src/gallium/drivers/freedreno/freedreno_state.c

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