src/gallium/drivers/panfrost/pan_job.c

   1 /*
   2  * Copyright (C) 2019 Alyssa Rosenzweig
   3  * Copyright (C) 2014-2017 Broadcom
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the "Software"),
   7  * to deal in the Software without restriction, including without limitation
   8  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   9  * and/or sell copies of the Software, and to permit persons to whom the
  10  * Software is furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice (including the next
  13  * paragraph) shall be included in all copies or substantial portions of the
  14  * Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  22  * SOFTWARE.
  23  *
  24  */
  25
  26 #include <assert.h>
  27
  28 #include "pan_context.h"
  29 #include "util/hash_table.h"
  30 #include "util/ralloc.h"
  31 #include "util/u_format.h"
  32 #include "util/u_pack_color.h"
  33
  34 struct panfrost_job *
  35 panfrost_create_job(struct panfrost_context *ctx)
  36 {
  37         struct panfrost_job *job = rzalloc(ctx, struct panfrost_job);
  38
  39         job->ctx = ctx;
  40
  41         job->bos = _mesa_set_create(job,
  42                                     _mesa_hash_pointer,
  43                                     _mesa_key_pointer_equal);
  44
  45         job->minx = job->miny = ~0;
  46         job->maxx = job->maxy = 0;
  47         job->transient_offset = 0;
  48
  49         util_dynarray_init(&job->headers, job);
  50         util_dynarray_init(&job->gpu_headers, job);
  51         util_dynarray_init(&job->transient_indices, job);
  52
  53         return job;
  54 }
  55
  56 void
  57 panfrost_free_job(struct panfrost_context *ctx, struct panfrost_job *job)
  58 {
  59         if (!job)
  60                 return;
  61
  62         set_foreach(job->bos, entry) {
  63                 struct panfrost_bo *bo = (struct panfrost_bo *)entry->key;
  64                 panfrost_bo_unreference(ctx->base.screen, bo);
  65         }
  66
  67         /* Free up the transient BOs we're sitting on */
  68         struct panfrost_screen *screen = pan_screen(ctx->base.screen);
  69
  70         util_dynarray_foreach(&job->transient_indices, unsigned, index) {
  71                 /* Mark it free */
  72                 BITSET_SET(screen->free_transient, *index);
  73         }
  74
  75         _mesa_hash_table_remove_key(ctx->jobs, &job->key);
  76
  77         if (ctx->job == job)
  78                 ctx->job = NULL;
  79
  80         ralloc_free(job);
  81 }
  82
  83 struct panfrost_job *
  84 panfrost_get_job(struct panfrost_context *ctx,
  85                  struct pipe_surface **cbufs, struct pipe_surface *zsbuf)
  86 {
  87         /* Lookup the job first */
  88
  89         struct panfrost_job_key key = {
  90                 .cbufs = {
  91                         cbufs[0],
  92                         cbufs[1],
  93                         cbufs[2],
  94                         cbufs[3],
  95                 },
  96                 .zsbuf = zsbuf
  97         };
  98
  99         struct hash_entry *entry = _mesa_hash_table_search(ctx->jobs, &key);
 100
 101         if (entry)
 102                 return entry->data;
 103
 104         /* Otherwise, let's create a job */
 105
 106         struct panfrost_job *job = panfrost_create_job(ctx);
 107
 108         /* Save the created job */
 109
 110         memcpy(&job->key, &key, sizeof(key));
 111         _mesa_hash_table_insert(ctx->jobs, &job->key, job);
 112
 113         return job;
 114 }
 115
 116 /* Get the job corresponding to the FBO we're currently rendering into */
 117
 118 struct panfrost_job *
 119 panfrost_get_job_for_fbo(struct panfrost_context *ctx)
 120 {
 121         /* If we're wallpapering, we special case to workaround
 122          * u_blitter abuse */
 123
 124         if (ctx->wallpaper_batch)
 125                 return ctx->wallpaper_batch;
 126
 127         /* If we already began rendering, use that */
 128
 129         if (ctx->job) {
 130                 assert(ctx->job->key.zsbuf == ctx->pipe_framebuffer.zsbuf &&
 131                        !memcmp(ctx->job->key.cbufs,
 132                                ctx->pipe_framebuffer.cbufs,
 133                                sizeof(ctx->job->key.cbufs)));
 134                 return ctx->job;
 135         }
 136
 137         /* If not, look up the job */
 138
 139         struct pipe_surface **cbufs = ctx->pipe_framebuffer.cbufs;
 140         struct pipe_surface *zsbuf = ctx->pipe_framebuffer.zsbuf;
 141         struct panfrost_job *job = panfrost_get_job(ctx, cbufs, zsbuf);
 142
 143         /* Set this job as the current FBO job. Will be reset when updating the
 144          * FB state and when submitting or releasing a job.
