xref: /haiku/src/add-ons/accelerants/intel_extreme/overlay.cpp (revision 2600324b57fa31cdea1627d584d314f2a579c4a8)
1 /*
2  * Copyright 2006-2008, Haiku, Inc. All Rights Reserved.
3  * Distributed under the terms of the MIT License.
4  *
5  * Authors:
6  *		Axel Dörfler, axeld@pinc-software.de
7  *
8  * The phase coefficient computation was taken from the X driver written by
9  * Alan Hourihane and David Dawes.
10  */
11 
12 
13 #include "accelerant.h"
14 #include "accelerant_protos.h"
15 #include "commands.h"
16 
17 #include <math.h>
18 #include <stdlib.h>
19 #include <string.h>
20 
21 #include <AGP.h>
22 
23 
24 //#define TRACE_OVERLAY
25 #ifdef TRACE_OVERLAY
26 extern "C" void _sPrintf(const char *format, ...);
27 #	define TRACE(x) _sPrintf x
28 #else
29 #	define TRACE(x) ;
30 #endif
31 
32 
33 #define NUM_HORIZONTAL_TAPS		5
34 #define NUM_VERTICAL_TAPS		3
35 #define NUM_HORIZONTAL_UV_TAPS	3
36 #define NUM_VERTICAL_UV_TAPS	3
37 #define NUM_PHASES				17
38 #define MAX_TAPS				5
39 
40 struct phase_coefficient {
41 	uint8	sign;
42 	uint8	exponent;
43 	uint16	mantissa;
44 };
45 
46 
47 /*!	Splits the coefficient floating point value into the 3 components
48 	sign, mantissa, and exponent.
49 */
50 static bool
51 split_coefficient(double &coefficient, int32 mantissaSize,
52 	phase_coefficient &splitCoefficient)
53 {
54 	double absCoefficient = fabs(coefficient);
55 
56 	int sign;
57 	if (coefficient < 0.0)
58 		sign = 1;
59 	else
60 		sign = 0;
61 
62 	int32 intCoefficient, res;
63 	int32 maxValue = 1 << mantissaSize;
64 	res = 12 - mantissaSize;
65 
66 	if ((intCoefficient = (int)(absCoefficient * 4 * maxValue + 0.5)) < maxValue) {
67 		splitCoefficient.exponent = 3;
68 		splitCoefficient.mantissa = intCoefficient << res;
69 		coefficient = (double)intCoefficient / (double)(4 * maxValue);
70 	} else if ((intCoefficient = (int)(absCoefficient * 2 * maxValue + 0.5)) < maxValue) {
71 		splitCoefficient.exponent = 2;
72 		splitCoefficient.mantissa = intCoefficient << res;
73 		coefficient = (double)intCoefficient / (double)(2 * maxValue);
74 	} else if ((intCoefficient = (int)(absCoefficient * maxValue + 0.5)) < maxValue) {
75 		splitCoefficient.exponent = 1;
76 		splitCoefficient.mantissa = intCoefficient << res;
77 		coefficient = (double)intCoefficient / (double)maxValue;
78 	} else if ((intCoefficient = (int)(absCoefficient * maxValue * 0.5 + 0.5)) < maxValue) {
79 		splitCoefficient.exponent = 0;
80 		splitCoefficient.mantissa = intCoefficient << res;
81 		coefficient = (double)intCoefficient / (double)(maxValue / 2);
82 	} else {
83 		// coefficient out of range
84 		return false;
85 	}
86 
87 	splitCoefficient.sign = sign;
88 	if (sign)
89 		coefficient = -coefficient;
90 
91 	return true;
92 }
93 
94 
95 static void
96 update_coefficients(int32 taps, double filterCutOff, bool horizontal, bool isY,
97 	phase_coefficient *splitCoefficients)
98 {
99 	if (filterCutOff < 1)
100 		filterCutOff = 1;
101 	if (filterCutOff > 3)
102 		filterCutOff = 3;
103 
104 	bool isVerticalUV = !horizontal && !isY;
105 	int32 mantissaSize = horizontal ? 7 : 6;
106 
