intel_blit.c revision eb10cdc838fc31ea2cf59f556f6f7d8b072f5bae
1/**************************************************************************
2 *
3 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
13 *
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
16 * of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 *
26 **************************************************************************/
27
28
29#include <stdio.h>
30#include <errno.h>
31
32#include "mtypes.h"
33#include "context.h"
34#include "enums.h"
35
36#include "intel_batchbuffer.h"
37#include "intel_blit.h"
38#include "intel_buffers.h"
39#include "intel_context.h"
40#include "intel_fbo.h"
41#include "intel_reg.h"
42#include "intel_regions.h"
43
44#define FILE_DEBUG_FLAG DEBUG_BLIT
45
46/**
47 * Copy the back color buffer to the front color buffer.
48 * Used for SwapBuffers().
49 */
50void
51intelCopyBuffer(const __DRIdrawablePrivate * dPriv,
52                const drm_clip_rect_t * rect)
53{
54
55   struct intel_context *intel;
56   const intelScreenPrivate *intelScreen;
57   int ret;
58
59   DBG("%s\n", __FUNCTION__);
60
61   assert(dPriv);
62
63   intel = intelScreenContext(dPriv->driScreenPriv->private);
64   if (!intel)
65      return;
66
67   intelScreen = intel->intelScreen;
68
69   /* The LOCK_HARDWARE is required for the cliprects.  Buffer offsets
70    * should work regardless.
71    */
72   LOCK_HARDWARE(intel);
73
74   if (dPriv && dPriv->numClipRects) {
75      struct intel_framebuffer *intel_fb = dPriv->driverPrivate;
76      struct intel_region *src, *dst;
77      int nbox = dPriv->numClipRects;
78      drm_clip_rect_t *pbox = dPriv->pClipRects;
79      int cpp;
80      int src_pitch, dst_pitch;
81      unsigned short src_x, src_y;
82      int BR13, CMD;
83      int i;
84
85      src = intel_get_rb_region(&intel_fb->Base, BUFFER_BACK_LEFT);
86      dst = intel_get_rb_region(&intel_fb->Base, BUFFER_FRONT_LEFT);
87
88      src_pitch = src->pitch * src->cpp;
89      dst_pitch = dst->pitch * dst->cpp;
90
91      cpp = src->cpp;
92
93      ASSERT(intel_fb);
94      ASSERT(intel_fb->Base.Name == 0);    /* Not a user-created FBO */
95      ASSERT(src);
96      ASSERT(dst);
97      ASSERT(src->cpp == dst->cpp);
98
99      if (cpp == 2) {
100	 BR13 = (0xCC << 16) | (1 << 24);
101	 CMD = XY_SRC_COPY_BLT_CMD;
102      }
103      else {
104	 BR13 = (0xCC << 16) | (1 << 24) | (1 << 25);
105	 CMD = XY_SRC_COPY_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
106      }
107
108#ifndef I915
109      if (src->tiled) {
110	 CMD |= XY_SRC_TILED;
111	 src_pitch /= 4;
112      }
113      if (dst->tiled) {
114	 CMD |= XY_DST_TILED;
115	 dst_pitch /= 4;
116      }
117#endif
118
119   again:
120      ret = dri_bufmgr_check_aperture_space(dst->buffer);
121      ret |= dri_bufmgr_check_aperture_space(src->buffer);
122
123      if (ret) {
124	intel_batchbuffer_flush(intel->batch);
125	goto again;
126      }
127
128      for (i = 0; i < nbox; i++, pbox++) {
129	 drm_clip_rect_t box = *pbox;
130
131	 if (rect) {
132	    if (!intel_intersect_cliprects(&box, &box, rect))
133	       continue;
134	 }
135
136	 if (box.x1 >= box.x2 ||
137	     box.y1 >= box.y2)
138	    continue;
139
140	 assert(box.x1 < box.x2);
141	 assert(box.y1 < box.y2);
142	 src_x = box.x1 - dPriv->x + dPriv->backX;
143	 src_y = box.y1 - dPriv->y + dPriv->backY;
144
145	 BEGIN_BATCH(8, REFERENCES_CLIPRECTS);
146	 OUT_BATCH(CMD);
