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         rgn->one_bits = 0;
  52
  53         switch(bits)
  54         {
  55         case 8:
  56                 rgn->bpps = 0;
  57                 break;
  58         case 16:
  59                 rgn->bpps = 1;
  60                 if(format == PIPE_FORMAT_B5G5R5X1_UNORM)
  61                         rgn->one_bits = 1;
  62                 break;
  63         case 32:
  64                 rgn->bpps = 2;
  65                 if(format == PIPE_FORMAT_R8G8B8X8_UNORM || format == PIPE_FORMAT_B8G8R8X8_UNORM)
  66                         rgn->one_bits = 8;
  67                 break;
  68         case 64:
  69                 rgn->bpps = 2;
  70                 shift = 1;
  71                 break;
  72         case 128:
  73                 rgn->bpps = 2;
  74                 shift = 2;
  75                 break;
  76         }
  77
  78         if(shift) {
  79                 rgn->x = util_format_get_nblocksx(format, rgn->x) << shift;
  80                 rgn->y = util_format_get_nblocksy(format, rgn->y);
  81                 rgn->w <<= shift;
  82         }
  83 }
  84
  85 static INLINE void
  86 nvfx_region_init_for_surface(struct nv04_region* rgn, struct nvfx_surface* surf, unsigned x, unsigned y, bool for_write)
  87 {
  88         rgn->x = x;
  89         rgn->y = y;
  90         rgn->z = 0;
  91
  92         if(surf->temp)
  93         {
  94                 rgn->bo = surf->temp->base.bo;
  95                 rgn->offset = 0;
  96                 rgn->pitch = surf->temp->linear_pitch;
  97
  98                 if(for_write)
  99                         util_dirty_surface_set_dirty(nvfx_surface_get_dirty_surfaces(&surf->base.base), &surf->base);
 100         } else {
 101                 rgn->bo = ((struct nvfx_resource*)surf->base.base.texture)->bo;
 102                 rgn->offset = surf->base.base.offset;
 103
 104                 if(surf->base.base.texture->flags & NVFX_RESOURCE_FLAG_LINEAR)
 105                         rgn->pitch = surf->pitch;
 106                 else
 107                 {
 108                         rgn->pitch = 0;
 109                         rgn->z = surf->base.base.zslice;
 110                         rgn->w = surf->base.base.width;
 111                         rgn->h = surf->base.base.height;
 112                         rgn->d = u_minify(surf->base.base.texture->depth0, surf->base.base.level);
 113                 }
 114         }
 115
 116         nvfx_region_set_format(rgn, surf->base.base.format);
 117         if(!rgn->pitch)
 118                 nv04_region_try_to_linearize(rgn);
 119 }
 120
 121 static INLINE void
 122 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)
 123 {
 124         if(pt->target != PIPE_BUFFER)
 125         {
 126                 struct nvfx_surface* ns = (struct nvfx_surface*)util_surfaces_peek(&((struct nvfx_miptree*)pt)->surfaces, pt, sub.face, sub.level, z);
 127                 if(ns && util_dirty_surface_is_dirty(&ns->base))
 128                 {
 129                         nvfx_region_init_for_surface(rgn, ns, x, y, for_write);
 130                         return;
 131                 }
 132         }
 133
 134         rgn->bo = ((struct nvfx_resource*)pt)->bo;
 135         rgn->offset = nvfx_subresource_offset(pt, sub.face, sub.level, z);
 136         rgn->x = x;
 137         rgn->y = y;
 138
 139         if(pt->flags & NVFX_RESOURCE_FLAG_LINEAR)
 140         {
 141                 rgn->pitch = nvfx_subresource_pitch(pt, sub.level);
 142                 rgn->z = 0;
 143         }
 144         else
 145         {
 146                 rgn->pitch = 0;
 147                 rgn->z = z;
 148                 rgn->w = u_minify(pt->width0, sub.level);
 149                 rgn->h = u_minify(pt->height0, sub.level);
 150                 rgn->d = u_minify(pt->depth0, sub.level);
 151         }
 152
 153         nvfx_region_set_format(rgn, pt->format);
 154         if(!rgn->pitch)
 155                 nv04_region_try_to_linearize(rgn);
 156 }
 157
 158 // don't save index buffer because blitter doesn't setit
 159 static struct blitter_context*
 160 nvfx_get_blitter(struct pipe_context* pipe, int copy)
 161 {
 162         struct nvfx_context* nvfx = nvfx_context(pipe);
 163
 164         assert(nvfx->blitters_in_use < Elements(nvfx->blitter));
 165
 166         struct blitter_context** pblitter = &nvfx->blitter[nvfx->blitters_in_use++];
 167         if(!*pblitter)