107 	double rawCoefficients[MAX_TAPS * 32], coefficients[NUM_PHASES][MAX_TAPS];
108 
109 	int32 num = taps * 16;
110 	for (int32 i = 0; i < num * 2; i++) {
111 		double sinc;
112 		double value = (1.0 / filterCutOff) * taps * PI * (i - num) / (2 * num);
113 		if (value == 0.0)
114 			sinc = 1.0;
115 		else
116 			sinc = sin(value) / value;
117 
118 		// Hamming window
119 		double window = (0.5 - 0.5 * cos(i * PI / num));
120 		rawCoefficients[i] = sinc * window;
121 	}
122 
123 	for (int32 i = 0; i < NUM_PHASES; i++) {
124 		// Normalise the coefficients
125 		double sum = 0.0;
126 		int32 pos;
127 		for (int32 j = 0; j < taps; j++) {
128 			pos = i + j * 32;
129 			sum += rawCoefficients[pos];
130 		}
131 		for (int32 j = 0; j < taps; j++) {
132 			pos = i + j * 32;
133 			coefficients[i][j] = rawCoefficients[pos] / sum;
134 		}
135 
136 		// split them into sign/mantissa/exponent
137 		for (int32 j = 0; j < taps; j++) {
138 			pos = j + i * taps;
139 
140 			split_coefficient(coefficients[i][j], mantissaSize
141 				+ (((j == (taps - 1) / 2) && !isVerticalUV) ? 2 : 0),
142 				splitCoefficients[pos]);
143 		}
144 
145 		int32 tapAdjust[MAX_TAPS];
146 		tapAdjust[0] = (taps - 1) / 2;
147 		for (int32 j = 1, k = 1; j <= tapAdjust[0]; j++, k++) {
148 			tapAdjust[k] = tapAdjust[0] - j;
149 			tapAdjust[++k] = tapAdjust[0] + j;
150 		}
151 
152 		// Adjust the coefficients
153 		sum = 0.0;
154 		for (int32 j = 0; j < taps; j++) {
155 			sum += coefficients[i][j];
156 		}
157 
158 		if (sum != 1.0) {
159 			for (int32 k = 0; k < taps; k++) {
160 				int32 tap2Fix = tapAdjust[k];
161 				double diff = 1.0 - sum;
162 
163 				coefficients[i][tap2Fix] += diff;
164 				pos = tap2Fix + i * taps;
165 
166 				split_coefficient(coefficients[i][tap2Fix], mantissaSize
167 					+ (((tap2Fix == (taps - 1) / 2) && !isVerticalUV) ? 2 : 0),
168 					splitCoefficients[pos]);
169 
170 				sum = 0.0;
171 				for (int32 j = 0; j < taps; j++) {
172 					sum += coefficients[i][j];
173 				}
174 				if (sum == 1.0)
175 					break;
176 			}
177 		}
178 	}
179 }
180 
181 
182 static void
183 set_color_key(uint8 red, uint8 green, uint8 blue, uint8 redMask,
184 	uint8 greenMask, uint8 blueMask)
185 {
186 	overlay_registers *registers = gInfo->overlay_registers;
187 
188 	registers->color_key_red = red;
189 	registers->color_key_green = green;
190 	registers->color_key_blue = blue;
191 	registers->color_key_mask_red = ~redMask;
192 	registers->color_key_mask_green = ~greenMask;
193 	registers->color_key_mask_blue = ~blueMask;
194 	registers->color_key_enabled = true;
195 }
196 
197 
198 static void
199 set_color_key(const overlay_window *window)
200 {
201 	switch (gInfo->shared_info->current_mode.space) {
202 		case B_CMAP8:
203 			set_color_key(0, 0, window->blue.value, 0x0, 0x0, 0xff);
204 			break;
205 		case B_RGB15:
206 			set_color_key(window->red.value << 3, window->green.value << 3,
207 				window->blue.value << 3, window->red.mask << 3, window->green.mask << 3,
208 				window->blue.mask << 3);
209 			break;
210 		case B_RGB16:
211 			set_color_key(window->red.value << 3, window->green.value << 2,
