src/gallium/drivers/vc4/vc4_uniforms.c

   1 /*
   2  * Copyright © 2014-2015 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 "vc4_context.h"
  28 #include "vc4_qir.h"
  29
  30 static void
  31 write_texture_p0(struct vc4_job *job,
  32                  struct vc4_cl_out **uniforms,
  33                  struct vc4_texture_stateobj *texstate,
  34                  uint32_t unit)
  35 {
  36         struct vc4_sampler_view *sview =
  37                 vc4_sampler_view(texstate->textures[unit]);
  38         struct vc4_resource *rsc = vc4_resource(sview->texture);
  39
  40         cl_reloc(job, &job->uniforms, uniforms, rsc->bo, sview->texture_p0);
  41 }
  42
  43 static void
  44 write_texture_p1(struct vc4_job *job,
  45                  struct vc4_cl_out **uniforms,
  46                  struct vc4_texture_stateobj *texstate,
  47                  uint32_t unit)
  48 {
  49         struct vc4_sampler_view *sview =
  50                 vc4_sampler_view(texstate->textures[unit]);
  51         struct vc4_sampler_state *sampler =
  52                 vc4_sampler_state(texstate->samplers[unit]);
  53
  54         cl_aligned_u32(uniforms, sview->texture_p1 | sampler->texture_p1);
  55 }
  56
  57 static void
  58 write_texture_p2(struct vc4_job *job,
  59                  struct vc4_cl_out **uniforms,
  60                  struct vc4_texture_stateobj *texstate,
  61                  uint32_t data)
  62 {
  63         uint32_t unit = data & 0xffff;
  64         struct pipe_sampler_view *texture = texstate->textures[unit];
  65         struct vc4_resource *rsc = vc4_resource(texture->texture);
  66
  67         cl_aligned_u32(uniforms,
  68                VC4_SET_FIELD(VC4_TEX_P2_PTYPE_CUBE_MAP_STRIDE,
  69                              VC4_TEX_P2_PTYPE) |
  70                VC4_SET_FIELD(rsc->cube_map_stride >> 12, VC4_TEX_P2_CMST) |
  71                VC4_SET_FIELD((data >> 16) & 1, VC4_TEX_P2_BSLOD));
  72 }
  73
  74 static void
  75 write_texture_first_level(struct vc4_job *job,
  76                           struct vc4_cl_out **uniforms,
  77                           struct vc4_texture_stateobj *texstate,
  78                           uint32_t data)
  79 {
  80         uint32_t unit = data & 0xffff;
  81         struct pipe_sampler_view *texture = texstate->textures[unit];
  82
  83         cl_aligned_f(uniforms, texture->u.tex.first_level);
  84 }
  85
  86 static void
  87 write_texture_msaa_addr(struct vc4_job *job,
  88                  struct vc4_cl_out **uniforms,
  89                         struct vc4_texture_stateobj *texstate,
  90                         uint32_t unit)
  91 {
  92         struct pipe_sampler_view *texture = texstate->textures[unit];
  93         struct vc4_resource *rsc = vc4_resource(texture->texture);
  94
  95         cl_aligned_reloc(job, &job->uniforms, uniforms, rsc->bo, 0);
  96 }
  97
  98
  99 #define SWIZ(x,y,z,w) {          \
 100         PIPE_SWIZZLE_##x, \
 101         PIPE_SWIZZLE_##y, \
 102         PIPE_SWIZZLE_##z, \
 103         PIPE_SWIZZLE_##w  \
 104 }
 105
 106 static void
 107 write_texture_border_color(struct vc4_job *job,
 108                            struct vc4_cl_out **uniforms,
 109                            struct vc4_texture_stateobj *texstate,
 110                            uint32_t unit)
 111 {
 112         struct pipe_sampler_state *sampler = texstate->samplers[unit];
 113         struct pipe_sampler_view *texture = texstate->textures[unit];
 114         struct vc4_resource *rsc = vc4_resource(texture->texture);
 115         union util_color uc;
 116
 117         const struct util_format_description *tex_format_desc =
 118                 util_format_description(texture->format);
 119
 120         float border_color[4];
 121         for (int i = 0; i < 4; i++)
 122                 border_color[i] = sampler->border_color.f[i];
 123         if (util_format_is_srgb(texture->format)) {
 124                 for (int i = 0; i < 3; i++)
 125                         border_color[i] =
 126                                 util_format_linear_to_srgb_float(border_color[i]);
 127         }
 128
 129         /* Turn the border color into the layout of channels that it would
 130          * have when stored as texture contents.
