src/gallium/drivers/v3d/v3d_uniforms.c

   1 /*
   2  * Copyright © 2014-2017 Broadcom
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include "util/u_pack_color.h"
  25 #include "util/format_srgb.h"
  26
  27 #include "v3d_context.h"
  28 #include "compiler/v3d_compiler.h"
  29 #include "broadcom/cle/v3d_packet_v33_pack.h"
  30
  31 static uint32_t
  32 get_texrect_scale(struct v3d_texture_stateobj *texstate,
  33                   enum quniform_contents contents,
  34                   uint32_t data)
  35 {
  36         struct pipe_sampler_view *texture = texstate->textures[data];
  37         uint32_t dim;
  38
  39         if (contents == QUNIFORM_TEXRECT_SCALE_X)
  40                 dim = texture->texture->width0;
  41         else
  42                 dim = texture->texture->height0;
  43
  44         return fui(1.0f / dim);
  45 }
  46
  47 static uint32_t
  48 get_texture_size(struct v3d_texture_stateobj *texstate,
  49                  enum quniform_contents contents,
  50                  uint32_t data)
  51 {
  52         struct pipe_sampler_view *texture = texstate->textures[data];
  53
  54         switch (contents) {
  55         case QUNIFORM_TEXTURE_WIDTH:
  56                 return u_minify(texture->texture->width0,
  57                                 texture->u.tex.first_level);
  58         case QUNIFORM_TEXTURE_HEIGHT:
  59                 return u_minify(texture->texture->height0,
  60                                 texture->u.tex.first_level);
  61         case QUNIFORM_TEXTURE_DEPTH:
  62                 return u_minify(texture->texture->depth0,
  63                                 texture->u.tex.first_level);
  64         case QUNIFORM_TEXTURE_ARRAY_SIZE:
  65                 return texture->texture->array_size;
  66         case QUNIFORM_TEXTURE_LEVELS:
  67                 return (texture->u.tex.last_level -
  68                         texture->u.tex.first_level) + 1;
  69         default:
  70                 unreachable("Bad texture size field");
  71         }
  72 }
  73
  74 static uint32_t
  75 get_image_size(struct v3d_shaderimg_stateobj *shaderimg,
  76                enum quniform_contents contents,
  77                uint32_t data)
  78 {
  79         struct v3d_image_view *image = &shaderimg->si[data];
  80
  81         switch (contents) {
  82         case QUNIFORM_IMAGE_WIDTH:
  83                 return u_minify(image->base.resource->width0,
  84                                 image->base.u.tex.level);
  85         case QUNIFORM_IMAGE_HEIGHT:
  86                 return u_minify(image->base.resource->height0,
  87                                 image->base.u.tex.level);
  88         case QUNIFORM_IMAGE_DEPTH:
  89                 return u_minify(image->base.resource->depth0,
  90                                 image->base.u.tex.level);
  91         case QUNIFORM_IMAGE_ARRAY_SIZE:
  92                 return image->base.resource->array_size;
  93         default:
  94                 unreachable("Bad texture size field");
  95         }
  96 }
  97
  98 static struct v3d_bo *
  99 v3d_upload_ubo(struct v3d_context *v3d,
 100                struct v3d_compiled_shader *shader,
 101                const uint32_t *gallium_uniforms)
 102 {
 103         if (!shader->prog_data.base->ubo_size)
 104                 return NULL;
 105
 106         struct v3d_bo *ubo = v3d_bo_alloc(v3d->screen,
 107                                           shader->prog_data.base->ubo_size,
 108                                           "ubo");
 109         void *data = v3d_bo_map(ubo);
 110         for (uint32_t i = 0; i < shader->prog_data.base->num_ubo_ranges; i++) {
 111                 memcpy(data + shader->prog_data.base->ubo_ranges[i].dst_offset,
 112                        ((const void *)gallium_uniforms +
 113                         shader->prog_data.base->ubo_ranges[i].src_offset),
 114                        shader->prog_data.base->ubo_ranges[i].size);
 115         }
 116
 117         return ubo;
 118 }
 119
 120 /**
 121  *  Writes the V3D 3.x P0 (CFG_MODE=1) texture parameter.
