intel_blit.c revision ecadb51bbcb972a79f3ed79e65a7986b9396e757
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 "main/mtypes.h"
33#include "main/context.h"
34#include "main/enums.h"
35
36#include "intel_blit.h"
37#include "intel_buffers.h"
38#include "intel_context.h"
39#include "intel_fbo.h"
40#include "intel_reg.h"
41#include "intel_regions.h"
42#include "intel_batchbuffer.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
58   DBG("%s\n", __FUNCTION__);
59
60   assert(dPriv);
61
62   intel = intelScreenContext(dPriv->driScreenPriv->private);
63   if (!intel)
64      return;
65
66   intelScreen = intel->intelScreen;
67
68   /* The LOCK_HARDWARE is required for the cliprects.  Buffer offsets
69    * should work regardless.
70    */
71   LOCK_HARDWARE(intel);
72
73   if (dPriv && dPriv->numClipRects) {
74      struct intel_framebuffer *intel_fb = dPriv->driverPrivate;
75      struct intel_region *src, *dst;
76      int nbox = dPriv->numClipRects;
77      drm_clip_rect_t *pbox = dPriv->pClipRects;
78      int cpp;
79      int src_pitch, dst_pitch;
80      unsigned short src_x, src_y;
81      int BR13, CMD;
82      int i;
83      dri_bo *aper_array[3];
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->tiling != I915_TILING_NONE) {
110	 CMD |= XY_SRC_TILED;
111	 src_pitch /= 4;
112      }
113      if (dst->tiling != I915_TILING_NONE) {
114	 CMD |= XY_DST_TILED;
115	 dst_pitch /= 4;
116      }
117#endif
118      /* do space/cliprects check before going any further */
119      intel_batchbuffer_require_space(intel->batch, 8 * 4,
120				      REFERENCES_CLIPRECTS);
121   again:
122      aper_array[0] = intel->batch->buf;
123      aper_array[1] = dst->buffer;
124      aper_array[2] = src->buffer;
125
126      if (dri_bufmgr_check_aperture_space(aper_array, 3) != 0) {
127	intel_batchbuffer_flush(intel->batch);
128	goto again;
129      }
130
131      for (i = 0; i < nbox; i++, pbox++) {
132	 drm_clip_rect_t box = *pbox;
133
134	 if (rect) {
135	    if (!intel_intersect_cliprects(&box, &box, rect))
136	       continue;
137	 }
138
139	 if (box.x1 >= box.x2 ||
140	     box.y1 >= box.y2)
141	    continue;
142
143	 assert(box.x1 < box.x2);
144	 assert(box.y1 < box.y2);
145	 src_x = box.x1 - dPriv->x + dPriv->backX;
146	 src_y = box.y1 - dPriv->y + dPriv->backY;
147
148	 BEGIN_BATCH(8, REFERENCES_CLIPRECTS);
149	 OUT_BATCH(CMD);
150	 OUT_BATCH(BR13 | dst_pitch);
151	 OUT_BATCH((box.y1 << 16) | box.x1);
152	 OUT_BATCH((box.y2 << 16) | box.x2);
153
154	 OUT_RELOC(dst->buffer,
155		   I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
156		   0);
157	 OUT_BATCH((src_y << 16) | src_x);
158	 OUT_BATCH(src_pitch);
159	 OUT_RELOC(src->buffer,
160		   I915_GEM_DOMAIN_RENDER, 0,
161		   0);
162	 ADVANCE_BATCH();
163      }
164
165      /* Flush the rendering and the batch so that the results all land on the
166       * screen in a timely fashion.
