xref: /haiku/src/add-ons/accelerants/neomagic/engine/nm_bes.c (revision 9eb55bc1d104b8fda80898f8b25c94d8000c8255)
1 /* NeoMagic Back End Scaler functions */
2 /* Written by Rudolf Cornelissen 05/2002-06/2003 */
3 
4 #define MODULE_BIT 0x00000200
5 
6 #include "nm_std.h"
7 
8 //fixme: implement: (used for virtual screens!)
9 //void move_overlay(uint16 hdisp_start, uint16 vdisp_start);
10 
11 status_t nm_configure_bes
12 	(const overlay_buffer *ob, const overlay_window *ow, const overlay_view *ov, int offset)
13 {
14 	/* yuy2 (4:2:2) colorspace calculations */
15 	/* Note: Some calculations will have to be modified for other colorspaces if they are incorporated. */
16 
17 	/* Note:
18 	 * in BeOS R5.0.3 and DANO:
19 	 * 'ow->offset_xxx' is always 0, so not used;
20 	 * 'ow->width' and 'ow->height' are the output window size: does not change
21 	 * if window is clipping;
22 	 * 'ow->h_start' and 'ow->v_start' are the left-top position of the output
23 	 * window. These values can be negative: this means the window is clipping
24 	 * at the left or the top of the display, respectively. */
25 
26 	/* 'ov' is the view in the source bitmap, so which part of the bitmap is actually
27 	 * displayed on screen. This is used for the 'hardware zoom' function. */
28 
29 	/* bes setup data */
30 	nm_bes_data bi;
31 	/* misc used variables */
32 	uint16 temp1, temp2;
33 	/* BES output coordinate system for virtual workspaces */
34 	uint16 crtc_hstart, crtc_vstart, crtc_hend, crtc_vend;
35 	/* inverse scaling factor, used for source positioning */
36 	uint32 ifactor;
37 	/* copy of overlay view which has checked valid values */
38 	overlay_view my_ov;
39 
40 
41 	/**************************************************************************************
42 	 *** copy, check and limit if needed the user-specified view into the intput bitmap ***
43 	 **************************************************************************************/
44 	my_ov = *ov;
45 	/* check for valid 'coordinates' */
46 	if (my_ov.width == 0) my_ov.width++;
47 	if (my_ov.height == 0) my_ov.height++;
48 	if (my_ov.h_start > ((ob->width - si->overlay.myBufInfo[offset].slopspace) - 1))
49 		my_ov.h_start = ((ob->width - si->overlay.myBufInfo[offset].slopspace) - 1);
50 	if (((my_ov.h_start + my_ov.width) - 1) > ((ob->width - si->overlay.myBufInfo[offset].slopspace) - 1))
51 		my_ov.width = ((((ob->width - si->overlay.myBufInfo[offset].slopspace) - 1) - my_ov.h_start) + 1);
52 	if (my_ov.v_start > (ob->height - 1))
53 		my_ov.v_start = (ob->height - 1);
54 	if (((my_ov.v_start + my_ov.height) - 1) > (ob->height - 1))
55 		my_ov.height = (((ob->height - 1) - my_ov.v_start) + 1);
56 
57 	LOG(6,("Overlay: inputbuffer view (zoom) left %d, top %d, width %d, height %d\n",
58 		my_ov.h_start, my_ov.v_start, my_ov.width, my_ov.height));
59 
60 	/* the BES does not respect virtual_workspaces, but adheres to CRTC
61 	 * constraints only */
62 	crtc_hstart = si->dm.h_display_start;
63 	/* horizontal end is the first position beyond the displayed range on the CRTC */
64 	crtc_hend = crtc_hstart + si->dm.timing.h_display;
65 	crtc_vstart = si->dm.v_display_start;
66 	/* vertical end is the first position beyond the displayed range on the CRTC */
67 	crtc_vend = crtc_vstart + si->dm.timing.v_display;
68 
69 
70 	/****************************************
71 	 *** setup all edges of output window ***
72 	 ****************************************/
73 
74 	/* setup left and right edges of output window */
75 	bi.hcoordv = 0;
76 	/* left edge coordinate of output window, must be inside desktop */
77 	/* clipping on the left side */
78 	if (ow->h_start < crtc_hstart)
79 	{
80 		temp1 = 0;
81 	}
82 	else
83 	{
84 		/* clipping on the right side */
85 		if (ow->h_start >= (crtc_hend - 1))
86 		{
87 			/* width < 2 is not allowed */
88 			temp1 = (crtc_hend - crtc_hstart - 2);
89 		}
90 		else
91 		/* no clipping here */
92 		{
93 			temp1 = (ow->h_start - crtc_hstart);
94 		}
95 	}
