Revert "gallium: make handles of set_global_binding 64 bit"
[mesa.git] / src / gallium / drivers / freedreno / freedreno_state.c
1 /*
2 * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 * SOFTWARE.
22 *
23 * Authors:
24 * Rob Clark <robclark@freedesktop.org>
25 */
26
27 #include "pipe/p_state.h"
28 #include "util/u_dual_blend.h"
29 #include "util/u_string.h"
30 #include "util/u_memory.h"
31 #include "util/u_helpers.h"
32
33 #include "freedreno_state.h"
34 #include "freedreno_context.h"
35 #include "freedreno_resource.h"
36 #include "freedreno_texture.h"
37 #include "freedreno_gmem.h"
38 #include "freedreno_query_hw.h"
39 #include "freedreno_util.h"
40
41 /* All the generic state handling.. In case of CSO's that are specific
42 * to the GPU version, when the bind and the delete are common they can
43 * go in here.
44 */
45
46 static void
47 fd_set_blend_color(struct pipe_context *pctx,
48 const struct pipe_blend_color *blend_color)
49 {
50 struct fd_context *ctx = fd_context(pctx);
51 ctx->blend_color = *blend_color;
52 ctx->dirty |= FD_DIRTY_BLEND_COLOR;
53 }
54
55 static void
56 fd_set_stencil_ref(struct pipe_context *pctx,
57 const struct pipe_stencil_ref *stencil_ref)
58 {
59 struct fd_context *ctx = fd_context(pctx);
60 ctx->stencil_ref =* stencil_ref;
61 ctx->dirty |= FD_DIRTY_STENCIL_REF;
62 }
63
64 static void
65 fd_set_clip_state(struct pipe_context *pctx,
66 const struct pipe_clip_state *clip)
67 {
68 struct fd_context *ctx = fd_context(pctx);
69 ctx->ucp = *clip;
70 ctx->dirty |= FD_DIRTY_UCP;
71 }
72
73 static void
74 fd_set_sample_mask(struct pipe_context *pctx, unsigned sample_mask)
75 {
76 struct fd_context *ctx = fd_context(pctx);
77 ctx->sample_mask = (uint16_t)sample_mask;
78 ctx->dirty |= FD_DIRTY_SAMPLE_MASK;
79 }
80
81 static void
82 fd_set_min_samples(struct pipe_context *pctx, unsigned min_samples)
83 {
84 struct fd_context *ctx = fd_context(pctx);
85 ctx->min_samples = min_samples;
86 ctx->dirty |= FD_DIRTY_MIN_SAMPLES;
87 }
88
89 /* notes from calim on #dri-devel:
90 * index==0 will be non-UBO (ie. glUniformXYZ()) all packed together padded
91 * out to vec4's
92 * I should be able to consider that I own the user_ptr until the next
93 * set_constant_buffer() call, at which point I don't really care about the
94 * previous values.
95 * index>0 will be UBO's.. well, I'll worry about that later
96 */
97 static void
98 fd_set_constant_buffer(struct pipe_context *pctx,
99 enum pipe_shader_type shader, uint index,
100 const struct pipe_constant_buffer *cb)
101 {
102 struct fd_context *ctx = fd_context(pctx);
103 struct fd_constbuf_stateobj *so = &ctx->constbuf[shader];
104
105 util_copy_constant_buffer(&so->cb[index], cb);
106
107 /* Note that the state tracker can unbind constant buffers by
108 * passing NULL here.
