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/u_upload_mgr.h"
  26 #include "util/format_srgb.h"
  27
  28 #include "v3d_context.h"
  29 #include "compiler/v3d_compiler.h"
  30 #include "broadcom/cle/v3d_packet_v33_pack.h"
  31
  32 static uint32_t
  33 get_texrect_scale(struct v3d_texture_stateobj *texstate,
  34                   enum quniform_contents contents,
  35                   uint32_t data)
  36 {
  37         struct pipe_sampler_view *texture = texstate->textures[data];
  38         uint32_t dim;
  39
  40         if (contents == QUNIFORM_TEXRECT_SCALE_X)
  41                 dim = texture->texture->width0;
  42         else
  43                 dim = texture->texture->height0;
  44
  45         return fui(1.0f / dim);
  46 }
  47
  48 static uint32_t
  49 get_texture_size(struct v3d_texture_stateobj *texstate,
  50                  enum quniform_contents contents,
  51                  uint32_t data)
  52 {
  53         struct pipe_sampler_view *texture = texstate->textures[data];
  54
  55         switch (contents) {
  56         case QUNIFORM_TEXTURE_WIDTH:
  57                 return u_minify(texture->texture->width0,
  58                                 texture->u.tex.first_level);
  59         case QUNIFORM_TEXTURE_HEIGHT:
  60                 return u_minify(texture->texture->height0,
  61                                 texture->u.tex.first_level);
  62         case QUNIFORM_TEXTURE_DEPTH:
  63                 return u_minify(texture->texture->depth0,
  64                                 texture->u.tex.first_level);
  65         case QUNIFORM_TEXTURE_ARRAY_SIZE:
  66                 return texture->texture->array_size;
  67         case QUNIFORM_TEXTURE_LEVELS:
  68                 return (texture->u.tex.last_level -
  69                         texture->u.tex.first_level) + 1;
  70         default:
  71                 unreachable("Bad texture size field");
  72         }
  73 }
  74
  75 static uint32_t
  76 get_image_size(struct v3d_shaderimg_stateobj *shaderimg,
  77                enum quniform_contents contents,
  78                uint32_t data)
  79 {
  80         struct v3d_image_view *image = &shaderimg->si[data];
  81
  82         switch (contents) {
  83         case QUNIFORM_IMAGE_WIDTH:
  84                 return u_minify(image->base.resource->width0,
  85                                 image->base.u.tex.level);
  86         case QUNIFORM_IMAGE_HEIGHT:
  87                 return u_minify(image->base.resource->height0,
  88                                 image->base.u.tex.level);
  89         case QUNIFORM_IMAGE_DEPTH:
  90                 return u_minify(image->base.resource->depth0,
  91                                 image->base.u.tex.level);
  92         case QUNIFORM_IMAGE_ARRAY_SIZE:
  93                 return image->base.resource->array_size;
  94         default:
  95                 unreachable("Bad texture size field");
  96         }
  97 }
  98
  99 /**
 100  *  Writes the V3D 3.x P0 (CFG_MODE=1) texture parameter.
 101  *
 102  * Some bits of this field are dependent on the type of sample being done by
 103  * the shader, while other bits are dependent on the sampler state.  We OR the
 104  * two together here.
 105  */
 106 static void
 107 write_texture_p0(struct v3d_job *job,
 108                  struct v3d_cl_out **uniforms,
 109                  struct v3d_texture_stateobj *texstate,
 110                  uint32_t unit,
 111                  uint32_t shader_data)
 112 {
 113         struct pipe_sampler_state *psampler = texstate->samplers[unit];
 114         struct v3d_sampler_state *sampler = v3d_sampler_state(psampler);
 115
 116         cl_aligned_u32(uniforms, shader_data | sampler->p0);
 117 }
 118
 119 /** Writes the V3D 3.x P1 (CFG_MODE=1) texture parameter. */
 120 static void
 121 write_texture_p1(struct v3d_job *job,
 122                  struct v3d_cl_out **uniforms,
 123                  struct v3d_texture_stateobj *texstate,
 124                  uint32_t data)
 125 {
 126         /* Extract the texture unit from the top bits, and the compiler's
 127          * packed p1 from the bottom.