96 	bi.hcoordv |= temp1 << 16;
97 	/* right edge coordinate of output window, must be inside desktop */
98 	/* width < 2 is not allowed */
99 	if (ow->width < 2)
100 	{
101 		temp2 = (temp1 + 1);
102 	}
103 	else
104 	{
105 		/* clipping on the right side */
106 		if ((ow->h_start + ow->width - 1) > (crtc_hend - 1))
107 		{
108 			temp2 = (crtc_hend - crtc_hstart - 1);
109 		}
110 		else
111 		{
112 			/* clipping on the left side */
113 			if ((ow->h_start + ow->width - 1) < (crtc_hstart + 1))
114 			{
115 				/* width < 2 is not allowed */
116 				temp2 = 1;
117 			}
118 			else
119 			/* no clipping here */
120 			{
121 				temp2 = ((uint16)(ow->h_start + ow->width - crtc_hstart - 1));
122 			}
123 		}
124 	}
125 	bi.hcoordv |= temp2 << 0;
126 	LOG(4,("Overlay: CRTC left-edge output %d, right-edge output %d\n",temp1, temp2));
127 
128 	/* setup top and bottom edges of output window */
129 	bi.vcoordv = 0;
130 	/* top edge coordinate of output window, must be inside desktop */
131 	/* clipping on the top side */
132 	if (ow->v_start < crtc_vstart)
133 	{
134 		temp1 = 0;
135 	}
136 	else
137 	{
138 		/* clipping on the bottom side */
139 		if (ow->v_start >= (crtc_vend - 1))
140 		{
141 			/* height < 2 is not allowed */
142 			temp1 = (crtc_vend - crtc_vstart - 2);
143 		}
144 		else
145 		/* no clipping here */
146 		{
147 			temp1 = (ow->v_start - crtc_vstart);
148 		}
149 	}
150 	bi.vcoordv |= temp1 << 16;
151 	/* bottom edge coordinate of output window, must be inside desktop */
152 	/* height < 2 is not allowed */
153 	if (ow->height < 2)
154 	{
155 		temp2 = (temp1 + 1);
156 	}
157 	else
158 	{
159 		/* clipping on the bottom side */
160 		if ((ow->v_start + ow->height - 1) > (crtc_vend - 1))
161 		{
162 			temp2 = (crtc_vend - crtc_vstart - 1);
163 		}
164 		else
165 		{
166 			/* clipping on the top side */
167 			if ((ow->v_start + ow->height - 1) < (crtc_vstart + 1))
168 			{
169 				/* height < 2 is not allowed */
170 				temp2 = 1;
171 			}
172 			else
173 			/* no clipping here */
174 			{
175 				temp2 = ((uint16)(ow->v_start + ow->height - crtc_vstart - 1));
176 			}
177 		}
178 	}
179 	bi.vcoordv |= temp2 << 0;
180 	LOG(4,("Overlay: CRTC top-edge output %d, bottom-edge output %d\n",temp1, temp2));
181 
182 
183 	/*********************************************
184 	 *** setup horizontal scaling and clipping ***
185 	 *********************************************/
186 
187 	LOG(6,("Overlay: total input picture width = %d, height = %d\n",
188 			(ob->width - si->overlay.myBufInfo[offset].slopspace), ob->height));
189 	LOG(6,("Overlay: output picture width = %d, height = %d\n", ow->width, ow->height));
190 
191 	/* calculate inverse horizontal scaling factor, taking zoom into account */
192 	ifactor = ((((uint32)my_ov.width) << 12) / ow->width);
193 
194 	/* correct factor to prevent most-right visible 'line' from distorting */
195 	ifactor -= 1;
196 	bi.hiscalv = ifactor;
197 	LOG(4,("Overlay: horizontal scaling factor is %f\n", (float)4096 / ifactor));
198 
199 	/* check scaling factor (and modify if needed) to be within scaling limits */
200 	if (((((uint32)my_ov.width) << 12) / 1024) > bi.hiscalv)
201 	{
202 		/* (non-inverse) factor too large, set factor to max. valid value */
203 		bi.hiscalv = ((((uint32)my_ov.width) << 12) / 1024);
204 		LOG(4,("Overlay: horizontal scaling factor too large, clamping at %f\n", (float)4096 / bi.hiscalv));
205 	}
206 	/* horizontal downscaling cannot be done by NM BES hardware */
207 	if (bi.hiscalv > (1 << 12))
208 	{
209 		/* (non-inverse) factor too small, set factor to min. valid value */
210 		bi.hiscalv = 0x1000;
211 		LOG(4,("Overlay: horizontal scaling factor too small, clamping at %f\n", (float)4096 / bi.hiscalv));
212 	}
213 
214 
215 	/* do horizontal clipping... */
216 	/* Setup horizontal source start: first (sub)pixel contributing to output picture */
217 	/* Note:
218 	 * The method is to calculate, based on 1:1 scaling, based on the output window.
