src/gallium/drivers/nvfx/nvfx_surface.c

   1
   2 /**************************************************************************
   3  *
   4  * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
   5  * All Rights Reserved.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a
   8  * copy of this software and associated documentation files (the
   9  * "Software"), to deal in the Software without restriction, including
  10  * without limitation the rights to use, copy, modify, merge, publish,
  11  * distribute, sub license, and/or sell copies of the Software, and to
  12  * permit persons to whom the Software is furnished to do so, subject to
  13  * the following conditions:
  14  *
  15  * The above copyright notice and this permission notice (including the
  16  * next paragraph) shall be included in all copies or substantial portions
  17  * of the Software.
  18  *
  19  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  20  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  21  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  22  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  23  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  24  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  25  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  26  *
  27  **************************************************************************/
  28
  29 #include "pipe/p_context.h"
  30 #include "pipe/p_format.h"
  31 #include "util/u_format.h"
  32 #include "util/u_math.h"
  33 #include "util/u_memory.h"
  34 #include "util/u_pack_color.h"
  35 #include "util/u_blitter.h"
  36
  37 #include "nouveau/nouveau_winsys.h"
  38 #include "nouveau/nouveau_screen.h"
  39 #include "nvfx_context.h"
  40 #include "nvfx_screen.h"
  41 #include "nvfx_resource.h"
  42 #include "nv04_2d.h"
  43
  44 #include <nouveau/nouveau_bo.h>
  45
  46 static INLINE void
  47 nvfx_region_set_format(struct nv04_region* rgn, enum pipe_format format)
  48 {
  49         unsigned bits = util_format_get_blocksizebits(format);
  50         unsigned shift = 0;
  51         switch(bits)
  52         {
  53         case 8:
  54                 rgn->bpps = 0;
  55                 break;
  56         case 16:
  57                 rgn->bpps = 1;
  58                 break;
  59         case 32:
  60                 rgn->bpps = 2;
  61                 break;
  62         case 64:
  63                 rgn->bpps = 2;
  64                 shift = 1;
  65                 break;
  66         case 128:
  67                 rgn->bpps = 2;
  68                 shift = 2;
  69                 break;
  70         }
  71
  72         if(shift) {
  73                 rgn->x = util_format_get_nblocksx(format, rgn->x) << shift;
  74                 rgn->y = util_format_get_nblocksy(format, rgn->y);
  75                 rgn->w <<= shift;
  76         }
  77 }
  78
  79 static INLINE void
  80 nvfx_region_init_for_surface(struct nv04_region* rgn, struct nvfx_surface* surf, unsigned x, unsigned y, bool for_write)
  81 {
  82         rgn->x = x;
  83         rgn->y = y;
  84         rgn->z = 0;
  85
  86         if(surf->temp)
  87         {
  88                 rgn->bo = surf->temp->base.bo;
  89                 rgn->offset = 0;
  90                 rgn->pitch = surf->temp->linear_pitch;
  91
  92                 if(for_write)
  93                         util_dirty_surface_set_dirty(nvfx_surface_get_dirty_surfaces(&surf->base.base), &surf->base);
  94         } else {
  95                 rgn->bo = ((struct nvfx_resource*)surf->base.base.texture)->bo;
  96                 rgn->offset = surf->base.base.offset;
  97
  98                 if(surf->base.base.texture->flags & NVFX_RESOURCE_FLAG_LINEAR)
  99                         rgn->pitch = surf->pitch;
 100                 else
 101                 {
 102                         rgn->pitch = 0;
 103                         rgn->z = surf->base.base.zslice;
 104                         rgn->w = surf->base.base.width;
 105                         rgn->h = surf->base.base.height;
 106                         rgn->d = u_minify(surf->base.base.texture->depth0, surf->base.base.level);
 107                 }
 108         }
 109
 110         nvfx_region_set_format(rgn, surf->base.base.format);
 111         if(!rgn->pitch)
 112                 nv04_region_try_to_linearize(rgn);
 113 }
 114
 115 static INLINE void
 116 nvfx_region_init_for_subresource(struct nv04_region* rgn, struct pipe_resource* pt, struct pipe_subresource sub, unsigned x, unsigned y, unsigned z, bool for_write)
 117 {
 118         if(pt->target != PIPE_BUFFER)
 119         {
 120                 struct nvfx_surface* ns = (struct nvfx_surface*)util_surfaces_peek(&((struct nvfx_miptree*)pt)->surfaces, pt, sub.face, sub.level, z);
 121                 if(ns && util_dirty_surface_is_dirty(&ns->base))
 122                 {
 123                         nvfx_region_init_for_surface(rgn, ns, x, y, for_write);
 124                         return;
 125                 }
 126         }
 127
 128         rgn->bo = ((struct nvfx_resource*)pt)->bo;
 129         rgn->offset = nvfx_subresource_offset(pt, sub.face, sub.level, z);
 130         rgn->x = x;
 131         rgn->y = y;
 132
 133         if(pt->flags & NVFX_RESOURCE_FLAG_LINEAR)
 134         {
 135                 rgn->pitch = nvfx_subresource_pitch(pt, sub.level);
 136                 rgn->z = 0;
 137         }
 138         else
 139         {
 140                 rgn->pitch = 0;
 141                 rgn->z = z;
 142                 rgn->w = u_minify(pt->width0, sub.level);
 143                 rgn->h = u_minify(pt->height0, sub.level);
 144                 rgn->d = u_minify(pt->depth0, sub.level);
 145         }
 146
 147         nvfx_region_set_format(rgn, pt->format);
 148         if(!rgn->pitch)
 149                 nv04_region_try_to_linearize(rgn);
 150 }
 151
 152 // TODO: actually test this for all formats, it's probably wrong for some...
