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_framebuffer_state(struct pipe_context *pctx,
 166                 const struct pipe_framebuffer_state *framebuffer)
 167 {
 168         struct fd_context *ctx = fd_context(pctx);
 169         struct pipe_framebuffer_state *cso;
 170
 171         if (ctx->screen->reorder) {
 172                 struct fd_batch *batch, *old_batch = NULL;
 173
 174                 fd_batch_reference(&old_batch, ctx->batch);
 175
 176                 if (likely(old_batch))
 177                         fd_batch_set_stage(old_batch, FD_STAGE_NULL);
 178
 179                 batch = fd_batch_from_fb(&ctx->screen->batch_cache, ctx, framebuffer);
 180                 fd_batch_reference(&ctx->batch, NULL);
 181                 fd_reset_wfi(batch);
 182                 ctx->batch = batch;
 183                 fd_context_all_dirty(ctx);
 184
 185                 if (old_batch && old_batch->blit && !old_batch->back_blit) {
 186                         /* for blits, there is not really much point in hanging on
 187                          * to the uncommitted batch (ie. you probably don't blit
 188                          * multiple times to the same surface), so we might as
 189                          * well go ahead and flush this one:
 190                          */
 191                         fd_batch_flush(old_batch, false);
 192                 }
 193
 194                 fd_batch_reference(&old_batch, NULL);
 195         } else {
 196                 DBG("%d: cbufs[0]=%p, zsbuf=%p", ctx->batch->needs_flush,
 197                                 framebuffer->cbufs[0], framebuffer->zsbuf);
 198                 fd_batch_flush(ctx->batch, false);
 199         }
 200
 201         cso = &ctx->batch->framebuffer;
 202
 203         util_copy_framebuffer_state(cso, framebuffer);
 204
 205         ctx->dirty |= FD_DIRTY_FRAMEBUFFER;
 206
 207         ctx->disabled_scissor.minx = 0;
 208         ctx->disabled_scissor.miny = 0;
 209         ctx->disabled_scissor.maxx = cso->width;
 210         ctx->disabled_scissor.maxy = cso->height;
 211
 212         ctx->dirty |= FD_DIRTY_SCISSOR;
 213 }
 214
 215 static void
 216 fd_set_polygon_stipple(struct pipe_context *pctx,
 217                 const struct pipe_poly_stipple *stipple)
 218 {
 219         struct fd_context *ctx = fd_context(pctx);
 220         ctx->stipple = *stipple;
 221         ctx->dirty |= FD_DIRTY_STIPPLE;
 222 }
 223
 224 static void
 225 fd_set_scissor_states(struct pipe_context *pctx,
 226                 unsigned start_slot,
 227                 unsigned num_scissors,
 228                 const struct pipe_scissor_state *scissor)
 229 {
 230         struct fd_context *ctx = fd_context(pctx);
 231
 232         ctx->scissor = *scissor;
 233         ctx->dirty |= FD_DIRTY_SCISSOR;
 234 }
 235
 236 static void
 237 fd_set_viewport_states(struct pipe_context *pctx,
 238                 unsigned start_slot,
 239                 unsigned num_viewports,
 240                 const struct pipe_viewport_state *viewport)
 241 {
 242         struct fd_context *ctx = fd_context(pctx);
 243         ctx->viewport = *viewport;
 244         ctx->dirty |= FD_DIRTY_VIEWPORT;
 245 }
 246
 247 static void
 248 fd_set_vertex_buffers(struct pipe_context *pctx,
 249                 unsigned start_slot, unsigned count,
 250                 const struct pipe_vertex_buffer *vb)
 251 {
 252         struct fd_context *ctx = fd_context(pctx);
 253         struct fd_vertexbuf_stateobj *so = &ctx->vtx.vertexbuf;
 254         int i;
 255
 256         /* on a2xx, pitch is encoded in the vtx fetch instruction, so
 257          * we need to mark VTXSTATE as dirty as well to trigger patching
 258          * and re-emitting the vtx shader:
 259          */
 260         if (ctx->screen->gpu_id < 300) {
 261                 for (i = 0; i < count; i++) {
 262                         bool new_enabled = vb && vb[i].buffer.resource;
 263                         bool old_enabled = so->vb[i].buffer.resource != NULL;
 264                         uint32_t new_stride = vb ? vb[i].stride : 0;
 265                         uint32_t old_stride = so->vb[i].stride;
 266                         if ((new_enabled != old_enabled) || (new_stride != old_stride)) {
 267                                 ctx->dirty |= FD_DIRTY_VTXSTATE;
 268                                 break;
 269                         }
 270                 }
 271         }
 272
 273         util_set_vertex_buffers_mask(so->vb, &so->enabled_mask, vb, start_slot, count);
 274         so->count = util_last_bit(so->enabled_mask);
 275
 276         ctx->dirty |= FD_DIRTY_VTXBUF;
 277 }
 278
 279 static void
 280 fd_blend_state_bind(struct pipe_context *pctx, void *hwcso)
 281 {
 282         struct fd_context *ctx = fd_context(pctx);
 283         struct pipe_blend_state *cso = hwcso;
 284         bool old_is_dual = ctx->blend ?
