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