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 "pan_context.h"
  27 #include "util/hash_table.h"
  28 #include "util/ralloc.h"
  29 #include "util/u_format.h"
  30 #include "util/u_pack_color.h"
  31
  32 struct panfrost_job *
  33 panfrost_create_job(struct panfrost_context *ctx)
  34 {
  35         struct panfrost_job *job = rzalloc(ctx, struct panfrost_job);
  36
  37         job->ctx = ctx;
  38
  39         job->bos = _mesa_set_create(job,
  40                                     _mesa_hash_pointer,
  41                                     _mesa_key_pointer_equal);
  42
  43         job->minx = job->miny = ~0;
  44         job->maxx = job->maxy = 0;
  45
  46         util_dynarray_init(&job->headers, job);
  47         util_dynarray_init(&job->gpu_headers, job);
  48         util_dynarray_init(&job->transient_indices, job);
  49
  50         return job;
  51 }
  52
  53 void
  54 panfrost_free_job(struct panfrost_context *ctx, struct panfrost_job *job)
  55 {
  56         if (!job)
  57                 return;
  58
  59         set_foreach(job->bos, entry) {
  60                 struct panfrost_bo *bo = (struct panfrost_bo *)entry->key;
  61                 panfrost_bo_unreference(ctx->base.screen, bo);
  62         }
  63
  64         _mesa_hash_table_remove_key(ctx->jobs, &job->key);
  65
  66         if (ctx->job == job)
  67                 ctx->job = NULL;
  68
  69         ralloc_free(job);
  70 }
  71
  72 struct panfrost_job *
  73 panfrost_get_job(struct panfrost_context *ctx,
  74                  struct pipe_surface **cbufs, struct pipe_surface *zsbuf)
  75 {
  76         /* Lookup the job first */
  77
  78         struct panfrost_job_key key = {
  79                 .cbufs = {
  80                         cbufs[0],
  81                         cbufs[1],
  82                         cbufs[2],
  83                         cbufs[3],
  84                 },
  85                 .zsbuf = zsbuf
  86         };
  87
  88         struct hash_entry *entry = _mesa_hash_table_search(ctx->jobs, &key);
  89
  90         if (entry)
  91                 return entry->data;
  92
  93         /* Otherwise, let's create a job */
  94
  95         struct panfrost_job *job = panfrost_create_job(ctx);
  96
  97         /* Save the created job */
  98
  99         memcpy(&job->key, &key, sizeof(key));
 100         _mesa_hash_table_insert(ctx->jobs, &job->key, job);
 101
 102         return job;
 103 }
 104
 105 /* Get the job corresponding to the FBO we're currently rendering into */
 106
 107 struct panfrost_job *
 108 panfrost_get_job_for_fbo(struct panfrost_context *ctx)
 109 {
 110         /* If we're wallpapering, we special case to workaround
 111          * u_blitter abuse */
 112
 113         if (ctx->wallpaper_batch)
 114                 return ctx->wallpaper_batch;
 115
 116         /* If we already began rendering, use that */
 117
 118         if (ctx->job)
 119                 return ctx->job;
 120
 121         /* If not, look up the job */
 122
 123         struct pipe_surface **cbufs = ctx->pipe_framebuffer.cbufs;
 124         struct pipe_surface *zsbuf = ctx->pipe_framebuffer.zsbuf;
 125         struct panfrost_job *job = panfrost_get_job(ctx, cbufs, zsbuf);
 126
 127         return job;
 128 }
 129
 130 void
 131 panfrost_job_add_bo(struct panfrost_job *job, struct panfrost_bo *bo)
 132 {
 133         if (!bo)
 134                 return;
 135
 136         if (_mesa_set_search(job->bos, bo))
 137                 return;
 138
 139         panfrost_bo_reference(bo);
 140         _mesa_set_add(job->bos, bo);
 141 }
 142
 143 void
 144 panfrost_flush_jobs_writing_resource(struct panfrost_context *panfrost,
 145                                      struct pipe_resource *prsc)
 146 {
 147 #if 0
 148         struct hash_entry *entry = _mesa_hash_table_search(panfrost->write_jobs,
 149                                    prsc);
 150         if (entry) {
 151                 struct panfrost_job *job = entry->data;
 152                 panfrost_job_submit(panfrost, job);
 153         }
 154 #endif
 155         /* TODO stub */
 156 }
 157
 158 void
 159 panfrost_job_submit(struct panfrost_context *ctx, struct panfrost_job *job)
 160 {
 161         int ret;
 162
 163         panfrost_scoreboard_link_batch(job);
 164
 165         bool has_draws = job->last_job.gpu;