109 */
110 if (unlikely(!cb)) {
111 so->enabled_mask &= ~(1 << index);
112 return;
113 }
114
115 so->enabled_mask |= 1 << index;
116 ctx->dirty_shader[shader] |= FD_DIRTY_SHADER_CONST;
117 ctx->dirty |= FD_DIRTY_CONST;
118 }
119
120 static void
121 fd_set_shader_buffers(struct pipe_context *pctx,
122 enum pipe_shader_type shader,
123 unsigned start, unsigned count,
124 const struct pipe_shader_buffer *buffers,
125 unsigned writable_bitmask)
126 {
127 struct fd_context *ctx = fd_context(pctx);
128 struct fd_shaderbuf_stateobj *so = &ctx->shaderbuf[shader];
129 unsigned mask = 0;
130
131 if (buffers) {
132 for (unsigned i = 0; i < count; i++) {
133 unsigned n = i + start;
134 struct pipe_shader_buffer *buf = &so->sb[n];
135
136 if ((buf->buffer == buffers[i].buffer) &&
137 (buf->buffer_offset == buffers[i].buffer_offset) &&
138 (buf->buffer_size == buffers[i].buffer_size))
139 continue;
140
141 mask |= BIT(n);
142
143 buf->buffer_offset = buffers[i].buffer_offset;
144 buf->buffer_size = buffers[i].buffer_size;
145 pipe_resource_reference(&buf->buffer, buffers[i].buffer);
146
147 if (buf->buffer)
148 so->enabled_mask |= BIT(n);
149 else
150 so->enabled_mask &= ~BIT(n);
151 }
152 } else {
153 mask = (BIT(count) - 1) << start;
154
155 for (unsigned i = 0; i < count; i++) {
156 unsigned n = i + start;
157 struct pipe_shader_buffer *buf = &so->sb[n];
158
159 pipe_resource_reference(&buf->buffer, NULL);
160 }
161
162 so->enabled_mask &= ~mask;
163 }
164
165 ctx->dirty_shader[shader] |= FD_DIRTY_SHADER_SSBO;
166 }
167
168 void
169 fd_set_shader_images(struct pipe_context *pctx,
170 enum pipe_shader_type shader,
171 unsigned start, unsigned count,
172 const struct pipe_image_view *images)
173 {
174 struct fd_context *ctx = fd_context(pctx);
175 struct fd_shaderimg_stateobj *so = &ctx->shaderimg[shader];
176
177 unsigned mask = 0;
178
179 if (images) {
180 for (unsigned i = 0; i < count; i++) {
181 unsigned n = i + start;
182 struct pipe_image_view *buf = &so->si[n];
183
184 if ((buf->resource == images[i].resource) &&
185 (buf->format == images[i].format) &&
186 (buf->access == images[i].access) &&
187 !memcmp(&buf->u, &images[i].u, sizeof(buf->u)))
188 continue;
189
190 mask |= BIT(n);
191 util_copy_image_view(buf, &images[i]);
192
193 if (buf->resource)
194 so->enabled_mask |= BIT(n);
195 else
196 so->enabled_mask &= ~BIT(n);
197 }
198 } else {
199 mask = (BIT(count) - 1) << start;
200
201 for (unsigned i = 0; i < count; i++) {
202 unsigned n = i + start;
203 struct pipe_image_view *img = &so->si[n];
204
205 pipe_resource_reference(&img->resource, NULL);
206 }
207
208 so->enabled_mask &= ~mask;
209 }
210
211 ctx->dirty_shader[shader] |= FD_DIRTY_SHADER_IMAGE;
212 }
213
214 static void
215 fd_set_framebuffer_state(struct pipe_context *pctx,
216 const struct pipe_framebuffer_state *framebuffer)
217 {
218 struct fd_context *ctx = fd_context(pctx);
219 struct pipe_framebuffer_state *cso;
220
221 DBG("%ux%u, %u layers, %u samples",
222 framebuffer->width, framebuffer->height,
223 framebuffer->layers, framebuffer->samples);
224
225 cso = &ctx->framebuffer;
226
227 if (util_framebuffer_state_equal(cso, framebuffer))
228 return;
229
230 util_copy_framebuffer_state(cso, framebuffer);
231
232 cso->samples = util_framebuffer_get_num_samples(cso);
233
234 if (ctx->screen->reorder) {
235 struct fd_batch *old_batch = NULL;
236
237 fd_batch_reference(&old_batch, ctx->batch);
238
239 if (likely(old_batch))
240 fd_batch_set_stage(old_batch, FD_STAGE_NULL);
241
242 fd_batch_reference(&ctx->batch, NULL);
243 fd_context_all_dirty(ctx);
244
245 if (old_batch && old_batch->blit && !old_batch->back_blit) {
246 /* for blits, there is not really much point in hanging on
