[9fans] neomagic acceleration

[Micah Stetson <micah@cnm-vra.com>]

[-- Attachment #1: Type: text/plain, Size: 381 bytes --]

With the help of Russ's userlevel scaffolding, I've gotten
hardware acceleration to work on my NeoMagic MagicGraph
128XD (IBM Thinkpad 600).  I'm not sure if it works with
other neomagic chipsets (although the drivers I looked
at seemed to indicate so) and it doesn't work at 24 bpp,
but for my purposes, it's complete.  Anyway, here's the
changed vganeomagic.c.

Micah


[-- Attachment #2.1: Type: text/plain, Size: 280 bytes --]

The following attachment had content that we can't
prove to be harmless.  To avoid possible automatic
execution, we changed the content headers.
The original header was:

	Content-Type: text/x-csrc; charset=us-ascii
	Content-Disposition: attachment; filename="vganeomagic.c"

[-- Attachment #2.2: vganeomagic.c.suspect --]
[-- Type: application/octet-stream, Size: 11231 bytes --]

#include "u.h"
#include "../port/lib.h"
#include "mem.h"
#include "dat.h"
#include "fns.h"
#include "io.h"
#include "../port/error.h"

#define	Image	IMAGE
#include <draw.h>
#include <memdraw.h>
#include <cursor.h>
#include "screen.h"

typedef struct {
	int	enable;
	int	x;
	int	y;
	int	colour1;
	int	colour2;
	int	addr;
} CursorNM;

static ulong
neomagiclinear(VGAscr* scr, int* size, int* align)
{
	ulong aperture, oaperture;
	int oapsize, wasupamem;
	Pcidev *p;

	oaperture = scr->aperture;
	oapsize = scr->apsize;
	wasupamem = scr->isupamem;

	aperture = 0;
	if(p = pcimatch(nil, 0x10C8, 0)){
		switch(p->did){
		case 0x0004:		/* MagicGraph 128XD */
		case 0x0005:		/* MagicMedia 256AV */
		case 0x0006:		/* MagicMedia 256ZX */
			aperture = p->mem[0].bar & ~0x0F;
			*size = p->mem[0].size;
			break;
		default:
			break;
		}
	}

	if(wasupamem){
		if(oaperture == aperture)
			return oaperture;
		upafree(oaperture, oapsize);
	}
	scr->isupamem = 0;

	aperture = upamalloc(aperture, *size, *align);
	if(aperture == 0){
		if(wasupamem && upamalloc(oaperture, oapsize, 0)){
			aperture = oaperture;
			scr->isupamem = 1;
		}
		else
			scr->isupamem = 0;
	}
	else
		scr->isupamem = 1;

	return aperture;
}

static void
neomagicenable(VGAscr* scr)
{
	Pcidev *p;
	int align, curoff, size, vmsize;
	ulong aperture;

	/*
	 * Only once, can't be disabled for now.
	 * scr->io holds the physical address of the cursor registers
	 * in the MMIO space. This may need to change for older chips
	 * which have the MMIO space offset in the framebuffer region.
	 */
	if(scr->io)
		return;
	if(p = pcimatch(nil, 0x10C8, 0)){
		switch(p->did){
		case 0x0004:		/* MagicGraph 128XD */
			curoff = 0x100;
			vmsize = 2048*1024;
			break;
		case 0x0005:		/* MagicMedia 256AV */
			curoff = 0x1000;
			vmsize = 2560*1024;
			break;
		case 0x0006:		/* MagicMedia 256ZX */
			curoff = 0x1000;
			vmsize = 4096*1024;
			break;
		default:
			return;
		}
	}
	else
		return;
	scr->io = upamalloc(p->mem[1].bar & ~0x0F, p->mem[1].size, 0);
	if(scr->io == 0)
		return;
	addvgaseg("neomagicmmio", scr->io, p->mem[1].size);
	scr->mmio = KADDR(scr->io);

