| 1 |
On 25/02/2013 22:57, Rich Freeman wrote: |
| 2 |
> A sword that cuts two ways - judging understanding by an email is a |
| 3 |
> dubious proposition, otherwise we wouldn't need job interviews. :) |
| 4 |
> It is just as likely that we're simply miscommunicating. |
| 5 |
|
| 6 |
Did you not just say there: |
| 7 |
|
| 8 |
"Calculating scroll bar movement is |
| 9 |
exactly the sort of thing that this flag was actually designed for - |
| 10 |
you don't care if it is off by 1/100th of a pixel." |
| 11 |
|
| 12 |
or am I mistaken? If I'm not mistaken, that phrase is really not |
| 13 |
understanding it. |
| 14 |
|
| 15 |
The calculation that goes in painting on screen a scrollbar are hardly |
| 16 |
something you expect -ffast-math to be designed for. Can you defend your |
| 17 |
statement in any way? |
| 18 |
|
| 19 |
> Define what you mean by "building *by itself*" - I don't want to |
| 20 |
> assume I understand what you're getting at here. I certainly wasn't |
| 21 |
> suggesting that you could be able to run CFLAGS="-O2 -ffast-math" |
| 22 |
> emerge chromium and get anything usable. |
| 23 |
|
| 24 |
Which is exactly what Tom complained about. |
| 25 |
|
| 26 |
> -ffast-math is a flag that |
| 27 |
> should be applied to specific functions or even parts of functions |
| 28 |
> where there is an understood performance vs accuracy tradeoff. |
| 29 |
|
| 30 |
Of course dealing with flags _per functions_ is not possible, as flags |
| 31 |
apply at the very least to a translation unit... |
| 32 |
|
| 33 |
> If you're just using it to calculate how many pixels down it is, it |
| 34 |
> certainly shouldn't be that inaccurate. |
| 35 |
|
| 36 |
But you're not just calculating how many pixels down to draw it... |
| 37 |
you're calculating a bunch of parameters, including shades, shadows, |
| 38 |
sub-pixel positioning, .... |
| 39 |
|
| 40 |
> If you're using it to do |
| 41 |
> pointer arithmetic or something then you're just going to get |
| 42 |
> segfaults. |
| 43 |
|
| 44 |
Uh.. no. Pointer arithmetic is, by and of itself, integer arithmetic. |
| 45 |
That's not going to be influenced by -ffast-math. |
| 46 |
|
| 47 |
Vastly, -ffast-math deals with floating-point arithmetic, which can be |
| 48 |
sped up significantly ignoring some of the rules imposed by |
| 49 |
floating-point arithmetic by IEEE/ISO standards. Breaking which, though, |
| 50 |
can lead to seriously messed up results. |
| 51 |
|
| 52 |
> There are arithmetic functions in computing that are |
| 53 |
> discrete/functional in nature, and there are those which relate more |
| 54 |
> to real-world measurements. Adding a 0.001% error to a hash |
| 55 |
> calculation breaks a hash. Adding 0.001% error to a scientific |
| 56 |
> calculation where all the components have 1% measurement error is |
| 57 |
> insignificant. |
| 58 |
|
| 59 |
But if you add 1% error to hundreds of small calculations ... well, you |
| 60 |
should get the point, don't you? |
| 61 |
|
| 62 |
There are decent use cases for -ffast-math... none of which involve a |
| 63 |
desktop system, in my opinion. |
| 64 |
|
| 65 |
-- |
| 66 |
Diego Elio Pettenò — Flameeyes |
| 67 |
flameeyes@×××××××××.eu — http://blog.flameeyes.eu/ |
Archives