 285                 ctx->blend->rt[0].blend_enable && util_blend_state_is_dual(ctx->blend, 0) :
 286                 false;
 287         bool new_is_dual = cso ?
 288                 cso->rt[0].blend_enable && util_blend_state_is_dual(cso, 0) :
 289                 false;
 290         ctx->blend = hwcso;
 291         ctx->dirty |= FD_DIRTY_BLEND;
 292         if (old_is_dual != new_is_dual)
 293                 ctx->dirty |= FD_DIRTY_BLEND_DUAL;
 294 }
 295
 296 static void
 297 fd_blend_state_delete(struct pipe_context *pctx, void *hwcso)
 298 {
 299         FREE(hwcso);
 300 }
 301
 302 static void
 303 fd_rasterizer_state_bind(struct pipe_context *pctx, void *hwcso)
 304 {
 305         struct fd_context *ctx = fd_context(pctx);
 306         struct pipe_scissor_state *old_scissor = fd_context_get_scissor(ctx);
 307
 308         ctx->rasterizer = hwcso;
 309         ctx->dirty |= FD_DIRTY_RASTERIZER;
 310
 311         /* if scissor enable bit changed we need to mark scissor
 312          * state as dirty as well:
 313          * NOTE: we can do a shallow compare, since we only care
 314          * if it changed to/from &ctx->disable_scissor
 315          */
 316         if (old_scissor != fd_context_get_scissor(ctx))
 317                 ctx->dirty |= FD_DIRTY_SCISSOR;
 318 }
 319
 320 static void
 321 fd_rasterizer_state_delete(struct pipe_context *pctx, void *hwcso)
 322 {
 323         FREE(hwcso);
 324 }
 325
 326 static void
 327 fd_zsa_state_bind(struct pipe_context *pctx, void *hwcso)
 328 {
 329         struct fd_context *ctx = fd_context(pctx);
 330         ctx->zsa = hwcso;
 331         ctx->dirty |= FD_DIRTY_ZSA;
 332 }
 333
 334 static void
 335 fd_zsa_state_delete(struct pipe_context *pctx, void *hwcso)
 336 {
 337         FREE(hwcso);
 338 }
 339
 340 static void *
 341 fd_vertex_state_create(struct pipe_context *pctx, unsigned num_elements,
 342                 const struct pipe_vertex_element *elements)
 343 {
 344         struct fd_vertex_stateobj *so = CALLOC_STRUCT(fd_vertex_stateobj);
 345
 346         if (!so)
 347                 return NULL;
 348
 349         memcpy(so->pipe, elements, sizeof(*elements) * num_elements);
 350         so->num_elements = num_elements;
 351
 352         return so;
 353 }
 354
 355 static void
 356 fd_vertex_state_delete(struct pipe_context *pctx, void *hwcso)
 357 {
 358         FREE(hwcso);
 359 }
 360
 361 static void
 362 fd_vertex_state_bind(struct pipe_context *pctx, void *hwcso)
 363 {
 364         struct fd_context *ctx = fd_context(pctx);
 365         ctx->vtx.vtx = hwcso;
 366         ctx->dirty |= FD_DIRTY_VTXSTATE;
 367 }
 368
 369 static struct pipe_stream_output_target *
 370 fd_create_stream_output_target(struct pipe_context *pctx,
 371                 struct pipe_resource *prsc, unsigned buffer_offset,
 372                 unsigned buffer_size)
 373 {
 374         struct pipe_stream_output_target *target;
 375         struct fd_resource *rsc = fd_resource(prsc);
 376
 377         target = CALLOC_STRUCT(pipe_stream_output_target);
 378         if (!target)
 379                 return NULL;
 380
 381         pipe_reference_init(&target->reference, 1);
 382         pipe_resource_reference(&target->buffer, prsc);
 383
 384         target->context = pctx;
 385         target->buffer_offset = buffer_offset;
 386         target->buffer_size = buffer_size;
 387
 388         assert(rsc->base.b.target == PIPE_BUFFER);
 389         util_range_add(&rsc->valid_buffer_range,
 390                 buffer_offset, buffer_offset + buffer_size);
 391
 392         return target;
 393 }
 394
 395 static void
 396 fd_stream_output_target_destroy(struct pipe_context *pctx,
 397                 struct pipe_stream_output_target *target)
 398 {
 399         pipe_resource_reference(&target->buffer, NULL);
 400         FREE(target);
 401 }
 402
 403 static void
 404 fd_set_stream_output_targets(struct pipe_context *pctx,
