gallium/util: replace pipe_mutex_lock() with mtx_lock()
[mesa.git] / src / gallium / auxiliary / util / u_blit.c
1 /**************************************************************************
2 *
3 * Copyright 2008 VMware, Inc.
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 VMWARE 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 * @file
30 * Copy/blit pixel rect between surfaces
31 *
32 * @author Brian Paul
33 */
34
35
36 #include "pipe/p_context.h"
37 #include "util/u_debug.h"
38 #include "pipe/p_defines.h"
39 #include "util/u_inlines.h"
40 #include "pipe/p_shader_tokens.h"
41 #include "pipe/p_state.h"
42
43 #include "util/u_blit.h"
44 #include "util/u_draw_quad.h"
45 #include "util/u_format.h"
46 #include "util/u_math.h"
47 #include "util/u_memory.h"
48 #include "util/u_sampler.h"
49 #include "util/u_texture.h"
50 #include "util/u_simple_shaders.h"
51
52 #include "cso_cache/cso_context.h"
53
54
55 struct blit_state
56 {
57 struct pipe_context *pipe;
58 struct cso_context *cso;
59
60 struct pipe_blend_state blend_write_color;
61 struct pipe_depth_stencil_alpha_state dsa_keep_depthstencil;
62 struct pipe_rasterizer_state rasterizer;
63 struct pipe_sampler_state sampler;
64 struct pipe_viewport_state viewport;
65 struct pipe_vertex_element velem[2];
66
67 void *vs;
68 void *fs[PIPE_MAX_TEXTURE_TYPES][TGSI_WRITEMASK_XYZW + 1][3];
69
70 struct pipe_resource *vbuf; /**< quad vertices */
71 unsigned vbuf_slot;
72
73 float vertices[4][2][4]; /**< vertex/texcoords for quad */
74 };
75
76
77 /**
78 * Create state object for blit.
79 * Intended to be created once and re-used for many blit() calls.
80 */
81 struct blit_state *
82 util_create_blit(struct pipe_context *pipe, struct cso_context *cso)
83 {
84 struct blit_state *ctx;
85 uint i;
86
87 ctx = CALLOC_STRUCT(blit_state);
88 if (!ctx)
89 return NULL;
90
91 ctx->pipe = pipe;
92 ctx->cso = cso;
93
94 /* disabled blending/masking */
95 ctx->blend_write_color.rt[0].colormask = PIPE_MASK_RGBA;
96
97 /* rasterizer */
98 ctx->rasterizer.cull_face = PIPE_FACE_NONE;
99 ctx->rasterizer.half_pixel_center = 1;
100 ctx->rasterizer.bottom_edge_rule = 1;
101 ctx->rasterizer.depth_clip = 1;
102
103 /* samplers */
104 ctx->sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
105 ctx->sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
106 ctx->sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
107 ctx->sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
108 ctx->sampler.min_img_filter = 0; /* set later */
109 ctx->sampler.mag_img_filter = 0; /* set later */
110
111 /* vertex elements state */
112 for (i = 0; i < 2; i++) {
113 ctx->velem[i].src_offset = i * 4 * sizeof(float);
114 ctx->velem[i].instance_divisor = 0;
115 ctx->velem[i].vertex_buffer_index = cso_get_aux_vertex_buffer_slot(cso);
116 ctx->velem[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
117 }
118
119 ctx->vbuf = NULL;
120
121 /* init vertex data that doesn't change */
122 for (i = 0; i < 4; i++) {
123 ctx->vertices[i][0][3] = 1.0f; /* w */
124 ctx->vertices[i][1][3] = 1.0f; /* q */
125 }
126
127 return ctx;
128 }
129
130
131 /**
132 * Destroy a blit context
133 */
134 void
135 util_destroy_blit(struct blit_state *ctx)
136 {
137 struct pipe_context *pipe = ctx->pipe;
138 unsigned i, j, k;
139
140 if (ctx->vs)
141 pipe->delete_vs_state(pipe, ctx->vs);
142
143 for (i = 0; i < ARRAY_SIZE(ctx->fs); i++) {
144 for (j = 0; j < ARRAY_SIZE(ctx->fs[i]); j++) {
145 for (k = 0; k < ARRAY_SIZE(ctx->fs[i][j]); k++) {
146 if (ctx->fs[i][j][k])
147 pipe->delete_fs_state(pipe, ctx->fs[i][j][k]);
148 }
149 }
150 }
151
152 pipe_resource_reference(&ctx->vbuf, NULL);
153
154 FREE(ctx);
155 }
156
157
158 /**
159 * Helper function to set the fragment shaders.