247 * to the uncommitted batch (ie. you probably don't blit
248 * multiple times to the same surface), so we might as
249 * well go ahead and flush this one:
250 */
251 fd_batch_flush(old_batch);
252 }
253
254 fd_batch_reference(&old_batch, NULL);
255 } else if (ctx->batch) {
256 DBG("%d: cbufs[0]=%p, zsbuf=%p", ctx->batch->needs_flush,
257 framebuffer->cbufs[0], framebuffer->zsbuf);
258 fd_batch_flush(ctx->batch);
259 }
260
261 ctx->dirty |= FD_DIRTY_FRAMEBUFFER;
262
263 ctx->disabled_scissor.minx = 0;
264 ctx->disabled_scissor.miny = 0;
265 ctx->disabled_scissor.maxx = cso->width;
266 ctx->disabled_scissor.maxy = cso->height;
267
268 ctx->dirty |= FD_DIRTY_SCISSOR;
269 }
270
271 static void
272 fd_set_polygon_stipple(struct pipe_context *pctx,
273 const struct pipe_poly_stipple *stipple)
274 {
275 struct fd_context *ctx = fd_context(pctx);
276 ctx->stipple = *stipple;
277 ctx->dirty |= FD_DIRTY_STIPPLE;
278 }
279
280 static void
281 fd_set_scissor_states(struct pipe_context *pctx,
282 unsigned start_slot,
283 unsigned num_scissors,
284 const struct pipe_scissor_state *scissor)
285 {
286 struct fd_context *ctx = fd_context(pctx);
287
288 ctx->scissor = *scissor;
289 ctx->dirty |= FD_DIRTY_SCISSOR;
290 }
291
292 static void
293 fd_set_viewport_states(struct pipe_context *pctx,
294 unsigned start_slot,
295 unsigned num_viewports,
296 const struct pipe_viewport_state *viewport)
297 {
298 struct fd_context *ctx = fd_context(pctx);
299 struct pipe_scissor_state *scissor = &ctx->viewport_scissor;
300 float minx, miny, maxx, maxy;
301
302 ctx->viewport = *viewport;
303
304 /* see si_get_scissor_from_viewport(): */
305
306 /* Convert (-1, -1) and (1, 1) from clip space into window space. */
307 minx = -viewport->scale[0] + viewport->translate[0];
308 miny = -viewport->scale[1] + viewport->translate[1];
309 maxx = viewport->scale[0] + viewport->translate[0];
310 maxy = viewport->scale[1] + viewport->translate[1];
311
312 /* Handle inverted viewports. */
313 if (minx > maxx) {
314 swap(minx, maxx);
315 }
316 if (miny > maxy) {
317 swap(miny, maxy);
318 }
319
320 debug_assert(miny >= 0);
321 debug_assert(maxy >= 0);
322
323 /* Convert to integer and round up the max bounds. */
324 scissor->minx = minx;
325 scissor->miny = miny;
326 scissor->maxx = ceilf(maxx);
327 scissor->maxy = ceilf(maxy);
328
329 ctx->dirty |= FD_DIRTY_VIEWPORT;
330 }
331
332 static void
333 fd_set_vertex_buffers(struct pipe_context *pctx,
334 unsigned start_slot, unsigned count,
335 const struct pipe_vertex_buffer *vb)
336 {
337 struct fd_context *ctx = fd_context(pctx);
338 struct fd_vertexbuf_stateobj *so = &ctx->vtx.vertexbuf;
339 int i;
340
341 /* on a2xx, pitch is encoded in the vtx fetch instruction, so
342 * we need to mark VTXSTATE as dirty as well to trigger patching
343 * and re-emitting the vtx shader:
344 */
345 if (ctx->screen->gpu_id < 300) {
346 for (i = 0; i < count; i++) {
347 bool new_enabled = vb && vb[i].buffer.resource;
348 bool old_enabled = so->vb[i].buffer.resource != NULL;
349 uint32_t new_stride = vb ? vb[i].stride : 0;
350 uint32_t old_stride = so->vb[i].stride;
351 if ((new_enabled != old_enabled) || (new_stride != old_stride)) {
352 ctx->dirty |= FD_DIRTY_VTXSTATE;
353 break;
354 }
355 }
356 }
357
358 util_set_vertex_buffers_mask(so->vb, &so->enabled_mask, vb, start_slot, count);
359 so->count = util_last_bit(so->enabled_mask);
360
361 ctx->dirty |= FD_DIRTY_VTXBUF;
362 }
363
364 static void
365 fd_blend_state_bind(struct pipe_context *pctx, void *hwcso)
366 {
367 struct fd_context *ctx = fd_context(pctx);
368 struct pipe_blend_state *cso = hwcso;
369 bool old_is_dual = ctx->blend ?
370 ctx->blend->rt[0].blend_enable && util_blend_state_is_dual(ctx->blend, 0) :
371 false;
372 bool new_is_dual = cso ?