212 				window->blue.value << 3, window->red.mask << 3, window->green.mask << 2,
213 				window->blue.mask << 3);
214 			break;
215 
216 		default:
217 			set_color_key(window->red.value, window->green.value,
218 				window->blue.value, window->red.mask, window->green.mask,
219 				window->blue.mask);
220 			break;
221 	}
222 }
223 
224 
225 static void
226 update_overlay(bool updateCoefficients)
227 {
228 	if (!gInfo->shared_info->overlay_active
229 		|| gInfo->shared_info->device_type == INTEL_TYPE_965)
230 		return;
231 
232 	QueueCommands queue(gInfo->shared_info->primary_ring_buffer);
233 	queue.PutFlush();
234 	queue.PutWaitFor(COMMAND_WAIT_FOR_OVERLAY_FLIP);
235 	queue.PutOverlayFlip(COMMAND_OVERLAY_CONTINUE, updateCoefficients);
236 
237 	// make sure the flip is done now
238 	queue.PutWaitFor(COMMAND_WAIT_FOR_OVERLAY_FLIP);
239 	queue.PutFlush();
240 
241 	TRACE(("update overlay: UP: %lx, TST: %lx, ST: %lx, CMD: %lx (%lx), ERR: %lx\n",
242 		read32(INTEL_OVERLAY_UPDATE), read32(INtEL_OVERLAY_TEST), read32(INTEL_OVERLAY_STATUS),
243 		*(((uint32 *)gInfo->overlay_registers) + 0x68/4), read32(0x30168), read32(0x2024)));
244 }
245 
246 
247 static void
248 show_overlay(void)
249 {
250 	if (gInfo->shared_info->overlay_active
251 		|| gInfo->shared_info->device_type == INTEL_TYPE_965)
252 		return;
253 
254 	gInfo->shared_info->overlay_active = true;
255 	gInfo->overlay_registers->overlay_enabled = true;
256 
257 	QueueCommands queue(gInfo->shared_info->primary_ring_buffer);
258 	queue.PutOverlayFlip(COMMAND_OVERLAY_ON, true);
259 	queue.PutFlush();
260 
261 	TRACE(("show overlay: UP: %lx, TST: %lx, ST: %lx, CMD: %lx (%lx), ERR: %lx\n",
262 		read32(INTEL_OVERLAY_UPDATE), read32(INTEL_OVERLAY_TEST), read32(INTEL_OVERLAY_STATUS),
263 		*(((uint32 *)gInfo->overlay_registers) + 0x68/4), read32(0x30168), read32(0x2024)));
264 }
265 
266 
267 static void
268 hide_overlay(void)
269 {
270 	if (!gInfo->shared_info->overlay_active
271 		|| gInfo->shared_info->device_type == INTEL_TYPE_965)
272 		return;
273 
274 	overlay_registers *registers = gInfo->overlay_registers;
275 
276 	gInfo->shared_info->overlay_active = false;
277 	registers->overlay_enabled = false;
278 
279 	QueueCommands queue(gInfo->shared_info->primary_ring_buffer);
280 
281 	// flush pending commands
282 	queue.PutFlush();
283 	queue.PutWaitFor(COMMAND_WAIT_FOR_OVERLAY_FLIP);
284 
285 	// clear overlay enabled bit
286 	queue.PutOverlayFlip(COMMAND_OVERLAY_CONTINUE, false);
287 	queue.PutWaitFor(COMMAND_WAIT_FOR_OVERLAY_FLIP);
288 
289 	// turn off overlay engine
290 	queue.PutOverlayFlip(COMMAND_OVERLAY_OFF, false);
291 	queue.PutWaitFor(COMMAND_WAIT_FOR_OVERLAY_FLIP);
292 
293 	gInfo->current_overlay = NULL;
294 }
295 
296 
297 //	#pragma mark -
298 
299 
300 uint32
301 intel_overlay_count(const display_mode *mode)
302 {
303 	// TODO: make this depending on the amount of RAM and the screen mode
304 	// (and we could even have more than one when using 3D as well)
305 	return 1;
306 }
307 
308 
309 const uint32 *
310 intel_overlay_supported_spaces(const display_mode *mode)
311 {
