Revert "mesa/st: Passthrough scissor when clearing by quad"
[mesa.git] / src / mesa / state_tracker / st_cb_clear.c
1 /**************************************************************************
2 *
3 * Copyright 2007 VMware, Inc.
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 VMWARE 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 <keithw@vmware.com>
32 * Brian Paul
33 * Michel Dänzer
34 */
35
36 #include "main/errors.h"
37 #include "main/glheader.h"
38 #include "main/accum.h"
39 #include "main/formats.h"
40 #include "main/framebuffer.h"
41 #include "main/macros.h"
42 #include "main/glformats.h"
43 #include "program/prog_instruction.h"
44 #include "st_context.h"
45 #include "st_atom.h"
46 #include "st_cb_bitmap.h"
47 #include "st_cb_clear.h"
48 #include "st_cb_fbo.h"
49 #include "st_draw.h"
50 #include "st_format.h"
51 #include "st_nir.h"
52 #include "st_program.h"
53 #include "st_util.h"
54
55 #include "pipe/p_context.h"
56 #include "pipe/p_shader_tokens.h"
57 #include "pipe/p_state.h"
58 #include "pipe/p_defines.h"
59 #include "util/u_format.h"
60 #include "util/u_inlines.h"
61 #include "util/u_simple_shaders.h"
62
63 #include "cso_cache/cso_context.h"
64
65
66 /**
67 * Do per-context initialization for glClear.
68 */
69 void
70 st_init_clear(struct st_context *st)
71 {
72 memset(&st->clear, 0, sizeof(st->clear));
73
74 st->clear.raster.half_pixel_center = 1;
75 st->clear.raster.bottom_edge_rule = 1;
76 st->clear.raster.depth_clip_near = 1;
77 st->clear.raster.depth_clip_far = 1;
78 }
79
80
81 /**
82 * Free per-context state for glClear.
83 */
84 void
85 st_destroy_clear(struct st_context *st)
86 {
87 if (st->clear.fs) {
88 cso_delete_fragment_shader(st->cso_context, st->clear.fs);
89 st->clear.fs = NULL;
90 }
91 if (st->clear.vs) {
92 cso_delete_vertex_shader(st->cso_context, st->clear.vs);
93 st->clear.vs = NULL;
94 }
95 if (st->clear.vs_layered) {
96 cso_delete_vertex_shader(st->cso_context, st->clear.vs_layered);
97 st->clear.vs_layered = NULL;
98 }
99 if (st->clear.gs_layered) {
100 cso_delete_geometry_shader(st->cso_context, st->clear.gs_layered);
101 st->clear.gs_layered = NULL;
102 }
103 }
104
105
106 /**
107 * Helper function to set the fragment shaders.
108 */
109 static inline void
110 set_fragment_shader(struct st_context *st)
111 {
112 struct pipe_screen *pscreen = st->pipe->screen;
113 bool use_nir = PIPE_SHADER_IR_NIR ==
114 pscreen->get_shader_param(pscreen, PIPE_SHADER_VERTEX,
115 PIPE_SHADER_CAP_PREFERRED_IR);
116
117 if (!st->clear.fs) {
118 if (use_nir) {
119 unsigned inputs[] = { VARYING_SLOT_VAR0 };
120 unsigned outputs[] = { FRAG_RESULT_COLOR };
121 unsigned interpolation[] = { INTERP_MODE_FLAT };
122 st->clear.fs = st_nir_make_passthrough_shader(st, "clear FS",
123 MESA_SHADER_FRAGMENT,
124 1, inputs, outputs,
125 interpolation, 0);
126 } else {
127 st->clear.fs =
128 util_make_fragment_passthrough_shader(st->pipe,
129 TGSI_SEMANTIC_GENERIC,
130 TGSI_INTERPOLATE_CONSTANT,
131 TRUE);
132 }
133 }
134
135 cso_set_fragment_shader_handle(st->cso_context, st->clear.fs);
136 }
137
138
139 static void *
140 make_nir_clear_vertex_shader(struct st_context *st, bool layered)
141 {
142 const char *shader_name = layered ? "layered clear VS" : "clear VS";
143 unsigned inputs[] = {
144 VERT_ATTRIB_POS,
145 VERT_ATTRIB_GENERIC0,
146 SYSTEM_VALUE_INSTANCE_ID,
147 };
148 unsigned outputs[] = {
149 VARYING_SLOT_POS,
150 VARYING_SLOT_VAR0,
151 VARYING_SLOT_LAYER
152 };
153
154 return st_nir_make_passthrough_shader(st, shader_name, MESA_SHADER_VERTEX,
155 layered ? 3 : 2, inputs, outputs,
156 NULL, (1 << 2));
157 }
158
159
160 /**
161 * Helper function to set the vertex shader.