 128          */
 129         uint32_t unit = data >> 5;
 130         uint32_t p1 = data & 0x1f;
 131
 132         struct pipe_sampler_view *psview = texstate->textures[unit];
 133         struct v3d_sampler_view *sview = v3d_sampler_view(psview);
 134
 135         struct V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1 unpacked = {
 136                 .texture_state_record_base_address = texstate->texture_state[unit],
 137         };
 138
 139         uint32_t packed;
 140         V3D33_TEXTURE_UNIFORM_PARAMETER_1_CFG_MODE1_pack(&job->indirect,
 141                                                          (uint8_t *)&packed,
 142                                                          &unpacked);
 143
 144         cl_aligned_u32(uniforms, p1 | packed | sview->p1);
 145 }
 146
 147 /** Writes the V3D 4.x TMU configuration parameter 0. */
 148 static void
 149 write_tmu_p0(struct v3d_job *job,
 150              struct v3d_cl_out **uniforms,
 151              struct v3d_texture_stateobj *texstate,
 152              uint32_t data)
 153 {
 154         int unit = v3d_unit_data_get_unit(data);
 155         struct pipe_sampler_view *psview = texstate->textures[unit];
 156         struct v3d_sampler_view *sview = v3d_sampler_view(psview);
 157         struct v3d_resource *rsc = v3d_resource(sview->texture);
 158
 159         cl_aligned_reloc(&job->indirect, uniforms, sview->bo,
 160                          v3d_unit_data_get_offset(data));
 161         v3d_job_add_bo(job, rsc->bo);
 162 }
 163
 164 static void
 165 write_image_tmu_p0(struct v3d_job *job,
 166                    struct v3d_cl_out **uniforms,
 167                    struct v3d_shaderimg_stateobj *img,
 168                    uint32_t data)
 169 {
 170         /* Extract the image unit from the top bits, and the compiler's
 171          * packed p0 from the bottom.
 172          */
 173         uint32_t unit = data >> 24;
 174         uint32_t p0 = data & 0x00ffffff;
 175
 176         struct v3d_image_view *iview = &img->si[unit];
 177         struct v3d_resource *rsc = v3d_resource(iview->base.resource);
 178
 179         cl_aligned_reloc(&job->indirect, uniforms,
 180                          v3d_resource(iview->tex_state)->bo,
 181                          iview->tex_state_offset | p0);
 182         v3d_job_add_bo(job, rsc->bo);
 183 }
 184
 185 /** Writes the V3D 4.x TMU configuration parameter 1. */
 186 static void
 187 write_tmu_p1(struct v3d_job *job,
 188              struct v3d_cl_out **uniforms,
 189              struct v3d_texture_stateobj *texstate,
 190              uint32_t data)
 191 {
 192         uint32_t unit = v3d_unit_data_get_unit(data);
 193         struct pipe_sampler_state *psampler = texstate->samplers[unit];
 194         struct v3d_sampler_state *sampler = v3d_sampler_state(psampler);
 195         struct pipe_sampler_view *psview = texstate->textures[unit];
 196         struct v3d_sampler_view *sview = v3d_sampler_view(psview);
 197         int variant = 0;
 198
 199         if (sampler->border_color_variants)
 200                 variant = sview->sampler_variant;
 201
 202         cl_aligned_reloc(&job->indirect, uniforms,
 203                          v3d_resource(sampler->sampler_state)->bo,
 204                          sampler->sampler_state_offset[variant] |
 205                          v3d_unit_data_get_offset(data));
 206 }
 207
 208 struct v3d_cl_reloc
 209 v3d_write_uniforms(struct v3d_context *v3d, struct v3d_compiled_shader *shader,
 210                    enum pipe_shader_type stage)
 211 {
 212         struct v3d_constbuf_stateobj *cb = &v3d->constbuf[stage];
 213         struct v3d_texture_stateobj *texstate = &v3d->tex[stage];
 214         struct v3d_uniform_list *uinfo = &shader->prog_data.base->uniforms;
 215         struct v3d_job *job = v3d->job;
 216         const uint32_t *gallium_uniforms = cb->cb[0].user_buffer;
 217
 218         /* We always need to return some space for uniforms, because the HW
 219          * will be prefetching, even if we don't read any in the program.
