)
>> doesn't change within a single call to DELTAP or DELTAC. Good catch!
>
> I found it when I was collecting FT_MulFix statistics and saw way too
> many trivial multiplications by 1.0 in large batches. Just committed.
Furthermore, it turns out that MPS instruction is *very* *very*
common. Why do we call CURRENT_Ppem here? The comment suggest that the
decision was arbitrary and, perhaps, wasteful. I see a huge number of
trivial FT_MulFix's by 1.0 still.