 131          */
 132         float storage_color[4];
 133         util_format_unswizzle_4f(storage_color,
 134                                  border_color,
 135                                  tex_format_desc->swizzle);
 136
 137         /* Now, pack so that when the vc4_format-sampled texture contents are
 138          * replaced with our border color, the vc4_get_format_swizzle()
 139          * swizzling will get the right channels.
 140          */
 141         if (util_format_is_depth_or_stencil(texture->format)) {
 142                 uc.ui[0] = util_pack_z(PIPE_FORMAT_Z24X8_UNORM,
 143                                        sampler->border_color.f[0]) << 8;
 144         } else {
 145                 switch (rsc->vc4_format) {
 146                 default:
 147                 case VC4_TEXTURE_TYPE_RGBA8888:
 148                         util_pack_color(storage_color,
 149                                         PIPE_FORMAT_R8G8B8A8_UNORM, &uc);
 150                         break;
 151                 case VC4_TEXTURE_TYPE_RGBA4444:
 152                 case VC4_TEXTURE_TYPE_RGBA5551:
 153                         util_pack_color(storage_color,
 154                                         PIPE_FORMAT_A8B8G8R8_UNORM, &uc);
 155                         break;
 156                 case VC4_TEXTURE_TYPE_RGB565:
 157                         util_pack_color(storage_color,
 158                                         PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
 159                         break;
 160                 case VC4_TEXTURE_TYPE_ALPHA:
 161                         uc.ui[0] = float_to_ubyte(storage_color[0]) << 24;
 162                         break;
 163                 case VC4_TEXTURE_TYPE_LUMALPHA:
 164                         uc.ui[0] = ((float_to_ubyte(storage_color[1]) << 24) |
 165                                     (float_to_ubyte(storage_color[0]) << 0));
 166                         break;
 167                 }
 168         }
 169
 170         cl_aligned_u32(uniforms, uc.ui[0]);
 171 }
 172
 173 static uint32_t
 174 get_texrect_scale(struct vc4_texture_stateobj *texstate,
 175                   enum quniform_contents contents,
 176                   uint32_t data)
 177 {
 178         struct pipe_sampler_view *texture = texstate->textures[data];
 179         uint32_t dim;
 180
 181         if (contents == QUNIFORM_TEXRECT_SCALE_X)
 182                 dim = texture->texture->width0;
 183         else
 184                 dim = texture->texture->height0;
 185
 186         return fui(1.0f / dim);
 187 }
 188
 189 static struct vc4_bo *
 190 vc4_upload_ubo(struct vc4_context *vc4,
 191                struct vc4_compiled_shader *shader,
 192                const uint32_t *gallium_uniforms)
 193 {
 194         if (!shader->ubo_size)
 195                 return NULL;
 196
 197         struct vc4_bo *ubo = vc4_bo_alloc(vc4->screen, shader->ubo_size, "ubo");
 198         void *data = vc4_bo_map(ubo);
 199         for (uint32_t i = 0; i < shader->num_ubo_ranges; i++) {
 200                 memcpy(data + shader->ubo_ranges[i].dst_offset,
 201                        ((const void *)gallium_uniforms +
 202                         shader->ubo_ranges[i].src_offset),
 203                        shader->ubo_ranges[i].size);
 204         }
 205
 206         return ubo;
 207 }
 208
 209 void
 210 vc4_write_uniforms(struct vc4_context *vc4, struct vc4_compiled_shader *shader,
 211                    struct vc4_constbuf_stateobj *cb,
 212                    struct vc4_texture_stateobj *texstate)
 213 {
 214         struct vc4_shader_uniform_info *uinfo = &shader->uniforms;
 215         struct vc4_job *job = vc4->job;
 216         const uint32_t *gallium_uniforms = cb->cb[0].user_buffer;
 217         struct vc4_bo *ubo = vc4_upload_ubo(vc4, shader, gallium_uniforms);
 218
 219         cl_ensure_space(&job->uniforms, (uinfo->count +
 220                                          uinfo->num_texture_samples) * 4);
 221
 222         struct vc4_cl_out *uniforms =
 223                 cl_start_shader_reloc(&job->uniforms,
 224                                       uinfo->num_texture_samples);
 225
 226         for (int i = 0; i < uinfo->count; i++) {
 227
 228                 switch (uinfo->contents[i]) {
 229                 case QUNIFORM_CONSTANT:
 230                         cl_aligned_u32(&uniforms, uinfo->data[i]);
 231                         break;
 232                 case QUNIFORM_UNIFORM:
 233                         cl_aligned_u32(&uniforms,
 234                                        gallium_uniforms[uinfo->data[i]]);
 235                         break;
 236                 case QUNIFORM_VIEWPORT_X_SCALE:
 237                         cl_aligned_f(&uniforms, vc4->viewport.scale[0] * 16.0f);
 238                         break;
 239                 case QUNIFORM_VIEWPORT_Y_SCALE:
 240                         cl_aligned_f(&uniforms, vc4->viewport.scale[1] * 16.0f);
 241                         break;
 242
 243                 case QUNIFORM_VIEWPORT_Z_OFFSET:
 244                         cl_aligned_f(&uniforms, vc4->viewport.translate[2]);
 245                         break;