 153
 154 static INLINE int
 155 nvfx_surface_format(enum pipe_format format)
 156 {
 157         switch(util_format_get_blocksize(format)) {
 158         case 1:
 159                 return NV04_CONTEXT_SURFACES_2D_FORMAT_Y8;
 160         case 2:
 161                 //return NV04_CONTEXT_SURFACES_2D_FORMAT_Y16;
 162                 return NV04_CONTEXT_SURFACES_2D_FORMAT_R5G6B5;
 163         case 4:
 164                 //if(format == PIPE_FORMAT_B8G8R8X8_UNORM || format == PIPE_FORMAT_B8G8R8A8_UNORM)
 165                         return NV04_CONTEXT_SURFACES_2D_FORMAT_A8R8G8B8;
 166                 //else
 167                 //      return NV04_CONTEXT_SURFACES_2D_FORMAT_Y32;
 168         default:
 169                 return -1;
 170         }
 171 }
 172
 173 static INLINE int
 174 nv04_scaled_image_format(enum pipe_format format)
 175 {
 176         switch(util_format_get_blocksize(format)) {
 177         case 1:
 178                 return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_Y8;
 179         case 2:
 180                 //if(format == PIPE_FORMAT_B5G5R5A1_UNORM)
 181                 //      return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_A1R5G5B5;
 182                 //else
 183                         return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_R5G6B5;
 184         case 4:
 185                 if(format == PIPE_FORMAT_B8G8R8X8_UNORM)
 186                         return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_X8R8G8B8;
 187                 else
 188                         return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_A8R8G8B8;
 189         default:
 190                 return -1;
 191         }
 192 }
 193
 194 // don't save index buffer because blitter doesn't setit
 195 static struct blitter_context*
 196 nvfx_get_blitter(struct pipe_context* pipe, int copy)
 197 {
 198         struct nvfx_context* nvfx = nvfx_context(pipe);
 199
 200         assert(nvfx->blitters_in_use < Elements(nvfx->blitter));
 201
 202         struct blitter_context** pblitter = &nvfx->blitter[nvfx->blitters_in_use++];
 203         if(!*pblitter)
 204                 *pblitter = util_blitter_create(pipe);
 205         struct blitter_context* blitter = *pblitter;
 206
 207         util_blitter_save_blend(blitter, nvfx->blend);
 208         util_blitter_save_depth_stencil_alpha(blitter, nvfx->zsa);
 209         util_blitter_save_stencil_ref(blitter, &nvfx->stencil_ref);
 210         util_blitter_save_rasterizer(blitter, nvfx->rasterizer);
 211         util_blitter_save_fragment_shader(blitter, nvfx->fragprog);
 212         util_blitter_save_vertex_shader(blitter, nvfx->vertprog);
 213         util_blitter_save_viewport(blitter, &nvfx->viewport);
 214         util_blitter_save_framebuffer(blitter, &nvfx->framebuffer);
 215         util_blitter_save_clip(blitter, &nvfx->clip);
 216         util_blitter_save_vertex_elements(blitter, nvfx->vtxelt);
 217         util_blitter_save_vertex_buffers(blitter, nvfx->vtxbuf_nr, nvfx->vtxbuf);
 218
 219         if(copy)
 220         {
 221                 util_blitter_save_fragment_sampler_states(blitter, nvfx->nr_samplers, (void**)nvfx->tex_sampler);
 222                 util_blitter_save_fragment_sampler_views(blitter, nvfx->nr_textures, nvfx->fragment_sampler_views);
 223         }
 224
 225         return blitter;
 226 }
 227
 228 static inline void
 229 nvfx_put_blitter(struct pipe_context* pipe, struct blitter_context* blitter)
 230 {
 231         struct nvfx_context* nvfx = nvfx_context(pipe);
 232         --nvfx->blitters_in_use;
 233         assert(nvfx->blitters_in_use >= 0);
 234 }
 235
 236 static unsigned
 237 nvfx_region_clone(struct nv04_2d_context* ctx, struct nv04_region* rgn, unsigned w, unsigned h, boolean for_read)