 122  *
 123  * Some bits of this field are dependent on the type of sample being done by
 124  * the shader, while other bits are dependent on the sampler state.  We OR the
 125  * two together here.
 126  */
 127 static void
 128 write_texture_p0(struct v3d_job *job,
 129                  struct v3d_cl_out **uniforms,
 130                  struct v3d_texture_stateobj *texstate,
 131                  uint32_t unit,
 132                  uint32_t shader_data)
 133 {
 134         struct pipe_sampler_state *psampler = texstate->samplers[unit];
 135         struct v3d_sampler_state *sampler = v3d_sampler_state(psampler);
 136
 137         cl_aligned_u32(uniforms, shader_data | sampler->p0);
 138 }
 139
 140 /** Writes the V3D 3.x P1 (CFG_MODE=1) texture parameter. */
 141 static void
 142 write_texture_p1(struct v3d_job *job,
 143                  struct v3d_cl_out **uniforms,
 144                  struct v3d_texture_stateobj *texstate,
 145                  uint32_t data)
 146 {
 147         /* Extract the texture unit from the top bits, and the compiler's
 148          * packed p1 from the bottom.
 149          */
 150         uint32_t unit = data >> 5;
 151         uint32_t p1 = data & 0x1f;
 152
 153         struct pipe_sampler_view *psview = texstate->textures[unit];
 154         struct v3d_sampler_view *sview = v3d_sampler_view(psview);
 155
 156         struct V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1 unpacked = {
 157                 .texture_state_record_base_address = texstate->texture_state[unit],
 158         };
 159
 160         uint32_t packed;
 161         V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1_pack(&job->indirect,
 162                                                          (uint8_t *)&packed,
 163                                                          &unpacked);
 164
 165         cl_aligned_u32(uniforms, p1 | packed | sview->p1);
 166 }
 167
 168 /** Writes the V3D 4.x TMU configuration parameter 0. */
 169 static void
 170 write_tmu_p0(struct v3d_job *job,
 171              struct v3d_cl_out **uniforms,
 172              struct v3d_texture_stateobj *texstate,
 173              uint32_t data)
 174 {
 175         int unit  = v3d_tmu_config_data_get_unit(data);
 176         struct pipe_sampler_view *psview = texstate->textures[unit];
 177         struct v3d_sampler_view *sview = v3d_sampler_view(psview);
 178         struct v3d_resource *rsc = v3d_resource(sview->texture);
 179
 180         cl_aligned_reloc(&job->indirect, uniforms, sview->bo,
 181                          v3d_tmu_config_data_get_value(data));
 182         v3d_job_add_bo(job, rsc->bo);
 183 }
 184
 185 static void
 186 write_image_tmu_p0(struct v3d_job *job,
 187                    struct v3d_cl_out **uniforms,
 188                    struct v3d_shaderimg_stateobj *img,
 189                    uint32_t data)
 190 {
 191         /* Extract the image unit from the top bits, and the compiler's
 192          * packed p0 from the bottom.