	/*
	 * Find a place for the cursor data in display memory.
	 * 2 cursor images might be needed, 1KB each so use the
	 * last 2KB of the framebuffer.
	 */
	scr->storage = vmsize-2*1024;
	scr->io += curoff;

	size = p->mem[0].size;
	align = 0;
	aperture = neomagiclinear(scr, &size, &align);
	if(aperture) {
		scr->aperture = aperture;
		scr->apsize = size;
		addvgaseg("neomagicscreen", aperture, size);
	}
}

static void
neomagiccurdisable(VGAscr* scr)
{
	CursorNM *cursornm;

	if(scr->io == 0)
		return;
	cursornm = KADDR(scr->io);
	cursornm->enable = 0;
}

static void
neomagicinitcursor(VGAscr* scr, int xo, int yo, int index)
{
	uchar *p;
	uint p0, p1;
	int x, y;

	p = KADDR(scr->aperture);
	p += scr->storage + index*1024;

	for(y = yo; y < 16; y++){
		p0 = scr->set[2*y];
		p1 = scr->set[2*y+1];
		if(xo){
			p0 = (p0<<xo)|(p1>>(8-xo));
			p1 <<= xo;
		}
		*p++ = p0;
		*p++ = p1;

		for(x = 16; x < 64; x += 8)
			*p++ = 0x00;

		p0 = scr->clr[2*y]|scr->set[2*y];
		p1 = scr->clr[2*y+1]|scr->set[2*y+1];
		if(xo){
			p0 = (p0<<xo)|(p1>>(8-xo));
			p1 <<= xo;
		}
		*p++ = p0;
		*p++ = p1;

		for(x = 16; x < 64; x += 8)
			*p++ = 0x00;
	}
	while(y < 64+yo){
		for(x = 0; x < 64; x += 8){
			*p++ = 0x00;
			*p++ = 0x00;
		}
		y++;
	}
}

static void
neomagiccurload(VGAscr* scr, Cursor* curs)
{
	CursorNM *cursornm;

	if(scr->io == 0)
		return;
	cursornm = KADDR(scr->io);

	cursornm->enable = 0;
	memmove(&scr->Cursor, curs, sizeof(Cursor));
	neomagicinitcursor(scr, 0, 0, 0);
	cursornm->enable = 1;
}

static int
neomagiccurmove(VGAscr* scr, Point p)
{
	CursorNM *cursornm;
	int addr, index, x, xo, y, yo;

	if(scr->io == 0)
		return 1;
	cursornm = KADDR(scr->io);

	index = 0;
	if((x = p.x+scr->offset.x) < 0){
		xo = -x;
		x = 0;
	}
	else
		xo = 0;
	if((y = p.y+scr->offset.y) < 0){
		yo = -y;
		y = 0;
	}
	else
		yo = 0;

	if(xo || yo){
		index = 1;
		neomagicinitcursor(scr, xo, yo, index);
	}
	addr = ((scr->storage+(1024*index))>>10) & 0xFFF;
	addr = ((addr & 0x00F)<<8)|((addr>>4) & 0xFF);
	if(cursornm->addr != addr)
		cursornm->addr = addr;

	cursornm->x = x;
	cursornm->y = y;

	return 0;
}

static void
neomagiccurenable(VGAscr* scr)
{
	CursorNM *cursornm;

	neomagicenable(scr);
	if(scr->io == 0)
		return;
	cursornm = KADDR(scr->io);
	cursornm->enable = 0;

	/*
	 * Cursor colours.
	 */
	cursornm->colour1 = (Pblack<<16)|(Pblack<<8)|Pblack;
	cursornm->colour2 = (Pwhite<<16)|(Pwhite<<8)|Pwhite;

	/*
	 * Load, locate and enable the 64x64 cursor.
	 */
	neomagiccurload(scr, &arrow);
	neomagiccurmove(scr, ZP);
	cursornm->enable = 1;
}

static int neomagicbltflags;

/* registers */
enum {
	BltStat = 0,
	BltCntl = 1,
	XPColor = 2,
	FGColor = 3,
	BGColor = 4,
	Pitch = 5,
	ClipLT = 6,
	ClipRB = 7,
	SrcBitOff = 8,
	SrcStartOff = 9,

	DstStartOff = 11,
	XYExt = 12,

	PageCntl = 20,
	PageBase,
	PostBase,
	PostPtr,
	DataPtr,
};

/* flags */
enum {
	NEO_BS0_BLT_BUSY =	0x00000001,
	NEO_BS0_FIFO_AVAIL =	0x00000002,
	NEO_BS0_FIFO_PEND =	0x00000004,

	NEO_BC0_DST_Y_DEC =	0x00000001,
	NEO_BC0_X_DEC =		0x00000002,
	NEO_BC0_SRC_TRANS =	0x00000004,
	NEO_BC0_SRC_IS_FG =	0x00000008,
	NEO_BC0_SRC_Y_DEC =	0x00000010,
	NEO_BC0_FILL_PAT =	0x00000020,
	NEO_BC0_SRC_MONO =	0x00000040,
	NEO_BC0_SYS_TO_VID =	0x00000080,

