a while a go I tried to get alpha blitting of per pixel images going faster.
Some c improvements were made, and some asm mmx/ prefetching stuff was
done as well. The asm uses gcc inline. It’s attached to this email.
I don’t really have time to add in the detection of mmx/prefetching
instructions the way sdl does it. I’d appreciate some help doing it The
SDL Way. Please Pretty please with honey on top(you don’t want to
be eating too much sugar).
The other things left to do are to optionally use SSE prefetching
instead of 3dnow prefetching. Also using femms on k6 processors instead
of emms would be nice. Also need someone to translate it into msvc
inline asm.
If no one helps me with integrating the asm then just doing the c
optimization can speed things up by quite a bit. The major speed boost
from the asm came from the prefetching, not the mmx. Inlining the
function gave an ok speed boost too(1-2 fps).
The function to look for is BlitRGBtoRGBPixelAlpha.
These are FPS of two of my games I tested with, and then a comparison
between the new version and old. Other apps I tried this with had
similar increases in speed. To test use a game which uses 32 bit per
pixel alpha surfaces. These results are from my duron 850 with gcc
3.2.3. Haven’t tried it on anything else, so any bug reports/benchmarks
are welcomed.
With no optimizations.
bleten | dohfighters
average score: 21.7745967561 | 37.2473964489
Final results with all optimizations:
bleten | dohfighters
lowest score: 34.3333557719 |48.1210294823
highest score: 34.4127871498 |50.6342458682
Hi, it’s very cool!!
I think that the alpha blitting is the slowest task of SDL (in graphic), I
will try to port it to Visual C.
cheers> ----- Original Message -----
From: illumen@yahoo.com (Rene Dudfield)
To:
Sent: Friday, April 18, 2003 3:00 AM
Subject: [SDL] some faster alpha blitting code.
Hi,
a while a go I tried to get alpha blitting of per pixel images going
faster.
Some c improvements were made, and some asm mmx/ prefetching stuff was
done as well. The asm uses gcc inline. It’s attached to this email.
I don’t really have time to add in the detection of mmx/prefetching
instructions the way sdl does it. I’d appreciate some help doing it The
SDL Way. Please Pretty please with honey on top(you don’t want to
be eating too much sugar).
The other things left to do are to optionally use SSE prefetching
instead of 3dnow prefetching. Also using femms on k6 processors instead
of emms would be nice. Also need someone to translate it into msvc
inline asm.
If no one helps me with integrating the asm then just doing the c
optimization can speed things up by quite a bit. The major speed boost
from the asm came from the prefetching, not the mmx. Inlining the
function gave an ok speed boost too(1-2 fps).
The function to look for is BlitRGBtoRGBPixelAlpha.
These are FPS of two of my games I tested with, and then a comparison
between the new version and old. Other apps I tried this with had
similar increases in speed. To test use a game which uses 32 bit per
pixel alpha surfaces. These results are from my duron 850 with gcc
3.2.3. Haven’t tried it on anything else, so any bug reports/benchmarks
are welcomed.
With no optimizations.
bleten | dohfighters
average score: 21.7745967561 | 37.2473964489
Final results with all optimizations:
bleten | dohfighters
lowest score: 34.3333557719 |48.1210294823
highest score: 34.4127871498 |50.6342458682
Well, you need to check your percentage calculations. Correct
calculations paint an even better picture:
50.6342458682 / 37.2473964489 * 100 - 100 = 35.94% increase
34.4127871498 / 21.7745967561 * 100 - 100 = 58.04% increase–
[ Below is a random fortune, which is unrelated to the above message. ]
Let him choose out of my files, his projects to accomplish.
– Shakespeare, “Coriolanus”
Well, If I put, for example, any asm instrucction in DUFFS_LOOP4, the
visual
C give me a lot of errors, maybe I could delete macro and to code all but…
any idea about this?
Well, If I put, for example, any asm instrucction in DUFFS_LOOP4, the
visual
C give me a lot of errors, maybe I could delete macro and to code all but…
any idea about this?
Hi,
Sorry no ideas apart from expanding the macro
Perhaps look for some other msvc code in sdl, which may have similar macros.
What are your errors, perhaps a msvc asm coder will know how to fix them?