162 */
163 static inline void
164 set_vertex_shader(struct st_context *st)
165 {
166 struct pipe_screen *pscreen = st->pipe->screen;
167 bool use_nir = PIPE_SHADER_IR_NIR ==
168 pscreen->get_shader_param(pscreen, PIPE_SHADER_VERTEX,
169 PIPE_SHADER_CAP_PREFERRED_IR);
170
171 /* vertex shader - still required to provide the linkage between
172 * fragment shader input semantics and vertex_element/buffers.
173 */
174 if (!st->clear.vs)
175 {
176 if (use_nir) {
177 st->clear.vs = make_nir_clear_vertex_shader(st, false);
178 } else {
179 const uint semantic_names[] = { TGSI_SEMANTIC_POSITION,
180 TGSI_SEMANTIC_GENERIC };
181 const uint semantic_indexes[] = { 0, 0 };
182 st->clear.vs = util_make_vertex_passthrough_shader(st->pipe, 2,
183 semantic_names,
184 semantic_indexes,
185 FALSE);
186 }
187 }
188
189 cso_set_vertex_shader_handle(st->cso_context, st->clear.vs);
190 cso_set_geometry_shader_handle(st->cso_context, NULL);
191 }
192
193
194 static void
195 set_vertex_shader_layered(struct st_context *st)
196 {
197 struct pipe_context *pipe = st->pipe;
198 struct pipe_screen *pscreen = pipe->screen;
199 bool use_nir = PIPE_SHADER_IR_NIR ==
200 pscreen->get_shader_param(pscreen, PIPE_SHADER_VERTEX,
201 PIPE_SHADER_CAP_PREFERRED_IR);
202
203 if (!pipe->screen->get_param(pipe->screen, PIPE_CAP_TGSI_INSTANCEID)) {
204 assert(!"Got layered clear, but VS instancing is unsupported");
205 set_vertex_shader(st);
206 return;
207 }
208
209 if (!st->clear.vs_layered) {
210 bool vs_layer =
211 pipe->screen->get_param(pipe->screen, PIPE_CAP_TGSI_VS_LAYER_VIEWPORT);
212 if (vs_layer) {
213 st->clear.vs_layered =
214 use_nir ? make_nir_clear_vertex_shader(st, true)
215 : util_make_layered_clear_vertex_shader(pipe);
216 } else {
217 st->clear.vs_layered = util_make_layered_clear_helper_vertex_shader(pipe);
218 st->clear.gs_layered = util_make_layered_clear_geometry_shader(pipe);
219 }
220 }
221
222 cso_set_vertex_shader_handle(st->cso_context, st->clear.vs_layered);
223 cso_set_geometry_shader_handle(st->cso_context, st->clear.gs_layered);
224 }
225
226
227 /**
228 * Do glClear by drawing a quadrilateral.
229 * The vertices of the quad will be computed from the
230 * ctx->DrawBuffer->_X/Ymin/max fields.
