gallium: always save and restore vertex buffers using cso_cache
[mesa.git] / src / mesa / state_tracker / st_cb_clear.c
1 /**************************************************************************
2 *
3 * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
4 * All Rights Reserved.
5 * Copyright 2009 VMware, Inc. All Rights Reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sub license, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
14 *
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
17 * of the Software.
18 *
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
22 * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
23 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
24 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
25 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 *
27 **************************************************************************/
28
29 /*
30 * Authors:
31 * Keith Whitwell <keith@tungstengraphics.com>
32 * Brian Paul
33 * Michel Dänzer
34 */
35
36 #include "main/glheader.h"
37 #include "main/formats.h"
38 #include "main/macros.h"
39 #include "program/prog_instruction.h"
40 #include "st_context.h"
41 #include "st_atom.h"
42 #include "st_cb_accum.h"
43 #include "st_cb_clear.h"
44 #include "st_cb_fbo.h"
45 #include "st_program.h"
46
47 #include "pipe/p_context.h"
48 #include "pipe/p_shader_tokens.h"
49 #include "pipe/p_state.h"
50 #include "pipe/p_defines.h"
51 #include "util/u_format.h"
52 #include "util/u_inlines.h"
53 #include "util/u_simple_shaders.h"
54 #include "util/u_draw_quad.h"
55
56 #include "cso_cache/cso_context.h"
57
58
59 /**
60 * Do per-context initialization for glClear.
61 */
62 void
63 st_init_clear(struct st_context *st)
64 {
65 struct pipe_context *pipe = st->pipe;
66 struct pipe_screen *pscreen = st->pipe->screen;
67
68 memset(&st->clear, 0, sizeof(st->clear));
69
70 st->clear.raster.gl_rasterization_rules = 1;
71 st->clear.enable_ds_separate = pscreen->get_param(pscreen, PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE);
72
73 /* fragment shader state: color pass-through program */
74 st->clear.fs = util_make_fragment_passthrough_shader(pipe);
75
76 /* vertex shader state: color/position pass-through */
77 {
78 const uint semantic_names[] = { TGSI_SEMANTIC_POSITION,
79 TGSI_SEMANTIC_COLOR };
80 const uint semantic_indexes[] = { 0, 0 };
81 st->clear.vs = util_make_vertex_passthrough_shader(pipe, 2,
82 semantic_names,
83 semantic_indexes);
84 }
85 }
86
87
88 /**
89 * Free per-context state for glClear.
90 */
91 void
92 st_destroy_clear(struct st_context *st)
93 {
94 if (st->clear.fs) {
95 cso_delete_fragment_shader(st->cso_context, st->clear.fs);
96 st->clear.fs = NULL;
97 }
98 if (st->clear.vs) {
99 cso_delete_vertex_shader(st->cso_context, st->clear.vs);
100 st->clear.vs = NULL;
101 }
102 if (st->clear.vbuf) {
103 pipe_resource_reference(&st->clear.vbuf, NULL);
104 st->clear.vbuf = NULL;
105 }
106 }
107
108
109 /**
110 * Draw a screen-aligned quadrilateral.
111 * Coords are clip coords with y=0=bottom.
112 */
113 static void
114 draw_quad(struct st_context *st,
115 float x0, float y0, float x1, float y1, GLfloat z,
116 const GLfloat color[4])
117 {
118 struct pipe_context *pipe = st->pipe;
119
120 /* XXX: Need to improve buffer_write to allow NO_WAIT (as well as
121 * no_flush) updates to buffers where we know there is no conflict
122 * with previous data. Currently using max_slots > 1 will cause
123 * synchronous rendering if the driver flushes its command buffers
124 * between one bitmap and the next. Our flush hook below isn't
125 * sufficient to catch this as the driver doesn't tell us when it
126 * flushes its own command buffers. Until this gets fixed, pay the
127 * price of allocating a new buffer for each bitmap cache-flush to
128 * avoid synchronous rendering.
