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 /* notes from calim on #dri-devel:
  82  * index==0 will be non-UBO (ie. glUniformXYZ()) all packed together padded
  83  * out to vec4's
  84  * I should be able to consider that I own the user_ptr until the next
  85  * set_constant_buffer() call, at which point I don't really care about the
  86  * previous values.
  87  * index>0 will be UBO's.. well, I'll worry about that later
  88  */
  89 static void
  90 fd_set_constant_buffer(struct pipe_context *pctx,
  91                 enum pipe_shader_type shader, uint index,
  92                 const struct pipe_constant_buffer *cb)
  93 {
  94         struct fd_context *ctx = fd_context(pctx);
  95         struct fd_constbuf_stateobj *so = &ctx->constbuf[shader];
  96
  97         util_copy_constant_buffer(&so->cb[index], cb);
  98
  99         /* Note that the state tracker can unbind constant buffers by
 100          * passing NULL here.
 101          */
 102         if (unlikely(!cb)) {
 103                 so->enabled_mask &= ~(1 << index);
 104                 return;
 105         }
 106
 107         so->enabled_mask |= 1 << index;
 108         ctx->dirty_shader[shader] |= FD_DIRTY_SHADER_CONST;
 109         ctx->dirty |= FD_DIRTY_CONST;
 110 }
 111
 112 static void
 113 fd_set_shader_buffers(struct pipe_context *pctx,
 114                 enum pipe_shader_type shader,
 115                 unsigned start, unsigned count,
 116                 const struct pipe_shader_buffer *buffers)
 117 {
 118         struct fd_context *ctx = fd_context(pctx);
 119         struct fd_shaderbuf_stateobj *so = &ctx->shaderbuf[shader];
 120         unsigned mask = 0;
 121
 122         if (buffers) {
 123                 for (unsigned i = 0; i < count; i++) {
 124                         unsigned n = i + start;
 125                         struct pipe_shader_buffer *buf = &so->sb[n];
 126
 127                         if ((buf->buffer == buffers[i].buffer) &&
 128                                         (buf->buffer_offset == buffers[i].buffer_offset) &&
 129                                         (buf->buffer_size == buffers[i].buffer_size))
 130                                 continue;
 131
 132                         mask |= BIT(n);
 133
 134                         buf->buffer_offset = buffers[i].buffer_offset;
 135                         buf->buffer_size = buffers[i].buffer_size;
 136                         pipe_resource_reference(&buf->buffer, buffers[i].buffer);
 137
 138                         if (buf->buffer)
 139                                 so->enabled_mask |= BIT(n);
 140                         else
 141                                 so->enabled_mask &= ~BIT(n);
 142                 }
 143         } else {
 144                 mask = (BIT(count) - 1) << start;
 145
 146                 for (unsigned i = 0; i < count; i++) {
 147                         unsigned n = i + start;
 148                         struct pipe_shader_buffer *buf = &so->sb[n];
 149
 150                         pipe_resource_reference(&buf->buffer, NULL);
 151                 }
 152
 153                 so->enabled_mask &= ~mask;
 154         }
 155
 156         ctx->dirty_shader[shader] |= FD_DIRTY_SHADER_SSBO;
 157 }
 158
 159 static void
 160 fd_set_shader_images(struct pipe_context *pctx,
 161                 enum pipe_shader_type shader,
 162                 unsigned start, unsigned count,
 163                 const struct pipe_image_view *images)
 164 {
 165         struct fd_context *ctx = fd_context(pctx);
 166         struct fd_shaderimg_stateobj *so = &ctx->shaderimg[shader];
 167
 168         unsigned mask = 0;
 169
 170         if (images) {
 171                 for (unsigned i = 0; i < count; i++) {
 172                         unsigned n = i + start;
 173                         struct pipe_image_view *buf = &so->si[n];
 174
 175                         if ((buf->resource == images[i].resource) &&
 176                                         (buf->format == images[i].format) &&
 177                                         (buf->access == images[i].access) &&