 246                 case QUNIFORM_VIEWPORT_Z_SCALE:
 247                         cl_aligned_f(&uniforms, vc4->viewport.scale[2]);
 248                         break;
 249
 250                 case QUNIFORM_USER_CLIP_PLANE:
 251                         cl_aligned_f(&uniforms,
 252                                      vc4->clip.ucp[uinfo->data[i] / 4][uinfo->data[i] % 4]);
 253                         break;
 254
 255                 case QUNIFORM_TEXTURE_CONFIG_P0:
 256                         write_texture_p0(job, &uniforms, texstate,
 257                                          uinfo->data[i]);
 258                         break;
 259
 260                 case QUNIFORM_TEXTURE_CONFIG_P1:
 261                         write_texture_p1(job, &uniforms, texstate,
 262                                          uinfo->data[i]);
 263                         break;
 264
 265                 case QUNIFORM_TEXTURE_CONFIG_P2:
 266                         write_texture_p2(job, &uniforms, texstate,
 267                                          uinfo->data[i]);
 268                         break;
 269
 270                 case QUNIFORM_TEXTURE_FIRST_LEVEL:
 271                         write_texture_first_level(job, &uniforms, texstate,
 272                                                   uinfo->data[i]);
 273                         break;
 274
 275                 case QUNIFORM_UBO_ADDR:
 276                         if (uinfo->data[i] == 0) {
 277                                 cl_aligned_reloc(job, &job->uniforms,
 278                                                  &uniforms, ubo, 0);
 279                         } else {
 280                                 struct pipe_constant_buffer *c =
 281                                         &cb->cb[uinfo->data[i]];
 282                                 struct vc4_resource *rsc =
 283                                         vc4_resource(c->buffer);
 284
 285                                 cl_aligned_reloc(job, &job->uniforms,
 286                                                  &uniforms,
 287                                                  rsc->bo, c->buffer_offset);
 288                         }
 289                         break;
 290
 291                 case QUNIFORM_TEXTURE_MSAA_ADDR:
 292                         write_texture_msaa_addr(job, &uniforms,
 293                                                 texstate, uinfo->data[i]);
 294                         break;
 295
 296                 case QUNIFORM_TEXTURE_BORDER_COLOR:
 297                         write_texture_border_color(job, &uniforms,
 298                                                    texstate, uinfo->data[i]);
 299                         break;
 300
 301                 case QUNIFORM_TEXRECT_SCALE_X:
 302                 case QUNIFORM_TEXRECT_SCALE_Y:
 303                         cl_aligned_u32(&uniforms,
 304                                        get_texrect_scale(texstate,
 305                                                          uinfo->contents[i],
 306                                                          uinfo->data[i]));
 307                         break;
 308
 309                 case QUNIFORM_BLEND_CONST_COLOR_X:
 310                 case QUNIFORM_BLEND_CONST_COLOR_Y:
 311                 case QUNIFORM_BLEND_CONST_COLOR_Z:
 312                 case QUNIFORM_BLEND_CONST_COLOR_W:
 313                         cl_aligned_f(&uniforms,
 314                                      CLAMP(vc4->blend_color.f.color[uinfo->contents[i] -
 315                                                                     QUNIFORM_BLEND_CONST_COLOR_X],
 316                                            0, 1));
 317                         break;
 318
 319                 case QUNIFORM_BLEND_CONST_COLOR_RGBA: {
 320                         const uint8_t *format_swiz =
 321                                 vc4_get_format_swizzle(vc4->framebuffer.cbufs[0]->format);
 322                         uint32_t color = 0;
 323                         for (int i = 0; i < 4; i++) {
 324                                 if (format_swiz[i] >= 4)
 325                                         continue;
 326
 327                                 color |= (vc4->blend_color.ub[format_swiz[i]] <<
 328                                           (i * 8));
 329                         }
 330                         cl_aligned_u32(&uniforms, color);
 331                         break;
 332                 }
 333
 334                 case QUNIFORM_BLEND_CONST_COLOR_AAAA: {
 335                         uint8_t a = vc4->blend_color.ub[3];
 336                         cl_aligned_u32(&uniforms, ((a) |
 337                                                    (a << 8) |
 338                                                    (a << 16) |
 339                                                    (a << 24)));
 340                         break;
 341                 }
 342
 343                 case QUNIFORM_STENCIL:
 344                         cl_aligned_u32(&uniforms,
 345                                        vc4->zsa->stencil_uniforms[uinfo->data[i]] |
 346                                        (uinfo->data[i] <= 1 ?
