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         switch(bits)
  51         {
  52         case 8:
  53                 rgn->bpps = 0;
  54                 break;
  55         case 16:
  56                 rgn->bpps = 1;
  57                 break;
  58         case 32:
  59                 rgn->bpps = 2;
  60                 break;
  61         default:
  62                 {
  63                         int shift;
  64                         assert(util_is_power_of_two(bits));
  65                         shift = util_logbase2(bits) - 3;
  66                         assert(shift >= 2);
  67                         rgn->bpps = 2;
  68                         shift -= 2;
  69
  70                         rgn->x = util_format_get_nblocksx(format, rgn->x) << shift;
  71                         rgn->y = util_format_get_nblocksy(format, rgn->y);
  72                 }
  73         }
  74 }
  75
  76 static INLINE void
  77 nvfx_region_fixup_swizzled(struct nv04_region* rgn, unsigned zslice, unsigned width, unsigned height, unsigned depth)
  78 {
  79         // TODO: move this code to surface creation?
  80         if((depth <= 1) && (height <= 1 || width <= 2))
  81                 rgn->pitch = width << rgn->bpps;
  82         else if(depth > 1 && height <= 2 && width <= 2)
  83         {
  84                 rgn->pitch = width << rgn->bpps;
  85                 rgn->offset += (zslice * width * height) << rgn->bpps;
  86         }
  87         else
  88         {
  89                 rgn->pitch = 0;
  90                 rgn->z = zslice;
  91                 rgn->w = width;
  92                 rgn->h = height;
  93                 rgn->d = depth;
  94         }
  95 }
  96
  97 static INLINE void
  98 nvfx_region_init_for_surface(struct nv04_region* rgn, struct nvfx_surface* surf, unsigned x, unsigned y, bool for_write)
  99 {
 100         rgn->x = x;
 101         rgn->y = y;
 102         rgn->z = 0;
 103         nvfx_region_set_format(rgn, surf->base.base.format);
 104
 105         if(surf->temp)
 106         {
 107                 rgn->bo = surf->temp->base.bo;
 108                 rgn->offset = 0;
 109                 rgn->pitch = surf->temp->linear_pitch;
 110
 111                 if(for_write)
 112                         util_dirty_surface_set_dirty(nvfx_surface_get_dirty_surfaces(&surf->base.base), &surf->base);
 113         } else {
 114                 rgn->bo = ((struct nvfx_resource*)surf->base.base.texture)->bo;
 115                 rgn->offset = surf->base.base.offset;
 116                 rgn->pitch = surf->pitch;
 117
 118                 if(!(surf->base.base.texture->flags & NVFX_RESOURCE_FLAG_LINEAR))
 119                         nvfx_region_fixup_swizzled(rgn, surf->base.base.zslice, surf->base.base.width, surf->base.base.height, u_minify(surf->base.base.texture->depth0, surf->base.base.level));
 120         }
 121 }
 122
 123 static INLINE void
 124 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)
 125 {
 126         if(pt->target != PIPE_BUFFER)
 127         {
 128                 struct nvfx_surface* ns = (struct nvfx_surface*)util_surfaces_peek(&((struct nvfx_miptree*)pt)->surfaces, pt, sub.face, sub.level, z);
 129                 if(ns && util_dirty_surface_is_dirty(&ns->base))
 130                 {
 131                         nvfx_region_init_for_surface(rgn, ns, x, y, for_write);
 132                         return;
 133                 }
 134         }
 135
 136         rgn->bo = ((struct nvfx_resource*)pt)->bo;
 137         rgn->offset = nvfx_subresource_offset(pt, sub.face, sub.level, z);
 138         rgn->pitch = nvfx_subresource_pitch(pt, sub.level);
 139         rgn->x = x;
 140         rgn->y = y;
 141         rgn->z = 0;
 142
 143         nvfx_region_set_format(rgn, pt->format);
 144         if(!(pt->flags & NVFX_RESOURCE_FLAG_LINEAR))
 145                 nvfx_region_fixup_swizzled(rgn, z, u_minify(pt->width0, sub.level), u_minify(pt->height0, sub.level), u_minify(pt->depth0, sub.level));
 146 }
 147
 148 // TODO: actually test this for all formats, it's probably wrong for some...