312 	static const uint32 kSupportedSpaces[] = {B_RGB15, B_RGB16, B_RGB32,
313 		B_YCbCr422, 0};
314 	static const uint32 kSupportedi965Spaces[] = {B_YCbCr422, 0};
315 	intel_shared_info &sharedInfo = *gInfo->shared_info;
316 
317 	if (sharedInfo.device_type == INTEL_TYPE_965)
318 		return kSupportedi965Spaces;
319 
320 	return kSupportedSpaces;
321 }
322 
323 
324 uint32
325 intel_overlay_supported_features(uint32 colorSpace)
326 {
327 	return B_OVERLAY_COLOR_KEY
328 		| B_OVERLAY_HORIZONTAL_FILTERING
329 		| B_OVERLAY_VERTICAL_FILTERING
330 		| B_OVERLAY_HORIZONTAL_MIRRORING;
331 }
332 
333 
334 const overlay_buffer *
335 intel_allocate_overlay_buffer(color_space colorSpace, uint16 width,
336 	uint16 height)
337 {
338 	TRACE(("intel_allocate_overlay_buffer(width %u, height %u, colorSpace %lu)\n",
339 		width, height, colorSpace));
340 
341 	intel_shared_info &sharedInfo = *gInfo->shared_info;
342 	uint32 bytesPerPixel;
343 
344 	switch (colorSpace) {
345 		case B_RGB15:
346 			bytesPerPixel = 2;
347 			break;
348 		case B_RGB16:
349 			bytesPerPixel = 2;
350 			break;
351 		case B_RGB32:
352 			bytesPerPixel = 4;
353 			break;
354 		case B_YCbCr422:
355 			bytesPerPixel = 2;
356 			break;
357 		default:
358 			return NULL;
359 	}
360 
361 	struct overlay *overlay = (struct overlay *)malloc(sizeof(struct overlay));
362 	if (overlay == NULL)
363 		return NULL;
364 
365 	// TODO: locking!
366 
367 	// alloc graphics mem
368 
369 	int32 alignment = 0x3f;
370 	if (sharedInfo.device_type == INTEL_TYPE_965)
371 		alignment = 0xff;
372 
373 	overlay_buffer *buffer = &overlay->buffer;
374 	buffer->space = colorSpace;
375 	buffer->width = width;
376 	buffer->height = height;
377 	buffer->bytes_per_row = (width * bytesPerPixel + alignment) & ~alignment;
378 
379 	status_t status = intel_allocate_memory(buffer->bytes_per_row * height,
380 		0, overlay->buffer_base);
381 	if (status < B_OK) {
382 		free(overlay);
383 		return NULL;
384 	}
385 
386 	if (sharedInfo.device_type == INTEL_TYPE_965) {
387 		status = intel_allocate_memory(INTEL_i965_OVERLAY_STATE_SIZE,
388 			B_APERTURE_NON_RESERVED, overlay->state_base);
389 		if (status < B_OK) {
390 			intel_free_memory(overlay->buffer_base);
391 			free(overlay);
392 			return NULL;
393 		}
394 
395 		overlay->state_offset = overlay->state_base
396 			- (addr_t)gInfo->shared_info->graphics_memory;
397 	}
398 
399 	overlay->buffer_offset = overlay->buffer_base
400 		- (addr_t)gInfo->shared_info->graphics_memory;
401 
402 	buffer->buffer = (uint8 *)overlay->buffer_base;
403 	buffer->buffer_dma = (uint8 *)gInfo->shared_info->physical_graphics_memory
404 		+ overlay->buffer_offset;
405 
406 	TRACE(("allocated overlay buffer: base=%x, offset=%x, address=%x, "
407 		"physical address=%x\n", overlay->buffer_base, overlay->buffer_offset,
408 		buffer->buffer, buffer->buffer_dma));
409 
410 	return buffer;
411 }
412 
413 
414 status_t
415 intel_release_overlay_buffer(const overlay_buffer *buffer)
416 {
417 	TRACE(("intel_release_overlay_buffer(buffer %p)\n", buffer));
418 
419 	struct overlay *overlay = (struct overlay *)buffer;
420 
421 	// TODO: locking!