 193          */
 194         uint32_t unit = data >> 24;
 195         uint32_t p0 = data & 0x00ffffff;
 196
 197         struct v3d_image_view *iview = &img->si[unit];
 198         struct v3d_resource *rsc = v3d_resource(iview->base.resource);
 199
 200         cl_aligned_reloc(&job->indirect, uniforms,
 201                          v3d_resource(iview->tex_state)->bo,
 202                          iview->tex_state_offset | p0);
 203         v3d_job_add_bo(job, rsc->bo);
 204 }
 205
 206 /** Writes the V3D 4.x TMU configuration parameter 1. */
 207 static void
 208 write_tmu_p1(struct v3d_job *job,
 209              struct v3d_cl_out **uniforms,
 210              struct v3d_texture_stateobj *texstate,
 211              uint32_t data)
 212 {
 213         uint32_t unit = v3d_tmu_config_data_get_unit(data);
 214         struct pipe_sampler_state *psampler = texstate->samplers[unit];
 215         struct v3d_sampler_state *sampler = v3d_sampler_state(psampler);
 216
 217         cl_aligned_reloc(&job->indirect, uniforms, sampler->bo,
 218                          v3d_tmu_config_data_get_value(data));
 219 }
 220
 221 struct v3d_cl_reloc
 222 v3d_write_uniforms(struct v3d_context *v3d, struct v3d_compiled_shader *shader,
 223                    enum pipe_shader_type stage)
 224 {
 225         struct v3d_constbuf_stateobj *cb = &v3d->constbuf[stage];
 226         struct v3d_texture_stateobj *texstate = &v3d->tex[stage];
 227         struct v3d_uniform_list *uinfo = &shader->prog_data.base->uniforms;
 228         struct v3d_job *job = v3d->job;
 229         const uint32_t *gallium_uniforms = cb->cb[0].user_buffer;
 230         struct v3d_bo *ubo = v3d_upload_ubo(v3d, shader, gallium_uniforms);
 231
 232         /* We always need to return some space for uniforms, because the HW
 233          * will be prefetching, even if we don't read any in the program.
 234          */
 235         v3d_cl_ensure_space(&job->indirect, MAX2(uinfo->count, 1) * 4, 4);
 236
 237         struct v3d_cl_reloc uniform_stream = cl_get_address(&job->indirect);
 238         v3d_bo_reference(uniform_stream.bo);
 239
 240         struct v3d_cl_out *uniforms =
 241                 cl_start(&job->indirect);
 242
 243         for (int i = 0; i < uinfo->count; i++) {
 244                 uint32_t data = uinfo->data[i];
 245
 246                 switch (uinfo->contents[i]) {
 247                 case QUNIFORM_CONSTANT:
 248                         cl_aligned_u32(&uniforms, data);
 249                         break;
 250                 case QUNIFORM_UNIFORM:
 251                         cl_aligned_u32(&uniforms, gallium_uniforms[data]);
 252                         break;
 253                 case QUNIFORM_VIEWPORT_X_SCALE:
 254                         cl_aligned_f(&uniforms, v3d->viewport.scale[0] * 256.0f);
 255                         break;
 256                 case QUNIFORM_VIEWPORT_Y_SCALE:
 257                         cl_aligned_f(&uniforms, v3d->viewport.scale[1] * 256.0f);
 258                         break;
 259
 260                 case QUNIFORM_VIEWPORT_Z_OFFSET:
 261                         cl_aligned_f(&uniforms, v3d->viewport.translate[2]);
 262                         break;
 263                 case QUNIFORM_VIEWPORT_Z_SCALE:
 264                         cl_aligned_f(&uniforms, v3d->viewport.scale[2]);
 265                         break;
 266
 267                 case QUNIFORM_USER_CLIP_PLANE:
 268                         cl_aligned_f(&uniforms,
 269                                      v3d->clip.ucp[data / 4][data % 4]);
 270                         break;
 271
 272                 case QUNIFORM_TMU_CONFIG_P0:
 273                         write_tmu_p0(job, &uniforms, texstate, data);
 274                         break;
 275
 276                 case QUNIFORM_TMU_CONFIG_P1:
 277                         write_tmu_p1(job, &uniforms, texstate, data);
 278                         break;
 279
 280                 case QUNIFORM_IMAGE_TMU_CONFIG_P0:
 281                         write_image_tmu_p0(job, &uniforms,
 282                                            &v3d->shaderimg[stage], data);
