Blitter emulator for 68k

[Wepl]

Originally Posted by BastyCDGS

How does, e.g. WHDLoad do it with MMU snoop options enabled?

On Snoop the custom/cia memory is declared write-protected, on SnoopOCS/ECS/AGA invalid.
All work is then done in the exception handler which differs for 68030, 68040 and 68060. The access is validated regarding size (b/w/l), type of custom register (read-only, write-only, strobe) and several special checks (blitter, copper, ...). Valid writes are saved internally and performed. Valid reads are performed.

68030: all data is contained in the exception stackframe, whdload uses a transparent translation and moveS to write the original data (works only on boards where a usable FC is available, this avoids to change the mmu setup

68040: Write data is contained in the exception stackframe. On writes mmu setup is adapted, data written and mmu setup restored. On reads mmu setup is adapted, instruction executed in trace mode and after mmu setup restored. Special handling for movem with full ea decoder.

68060: No read/write data in the stackframe. On reads adapt mmu, execute in trace mode and restore mmu. On writes adapt MMU, execute in trace mode to save the written data to the internal memory, adapt mmu, execute the same instruction a second time to write the data to the original location, restore mmu.

[jotd]

is there some available 68k source that emulates blitter from cpu?

WinUAE of course does that, but something simpler maybe? asm or C? which could take blitter registers in input can write data the way the blitter does?

I'm interested in replicating some polygon fill code found in Killing Cloud.

	; mode: 0000 0111 1100 1010
	; - no shifting
	; - use b,c,d
	; - LF7 LF6 on, LF5, LF4 off
	; - LF3 LF1 on, FCI & LINE off
	MOVE.W	#$07ca,bltcon0(A5)		;0d20c: 3b7c07ca0040
	MOVE.W	#$0000,bltcon1(A5)		;0d212: 3b7c00000042
	MOVE.W	#$ffff,bltadat(A5)		;0d218: 3b7cffff0074
	TST.B	16(A0)			;0d21e: 4a280010
	BNE.W	lb_0d29c		;0d222: 66000078
	MOVEA.L	18(A0),A2		;0d226: 24680012
	MOVEA.L	22(A0),A3		;0d22a: 26680016
lb_0d22e:
	MOVE.W	(A2)+,D2		;0d22e: 341a
	MOVE.W	(A3)+,D3		;0d230: 361b
	MOVEQ	#-16,D4			;0d232: 78f0
	MOVE.W	D2,D1			;0d234: 3202
	AND.W	D4,D2			;0d236: c444
	AND.W	D3,D4			;0d238: c843
	EOR.W	D2,D1			;0d23a: b541
	EOR.W	D4,D3			;0d23c: b943
	SUB.W	D2,D4			;0d23e: 9842
	BLT.S	lb_0d290		;0d240: 6d4e
	MOVEA.L	A1,A0			;0d242: 2049
	LSR.W	#3,D2			;0d244: e64a
	ADDA.W	D2,A0			;0d246: d0c2
	LSR.W	#4,D4			;0d248: e84c
	ADDQ.W	#1,D4			;0d24a: 5244
	MOVEQ	#20,D0			;0d24c: 7014
	SUB.W	D4,D0			;0d24e: 9044
	ADD.W	D0,D0			;0d250: d040
	ORI.W	#$0100,D4		;0d252: 00440100
	ADD.W	D3,D3			;0d256: d643
	ADD.W	D1,D1			;0d258: d241
	BTST	#6,sd_ctl(A5)		;0d25a: 082d00060002
lb_0d260:
	BTST	#6,sd_ctl(A5)		;0d260: 082d00060002
	BNE.S	lb_0d260		;0d266: 66f8
	MOVE.W	0(A6,D1.W),bltafwm(A5)	;0d268: 3b7610000044
	MOVE.W	32(A6,D3.W),bltalwm(A5)	;0d26e: 3b7630200046
	MOVE.L	A4,bltbpt(A5)		;0d274: 2b4c004c
	MOVE.L	A0,bltcpt(A5)		;0d278: 2b480048
	MOVE.L	A0,bltdpt(A5)		;0d27c: 2b480054
	MOVE.W	D0,bltbmod(A5)		;0d280: 3b400062
	MOVE.W	D0,bltcmod(A5)		;0d284: 3b400060
	MOVE.W	D0,bltdmod(A5)		;0d288: 3b400066
	; main blit of most 3D (filled polygons)
	; without that write, only cockpit and bitmaps are displayed
	; nothing is filled or erased,
	; only lines appear on the display
	MOVE.W	D4,bltsize(A5)		;0d28c: 3b440058 blit (in-game)
lb_0d290:
	LEA	-160(A1),A1		;0d290: 43e9ff60
	DBF	D5,lb_0d22e		;0d294: 51cdff98

[ross]

Originally Posted by jotd

I'm interested in replicating some polygon fill code found in Killing Cloud.

It doesn't seem like a polygon filling routine to me, at least not in the usual ways and terms.

What it seems to do is extract a series of columns (of constant height and variable width) from the source 'bitmap' and insert it into the destination bitmap.
Basically a limited cookie-cut blit where the mask is fixed and 'trimmed'.

So it would be more appropriate to designate it as a 'columns renderer'.
Of course I'm not fully sure of it, you should check the main code flow to be sure

[paraj]

For prototyping simple area blits I use the attached C++ code. Should be easy enough to turn into C or asm if you need to. It's not optimized in any way, doesn't handle line/fill modes and probably has issues with some corner cases.

[jotd]

that looks neat and exactly what I'm looking for! I think I'll have to convert it to C to be able to link to asm easily for a test. Shouldn't be a problem.