 220          */
 221         v3d_cl_ensure_space(&job->indirect, MAX2(uinfo->count, 1) * 4, 4);
 222
 223         struct v3d_cl_reloc uniform_stream = cl_get_address(&job->indirect);
 224         v3d_bo_reference(uniform_stream.bo);
 225
 226         struct v3d_cl_out *uniforms =
 227                 cl_start(&job->indirect);
 228
 229         for (int i = 0; i < uinfo->count; i++) {
 230                 uint32_t data = uinfo->data[i];
 231
 232                 switch (uinfo->contents[i]) {
 233                 case QUNIFORM_CONSTANT:
 234                         cl_aligned_u32(&uniforms, data);
 235                         break;
 236                 case QUNIFORM_UNIFORM:
 237                         cl_aligned_u32(&uniforms, gallium_uniforms[data]);
 238                         break;
 239                 case QUNIFORM_VIEWPORT_X_SCALE:
 240                         cl_aligned_f(&uniforms, v3d->viewport.scale[0] * 256.0f);
 241                         break;
 242                 case QUNIFORM_VIEWPORT_Y_SCALE:
 243                         cl_aligned_f(&uniforms, v3d->viewport.scale[1] * 256.0f);
 244                         break;
 245
 246                 case QUNIFORM_VIEWPORT_Z_OFFSET:
 247                         cl_aligned_f(&uniforms, v3d->viewport.translate[2]);
 248                         break;
 249                 case QUNIFORM_VIEWPORT_Z_SCALE:
 250                         cl_aligned_f(&uniforms, v3d->viewport.scale[2]);
 251                         break;
 252
 253                 case QUNIFORM_USER_CLIP_PLANE:
 254                         cl_aligned_f(&uniforms,
 255                                      v3d->clip.ucp[data / 4][data % 4]);
 256                         break;
 257
 258                 case QUNIFORM_TMU_CONFIG_P0:
 259                         write_tmu_p0(job, &uniforms, texstate, data);
 260                         break;
 261
 262                 case QUNIFORM_TMU_CONFIG_P1:
 263                         write_tmu_p1(job, &uniforms, texstate, data);
 264                         break;
 265
 266                 case QUNIFORM_IMAGE_TMU_CONFIG_P0:
 267                         write_image_tmu_p0(job, &uniforms,
 268                                            &v3d->shaderimg[stage], data);
 269                         break;
 270
 271                 case QUNIFORM_TEXTURE_CONFIG_P1:
 272                         write_texture_p1(job, &uniforms, texstate,
 273                                          data);
 274                         break;
 275
 276                 case QUNIFORM_TEXRECT_SCALE_X:
 277                 case QUNIFORM_TEXRECT_SCALE_Y:
 278                         cl_aligned_u32(&uniforms,
 279                                        get_texrect_scale(texstate,
 280                                                          uinfo->contents[i],
 281                                                          data));
 282                         break;
 283
 284                 case QUNIFORM_TEXTURE_WIDTH:
 285                 case QUNIFORM_TEXTURE_HEIGHT:
 286                 case QUNIFORM_TEXTURE_DEPTH:
 287                 case QUNIFORM_TEXTURE_ARRAY_SIZE:
 288                 case QUNIFORM_TEXTURE_LEVELS:
 289                         cl_aligned_u32(&uniforms,
 290                                        get_texture_size(texstate,
 291                                                         uinfo->contents[i],
 292                                                         data));
 293                         break;
 294
 295                 case QUNIFORM_IMAGE_WIDTH:
 296                 case QUNIFORM_IMAGE_HEIGHT:
 297                 case QUNIFORM_IMAGE_DEPTH:
 298                 case QUNIFORM_IMAGE_ARRAY_SIZE:
 299                         cl_aligned_u32(&uniforms,
 300                                        get_image_size(&v3d->shaderimg[stage],
 301                                                       uinfo->contents[i],
 302                                                       data));
 303                         break;
 304
 305                 case QUNIFORM_ALPHA_REF:
 306                         cl_aligned_f(&uniforms,
 307                                      v3d->zsa->base.alpha.ref_value);
 308                         break;
 309
 310                 case QUNIFORM_UBO_ADDR: {
 311                         uint32_t unit = v3d_unit_data_get_unit(data);
 312                         /* Constant buffer 0 may be a system memory pointer,
 313                          * in which case we want to upload a shadow copy to
 314                          * the GPU.