 178                                         !memcmp(&buf->u, &images[i].u, sizeof(buf->u)))
 179                                 continue;
 180
 181                         mask |= BIT(n);
 182                         util_copy_image_view(buf, &images[i]);
 183
 184                         if (buf->resource)
 185                                 so->enabled_mask |= BIT(n);
 186                         else
 187                                 so->enabled_mask &= ~BIT(n);
 188                 }
 189         } else {
 190                 mask = (BIT(count) - 1) << start;
 191
 192                 for (unsigned i = 0; i < count; i++) {
 193                         unsigned n = i + start;
 194                         struct pipe_image_view *img = &so->si[n];
 195
 196                         pipe_resource_reference(&img->resource, NULL);
 197                 }
 198
 199                 so->enabled_mask &= ~mask;
 200         }
 201
 202         ctx->dirty_shader[shader] |= FD_DIRTY_SHADER_IMAGE;
 203 }
 204
 205 static void
 206 fd_set_framebuffer_state(struct pipe_context *pctx,
 207                 const struct pipe_framebuffer_state *framebuffer)
 208 {
 209         struct fd_context *ctx = fd_context(pctx);
 210         struct pipe_framebuffer_state *cso;
 211
 212         DBG("%ux%u, %u layers, %u samples",
 213                 framebuffer->width, framebuffer->height,
 214                 framebuffer->layers, framebuffer->samples);
 215
 216         cso = &ctx->framebuffer;
 217
 218         if (util_framebuffer_state_equal(cso, framebuffer))
 219                 return;
 220
 221         util_copy_framebuffer_state(cso, framebuffer);
 222
 223         cso->samples = util_framebuffer_get_num_samples(cso);
 224
 225         if (ctx->screen->reorder) {
 226                 struct fd_batch *old_batch = NULL;
 227
 228                 fd_batch_reference(&old_batch, ctx->batch);
 229
 230                 if (likely(old_batch))
 231                         fd_batch_set_stage(old_batch, FD_STAGE_NULL);
 232
 233                 fd_batch_reference(&ctx->batch, NULL);
 234                 fd_context_all_dirty(ctx);
 235
 236                 if (old_batch && old_batch->blit && !old_batch->back_blit) {
 237                         /* for blits, there is not really much point in hanging on
 238                          * to the uncommitted batch (ie. you probably don't blit
 239                          * multiple times to the same surface), so we might as
 240                          * well go ahead and flush this one:
 241                          */
 242                         fd_batch_flush(old_batch, false, false);
 243                 }
 244
 245                 fd_batch_reference(&old_batch, NULL);
 246         } else {
 247                 DBG("%d: cbufs[0]=%p, zsbuf=%p", ctx->batch->needs_flush,
 248                                 framebuffer->cbufs[0], framebuffer->zsbuf);
 249                 fd_batch_flush(ctx->batch, false, false);
 250                 util_copy_framebuffer_state(&ctx->batch->framebuffer, cso);
 251         }
 252
 253         ctx->dirty |= FD_DIRTY_FRAMEBUFFER;
 254
 255         ctx->disabled_scissor.minx = 0;
 256         ctx->disabled_scissor.miny = 0;
 257         ctx->disabled_scissor.maxx = cso->width;
 258         ctx->disabled_scissor.maxy = cso->height;
 259
 260         ctx->dirty |= FD_DIRTY_SCISSOR;
 261 }
 262
 263 static void
 264 fd_set_polygon_stipple(struct pipe_context *pctx,
 265                 const struct pipe_poly_stipple *stipple)
 266 {
 267         struct fd_context *ctx = fd_context(pctx);
 268         ctx->stipple = *stipple;
 269         ctx->dirty |= FD_DIRTY_STIPPLE;
 270 }
 271
 272 static void
 273 fd_set_scissor_states(struct pipe_context *pctx,
 274                 unsigned start_slot,
 275                 unsigned num_scissors,
 276                 const struct pipe_scissor_state *scissor)
 277 {
 278         struct fd_context *ctx = fd_context(pctx);
 279
 280         ctx->scissor = *scissor;
 281         ctx->dirty |= FD_DIRTY_SCISSOR;
 282 }
 283
 284 static void
 285 fd_set_viewport_states(struct pipe_context *pctx,
 286                 unsigned start_slot,