147	 OUT_BATCH(BR13 | dst_pitch);
148	 OUT_BATCH((box.y1 << 16) | box.x1);
149	 OUT_BATCH((box.y2 << 16) | box.x2);
150
151	 OUT_RELOC(dst->buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE, 0);
152	 OUT_BATCH((src_y << 16) | src_x);
153	 OUT_BATCH(src_pitch);
154	 OUT_RELOC(src->buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, 0);
155	 ADVANCE_BATCH();
156      }
157
158      intel_batchbuffer_flush(intel->batch);
159   }
160
161   UNLOCK_HARDWARE(intel);
162}
163
164
165
166
167void
168intelEmitFillBlit(struct intel_context *intel,
169		  GLuint cpp,
170		  GLshort dst_pitch,
171		  dri_bo *dst_buffer,
172		  GLuint dst_offset,
173		  GLboolean dst_tiled,
174		  GLshort x, GLshort y,
175		  GLshort w, GLshort h,
176		  GLuint color)
177{
178   GLuint BR13, CMD;
179   BATCH_LOCALS;
180
181   dst_pitch *= cpp;
182
183   switch (cpp) {
184   case 1:
185   case 2:
186   case 3:
187      BR13 = (0xF0 << 16) | (1 << 24);
188      CMD = XY_COLOR_BLT_CMD;
189      break;
190   case 4:
191      BR13 = (0xF0 << 16) | (1 << 24) | (1 << 25);
192      CMD = XY_COLOR_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
193      break;
194   default:
195      return;
196   }
197#ifndef I915
198   if (dst_tiled) {
199      CMD |= XY_DST_TILED;
200      dst_pitch /= 4;
201   }
202#endif
203
204   DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
205       __FUNCTION__, dst_buffer, dst_pitch, dst_offset, x, y, w, h);
206
207   assert(w > 0);
208   assert(h > 0);
209
210   BEGIN_BATCH(6, NO_LOOP_CLIPRECTS);
211   OUT_BATCH(CMD);
212   OUT_BATCH(BR13 | dst_pitch);
213   OUT_BATCH((y << 16) | x);
214   OUT_BATCH(((y + h) << 16) | (x + w));
215   OUT_RELOC(dst_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE, dst_offset);
216   OUT_BATCH(color);
217   ADVANCE_BATCH();
218}
219
220static GLuint translate_raster_op(GLenum logicop)
221{
222   switch(logicop) {
223   case GL_CLEAR: return 0x00;
224   case GL_AND: return 0x88;
225   case GL_AND_REVERSE: return 0x44;
226   case GL_COPY: return 0xCC;
227   case GL_AND_INVERTED: return 0x22;
228   case GL_NOOP: return 0xAA;
229   case GL_XOR: return 0x66;
230   case GL_OR: return 0xEE;
231   case GL_NOR: return 0x11;
232   case GL_EQUIV: return 0x99;
233   case GL_INVERT: return 0x55;
234   case GL_OR_REVERSE: return 0xDD;
235   case GL_COPY_INVERTED: return 0x33;
236   case GL_OR_INVERTED: return 0xBB;
237   case GL_NAND: return 0x77;
238   case GL_SET: return 0xFF;
239   default: return 0;
240   }
241}
242
243
244/* Copy BitBlt
245 */
246void
247intelEmitCopyBlit(struct intel_context *intel,
248		  GLuint cpp,
249		  GLshort src_pitch,
250		  dri_bo *src_buffer,
251		  GLuint src_offset,
252		  GLboolean src_tiled,
253		  GLshort dst_pitch,
254		  dri_bo *dst_buffer,
255		  GLuint dst_offset,
256		  GLboolean dst_tiled,
257		  GLshort src_x, GLshort src_y,
258		  GLshort dst_x, GLshort dst_y,
259		  GLshort w, GLshort h,
260		  GLenum logic_op)
261{
262   GLuint CMD, BR13;
263   int dst_y2 = dst_y + h;
264   int dst_x2 = dst_x + w;
265   int ret;
266   BATCH_LOCALS;
267
268 again:
269   ret = dri_bufmgr_check_aperture_space(dst_buffer);
270   ret |= dri_bufmgr_check_aperture_space(src_buffer);
271   if (ret) {
272     intel_batchbuffer_flush(intel->batch);
273     goto again;
274   }
275
276   DBG("%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
277       __FUNCTION__,
278       src_buffer, src_pitch, src_offset, src_x, src_y,