219 	 * After this is done, include the scaling factor so you get a value based on the input bitmap.
220 	 * Then add the left starting position of the bitmap's view (zoom function) to get the final value needed.
221 	 * Note: The input bitmaps slopspace is automatically excluded from the calculations this way! */
222 	/* Note also:
223 	 * Even if the scaling factor is clamping we instruct the BES to use the correct source start pos.! */
224 	bi.hsrcstv = 0;
225 	/* check for destination horizontal clipping at left side */
226 	if (ow->h_start < crtc_hstart)
227 	{
228 		/* check if entire destination picture is clipping left:
229 		 * (2 pixels will be clamped onscreen at least) */
230 		if ((ow->h_start + ow->width - 1) < (crtc_hstart + 1))
231 		{
232 			/* increase 'first contributing pixel' with 'fixed value': (total dest. width - 2) */
233 			bi.hsrcstv += (ow->width - 2);
234 		}
235 		else
236 		{
237 			/* increase 'first contributing pixel' with actual number of dest. clipping pixels */
238 			bi.hsrcstv += (crtc_hstart - ow->h_start);
239 		}
240 		LOG(4,("Overlay: clipping left...\n"));
241 
242 		/* The calculated value is based on scaling = 1x. So we now compensate for scaling.
243 		 * Note that this also already takes care of aligning the value to the BES register! */
244 		bi.hsrcstv *= (ifactor << 4);
245 	}
246 	/* take zoom into account */
247 	bi.hsrcstv += ((uint32)my_ov.h_start) << 16;
248 	/* AND below required by hardware */
249 	bi.hsrcstv &= 0x03fffffc;
250 	LOG(4,("Overlay: first hor. (sub)pixel of input bitmap contributing %f\n", bi.hsrcstv / (float)65536));
251 
252 
253 	/* Setup horizontal source end: last (sub)pixel contributing to output picture */
254 	/* Note:
255 	 * The method is to calculate, based on 1:1 scaling, based on the output window.
256 	 * After this is done, include the scaling factor so you get a value based on the input bitmap.
257 	 * Then add the right ending position of the bitmap's view (zoom function) to get the final value needed. */
258 	/* Note also:
259 	 * Even if the scaling factor is clamping we instruct the BES to use the correct source end pos.! */
260 
261 	bi.hsrcendv = 0;
262 	/* check for destination horizontal clipping at right side */
263 	if ((ow->h_start + ow->width - 1) > (crtc_hend - 1))
264 	{
265 		/* check if entire destination picture is clipping right:
266 		 * (2 pixels will be clamped onscreen at least) */
267 		if (ow->h_start > (crtc_hend - 2))
268 		{
269 			/* increase 'number of clipping pixels' with 'fixed value': (total dest. width - 2) */
270 			bi.hsrcendv += (ow->width - 2);
271 		}
272 		else
273 		{
274 			/* increase 'number of clipping pixels' with actual number of dest. clipping pixels */
275 			bi.hsrcendv += ((ow->h_start + ow->width - 1) - (crtc_hend - 1));
276 		}
277 		LOG(4,("Overlay: clipping right...\n"));
278 
279 		/* The calculated value is based on scaling = 1x. So we now compensate for scaling.