 287                 unsigned num_viewports,
 288                 const struct pipe_viewport_state *viewport)
 289 {
 290         struct fd_context *ctx = fd_context(pctx);
 291         struct pipe_scissor_state *scissor = &ctx->viewport_scissor;
 292         float minx, miny, maxx, maxy;
 293
 294         ctx->viewport = *viewport;
 295
 296         /* see si_get_scissor_from_viewport(): */
 297
 298         /* Convert (-1, -1) and (1, 1) from clip space into window space. */
 299         minx = -viewport->scale[0] + viewport->translate[0];
 300         miny = -viewport->scale[1] + viewport->translate[1];
 301         maxx = viewport->scale[0] + viewport->translate[0];
 302         maxy = viewport->scale[1] + viewport->translate[1];
 303
 304         /* Handle inverted viewports. */
 305         if (minx > maxx) {
 306                 swap(minx, maxx);
 307         }
 308         if (miny > maxy) {
 309                 swap(miny, maxy);
 310         }
 311
 312         debug_assert(miny >= 0);
 313         debug_assert(maxy >= 0);
 314
 315         /* Convert to integer and round up the max bounds. */
 316         scissor->minx = minx;
 317         scissor->miny = miny;
 318         scissor->maxx = ceilf(maxx);
 319         scissor->maxy = ceilf(maxy);
 320
 321         ctx->dirty |= FD_DIRTY_VIEWPORT;
 322 }
 323
 324 static void
 325 fd_set_vertex_buffers(struct pipe_context *pctx,
 326                 unsigned start_slot, unsigned count,
 327                 const struct pipe_vertex_buffer *vb)
 328 {
 329         struct fd_context *ctx = fd_context(pctx);
 330         struct fd_vertexbuf_stateobj *so = &ctx->vtx.vertexbuf;
 331         int i;
 332
 333         /* on a2xx, pitch is encoded in the vtx fetch instruction, so
 334          * we need to mark VTXSTATE as dirty as well to trigger patching
 335          * and re-emitting the vtx shader:
 336          */
 337         if (ctx->screen->gpu_id < 300) {
 338                 for (i = 0; i < count; i++) {
 339                         bool new_enabled = vb && vb[i].buffer.resource;
 340                         bool old_enabled = so->vb[i].buffer.resource != NULL;
 341                         uint32_t new_stride = vb ? vb[i].stride : 0;
 342                         uint32_t old_stride = so->vb[i].stride;
 343                         if ((new_enabled != old_enabled) || (new_stride != old_stride)) {
 344                                 ctx->dirty |= FD_DIRTY_VTXSTATE;
 345                                 break;
 346                         }
 347                 }
 348         }
 349
 350         util_set_vertex_buffers_mask(so->vb, &so->enabled_mask, vb, start_slot, count);
 351         so->count = util_last_bit(so->enabled_mask);
 352
 353         ctx->dirty |= FD_DIRTY_VTXBUF;
 354 }
 355
 356 static void
 357 fd_blend_state_bind(struct pipe_context *pctx, void *hwcso)
 358 {
 359         struct fd_context *ctx = fd_context(pctx);
 360         struct pipe_blend_state *cso = hwcso;
 361         bool old_is_dual = ctx->blend ?
 362                 ctx->blend->rt[0].blend_enable && util_blend_state_is_dual(ctx->blend, 0) :
 363                 false;
 364         bool new_is_dual = cso ?
 365                 cso->rt[0].blend_enable && util_blend_state_is_dual(cso, 0) :
 366                 false;
 367         ctx->blend = hwcso;
 368         ctx->dirty |= FD_DIRTY_BLEND;
 369         if (old_is_dual != new_is_dual)
 370                 ctx->dirty |= FD_DIRTY_BLEND_DUAL;
 371 }
 372
 373 static void
 374 fd_blend_state_delete(struct pipe_context *pctx, void *hwcso)
 375 {
 376         FREE(hwcso);
 377 }
 378
 379 static void
 380 fd_rasterizer_state_bind(struct pipe_context *pctx, void *hwcso)
 381 {
 382         struct fd_context *ctx = fd_context(pctx);
 383         struct pipe_scissor_state *old_scissor = fd_context_get_scissor(ctx);
 384
 385         ctx->rasterizer = hwcso;
 386         ctx->dirty |= FD_DIRTY_RASTERIZER;
 387
 388         /* if scissor enable bit changed we need to mark scissor
 389          * state as dirty as well:
 390          * NOTE: we can do a shallow compare, since we only care
 391          * if it changed to/from &ctx->disable_scissor