129 */
130 const GLuint max_slots = 1; /* 1024 / sizeof(st->clear.vertices); */
131 GLuint i;
132
133 if (st->clear.vbuf_slot >= max_slots) {
134 pipe_resource_reference(&st->clear.vbuf, NULL);
135 st->clear.vbuf_slot = 0;
136 }
137
138 if (!st->clear.vbuf) {
139 st->clear.vbuf = pipe_buffer_create(pipe->screen,
140 PIPE_BIND_VERTEX_BUFFER,
141 max_slots * sizeof(st->clear.vertices));
142 }
143
144 /* positions */
145 st->clear.vertices[0][0][0] = x0;
146 st->clear.vertices[0][0][1] = y0;
147
148 st->clear.vertices[1][0][0] = x1;
149 st->clear.vertices[1][0][1] = y0;
150
151 st->clear.vertices[2][0][0] = x1;
152 st->clear.vertices[2][0][1] = y1;
153
154 st->clear.vertices[3][0][0] = x0;
155 st->clear.vertices[3][0][1] = y1;
156
157 /* same for all verts: */
158 for (i = 0; i < 4; i++) {
159 st->clear.vertices[i][0][2] = z;
160 st->clear.vertices[i][0][3] = 1.0;
161 st->clear.vertices[i][1][0] = color[0];
162 st->clear.vertices[i][1][1] = color[1];
163 st->clear.vertices[i][1][2] = color[2];
164 st->clear.vertices[i][1][3] = color[3];
165 }
166
167 /* put vertex data into vbuf */
168 pipe_buffer_write_nooverlap(st->pipe, st->clear.vbuf,
169 st->clear.vbuf_slot
170 * sizeof(st->clear.vertices),
171 sizeof(st->clear.vertices),
172 st->clear.vertices);
173
174 /* draw */
175 util_draw_vertex_buffer(pipe,
176 st->cso_context,
177 st->clear.vbuf,
178 st->clear.vbuf_slot * sizeof(st->clear.vertices),
179 PIPE_PRIM_TRIANGLE_FAN,
180 4, /* verts */
181 2); /* attribs/vert */
182
183 /* Increment slot */
184 st->clear.vbuf_slot++;
185 }
186
187
188
189 /**
190 * Do glClear by drawing a quadrilateral.
191 * The vertices of the quad will be computed from the
192 * ctx->DrawBuffer->_X/Ymin/max fields.
193 */
194 static void
195 clear_with_quad(struct gl_context *ctx,
196 GLboolean color, GLboolean depth, GLboolean stencil)
197 {
198 struct st_context *st = st_context(ctx);
199 const struct gl_framebuffer *fb = ctx->DrawBuffer;
200 const GLfloat fb_width = (GLfloat) fb->Width;
201 const GLfloat fb_height = (GLfloat) fb->Height;
202 const GLfloat x0 = (GLfloat) ctx->DrawBuffer->_Xmin / fb_width * 2.0f - 1.0f;
203 const GLfloat x1 = (GLfloat) ctx->DrawBuffer->_Xmax / fb_width * 2.0f - 1.0f;
204 const GLfloat y0 = (GLfloat) ctx->DrawBuffer->_Ymin / fb_height * 2.0f - 1.0f;
205 const GLfloat y1 = (GLfloat) ctx->DrawBuffer->_Ymax / fb_height * 2.0f - 1.0f;
206
207 /*
208 printf("%s %s%s%s %f,%f %f,%f\n", __FUNCTION__,
209 color ? "color, " : "",
210 depth ? "depth, " : "",
211 stencil ? "stencil" : "",
212 x0, y0,
213 x1, y1);
214 */
215
216 cso_save_blend(st->cso_context);
217 cso_save_stencil_ref(st->cso_context);
218 cso_save_depth_stencil_alpha(st->cso_context);
219 cso_save_rasterizer(st->cso_context);
220 cso_save_viewport(st->cso_context);
221 cso_save_clip(st->cso_context);
222 cso_save_fragment_shader(st->cso_context);
223 cso_save_vertex_shader(st->cso_context);
224 cso_save_vertex_elements(st->cso_context);
225 cso_save_vertex_buffers(st->cso_context);
226
227 /* blend state: RGBA masking */
228 {
229 struct pipe_blend_state blend;
230 memset(&blend, 0, sizeof(blend));
231 blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
232 blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
233 blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
234 blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
235 if (color) {
236 if (ctx->Color.ColorMask[0][0])
237 blend.rt[0].colormask |= PIPE_MASK_R;
238 if (ctx->Color.ColorMask[0][1])
239 blend.rt[0].colormask |= PIPE_MASK_G;