160 */
161 static inline void
162 set_fragment_shader(struct blit_state *ctx, uint writemask,
163 enum pipe_format format,
164 enum pipe_texture_target pipe_tex)
165 {
166 enum tgsi_return_type stype;
167 unsigned idx;
168
169 if (util_format_is_pure_uint(format)) {
170 stype = TGSI_RETURN_TYPE_UINT;
171 idx = 0;
172 } else if (util_format_is_pure_sint(format)) {
173 stype = TGSI_RETURN_TYPE_SINT;
174 idx = 1;
175 } else {
176 stype = TGSI_RETURN_TYPE_FLOAT;
177 idx = 2;
178 }
179
180 if (!ctx->fs[pipe_tex][writemask][idx]) {
181 unsigned tgsi_tex = util_pipe_tex_to_tgsi_tex(pipe_tex, 0);
182
183 /* OpenGL does not allow blits from signed to unsigned integer
184 * or vice versa. */
185 ctx->fs[pipe_tex][writemask][idx] =
186 util_make_fragment_tex_shader_writemask(ctx->pipe, tgsi_tex,
187 TGSI_INTERPOLATE_LINEAR,
188 writemask,
189 stype, stype);
190 }
191
192 cso_set_fragment_shader_handle(ctx->cso, ctx->fs[pipe_tex][writemask][idx]);
193 }
194
195
196 /**
197 * Helper function to set the vertex shader.
198 */
199 static inline void
200 set_vertex_shader(struct blit_state *ctx)
201 {
202 /* vertex shader - still required to provide the linkage between
203 * fragment shader input semantics and vertex_element/buffers.
204 */
205 if (!ctx->vs) {
206 const uint semantic_names[] = { TGSI_SEMANTIC_POSITION,
207 TGSI_SEMANTIC_GENERIC };
208 const uint semantic_indexes[] = { 0, 0 };
209 ctx->vs = util_make_vertex_passthrough_shader(ctx->pipe, 2,
210 semantic_names,
211 semantic_indexes, FALSE);
212 }
213
214 cso_set_vertex_shader_handle(ctx->cso, ctx->vs);
215 }
216
217
218 /**
219 * Get offset of next free slot in vertex buffer for quad vertices.
220 */
221 static unsigned
222 get_next_slot( struct blit_state *ctx )
223 {
224 const unsigned max_slots = 4096 / sizeof ctx->vertices;
225
226 if (ctx->vbuf_slot >= max_slots) {
227 pipe_resource_reference(&ctx->vbuf, NULL);
228 ctx->vbuf_slot = 0;
229 }
230
231 if (!ctx->vbuf) {
232 ctx->vbuf = pipe_buffer_create(ctx->pipe->screen,
233 PIPE_BIND_VERTEX_BUFFER,
234 PIPE_USAGE_STREAM,
235 max_slots * sizeof ctx->vertices);
236 }
237
238 return ctx->vbuf_slot++ * sizeof ctx->vertices;
239 }
240
241
242
243
244 /**
245 * Setup vertex data for the textured quad we'll draw.
246 * Note: y=0=top
247 *
248 * FIXME: We should call util_map_texcoords2d_onto_cubemap
249 * for cubemaps.