279       dst_buffer, dst_pitch, dst_offset, dst_x, dst_y, w, h);
280
281   src_pitch *= cpp;
282   dst_pitch *= cpp;
283
284   BR13 = translate_raster_op(logic_op) << 16;
285
286   switch (cpp) {
287   case 1:
288   case 2:
289   case 3:
290      BR13 |= (1 << 24);
291      CMD = XY_SRC_COPY_BLT_CMD;
292      break;
293   case 4:
294      BR13 |= (1 << 24) | (1 << 25);
295      CMD = XY_SRC_COPY_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
296      break;
297   default:
298      return;
299   }
300
301#ifndef I915
302   if (dst_tiled) {
303      CMD |= XY_DST_TILED;
304      dst_pitch /= 4;
305   }
306   if (src_tiled) {
307      CMD |= XY_SRC_TILED;
308      src_pitch /= 4;
309   }
310#endif
311
312   if (dst_y2 <= dst_y || dst_x2 <= dst_x) {
313      return;
314   }
315
316   dst_pitch &= 0xffff;
317   src_pitch &= 0xffff;
318
319   /* Initial y values don't seem to work with negative pitches.  If
320    * we adjust the offsets manually (below), it seems to work fine.
321    *
322    * On the other hand, if we always adjust, the hardware doesn't
323    * know which blit directions to use, so overlapping copypixels get
324    * the wrong result.
325    */
326   if (dst_pitch > 0 && src_pitch > 0) {
327      assert(dst_x < dst_x2);
328      assert(dst_y < dst_y2);
329
330      BEGIN_BATCH(8, NO_LOOP_CLIPRECTS);
331      OUT_BATCH(CMD);
332      OUT_BATCH(BR13 | dst_pitch);
333      OUT_BATCH((dst_y << 16) | dst_x);
334      OUT_BATCH((dst_y2 << 16) | dst_x2);
335      OUT_RELOC(dst_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
336		dst_offset);
337      OUT_BATCH((src_y << 16) | src_x);
338      OUT_BATCH(src_pitch);
339      OUT_RELOC(src_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
340		src_offset);
341      ADVANCE_BATCH();
342   }
343   else {
344      assert(dst_x < dst_x2);
345      assert(h > 0);
346
347      BEGIN_BATCH(8, NO_LOOP_CLIPRECTS);
348      OUT_BATCH(CMD);
349      OUT_BATCH(BR13 | dst_pitch);
350      OUT_BATCH((0 << 16) | dst_x);
351      OUT_BATCH((h << 16) | dst_x2);
352      OUT_RELOC(dst_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
353		dst_offset + dst_y * dst_pitch);
354      OUT_BATCH((0 << 16) | src_x);
355      OUT_BATCH(src_pitch);
356      OUT_RELOC(src_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
357		src_offset + src_y * src_pitch);
358      ADVANCE_BATCH();
359   }
360}
361
362
363/**
364 * Use blitting to clear the renderbuffers named by 'flags'.
365 * Note: we can't use the ctx->DrawBuffer->_ColorDrawBufferIndexes field
366 * since that might include software renderbuffers or renderbuffers
367 * which we're clearing with triangles.
368 * \param mask  bitmask of BUFFER_BIT_* values indicating buffers to clear
369 */
370void
371intelClearWithBlit(GLcontext *ctx, GLbitfield mask)
372{
373   struct intel_context *intel = intel_context(ctx);
374   struct gl_framebuffer *fb = ctx->DrawBuffer;
375   GLuint clear_depth;
376   GLbitfield skipBuffers = 0;
377   BATCH_LOCALS;
378
379   /*
380    * Compute values for clearing the buffers.
381    */
382   clear_depth = 0;
383   if (mask & BUFFER_BIT_DEPTH) {
384      clear_depth = (GLuint) (fb->_DepthMax * ctx->Depth.Clear);
385   }
386   if (mask & BUFFER_BIT_STENCIL) {
387      clear_depth |= (ctx->Stencil.Clear & 0xff) << 24;
388   }
389
390   /* If clearing both depth and stencil, skip BUFFER_BIT_STENCIL in
391    * the loop below.
