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