280 		 * Note that this also already takes care of aligning the value to the BES register! */
281 		bi.hsrcendv *= (ifactor << 4);
282 		/* now subtract this value from the last used pixel in (zoomed) inputbuffer, aligned to BES */
283 		bi.hsrcendv = (((uint32)((my_ov.h_start + my_ov.width) - 1)) << 16) - bi.hsrcendv;
284 	}
285 	else
286 	{
287 		/* set last contributing pixel to last used pixel in (zoomed) inputbuffer, aligned to BES */
288 		bi.hsrcendv = (((uint32)((my_ov.h_start + my_ov.width) - 1)) << 16);
289 	}
290 	/* AND below required by hardware */
291 	bi.hsrcendv &= 0x03fffffc;
292 	LOG(4,("Overlay: last horizontal (sub)pixel of input bitmap contributing %f\n", bi.hsrcendv / (float)65536));
293 
294 
295 	/* setup horizontal source last position excluding slopspace:
296 	 * this is the last pixel that will be used for calculating interpolated pixels */
297 	bi.hsrclstv = ((ob->width - 1) - si->overlay.myBufInfo[offset].slopspace) << 16;
298 	/* AND below required by hardware */
299 	bi.hsrclstv &= 0x03ff0000;
300 
301 
302 	/*******************************************
303 	 *** setup vertical scaling and clipping ***
304 	 *******************************************/
305 
306 	/* calculate inverse vertical scaling factor, taking zoom into account */
307 	ifactor = ((((uint32)my_ov.height) << 12) / ow->height);
308 
309 	/* correct factor to prevent lowest visible line from distorting */
310 	ifactor -= 1;
311 	LOG(4,("Overlay: vertical scaling factor is %f\n", (float)4096 / ifactor));
312 
313 	/* preserve ifactor for source positioning calculations later on */
314 	bi.viscalv = ifactor;
315 
316 	/* check scaling factor (and modify if needed) to be within scaling limits */
317 	if (((((uint32)my_ov.height) << 12) / 1024) > bi.viscalv)
318 	{
319 		/* (non-inverse) factor too large, set factor to max. valid value */
320 		bi.viscalv = ((((uint32)my_ov.height) << 12) / 1024);
321 		LOG(4,("Overlay: vertical scaling factor too large, clamping at %f\n", (float)4096 / bi.viscalv));
322 	}
323 	/* vertical downscaling cannot be done by NM BES hardware */
324 	if (bi.viscalv > (1 << 12))
325 	{
326 		/* (non-inverse) factor too small, set factor to min. valid value */
327 		bi.viscalv = 0x1000;
328 		LOG(4,("Overlay: vertical scaling factor too small, clamping at %f\n", (float)4096 / bi.viscalv));
329 	}
330 
331 
332 	/* do vertical clipping... */
333 	/* Setup vertical source start: first (sub)pixel contributing to output picture.
334 	 * Note: this exists of two parts:
335 	 * 1. setup fractional part (sign is always 'positive');
336 	 * 2. setup relative base_adress, taking clipping on top (and zoom) into account.
337 	 * Both parts are done intertwined below. */
338 	/* Note:
339 	 * The method is to calculate, based on 1:1 scaling, based on the output window.
340 	 * 'After' this is done, include the scaling factor so you get a value based on the input bitmap.