	NEO_BC1_DEPTH8 =	0x00000100,
	NEO_BC1_DEPTH16 =	0x00000200,
	NEO_BC1_DEPTH24 =	0x00000300,
	NEO_BC1_X_320 =		0x00000400,
	NEO_BC1_X_640 =		0x00000800,
	NEO_BC1_X_800 =		0x00000c00,
	NEO_BC1_X_1024 =		0x00001000,
	NEO_BC1_X_1152 =		0x00001400,
	NEO_BC1_X_1280 =		0x00001800,
	NEO_BC1_X_1600 =		0x00001c00,
	NEO_BC1_DST_TRANS =	0x00002000,
	NEO_BC1_MSTR_BLT =	0x00004000,
	NEO_BC1_FILTER_Z =	0x00008000,

	NEO_BC2_WR_TR_DST =	0x00800000,

	NEO_BC3_SRC_XY_ADDR =	0x01000000,
	NEO_BC3_DST_XY_ADDR =	0x02000000,
	NEO_BC3_CLIP_ON =		0x04000000,
	NEO_BC3_FIFO_EN =		0x08000000,
	NEO_BC3_BLT_ON_ADDR =	0x10000000,
	NEO_BC3_SKIP_MAPPING =	0x80000000,

	NEO_MODE1_DEPTH8 =			0x0100,
	NEO_MODE1_DEPTH16 =			0x0200,
	NEO_MODE1_DEPTH24 =			0x0300,
	NEO_MODE1_X_320 =			0x0400,
	NEO_MODE1_X_640 =			0x0800,
	NEO_MODE1_X_800 =			0x0c00,
	NEO_MODE1_X_1024 =			0x1000,
	NEO_MODE1_X_1152 =			0x1400,
	NEO_MODE1_X_1280 =			0x1800,
	NEO_MODE1_X_1600 =			0x1c00,
	NEO_MODE1_BLT_ON_ADDR =	0x2000,
};

/* Raster Operations */
enum {
	GXclear =			0x000000,	/* 0x0000 */
	GXand =			0x080000,	/* 0x1000 */
	GXandReverse =	0x040000,	/* 0x0100 */
	GXcopy =			0x0c0000,	/* 0x1100 */
	GXandInvert =		0x020000,	/* 0x0010 */
	GXnoop =			0x0a0000,	/* 0x1010 */
	GXxor =			0x060000,	/* 0x0110 */
	GXor =			0x0e0000,	/* 0x1110 */
	GXnor =			0x010000,	/* 0x0001 */
	GXequiv =			0x090000,	/* 0x1001 */
	GXinvert =		0x050000,	/* 0x0101 */
	GXorReverse =		0x0d0000,	/* 0x1101 */
	GXcopyInvert =		0x030000,	/* 0x0011 */
	GXorInverted =		0x0b0000,	/* 0x1011 */
	GXnand =			0x070000,	/* 0x0111 */
	GXset =			0x0f0000,	/* 0x1111 */
};

static void
waitforidle(VGAscr *scr)
{
	ulong *mmio;
	long x;

	mmio = scr->mmio;
	x = 0;
	while((mmio[BltStat] & NEO_BS0_BLT_BUSY) && x++ < 1000000)
		;
	//if(x >= 1000000)
	//	iprint("idle stat %lud scrmmio %.8lux scr %p pc %luX\n", mmio[BltStat], scr->mmio, scr, getcallerpc(&scr));
}

static void
waitforfifo(VGAscr *scr, int entries)
{
	ulong *mmio;
	long x;

	mmio = scr->mmio;
	x = 0;
	while(((mmio[BltStat]>>8) < entries) && x++ < 1000000)
		;
	//if(x >= 1000000)
	//	iprint("fifo stat %d scrmmio %.8lux scr %p pc %luX\n", mmio[BltStat]>>8, scr->mmio, scr, getcallerpc(&scr));
	/* DirectFB says the above doesn't work.  if so... */
	/* waitforidle(scr); */
}

static int
neomagichwfill(VGAscr *scr, Rectangle r, ulong sval)
{
	ulong *mmio;

	mmio = scr->mmio;

