[AD] Allegro x86 clear and blit optimizations - update

[Bob <ohannessian@xxxxxxxxxx>]

That didn't come out very well. Ok, here's the 'better' table.

Here are my test results:

Test system: 
CPU: Pentium 3, 800MHz EB, 133MHz FSB
Mobo: Soyo 7VCA VIA @ 133MHz
RAM: 64MB, 100MHz, cas3
Video: Asus V3900 TNT2 + TV in/out, AGP 2X
OS: Windows 95
Compiler: DJGPP (gcc -v = 2.95.2)
Compilation command: 
	make I686ONLY=1
Test program: tests/test.exe


Resolution:	Test:	1	2	3	4	Imp	Imp2
640x480x8
	sc->sc  	3715	3700	3729	3749.5	0.37%	1.33%
	sc->sc(al)	3703	3734	3746.5	3726.5	1.17%	-0.21%
	mem->sc 	29560	29527.5	29470	29551	-0.31%	0.07%
	mem->sc(al)	31147.5	31092	30751	31204.5	-1.28%	0.36%
	masked  	31713	31670.5	31707	31524	-0.02%	-0.47%
	clear   	25107	14078	29357	15983	16.92%	13.53%

640x480x16
	sc->sc  	2637.5	1872	2639.5	1866.5	0.07%	-0.30%
	sc->sc(al)	2654.5	1874.5	2663	1873	0.32%	-0.09%
	mem->sc 	15983	15907	16097	16004	0.71%	0.60%
	mem->sc(al)	16707.5	16780	16891	16895	1.09%	0.68%
	masked  	21178	21106	21333.5	21608	0.73%	2.37%
	clear   	13813	10558.5	13830	10244	0.12%	-2.98%

640x480x32
	sc->sc  	N/A	961.5	N/A	961.5	N/A	0.00%
	sc->sc(al)	N/A	1029	N/A	1032.5	N/A	0.34%
	mem->sc 	N/A	8379	N/A	8323.5	N/A	-0.67%
	mem->sc(al)	N/A	8524.5	N/A	8522.5	N/A	-0.03%
	masked  	N/A	13189	N/A	13116.5	N/A	-0.55%
	clear   	N/A	3771	N/A	3525.5		-6.52%

Test1: old code (MMX)
Test2: old code (non-MMX)
Test3: blit patch (MMX)
Test4: blit patch (non-MMX)
Imp: Improvement over MMX
Imp2: Improvement over non-MMX

Note that 32bpp doesn't have MMX code.

Conclusions: There is virtually no improvement in the blit code. The
only winner is the 8bpp clear.
As always, YMMV.

--
- Robert J Ohannessian

"Microsoft code is probably O(n^20)" (My CS teacher)