 392          */
 393         if (old_scissor != fd_context_get_scissor(ctx))
 394                 ctx->dirty |= FD_DIRTY_SCISSOR;
 395 }
 396
 397 static void
 398 fd_rasterizer_state_delete(struct pipe_context *pctx, void *hwcso)
 399 {
 400         FREE(hwcso);
 401 }
 402
 403 static void
 404 fd_zsa_state_bind(struct pipe_context *pctx, void *hwcso)
 405 {
 406         struct fd_context *ctx = fd_context(pctx);
 407         ctx->zsa = hwcso;
 408         ctx->dirty |= FD_DIRTY_ZSA;
 409 }
 410
 411 static void
 412 fd_zsa_state_delete(struct pipe_context *pctx, void *hwcso)
 413 {
 414         FREE(hwcso);
 415 }
 416
 417 static void *
 418 fd_vertex_state_create(struct pipe_context *pctx, unsigned num_elements,
 419                 const struct pipe_vertex_element *elements)
 420 {
 421         struct fd_vertex_stateobj *so = CALLOC_STRUCT(fd_vertex_stateobj);
 422
 423         if (!so)
 424                 return NULL;
 425
 426         memcpy(so->pipe, elements, sizeof(*elements) * num_elements);
 427         so->num_elements = num_elements;
 428
 429         return so;
 430 }
 431
 432 static void
 433 fd_vertex_state_delete(struct pipe_context *pctx, void *hwcso)
 434 {
 435         FREE(hwcso);
 436 }
 437
 438 static void
 439 fd_vertex_state_bind(struct pipe_context *pctx, void *hwcso)
 440 {
 441         struct fd_context *ctx = fd_context(pctx);
 442         ctx->vtx.vtx = hwcso;
 443         ctx->dirty |= FD_DIRTY_VTXSTATE;
 444 }
 445
 446 static struct pipe_stream_output_target *
 447 fd_create_stream_output_target(struct pipe_context *pctx,
 448                 struct pipe_resource *prsc, unsigned buffer_offset,
 449                 unsigned buffer_size)
 450 {
 451         struct pipe_stream_output_target *target;
 452         struct fd_resource *rsc = fd_resource(prsc);
 453
 454         target = CALLOC_STRUCT(pipe_stream_output_target);
 455         if (!target)
 456                 return NULL;
 457
 458         pipe_reference_init(&target->reference, 1);
 459         pipe_resource_reference(&target->buffer, prsc);
 460
 461         target->context = pctx;
 462         target->buffer_offset = buffer_offset;
 463         target->buffer_size = buffer_size;
 464
 465         assert(rsc->base.target == PIPE_BUFFER);
 466         util_range_add(&rsc->valid_buffer_range,
 467                 buffer_offset, buffer_offset + buffer_size);
 468
 469         return target;
 470 }
 471
 472 static void
 473 fd_stream_output_target_destroy(struct pipe_context *pctx,
 474                 struct pipe_stream_output_target *target)
 475 {
 476         pipe_resource_reference(&target->buffer, NULL);
 477         FREE(target);
 478 }
 479
 480 static void
 481 fd_set_stream_output_targets(struct pipe_context *pctx,
 482                 unsigned num_targets, struct pipe_stream_output_target **targets,
 483                 const unsigned *offsets)
 484 {
 485         struct fd_context *ctx = fd_context(pctx);
 486         struct fd_streamout_stateobj *so = &ctx->streamout;
 487         unsigned i;
 488
 489         debug_assert(num_targets <= ARRAY_SIZE(so->targets));
 490
 491         for (i = 0; i < num_targets; i++) {
 492                 boolean changed = targets[i] != so->targets[i];
 493                 boolean append = (offsets[i] == (unsigned)-1);
 494
 495                 if (!changed && append)
 496                         continue;
 497
 498                 if (!append)
 499                         so->offsets[i] = offsets[i];
 500
 501                 pipe_so_target_reference(&so->targets[i], targets[i]);
 502         }
 503
 504         for (; i < so->num_targets; i++) {
 505                 pipe_so_target_reference(&so->targets[i], NULL);
 506         }
 507
 508         so->num_targets = num_targets;
 509
 510         ctx->dirty |= FD_DIRTY_STREAMOUT;
 511 }
 512
 513 static void
 514 fd_bind_compute_state(struct pipe_context *pctx, void *state)
 515 {
 516         struct fd_context *ctx = fd_context(pctx);
 517         ctx->compute = state;
 518         ctx->dirty_shader[PIPE_SHADER_COMPUTE] |= FD_DIRTY_SHADER_PROG;
 519 }
 520
 521 static void
