src/gallium/drivers/v3d/v3d_blit.c

   1 /*
   2  * Copyright © 2015-2017 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_format.h"
  25 #include "util/u_surface.h"
  26 #include "util/u_blitter.h"
  27 #include "v3d_context.h"
  28 #include "v3d_tiling.h"
  29
  30 #if 0
  31 static struct pipe_surface *
  32 v3d_get_blit_surface(struct pipe_context *pctx,
  33                      struct pipe_resource *prsc, unsigned level)
  34 {
  35         struct pipe_surface tmpl;
  36
  37         memset(&tmpl, 0, sizeof(tmpl));
  38         tmpl.format = prsc->format;
  39         tmpl.u.tex.level = level;
  40         tmpl.u.tex.first_layer = 0;
  41         tmpl.u.tex.last_layer = 0;
  42
  43         return pctx->create_surface(pctx, prsc, &tmpl);
  44 }
  45
  46 static bool
  47 is_tile_unaligned(unsigned size, unsigned tile_size)
  48 {
  49         return size & (tile_size - 1);
  50 }
  51
  52 static bool
  53 v3d_tile_blit(struct pipe_context *pctx, const struct pipe_blit_info *info)
  54 {
  55         struct v3d_context *v3d = v3d_context(pctx);
  56         bool msaa = (info->src.resource->nr_samples > 1 ||
  57                      info->dst.resource->nr_samples > 1);
  58         int tile_width = msaa ? 32 : 64;
  59         int tile_height = msaa ? 32 : 64;
  60
  61         if (util_format_is_depth_or_stencil(info->dst.resource->format))
  62                 return false;
  63
  64         if (info->scissor_enable)
  65                 return false;
  66
  67         if ((info->mask & PIPE_MASK_RGBA) == 0)
  68                 return false;
  69
  70         if (info->dst.box.x != info->src.box.x ||
  71             info->dst.box.y != info->src.box.y ||
  72             info->dst.box.width != info->src.box.width ||
  73             info->dst.box.height != info->src.box.height) {
  74                 return false;
  75         }
  76
  77         int dst_surface_width = u_minify(info->dst.resource->width0,
  78                                          info->dst.level);
  79         int dst_surface_height = u_minify(info->dst.resource->height0,
  80                                          info->dst.level);
  81         if (is_tile_unaligned(info->dst.box.x, tile_width) ||
  82             is_tile_unaligned(info->dst.box.y, tile_height) ||
  83             (is_tile_unaligned(info->dst.box.width, tile_width) &&
  84              info->dst.box.x + info->dst.box.width != dst_surface_width) ||
  85             (is_tile_unaligned(info->dst.box.height, tile_height) &&
  86              info->dst.box.y + info->dst.box.height != dst_surface_height)) {
  87                 return false;
  88         }
  89
  90         /* VC5_PACKET_LOAD_TILE_BUFFER_GENERAL uses the
  91          * VC5_PACKET_TILE_RENDERING_MODE_CONFIG's width (determined by our
  92          * destination surface) to determine the stride.  This may be wrong
  93          * when reading from texture miplevels > 0, which are stored in
  94          * POT-sized areas.  For MSAA, the tile addresses are computed
  95          * explicitly by the RCL, but still use the destination width to
  96          * determine the stride (which could be fixed by explicitly supplying
  97          * it in the ABI).