167       */
168      intel_batchbuffer_emit_mi_flush(intel->batch);
169      intel_batchbuffer_flush(intel->batch);
170   }
171
172   UNLOCK_HARDWARE(intel);
173}
174
175
176
177
178void
179intelEmitFillBlit(struct intel_context *intel,
180		  GLuint cpp,
181		  GLshort dst_pitch,
182		  dri_bo *dst_buffer,
183		  GLuint dst_offset,
184		  uint32_t dst_tiling,
185		  GLshort x, GLshort y,
186		  GLshort w, GLshort h,
187		  GLuint color)
188{
189   GLuint BR13, CMD;
190   BATCH_LOCALS;
191
192   dst_pitch *= cpp;
193
194   switch (cpp) {
195   case 1:
196   case 2:
197   case 3:
198      BR13 = (0xF0 << 16) | (1 << 24);
199      CMD = XY_COLOR_BLT_CMD;
200      break;
201   case 4:
202      BR13 = (0xF0 << 16) | (1 << 24) | (1 << 25);
203      CMD = XY_COLOR_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
204      break;
205   default:
206      return;
207   }
208#ifndef I915
209   if (dst_tiling != I915_TILING_NONE) {
210      CMD |= XY_DST_TILED;
211      dst_pitch /= 4;
212   }
213#endif
214
215   DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
216       __FUNCTION__, dst_buffer, dst_pitch, dst_offset, x, y, w, h);
217
218   assert(w > 0);
219   assert(h > 0);
220
221   BEGIN_BATCH(6, NO_LOOP_CLIPRECTS);
222   OUT_BATCH(CMD);
223   OUT_BATCH(BR13 | dst_pitch);
224   OUT_BATCH((y << 16) | x);
225   OUT_BATCH(((y + h) << 16) | (x + w));
226   OUT_RELOC(dst_buffer,
227	     I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
228	     dst_offset);
229   OUT_BATCH(color);
230   ADVANCE_BATCH();
231}
232
233static GLuint translate_raster_op(GLenum logicop)
234{
235   switch(logicop) {
236   case GL_CLEAR: return 0x00;
237   case GL_AND: return 0x88;
238   case GL_AND_REVERSE: return 0x44;
239   case GL_COPY: return 0xCC;
240   case GL_AND_INVERTED: return 0x22;
241   case GL_NOOP: return 0xAA;
242   case GL_XOR: return 0x66;
243   case GL_OR: return 0xEE;
244   case GL_NOR: return 0x11;
245   case GL_EQUIV: return 0x99;
246   case GL_INVERT: return 0x55;
247   case GL_OR_REVERSE: return 0xDD;
248   case GL_COPY_INVERTED: return 0x33;
249   case GL_OR_INVERTED: return 0xBB;
250   case GL_NAND: return 0x77;
251   case GL_SET: return 0xFF;
252   default: return 0;
253   }
254}
255
256
257/* Copy BitBlt
258 */
259void
260intelEmitCopyBlit(struct intel_context *intel,
261		  GLuint cpp,
262		  GLshort src_pitch,
263		  dri_bo *src_buffer,
264		  GLuint src_offset,
265		  uint32_t src_tiling,
266		  GLshort dst_pitch,
267		  dri_bo *dst_buffer,
268		  GLuint dst_offset,
269		  uint32_t dst_tiling,
270		  GLshort src_x, GLshort src_y,
271		  GLshort dst_x, GLshort dst_y,
272		  GLshort w, GLshort h,
273		  GLenum logic_op)
274{
275   GLuint CMD, BR13;
276   int dst_y2 = dst_y + h;
277   int dst_x2 = dst_x + w;
278   dri_bo *aper_array[3];
279   BATCH_LOCALS;
280
281   /* do space/cliprects check before going any further */
282   intel_batchbuffer_require_space(intel->batch, 8 * 4, NO_LOOP_CLIPRECTS);
283 again:
284   aper_array[0] = intel->batch->buf;
285   aper_array[1] = dst_buffer;
286   aper_array[2] = src_buffer;
287
288   if (dri_bufmgr_check_aperture_space(aper_array, 3) != 0) {
289      intel_batchbuffer_flush(intel->batch);
290      goto again;
291   }
292
293   DBG("%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
294       __FUNCTION__,
295       src_buffer, src_pitch, src_offset, src_x, src_y,
296       dst_buffer, dst_pitch, dst_offset, dst_x, dst_y, w, h);
297
298   src_pitch *= cpp;
299   dst_pitch *= cpp;
300
301   BR13 = translate_raster_op(logic_op) << 16;
302
303   switch (cpp) {
304   case 1:
305   case 2:
306   case 3:
307      BR13 |= (1 << 24);
308      CMD = XY_SRC_COPY_BLT_CMD;
309      break;
310   case 4:
311      BR13 |= (1 << 24) | (1 << 25);
312      CMD = XY_SRC_COPY_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
313      break;
314   default:
315      return;
316   }
317
318#ifndef I915
319   if (dst_tiling != I915_TILING_NONE) {
320      CMD |= XY_DST_TILED;
321      dst_pitch /= 4;
322   }
323   if (src_tiling != I915_TILING_NONE) {
324      CMD |= XY_SRC_TILED;
325      src_pitch /= 4;
326   }
327#endif
328
329   if (dst_y2 <= dst_y || dst_x2 <= dst_x) {
330      return;
331   }
332
333   /* Initial y values don't seem to work with negative pitches.  If
334    * we adjust the offsets manually (below), it seems to work fine.