 283                         break;
 284
 285                 case QUNIFORM_TEXTURE_CONFIG_P1:
 286                         write_texture_p1(job, &uniforms, texstate,
 287                                          data);
 288                         break;
 289
 290                 case QUNIFORM_TEXRECT_SCALE_X:
 291                 case QUNIFORM_TEXRECT_SCALE_Y:
 292                         cl_aligned_u32(&uniforms,
 293                                        get_texrect_scale(texstate,
 294                                                          uinfo->contents[i],
 295                                                          data));
 296                         break;
 297
 298                 case QUNIFORM_TEXTURE_WIDTH:
 299                 case QUNIFORM_TEXTURE_HEIGHT:
 300                 case QUNIFORM_TEXTURE_DEPTH:
 301                 case QUNIFORM_TEXTURE_ARRAY_SIZE:
 302                 case QUNIFORM_TEXTURE_LEVELS:
 303                         cl_aligned_u32(&uniforms,
 304                                        get_texture_size(texstate,
 305                                                         uinfo->contents[i],
 306                                                         data));
 307                         break;
 308
 309                 case QUNIFORM_IMAGE_WIDTH:
 310                 case QUNIFORM_IMAGE_HEIGHT:
 311                 case QUNIFORM_IMAGE_DEPTH:
 312                 case QUNIFORM_IMAGE_ARRAY_SIZE:
 313                         cl_aligned_u32(&uniforms,
 314                                        get_image_size(&v3d->shaderimg[stage],
 315                                                       uinfo->contents[i],
 316                                                       data));
 317                         break;
 318
 319                 case QUNIFORM_ALPHA_REF:
 320                         cl_aligned_f(&uniforms,
 321                                      v3d->zsa->base.alpha.ref_value);
 322                         break;
 323
 324                 case QUNIFORM_UBO_ADDR:
 325                         if (data == 0) {
 326                                 cl_aligned_reloc(&job->indirect, &uniforms,
 327                                                  ubo, 0);
 328                         } else {
 329                                 int ubo_index = data;
 330                                 struct v3d_resource *rsc =
 331                                         v3d_resource(cb->cb[ubo_index].buffer);
 332
 333                                 cl_aligned_reloc(&job->indirect, &uniforms,
 334                                                  rsc->bo,
 335                                                  cb->cb[ubo_index].buffer_offset);
 336                         }
 337                         break;
 338
 339                 case QUNIFORM_SSBO_OFFSET: {
 340                         struct pipe_shader_buffer *sb =
 341                                 &v3d->ssbo[stage].sb[data];
 342
 343                         cl_aligned_reloc(&job->indirect, &uniforms,
 344                                          v3d_resource(sb->buffer)->bo,
 345                                          sb->buffer_offset);
 346                         break;
 347                 }
 348
 349                 case QUNIFORM_GET_BUFFER_SIZE:
 350                         cl_aligned_u32(&uniforms,
 351                                        v3d->ssbo[stage].sb[data].buffer_size);
 352                         break;
 353
 354                 case QUNIFORM_TEXTURE_FIRST_LEVEL:
 355                         cl_aligned_f(&uniforms,
 356                                      texstate->textures[data]->u.tex.first_level);
 357                         break;
 358
 359                 case QUNIFORM_SPILL_OFFSET:
 360                         cl_aligned_reloc(&job->indirect, &uniforms,
 361                                          v3d->prog.spill_bo, 0);
 362                         break;
 363
 364                 case QUNIFORM_SPILL_SIZE_PER_THREAD:
 365                         cl_aligned_u32(&uniforms,
 366                                        v3d->prog.spill_size_per_thread);
