i965: Add support for using the BLT ring on gen6.
[mesa.git] / src / mesa / drivers / dri / intel / intel_blit.c
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 "main/mtypes.h"
30 #include "main/context.h"
31 #include "main/enums.h"
32 #include "main/colormac.h"
33
34 #include "intel_blit.h"
35 #include "intel_buffers.h"
36 #include "intel_context.h"
37 #include "intel_fbo.h"
38 #include "intel_reg.h"
39 #include "intel_regions.h"
40 #include "intel_batchbuffer.h"
41
42 #define FILE_DEBUG_FLAG DEBUG_BLIT
43
44 static GLuint translate_raster_op(GLenum logicop)
45 {
46 switch(logicop) {
47 case GL_CLEAR: return 0x00;
48 case GL_AND: return 0x88;
49 case GL_AND_REVERSE: return 0x44;
50 case GL_COPY: return 0xCC;
51 case GL_AND_INVERTED: return 0x22;
52 case GL_NOOP: return 0xAA;
53 case GL_XOR: return 0x66;
54 case GL_OR: return 0xEE;
55 case GL_NOR: return 0x11;
56 case GL_EQUIV: return 0x99;
57 case GL_INVERT: return 0x55;
58 case GL_OR_REVERSE: return 0xDD;
59 case GL_COPY_INVERTED: return 0x33;
60 case GL_OR_INVERTED: return 0xBB;
61 case GL_NAND: return 0x77;
62 case GL_SET: return 0xFF;
63 default: return 0;
64 }
65 }
66
67 static uint32_t
68 br13_for_cpp(int cpp)
69 {
70 switch (cpp) {
71 case 4:
72 return BR13_8888;
73 break;
74 case 2:
75 return BR13_565;
76 break;
77 case 1:
78 return BR13_8;
79 break;
80 default:
81 assert(0);
82 return 0;
83 }
84 }
85
86 /* Copy BitBlt
87 */
88 GLboolean
89 intelEmitCopyBlit(struct intel_context *intel,
90 GLuint cpp,
91 GLshort src_pitch,
92 drm_intel_bo *src_buffer,
93 GLuint src_offset,
94 uint32_t src_tiling,
95 GLshort dst_pitch,
96 drm_intel_bo *dst_buffer,
97 GLuint dst_offset,
98 uint32_t dst_tiling,
99 GLshort src_x, GLshort src_y,
100 GLshort dst_x, GLshort dst_y,
101 GLshort w, GLshort h,
102 GLenum logic_op)
103 {
104 GLuint CMD, BR13, pass = 0;
105 int dst_y2 = dst_y + h;
106 int dst_x2 = dst_x + w;
107 drm_intel_bo *aper_array[3];
108 BATCH_LOCALS;
109
110 if (dst_tiling != I915_TILING_NONE) {
111 if (dst_offset & 4095)
112 return GL_FALSE;
113 if (dst_tiling == I915_TILING_Y)
114 return GL_FALSE;
115 }
116 if (src_tiling != I915_TILING_NONE) {
117 if (src_offset & 4095)
118 return GL_FALSE;
119 if (src_tiling == I915_TILING_Y)
120 return GL_FALSE;
121 }
122
123 /* do space check before going any further */
124 do {
125 aper_array[0] = intel->batch->buf;
126 aper_array[1] = dst_buffer;
127 aper_array[2] = src_buffer;
128
129 if (dri_bufmgr_check_aperture_space(aper_array, 3) != 0) {
130 intel_batchbuffer_flush(intel->batch);
131 pass++;
132 } else
133 break;
134 } while (pass < 2);
135
136 if (pass >= 2)
137 return GL_FALSE;
138
139 intel_batchbuffer_require_space(intel->batch, 8 * 4, true);
140 DBG("%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
141 __FUNCTION__,
142 src_buffer, src_pitch, src_offset, src_x, src_y,
143 dst_buffer, dst_pitch, dst_offset, dst_x, dst_y, w, h);
144
145 src_pitch *= cpp;
146 dst_pitch *= cpp;
147
148 BR13 = br13_for_cpp(cpp) | translate_raster_op(logic_op) << 16;
149
150 switch (cpp) {
151 case 1:
152 case 2:
153 CMD = XY_SRC_COPY_BLT_CMD;
154 break;
155 case 4:
156 CMD = XY_SRC_COPY_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