 238 {
 239         unsigned begin = nv04_region_begin(rgn, w, h);
 240         unsigned end = nv04_region_end(rgn, w, h);
 241         unsigned size = end - begin;
 242         struct nouveau_bo* bo = 0;
 243         nouveau_bo_new(rgn->bo->device, NOUVEAU_BO_MAP | NOUVEAU_BO_GART, 256, size, &bo);
 244
 245         if(for_read || (size > ((w * h) << rgn->bpps)))
 246                 nv04_memcpy(ctx, bo, 0, rgn->bo, rgn->offset + begin, size);
 247
 248         rgn->bo = bo;
 249         rgn->offset = -begin;
 250         return begin;
 251 }
 252
 253 static void
 254 nvfx_resource_copy_region(struct pipe_context *pipe,
 255                   struct pipe_resource *dstr, struct pipe_subresource subdst,
 256                   unsigned dstx, unsigned dsty, unsigned dstz,
 257                   struct pipe_resource *srcr, struct pipe_subresource subsrc,
 258                   unsigned srcx, unsigned srcy, unsigned srcz,
 259                   unsigned w, unsigned h)
 260 {
 261         static int copy_threshold = -1;
 262         struct nv04_2d_context *ctx = nvfx_screen(pipe->screen)->eng2d;
 263         struct nv04_region dst, src;
 264         int dst_to_gpu;
 265         int src_on_gpu;
 266         boolean small;
 267         int ret;
 268
 269         if(!w || !h)
 270                 return;
 271
 272         if(copy_threshold < 0)
 273                 copy_threshold = debug_get_num_option("NOUVEAU_COPY_THRESHOLD", 4);
 274
 275         dst_to_gpu = dstr->usage != PIPE_USAGE_DYNAMIC && dstr->usage != PIPE_USAGE_STAGING;
 276         src_on_gpu = nvfx_resource_on_gpu(srcr);
 277
 278         nvfx_region_init_for_subresource(&dst, dstr, subdst, dstx, dsty, dstz, TRUE);
 279         nvfx_region_init_for_subresource(&src, srcr, subsrc, srcx, srcy, srcz, FALSE);
 280         w = util_format_get_stride(dstr->format, w) >> dst.bpps;
 281         h = util_format_get_nblocksy(dstr->format, h);
 282
 283         small = (w * h <= copy_threshold);
 284         if((!dst_to_gpu || !src_on_gpu) && small)
 285                 ret = -1; /* use the CPU */
 286         else
 287                 ret = nv04_region_copy_2d(ctx, &dst, &src, w, h,
 288                         dstr->target == PIPE_BUFFER ? -1 : nvfx_surface_format(dstr->format),
 289                         dstr->target == PIPE_BUFFER ? -1 : nv04_scaled_image_format(dstr->format),
 290                         dst_to_gpu, src_on_gpu);
 291         if(!ret)
 292         {}
 293         else if(ret > 0 && dstr->bind & PIPE_BIND_RENDER_TARGET && srcr->bind & PIPE_BIND_SAMPLER_VIEW)
 294         {
 295                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 1);
 296                 util_blitter_copy_region(blitter, dstr, subdst, dstx, dsty, dstz, srcr, subsrc, srcx, srcy, srcz, w, h, TRUE);
 297                 nvfx_put_blitter(pipe, blitter);
 298         }
 299         else
 300         {
 301                 struct nv04_region dstt = dst;
 302                 struct nv04_region srct = src;
 303                 unsigned dstbegin = 0;
 304
 305                 if(!small)
 306                 {
 307                         if(src_on_gpu)
 308                                 nvfx_region_clone(ctx, &srct, w, h, TRUE);
 309
 310                         if(dst_to_gpu)
 311                                 dstbegin = nvfx_region_clone(ctx, &dstt, w, h, FALSE);
 312                 }
 313
 314                 nv04_region_copy_cpu(&dstt, &srct, w, h);
 315
 316                 if(srct.bo != src.bo)
 317                         nouveau_screen_bo_release(pipe->screen, srct.bo);
 318
 319                 if(dstt.bo != dst.bo)
 320                 {
 321                         nv04_memcpy(ctx, dst.bo, dst.offset + dstbegin, dstt.bo, 0, dstt.bo->size);