 166         bool is_scanout = panfrost_is_scanout(ctx);
 167
 168         if (!job)
 169                 return;
 170
 171         ret = panfrost_drm_submit_vs_fs_job(ctx, has_draws, is_scanout);
 172
 173         if (ret)
 174                 fprintf(stderr, "panfrost_job_submit failed: %d\n", ret);
 175 }
 176
 177 void
 178 panfrost_job_set_requirements(struct panfrost_context *ctx,
 179                               struct panfrost_job *job)
 180 {
 181         if (ctx->rasterizer && ctx->rasterizer->base.multisample)
 182                 job->requirements |= PAN_REQ_MSAA;
 183
 184         if (ctx->depth_stencil && ctx->depth_stencil->depth.writemask)
 185                 job->requirements |= PAN_REQ_DEPTH_WRITE;
 186 }
 187
 188 /* Helper to smear a 32-bit color across 128-bit components */
 189
 190 static void
 191 pan_pack_color_32(uint32_t *packed, uint32_t v)
 192 {
 193         for (unsigned i = 0; i < 4; ++i)
 194                 packed[i] = v;
 195 }
 196
 197 static void
 198 pan_pack_color_64(uint32_t *packed, uint32_t lo, uint32_t hi)
 199 {
 200         for (unsigned i = 0; i < 4; i += 2) {
 201                 packed[i + 0] = lo;
 202                 packed[i + 1] = hi;
 203         }
 204 }
 205
 206 static void
 207 pan_pack_color(uint32_t *packed, const union pipe_color_union *color, enum pipe_format format)
 208 {
 209         /* Alpha magicked to 1.0 if there is no alpha */
 210
 211         bool has_alpha = util_format_has_alpha(format);
 212         float clear_alpha = has_alpha ? color->f[3] : 1.0f;
 213
 214         /* Packed color depends on the framebuffer format */
 215
 216         const struct util_format_description *desc =
 217                 util_format_description(format);
 218
 219         if (util_format_is_rgba8_variant(desc)) {
 220                 pan_pack_color_32(packed,
 221                                   (float_to_ubyte(clear_alpha) << 24) |
 222                                   (float_to_ubyte(color->f[2]) << 16) |
 223                                   (float_to_ubyte(color->f[1]) <<  8) |
 224                                   (float_to_ubyte(color->f[0]) <<  0));
 225         } else if (format == PIPE_FORMAT_B5G6R5_UNORM) {
 226                 /* First, we convert the components to R5, G6, B5 separately */
 227                 unsigned r5 = CLAMP(color->f[0], 0.0, 1.0) * 31.0;
 228                 unsigned g6 = CLAMP(color->f[1], 0.0, 1.0) * 63.0;
 229                 unsigned b5 = CLAMP(color->f[2], 0.0, 1.0) * 31.0;
 230
 231                 /* Then we pack into a sparse u32. TODO: Why these shifts? */
 232                 pan_pack_color_32(packed, (b5 << 25) | (g6 << 14) | (r5 << 5));
 233         } else if (format == PIPE_FORMAT_B4G4R4A4_UNORM) {
 234                 /* We scale the components against 0xF0 (=240.0), rather than 0xFF */
 235                 unsigned r4 = CLAMP(color->f[0], 0.0, 1.0) * 240.0;
 236                 unsigned g4 = CLAMP(color->f[1], 0.0, 1.0) * 240.0;
 237                 unsigned b4 = CLAMP(color->f[2], 0.0, 1.0) * 240.0;
 238                 unsigned a4 = CLAMP(clear_alpha, 0.0, 1.0) * 240.0;
 239
 240                 /* Pack on *byte* intervals */
 241                 pan_pack_color_32(packed, (a4 << 24) | (b4 << 16) | (g4 << 8) | r4);
 242         } else if (format == PIPE_FORMAT_B5G5R5A1_UNORM) {
 243                 /* Scale as expected but shift oddly */
 244                 unsigned r5 = round(CLAMP(color->f[0], 0.0, 1.0)) * 31.0;
 245                 unsigned g5 = round(CLAMP(color->f[1], 0.0, 1.0)) * 31.0;
 246                 unsigned b5 = round(CLAMP(color->f[2], 0.0, 1.0)) * 31.0;
 247                 unsigned a1 = round(CLAMP(clear_alpha, 0.0, 1.0)) * 1.0;
 248
 249                 pan_pack_color_32(packed, (a1 << 31) | (b5 << 25) | (g5 << 15) | (r5 << 5));
 250         } else {
 251                 /* Try Gallium's generic default path. Doesn't work for all
 252                  * formats but it's a good guess. */
 253
 254                 union util_color out;
 255
 256                 if (util_format_is_pure_integer(format)) {
 257                         memcpy(out.ui, color->ui, 16);
 258                 } else {
 259                         util_pack_color(color->f, format, &out);
 260                 }
 261
 262                 unsigned size = util_format_get_blocksize(format);
 263
 264                 if (size == 1) {