 405                 unsigned num_targets, struct pipe_stream_output_target **targets,
 406                 const unsigned *offsets)
 407 {
 408         struct fd_context *ctx = fd_context(pctx);
 409         struct fd_streamout_stateobj *so = &ctx->streamout;
 410         unsigned i;
 411
 412         debug_assert(num_targets <= ARRAY_SIZE(so->targets));
 413
 414         for (i = 0; i < num_targets; i++) {
 415                 boolean changed = targets[i] != so->targets[i];
 416                 boolean append = (offsets[i] == (unsigned)-1);
 417
 418                 if (!changed && append)
 419                         continue;
 420
 421                 if (!append)
 422                         so->offsets[i] = offsets[i];
 423
 424                 pipe_so_target_reference(&so->targets[i], targets[i]);
 425         }
 426
 427         for (; i < so->num_targets; i++) {
 428                 pipe_so_target_reference(&so->targets[i], NULL);
 429         }
 430
 431         so->num_targets = num_targets;
 432
 433         ctx->dirty |= FD_DIRTY_STREAMOUT;
 434 }
 435
 436 static void
 437 fd_bind_compute_state(struct pipe_context *pctx, void *state)
 438 {
 439         struct fd_context *ctx = fd_context(pctx);
 440         ctx->compute = state;
 441         ctx->dirty_shader[PIPE_SHADER_COMPUTE] |= FD_DIRTY_SHADER_PROG;
 442 }
 443
 444 static void
 445 fd_set_compute_resources(struct pipe_context *pctx,
 446                 unsigned start, unsigned count, struct pipe_surface **prscs)
 447 {
 448         // TODO
 449 }
 450
 451 static void
 452 fd_set_global_binding(struct pipe_context *pctx,
 453                 unsigned first, unsigned count, struct pipe_resource **prscs,
 454                 uint32_t **handles)
 455 {
 456         /* TODO only used by clover.. seems to need us to return the actual
 457          * gpuaddr of the buffer.. which isn't really exposed to mesa atm.
 458          * How is this used?
 459          */
 460 }
 461
 462 void
 463 fd_state_init(struct pipe_context *pctx)
 464 {
 465         pctx->set_blend_color = fd_set_blend_color;
 466         pctx->set_stencil_ref = fd_set_stencil_ref;
 467         pctx->set_clip_state = fd_set_clip_state;
 468         pctx->set_sample_mask = fd_set_sample_mask;
 469         pctx->set_constant_buffer = fd_set_constant_buffer;
 470         pctx->set_shader_buffers = fd_set_shader_buffers;
 471         pctx->set_framebuffer_state = fd_set_framebuffer_state;
 472         pctx->set_polygon_stipple = fd_set_polygon_stipple;
 473         pctx->set_scissor_states = fd_set_scissor_states;
 474         pctx->set_viewport_states = fd_set_viewport_states;
 475
 476         pctx->set_vertex_buffers = fd_set_vertex_buffers;
 477
 478         pctx->bind_blend_state = fd_blend_state_bind;
 479         pctx->delete_blend_state = fd_blend_state_delete;
 480
 481         pctx->bind_rasterizer_state = fd_rasterizer_state_bind;
 482         pctx->delete_rasterizer_state = fd_rasterizer_state_delete;
 483
 484         pctx->bind_depth_stencil_alpha_state = fd_zsa_state_bind;
 485         pctx->delete_depth_stencil_alpha_state = fd_zsa_state_delete;
 486
 487         pctx->create_vertex_elements_state = fd_vertex_state_create;
 488         pctx->delete_vertex_elements_state = fd_vertex_state_delete;
 489         pctx->bind_vertex_elements_state = fd_vertex_state_bind;
 490
 491         pctx->create_stream_output_target = fd_create_stream_output_target;
 492         pctx->stream_output_target_destroy = fd_stream_output_target_destroy;
 493         pctx->set_stream_output_targets = fd_set_stream_output_targets;
 494
 495         if (has_compute(fd_screen(pctx->screen))) {
 496                 pctx->bind_compute_state = fd_bind_compute_state;
 497                 pctx->set_compute_resources = fd_set_compute_resources;
 498                 pctx->set_global_binding = fd_set_global_binding;
 499         }
 500 }