  98          */
  99         struct v3d_resource *rsc = v3d_resource(info->src.resource);
 100
 101         uint32_t stride;
 102
 103         if (info->src.resource->nr_samples > 1)
 104                 stride = align(dst_surface_width, 32) * 4 * rsc->cpp;
 105         /* XXX else if (rsc->slices[info->src.level].tiling == VC5_TILING_FORMAT_T)
 106            stride = align(dst_surface_width * rsc->cpp, 128); */
 107         else
 108                 stride = align(dst_surface_width * rsc->cpp, 16);
 109
 110         if (stride != rsc->slices[info->src.level].stride)
 111                 return false;
 112
 113         if (info->dst.resource->format != info->src.resource->format)
 114                 return false;
 115
 116         if (false) {
 117                 fprintf(stderr, "RCL blit from %d,%d to %d,%d (%d,%d)\n",
 118                         info->src.box.x,
 119                         info->src.box.y,
 120                         info->dst.box.x,
 121                         info->dst.box.y,
 122                         info->dst.box.width,
 123                         info->dst.box.height);
 124         }
 125
 126         struct pipe_surface *dst_surf =
 127                 v3d_get_blit_surface(pctx, info->dst.resource, info->dst.level);
 128         struct pipe_surface *src_surf =
 129                 v3d_get_blit_surface(pctx, info->src.resource, info->src.level);
 130
 131         v3d_flush_jobs_reading_resource(v3d, info->src.resource);
 132
 133         struct v3d_job *job = v3d_get_job(v3d, dst_surf, NULL);
 134         pipe_surface_reference(&job->color_read, src_surf);
 135
 136         /* If we're resolving from MSAA to single sample, we still need to run
 137          * the engine in MSAA mode for the load.
 138          */
 139         if (!job->msaa && info->src.resource->nr_samples > 1) {
 140                 job->msaa = true;
 141                 job->tile_width = 32;
 142                 job->tile_height = 32;
 143         }
 144
 145         job->draw_min_x = info->dst.box.x;
 146         job->draw_min_y = info->dst.box.y;
 147         job->draw_max_x = info->dst.box.x + info->dst.box.width;
 148         job->draw_max_y = info->dst.box.y + info->dst.box.height;
 149         job->draw_width = dst_surf->width;
 150         job->draw_height = dst_surf->height;
 151
 152         job->tile_width = tile_width;
 153         job->tile_height = tile_height;
 154         job->msaa = msaa;
 155         job->needs_flush = true;
 156         job->resolve |= PIPE_CLEAR_COLOR;
 157
 158         v3d_job_submit(v3d, job);
 159
 160         pipe_surface_reference(&dst_surf, NULL);
 161         pipe_surface_reference(&src_surf, NULL);
 162
 163         return true;
 164 }
 165 #endif
 166
 167 void
 168 v3d_blitter_save(struct v3d_context *v3d)
 169 {
 170         util_blitter_save_fragment_constant_buffer_slot(v3d->blitter,
 171                                                         v3d->constbuf[PIPE_SHADER_FRAGMENT].cb);
 172         util_blitter_save_vertex_buffer_slot(v3d->blitter, v3d->vertexbuf.vb);
 173         util_blitter_save_vertex_elements(v3d->blitter, v3d->vtx);
 174         util_blitter_save_vertex_shader(v3d->blitter, v3d->prog.bind_vs);
 175         util_blitter_save_so_targets(v3d->blitter, v3d->streamout.num_targets,
 176                                      v3d->streamout.targets);
 177         util_blitter_save_rasterizer(v3d->blitter, v3d->rasterizer);
 178         util_blitter_save_viewport(v3d->blitter, &v3d->viewport);
 179         util_blitter_save_scissor(v3d->blitter, &v3d->scissor);
 180         util_blitter_save_fragment_shader(v3d->blitter, v3d->prog.bind_fs);
 181         util_blitter_save_blend(v3d->blitter, v3d->blend);
 182         util_blitter_save_depth_stencil_alpha(v3d->blitter, v3d->zsa);
 183         util_blitter_save_stencil_ref(v3d->blitter, &v3d->stencil_ref);
 184         util_blitter_save_sample_mask(v3d->blitter, v3d->sample_mask);
 185         util_blitter_save_framebuffer(v3d->blitter, &v3d->framebuffer);
 186         util_blitter_save_fragment_sampler_states(v3d->blitter,
 187                         v3d->fragtex.num_samplers,
 188                         (void **)v3d->fragtex.samplers);
 189         util_blitter_save_fragment_sampler_views(v3d->blitter,
 190                         v3d->fragtex.num_textures, v3d->fragtex.textures);
 191         util_blitter_save_so_targets(v3d->blitter, v3d->streamout.num_targets,