 168                 *pblitter = util_blitter_create(pipe);
 169         struct blitter_context* blitter = *pblitter;
 170
 171         util_blitter_save_blend(blitter, nvfx->blend);
 172         util_blitter_save_depth_stencil_alpha(blitter, nvfx->zsa);
 173         util_blitter_save_stencil_ref(blitter, &nvfx->stencil_ref);
 174         util_blitter_save_rasterizer(blitter, nvfx->rasterizer);
 175         util_blitter_save_fragment_shader(blitter, nvfx->fragprog);
 176         util_blitter_save_vertex_shader(blitter, nvfx->vertprog);
 177         util_blitter_save_viewport(blitter, &nvfx->viewport);
 178         util_blitter_save_framebuffer(blitter, &nvfx->framebuffer);
 179         util_blitter_save_clip(blitter, &nvfx->clip);
 180         util_blitter_save_vertex_elements(blitter, nvfx->vtxelt);
 181         util_blitter_save_vertex_buffers(blitter, nvfx->vtxbuf_nr, nvfx->vtxbuf);
 182
 183         if(copy)
 184         {
 185                 util_blitter_save_fragment_sampler_states(blitter, nvfx->nr_samplers, (void**)nvfx->tex_sampler);
 186                 util_blitter_save_fragment_sampler_views(blitter, nvfx->nr_textures, nvfx->fragment_sampler_views);
 187         }
 188
 189         return blitter;
 190 }
 191
 192 static inline void
 193 nvfx_put_blitter(struct pipe_context* pipe, struct blitter_context* blitter)
 194 {
 195         struct nvfx_context* nvfx = nvfx_context(pipe);
 196         --nvfx->blitters_in_use;
 197         assert(nvfx->blitters_in_use >= 0);
 198 }
 199
 200 static unsigned
 201 nvfx_region_clone(struct nv04_2d_context* ctx, struct nv04_region* rgn, unsigned w, unsigned h, boolean for_read)
 202 {
 203         unsigned begin = nv04_region_begin(rgn, w, h);
 204         unsigned end = nv04_region_end(rgn, w, h);
 205         unsigned size = end - begin;
 206         struct nouveau_bo* bo = 0;
 207         nouveau_bo_new(rgn->bo->device, NOUVEAU_BO_MAP | NOUVEAU_BO_GART, 256, size, &bo);
 208
 209         if(for_read || (size > ((w * h) << rgn->bpps)))
 210                 nv04_memcpy(ctx, bo, 0, rgn->bo, rgn->offset + begin, size);
 211
 212         rgn->bo = bo;
 213         rgn->offset = -begin;
 214         return begin;
 215 }
 216
 217 static void
 218 nvfx_resource_copy_region(struct pipe_context *pipe,
 219                   struct pipe_resource *dstr, struct pipe_subresource subdst,
 220                   unsigned dstx, unsigned dsty, unsigned dstz,
 221                   struct pipe_resource *srcr, struct pipe_subresource subsrc,
 222                   unsigned srcx, unsigned srcy, unsigned srcz,
 223                   unsigned w, unsigned h)
 224 {
 225         static int copy_threshold = -1;
 226         struct nv04_2d_context *ctx = nvfx_screen(pipe->screen)->eng2d;
 227         struct nv04_region dst, src;
 228         int dst_to_gpu;
 229         int src_on_gpu;
 230         boolean small;
 231         int ret;
 232
 233         if(!w || !h)
 234                 return;
 235
 236         if(copy_threshold < 0)
 237                 copy_threshold = debug_get_num_option("NOUVEAU_COPY_THRESHOLD", 4);
 238
 239         dst_to_gpu = dstr->usage != PIPE_USAGE_DYNAMIC && dstr->usage != PIPE_USAGE_STAGING;
 240         src_on_gpu = nvfx_resource_on_gpu(srcr);
 241
 242         nvfx_region_init_for_subresource(&dst, dstr, subdst, dstx, dsty, dstz, TRUE);
 243         nvfx_region_init_for_subresource(&src, srcr, subsrc, srcx, srcy, srcz, FALSE);
 244         w = util_format_get_stride(dstr->format, w) >> dst.bpps;
 245         h = util_format_get_nblocksy(dstr->format, h);
 246
 247         small = (w * h <= copy_threshold);
 248         if((!dst_to_gpu || !src_on_gpu) && small)
 249                 ret = -1; /* use the CPU */
 250         else
 251                 ret = nv04_region_copy_2d(ctx, &dst, &src, w, h, dst_to_gpu, src_on_gpu);
 252         if(!ret)
 253         {}
 254         else if(ret > 0 && dstr->bind & PIPE_BIND_RENDER_TARGET && srcr->bind & PIPE_BIND_SAMPLER_VIEW)
 255         {
 256                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 1);
 257                 util_blitter_copy_region(blitter, dstr, subdst, dstx, dsty, dstz, srcr, subsrc, srcx, srcy, srcz, w, h, TRUE);
 258                 nvfx_put_blitter(pipe, blitter);
 259         }
 260         else
 261         {
 262                 struct nv04_region dstt = dst;
 263                 struct nv04_region srct = src;
 264                 unsigned dstbegin = 0;
 265
 266                 if(!small)