 265                         unsigned b = out.ui[0];
 266                         unsigned s = b | (b << 8);
 267                         pan_pack_color_32(packed, s | (s << 16));
 268                 } else if (size == 2)
 269                         pan_pack_color_32(packed, out.ui[0] | (out.ui[0] << 16));
 270                 else if (size == 4)
 271                         pan_pack_color_32(packed, out.ui[0]);
 272                 else if (size == 8)
 273                         pan_pack_color_64(packed, out.ui[0], out.ui[1]);
 274                 else if (size == 16)
 275                         memcpy(packed, out.ui, 16);
 276                 else
 277                         unreachable("Unknown generic format size packing clear colour");
 278         }
 279 }
 280
 281 void
 282 panfrost_job_clear(struct panfrost_context *ctx,
 283                    struct panfrost_job *job,
 284                    unsigned buffers,
 285                    const union pipe_color_union *color,
 286                    double depth, unsigned stencil)
 287
 288 {
 289         if (buffers & PIPE_CLEAR_COLOR) {
 290                 for (unsigned i = 0; i < PIPE_MAX_COLOR_BUFS; ++i) {
 291                         if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
 292                                 continue;
 293
 294                         enum pipe_format format = ctx->pipe_framebuffer.cbufs[i]->format;
 295                         pan_pack_color(job->clear_color[i], color, format);
 296                 }
 297         }
 298
 299         if (buffers & PIPE_CLEAR_DEPTH) {
 300                 job->clear_depth = depth;
 301         }
 302
 303         if (buffers & PIPE_CLEAR_STENCIL) {
 304                 job->clear_stencil = stencil;
 305         }
 306
 307         job->clear |= buffers;
 308
 309         /* Clearing affects the entire framebuffer (by definition -- this is
 310          * the Gallium clear callback, which clears the whole framebuffer. If
 311          * the scissor test were enabled from the GL side, the state tracker
 312          * would emit a quad instead and we wouldn't go down this code path) */
 313
 314         panfrost_job_union_scissor(job, 0, 0,
 315                                    ctx->pipe_framebuffer.width,
 316                                    ctx->pipe_framebuffer.height);
 317 }
 318
 319 void
 320 panfrost_flush_jobs_reading_resource(struct panfrost_context *panfrost,
 321                                      struct pipe_resource *prsc)
 322 {
 323         struct panfrost_resource *rsc = pan_resource(prsc);
 324
 325         panfrost_flush_jobs_writing_resource(panfrost, prsc);
 326
 327         hash_table_foreach(panfrost->jobs, entry) {
 328                 struct panfrost_job *job = entry->data;
 329
 330                 if (_mesa_set_search(job->bos, rsc->bo)) {
 331                         printf("TODO: submit job for flush\n");
 332                         //panfrost_job_submit(panfrost, job);
 333                         continue;
 334                 }
 335         }
 336 }
 337
 338 static bool
 339 panfrost_job_compare(const void *a, const void *b)
 340 {
 341         return memcmp(a, b, sizeof(struct panfrost_job_key)) == 0;
 342 }
 343
 344 static uint32_t
 345 panfrost_job_hash(const void *key)
 346 {
 347         return _mesa_hash_data(key, sizeof(struct panfrost_job_key));
 348 }
 349
 350 /* Given a new bounding rectangle (scissor), let the job cover the union of the
 351  * new and old bounding rectangles */
 352
 353 void
 354 panfrost_job_union_scissor(struct panfrost_job *job,
 355                            unsigned minx, unsigned miny,
 356                            unsigned maxx, unsigned maxy)
 357 {
 358         job->minx = MIN2(job->minx, minx);
 359         job->miny = MIN2(job->miny, miny);
 360         job->maxx = MAX2(job->maxx, maxx);
 361         job->maxy = MAX2(job->maxy, maxy);
 362 }
 363
 364 void
 365 panfrost_job_init(struct panfrost_context *ctx)
 366 {
 367         ctx->jobs = _mesa_hash_table_create(ctx,
 368                                             panfrost_job_hash,
 369                                             panfrost_job_compare);
 370
 371         ctx->write_jobs = _mesa_hash_table_create(ctx,
 372                           _mesa_hash_pointer,
 373                           _mesa_key_pointer_equal);
 374 }