 367                         break;
 368
 369                 default:
 370                         assert(quniform_contents_is_texture_p0(uinfo->contents[i]));
 371
 372                         write_texture_p0(job, &uniforms, texstate,
 373                                          uinfo->contents[i] -
 374                                          QUNIFORM_TEXTURE_CONFIG_P0_0,
 375                                          data);
 376                         break;
 377
 378                 }
 379 #if 0
 380                 uint32_t written_val = *((uint32_t *)uniforms - 1);
 381                 fprintf(stderr, "shader %p[%d]: 0x%08x / 0x%08x (%f) ",
 382                         shader, i, __gen_address_offset(&uniform_stream) + i * 4,
 383                         written_val, uif(written_val));
 384                 vir_dump_uniform(uinfo->contents[i], data);
 385                 fprintf(stderr, "\n");
 386 #endif
 387         }
 388
 389         cl_end(&job->indirect, uniforms);
 390
 391         v3d_bo_unreference(&ubo);
 392
 393         return uniform_stream;
 394 }
 395
 396 void
 397 v3d_set_shader_uniform_dirty_flags(struct v3d_compiled_shader *shader)
 398 {
 399         uint32_t dirty = 0;
 400
 401         for (int i = 0; i < shader->prog_data.base->uniforms.count; i++) {
 402                 switch (shader->prog_data.base->uniforms.contents[i]) {
 403                 case QUNIFORM_CONSTANT:
 404                         break;
 405                 case QUNIFORM_UNIFORM:
 406                 case QUNIFORM_UBO_ADDR:
 407                         dirty |= VC5_DIRTY_CONSTBUF;
 408                         break;
 409
 410                 case QUNIFORM_VIEWPORT_X_SCALE:
 411                 case QUNIFORM_VIEWPORT_Y_SCALE:
 412                 case QUNIFORM_VIEWPORT_Z_OFFSET:
 413                 case QUNIFORM_VIEWPORT_Z_SCALE:
 414                         dirty |= VC5_DIRTY_VIEWPORT;
 415                         break;
 416
 417                 case QUNIFORM_USER_CLIP_PLANE:
 418                         dirty |= VC5_DIRTY_CLIP;
 419                         break;
 420
 421                 case QUNIFORM_TMU_CONFIG_P0:
 422                 case QUNIFORM_TMU_CONFIG_P1:
 423                 case QUNIFORM_TEXTURE_CONFIG_P1:
 424                 case QUNIFORM_TEXTURE_FIRST_LEVEL:
 425                 case QUNIFORM_TEXRECT_SCALE_X:
 426                 case QUNIFORM_TEXRECT_SCALE_Y:
 427                 case QUNIFORM_TEXTURE_WIDTH:
 428                 case QUNIFORM_TEXTURE_HEIGHT:
 429                 case QUNIFORM_TEXTURE_DEPTH:
 430                 case QUNIFORM_TEXTURE_ARRAY_SIZE:
 431                 case QUNIFORM_TEXTURE_LEVELS:
 432                 case QUNIFORM_SPILL_OFFSET:
 433                 case QUNIFORM_SPILL_SIZE_PER_THREAD:
 434                         /* We could flag this on just the stage we're
 435                          * compiling for, but it's not passed in.
 436                          */
 437                         dirty |= VC5_DIRTY_FRAGTEX | VC5_DIRTY_VERTTEX;
 438                         break;
 439
 440                 case QUNIFORM_SSBO_OFFSET:
 441                 case QUNIFORM_GET_BUFFER_SIZE:
 442                         dirty |= VC5_DIRTY_SSBO;
 443                         break;
 444
 445                 case QUNIFORM_IMAGE_TMU_CONFIG_P0:
 446                 case QUNIFORM_IMAGE_WIDTH:
 447                 case QUNIFORM_IMAGE_HEIGHT:
 448                 case QUNIFORM_IMAGE_DEPTH:
 449                 case QUNIFORM_IMAGE_ARRAY_SIZE:
 450                         dirty |= VC5_DIRTY_SHADER_IMAGE;
 451                         break;
 452
 453                 case QUNIFORM_ALPHA_REF:
 454                         dirty |= VC5_DIRTY_ZSA;
 455                         break;
 456
 457                 default:
 458                         assert(quniform_contents_is_texture_p0(shader->prog_data.base->uniforms.contents[i]));
 459                         dirty |= VC5_DIRTY_FRAGTEX | VC5_DIRTY_VERTTEX;
 460                         break;
 461                 }
 462         }
 463
 464         shader->uniform_dirty_bits = dirty;
 465 }