 267                 {
 268                         if(src_on_gpu)
 269                                 nvfx_region_clone(ctx, &srct, w, h, TRUE);
 270
 271                         if(dst_to_gpu)
 272                                 dstbegin = nvfx_region_clone(ctx, &dstt, w, h, FALSE);
 273                 }
 274
 275                 nv04_region_copy_cpu(&dstt, &srct, w, h);
 276
 277                 if(srct.bo != src.bo)
 278                         nouveau_screen_bo_release(pipe->screen, srct.bo);
 279
 280                 if(dstt.bo != dst.bo)
 281                 {
 282                         nv04_memcpy(ctx, dst.bo, dst.offset + dstbegin, dstt.bo, 0, dstt.bo->size);
 283                         nouveau_screen_bo_release(pipe->screen, dstt.bo);
 284                 }
 285         }
 286 }
 287
 288 static int
 289 nvfx_surface_fill(struct pipe_context* pipe, struct pipe_surface *dsts,
 290                   unsigned dx, unsigned dy, unsigned w, unsigned h, unsigned value)
 291 {
 292         struct nv04_2d_context *ctx = nvfx_screen(pipe->screen)->eng2d;
 293         struct nv04_region dst;
 294         int ret;
 295         /* Always try to use the GPU right now, if possible
 296          * If the user wanted the surface data on the CPU, he would have cleared with memset (hopefully) */
 297
 298         // we don't care about interior pixel order since we set all them to the same value
 299         nvfx_region_init_for_surface(&dst, (struct nvfx_surface*)dsts, dx, dy, TRUE);
 300
 301         w = util_format_get_stride(dsts->format, w) >> dst.bpps;
 302         h = util_format_get_nblocksy(dsts->format, h);
 303
 304         ret = nv04_region_fill_2d(ctx, &dst, w, h, value);
 305         if(ret > 0 && dsts->texture->bind & PIPE_BIND_RENDER_TARGET)
 306                 return 1;
 307         else if(ret)
 308         {
 309                 struct nv04_region dstt = dst;
 310                 unsigned dstbegin = 0;
 311
 312                 if(nvfx_resource_on_gpu(dsts->texture))
 313                         dstbegin = nvfx_region_clone(ctx, &dstt, w, h, FALSE);
 314
 315                 nv04_region_fill_cpu(&dstt, w, h, value);
 316
 317                 if(dstt.bo != dst.bo)
 318                 {
 319                         nv04_memcpy(ctx, dst.bo, dst.offset + dstbegin, dstt.bo, 0, dstt.bo->size);
 320                         nouveau_screen_bo_release(pipe->screen, dstt.bo);
 321                 }
 322         }
 323
 324         return 0;
 325 }
 326
 327
 328 void
 329 nvfx_screen_surface_takedown(struct pipe_screen *pscreen)
 330 {
 331         nv04_2d_context_takedown(nvfx_screen(pscreen)->eng2d);
 332         nvfx_screen(pscreen)->eng2d = 0;
 333 }
 334
 335 int
 336 nvfx_screen_surface_init(struct pipe_screen *pscreen)
 337 {
 338         struct nv04_2d_context* ctx = nv04_2d_context_init(nouveau_screen(pscreen)->channel);
 339         if(!ctx)
 340                 return -1;
 341         nvfx_screen(pscreen)->eng2d = ctx;
 342         return 0;
 343 }
 344
 345 static void
 346 nvfx_surface_copy_temp(struct pipe_context* pipe, struct pipe_surface* surf, int to_temp)
 347 {
 348         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 349         struct pipe_subresource tempsr, surfsr;
 350         struct nvfx_context* nvfx = nvfx_context(pipe);
 351
 352         /* temporarily detach the temp, so it isn't used in place of the actual resource */
 353         struct nvfx_miptree* temp = ns->temp;
 354         ns->temp = 0;
 355
 356         // TODO: we really should do this validation before setting these variable in draw calls
 357         unsigned use_vertex_buffers = nvfx->use_vertex_buffers;
 358         boolean use_index_buffer = nvfx->use_index_buffer;
 359         unsigned base_vertex = nvfx->base_vertex;
 360
 361         tempsr.face = 0;
 362         tempsr.level = 0;
 363         surfsr.face = surf->face;
 364         surfsr.level = surf->level;
 365
 366         if(to_temp)
 367                 nvfx_resource_copy_region(pipe, &temp->base.base, tempsr, 0, 0, 0, surf->texture, surfsr, 0, 0, surf->zslice, surf->width, surf->height);
 368         else
 369                 nvfx_resource_copy_region(pipe, surf->texture, surfsr, 0, 0, surf->zslice, &temp->base.base, tempsr, 0, 0, 0, surf->width, surf->height);
 370
 371         /* If this triggers, it probably means we attempted to use the blitter
 372          * but failed due to non-renderability of the target.