157 break;
158 default:
159 return GL_FALSE;
160 }
161
162 #ifndef I915
163 if (dst_tiling != I915_TILING_NONE) {
164 CMD |= XY_DST_TILED;
165 dst_pitch /= 4;
166 }
167 if (src_tiling != I915_TILING_NONE) {
168 CMD |= XY_SRC_TILED;
169 src_pitch /= 4;
170 }
171 #endif
172
173 if (dst_y2 <= dst_y || dst_x2 <= dst_x) {
174 return GL_TRUE;
175 }
176
177 assert(dst_x < dst_x2);
178 assert(dst_y < dst_y2);
179
180 BEGIN_BATCH_BLT(8);
181 OUT_BATCH(CMD);
182 OUT_BATCH(BR13 | (uint16_t)dst_pitch);
183 OUT_BATCH((dst_y << 16) | dst_x);
184 OUT_BATCH((dst_y2 << 16) | dst_x2);
185 OUT_RELOC_FENCED(dst_buffer,
186 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
187 dst_offset);
188 OUT_BATCH((src_y << 16) | src_x);
189 OUT_BATCH((uint16_t)src_pitch);
190 OUT_RELOC_FENCED(src_buffer,
191 I915_GEM_DOMAIN_RENDER, 0,
192 src_offset);
193 ADVANCE_BATCH();
194
195 intel_batchbuffer_emit_mi_flush(intel->batch);
196
197 return GL_TRUE;
198 }
199
200
201 /**
202 * Use blitting to clear the renderbuffers named by 'flags'.
203 * Note: we can't use the ctx->DrawBuffer->_ColorDrawBufferIndexes field
204 * since that might include software renderbuffers or renderbuffers
205 * which we're clearing with triangles.
206 * \param mask bitmask of BUFFER_BIT_* values indicating buffers to clear
207 */
208 void
209 intelClearWithBlit(struct gl_context *ctx, GLbitfield mask)
210 {
211 struct intel_context *intel = intel_context(ctx);
212 struct gl_framebuffer *fb = ctx->DrawBuffer;
213 GLuint clear_depth;
214 GLboolean all;
215 GLint cx, cy, cw, ch;
216 BATCH_LOCALS;
217
218 /*
219 * Compute values for clearing the buffers.
220 */
221 clear_depth = 0;
222 if (mask & BUFFER_BIT_DEPTH) {
223 clear_depth = (GLuint) (fb->_DepthMax * ctx->Depth.Clear);
224 }
225 if (mask & BUFFER_BIT_STENCIL) {
226 clear_depth |= (ctx->Stencil.Clear & 0xff) << 24;
227 }
228
229 cx = fb->_Xmin;
230 if (fb->Name == 0)
231 cy = ctx->DrawBuffer->Height - fb->_Ymax;
232 else
233 cy = fb->_Ymin;
234 cw = fb->_Xmax - fb->_Xmin;
235 ch = fb->_Ymax - fb->_Ymin;
236
237 if (cw == 0 || ch == 0)
238 return;
239
240 GLuint buf;
241 all = (cw == fb->Width && ch == fb->Height);
242
243 /* Loop over all renderbuffers */
244 for (buf = 0; buf < BUFFER_COUNT && mask; buf++) {
245 const GLbitfield bufBit = 1 << buf;
246 struct intel_renderbuffer *irb;
247 drm_intel_bo *write_buffer;
248 int x1, y1, x2, y2;
249 uint32_t clear_val;
250 uint32_t BR13, CMD;
251 int pitch, cpp;
252 drm_intel_bo *aper_array[2];
253
254 if (!(mask & bufBit))
255 continue;
256
257 /* OK, clear this renderbuffer */
258 irb = intel_get_renderbuffer(fb, buf);
259 write_buffer = intel_region_buffer(intel, irb->region,
260 all ? INTEL_WRITE_FULL :
261 INTEL_WRITE_PART);
262 x1 = cx + irb->region->draw_x;
263 y1 = cy + irb->region->draw_y;
264 x2 = cx + cw + irb->region->draw_x;
265 y2 = cy + ch + irb->region->draw_y;
266
267 pitch = irb->region->pitch;
268 cpp = irb->region->cpp;
269
270 DBG("%s dst:buf(%p)/%d %d,%d sz:%dx%d\n",
271 __FUNCTION__,
272 irb->region->buffer, (pitch * cpp),
273 x1, y1, x2 - x1, y2 - y1);
274
275 BR13 = br13_for_cpp(cpp) | 0xf0 << 16;