 322                         nouveau_screen_bo_release(pipe->screen, dstt.bo);
 323                 }
 324         }
 325 }
 326
 327 static int
 328 nvfx_surface_fill(struct pipe_context* pipe, struct pipe_surface *dsts,
 329                   unsigned dx, unsigned dy, unsigned w, unsigned h, unsigned value)
 330 {
 331         struct nv04_2d_context *ctx = nvfx_screen(pipe->screen)->eng2d;
 332         struct nv04_region dst;
 333         int ret;
 334         /* Always try to use the GPU right now, if possible
 335          * If the user wanted the surface data on the CPU, he would have cleared with memset (hopefully) */
 336
 337         // we don't care about interior pixel order since we set all them to the same value
 338         nvfx_region_init_for_surface(&dst, (struct nvfx_surface*)dsts, dx, dy, TRUE);
 339
 340         w = util_format_get_stride(dsts->format, w) >> dst.bpps;
 341         h = util_format_get_nblocksy(dsts->format, h);
 342
 343         ret = nv04_region_fill_2d(ctx, &dst, w, h, value);
 344         if(ret > 0 && dsts->texture->bind & PIPE_BIND_RENDER_TARGET)
 345                 return 1;
 346         else if(ret)
 347         {
 348                 struct nv04_region dstt = dst;
 349                 unsigned dstbegin = 0;
 350
 351                 if(nvfx_resource_on_gpu(dsts->texture))
 352                         dstbegin = nvfx_region_clone(ctx, &dstt, w, h, FALSE);
 353
 354                 nv04_region_fill_cpu(&dstt, w, h, value);
 355
 356                 if(dstt.bo != dst.bo)
 357                 {
 358                         nv04_memcpy(ctx, dst.bo, dst.offset + dstbegin, dstt.bo, 0, dstt.bo->size);
 359                         nouveau_screen_bo_release(pipe->screen, dstt.bo);
 360                 }
 361         }
 362
 363         return 0;
 364 }
 365
 366
 367 void
 368 nvfx_screen_surface_takedown(struct pipe_screen *pscreen)
 369 {
 370         nv04_2d_context_takedown(nvfx_screen(pscreen)->eng2d);
 371         nvfx_screen(pscreen)->eng2d = 0;
 372 }
 373
 374 int
 375 nvfx_screen_surface_init(struct pipe_screen *pscreen)
 376 {
 377         struct nv04_2d_context* ctx = nv04_2d_context_init(nouveau_screen(pscreen)->channel);
 378         if(!ctx)
 379                 return -1;
 380         nvfx_screen(pscreen)->eng2d = ctx;
 381         return 0;
 382 }
 383
 384 static void
 385 nvfx_surface_copy_temp(struct pipe_context* pipe, struct pipe_surface* surf, int to_temp)
 386 {
 387         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 388         struct pipe_subresource tempsr, surfsr;
 389         struct nvfx_context* nvfx = nvfx_context(pipe);
 390
 391         /* temporarily detach the temp, so it isn't used in place of the actual resource */
 392         struct nvfx_miptree* temp = ns->temp;
 393         ns->temp = 0;
 394
 395         // TODO: we really should do this validation before setting these variable in draw calls
 396         unsigned use_vertex_buffers = nvfx->use_vertex_buffers;
 397         boolean use_index_buffer = nvfx->use_index_buffer;
 398         unsigned base_vertex = nvfx->base_vertex;
 399
 400         tempsr.face = 0;
 401         tempsr.level = 0;
 402         surfsr.face = surf->face;
 403         surfsr.level = surf->level;
 404
 405         if(to_temp)
 406                 nvfx_resource_copy_region(pipe, &temp->base.base, tempsr, 0, 0, 0, surf->texture, surfsr, 0, 0, surf->zslice, surf->width, surf->height);
 407         else
 408                 nvfx_resource_copy_region(pipe, surf->texture, surfsr, 0, 0, surf->zslice, &temp->base.base, tempsr, 0, 0, 0, surf->width, surf->height);
 409
 410         /* If this triggers, it probably means we attempted to use the blitter
 411          * but failed due to non-renderability of the target.