 373          * Obviously, this would lead to infinite recursion if supported. */
 374         assert(!ns->temp);
 375
 376         ns->temp = temp;
 377
 378         nvfx->use_vertex_buffers = use_vertex_buffers;
 379         nvfx->use_index_buffer = use_index_buffer;
 380         nvfx->base_vertex = base_vertex;
 381
 382         nvfx->dirty |= NVFX_NEW_ARRAYS;
 383         nvfx->draw_dirty |= NVFX_NEW_ARRAYS;
 384 }
 385
 386 void
 387 nvfx_surface_create_temp(struct pipe_context* pipe, struct pipe_surface* surf)
 388 {
 389         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 390         struct pipe_resource template;
 391         memset(&template, 0, sizeof(struct pipe_resource));
 392         template.target = PIPE_TEXTURE_2D;
 393         template.format = surf->format;
 394         template.width0 = surf->width;
 395         template.height0 = surf->height;
 396         template.depth0 = 1;
 397         template.nr_samples = surf->texture->nr_samples;
 398         template.flags = NVFX_RESOURCE_FLAG_LINEAR;
 399
 400         assert(!ns->temp && !util_dirty_surface_is_dirty(&ns->base));
 401
 402         ns->temp = (struct nvfx_miptree*)nvfx_miptree_create(pipe->screen, &template);
 403         nvfx_surface_copy_temp(pipe, surf, 1);
 404 }
 405
 406 void
 407 nvfx_surface_flush(struct pipe_context* pipe, struct pipe_surface* surf)
 408 {
 409         struct nvfx_context* nvfx = (struct nvfx_context*)pipe;
 410         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 411         boolean bound = FALSE;
 412
 413         nvfx_surface_copy_temp(pipe, surf, 0);
 414
 415         util_dirty_surface_set_clean(nvfx_surface_get_dirty_surfaces(surf), &ns->base);
 416
 417         if(nvfx->framebuffer.zsbuf == surf)
 418                 bound = TRUE;
 419         else
 420         {
 421                 for(unsigned i = 0; i < nvfx->framebuffer.nr_cbufs; ++i)
 422                 {
 423                         if(nvfx->framebuffer.cbufs[i] == surf)
 424                         {
 425                                 bound = TRUE;
 426                                 break;
 427                         }
 428                 }
 429         }
 430
 431         if(!bound)
 432                 pipe_resource_reference((struct pipe_resource**)&ns->temp, 0);
 433 }
 434
 435 static void
 436 nvfx_clear_render_target(struct pipe_context *pipe,
 437                          struct pipe_surface *dst,
 438                          const float *rgba,
 439                          unsigned dstx, unsigned dsty,
 440                          unsigned width, unsigned height)
 441 {
 442         union util_color uc;
 443         util_pack_color(rgba, dst->format, &uc);
 444
 445         if(util_format_get_blocksizebits(dst->format) > 32
 446                 || nvfx_surface_fill(pipe, dst, dstx, dsty, width, height, uc.ui))
 447         {
 448                 // TODO: probably should use hardware clear here instead if possible
 449                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 0);
 450                 util_blitter_clear_render_target(blitter, dst, rgba, dstx, dsty, width, height);
 451                 nvfx_put_blitter(pipe, blitter);
 452         }
 453 }
 454
 455 static void
 456 nvfx_clear_depth_stencil(struct pipe_context *pipe,
 457                          struct pipe_surface *dst,
 458                          unsigned clear_flags,
 459                          double depth,
 460                          unsigned stencil,
 461                          unsigned dstx, unsigned dsty,
 462                          unsigned width, unsigned height)
 463 {
 464         if(util_format_get_blocksizebits(dst->format) > 32
 465                 || nvfx_surface_fill(pipe, dst, dstx, dsty, width, height, util_pack_z_stencil(dst->format, depth, stencil)))
 466         {
 467                 // TODO: probably should use hardware clear here instead if possible
 468                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 0);
 469                 util_blitter_clear_depth_stencil(blitter, dst, clear_flags, depth, stencil, dstx, dsty, width, height);
 470                 nvfx_put_blitter(pipe, blitter);
 471         }
 472 }
 473
 474
 475 void
 476 nvfx_init_surface_functions(struct nvfx_context *nvfx)
 477 {
 478         nvfx->pipe.resource_copy_region = nvfx_resource_copy_region;
 479         nvfx->pipe.clear_render_target = nvfx_clear_render_target;
 480         nvfx->pipe.clear_depth_stencil = nvfx_clear_depth_stencil;
 481 }