 315                         */
 316                         if (!cb->cb[unit].buffer) {
 317                                 u_upload_data(v3d->uploader, 0,
 318                                               cb->cb[unit].buffer_size, 16,
 319                                               cb->cb[unit].user_buffer,
 320                                               &cb->cb[unit].buffer_offset,
 321                                               &cb->cb[unit].buffer);
 322                         }
 323
 324                         cl_aligned_reloc(&job->indirect, &uniforms,
 325                                          v3d_resource(cb->cb[unit].buffer)->bo,
 326                                          cb->cb[unit].buffer_offset +
 327                                          v3d_unit_data_get_offset(data));
 328                         break;
 329                 }
 330
 331                 case QUNIFORM_SSBO_OFFSET: {
 332                         struct pipe_shader_buffer *sb =
 333                                 &v3d->ssbo[stage].sb[data];
 334
 335                         cl_aligned_reloc(&job->indirect, &uniforms,
 336                                          v3d_resource(sb->buffer)->bo,
 337                                          sb->buffer_offset);
 338                         break;
 339                 }
 340
 341                 case QUNIFORM_GET_BUFFER_SIZE:
 342                         cl_aligned_u32(&uniforms,
 343                                        v3d->ssbo[stage].sb[data].buffer_size);
 344                         break;
 345
 346                 case QUNIFORM_TEXTURE_FIRST_LEVEL:
 347                         cl_aligned_f(&uniforms,
 348                                      texstate->textures[data]->u.tex.first_level);
 349                         break;
 350
 351                 case QUNIFORM_SPILL_OFFSET:
 352                         cl_aligned_reloc(&job->indirect, &uniforms,
 353                                          v3d->prog.spill_bo, 0);
 354                         break;
 355
 356                 case QUNIFORM_SPILL_SIZE_PER_THREAD:
 357                         cl_aligned_u32(&uniforms,
 358                                        v3d->prog.spill_size_per_thread);
 359                         break;
 360
 361                 case QUNIFORM_NUM_WORK_GROUPS:
 362                         cl_aligned_u32(&uniforms,
 363                                        v3d->compute_num_workgroups[data]);
 364                         break;
 365
 366                 case QUNIFORM_SHARED_OFFSET:
 367                         cl_aligned_reloc(&job->indirect, &uniforms,
 368                                          v3d->compute_shared_memory, 0);
 369                         break;
 370
 371                 default:
 372                         assert(quniform_contents_is_texture_p0(uinfo->contents[i]));
 373
 374                         write_texture_p0(job, &uniforms, texstate,
 375                                          uinfo->contents[i] -
 376                                          QUNIFORM_TEXTURE_CONFIG_P0_0,
 377                                          data);
 378                         break;
 379
 380                 }
 381 #if 0
 382                 uint32_t written_val = *((uint32_t *)uniforms - 1);
 383                 fprintf(stderr, "shader %p[%d]: 0x%08x / 0x%08x (%f) ",
 384                         shader, i, __gen_address_offset(&uniform_stream) + i * 4,
 385                         written_val, uif(written_val));
 386                 vir_dump_uniform(uinfo->contents[i], data);
 387                 fprintf(stderr, "\n");
 388 #endif
 389         }
 390
 391         cl_end(&job->indirect, uniforms);
 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                            VC5_DIRTY_COMPTEX;
 439                         break;
 440
 441                 case QUNIFORM_SSBO_OFFSET:
 442                 case QUNIFORM_GET_BUFFER_SIZE:
 443                         dirty |= VC5_DIRTY_SSBO;
 444                         break;
 445
 446                 case QUNIFORM_IMAGE_TMU_CONFIG_P0:
 447                 case QUNIFORM_IMAGE_WIDTH:
 448                 case QUNIFORM_IMAGE_HEIGHT:
 449                 case QUNIFORM_IMAGE_DEPTH:
 450                 case QUNIFORM_IMAGE_ARRAY_SIZE:
 451                         dirty |= VC5_DIRTY_SHADER_IMAGE;
 452                         break;
 453
 454                 case QUNIFORM_ALPHA_REF:
 455                         dirty |= VC5_DIRTY_ZSA;
 456                         break;
 457
 458                 case QUNIFORM_NUM_WORK_GROUPS:
 459                 case QUNIFORM_SHARED_OFFSET:
 460                         /* Compute always recalculates uniforms. */
 461                         break;
 462
 463                 default:
 464                         assert(quniform_contents_is_texture_p0(shader->prog_data.base->uniforms.contents[i]));
 465                         dirty |= VC5_DIRTY_FRAGTEX | VC5_DIRTY_VERTTEX |
 466                            VC5_DIRTY_COMPTEX;
 467                         break;
 468                 }
 469         }
 470
 471         shader->uniform_dirty_bits = dirty;
 472 }