240 if (ctx->Color.ColorMask[0][2])
241 blend.rt[0].colormask |= PIPE_MASK_B;
242 if (ctx->Color.ColorMask[0][3])
243 blend.rt[0].colormask |= PIPE_MASK_A;
244 if (st->ctx->Color.DitherFlag)
245 blend.dither = 1;
246 }
247 cso_set_blend(st->cso_context, &blend);
248 }
249
250 /* depth_stencil state: always pass/set to ref value */
251 {
252 struct pipe_depth_stencil_alpha_state depth_stencil;
253 memset(&depth_stencil, 0, sizeof(depth_stencil));
254 if (depth) {
255 depth_stencil.depth.enabled = 1;
256 depth_stencil.depth.writemask = 1;
257 depth_stencil.depth.func = PIPE_FUNC_ALWAYS;
258 }
259
260 if (stencil) {
261 struct pipe_stencil_ref stencil_ref;
262 memset(&stencil_ref, 0, sizeof(stencil_ref));
263 depth_stencil.stencil[0].enabled = 1;
264 depth_stencil.stencil[0].func = PIPE_FUNC_ALWAYS;
265 depth_stencil.stencil[0].fail_op = PIPE_STENCIL_OP_REPLACE;
266 depth_stencil.stencil[0].zpass_op = PIPE_STENCIL_OP_REPLACE;
267 depth_stencil.stencil[0].zfail_op = PIPE_STENCIL_OP_REPLACE;
268 depth_stencil.stencil[0].valuemask = 0xff;
269 depth_stencil.stencil[0].writemask = ctx->Stencil.WriteMask[0] & 0xff;
270 stencil_ref.ref_value[0] = ctx->Stencil.Clear;
271 cso_set_stencil_ref(st->cso_context, &stencil_ref);
272 }
273
274 cso_set_depth_stencil_alpha(st->cso_context, &depth_stencil);
275 }
276
277 cso_set_vertex_elements(st->cso_context, 2, st->velems_util_draw);
278
279 cso_set_rasterizer(st->cso_context, &st->clear.raster);
280
281 /* viewport state: viewport matching window dims */
282 {
283 const GLboolean invert = (st_fb_orientation(fb) == Y_0_TOP);
284 struct pipe_viewport_state vp;
285 vp.scale[0] = 0.5f * fb_width;
286 vp.scale[1] = fb_height * (invert ? -0.5f : 0.5f);
287 vp.scale[2] = 1.0f;
288 vp.scale[3] = 1.0f;
289 vp.translate[0] = 0.5f * fb_width;
290 vp.translate[1] = 0.5f * fb_height;
291 vp.translate[2] = 0.0f;
292 vp.translate[3] = 0.0f;
293 cso_set_viewport(st->cso_context, &vp);
294 }
295
296 cso_set_clip(st->cso_context, &st->clear.clip);
297 cso_set_fragment_shader_handle(st->cso_context, st->clear.fs);
298 cso_set_vertex_shader_handle(st->cso_context, st->clear.vs);
299
300 /* draw quad matching scissor rect (XXX verify coord round-off) */
301 draw_quad(st, x0, y0, x1, y1,
302 (GLfloat) ctx->Depth.Clear, ctx->Color.ClearColor);
303
304 /* Restore pipe state */
305 cso_restore_blend(st->cso_context);
306 cso_restore_stencil_ref(st->cso_context);
307 cso_restore_depth_stencil_alpha(st->cso_context);
308 cso_restore_rasterizer(st->cso_context);
309 cso_restore_viewport(st->cso_context);
310 cso_restore_clip(st->cso_context);
311 cso_restore_fragment_shader(st->cso_context);
312 cso_restore_vertex_shader(st->cso_context);
313 cso_restore_vertex_elements(st->cso_context);
314 cso_restore_vertex_buffers(st->cso_context);
315 }
316
317
318 /**
319 * Determine if we need to clear the depth buffer by drawing a quad.
320 */
321 static INLINE GLboolean
322 check_clear_color_with_quad(struct gl_context *ctx, struct gl_renderbuffer *rb)
323 {
324 if (ctx->Scissor.Enabled &&
325 (ctx->Scissor.X != 0 ||
326 ctx->Scissor.Y != 0 ||
327 ctx->Scissor.Width < rb->Width ||
328 ctx->Scissor.Height < rb->Height))
329 return GL_TRUE;
330
331 if (!ctx->Color.ColorMask[0][0] ||
332 !ctx->Color.ColorMask[0][1] ||
333 !ctx->Color.ColorMask[0][2] ||
334 !ctx->Color.ColorMask[0][3])
335 return GL_TRUE;
336
337 return GL_FALSE;
338 }
339
340
341 /**
342 * Determine if we need to clear the combiend depth/stencil buffer by
343 * drawing a quad.