341 	 * Then add the top starting position of the bitmap's view (zoom function) to get the final value needed. */
342 	/* Note also:
343 	 * Even if the scaling factor is clamping we instruct the BES to use the correct source start pos.! */
344 
345 	/* calculate relative base_adress and 'vertical weight fractional part' */
346 	bi.weight = 0;
347 	bi.a1orgv = (uint32)((vuint32 *)ob->buffer);
348 	bi.a1orgv -= (uint32)((vuint32 *)si->framebuffer);
349 	/* calculate origin adress */
350 	LOG(4,("Overlay: topleft corner of input bitmap (cardRAM offset) $%08x\n",bi.a1orgv));
351 	/* check for destination vertical clipping at top side */
352 	if (ow->v_start < crtc_vstart)
353 	{
354 		/* check if entire destination picture is clipping at top:
355 		 * (2 pixels will be clamped onscreen at least) */
356 		if ((ow->v_start + ow->height - 1) < (crtc_vstart + 1))
357 		{
358 			/* increase source buffer origin with 'fixed value':
359 			 * (integer part of ('total height - 2' of dest. picture in pixels * inverse scaling factor)) *
360 			 * bytes per row source picture */
361 			bi.weight = (ow->height - 2) * (ifactor << 4);
362 			bi.a1orgv += ((bi.weight >> 16) * ob->bytes_per_row);
363 		}
364 		else
365 		{
366 			/* increase source buffer origin with:
367 			 * (integer part of (number of destination picture clipping pixels * inverse scaling factor)) *
368 			 * bytes per row source picture */
369 			bi.weight = (crtc_vstart - ow->v_start) * (ifactor << 4);
370 			bi.a1orgv += ((bi.weight >> 16) * ob->bytes_per_row);
371 		}
372 		LOG(4,("Overlay: clipping at top...\n"));
373 	}
374 	/* take zoom into account */
375 	bi.a1orgv += (my_ov.v_start * ob->bytes_per_row);
376 	bi.weight += (((uint32)my_ov.v_start) << 16);
377 	LOG(4,("Overlay: 'contributing part of buffer' origin is (cardRAM offset) $%08x\n",bi.a1orgv));
378 	LOG(4,("Overlay: first vert. (sub)pixel of input bitmap contributing %f\n", bi.weight / (float)65536));
379 
380 	/* AND below is required by NM hardware. */
381 	/* Buffer A topleft corner of field 1 (origin)(field 1 contains our full frames) */
382 	bi.a1orgv &= 0x00fffff0;
383 
384 	/* field 1 weight: AND below required by hardware, also make sure 'sign' is always 'positive' */
385 	bi.v1wghtv = bi.weight & 0x0000fffc;
386 
387 
388 	/* setup field 1 (is our complete frame) vertical source last position.
389 	 * this is the last pixel that will be used for calculating interpolated pixels */
390 	bi.v1srclstv = (ob->height - 1);
391 	/* AND below required by hardware */
392 	bi.v1srclstv &= 0x000003ff;
393 
394 
395 	/*****************************
396 	 *** log color keying info ***
397 	 *****************************/
398 
399 	LOG(6,("Overlay: key_red %d, key_green %d, key_blue %d, key_alpha %d\n",
400 		ow->red.value, ow->green.value, ow->blue.value, ow->alpha.value));
401 	LOG(6,("Overlay: mask_red %d, mask_green %d, mask_blue %d, mask_alpha %d\n",
402 		ow->red.mask, ow->green.mask, ow->blue.mask, ow->alpha.mask));
403 
404 
405 	/*************************
406 	 *** setup BES control ***
407 	 *************************/
408 
409 	/* BES global control: setup functions */
410 	bi.globctlv = 0;
411 
412 	/* enable BES */
413 	bi.globctlv |= 1 << 0;
414 	/* enable colorkeying */
415 	bi.globctlv |= 1 << 1;
416 	/* b3 = 1: distorts right-half of overlay output. Keeping it zero. */
417 	/* colorspace is YV12, I420 or YUY2 (no RV15 or RV16) */
418 	bi.globctlv |= 0 << 5;
419 
420 	/* enable auto-alternating hardware buffers if alternating buffers is enabled (NM2160) */
421 	bi.globctlv |= 1 << 8;
422 	/* ??? */
423 	bi.globctlv |= 1 << 13;
424 	/* display one buffer (no alternating buffers) (NM2160: no effect) */
425 	bi.globctlv |= 0 << 14;
426 	/* display frame (no field) (NM2160: no effect) */
427 	bi.globctlv |= 0 << 15;
428 
429 	/* BTW: horizontal and vertical filtering are always turned on in NM hardware. */
430 
431 
432 	/*************************************
433 	 *** sync to BES (Back End Scaler) ***
434 	 *************************************/
435 
436 	/* Make sure reprogramming the BES completes before the next retrace occurs,
437 	 * to prevent register-update glitches (double buffer feature). */
438 
439 	//fixme...
440 
441 //	LOG(3,("Overlay: starting register programming beyond Vcount %d\n", CR1R(VCOUNT)));
442 	/* Even at 1600x1200x90Hz, a single line still takes about 9uS to complete:
443 	 * this resolution will generate about 180Mhz pixelclock while we can do
444 	 * upto 360Mhz. So snooze about 4uS to prevent bus-congestion...