276 CMD = XY_COLOR_BLT_CMD;
277
278 /* Setup the blit command */
279 if (cpp == 4) {
280 if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL) {
281 if (mask & BUFFER_BIT_DEPTH)
282 CMD |= XY_BLT_WRITE_RGB;
283 if (mask & BUFFER_BIT_STENCIL)
284 CMD |= XY_BLT_WRITE_ALPHA;
285 } else {
286 /* clearing RGBA */
287 CMD |= XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
288 }
289 }
290
291 assert(irb->region->tiling != I915_TILING_Y);
292
293 #ifndef I915
294 if (irb->region->tiling != I915_TILING_NONE) {
295 CMD |= XY_DST_TILED;
296 pitch /= 4;
297 }
298 #endif
299 BR13 |= (pitch * cpp);
300
301 if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL) {
302 clear_val = clear_depth;
303 } else {
304 uint8_t clear[4];
305 GLclampf *color = ctx->Color.ClearColor;
306
307 CLAMPED_FLOAT_TO_UBYTE(clear[0], color[0]);
308 CLAMPED_FLOAT_TO_UBYTE(clear[1], color[1]);
309 CLAMPED_FLOAT_TO_UBYTE(clear[2], color[2]);
310 CLAMPED_FLOAT_TO_UBYTE(clear[3], color[3]);
311
312 switch (irb->Base.Format) {
313 case MESA_FORMAT_ARGB8888:
314 case MESA_FORMAT_XRGB8888:
315 clear_val = PACK_COLOR_8888(clear[3], clear[0],
316 clear[1], clear[2]);
317 break;
318 case MESA_FORMAT_RGB565:
319 clear_val = PACK_COLOR_565(clear[0], clear[1], clear[2]);
320 break;
321 case MESA_FORMAT_ARGB4444:
322 clear_val = PACK_COLOR_4444(clear[3], clear[0],
323 clear[1], clear[2]);
324 break;
325 case MESA_FORMAT_ARGB1555:
326 clear_val = PACK_COLOR_1555(clear[3], clear[0],
327 clear[1], clear[2]);
328 break;
329 case MESA_FORMAT_A8:
330 clear_val = PACK_COLOR_8888(clear[3], clear[3],
331 clear[3], clear[3]);
332 break;
333 default:
334 _mesa_problem(ctx, "Unexpected renderbuffer format: %d\n",
335 irb->Base.Format);
336 clear_val = 0;
337 }
338 }
339
340 assert(x1 < x2);
341 assert(y1 < y2);
342
343 /* do space check before going any further */
344 aper_array[0] = intel->batch->buf;
345 aper_array[1] = write_buffer;
346
347 if (drm_intel_bufmgr_check_aperture_space(aper_array,
348 ARRAY_SIZE(aper_array)) != 0) {
349 intel_batchbuffer_flush(intel->batch);
350 }
351
352 BEGIN_BATCH_BLT(6);
353 OUT_BATCH(CMD);
354 OUT_BATCH(BR13);
355 OUT_BATCH((y1 << 16) | x1);
356 OUT_BATCH((y2 << 16) | x2);
357 OUT_RELOC_FENCED(write_buffer,
358 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
359 0);
360 OUT_BATCH(clear_val);
361 ADVANCE_BATCH();
362
363 if (intel->always_flush_cache)
364 intel_batchbuffer_emit_mi_flush(intel->batch);
365
366 if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL)
367 mask &= ~(BUFFER_BIT_DEPTH | BUFFER_BIT_STENCIL);
368 else
369 mask &= ~bufBit; /* turn off bit, for faster loop exit */
370 }
371 }
372
373 GLboolean
374 intelEmitImmediateColorExpandBlit(struct intel_context *intel,
375 GLuint cpp,
376 GLubyte *src_bits, GLuint src_size,
377 GLuint fg_color,
378 GLshort dst_pitch,
379 drm_intel_bo *dst_buffer,
380 GLuint dst_offset,
381 uint32_t dst_tiling,
382 GLshort x, GLshort y,
383 GLshort w, GLshort h,
384 GLenum logic_op)
385 {
386 int dwords = ALIGN(src_size, 8) / 4;
387 uint32_t opcode, br13, blit_cmd;
388
389 if (dst_tiling != I915_TILING_NONE) {
390 if (dst_offset & 4095)
391 return GL_FALSE;