392    */
393   if ((mask & BUFFER_BIT_DEPTH) && (mask & BUFFER_BIT_STENCIL)) {
394      skipBuffers = BUFFER_BIT_STENCIL;
395   }
396
397   /* XXX Move this flush/lock into the following conditional? */
398   intelFlush(&intel->ctx);
399   LOCK_HARDWARE(intel);
400
401   if (intel->numClipRects) {
402      GLint cx, cy, cw, ch;
403      drm_clip_rect_t clear;
404      int i;
405
406      /* Get clear bounds after locking */
407      cx = fb->_Xmin;
408      cy = fb->_Ymin;
409      cw = fb->_Xmax - cx;
410      ch = fb->_Ymax - cy;
411
412      if (fb->Name == 0) {
413         /* clearing a window */
414
415         /* flip top to bottom */
416         clear.x1 = cx + intel->drawX;
417         clear.y1 = intel->driDrawable->y + intel->driDrawable->h - cy - ch;
418         clear.x2 = clear.x1 + cw;
419         clear.y2 = clear.y1 + ch;
420      }
421      else {
422         /* clearing FBO */
423         assert(intel->numClipRects == 1);
424         assert(intel->pClipRects == &intel->fboRect);
425         clear.x1 = cx;
426         clear.y1 = cy;
427         clear.x2 = clear.x1 + cw;
428         clear.y2 = clear.y1 + ch;
429         /* no change to mask */
430      }
431
432      for (i = 0; i < intel->numClipRects; i++) {
433         const drm_clip_rect_t *box = &intel->pClipRects[i];
434         drm_clip_rect_t b;
435         GLuint buf;
436         GLuint clearMask = mask;      /* use copy, since we modify it below */
437         GLboolean all = (cw == fb->Width && ch == fb->Height);
438
439         if (!all) {
440            intel_intersect_cliprects(&b, &clear, box);
441         }
442         else {
443            b = *box;
444         }
445
446         if (b.x1 >= b.x2 || b.y1 >= b.y2)
447            continue;
448
449         if (0)
450            _mesa_printf("clear %d,%d..%d,%d, mask %x\n",
451                         b.x1, b.y1, b.x2, b.y2, mask);
452
453         /* Loop over all renderbuffers */
454         for (buf = 0; buf < BUFFER_COUNT && clearMask; buf++) {
455            const GLbitfield bufBit = 1 << buf;
456            if ((clearMask & bufBit) && !(bufBit & skipBuffers)) {
457               /* OK, clear this renderbuffer */
458               struct intel_region *irb_region =
459		  intel_get_rb_region(fb, buf);
460               dri_bo *write_buffer =
461                  intel_region_buffer(intel, irb_region,
462                                      all ? INTEL_WRITE_FULL :
463                                      INTEL_WRITE_PART);
464
465               GLuint clearVal;
466               GLint pitch, cpp;
467               GLuint BR13, CMD;
468
469               ASSERT(irb_region);
470
471               pitch = irb_region->pitch;
472               cpp = irb_region->cpp;
473
474               DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
475                   __FUNCTION__,
476                   irb_region->buffer, (pitch * cpp),
477                   irb_region->draw_offset,
478                   b.x1, b.y1, b.x2 - b.x1, b.y2 - b.y1);
479
480	       BR13 = 0xf0 << 16;
481	       CMD = XY_COLOR_BLT_CMD;
482
483               /* Setup the blit command */
484               if (cpp == 4) {
485                  BR13 |= (1 << 24) | (1 << 25);
486                  if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL) {
487                     if (clearMask & BUFFER_BIT_DEPTH)
488                        CMD |= XY_BLT_WRITE_RGB;
489                     if (clearMask & BUFFER_BIT_STENCIL)
490                        CMD |= XY_BLT_WRITE_ALPHA;
491                  }
492                  else {
493                     /* clearing RGBA */
494                     CMD |= XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
495                  }
496               }