 192                                      v3d->streamout.targets);
 193 }
 194
 195 static bool
 196 v3d_render_blit(struct pipe_context *ctx, struct pipe_blit_info *info)
 197 {
 198         struct v3d_context *v3d = v3d_context(ctx);
 199         struct v3d_resource *src = v3d_resource(info->src.resource);
 200         struct pipe_resource *tiled = NULL;
 201
 202         if (!src->tiled) {
 203                 struct pipe_box box = {
 204                         .x = 0,
 205                         .y = 0,
 206                         .width = u_minify(info->src.resource->width0,
 207                                            info->src.level),
 208                         .height = u_minify(info->src.resource->height0,
 209                                            info->src.level),
 210                         .depth = 1,
 211                 };
 212                 struct pipe_resource tmpl = {
 213                         .target = info->src.resource->target,
 214                         .format = info->src.resource->format,
 215                         .width0 = box.width,
 216                         .height0 = box.height,
 217                         .depth0 = 1,
 218                         .array_size = 1,
 219                 };
 220                 tiled = ctx->screen->resource_create(ctx->screen, &tmpl);
 221                 if (!tiled) {
 222                         fprintf(stderr, "Failed to create tiled blit temp\n");
 223                         return false;
 224                 }
 225                 ctx->resource_copy_region(ctx,
 226                                           tiled, 0,
 227                                           0, 0, 0,
 228                                           info->src.resource, info->src.level,
 229                                           &box);
 230                 info->src.level = 0;
 231                 info->src.resource = tiled;
 232         }
 233
 234         if (!util_blitter_is_blit_supported(v3d->blitter, info)) {
 235                 fprintf(stderr, "blit unsupported %s -> %s\n",
 236                     util_format_short_name(info->src.resource->format),
 237                     util_format_short_name(info->dst.resource->format));
 238                 return false;
 239         }
 240
 241         v3d_blitter_save(v3d);
 242         util_blitter_blit(v3d->blitter, info);
 243
 244         pipe_resource_reference(&tiled, NULL);
 245
 246         return true;
 247 }
 248
 249 /* Implement stencil blits by reinterpreting the stencil data as an RGBA8888
 250  * or R8 texture.
 251  */
 252 static void
 253 v3d_stencil_blit(struct pipe_context *ctx, const struct pipe_blit_info *info)
 254 {
 255         struct v3d_context *v3d = v3d_context(ctx);
 256         struct v3d_resource *src = v3d_resource(info->src.resource);
 257         struct v3d_resource *dst = v3d_resource(info->dst.resource);
 258         enum pipe_format src_format, dst_format;
 259
 260         if (src->separate_stencil) {
 261                 src = src->separate_stencil;
 262                 src_format = PIPE_FORMAT_R8_UNORM;
 263         } else {
 264                 src_format = PIPE_FORMAT_RGBA8888_UNORM;
 265         }
 266
 267         if (dst->separate_stencil) {
 268                 dst = dst->separate_stencil;
 269                 dst_format = PIPE_FORMAT_R8_UNORM;
 270         } else {
 271                 dst_format = PIPE_FORMAT_RGBA8888_UNORM;
 272         }
 273
 274         /* Initialize the surface. */
 275         struct pipe_surface dst_tmpl = {
 276                 .u.tex = {
 277                         .level = info->dst.level,
 278                         .first_layer = info->dst.box.z,
 279                         .last_layer = info->dst.box.z,
 280                 },
 281                 .format = dst_format,
 282         };
 283         struct pipe_surface *dst_surf =
 284                 ctx->create_surface(ctx, &dst->base, &dst_tmpl);
 285
 286         /* Initialize the sampler view. */
 287         struct pipe_sampler_view src_tmpl = {
 288                 .target = src->base.target,
 289                 .format = src_format,
 290                 .u.tex = {
 291                         .first_level = info->src.level,
 292                         .last_level = info->src.level,
 293                         .first_layer = 0,
 294                         .last_layer = (PIPE_TEXTURE_3D ?