373 cso->rt[0].blend_enable && util_blend_state_is_dual(cso, 0) :
374 false;
375 ctx->blend = hwcso;
376 ctx->dirty |= FD_DIRTY_BLEND;
377 if (old_is_dual != new_is_dual)
378 ctx->dirty |= FD_DIRTY_BLEND_DUAL;
379 }
380
381 static void
382 fd_blend_state_delete(struct pipe_context *pctx, void *hwcso)
383 {
384 FREE(hwcso);
385 }
386
387 static void
388 fd_rasterizer_state_bind(struct pipe_context *pctx, void *hwcso)
389 {
390 struct fd_context *ctx = fd_context(pctx);
391 struct pipe_scissor_state *old_scissor = fd_context_get_scissor(ctx);
392
393 ctx->rasterizer = hwcso;
394 ctx->dirty |= FD_DIRTY_RASTERIZER;
395
396 /* if scissor enable bit changed we need to mark scissor
397 * state as dirty as well:
398 * NOTE: we can do a shallow compare, since we only care
399 * if it changed to/from &ctx->disable_scissor
400 */
401 if (old_scissor != fd_context_get_scissor(ctx))
402 ctx->dirty |= FD_DIRTY_SCISSOR;
403 }
404
405 static void
406 fd_rasterizer_state_delete(struct pipe_context *pctx, void *hwcso)
407 {
408 FREE(hwcso);
409 }
410
411 static void
412 fd_zsa_state_bind(struct pipe_context *pctx, void *hwcso)
413 {
414 struct fd_context *ctx = fd_context(pctx);
415 ctx->zsa = hwcso;
416 ctx->dirty |= FD_DIRTY_ZSA;
417 }
418
419 static void
420 fd_zsa_state_delete(struct pipe_context *pctx, void *hwcso)
421 {
422 FREE(hwcso);
423 }
424
425 static void *
426 fd_vertex_state_create(struct pipe_context *pctx, unsigned num_elements,
427 const struct pipe_vertex_element *elements)
428 {
429 struct fd_vertex_stateobj *so = CALLOC_STRUCT(fd_vertex_stateobj);
430
431 if (!so)
432 return NULL;
433
434 memcpy(so->pipe, elements, sizeof(*elements) * num_elements);
435 so->num_elements = num_elements;
436
437 return so;
438 }
439
440 static void
441 fd_vertex_state_delete(struct pipe_context *pctx, void *hwcso)
442 {
443 FREE(hwcso);
444 }
445
446 static void
447 fd_vertex_state_bind(struct pipe_context *pctx, void *hwcso)
448 {
449 struct fd_context *ctx = fd_context(pctx);
450 ctx->vtx.vtx = hwcso;
451 ctx->dirty |= FD_DIRTY_VTXSTATE;
452 }
453
454 static struct pipe_stream_output_target *
455 fd_create_stream_output_target(struct pipe_context *pctx,
456 struct pipe_resource *prsc, unsigned buffer_offset,
457 unsigned buffer_size)
458 {
459 struct pipe_stream_output_target *target;
460 struct fd_resource *rsc = fd_resource(prsc);
461
462 target = CALLOC_STRUCT(pipe_stream_output_target);
463 if (!target)
464 return NULL;
465
466 pipe_reference_init(&target->reference, 1);
467 pipe_resource_reference(&target->buffer, prsc);
468
469 target->context = pctx;
470 target->buffer_offset = buffer_offset;
471 target->buffer_size = buffer_size;
472
473 assert(rsc->base.target == PIPE_BUFFER);
474 util_range_add(&rsc->base, &rsc->valid_buffer_range,
475 buffer_offset, buffer_offset + buffer_size);
476
477 return target;
478 }
479
480 static void
481 fd_stream_output_target_destroy(struct pipe_context *pctx,
482 struct pipe_stream_output_target *target)
483 {
484 pipe_resource_reference(&target->buffer, NULL);
485 FREE(target);
486 }
487
488 static void
489 fd_set_stream_output_targets(struct pipe_context *pctx,
490 unsigned num_targets, struct pipe_stream_output_target **targets,
491 const unsigned *offsets)
492 {
493 struct fd_context *ctx = fd_context(pctx);
494 struct fd_streamout_stateobj *so = &ctx->streamout;
495 unsigned i;
496
497 debug_assert(num_targets <= ARRAY_SIZE(so->targets));
498
499 for (i = 0; i < num_targets; i++) {
500 boolean changed = targets[i] != so->targets[i];
501 boolean reset = (offsets[i] != (unsigned)-1);
502
503 so->reset |= (reset << i);
504
505 if (!changed && !reset)
506 continue;
507
508 if (reset)