422 
423 	if (gInfo->current_overlay == overlay)
424 		hide_overlay();
425 
426 	intel_free_memory(overlay->buffer_base);
427 	if (gInfo->shared_info->device_type == INTEL_TYPE_965)
428 		intel_free_memory(overlay->state_base);
429 	free(overlay);
430 
431 	return B_OK;
432 }
433 
434 
435 status_t
436 intel_get_overlay_constraints(const display_mode *mode,
437 	const overlay_buffer *buffer, overlay_constraints *constraints)
438 {
439 	TRACE(("intel_get_overlay_constraints(buffer %p)\n", buffer));
440 
441 	// taken from the Radeon driver...
442 
443 	// scaler input restrictions
444 	// TODO: check all these values; most of them are probably too restrictive
445 
446 	// position
447 	constraints->view.h_alignment = 0;
448 	constraints->view.v_alignment = 0;
449 
450 	// alignment
451 	switch (buffer->space) {
452 		case B_RGB15:
453 			constraints->view.width_alignment = 7;
454 			break;
455 		case B_RGB16:
456 			constraints->view.width_alignment = 7;
457 			break;
458 		case B_RGB32:
459 			constraints->view.width_alignment = 3;
460 			break;
461 		case B_YCbCr422:
462 			constraints->view.width_alignment = 7;
463 			break;
464 		case B_YUV12:
465 			constraints->view.width_alignment = 7;
466 		default:
467 			return B_BAD_VALUE;
468 	}
469 	constraints->view.height_alignment = 0;
470 
471 	// size
472 	constraints->view.width.min = 4;		// make 4-tap filter happy
473 	constraints->view.height.min = 4;
474 	constraints->view.width.max = buffer->width;
475 	constraints->view.height.max = buffer->height;
476 
477 	// scaler output restrictions
478 	constraints->window.h_alignment = 0;
479 	constraints->window.v_alignment = 0;
480 	constraints->window.width_alignment = 0;
481 	constraints->window.height_alignment = 0;
482 	constraints->window.width.min = 2;
483 	constraints->window.width.max = mode->virtual_width;
484 	constraints->window.height.min = 2;
485 	constraints->window.height.max = mode->virtual_height;
486 
487 	// TODO: the minimum values are not tested
488 	constraints->h_scale.min = 1.0f / (1 << 4);
489 	constraints->h_scale.max = buffer->width * 7;
490 	constraints->v_scale.min = 1.0f / (1 << 4);
491 	constraints->v_scale.max = buffer->height * 7;
492 
493 	return B_OK;
494 }
495 
496 
497 overlay_token
498 intel_allocate_overlay(void)
499 {
500 	TRACE(("intel_allocate_overlay()\n"));
501 
502 	// we only have a single overlay channel
503 	if (atomic_or(&gInfo->shared_info->overlay_channel_used, 1) != 0)
504 		return NULL;
505 
506 	return (overlay_token)++gInfo->shared_info->overlay_token;
507 }
508 
509 
510 status_t
511 intel_release_overlay(overlay_token overlayToken)
512 {
513 	TRACE(("intel_allocate_overlay(token %ld)\n", (uint32)overlayToken));
514 
515 	// we only have a single token, which simplifies this
516 	if (overlayToken != (overlay_token)gInfo->shared_info->overlay_token)
517 		return B_BAD_VALUE;
518 
519 	atomic_and(&gInfo->shared_info->overlay_channel_used, 0);
520 
521 	return B_OK;
522 }
523 
524 
525 status_t
526 intel_configure_overlay(overlay_token overlayToken, const overlay_buffer *buffer,
527 	const overlay_window *window, const overlay_view *view)
528 {
529 	TRACE(("intel_configure_overlay: buffer %p, window %p, view %p\n",
530 		buffer, window, view));
531 