 295                                        u_minify(src->base.depth0,
 296                                                 info->src.level) - 1 :
 297                                        src->base.array_size - 1),
 298                 },
 299                 .swizzle_r = PIPE_SWIZZLE_X,
 300                 .swizzle_g = PIPE_SWIZZLE_Y,
 301                 .swizzle_b = PIPE_SWIZZLE_Z,
 302                 .swizzle_a = PIPE_SWIZZLE_W,
 303         };
 304         struct pipe_sampler_view *src_view =
 305                 ctx->create_sampler_view(ctx, &src->base, &src_tmpl);
 306
 307         v3d_blitter_save(v3d);
 308         util_blitter_blit_generic(v3d->blitter, dst_surf, &info->dst.box,
 309                                   src_view, &info->src.box,
 310                                   src->base.width0, src->base.height0,
 311                                   PIPE_MASK_R,
 312                                   PIPE_TEX_FILTER_NEAREST,
 313                                   info->scissor_enable ? &info->scissor : NULL,
 314                                   info->alpha_blend);
 315
 316         pipe_surface_reference(&dst_surf, NULL);
 317         pipe_sampler_view_reference(&src_view, NULL);
 318 }
 319
 320 /* Disable level 0 write, just write following mipmaps */
 321 #define V3D_TFU_IOA_DIMTW (1 << 0)
 322 #define V3D_TFU_IOA_FORMAT_SHIFT 3
 323 #define V3D_TFU_IOA_FORMAT_LINEARTILE 3
 324 #define V3D_TFU_IOA_FORMAT_UBLINEAR_1_COLUMN 4
 325 #define V3D_TFU_IOA_FORMAT_UBLINEAR_2_COLUMN 5
 326 #define V3D_TFU_IOA_FORMAT_UIF_NO_XOR 6
 327 #define V3D_TFU_IOA_FORMAT_UIF_XOR 7
 328
 329 #define V3D_TFU_ICFG_NUMMM_SHIFT 5
 330 #define V3D_TFU_ICFG_TTYPE_SHIFT 9
 331
 332 #define V3D_TFU_ICFG_FORMAT_SHIFT 18
 333 #define V3D_TFU_ICFG_FORMAT_RASTER 0
 334 #define V3D_TFU_ICFG_FORMAT_SAND_128 1
 335 #define V3D_TFU_ICFG_FORMAT_SAND_256 2
 336 #define V3D_TFU_ICFG_FORMAT_LINEARTILE 11
 337 #define V3D_TFU_ICFG_FORMAT_UBLINEAR_1_COLUMN 12
 338 #define V3D_TFU_ICFG_FORMAT_UBLINEAR_2_COLUMN 13
 339 #define V3D_TFU_ICFG_FORMAT_UIF_NO_XOR 14
 340 #define V3D_TFU_ICFG_FORMAT_UIF_XOR 15
 341
 342 boolean
 343 v3d_generate_mipmap(struct pipe_context *pctx,
 344                     struct pipe_resource *prsc,
 345                     enum pipe_format format,
 346                     unsigned int base_level,
 347                     unsigned int last_level,
 348                     unsigned int first_layer,
 349                     unsigned int last_layer)
 350 {
 351         struct v3d_context *v3d = v3d_context(pctx);
 352         struct v3d_screen *screen = v3d->screen;
 353         struct v3d_resource *rsc = v3d_resource(prsc);
 354         struct v3d_resource_slice *base_slice = &rsc->slices[base_level];
 355         int width = u_minify(prsc->width0, base_level);
 356         int height = u_minify(prsc->height0, base_level);
 357         uint32_t tex_format = v3d_get_tex_format(&screen->devinfo,
 358                                                  prsc->format);
 359
 360         if (!v3d_tfu_supports_tex_format(&screen->devinfo, tex_format))
 361                 return false;
 362
 363         if (prsc->target != PIPE_TEXTURE_2D)
 364                 return false;
 365         /* Since we don't support array or 3D textures, there should be only
 366          * one layer.