231 */
232 static void
233 clear_with_quad(struct gl_context *ctx, unsigned clear_buffers)
234 {
235 struct st_context *st = st_context(ctx);
236 struct cso_context *cso = st->cso_context;
237 const struct gl_framebuffer *fb = ctx->DrawBuffer;
238 const GLfloat fb_width = (GLfloat) fb->Width;
239 const GLfloat fb_height = (GLfloat) fb->Height;
240
241 _mesa_update_draw_buffer_bounds(ctx, ctx->DrawBuffer);
242
243 const GLfloat x0 = (GLfloat) ctx->DrawBuffer->_Xmin / fb_width * 2.0f - 1.0f;
244 const GLfloat x1 = (GLfloat) ctx->DrawBuffer->_Xmax / fb_width * 2.0f - 1.0f;
245 const GLfloat y0 = (GLfloat) ctx->DrawBuffer->_Ymin / fb_height * 2.0f - 1.0f;
246 const GLfloat y1 = (GLfloat) ctx->DrawBuffer->_Ymax / fb_height * 2.0f - 1.0f;
247 unsigned num_layers = st->state.fb_num_layers;
248
249 /*
250 printf("%s %s%s%s %f,%f %f,%f\n", __func__,
251 color ? "color, " : "",
252 depth ? "depth, " : "",
253 stencil ? "stencil" : "",
254 x0, y0,
255 x1, y1);
256 */
257
258 cso_save_state(cso, (CSO_BIT_BLEND |
259 CSO_BIT_STENCIL_REF |
260 CSO_BIT_DEPTH_STENCIL_ALPHA |
261 CSO_BIT_RASTERIZER |
262 CSO_BIT_SAMPLE_MASK |
263 CSO_BIT_MIN_SAMPLES |
264 CSO_BIT_VIEWPORT |
265 CSO_BIT_STREAM_OUTPUTS |
266 CSO_BIT_VERTEX_ELEMENTS |
267 CSO_BIT_AUX_VERTEX_BUFFER_SLOT |
268 CSO_BIT_PAUSE_QUERIES |
269 CSO_BITS_ALL_SHADERS));
270
271 /* blend state: RGBA masking */
272 {
273 struct pipe_blend_state blend;
274 memset(&blend, 0, sizeof(blend));
275 if (clear_buffers & PIPE_CLEAR_COLOR) {
276 int num_buffers = ctx->Extensions.EXT_draw_buffers2 ?
277 ctx->DrawBuffer->_NumColorDrawBuffers : 1;
278 int i;
279
280 blend.independent_blend_enable = num_buffers > 1;
281
282 for (i = 0; i < num_buffers; i++) {
283 if (!(clear_buffers & (PIPE_CLEAR_COLOR0 << i)))
284 continue;
285
286 blend.rt[i].colormask = GET_COLORMASK(ctx->Color.ColorMask, i);
287 }
288
289 if (ctx->Color.DitherFlag)
290 blend.dither = 1;
291 }
292 cso_set_blend(cso, &blend);
293 }
294
295 /* depth_stencil state: always pass/set to ref value */
296 {
297 struct pipe_depth_stencil_alpha_state depth_stencil;
298 memset(&depth_stencil, 0, sizeof(depth_stencil));
299 if (clear_buffers & PIPE_CLEAR_DEPTH) {
300 depth_stencil.depth.enabled = 1;
301 depth_stencil.depth.writemask = 1;
302 depth_stencil.depth.func = PIPE_FUNC_ALWAYS;
303 }
304
305 if (clear_buffers & PIPE_CLEAR_STENCIL) {
306 struct pipe_stencil_ref stencil_ref;
307 memset(&stencil_ref, 0, sizeof(stencil_ref));
308 depth_stencil.stencil[0].enabled = 1;
309 depth_stencil.stencil[0].func = PIPE_FUNC_ALWAYS;
310 depth_stencil.stencil[0].fail_op = PIPE_STENCIL_OP_REPLACE;
311 depth_stencil.stencil[0].zpass_op = PIPE_STENCIL_OP_REPLACE;
312 depth_stencil.stencil[0].zfail_op = PIPE_STENCIL_OP_REPLACE;
313 depth_stencil.stencil[0].valuemask = 0xff;
314 depth_stencil.stencil[0].writemask = ctx->Stencil.WriteMask[0] & 0xff;
315 stencil_ref.ref_value[0] = ctx->Stencil.Clear;
316 cso_set_stencil_ref(cso, &stencil_ref);
317 }
318
319 cso_set_depth_stencil_alpha(cso, &depth_stencil);
320 }
321
322 cso_set_vertex_elements(cso, 2, st->util_velems);
323 cso_set_stream_outputs(cso, 0, NULL, NULL);
324 cso_set_sample_mask(cso, ~0);
325 cso_set_min_samples(cso, 1);
326 cso_set_rasterizer(cso, &st->clear.raster);
327
328 /* viewport state: viewport matching window dims */
329 cso_set_viewport_dims(st->cso_context, fb_width, fb_height,
330 st_fb_orientation(fb) == Y_0_TOP);
331
332 set_fragment_shader(st);
333 cso_set_tessctrl_shader_handle(cso, NULL);
334 cso_set_tesseval_shader_handle(cso, NULL);
335
336 if (num_layers > 1)
337 set_vertex_shader_layered(st);
338 else
339 set_vertex_shader(st);
340
341 /* draw quad matching scissor rect.