344 */
345 static INLINE GLboolean
346 check_clear_depth_stencil_with_quad(struct gl_context *ctx, struct gl_renderbuffer *rb)
347 {
348 const GLuint stencilMax = 0xff;
349 GLboolean maskStencil
350 = (ctx->Stencil.WriteMask[0] & stencilMax) != stencilMax;
351
352 assert(rb->Format == MESA_FORMAT_S8 ||
353 rb->Format == MESA_FORMAT_Z24_S8 ||
354 rb->Format == MESA_FORMAT_S8_Z24);
355
356 if (ctx->Scissor.Enabled &&
357 (ctx->Scissor.X != 0 ||
358 ctx->Scissor.Y != 0 ||
359 ctx->Scissor.Width < rb->Width ||
360 ctx->Scissor.Height < rb->Height))
361 return GL_TRUE;
362
363 if (maskStencil)
364 return GL_TRUE;
365
366 return GL_FALSE;
367 }
368
369
370 /**
371 * Determine if we need to clear the depth buffer by drawing a quad.
372 */
373 static INLINE GLboolean
374 check_clear_depth_with_quad(struct gl_context *ctx, struct gl_renderbuffer *rb,
375 boolean ds_separate)
376 {
377 const struct st_renderbuffer *strb = st_renderbuffer(rb);
378 const GLboolean isDS = util_format_is_depth_and_stencil(strb->surface->format);
379
380 if (ctx->Scissor.Enabled &&
381 (ctx->Scissor.X != 0 ||
382 ctx->Scissor.Y != 0 ||
383 ctx->Scissor.Width < rb->Width ||
384 ctx->Scissor.Height < rb->Height))
385 return GL_TRUE;
386
387 if (!ds_separate && isDS && ctx->DrawBuffer->Visual.stencilBits > 0)
388 return GL_TRUE;
389
390 return GL_FALSE;
391 }
392
393
394 /**
395 * Determine if we need to clear the stencil buffer by drawing a quad.
396 */
397 static INLINE GLboolean
398 check_clear_stencil_with_quad(struct gl_context *ctx, struct gl_renderbuffer *rb,
399 boolean ds_separate)
400 {
401 const struct st_renderbuffer *strb = st_renderbuffer(rb);
402 const GLboolean isDS = util_format_is_depth_and_stencil(strb->surface->format);
403 const GLuint stencilMax = 0xff;
404 const GLboolean maskStencil
405 = (ctx->Stencil.WriteMask[0] & stencilMax) != stencilMax;
406
407 assert(rb->Format == MESA_FORMAT_S8 ||
408 rb->Format == MESA_FORMAT_Z24_S8 ||
409 rb->Format == MESA_FORMAT_S8_Z24);
410
411 if (maskStencil)
412 return GL_TRUE;
413
414 if (ctx->Scissor.Enabled &&
415 (ctx->Scissor.X != 0 ||
416 ctx->Scissor.Y != 0 ||
417 ctx->Scissor.Width < rb->Width ||
418 ctx->Scissor.Height < rb->Height))
419 return GL_TRUE;
420
421 /* This is correct, but it is necessary to look at the depth clear
422 * value held in the surface when it comes time to issue the clear,
423 * rather than taking depth and stencil clear values from the
424 * current state.
425 */
426 if (!ds_separate && isDS && ctx->DrawBuffer->Visual.depthBits > 0)
427 return GL_TRUE;
428
429 return GL_FALSE;
430 }
431
432
433
434 /**
435 * Called when we need to flush.
436 */
437 void
438 st_flush_clear(struct st_context *st)
439 {
440 /* Release vertex buffer to avoid synchronous rendering if we were
441 * to map it in the next frame.