 367          */
 368         int layer = first_layer;
 369         assert(first_layer == last_layer);
 370
 371         /* Can't write to raster. */
 372         if (base_slice->tiling == VC5_TILING_RASTER)
 373                 return false;
 374
 375         v3d_flush_jobs_reading_resource(v3d, prsc);
 376
 377         struct drm_v3d_submit_tfu tfu = {
 378                 .ios = (height << 16) | width,
 379                 .bo_handles = { rsc->bo->handle },
 380                 .in_sync = v3d->out_sync,
 381                 .out_sync = v3d->out_sync,
 382         };
 383         uint32_t offset = (rsc->bo->offset +
 384                            v3d_layer_offset(prsc, base_level, layer));
 385         tfu.iia |= offset;
 386         tfu.icfg |= ((V3D_TFU_ICFG_FORMAT_LINEARTILE +
 387                       (base_slice->tiling - VC5_TILING_LINEARTILE)) <<
 388                      V3D_TFU_ICFG_FORMAT_SHIFT);
 389
 390         tfu.ioa |= offset;
 391         tfu.ioa |= V3D_TFU_IOA_DIMTW;
 392         tfu.ioa |= ((V3D_TFU_IOA_FORMAT_LINEARTILE +
 393                      (base_slice->tiling - VC5_TILING_LINEARTILE)) <<
 394                     V3D_TFU_IOA_FORMAT_SHIFT);
 395
 396         tfu.icfg |= tex_format << V3D_TFU_ICFG_TTYPE_SHIFT;
 397         tfu.icfg |= (last_level - base_level) << V3D_TFU_ICFG_NUMMM_SHIFT;
 398
 399         switch (base_slice->tiling) {
 400         case VC5_TILING_UIF_NO_XOR:
 401         case VC5_TILING_UIF_XOR:
 402                 tfu.iis |= (base_slice->padded_height /
 403                             (2 * v3d_utile_height(rsc->cpp)));
 404                 break;
 405         case VC5_TILING_RASTER:
 406                 tfu.iis |= base_slice->stride;
 407                 break;
 408         case VC5_TILING_LINEARTILE:
 409         case VC5_TILING_UBLINEAR_1_COLUMN:
 410         case VC5_TILING_UBLINEAR_2_COLUMN:
 411                 break;
 412        }
 413
 414         int ret = v3d_ioctl(screen->fd, DRM_IOCTL_V3D_SUBMIT_TFU, &tfu);
 415         if (ret != 0) {
 416                 fprintf(stderr, "Failed to submit TFU job: %d\n", ret);
 417                 return false;
 418         }
 419
 420         return true;
 421 }
 422
 423 /* Optimal hardware path for blitting pixels.
 424  * Scaling, format conversion, up- and downsampling (resolve) are allowed.
 425  */
 426 void
 427 v3d_blit(struct pipe_context *pctx, const struct pipe_blit_info *blit_info)
 428 {
 429         struct pipe_blit_info info = *blit_info;
 430
 431         if (info.mask & PIPE_MASK_S) {
 432                 v3d_stencil_blit(pctx, blit_info);
 433                 info.mask &= ~PIPE_MASK_S;
 434         }
 435
 436 #if 0
 437         if (v3d_tile_blit(pctx, blit_info))
 438                 return;
 439 #endif
 440
 441         v3d_render_blit(pctx, &info);
 442 }