497               else {
498                  ASSERT(cpp == 2 || cpp == 0);
499                  BR13 |= (1 << 24);
500               }
501
502#ifndef I915
503	       if (irb_region->tiled) {
504		  CMD |= XY_DST_TILED;
505		  pitch /= 4;
506	       }
507#endif
508	       BR13 |= (pitch * cpp);
509
510               if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL) {
511                  clearVal = clear_depth;
512               }
513               else {
514                  clearVal = (cpp == 4)
515                     ? intel->ClearColor8888 : intel->ClearColor565;
516               }
517               /*
518                  _mesa_debug(ctx, "hardware blit clear buf %d rb id %d\n",
519                  buf, irb->Base.Name);
520                */
521	       intel_wait_flips(intel);
522
523               assert(b.x1 < b.x2);
524               assert(b.y1 < b.y2);
525
526               BEGIN_BATCH(6, REFERENCES_CLIPRECTS);
527               OUT_BATCH(CMD);
528               OUT_BATCH(BR13);
529               OUT_BATCH((b.y1 << 16) | b.x1);
530               OUT_BATCH((b.y2 << 16) | b.x2);
531               OUT_RELOC(write_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
532                         irb_region->draw_offset);
533               OUT_BATCH(clearVal);
534               ADVANCE_BATCH();
535               clearMask &= ~bufBit;    /* turn off bit, for faster loop exit */
536            }
537         }
538      }
539      intel_batchbuffer_flush(intel->batch);
540   }
541
542   UNLOCK_HARDWARE(intel);
543}
544
545void
546intelEmitImmediateColorExpandBlit(struct intel_context *intel,
547				  GLuint cpp,
548				  GLubyte *src_bits, GLuint src_size,
549				  GLuint fg_color,
550				  GLshort dst_pitch,
551				  dri_bo *dst_buffer,
552				  GLuint dst_offset,
553				  GLboolean dst_tiled,
554				  GLshort x, GLshort y,
555				  GLshort w, GLshort h,
556				  GLenum logic_op)
557{
558   int dwords = ALIGN(src_size, 8) / 4;
559   uint32_t opcode, br13, blit_cmd;
560
561   assert( logic_op - GL_CLEAR >= 0 );
562   assert( logic_op - GL_CLEAR < 0x10 );
563
564   if (w < 0 || h < 0)
565      return;
566
567   dst_pitch *= cpp;
568
569   if (dst_tiled)
570      dst_pitch /= 4;
571
572   DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d, %d bytes %d dwords\n",
573       __FUNCTION__,
574       dst_buffer, dst_pitch, dst_offset, x, y, w, h, src_size, dwords);
575
576   intel_batchbuffer_require_space( intel->batch,
577				    (8 * 4) +
578				    (3 * 4) +
579				    dwords,
580				    NO_LOOP_CLIPRECTS );
581
582   opcode = XY_SETUP_BLT_CMD;
583   if (cpp == 4)
584      opcode |= XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
585   if (dst_tiled)
586      opcode |= XY_DST_TILED;
587
588   br13 = dst_pitch | (translate_raster_op(logic_op) << 16) | (1 << 29);
589   if (cpp == 2)
590      br13 |= BR13_565;
591   else
592      br13 |= BR13_8888;
593
594   blit_cmd = XY_TEXT_IMMEDIATE_BLIT_CMD | XY_TEXT_BYTE_PACKED; /* packing? */
595   if (dst_tiled)
596      blit_cmd |= XY_DST_TILED;
597
598   BEGIN_BATCH(8 + 3, NO_LOOP_CLIPRECTS);
599   OUT_BATCH(opcode);
600   OUT_BATCH(br13);
601   OUT_BATCH((0 << 16) | 0); /* clip x1, y1 */
602   OUT_BATCH((100 << 16) | 100); /* clip x2, y2 */
603   OUT_RELOC(dst_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE, dst_offset);
604   OUT_BATCH(0); /* bg */
605   OUT_BATCH(fg_color); /* fg */
606   OUT_BATCH(0); /* pattern base addr */
607
608   OUT_BATCH(blit_cmd | ((3 - 2) + dwords));
609   OUT_BATCH((y << 16) | x);
610   OUT_BATCH(((y + h) << 16) | (x + w));
611   ADVANCE_BATCH();
612
613   intel_batchbuffer_data( intel->batch,
614			   src_bits,
615			   dwords * 4,
616			   NO_LOOP_CLIPRECTS );
617}
618