509 so->offsets[i] = offsets[i];
510
511 pipe_so_target_reference(&so->targets[i], targets[i]);
512 }
513
514 for (; i < so->num_targets; i++) {
515 pipe_so_target_reference(&so->targets[i], NULL);
516 }
517
518 so->num_targets = num_targets;
519
520 ctx->dirty |= FD_DIRTY_STREAMOUT;
521 }
522
523 static void
524 fd_bind_compute_state(struct pipe_context *pctx, void *state)
525 {
526 struct fd_context *ctx = fd_context(pctx);
527 ctx->compute = state;
528 ctx->dirty_shader[PIPE_SHADER_COMPUTE] |= FD_DIRTY_SHADER_PROG;
529 }
530
531 static void
532 fd_set_compute_resources(struct pipe_context *pctx,
533 unsigned start, unsigned count, struct pipe_surface **prscs)
534 {
535 // TODO
536 }
537
538 /* used by clover to bind global objects, returning the bo address
539 * via handles[n]
540 */
541 static void
542 fd_set_global_binding(struct pipe_context *pctx,
543 unsigned first, unsigned count, struct pipe_resource **prscs,
544 uint32_t **handles)
545 {
546 struct fd_context *ctx = fd_context(pctx);
547 struct fd_global_bindings_stateobj *so = &ctx->global_bindings;
548 unsigned mask = 0;
549
550 if (prscs) {
551 for (unsigned i = 0; i < count; i++) {
552 unsigned n = i + first;
553
554 mask |= BIT(n);
555
556 pipe_resource_reference(&so->buf[n], prscs[i]);
557
558 if (so->buf[n]) {
559 struct fd_resource *rsc = fd_resource(so->buf[n]);
560 uint64_t iova = fd_bo_get_iova(rsc->bo);
561 // TODO need to scream if iova > 32b or fix gallium API..
562 *handles[i] += iova;
563 }
564
565 if (prscs[i])
566 so->enabled_mask |= BIT(n);
567 else
568 so->enabled_mask &= ~BIT(n);
569 }
570 } else {
571 mask = (BIT(count) - 1) << first;
572
573 for (unsigned i = 0; i < count; i++) {
574 unsigned n = i + first;
575 if (so->buf[n]) {
576 struct fd_resource *rsc = fd_resource(so->buf[n]);
577 fd_bo_put_iova(rsc->bo);
578 }
579 pipe_resource_reference(&so->buf[n], NULL);
580 }
581
582 so->enabled_mask &= ~mask;
583 }
584
585 }
586
587 void
588 fd_state_init(struct pipe_context *pctx)
589 {
590 pctx->set_blend_color = fd_set_blend_color;
591 pctx->set_stencil_ref = fd_set_stencil_ref;
592 pctx->set_clip_state = fd_set_clip_state;
593 pctx->set_sample_mask = fd_set_sample_mask;
594 pctx->set_min_samples = fd_set_min_samples;
595 pctx->set_constant_buffer = fd_set_constant_buffer;
596 pctx->set_shader_buffers = fd_set_shader_buffers;
597 pctx->set_shader_images = fd_set_shader_images;
598 pctx->set_framebuffer_state = fd_set_framebuffer_state;
599 pctx->set_polygon_stipple = fd_set_polygon_stipple;
600 pctx->set_scissor_states = fd_set_scissor_states;
601 pctx->set_viewport_states = fd_set_viewport_states;
602
603 pctx->set_vertex_buffers = fd_set_vertex_buffers;
604
605 pctx->bind_blend_state = fd_blend_state_bind;
606 pctx->delete_blend_state = fd_blend_state_delete;
607
608 pctx->bind_rasterizer_state = fd_rasterizer_state_bind;
609 pctx->delete_rasterizer_state = fd_rasterizer_state_delete;
610
611 pctx->bind_depth_stencil_alpha_state = fd_zsa_state_bind;
612 pctx->delete_depth_stencil_alpha_state = fd_zsa_state_delete;
613
614 pctx->create_vertex_elements_state = fd_vertex_state_create;
615 pctx->delete_vertex_elements_state = fd_vertex_state_delete;
616 pctx->bind_vertex_elements_state = fd_vertex_state_bind;
617
618 pctx->create_stream_output_target = fd_create_stream_output_target;
619 pctx->stream_output_target_destroy = fd_stream_output_target_destroy;
620 pctx->set_stream_output_targets = fd_set_stream_output_targets;
621
622 if (has_compute(fd_screen(pctx->screen))) {
623 pctx->bind_compute_state = fd_bind_compute_state;
624 pctx->set_compute_resources = fd_set_compute_resources;
625 pctx->set_global_binding = fd_set_global_binding;
626 }
627 }