442 */
443 pipe_resource_reference(&st->clear.vbuf, NULL);
444 st->clear.vbuf_slot = 0;
445 }
446
447
448
449 /**
450 * Called via ctx->Driver.Clear()
451 */
452 static void
453 st_Clear(struct gl_context *ctx, GLbitfield mask)
454 {
455 static const GLbitfield BUFFER_BITS_DS
456 = (BUFFER_BIT_DEPTH | BUFFER_BIT_STENCIL);
457 struct st_context *st = st_context(ctx);
458 struct gl_renderbuffer *depthRb
459 = ctx->DrawBuffer->Attachment[BUFFER_DEPTH].Renderbuffer;
460 struct gl_renderbuffer *stencilRb
461 = ctx->DrawBuffer->Attachment[BUFFER_STENCIL].Renderbuffer;
462 GLbitfield quad_buffers = 0x0;
463 GLbitfield clear_buffers = 0x0;
464 GLuint i;
465
466 /* This makes sure the pipe has the latest scissor, etc values */
467 st_validate_state( st );
468
469 if (mask & BUFFER_BITS_COLOR) {
470 for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) {
471 GLuint b = ctx->DrawBuffer->_ColorDrawBufferIndexes[i];
472
473 if (mask & (1 << b)) {
474 struct gl_renderbuffer *rb
475 = ctx->DrawBuffer->Attachment[b].Renderbuffer;
476 struct st_renderbuffer *strb = st_renderbuffer(rb);
477
478 if (!strb || !strb->surface)
479 continue;
480
481 if (check_clear_color_with_quad( ctx, rb ))
482 quad_buffers |= PIPE_CLEAR_COLOR;
483 else
484 clear_buffers |= PIPE_CLEAR_COLOR;
485 }
486 }
487 }
488
489 if ((mask & BUFFER_BITS_DS) == BUFFER_BITS_DS && depthRb == stencilRb) {
490 /* clearing combined depth + stencil */
491 struct st_renderbuffer *strb = st_renderbuffer(depthRb);
492
493 if (strb->surface) {
494 if (check_clear_depth_stencil_with_quad(ctx, depthRb))
495 quad_buffers |= PIPE_CLEAR_DEPTHSTENCIL;
496 else
497 clear_buffers |= PIPE_CLEAR_DEPTHSTENCIL;
498 }
499 }
500 else {
501 /* separate depth/stencil clears */
502 /* I don't think truly separate buffers are actually possible in gallium or hw? */
503 if (mask & BUFFER_BIT_DEPTH) {
504 struct st_renderbuffer *strb = st_renderbuffer(depthRb);
505
506 if (strb->surface) {
507 if (check_clear_depth_with_quad(ctx, depthRb,
508 st->clear.enable_ds_separate))
509 quad_buffers |= PIPE_CLEAR_DEPTH;
510 else
511 clear_buffers |= PIPE_CLEAR_DEPTH;
512 }
513 }
514 if (mask & BUFFER_BIT_STENCIL) {
515 struct st_renderbuffer *strb = st_renderbuffer(stencilRb);
516
517 if (strb->surface) {
518 if (check_clear_stencil_with_quad(ctx, stencilRb,
519 st->clear.enable_ds_separate))
520 quad_buffers |= PIPE_CLEAR_STENCIL;
521 else
522 clear_buffers |= PIPE_CLEAR_STENCIL;
523 }
524 }
525 }
526
527 /*
528 * If we're going to use clear_with_quad() for any reason, use it for
529 * everything possible.
530 */
531 if (quad_buffers) {
532 quad_buffers |= clear_buffers;
533 clear_with_quad(ctx,
534 quad_buffers & PIPE_CLEAR_COLOR,
535 quad_buffers & PIPE_CLEAR_DEPTH,
536 quad_buffers & PIPE_CLEAR_STENCIL);
537 } else if (clear_buffers) {
538 /* driver cannot know it can clear everything if the buffer
539 * is a combined depth/stencil buffer but this wasn't actually
540 * required from the visual. Hence fix this up to avoid potential
541 * read-modify-write in the driver.
542 */
543 if ((clear_buffers & PIPE_CLEAR_DEPTHSTENCIL) &&
544 ((clear_buffers & PIPE_CLEAR_DEPTHSTENCIL) != PIPE_CLEAR_DEPTHSTENCIL) &&
545 (depthRb == stencilRb) &&
546 (ctx->DrawBuffer->Visual.depthBits == 0 ||
547 ctx->DrawBuffer->Visual.stencilBits == 0))
548 clear_buffers |= PIPE_CLEAR_DEPTHSTENCIL;
549 st->pipe->clear(st->pipe, clear_buffers, ctx->Color.ClearColor,
550 ctx->Depth.Clear, ctx->Stencil.Clear);
551 }
552 if (mask & BUFFER_BIT_ACCUM)
553 st_clear_accum_buffer(ctx,
554 ctx->DrawBuffer->Attachment[BUFFER_ACCUM].Renderbuffer);
555 }
556
557
558 void
559 st_init_clear_functions(struct dd_function_table *functions)
560 {
561 functions->Clear = st_Clear;
562 }