335    *
336    * On the other hand, if we always adjust, the hardware doesn't
337    * know which blit directions to use, so overlapping copypixels get
338    * the wrong result.
339    */
340   if (dst_pitch > 0 && src_pitch > 0) {
341      assert(dst_x < dst_x2);
342      assert(dst_y < dst_y2);
343
344      BEGIN_BATCH(8, NO_LOOP_CLIPRECTS);
345      OUT_BATCH(CMD);
346      OUT_BATCH(BR13 | dst_pitch);
347      OUT_BATCH((dst_y << 16) | dst_x);
348      OUT_BATCH((dst_y2 << 16) | dst_x2);
349      OUT_RELOC(dst_buffer,
350		I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
351		dst_offset);
352      OUT_BATCH((src_y << 16) | src_x);
353      OUT_BATCH(src_pitch);
354      OUT_RELOC(src_buffer,
355		I915_GEM_DOMAIN_RENDER, 0,
356		src_offset);
357      ADVANCE_BATCH();
358   }
359   else {
360      assert(dst_x < dst_x2);
361      assert(h > 0);
362
363      BEGIN_BATCH(8, NO_LOOP_CLIPRECTS);
364      OUT_BATCH(CMD);
365      OUT_BATCH(BR13 | ((uint16_t)dst_pitch));
366      OUT_BATCH((0 << 16) | dst_x);
367      OUT_BATCH((h << 16) | dst_x2);
368      OUT_RELOC(dst_buffer,
369		I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
370		dst_offset + dst_y * dst_pitch);
371      OUT_BATCH((0 << 16) | src_x);
372      OUT_BATCH(src_pitch);
373      OUT_RELOC(src_buffer,
374		I915_GEM_DOMAIN_RENDER, 0,
375		src_offset + src_y * src_pitch);
376      ADVANCE_BATCH();
377   }
378   intel_batchbuffer_emit_mi_flush(intel->batch);
379}
380
381
382/**
383 * Use blitting to clear the renderbuffers named by 'flags'.
384 * Note: we can't use the ctx->DrawBuffer->_ColorDrawBufferIndexes field
385 * since that might include software renderbuffers or renderbuffers
386 * which we're clearing with triangles.
387 * \param mask  bitmask of BUFFER_BIT_* values indicating buffers to clear
388 */
389void
390intelClearWithBlit(GLcontext *ctx, GLbitfield mask)
391{
392   struct intel_context *intel = intel_context(ctx);
393   struct gl_framebuffer *fb = ctx->DrawBuffer;
394   GLuint clear_depth;
395   GLbitfield skipBuffers = 0;
396   BATCH_LOCALS;
397
398   /*
399    * Compute values for clearing the buffers.
400    */
401   clear_depth = 0;
402   if (mask & BUFFER_BIT_DEPTH) {
403      clear_depth = (GLuint) (fb->_DepthMax * ctx->Depth.Clear);
404   }
405   if (mask & BUFFER_BIT_STENCIL) {
406      clear_depth |= (ctx->Stencil.Clear & 0xff) << 24;
407   }
408
409   /* If clearing both depth and stencil, skip BUFFER_BIT_STENCIL in
410    * the loop below.