250 */
251 static unsigned
252 setup_vertex_data_tex(struct blit_state *ctx,
253 unsigned src_target,
254 unsigned src_face,
255 float x0, float y0, float x1, float y1,
256 float s0, float t0, float s1, float t1,
257 float z)
258 {
259 unsigned offset;
260
261 ctx->vertices[0][0][0] = x0;
262 ctx->vertices[0][0][1] = y0;
263 ctx->vertices[0][0][2] = z;
264 ctx->vertices[0][1][0] = s0; /*s*/
265 ctx->vertices[0][1][1] = t0; /*t*/
266 ctx->vertices[0][1][2] = 0; /*r*/
267
268 ctx->vertices[1][0][0] = x1;
269 ctx->vertices[1][0][1] = y0;
270 ctx->vertices[1][0][2] = z;
271 ctx->vertices[1][1][0] = s1; /*s*/
272 ctx->vertices[1][1][1] = t0; /*t*/
273 ctx->vertices[1][1][2] = 0; /*r*/
274
275 ctx->vertices[2][0][0] = x1;
276 ctx->vertices[2][0][1] = y1;
277 ctx->vertices[2][0][2] = z;
278 ctx->vertices[2][1][0] = s1;
279 ctx->vertices[2][1][1] = t1;
280 ctx->vertices[3][1][2] = 0;
281
282 ctx->vertices[3][0][0] = x0;
283 ctx->vertices[3][0][1] = y1;
284 ctx->vertices[3][0][2] = z;
285 ctx->vertices[3][1][0] = s0;
286 ctx->vertices[3][1][1] = t1;
287 ctx->vertices[3][1][2] = 0;
288
289 if (src_target == PIPE_TEXTURE_CUBE ||
290 src_target == PIPE_TEXTURE_CUBE_ARRAY) {
291 /* Map cubemap texture coordinates inplace. */
292 const unsigned stride = sizeof ctx->vertices[0] / sizeof ctx->vertices[0][0][0];
293 util_map_texcoords2d_onto_cubemap(src_face,
294 &ctx->vertices[0][1][0], stride,
295 &ctx->vertices[0][1][0], stride,
296 TRUE);
297 }
298
299 offset = get_next_slot( ctx );
300
301 if (ctx->vbuf) {
302 pipe_buffer_write_nooverlap(ctx->pipe, ctx->vbuf,
303 offset, sizeof(ctx->vertices), ctx->vertices);
304 }
305
306 return offset;
307 }
308
309
310 /**
311 * \return TRUE if two regions overlap, FALSE otherwise
312 */
313 static boolean
314 regions_overlap(int srcX0, int srcY0,
315 int srcX1, int srcY1,
316 int dstX0, int dstY0,
317 int dstX1, int dstY1)
318 {
319 if (MAX2(srcX0, srcX1) <= MIN2(dstX0, dstX1))
320 return FALSE; /* src completely left of dst */
321
322 if (MAX2(dstX0, dstX1) <= MIN2(srcX0, srcX1))
323 return FALSE; /* dst completely left of src */
324
325 if (MAX2(srcY0, srcY1) <= MIN2(dstY0, dstY1))
326 return FALSE; /* src completely above dst */
327
328 if (MAX2(dstY0, dstY1) <= MIN2(srcY0, srcY1))
329 return FALSE; /* dst completely above src */
330
331 return TRUE; /* some overlap */
332 }
333
334
335 /**
336 * Can we blit from src format to dest format with a simple copy?
337 */
338 static boolean
339 formats_compatible(enum pipe_format src_format,
340 enum pipe_format dst_format)
341 {
342 if (src_format == dst_format) {
343 return TRUE;
344 }
345 else {
346 const struct util_format_description *src_desc =
347 util_format_description(src_format);
348 const struct util_format_description *dst_desc =
349 util_format_description(dst_format);
350 return util_is_format_compatible(src_desc, dst_desc);
351 }
352 }
353
354
355 /**
356 * Copy pixel block from src surface to dst surface.