 347                                         (vc4->stencil_ref.ref_value[uinfo->data[i]] << 8) :
 348                                         0));
 349                         break;
 350
 351                 case QUNIFORM_ALPHA_REF:
 352                         cl_aligned_f(&uniforms,
 353                                      vc4->zsa->base.alpha.ref_value);
 354                         break;
 355
 356                 case QUNIFORM_SAMPLE_MASK:
 357                         cl_aligned_u32(&uniforms, vc4->sample_mask);
 358                         break;
 359
 360                 case QUNIFORM_UNIFORMS_ADDRESS:
 361                         /* This will be filled in by the kernel. */
 362                         cl_aligned_u32(&uniforms, 0xd0d0d0d0);
 363                         break;
 364                 }
 365 #if 0
 366                 uint32_t written_val = *((uint32_t *)uniforms - 1);
 367                 fprintf(stderr, "%p: %d / 0x%08x (%f)\n",
 368                         shader, i, written_val, uif(written_val));
 369 #endif
 370         }
 371
 372         cl_end(&job->uniforms, uniforms);
 373
 374         vc4_bo_unreference(&ubo);
 375 }
 376
 377 void
 378 vc4_set_shader_uniform_dirty_flags(struct vc4_compiled_shader *shader)
 379 {
 380         uint32_t dirty = 0;
 381
 382         for (int i = 0; i < shader->uniforms.count; i++) {
 383                 switch (shader->uniforms.contents[i]) {
 384                 case QUNIFORM_CONSTANT:
 385                 case QUNIFORM_UNIFORMS_ADDRESS:
 386                         break;
 387                 case QUNIFORM_UNIFORM:
 388                 case QUNIFORM_UBO_ADDR:
 389                         dirty |= VC4_DIRTY_CONSTBUF;
 390                         break;
 391
 392                 case QUNIFORM_VIEWPORT_X_SCALE:
 393                 case QUNIFORM_VIEWPORT_Y_SCALE:
 394                 case QUNIFORM_VIEWPORT_Z_OFFSET:
 395                 case QUNIFORM_VIEWPORT_Z_SCALE:
 396                         dirty |= VC4_DIRTY_VIEWPORT;
 397                         break;
 398
 399                 case QUNIFORM_USER_CLIP_PLANE:
 400                         dirty |= VC4_DIRTY_CLIP;
 401                         break;
 402
 403                 case QUNIFORM_TEXTURE_CONFIG_P0:
 404                 case QUNIFORM_TEXTURE_CONFIG_P1:
 405                 case QUNIFORM_TEXTURE_CONFIG_P2:
 406                 case QUNIFORM_TEXTURE_BORDER_COLOR:
 407                 case QUNIFORM_TEXTURE_FIRST_LEVEL:
 408                 case QUNIFORM_TEXTURE_MSAA_ADDR:
 409                 case QUNIFORM_TEXRECT_SCALE_X:
 410                 case QUNIFORM_TEXRECT_SCALE_Y:
 411                         /* We could flag this on just the stage we're
 412                          * compiling for, but it's not passed in.
 413                          */
 414                         dirty |= VC4_DIRTY_FRAGTEX | VC4_DIRTY_VERTTEX;
 415                         break;
 416
 417                 case QUNIFORM_BLEND_CONST_COLOR_X:
 418                 case QUNIFORM_BLEND_CONST_COLOR_Y:
 419                 case QUNIFORM_BLEND_CONST_COLOR_Z:
 420                 case QUNIFORM_BLEND_CONST_COLOR_W:
 421                 case QUNIFORM_BLEND_CONST_COLOR_RGBA:
 422                 case QUNIFORM_BLEND_CONST_COLOR_AAAA:
 423                         dirty |= VC4_DIRTY_BLEND_COLOR;
 424                         break;
 425
 426                 case QUNIFORM_STENCIL:
 427                 case QUNIFORM_ALPHA_REF:
 428                         dirty |= VC4_DIRTY_ZSA;
 429                         break;
 430
 431                 case QUNIFORM_SAMPLE_MASK:
 432                         dirty |= VC4_DIRTY_SAMPLE_MASK;
 433                         break;
 434                 }
 435         }
 436
 437         shader->uniform_dirty_bits = dirty;
 438 }