411    */
412   if ((mask & BUFFER_BIT_DEPTH) && (mask & BUFFER_BIT_STENCIL)) {
413      skipBuffers = BUFFER_BIT_STENCIL;
414   }
415
416   /* XXX Move this flush/lock into the following conditional? */
417   intelFlush(&intel->ctx);
418   LOCK_HARDWARE(intel);
419
420   if (intel->numClipRects) {
421      GLint cx, cy, cw, ch;
422      drm_clip_rect_t clear;
423      int i;
424
425      /* Get clear bounds after locking */
426      cx = fb->_Xmin;
427      cy = fb->_Ymin;
428      cw = fb->_Xmax - cx;
429      ch = fb->_Ymax - cy;
430
431      if (fb->Name == 0) {
432         /* clearing a window */
433
434         /* flip top to bottom */
435         clear.x1 = cx + intel->drawX;
436         clear.y1 = intel->driDrawable->y + intel->driDrawable->h - cy - ch;
437         clear.x2 = clear.x1 + cw;
438         clear.y2 = clear.y1 + ch;
439      }
440      else {
441         /* clearing FBO */
442         assert(intel->numClipRects == 1);
443         assert(intel->pClipRects == &intel->fboRect);
444         clear.x1 = cx;
445         clear.y1 = cy;
446         clear.x2 = clear.x1 + cw;
447         clear.y2 = clear.y1 + ch;
448         /* no change to mask */
449      }
450
451      for (i = 0; i < intel->numClipRects; i++) {
452         const drm_clip_rect_t *box = &intel->pClipRects[i];
453         drm_clip_rect_t b;
454         GLuint buf;
455         GLuint clearMask = mask;      /* use copy, since we modify it below */
456         GLboolean all = (cw == fb->Width && ch == fb->Height);
457
458         if (!all) {
459            intel_intersect_cliprects(&b, &clear, box);
460         }
461         else {
462            b = *box;
463         }
464
465         if (b.x1 >= b.x2 || b.y1 >= b.y2)
466            continue;
467
468         if (0)
469            _mesa_printf("clear %d,%d..%d,%d, mask %x\n",
470                         b.x1, b.y1, b.x2, b.y2, mask);
471
472         /* Loop over all renderbuffers */
473         for (buf = 0; buf < BUFFER_COUNT && clearMask; buf++) {
474            const GLbitfield bufBit = 1 << buf;
475            if ((clearMask & bufBit) && !(bufBit & skipBuffers)) {
476               /* OK, clear this renderbuffer */
477               struct intel_region *irb_region =
478		  intel_get_rb_region(fb, buf);
479               dri_bo *write_buffer =
480                  intel_region_buffer(intel, irb_region,
481                                      all ? INTEL_WRITE_FULL :
482                                      INTEL_WRITE_PART);
483
484               GLuint clearVal;
485               GLint pitch, cpp;
486               GLuint BR13, CMD;
487
488               ASSERT(irb_region);
489
490               pitch = irb_region->pitch;
491               cpp = irb_region->cpp;
492
493               DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
494                   __FUNCTION__,
495                   irb_region->buffer, (pitch * cpp),
496                   irb_region->draw_offset,
497                   b.x1, b.y1, b.x2 - b.x1, b.y2 - b.y1);
498
499	       BR13 = 0xf0 << 16;
500	       CMD = XY_COLOR_BLT_CMD;
501
502               /* Setup the blit command */
503               if (cpp == 4) {
504                  BR13 |= (1 << 24) | (1 << 25);
505                  if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL) {
506                     if (clearMask & BUFFER_BIT_DEPTH)
507                        CMD |= XY_BLT_WRITE_RGB;
508                     if (clearMask & BUFFER_BIT_STENCIL)
509                        CMD |= XY_BLT_WRITE_ALPHA;
510                  }
511                  else {
512                     /* clearing RGBA */
513                     CMD |= XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
514                  }
515               }
516               else {
517                  ASSERT(cpp == 2 || cpp == 0);