342 *
343 * Note: if we're only clearing depth/stencil we still setup vertices
344 * with color, but they'll be ignored.
345 *
346 * We can't translate the clear color to the colorbuffer format,
347 * because different colorbuffers may have different formats.
348 */
349 if (!st_draw_quad(st, x0, y0, x1, y1,
350 ctx->Depth.Clear * 2.0f - 1.0f,
351 0.0f, 0.0f, 0.0f, 0.0f,
352 (const float *) &ctx->Color.ClearColor.f,
353 num_layers)) {
354 _mesa_error(ctx, GL_OUT_OF_MEMORY, "glClear");
355 }
356
357 /* Restore pipe state */
358 cso_restore_state(cso);
359 }
360
361
362 /**
363 * Return if the scissor must be enabled during the clear.
364 */
365 static inline GLboolean
366 is_scissor_enabled(struct gl_context *ctx, struct gl_renderbuffer *rb)
367 {
368 const struct gl_scissor_rect *scissor = &ctx->Scissor.ScissorArray[0];
369
370 return (ctx->Scissor.EnableFlags & 1) &&
371 (scissor->X > 0 ||
372 scissor->Y > 0 ||
373 scissor->X + scissor->Width < (int)rb->Width ||
374 scissor->Y + scissor->Height < (int)rb->Height);
375 }
376
377 /**
378 * Return if window rectangles must be enabled during the clear.
379 */
380 static inline bool
381 is_window_rectangle_enabled(struct gl_context *ctx)
382 {
383 if (ctx->DrawBuffer == ctx->WinSysDrawBuffer)
384 return false;
385 return ctx->Scissor.NumWindowRects > 0 ||
386 ctx->Scissor.WindowRectMode == GL_INCLUSIVE_EXT;
387 }
388
389
390 /**
391 * Return if all of the stencil bits are masked.
392 */
393 static inline GLboolean
394 is_stencil_disabled(struct gl_context *ctx, struct gl_renderbuffer *rb)
395 {
396 const GLuint stencilMax = 0xff;
397
398 assert(_mesa_get_format_bits(rb->Format, GL_STENCIL_BITS) > 0);
399 return (ctx->Stencil.WriteMask[0] & stencilMax) == 0;
400 }
401
402
403 /**
404 * Return if any of the stencil bits are masked.