357 * Overlapping regions are acceptable.
358 * Flipping and stretching are supported.
359 * \param filter one of PIPE_TEX_FILTER_NEAREST/LINEAR
360 * \param writemask bitmask of PIPE_MASK_[RGBAZS]. Controls which channels
361 * in the dest surface are sourced from the src surface.
362 * Disabled color channels are sourced from (0,0,0,1).
363 */
364 void
365 util_blit_pixels(struct blit_state *ctx,
366 struct pipe_resource *src_tex,
367 unsigned src_level,
368 int srcX0, int srcY0,
369 int srcX1, int srcY1,
370 int srcZ0,
371 struct pipe_surface *dst,
372 int dstX0, int dstY0,
373 int dstX1, int dstY1,
374 float z, uint filter,
375 uint writemask)
376 {
377 struct pipe_context *pipe = ctx->pipe;
378 enum pipe_format src_format, dst_format;
379 const int srcW = abs(srcX1 - srcX0);
380 const int srcH = abs(srcY1 - srcY0);
381 boolean overlap;
382 boolean is_stencil, is_depth, blit_depth, blit_stencil;
383 const struct util_format_description *src_desc =
384 util_format_description(src_tex->format);
385 struct pipe_blit_info info;
386
387 assert(filter == PIPE_TEX_FILTER_NEAREST ||
388 filter == PIPE_TEX_FILTER_LINEAR);
389
390 assert(src_level <= src_tex->last_level);
391
392 /* do the regions overlap? */
393 overlap = src_tex == dst->texture &&
394 dst->u.tex.level == src_level &&
395 dst->u.tex.first_layer == srcZ0 &&
396 regions_overlap(srcX0, srcY0, srcX1, srcY1,
397 dstX0, dstY0, dstX1, dstY1);
398
399 src_format = util_format_linear(src_tex->format);
400 dst_format = util_format_linear(dst->texture->format);
401
402 /* See whether we will blit depth or stencil. */
403 is_depth = util_format_has_depth(src_desc);
404 is_stencil = util_format_has_stencil(src_desc);
405
406 blit_depth = is_depth && (writemask & PIPE_MASK_Z);
407 blit_stencil = is_stencil && (writemask & PIPE_MASK_S);
408
409 if (is_depth || is_stencil) {
410 assert((writemask & PIPE_MASK_RGBA) == 0);
411 assert(blit_depth || blit_stencil);
412 }
413 else {
414 assert((writemask & PIPE_MASK_ZS) == 0);
415 assert(!blit_depth);
416 assert(!blit_stencil);
417 }
418
419 /*
420 * XXX: z parameter is deprecated. dst->u.tex.first_layer
421 * specificies the destination layer.
422 */
423 assert(z == 0.0f);
424
425 /*
426 * Check for simple case: no format conversion, no flipping, no stretching,
427 * no overlapping, same number of samples.
428 * Filter mode should not matter since there's no stretching.