392 if (dst_tiling == I915_TILING_Y)
393 return GL_FALSE;
394 }
395
396 assert( logic_op - GL_CLEAR >= 0 );
397 assert( logic_op - GL_CLEAR < 0x10 );
398 assert(dst_pitch > 0);
399
400 if (w < 0 || h < 0)
401 return GL_TRUE;
402
403 dst_pitch *= cpp;
404
405 DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d, %d bytes %d dwords\n",
406 __FUNCTION__,
407 dst_buffer, dst_pitch, dst_offset, x, y, w, h, src_size, dwords);
408
409 intel_batchbuffer_require_space( intel->batch,
410 (8 * 4) +
411 (3 * 4) +
412 dwords * 4, true);
413
414 opcode = XY_SETUP_BLT_CMD;
415 if (cpp == 4)
416 opcode |= XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
417 #ifndef I915
418 if (dst_tiling != I915_TILING_NONE) {
419 opcode |= XY_DST_TILED;
420 dst_pitch /= 4;
421 }
422 #endif
423
424 br13 = dst_pitch | (translate_raster_op(logic_op) << 16) | (1 << 29);
425 br13 |= br13_for_cpp(cpp);
426
427 blit_cmd = XY_TEXT_IMMEDIATE_BLIT_CMD | XY_TEXT_BYTE_PACKED; /* packing? */
428 if (dst_tiling != I915_TILING_NONE)
429 blit_cmd |= XY_DST_TILED;
430
431 BEGIN_BATCH_BLT(8 + 3);
432 OUT_BATCH(opcode);
433 OUT_BATCH(br13);
434 OUT_BATCH((0 << 16) | 0); /* clip x1, y1 */
435 OUT_BATCH((100 << 16) | 100); /* clip x2, y2 */
436 OUT_RELOC_FENCED(dst_buffer,
437 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
438 dst_offset);
439 OUT_BATCH(0); /* bg */
440 OUT_BATCH(fg_color); /* fg */
441 OUT_BATCH(0); /* pattern base addr */
442
443 OUT_BATCH(blit_cmd | ((3 - 2) + dwords));
444 OUT_BATCH((y << 16) | x);
445 OUT_BATCH(((y + h) << 16) | (x + w));
446 ADVANCE_BATCH();
447
448 intel_batchbuffer_data(intel->batch,
449 src_bits,
450 dwords * 4, true);
451
452 intel_batchbuffer_emit_mi_flush(intel->batch);
453
454 return GL_TRUE;
455 }
456
457 /* We don't have a memmove-type blit like some other hardware, so we'll do a
458 * rectangular blit covering a large space, then emit 1-scanline blit at the
459 * end to cover the last if we need.
460 */
461 void
462 intel_emit_linear_blit(struct intel_context *intel,
463 drm_intel_bo *dst_bo,
464 unsigned int dst_offset,
465 drm_intel_bo *src_bo,
466 unsigned int src_offset,
467 unsigned int size)
468 {
469 GLuint pitch, height;
470 GLboolean ok;
471
472 /* The pitch given to the GPU must be DWORD aligned, and
473 * we want width to match pitch. Max width is (1 << 15 - 1),
474 * rounding that down to the nearest DWORD is 1 << 15 - 4
475 */
476 pitch = MIN2(size, (1 << 15) - 4);
477 height = size / pitch;
478 ok = intelEmitCopyBlit(intel, 1,
479 pitch, src_bo, src_offset, I915_TILING_NONE,
480 pitch, dst_bo, dst_offset, I915_TILING_NONE,
481 0, 0, /* src x/y */
482 0, 0, /* dst x/y */
483 pitch, height, /* w, h */
484 GL_COPY);
485 assert(ok);
486
487 src_offset += pitch * height;
488 dst_offset += pitch * height;
489 size -= pitch * height;
490 assert (size < (1 << 15));
491 assert ((size & 3) == 0); /* Pitch must be DWORD aligned */
492 if (size != 0) {
493 ok = intelEmitCopyBlit(intel, 1,
494 size, src_bo, src_offset, I915_TILING_NONE,
495 size, dst_bo, dst_offset, I915_TILING_NONE,
496 0, 0, /* src x/y */
497 0, 0, /* dst x/y */
498 size, 1, /* w, h */
499 GL_COPY);
500 assert(ok);
501 }
502 }