445 	 * Appr. 200 lines time will provide enough room even on a 100Mhz CPU if it's
446 	 * screen is set to the highest refreshrate/resolution possible. */
447 //	while (CR1R(VCOUNT) > (si->dm.timing.v_total - 200)) snooze(4);
448 
449 
450 	/**************************************
451 	 *** actually program the registers ***
452 	 **************************************/
453 
454 	if (si->ps.card_type >= NM2097)
455 	{
456 		/* PCI card */
457 		LOG(4,("Overlay: accelerant is programming BES\n"));
458 		/* unlock card overlay sequencer registers (b5 = 1) */
459 		PCIGRPHW(GENLOCK, (PCIGRPHR(GENLOCK) | 0x20));
460 		/* destination rectangle */
461 		PCIGRPHW(HDCOORD1L, ((bi.hcoordv >> 16) & 0xff));
462 		PCIGRPHW(HDCOORD2L, (bi.hcoordv & 0xff));
463 		PCIGRPHW(HDCOORD21H, (((bi.hcoordv >> 4) & 0xf0) | ((bi.hcoordv >> 24) & 0x0f)));
464 		PCIGRPHW(VDCOORD1L, ((bi.vcoordv >> 16) & 0xff));
465 		PCIGRPHW(VDCOORD2L, (bi.vcoordv & 0xff));
466 		PCIGRPHW(VDCOORD21H, (((bi.vcoordv >> 4) & 0xf0) | ((bi.vcoordv >> 24) & 0x0f)));
467 		/* scaling */
468 		PCIGRPHW(XSCALEL, (bi.hiscalv & 0xff));
469 		PCIGRPHW(XSCALEH, ((bi.hiscalv >> 8) & 0xff));
470 		PCIGRPHW(YSCALEL, (bi.viscalv & 0xff));
471 		PCIGRPHW(YSCALEH, ((bi.viscalv >> 8) & 0xff));
472 		/* inputbuffer #1 origin */
473 		/* (we don't program buffer #2 as it's unused.) */
474 		/* first include 'pixel precise' left clipping...
475 		 * (subpixel precision is not supported by NeoMagic cards) */
476 		bi.a1orgv += ((bi.hsrcstv >> 16) * 2);
477 		/* we need to step in 4-byte (2 pixel) granularity due to the nature of yuy2 */
478 		bi.a1orgv &= ~0x03;
479 		/* now setup buffer startadress and horizontal source end (minimizes used bandwidth) */
480 		if (si->ps.card_type < NM2200)
481 		{
482 			bi.a1orgv >>= 1;
483 			/* horizontal source end does not use subpixelprecision: granularity is 8 pixels */
484 			PCIGRPHW(0xbc, (((((bi.hsrcendv >> 16) + 7) & ~8) / 8) - 1));
485 		}
486 		else
487 		{
488 			/* horizontal source end does not use subpixelprecision: granularity is 16 pixels */
489 			//fixme? divide by 16 instead of 8 (if >= NM2200 owners report trouble then use 8!)
490 			//fixme? check if overlaybuffer width should also have granularity of 16 now!
491 			PCIGRPHW(0xbc, (((((bi.hsrcendv >> 16) + 15) & ~16) / 16) - 1));
492 		}
493 		PCIGRPHW(BUF1ORGL, (bi.a1orgv & 0xff));
494 		PCIGRPHW(BUF1ORGM, ((bi.a1orgv >> 8) & 0xff));
495 		PCIGRPHW(BUF1ORGH, ((bi.a1orgv >> 16) & 0xff));
496 		/* b2 = 0: don't use horizontal mirroring (NM2160) */
497 		/* other bits do ??? */
498 		PCIGRPHW(0xbf, 0x02);
499 		/* ??? */
500 		PCIGRPHW(0xbd, 0x02);
501 		PCIGRPHW(0xbe, 0x00);
502 		/* (subpixel precise) source rect clipping is not supported on NeoMagic cards;
503 		 * so we do 'pixel precise' left clipping via modification of buffer
504 		 * startadress above instead.
505 		 * (pixel precise top clipping is also done this way..) */
506 		//fixme: checkout real pixel precise clipping on NM2200 and later cards!!!