429 */
430 if (formats_compatible(src_format, dst_format) &&
431 src_tex->nr_samples == dst->texture->nr_samples &&
432 is_stencil == blit_stencil &&
433 is_depth == blit_depth &&
434 srcX0 < srcX1 &&
435 dstX0 < dstX1 &&
436 srcY0 < srcY1 &&
437 dstY0 < dstY1 &&
438 (dstX1 - dstX0) == (srcX1 - srcX0) &&
439 (dstY1 - dstY0) == (srcY1 - srcY0) &&
440 !overlap) {
441 struct pipe_box src_box;
442 src_box.x = srcX0;
443 src_box.y = srcY0;
444 src_box.z = srcZ0;
445 src_box.width = srcW;
446 src_box.height = srcH;
447 src_box.depth = 1;
448 pipe->resource_copy_region(pipe,
449 dst->texture, dst->u.tex.level,
450 dstX0, dstY0, dst->u.tex.first_layer,/* dest */
451 src_tex, src_level,
452 &src_box);
453 return;
454 }
455
456 memset(&info, 0, sizeof info);
457 info.dst.resource = dst->texture;
458 info.dst.level = dst->u.tex.level;
459 info.dst.box.x = dstX0;
460 info.dst.box.y = dstY0;
461 info.dst.box.z = dst->u.tex.first_layer;
462 info.dst.box.width = dstX1 - dstX0;
463 info.dst.box.height = dstY1 - dstY0;
464 assert(info.dst.box.width >= 0);
465 assert(info.dst.box.height >= 0);
466 info.dst.box.depth = 1;
467 info.dst.format = dst_format;
468 info.src.resource = src_tex;
469 info.src.level = src_level;
470 info.src.box.x = srcX0;
471 info.src.box.y = srcY0;
472 info.src.box.z = srcZ0;
473 info.src.box.width = srcX1 - srcX0;
474 info.src.box.height = srcY1 - srcY0;
475 info.src.box.depth = 1;
476 info.src.format = src_format;
477 info.mask = writemask;
478 info.filter = filter;
479 info.scissor_enable = 0;
480
481 pipe->blit(pipe, &info);
482 }
483
484
485 /**
486 * Copy pixel block from src sampler view to dst surface.
487 *
488 * The sampler view's first_level field indicates the source
489 * mipmap level to use.
490 *
491 * The sampler view's first_layer indicate the layer to use, but for
492 * cube maps it must point to the first face. Face is passed in src_face.
493 *
494 * The main advantage over util_blit_pixels is that it allows to specify swizzles in
495 * pipe_sampler_view::swizzle_?.
496 *
497 * But there is no control over blitting Z and/or stencil.
498 */
499 void
500 util_blit_pixels_tex(struct blit_state *ctx,
501 struct pipe_sampler_view *src_sampler_view,
502 int srcX0, int srcY0,
503 int srcX1, int srcY1,
504 unsigned src_face,
505 struct pipe_surface *dst,
506 int dstX0, int dstY0,
507 int dstX1, int dstY1,
508 float z, uint filter)
509 {
510 boolean normalized = src_sampler_view->texture->target != PIPE_TEXTURE_RECT;
511 struct pipe_framebuffer_state fb;
512 float s0, t0, s1, t1;
513 unsigned offset;
514 struct pipe_resource *tex = src_sampler_view->texture;
515
516 assert(filter == PIPE_TEX_FILTER_NEAREST ||
517 filter == PIPE_TEX_FILTER_LINEAR);
518
519 assert(tex);
520 assert(tex->width0 != 0);
521 assert(tex->height0 != 0);
522
523 s0 = (float) srcX0;
524 s1 = (float) srcX1;
525 t0 = (float) srcY0;
526 t1 = (float) srcY1;
527
528 if(normalized)
529 {
530 /* normalize according to the mipmap level's size */
531 int level = src_sampler_view->u.tex.first_level;
532 float w = (float) u_minify(tex->width0, level);
533 float h = (float) u_minify(tex->height0, level);