 149
 150 static INLINE int
 151 nvfx_surface_format(enum pipe_format format)
 152 {
 153         switch(util_format_get_blocksize(format)) {
 154         case 1:
 155                 return NV04_CONTEXT_SURFACES_2D_FORMAT_Y8;
 156         case 2:
 157                 //return NV04_CONTEXT_SURFACES_2D_FORMAT_Y16;
 158                 return NV04_CONTEXT_SURFACES_2D_FORMAT_R5G6B5;
 159         case 4:
 160                 //if(format == PIPE_FORMAT_B8G8R8X8_UNORM || format == PIPE_FORMAT_B8G8R8A8_UNORM)
 161                         return NV04_CONTEXT_SURFACES_2D_FORMAT_A8R8G8B8;
 162                 //else
 163                 //      return NV04_CONTEXT_SURFACES_2D_FORMAT_Y32;
 164         default:
 165                 return -1;
 166         }
 167 }
 168
 169 static INLINE int
 170 nv04_scaled_image_format(enum pipe_format format)
 171 {
 172         switch(util_format_get_blocksize(format)) {
 173         case 1:
 174                 return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_Y8;
 175         case 2:
 176                 //if(format == PIPE_FORMAT_B5G5R5A1_UNORM)
 177                 //      return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_A1R5G5B5;
 178                 //else
 179                         return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_R5G6B5;
 180         case 4:
 181                 if(format == PIPE_FORMAT_B8G8R8X8_UNORM)
 182                         return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_X8R8G8B8;
 183                 else
 184                         return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_A8R8G8B8;
 185         default:
 186                 return -1;
 187         }
 188 }
 189
 190 // don't save index buffer because blitter doesn't setit
 191 static struct blitter_context*
 192 nvfx_get_blitter(struct pipe_context* pipe, int copy)
 193 {
 194         struct nvfx_context* nvfx = nvfx_context(pipe);
 195
 196         assert(nvfx->blitters_in_use < Elements(nvfx->blitter));
 197
 198         struct blitter_context** pblitter = &nvfx->blitter[nvfx->blitters_in_use++];
 199         if(!*pblitter)
 200                 *pblitter = util_blitter_create(pipe);
 201         struct blitter_context* blitter = *pblitter;
 202
 203         util_blitter_save_blend(blitter, nvfx->blend);
 204         util_blitter_save_depth_stencil_alpha(blitter, nvfx->zsa);
 205         util_blitter_save_stencil_ref(blitter, &nvfx->stencil_ref);
 206         util_blitter_save_rasterizer(blitter, nvfx->rasterizer);
 207         util_blitter_save_fragment_shader(blitter, nvfx->fragprog);
 208         util_blitter_save_vertex_shader(blitter, nvfx->vertprog);
 209         util_blitter_save_viewport(blitter, &nvfx->viewport);
 210         util_blitter_save_framebuffer(blitter, &nvfx->framebuffer);
 211         util_blitter_save_clip(blitter, &nvfx->clip);
 212         util_blitter_save_vertex_elements(blitter, nvfx->vtxelt);
 213         util_blitter_save_vertex_buffers(blitter, nvfx->vtxbuf_nr, nvfx->vtxbuf);
 214
 215         if(copy)
 216         {
 217                 util_blitter_save_fragment_sampler_states(blitter, nvfx->nr_samplers, (void**)nvfx->tex_sampler);
 218                 util_blitter_save_fragment_sampler_views(blitter, nvfx->nr_textures, nvfx->fragment_sampler_views);
 219         }
 220
 221         return blitter;
 222 }
 223
 224 static inline void
 225 nvfx_put_blitter(struct pipe_context* pipe, struct blitter_context* blitter)
 226 {
 227         struct nvfx_context* nvfx = nvfx_context(pipe);
 228         --nvfx->blitters_in_use;
 229         assert(nvfx->blitters_in_use >= 0);
 230 }
 231
 232 static unsigned
 233 nvfx_region_clone(struct nv04_2d_context* ctx, struct nv04_region* rgn, unsigned w, unsigned h, boolean for_read)
 234 {
 235         unsigned begin = nv04_region_begin(rgn, w, h);