 522 fd_set_compute_resources(struct pipe_context *pctx,
 523                 unsigned start, unsigned count, struct pipe_surface **prscs)
 524 {
 525         // TODO
 526 }
 527
 528 /* used by clover to bind global objects, returning the bo address
 529  * via handles[n]
 530  */
 531 static void
 532 fd_set_global_binding(struct pipe_context *pctx,
 533                 unsigned first, unsigned count, struct pipe_resource **prscs,
 534                 uint32_t **handles)
 535 {
 536         struct fd_context *ctx = fd_context(pctx);
 537         struct fd_global_bindings_stateobj *so = &ctx->global_bindings;
 538         unsigned mask = 0;
 539
 540         if (prscs) {
 541                 for (unsigned i = 0; i < count; i++) {
 542                         unsigned n = i + first;
 543
 544                         mask |= BIT(n);
 545
 546                         pipe_resource_reference(&so->buf[n], prscs[i]);
 547
 548                         if (so->buf[n]) {
 549                                 struct fd_resource *rsc = fd_resource(so->buf[n]);
 550                                 uint64_t iova = fd_bo_get_iova(rsc->bo);
 551                                 // TODO need to scream if iova > 32b or fix gallium API..
 552                                 *handles[i] += iova;
 553                         }
 554
 555                         if (prscs[i])
 556                                 so->enabled_mask |= BIT(n);
 557                         else
 558                                 so->enabled_mask &= ~BIT(n);
 559                 }
 560         } else {
 561                 mask = (BIT(count) - 1) << first;
 562
 563                 for (unsigned i = 0; i < count; i++) {
 564                         unsigned n = i + first;
 565                         if (so->buf[n]) {
 566                                 struct fd_resource *rsc = fd_resource(so->buf[n]);
 567                                 fd_bo_put_iova(rsc->bo);
 568                         }
 569                         pipe_resource_reference(&so->buf[n], NULL);
 570                 }
 571
 572                 so->enabled_mask &= ~mask;
 573         }
 574
 575 }
 576
 577 void
 578 fd_state_init(struct pipe_context *pctx)
 579 {
 580         pctx->set_blend_color = fd_set_blend_color;
 581         pctx->set_stencil_ref = fd_set_stencil_ref;
 582         pctx->set_clip_state = fd_set_clip_state;
 583         pctx->set_sample_mask = fd_set_sample_mask;
 584         pctx->set_constant_buffer = fd_set_constant_buffer;
 585         pctx->set_shader_buffers = fd_set_shader_buffers;
 586         pctx->set_shader_images = fd_set_shader_images;
 587         pctx->set_framebuffer_state = fd_set_framebuffer_state;
 588         pctx->set_polygon_stipple = fd_set_polygon_stipple;
 589         pctx->set_scissor_states = fd_set_scissor_states;
 590         pctx->set_viewport_states = fd_set_viewport_states;
 591
 592         pctx->set_vertex_buffers = fd_set_vertex_buffers;
 593
 594         pctx->bind_blend_state = fd_blend_state_bind;
 595         pctx->delete_blend_state = fd_blend_state_delete;
 596
 597         pctx->bind_rasterizer_state = fd_rasterizer_state_bind;
 598         pctx->delete_rasterizer_state = fd_rasterizer_state_delete;
 599
 600         pctx->bind_depth_stencil_alpha_state = fd_zsa_state_bind;
 601         pctx->delete_depth_stencil_alpha_state = fd_zsa_state_delete;
 602
 603         pctx->create_vertex_elements_state = fd_vertex_state_create;
 604         pctx->delete_vertex_elements_state = fd_vertex_state_delete;
 605         pctx->bind_vertex_elements_state = fd_vertex_state_bind;
 606
 607         pctx->create_stream_output_target = fd_create_stream_output_target;
 608         pctx->stream_output_target_destroy = fd_stream_output_target_destroy;
 609         pctx->set_stream_output_targets = fd_set_stream_output_targets;
 610
 611         if (has_compute(fd_screen(pctx->screen))) {
 612                 pctx->bind_compute_state = fd_bind_compute_state;
 613                 pctx->set_compute_resources = fd_set_compute_resources;
 614                 pctx->set_global_binding = fd_set_global_binding;
 615         }
 616 }