Software Screen to Screen BitBlt Acceleration GRADD filter, r0.0.2

0. Table of Contents

0. Table of Contents
1. Welcome
- 1.1. Copyright and Disclaimer
- 1.2. System Requirements
- 1.3. Contact Author
- 1.4. Revision History
2. How to Use
- 2.1. In This Archive
- 2.2. Driver Installation
- 2.3. Configuration
- 2.4. Before and After
3. Credits

1. Welcome

SSSBBACC.DLL is a small filter GRADD driver that can intercept some sort of 'Screen to Screen' BitBlt software emulation responses from GHI layer and will process such responses using its own faster blitter.

both source and destination are VRAM
source-copy raster operation, with no pattern, no transparency, no stretching and no clipping

This simple, assembler-coded, MMX / SSEint-ready blitter is heavy optimized for uncached memory reads and write-combined memory writes, in order to eliminate unnecessary and much slower peripheral bus cycles as much as possible.
Typically, such slower software BitBlts may speed up to x1.5 ~ x2.5. Highly recommended for GENGRADD / poor-accelerated GHI (such as M64GRADD in 24bpp) users.

1.1. Copyright and Disclaimer

'THIS SOFTWARE (SSSBBACC.DLL) AND WHOLE SET OF DOCUMENTS' (CALLED 'THE PRODUCT') ARE ENTIRELY COPYRIGHTED BY TAKAYUKI 'JANUARY JUNE' SUWA/BELIEFIA PRODUCTIONS, AUTHOR/HOLDER OF THE PRODUCT.
THE PRODUCT IS STILL UNDER DEVELOPMENT AT THE DISTRIBUTION, AND MAY BE OCCURED ANY MODIFICATION / DELETION OF APPEARANCE / FUNCTION OR TERMINATION OF DEVELOPMENT/SERVICE IN FUTURE WITHOUT NOTICE.
THE PRODUCT IS PROVIDED 'AS IS' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
IN NO EVENT SHALL THE COPYRIGHT HOLDER BE LIABLE FOR ANY DAMAGES, WHETHER IN CONTRACT OR TORT (INCLUDED NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THE PRODUCT, EVEN THOUGH SUCH HOLDER HAS BEEN ADVISED OF THE POSSIBILITY OF ANY DAMAGE.

We are regarded you agree disclaimer above at the first installation of SSSBBACC.DLL. If you disagree above, you must discard all files of the product.

1.2. System Requirements

This software will be functioned in environment such as:

OS/2 version 3.x with FP35+, OS/2 version 4.0 with FP5+, WSeB, MCP, ACP or eCS. Both uniprocessor and multiprocessor version are supported.
Any GRADD-based display drivers, build E.083 equivalent or later. For example, IBM GRADD BBS package or SciTech Display Doctor 7 for OS/2.

1.3. Contact Author

Email to Takayuki 'January June' Suwa:

mailto:jjsuwa@sys3175.com

1.4. Revision History

r0.0.0 - August 31st,2001
1. Internal release. The idea seems true :)
r0.0.1 - September 4th,2001
1. First public release.
2. Added SSEint MMX blitter for SSEint-capable CPUs, and legacy MMX blitter for MMX-capable CPUs.
r0.0.2 - September 6th,2001
1. Modified that internal BitBlt buffer is allocated dynamically in order to reduce locked memory pages.

2. How to Use

2.1. In This Archive

SSSBBACC.dll: Software Screen to Screen BitBlt Acceleration GRADD filter driver file
README.htm: This document
source\*: Source code set of SSSBBACC.dll
source\fromIBM\*.h: Some C/C++ header files from DDK

2.2. Driver Installation

Unzip archive to where you desire. (e.g. C:\SSSBBACC)
Copy SSSBBACC.DLL to where LIBPATH reaches. (e.g. C:\OS2\DLL)
Find similar line
SET C1=GRADD1,GRADD2,GRADD3
from your CONFIG.SYS and append 'SSSBBACC' to it. For example:
SET C1=GRADD1,GRADD2,GRADD3,SSSBBACC
Optionally (but highly recommended), cast write-combining strategy to VRAM aperture.
- If using IBM GRADD BBS package with Intel P6 (PentiumPro, PentiumII, Celeron and PentiumIII) or AMD K7 (Athlon and Duron), simply add "SET GRADD_WC=Y" line to CONFIG.SYS (uniprocessor only)
- If using SDD/2, turn 'no write-combining' to off (gaoption nowcomb off)
- Another our product, Intel P6 / AMD K7 MTRRs management IOCtl PDD can also do. It now supports multiprocessor write-combining.
Restart your system.

Check GRADD.LOG on \OS2\INSTALL directory on boot drive. If installation successfully, you will see such as:

yyyymmddThhmmss VMAN:ProcessGraddList(): Module GRADD1 loaded successfully
yyyymmddThhmmss VMAN:ProcessGraddList(): Module GRADD2 loaded successfully
yyyymmddThhmmss VMAN:ProcessGraddList(): Module GRADD3 loaded successfully
yyyymmddThhmmss VMAN:ProcessGraddList(): Module SSSBBACC loaded successfully

2.3. Configuration

In default, SSSBBACC.DLL will detect MMX / SSEint and try to use them automatcally:
- first, uses SSEint MMX blitter if SSEint is available
- next, uses legacy MMX blitter if MMX is available
- last, uses generic FPU integer blitter
In order not to use MMX or SSEint, set some environment variables in CONFIG.SYS, with dummy value:

SSSBBACC_NOSSEINT

inhibits using SSEint

SSSBBACC_NOMMX

inhibits using MMX

2.4. Before and After

Sample case (our developping environment):

2-way Celeron-A 412MHz (75Mhz x5.5) multiprocessor, i440BX mainboard
ATI RagePro 2x AGP 4MB SDRAM, resolution of 1024 x 768, write-combining strategy applied by P6K7MTRR+VFBWC
OS/2 Warp4.0 Japanese NLV w/FP15 SMP-tweaked (revision 14.074h_SMP)
IBM GRADD BBS package build E.083

SysBench 'BitBlt S->S Copy' results:

ATI Mach64 GRADD [M64GRADD] (ill-slowed 24bpp left-to-right S2S BitBlt): 24bpp : 1.501 to 4.161 million pixels/second (x2.77 faster)
SVGA Generic GRADD [GENGRADD] (totally software processing): 8bpp : 4.616 to 11.551 million pixels/second (x2.50 faster); 16bpp : 2.548 to 5.660 million pixels/second (x2.22 faster); 24bpp : 1.471 to 3.687 million pixels/second (x2.51 faster)

3. Credits

Development
Special thanks to

Copyright 2001 Takayuki 'January June' Suwa / Beliefia Productions. All rights reserved.
Beliefia Productions and its logo design are trademarks of Beliefia Productions. All rights reserved.
Other copyrighted works(images, musics, programs, and so on), trademarks and servicemarks in this document are owned by their respective owners.

[This document is HTML3.2(Wilbur)-compliant.]