532 	if (overlayToken != (overlay_token)gInfo->shared_info->overlay_token)
533 		return B_BAD_VALUE;
534 
535 	if (window == NULL && view == NULL) {
536 		hide_overlay();
537 		return B_OK;
538 	}
539 
540 	struct overlay *overlay = (struct overlay *)buffer;
541 	overlay_registers *registers = gInfo->overlay_registers;
542 	bool updateCoefficients = false;
543 	uint32 bytesPerPixel = 2;
544 
545 	switch (buffer->space) {
546 		case B_RGB15:
547 			registers->source_format = OVERLAY_FORMAT_RGB15;
548 			break;
549 		case B_RGB16:
550 			registers->source_format = OVERLAY_FORMAT_RGB16;
551 			break;
552 		case B_RGB32:
553 			registers->source_format = OVERLAY_FORMAT_RGB32;
554 			bytesPerPixel = 4;
555 			break;
556 		case B_YCbCr422:
557 			registers->source_format = OVERLAY_FORMAT_YCbCr422;
558 			break;
559 	}
560 
561 	if (!gInfo->shared_info->overlay_active
562 		|| memcmp(&gInfo->last_overlay_view, view, sizeof(overlay_view))
563 		|| memcmp(&gInfo->last_overlay_frame, window, sizeof(overlay_frame))) {
564 		// scaling has changed, program window and scaling factor
565 
566 		// clip the window to on screen bounds
567 		// TODO: this is not yet complete or correct - especially if we start
568 		// to support moving the display!
569 		int32 left, top, right, bottom;
570 		left = window->h_start;
571 		right = window->h_start + window->width;
572 		top = window->v_start;
573 		bottom = window->v_start + window->height;
574 		if (left < 0)
575 			left = 0;
576 		if (top < 0)
577 			top = 0;
578 		if (right > gInfo->shared_info->current_mode.timing.h_display)
579 			right = gInfo->shared_info->current_mode.timing.h_display;
580 		if (bottom > gInfo->shared_info->current_mode.timing.v_display)
581 			bottom = gInfo->shared_info->current_mode.timing.v_display;
582 		if (left >= right || top >= bottom) {
583 			// overlay is not within visible bounds
584 			hide_overlay();
585 			return B_OK;
586 		}
587 
588 		registers->window_left = left;
589 		registers->window_top = top;
590 		registers->window_width = right - left;
591 		registers->window_height = bottom - top;
592 
593 		uint32 horizontalScale = (view->width << 12) / window->width;
594 		uint32 verticalScale = (view->height << 12) / window->height;
595 		uint32 horizontalScaleUV = horizontalScale >> 1;
596 		uint32 verticalScaleUV = verticalScale >> 1;
597 		horizontalScale = horizontalScaleUV << 1;
598 		verticalScale = verticalScaleUV << 1;
599 
600 		// we need to offset the overlay view to adapt it to the clipping
601 		// (in addition to whatever offset is desired already)
602 		left = view->h_start - (int32)((window->h_start - left) * (horizontalScale / 4096.0) + 0.5);
603 		top = view->v_start - (int32)((window->v_start - top) * (verticalScale / 4096.0) + 0.5);
604 		right = view->h_start + view->width;
605 		bottom = view->v_start + view->height;
606 
607 		gInfo->overlay_position_buffer_offset = buffer->bytes_per_row * top
608 			+ left * bytesPerPixel;
609 
610 		// Note: in non-planar mode, you *must* not program the source width/height
611 		// UV registers - they must stay cleared, or the chip is doing strange stuff.
612 		// On the other hand, you have to program the UV scaling registers, or the
613 		// result will be wrong, too.