534 s0 /= w;
535 s1 /= w;
536 t0 /= h;
537 t1 /= h;
538 }
539
540 assert(ctx->pipe->screen->is_format_supported(ctx->pipe->screen, dst->format,
541 PIPE_TEXTURE_2D,
542 dst->texture->nr_samples,
543 PIPE_BIND_RENDER_TARGET));
544
545 /* save state (restored below) */
546 cso_save_state(ctx->cso, (CSO_BIT_BLEND |
547 CSO_BIT_DEPTH_STENCIL_ALPHA |
548 CSO_BIT_RASTERIZER |
549 CSO_BIT_SAMPLE_MASK |
550 CSO_BIT_MIN_SAMPLES |
551 CSO_BIT_FRAGMENT_SAMPLERS |
552 CSO_BIT_FRAGMENT_SAMPLER_VIEWS |
553 CSO_BIT_STREAM_OUTPUTS |
554 CSO_BIT_VIEWPORT |
555 CSO_BIT_FRAMEBUFFER |
556 CSO_BIT_PAUSE_QUERIES |
557 CSO_BIT_FRAGMENT_SHADER |
558 CSO_BIT_VERTEX_SHADER |
559 CSO_BIT_TESSCTRL_SHADER |
560 CSO_BIT_TESSEVAL_SHADER |
561 CSO_BIT_GEOMETRY_SHADER |
562 CSO_BIT_VERTEX_ELEMENTS |
563 CSO_BIT_AUX_VERTEX_BUFFER_SLOT));
564
565 /* set misc state we care about */
566 cso_set_blend(ctx->cso, &ctx->blend_write_color);
567 cso_set_depth_stencil_alpha(ctx->cso, &ctx->dsa_keep_depthstencil);
568 cso_set_sample_mask(ctx->cso, ~0);
569 cso_set_min_samples(ctx->cso, 1);
570 cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
571 cso_set_vertex_elements(ctx->cso, 2, ctx->velem);
572 cso_set_stream_outputs(ctx->cso, 0, NULL, NULL);
573
574 /* sampler */
575 ctx->sampler.normalized_coords = normalized;
576 ctx->sampler.min_img_filter = filter;
577 ctx->sampler.mag_img_filter = filter;
578 {
579 const struct pipe_sampler_state *samplers[] = {&ctx->sampler};
580 cso_set_samplers(ctx->cso, PIPE_SHADER_FRAGMENT, 1, samplers);
581 }
582
583 /* viewport */
584 ctx->viewport.scale[0] = 0.5f * dst->width;
585 ctx->viewport.scale[1] = 0.5f * dst->height;
586 ctx->viewport.scale[2] = 0.5f;
587 ctx->viewport.translate[0] = 0.5f * dst->width;
588 ctx->viewport.translate[1] = 0.5f * dst->height;
589 ctx->viewport.translate[2] = 0.5f;
590 cso_set_viewport(ctx->cso, &ctx->viewport);
591
592 /* texture */
593 cso_set_sampler_views(ctx->cso, PIPE_SHADER_FRAGMENT, 1, &src_sampler_view);
594
595 /* shaders */
596 set_fragment_shader(ctx, TGSI_WRITEMASK_XYZW,
597 src_sampler_view->format,
598 src_sampler_view->texture->target);
599 set_vertex_shader(ctx);
600 cso_set_tessctrl_shader_handle(ctx->cso, NULL);
601 cso_set_tesseval_shader_handle(ctx->cso, NULL);
602 cso_set_geometry_shader_handle(ctx->cso, NULL);
603
604 /* drawing dest */
605 memset(&fb, 0, sizeof(fb));
606 fb.width = dst->width;
607 fb.height = dst->height;
608 fb.nr_cbufs = 1;
609 fb.cbufs[0] = dst;
610 cso_set_framebuffer(ctx->cso, &fb);
611
612 /* draw quad */
613 offset = setup_vertex_data_tex(ctx,
614 src_sampler_view->texture->target,
615 src_face,
616 (float) dstX0 / dst->width * 2.0f - 1.0f,
617 (float) dstY0 / dst->height * 2.0f - 1.0f,
618 (float) dstX1 / dst->width * 2.0f - 1.0f,
619 (float) dstY1 / dst->height * 2.0f - 1.0f,
620 s0, t0, s1, t1,
621 z);
622
623 util_draw_vertex_buffer(ctx->pipe, ctx->cso, ctx->vbuf,
624 cso_get_aux_vertex_buffer_slot(ctx->cso),
625 offset,
626 PIPE_PRIM_TRIANGLE_FAN,
627 4, /* verts */
628 2); /* attribs/vert */
629
630 /* restore state we changed */
631 cso_restore_state(ctx->cso);
632 }