405 */
406 static inline GLboolean
407 is_stencil_masked(struct gl_context *ctx, struct gl_renderbuffer *rb)
408 {
409 const GLuint stencilMax = 0xff;
410
411 assert(_mesa_get_format_bits(rb->Format, GL_STENCIL_BITS) > 0);
412 return (ctx->Stencil.WriteMask[0] & stencilMax) != stencilMax;
413 }
414
415
416 /**
417 * Called via ctx->Driver.Clear()
418 */
419 static void
420 st_Clear(struct gl_context *ctx, GLbitfield mask)
421 {
422 struct st_context *st = st_context(ctx);
423 struct gl_renderbuffer *depthRb
424 = ctx->DrawBuffer->Attachment[BUFFER_DEPTH].Renderbuffer;
425 struct gl_renderbuffer *stencilRb
426 = ctx->DrawBuffer->Attachment[BUFFER_STENCIL].Renderbuffer;
427 GLbitfield quad_buffers = 0x0;
428 GLbitfield clear_buffers = 0x0;
429 GLuint i;
430
431 st_flush_bitmap_cache(st);
432 st_invalidate_readpix_cache(st);
433
434 /* This makes sure the pipe has the latest scissor, etc values */
435 st_validate_state(st, ST_PIPELINE_CLEAR);
436
437 if (mask & BUFFER_BITS_COLOR) {
438 for (i = 0; i < ctx->DrawBuffer->_NumColorDrawBuffers; i++) {
439 gl_buffer_index b = ctx->DrawBuffer->_ColorDrawBufferIndexes[i];
440
441 if (b != BUFFER_NONE && mask & (1 << b)) {
442 struct gl_renderbuffer *rb
443 = ctx->DrawBuffer->Attachment[b].Renderbuffer;
444 struct st_renderbuffer *strb = st_renderbuffer(rb);
445 int colormask_index = ctx->Extensions.EXT_draw_buffers2 ? i : 0;
446
447 if (!strb || !strb->surface)
448 continue;
449
450 unsigned colormask =
451 GET_COLORMASK(ctx->Color.ColorMask, colormask_index);
452
453 if (!colormask)
454 continue;
455
456 unsigned surf_colormask =
457 util_format_colormask(util_format_description(strb->surface->format));
458
459 if (is_scissor_enabled(ctx, rb) ||
460 is_window_rectangle_enabled(ctx) ||
461 ((colormask & surf_colormask) != surf_colormask))
462 quad_buffers |= PIPE_CLEAR_COLOR0 << i;
463 else
464 clear_buffers |= PIPE_CLEAR_COLOR0 << i;
465 }
466 }
467 }
468
469 if (mask & BUFFER_BIT_DEPTH) {
470 struct st_renderbuffer *strb = st_renderbuffer(depthRb);
471
472 if (strb->surface && ctx->Depth.Mask) {
473 if (is_scissor_enabled(ctx, depthRb) ||
474 is_window_rectangle_enabled(ctx))
475 quad_buffers |= PIPE_CLEAR_DEPTH;
476 else
477 clear_buffers |= PIPE_CLEAR_DEPTH;
478 }
479 }
480 if (mask & BUFFER_BIT_STENCIL) {
481 struct st_renderbuffer *strb = st_renderbuffer(stencilRb);
482
483 if (strb->surface && !is_stencil_disabled(ctx, stencilRb)) {
484 if (is_scissor_enabled(ctx, stencilRb) ||
485 is_window_rectangle_enabled(ctx) ||
486 is_stencil_masked(ctx, stencilRb))
487 quad_buffers |= PIPE_CLEAR_STENCIL;
488 else
489 clear_buffers |= PIPE_CLEAR_STENCIL;
490 }
491 }
492
493 /* Always clear depth and stencil together.
494 * This can only happen when the stencil writemask is not a full mask.
495 */
496 if (quad_buffers & PIPE_CLEAR_DEPTHSTENCIL &&
497 clear_buffers & PIPE_CLEAR_DEPTHSTENCIL) {
498 quad_buffers |= clear_buffers & PIPE_CLEAR_DEPTHSTENCIL;
499 clear_buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
500 }
501
502 /* Only use quad-based clearing for the renderbuffers which cannot
503 * use pipe->clear. We want to always use pipe->clear for the other
504 * renderbuffers, because it's likely to be faster.
505 */
506 if (clear_buffers) {
507 /* We can't translate the clear color to the colorbuffer format,
508 * because different colorbuffers may have different formats.
509 */
510 st->pipe->clear(st->pipe, clear_buffers,
511 (union pipe_color_union*)&ctx->Color.ClearColor,
512 ctx->Depth.Clear, ctx->Stencil.Clear);
513 }
514 if (quad_buffers) {
515 clear_with_quad(ctx, quad_buffers);
516 }
517 if (mask & BUFFER_BIT_ACCUM)
518 _mesa_clear_accum_buffer(ctx);
519 }
520
521
522 void
523 st_init_clear_functions(struct dd_function_table *functions)
524 {
525 functions->Clear = st_Clear;
526 }