	waitforfifo(scr, 1);
	mmio[FGColor] = sval;
	waitforfifo(scr, 3);
	mmio[BltCntl] = neomagicbltflags
		| NEO_BC3_FIFO_EN
		| NEO_BC0_SRC_IS_FG
		| NEO_BC3_SKIP_MAPPING
		| GXcopy;
	mmio[DstStartOff] = scr->aperture
		+ r.min.y*scr->gscreen->width*BY2WD
		+ r.min.x*scr->gscreen->depth/BI2BY;
	mmio[XYExt] = (Dy(r) << 16) | (Dx(r) & 0xffff);
	waitforidle(scr);
	return 1;
}

static int
neomagichwscroll(VGAscr *scr, Rectangle r, Rectangle sr)
{
	ulong *mmio;
	int pitch, pixel;

	mmio = scr->mmio;

	pitch = scr->gscreen->width*BY2WD;
	pixel = scr->gscreen->depth/BI2BY;

	waitforfifo(scr, 4);
	if (r.min.y < sr.min.y || (r.min.y == sr.min.y && r.min.x < sr.min.x)) {
		/* start from upper-left */
		mmio[BltCntl] = neomagicbltflags
			| NEO_BC3_FIFO_EN
			| NEO_BC3_SKIP_MAPPING
			| GXcopy;
		mmio[SrcStartOff] = scr->aperture
			+ sr.min.y*pitch + sr.min.x*pixel;
		mmio[DstStartOff] = scr->aperture
			+ r.min.y*pitch + r.min.x*pixel;
	} else {
		/* start from lower-right */
		mmio[BltCntl] = neomagicbltflags
			| NEO_BC0_X_DEC
			| NEO_BC0_DST_Y_DEC
			| NEO_BC0_SRC_Y_DEC
			| NEO_BC3_FIFO_EN
			| NEO_BC3_SKIP_MAPPING
			| GXcopy;
		mmio[SrcStartOff] = scr->aperture
			+ (sr.max.y-1)*pitch + (sr.max.x-1)*pixel;
		mmio[DstStartOff] = scr->aperture
			+ (r.max.y-1)*pitch + (r.max.x-1)*pixel;
	}
	mmio[XYExt] = (Dy(r) << 16) | (Dx(r) & 0xffff);
	waitforidle(scr);
	return 1;
}

static void
neomagicdrawinit(VGAscr *scr)
{
	ulong *mmio;
	uint bltmode, pitch;

	mmio = scr->mmio;

	pitch = scr->gscreen->width*BY2WD;

	neomagicbltflags = bltmode = 0;

	switch(scr->gscreen->depth) {
	case 8:
		bltmode |= NEO_MODE1_DEPTH8;
		neomagicbltflags |= NEO_BC1_DEPTH8;
		break;
	case 16:
		bltmode |= NEO_MODE1_DEPTH16;
		neomagicbltflags |= NEO_BC1_DEPTH16;
		break;
	case 24:	/* I can't get it to work, and XFree86 doesn't either. */
	default:	/* give up */
		return;
	}

	switch(Dx(scr->gscreen->r)) {
	case 320:
		bltmode |= NEO_MODE1_X_320;
		neomagicbltflags |= NEO_BC1_X_320;
		break;
	case 640:
		bltmode |= NEO_MODE1_X_640;
		neomagicbltflags |= NEO_BC1_X_640;
		break;
	case 800:
		bltmode |= NEO_MODE1_X_800;
		neomagicbltflags |= NEO_BC1_X_800;
		break;
	case 1024:
		bltmode |= NEO_MODE1_X_1024;
		neomagicbltflags |= NEO_BC1_X_1024;
		break;
	case 1152:
		bltmode |= NEO_MODE1_X_1152;
		neomagicbltflags |= NEO_BC1_X_1152;
		break;
	case 1280:
		bltmode |= NEO_MODE1_X_1280;
		neomagicbltflags |= NEO_BC1_X_1280;
		break;
	case 1600:
		bltmode |= NEO_MODE1_X_1600;
		neomagicbltflags |= NEO_BC1_X_1600;
		break;
	default:
		/* don't worry about it */
		break;
	}

	waitforidle(scr);
	mmio[BltStat] = bltmode << 16;
	mmio[Pitch] = (pitch << 16) | (pitch & 0xffff);

	scr->fill = neomagichwfill;
	scr->scroll = neomagichwscroll;
}

VGAdev vganeomagicdev = {
	"neomagic",

	neomagicenable,
	nil,
	nil,
	neomagiclinear,
	neomagicdrawinit,
};

VGAcur vganeomagiccur = {
	"neomagichwgc",

	neomagiccurenable,
	neomagiccurdisable,
	neomagiccurload,
	neomagiccurmove,
};