 412          * Obviously, this would lead to infinite recursion if supported. */
 413         assert(!ns->temp);
 414
 415         ns->temp = temp;
 416
 417         nvfx->use_vertex_buffers = use_vertex_buffers;
 418         nvfx->use_index_buffer = use_index_buffer;
 419         nvfx->base_vertex = base_vertex;
 420
 421         nvfx->dirty |= NVFX_NEW_ARRAYS;
 422         nvfx->draw_dirty |= NVFX_NEW_ARRAYS;
 423 }
 424
 425 void
 426 nvfx_surface_create_temp(struct pipe_context* pipe, struct pipe_surface* surf)
 427 {
 428         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 429         struct pipe_resource template;
 430         memset(&template, 0, sizeof(struct pipe_resource));
 431         template.target = PIPE_TEXTURE_2D;
 432         template.format = surf->format;
 433         template.width0 = surf->width;
 434         template.height0 = surf->height;
 435         template.depth0 = 1;
 436         template.nr_samples = surf->texture->nr_samples;
 437         template.flags = NVFX_RESOURCE_FLAG_LINEAR;
 438
 439         assert(!ns->temp && !util_dirty_surface_is_dirty(&ns->base));
 440
 441         ns->temp = (struct nvfx_miptree*)nvfx_miptree_create(pipe->screen, &template);
 442         nvfx_surface_copy_temp(pipe, surf, 1);
 443 }
 444
 445 void
 446 nvfx_surface_flush(struct pipe_context* pipe, struct pipe_surface* surf)
 447 {
 448         struct nvfx_context* nvfx = (struct nvfx_context*)pipe;
 449         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 450         boolean bound = FALSE;
 451
 452         nvfx_surface_copy_temp(pipe, surf, 0);
 453
 454         util_dirty_surface_set_clean(nvfx_surface_get_dirty_surfaces(surf), &ns->base);
 455
 456         if(nvfx->framebuffer.zsbuf == surf)
 457                 bound = TRUE;
 458         else
 459         {
 460                 for(unsigned i = 0; i < nvfx->framebuffer.nr_cbufs; ++i)
 461                 {
 462                         if(nvfx->framebuffer.cbufs[i] == surf)
 463                         {
 464                                 bound = TRUE;
 465                                 break;
 466                         }
 467                 }
 468         }
 469
 470         if(!bound)
 471                 pipe_resource_reference((struct pipe_resource**)&ns->temp, 0);
 472 }
 473
 474 static void
 475 nvfx_clear_render_target(struct pipe_context *pipe,
 476                          struct pipe_surface *dst,
 477                          const float *rgba,
 478                          unsigned dstx, unsigned dsty,
 479                          unsigned width, unsigned height)
 480 {
 481         union util_color uc;
 482         util_pack_color(rgba, dst->format, &uc);
 483
 484         if(util_format_get_blocksizebits(dst->format) > 32
 485                 || nvfx_surface_fill(pipe, dst, dstx, dsty, width, height, uc.ui))
 486         {
 487                 // TODO: probably should use hardware clear here instead if possible
 488                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 0);
 489                 util_blitter_clear_render_target(blitter, dst, rgba, dstx, dsty, width, height);
 490                 nvfx_put_blitter(pipe, blitter);
 491         }
 492 }
 493
 494 static void
 495 nvfx_clear_depth_stencil(struct pipe_context *pipe,
 496                          struct pipe_surface *dst,
 497                          unsigned clear_flags,
 498                          double depth,
 499                          unsigned stencil,
 500                          unsigned dstx, unsigned dsty,
 501                          unsigned width, unsigned height)
 502 {
 503         if(util_format_get_blocksizebits(dst->format) > 32
 504                 || nvfx_surface_fill(pipe, dst, dstx, dsty, width, height, util_pack_z_stencil(dst->format, depth, stencil)))
 505         {
 506                 // TODO: probably should use hardware clear here instead if possible
 507                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 0);
 508                 util_blitter_clear_depth_stencil(blitter, dst, clear_flags, depth, stencil, dstx, dsty, width, height);
 509                 nvfx_put_blitter(pipe, blitter);
 510         }
 511 }
 512
 513
 514 void
 515 nvfx_init_surface_functions(struct nvfx_context *nvfx)
 516 {
 517         nvfx->pipe.resource_copy_region = nvfx_resource_copy_region;
 518         nvfx->pipe.clear_render_target = nvfx_clear_render_target;
 519         nvfx->pipe.clear_depth_stencil = nvfx_clear_depth_stencil;
 520 }