518                  BR13 |= (1 << 24);
519               }
520
521#ifndef I915
522	       if (irb_region->tiling != I915_TILING_NONE) {
523		  CMD |= XY_DST_TILED;
524		  pitch /= 4;
525	       }
526#endif
527	       BR13 |= (pitch * cpp);
528
529               if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL) {
530                  clearVal = clear_depth;
531               }
532               else {
533                  clearVal = (cpp == 4)
534                     ? intel->ClearColor8888 : intel->ClearColor565;
535               }
536               /*
537                  _mesa_debug(ctx, "hardware blit clear buf %d rb id %d\n",
538                  buf, irb->Base.Name);
539                */
540	       intel_wait_flips(intel);
541
542               assert(b.x1 < b.x2);
543               assert(b.y1 < b.y2);
544
545               BEGIN_BATCH(6, REFERENCES_CLIPRECTS);
546               OUT_BATCH(CMD);
547               OUT_BATCH(BR13);
548               OUT_BATCH((b.y1 << 16) | b.x1);
549               OUT_BATCH((b.y2 << 16) | b.x2);
550               OUT_RELOC(write_buffer,
551			 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
552                         irb_region->draw_offset);
553               OUT_BATCH(clearVal);
554               ADVANCE_BATCH();
555               clearMask &= ~bufBit;    /* turn off bit, for faster loop exit */
556            }
557         }
558      }
559      intel_batchbuffer_emit_mi_flush(intel->batch);
560   }
561
562   UNLOCK_HARDWARE(intel);
563}
564
565void
566intelEmitImmediateColorExpandBlit(struct intel_context *intel,
567				  GLuint cpp,
568				  GLubyte *src_bits, GLuint src_size,
569				  GLuint fg_color,
570				  GLshort dst_pitch,
571				  dri_bo *dst_buffer,
572				  GLuint dst_offset,
573				  uint32_t dst_tiling,
574				  GLshort x, GLshort y,
575				  GLshort w, GLshort h,
576				  GLenum logic_op)
577{
578   int dwords = ALIGN(src_size, 8) / 4;
579   uint32_t opcode, br13, blit_cmd;
580
581   assert( logic_op - GL_CLEAR >= 0 );
582   assert( logic_op - GL_CLEAR < 0x10 );
583
584   if (w < 0 || h < 0)
585      return;
586
587   dst_pitch *= cpp;
588
589   DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d, %d bytes %d dwords\n",
590       __FUNCTION__,
591       dst_buffer, dst_pitch, dst_offset, x, y, w, h, src_size, dwords);
592
593   intel_batchbuffer_require_space( intel->batch,
594				    (8 * 4) +
595				    (3 * 4) +
596				    dwords,
597				    REFERENCES_CLIPRECTS );
598
599   opcode = XY_SETUP_BLT_CMD;
600   if (cpp == 4)
601      opcode |= XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
602#ifndef I915
603   if (dst_tiling != I915_TILING_NONE) {
604      opcode |= XY_DST_TILED;
605      dst_pitch /= 4;
606   }
607#endif
608
609   br13 = dst_pitch | (translate_raster_op(logic_op) << 16) | (1 << 29);
610   if (cpp == 2)
611      br13 |= BR13_565;
612   else
613      br13 |= BR13_8888;
614
615   blit_cmd = XY_TEXT_IMMEDIATE_BLIT_CMD | XY_TEXT_BYTE_PACKED; /* packing? */
616   if (dst_tiling != I915_TILING_NONE)
617      blit_cmd |= XY_DST_TILED;
618
619   BEGIN_BATCH(8 + 3, REFERENCES_CLIPRECTS);
620   OUT_BATCH(opcode);
621   OUT_BATCH(br13);
622   OUT_BATCH((0 << 16) | 0); /* clip x1, y1 */
623   OUT_BATCH((100 << 16) | 100); /* clip x2, y2 */
624   OUT_RELOC(dst_buffer,
625	     I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
626	     dst_offset);
627   OUT_BATCH(0); /* bg */
628   OUT_BATCH(fg_color); /* fg */
629   OUT_BATCH(0); /* pattern base addr */
630
631   OUT_BATCH(blit_cmd | ((3 - 2) + dwords));
632   OUT_BATCH((y << 16) | x);
633   OUT_BATCH(((y + h) << 16) | (x + w));
634   ADVANCE_BATCH();
635
636   intel_batchbuffer_data( intel->batch,
637			   src_bits,
638			   dwords * 4,
639			   REFERENCES_CLIPRECTS );
640
641   intel_batchbuffer_emit_mi_flush(intel->batch);
642}
643