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