 145          */
 146         ctx->job = job;
 147         return job;
 148 }
 149
 150 void
 151 panfrost_job_add_bo(struct panfrost_job *job, struct panfrost_bo *bo)
 152 {
 153         if (!bo)
 154                 return;
 155
 156         if (_mesa_set_search(job->bos, bo))
 157                 return;
 158
 159         panfrost_bo_reference(bo);
 160         _mesa_set_add(job->bos, bo);
 161 }
 162
 163 void
 164 panfrost_flush_jobs_writing_resource(struct panfrost_context *panfrost,
 165                                      struct pipe_resource *prsc)
 166 {
 167 #if 0
 168         struct hash_entry *entry = _mesa_hash_table_search(panfrost->write_jobs,
 169                                    prsc);
 170         if (entry) {
 171                 struct panfrost_job *job = entry->data;
 172                 panfrost_job_submit(panfrost, job);
 173         }
 174 #endif
 175         /* TODO stub */
 176 }
 177
 178 void
 179 panfrost_job_submit(struct panfrost_context *ctx, struct panfrost_job *job)
 180 {
 181         int ret;
 182
 183         panfrost_scoreboard_link_batch(job);
 184
 185         bool has_draws = job->last_job.gpu;
 186         bool is_scanout = panfrost_is_scanout(ctx);
 187
 188         if (!job)
 189                 return;
 190
 191         ret = panfrost_drm_submit_vs_fs_job(ctx, has_draws, is_scanout);
 192
 193         if (ret)
 194                 fprintf(stderr, "panfrost_job_submit failed: %d\n", ret);
 195
 196         /* The job has been submitted, let's invalidate the current FBO job
 197          * cache.
 198          */
 199         assert(!ctx->job || job == ctx->job);
 200         ctx->job = NULL;
 201
 202         /* Remove the job from the ctx->jobs set so that future
 203          * panfrost_get_job() calls don't see it.
 204          * We must reset the job key to avoid removing another valid entry when
 205          * the job is freed.