614 		registers->source_width_rgb = right - left;
615 		registers->source_height_rgb = bottom - top;
616 		if ((gInfo->shared_info->device_type & INTEL_TYPE_8xx) != 0) {
617 			registers->source_bytes_per_row_rgb = (((overlay->buffer_offset + (view->width << 1)
618 				+ 0x1f) >> 5) - (overlay->buffer_offset >> 5) - 1) << 2;
619 		} else {
620 			registers->source_bytes_per_row_rgb = ((((overlay->buffer_offset + (view->width << 1)
621 				+ 0x3f) >> 6) - (overlay->buffer_offset >> 6) << 1) - 1) << 2;
622 		}
623 
624 		// horizontal scaling
625 		registers->scale_rgb.horizontal_downscale_factor = horizontalScale >> 12;
626 		registers->scale_rgb.horizontal_scale_fraction = horizontalScale & 0xfff;
627 		registers->scale_uv.horizontal_downscale_factor = horizontalScaleUV >> 12;
628 		registers->scale_uv.horizontal_scale_fraction = horizontalScaleUV & 0xfff;
629 
630 		// vertical scaling
631 		registers->scale_rgb.vertical_scale_fraction = verticalScale & 0xfff;
632 		registers->scale_uv.vertical_scale_fraction = verticalScaleUV & 0xfff;
633 		registers->vertical_scale_rgb = verticalScale >> 12;
634 		registers->vertical_scale_uv = verticalScaleUV >> 12;
635 
636 		TRACE(("scale: h = %ld.%ld, v = %ld.%ld\n", horizontalScale >> 12,
637 			horizontalScale & 0xfff, verticalScale >> 12, verticalScale & 0xfff));
638 
639 		if (verticalScale != gInfo->last_vertical_overlay_scale
640 			|| horizontalScale != gInfo->last_horizontal_overlay_scale) {
641 			// Recompute phase coefficients (taken from X driver)
642 			updateCoefficients = true;
643 
644 			phase_coefficient coefficients[NUM_HORIZONTAL_TAPS * NUM_PHASES];
645 			update_coefficients(NUM_HORIZONTAL_TAPS, horizontalScale / 4096.0,
646 				true, true, coefficients);
647 
648 			phase_coefficient coefficientsUV[NUM_HORIZONTAL_UV_TAPS * NUM_PHASES];
649 			update_coefficients(NUM_HORIZONTAL_UV_TAPS, horizontalScaleUV / 4096.0,
650 				true, false, coefficientsUV);
651 
652 			int32 pos = 0;
653 			for (int32 i = 0; i < NUM_PHASES; i++) {
654 				for (int32 j = 0; j < NUM_HORIZONTAL_TAPS; j++) {
655 					registers->horizontal_coefficients_rgb[pos] = coefficients[pos].sign << 15
656 						| coefficients[pos].exponent << 12
657 						| coefficients[pos].mantissa;
658 					pos++;
659 				}
660 			}
661 
662 			pos = 0;
663 			for (int32 i = 0; i < NUM_PHASES; i++) {
664 				for (int32 j = 0; j < NUM_HORIZONTAL_UV_TAPS; j++) {
665 					registers->horizontal_coefficients_uv[pos] = coefficientsUV[pos].sign << 15
666 						| coefficientsUV[pos].exponent << 12
667 						| coefficientsUV[pos].mantissa;
668 					pos++;
669 				}
670 			}
671 
672 			gInfo->last_vertical_overlay_scale = verticalScale;
673 			gInfo->last_horizontal_overlay_scale = horizontalScale;
674 		}
675 
676 		gInfo->last_overlay_view = *view;
677 		gInfo->last_overlay_frame = *(overlay_frame *)window;
678 	}
679 
680 	registers->color_control_output_mode = true;
681 	registers->select_pipe = 0;
682 
683 	// program buffer
684 
685 	registers->buffer_rgb0 = overlay->buffer_offset + gInfo->overlay_position_buffer_offset;
686 	registers->stride_rgb = buffer->bytes_per_row;
687 
688 	registers->mirroring_mode = (window->flags & B_OVERLAY_HORIZONTAL_MIRRORING) != 0
689 		? OVERLAY_MIRROR_HORIZONTAL : OVERLAY_MIRROR_NORMAL;
690 	registers->ycbcr422_order = 0;
691 
692 	if (!gInfo->shared_info->overlay_active) {
693 		// overlay is shown for the first time
694 		set_color_key(window);
695 		show_overlay();
696 	} else
697 		update_overlay(updateCoefficients);
698 
699 	gInfo->current_overlay = overlay;
700 	return B_OK;
701 }
702 
703