507 /*
508 	{
509 		uint16 left = 0;
510 		uint16 right = 128;
511 		uint16 top = 0;
512 		uint16 bottom = 128;
513 
514 		left = (bi.hsrcstv >> 16);
515 		right = (bi.hsrclstv >> 16);
516 		top = (bi.weight >> 16);
517 		bottom = (bi.v1srclstv >> 16);
518 
519 		PCISEQW(HSCOORD1L, (left & 0xff));
520 		PCISEQW(HSCOORD2L, (right & 0xff));
521 		PCISEQW(HSCOORD21H, (((right >> 4) & 0xf0) | ((left >> 8) & 0x0f)));
522 		PCISEQW(VSCOORD1L, (top & 0xff));
523 		PCISEQW(VSCOORD2L, (bottom & 0xff));
524 		PCISEQW(VSCOORD21H, (((bottom >> 4) & 0xf0) | ((top >> 8) & 0x0f)));
525 	}
526 */
527 		/* ??? */
528 	    PCISEQW(0x1c, 0xfb);
529     	PCISEQW(0x1d, 0x00);
530 		PCISEQW(0x1e, 0xe2);
531     	PCISEQW(0x1f, 0x02);
532  		/* b1 = 0: disable alternating hardware buffers (NM2160) */
533 		/* other bits do ??? */
534  		PCISEQW(0x09, 0x11);
535 		/* ??? */
536 		PCISEQW(0x0a, 0x00);
537 		/* global BES control */
538 		PCIGRPHW(BESCTRL1, (bi.globctlv & 0xff));
539 		PCISEQW(BESCTRL2, ((bi.globctlv >> 8) & 0xff));
540 
541 
542 		/**************************
543 		 *** setup color keying ***
544 		 **************************/
545 
546 		PCIGRPHW(COLKEY_R, (ow->red.value & ow->red.mask));
547 		PCIGRPHW(COLKEY_G, (ow->green.value & ow->green.mask));
548 		PCIGRPHW(COLKEY_B, (ow->blue.value & ow->blue.mask));
549 
550 
551 		/*************************
552 		 *** setup misc. stuff ***
553 		 *************************/
554 
555 		/* setup brightness to be 'neutral' (two's complement number) */
556 		PCIGRPHW(BRIGHTNESS, 0x00);
557 
558 		/* setup inputbuffer #1 pitch including slopspace (in pixels) */
559 		/* (we don't program the pitch for inputbuffer #2 as it's unused.) */
560 		PCIGRPHW(BUF1PITCHL, (ob->width & 0xff));
561 		PCIGRPHW(BUF1PITCHH, ((ob->width >> 8) & 0xff));
562 	}
563 	else
564 	{
565 		/* ISA card. Speed required, so:
566 		 * program entire sequence in kerneldriver in one context switch! */
567 		LOG(4,("Overlay: kerneldriver programs BES\n"));
568 
569 		/* complete BES info struct... */
570 		bi.card_type = si->ps.card_type;
571 		bi.colkey_r = (ow->red.value & ow->red.mask);
572 		bi.colkey_g = (ow->green.value & ow->green.mask);
573 		bi.colkey_b = (ow->blue.value & ow->blue.mask);
574 		bi.ob_width = ob->width;
575 		/* ... and call kerneldriver to program the BES */
576 		bi.magic = NM_PRIVATE_DATA_MAGIC;
577 		ioctl(fd, NM_PGM_BES, &bi, sizeof(bi));
578 	}
579 
580 	/* on a 500Mhz P3 CPU just logging a line costs 400uS (18-19 vcounts at 1024x768x60Hz)!
581 	 * programming the registers above actually costs 180uS here */
582 //	LOG(3,("Overlay: completed at Vcount %d\n", CR1R(VCOUNT)));
583 
584 	return B_OK;
585 }
586 
587 status_t nm_release_bes()
588 {
589 	/* setup BES control: disable scaler */
590 	if (si->ps.card_type >= NM2097)
591 	{
592 		/* PCI card */
593 		PCIGRPHW(BESCTRL1, 0x02);
594 		PCISEQW(BESCTRL2, 0xa0);
595 	}
596 	else
597 	{
598 		/* ISA card */
599 		ISAGRPHW(BESCTRL1, 0x02);
600 		ISASEQW(BESCTRL2, 0xa0);
601 	}
602 
603 	return B_OK;
604 }
605