 206          */
 207         _mesa_hash_table_remove_key(ctx->jobs, &job->key);
 208         memset(&job->key, 0, sizeof(job->key));
 209 }
 210
 211 void
 212 panfrost_job_set_requirements(struct panfrost_context *ctx,
 213                               struct panfrost_job *job)
 214 {
 215         if (ctx->rasterizer && ctx->rasterizer->base.multisample)
 216                 job->requirements |= PAN_REQ_MSAA;
 217
 218         if (ctx->depth_stencil && ctx->depth_stencil->depth.writemask)
 219                 job->requirements |= PAN_REQ_DEPTH_WRITE;
 220 }
 221
 222 /* Helper to smear a 32-bit color across 128-bit components */
 223
 224 static void
 225 pan_pack_color_32(uint32_t *packed, uint32_t v)
 226 {
 227         for (unsigned i = 0; i < 4; ++i)
 228                 packed[i] = v;
 229 }
 230
 231 static void
 232 pan_pack_color_64(uint32_t *packed, uint32_t lo, uint32_t hi)
 233 {
 234         for (unsigned i = 0; i < 4; i += 2) {
 235                 packed[i + 0] = lo;
 236                 packed[i + 1] = hi;
 237         }
 238 }
 239
 240 static void
 241 pan_pack_color(uint32_t *packed, const union pipe_color_union *color, enum pipe_format format)
 242 {
 243         /* Alpha magicked to 1.0 if there is no alpha */
 244
 245         bool has_alpha = util_format_has_alpha(format);
 246         float clear_alpha = has_alpha ? color->f[3] : 1.0f;
 247
 248         /* Packed color depends on the framebuffer format */
 249
 250         const struct util_format_description *desc =
 251                 util_format_description(format);
 252
 253         if (util_format_is_rgba8_variant(desc)) {
 254                 pan_pack_color_32(packed,
 255                                   (float_to_ubyte(clear_alpha) << 24) |
 256                                   (float_to_ubyte(color->f[2]) << 16) |
 257                                   (float_to_ubyte(color->f[1]) <<  8) |
 258                                   (float_to_ubyte(color->f[0]) <<  0));
 259         } else if (format == PIPE_FORMAT_B5G6R5_UNORM) {
 260                 /* First, we convert the components to R5, G6, B5 separately */
 261                 unsigned r5 = CLAMP(color->f[0], 0.0, 1.0) * 31.0;
 262                 unsigned g6 = CLAMP(color->f[1], 0.0, 1.0) * 63.0;
 263                 unsigned b5 = CLAMP(color->f[2], 0.0, 1.0) * 31.0;
 264
 265                 /* Then we pack into a sparse u32. TODO: Why these shifts? */
 266                 pan_pack_color_32(packed, (b5 << 25) | (g6 << 14) | (r5 << 5));
 267         } else if (format == PIPE_FORMAT_B4G4R4A4_UNORM) {
 268                 /* We scale the components against 0xF0 (=240.0), rather than 0xFF */
 269                 unsigned r4 = CLAMP(color->f[0], 0.0, 1.0) * 240.0;
 270                 unsigned g4 = CLAMP(color->f[1], 0.0, 1.0) * 240.0;
 271                 unsigned b4 = CLAMP(color->f[2], 0.0, 1.0) * 240.0;
 272                 unsigned a4 = CLAMP(clear_alpha, 0.0, 1.0) * 240.0;
 273
 274                 /* Pack on *byte* intervals */
 275                 pan_pack_color_32(packed, (a4 << 24) | (b4 << 16) | (g4 << 8) | r4);
 276         } else if (format == PIPE_FORMAT_B5G5R5A1_UNORM) {
 277                 /* Scale as expected but shift oddly */
 278                 unsigned r5 = round(CLAMP(color->f[0], 0.0, 1.0)) * 31.0;
 279                 unsigned g5 = round(CLAMP(color->f[1], 0.0, 1.0)) * 31.0;
 280                 unsigned b5 = round(CLAMP(color->f[2], 0.0, 1.0)) * 31.0;
 281                 unsigned a1 = round(CLAMP(clear_alpha, 0.0, 1.0)) * 1.0;
 282
 283                 pan_pack_color_32(packed, (a1 << 31) | (b5 << 25) | (g5 << 15) | (r5 << 5));
 284         } else {
 285                 /* Try Gallium's generic default path. Doesn't work for all
 286                  * formats but it's a good guess. */
 287
 288                 union util_color out;
 289
 290                 if (util_format_is_pure_integer(format)) {
 291                         memcpy(out.ui, color->ui, 16);
 292                 } else {
 293                         util_pack_color(color->f, format, &out);
 294                 }
 295
 296                 unsigned size = util_format_get_blocksize(format);
 297
 298                 if (size == 1) {
 299                         unsigned b = out.ui[0];