Well, If I put, for example, any asm instrucction in DUFFS_LOOP4, the
visual
C give me a lot of errors, maybe I could delete macro and to code all
but…
any idea about this?
Hi,
Sorry no ideas apart from expanding the macro
Perhaps look for some other msvc code in sdl, which may have similar
macros.
What are your errors, perhaps a msvc asm coder will know how to fix them?
this asm code was an example but it gave me a lot of errors, I am coding an
unroll macro, on the other hand… where is the best place to check the mmx
support?
also, I am coding mmx alpha blend for surface and a new SDL_memcpy to speed
up blit but I need to check mmx support, something like global var but maybe
that SDL already has it.> ----- Original Message -----
From: atrix2@cox.net (atrix2)
To:
Sent: Monday, April 21, 2003 6:34 PM
Subject: Re: [SDL] some faster alpha blitting code.
how are you putting the asm inside of there? like this?
__asm{
mov ax, 20;
mov es, ax;
};
----- Original Message -----
From: “Rene Dudfield”
To:
Sent: Monday, April 21, 2003 9:03 AM
Subject: Re: [SDL] some faster alpha blitting code.
Roberto Prieto wrote:
Hi Rene,
I was holidays and today I try to port to Visual C but I have a
problem
with
macros and inline asm:
in the function BlitRGBtoRGBPixelAlpha is the next while:
Well, If I put, for example, any asm instrucction in DUFFS_LOOP4, the
visual
C give me a lot of errors, maybe I could delete macro and to code all
but…
any idea about this?
Hi,
Sorry no ideas apart from expanding the macro
Perhaps look for some other msvc code in sdl, which may have similar
macros.
What are your errors, perhaps a msvc asm coder will know how to fix
them?
this asm code was an example but it gave me a lot of errors, I am coding an
unroll macro, on the other hand… where is the best place to check the mmx
support?
Not sure if you meant that “best place to check the mmx support” comment
with regard to the prefetch opcode. But IIRC prefetch was introduced with
SSE 1, not MMX.
Yes and no, we are going to code with mmx(and sse1 too) in SDL blit code, in
this example I used “prefetch” but it is only an example :), although we are
using the prefetch :). Whatever I search is the best place to check mmx (and
sse1 too) support, any idea about this?> ----- Original Message -----
From: dimitri@shortcut.nl (Dimitri)
To:
Sent: Tuesday, April 22, 2003 2:56 PM
Subject: Re: [SDL] some faster alpha blitting code.
this asm code was an example but it gave me a lot of errors, I am coding
an
unroll macro, on the other hand… where is the best place to check the
mmx
support?
Not sure if you meant that “best place to check the mmx support” comment
with regard to the prefetch opcode. But IIRC prefetch was introduced with
SSE 1, not MMX.
ive tried to use SIMD instructions in MSVC before for matrix/vector math but
it wouldnt take them. I think MSVC has a neutered inline asm component.> ----- Original Message -----
From: dm2@mi.madritel.es (Roberto Prieto)
To:
Sent: Tuesday, April 22, 2003 7:06 AM
Subject: Re: [SDL] some faster alpha blitting code.
Yes and no, we are going to code with mmx(and sse1 too) in SDL blit code,
in
this example I used “prefetch” but it is only an example :), although we
are
using the prefetch :). Whatever I search is the best place to check mmx
(and
sse1 too) support, any idea about this?
----- Original Message -----
From: “Dimitri”
To:
Sent: Tuesday, April 22, 2003 2:56 PM
Subject: Re: [SDL] some faster alpha blitting code.
Cant you just #ifdef your way out of the problem?On 22-Apr-2003, Atrix Wolfe wrote:
ive tried to use SIMD instructions in MSVC before for matrix/vector math but
it wouldnt take them. I think MSVC has a neutered inline asm component.
–
Patrick “Diablo-D3” McFarland || unknown at panax.com
"Computer games don’t affect kids; I mean if Pac-Man affected us as kids, we’d
all be running around in darkened rooms, munching magic pills and listening to
repetitive electronic music." – Kristian Wilson, Nintendo, Inc, 1989
Pretty please with honey on top(you don’t want to