 300                         unsigned s = b | (b << 8);
 301                         pan_pack_color_32(packed, s | (s << 16));
 302                 } else if (size == 2)
 303                         pan_pack_color_32(packed, out.ui[0] | (out.ui[0] << 16));
 304                 else if (size == 4)
 305                         pan_pack_color_32(packed, out.ui[0]);
 306                 else if (size == 8)
 307                         pan_pack_color_64(packed, out.ui[0], out.ui[1]);
 308                 else if (size == 16)
 309                         memcpy(packed, out.ui, 16);
 310                 else
 311                         unreachable("Unknown generic format size packing clear colour");
 312         }
 313 }
 314
 315 void
 316 panfrost_job_clear(struct panfrost_context *ctx,
 317                    struct panfrost_job *job,
 318                    unsigned buffers,
 319                    const union pipe_color_union *color,
 320                    double depth, unsigned stencil)
 321
 322 {
 323         if (buffers & PIPE_CLEAR_COLOR) {
 324                 for (unsigned i = 0; i < PIPE_MAX_COLOR_BUFS; ++i) {
 325                         if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
 326                                 continue;
 327
 328                         enum pipe_format format = ctx->pipe_framebuffer.cbufs[i]->format;
 329                         pan_pack_color(job->clear_color[i], color, format);
 330                 }
 331         }
 332
 333         if (buffers & PIPE_CLEAR_DEPTH) {
 334                 job->clear_depth = depth;
 335         }
 336
 337         if (buffers & PIPE_CLEAR_STENCIL) {
 338                 job->clear_stencil = stencil;
 339         }
 340
 341         job->clear |= buffers;
 342
 343         /* Clearing affects the entire framebuffer (by definition -- this is
 344          * the Gallium clear callback, which clears the whole framebuffer. If
 345          * the scissor test were enabled from the GL side, the state tracker
 346          * would emit a quad instead and we wouldn't go down this code path) */
 347
 348         panfrost_job_union_scissor(job, 0, 0,
 349                                    ctx->pipe_framebuffer.width,
 350                                    ctx->pipe_framebuffer.height);
 351 }
 352
 353 void
 354 panfrost_flush_jobs_reading_resource(struct panfrost_context *panfrost,
 355                                      struct pipe_resource *prsc)
 356 {
 357         struct panfrost_resource *rsc = pan_resource(prsc);
 358
 359         panfrost_flush_jobs_writing_resource(panfrost, prsc);
 360
 361         hash_table_foreach(panfrost->jobs, entry) {
 362                 struct panfrost_job *job = entry->data;
 363
 364                 if (_mesa_set_search(job->bos, rsc->bo)) {
 365                         printf("TODO: submit job for flush\n");
 366                         //panfrost_job_submit(panfrost, job);
 367                         continue;
 368                 }
 369         }
 370 }
 371
 372 static bool
 373 panfrost_job_compare(const void *a, const void *b)
 374 {
 375         return memcmp(a, b, sizeof(struct panfrost_job_key)) == 0;
 376 }
 377
 378 static uint32_t
 379 panfrost_job_hash(const void *key)
 380 {
 381         return _mesa_hash_data(key, sizeof(struct panfrost_job_key));
 382 }
 383
 384 /* Given a new bounding rectangle (scissor), let the job cover the union of the
 385  * new and old bounding rectangles */
 386
 387 void
 388 panfrost_job_union_scissor(struct panfrost_job *job,
 389                            unsigned minx, unsigned miny,
 390                            unsigned maxx, unsigned maxy)
 391 {
 392         job->minx = MIN2(job->minx, minx);
 393         job->miny = MIN2(job->miny, miny);
 394         job->maxx = MAX2(job->maxx, maxx);
 395         job->maxy = MAX2(job->maxy, maxy);
 396 }
 397
 398 void
 399 panfrost_job_init(struct panfrost_context *ctx)
 400 {
 401         ctx->jobs = _mesa_hash_table_create(ctx,
 402                                             panfrost_job_hash,
 403                                             panfrost_job_compare);
 404
 405         ctx->write_jobs = _mesa_hash_table_create(ctx,
 406                           _mesa_hash_pointer,
 407                           _mesa_key_pointer_equal);
 408 }