 236         unsigned end = nv04_region_end(rgn, w, h);
 237         unsigned size = end - begin;
 238         struct nouveau_bo* bo = 0;
 239         nouveau_bo_new(rgn->bo->device, NOUVEAU_BO_MAP | NOUVEAU_BO_GART, 256, size, &bo);
 240
 241         if(for_read || (size > ((w * h) << rgn->bpps)))
 242                 nv04_memcpy(ctx, bo, 0, rgn->bo, rgn->offset + begin, size);
 243
 244         rgn->bo = bo;
 245         rgn->offset = -begin;
 246         return begin;
 247 }
 248
 249 static void
 250 nvfx_resource_copy_region(struct pipe_context *pipe,
 251                   struct pipe_resource *dstr, struct pipe_subresource subdst,
 252                   unsigned dstx, unsigned dsty, unsigned dstz,
 253                   struct pipe_resource *srcr, struct pipe_subresource subsrc,
 254                   unsigned srcx, unsigned srcy, unsigned srcz,
 255                   unsigned w, unsigned h)
 256 {
 257         static int copy_threshold = -1;
 258         struct nv04_2d_context *ctx = nvfx_screen(pipe->screen)->eng2d;
 259         struct nv04_region dst, src;
 260         int dst_to_gpu;
 261         int src_on_gpu;
 262         boolean small;
 263         int ret;
 264
 265         if(!w || !h)
 266                 return;
 267
 268         if(copy_threshold < 0)
 269                 copy_threshold = debug_get_num_option("NOUVEAU_COPY_THRESHOLD", 4);
 270
 271         dst_to_gpu = dstr->usage != PIPE_USAGE_DYNAMIC && dstr->usage != PIPE_USAGE_STAGING;
 272         src_on_gpu = nvfx_resource_on_gpu(srcr);
 273
 274         nvfx_region_init_for_subresource(&dst, dstr, subdst, dstx, dsty, dstz, TRUE);
 275         nvfx_region_init_for_subresource(&src, srcr, subsrc, srcx, srcy, srcz, FALSE);
 276         w = util_format_get_stride(dstr->format, w) >> dst.bpps;
 277         h = util_format_get_nblocksy(dstr->format, h);
 278
 279         small = (w * h <= copy_threshold);
 280         if((!dst_to_gpu || !src_on_gpu) && small)
 281                 ret = -1; /* use the CPU */
 282         else
 283                 ret = nv04_region_copy_2d(ctx, &dst, &src, w, h,
 284                         dstr->target == PIPE_BUFFER ? -1 : nvfx_surface_format(dstr->format),
 285                         dstr->target == PIPE_BUFFER ? -1 : nv04_scaled_image_format(dstr->format),
 286                         dst_to_gpu, src_on_gpu);
 287         if(!ret)
 288         {}
 289         else if(ret > 0 && dstr->bind & PIPE_BIND_RENDER_TARGET && srcr->bind & PIPE_BIND_SAMPLER_VIEW)
 290         {
 291                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 1);
 292                 util_blitter_copy_region(blitter, dstr, subdst, dstx, dsty, dstz, srcr, subsrc, srcx, srcy, srcz, w, h, TRUE);
 293                 nvfx_put_blitter(pipe, blitter);
 294         }
 295         else
 296         {
 297                 struct nv04_region dstt = dst;
 298                 struct nv04_region srct = src;
 299                 unsigned dstbegin = 0;
 300
 301                 if(!small)
 302                 {
 303                         if(src_on_gpu)
 304                                 nvfx_region_clone(ctx, &srct, w, h, TRUE);
 305
 306                         if(dst_to_gpu)
 307                                 dstbegin = nvfx_region_clone(ctx, &dstt, w, h, FALSE);
 308                 }
 309
 310                 nv04_region_copy_cpu(&dstt, &srct, w, h);
 311
 312                 if(srct.bo != src.bo)
 313                         nouveau_screen_bo_release(pipe->screen, srct.bo);
 314
 315                 if(dstt.bo != dst.bo)
 316                 {
 317                         nv04_memcpy(ctx, dst.bo, dst.offset + dstbegin, dstt.bo, 0, dstt.bo->size);
 318                         nouveau_screen_bo_release(pipe->screen, dstt.bo);
 319                 }
 320         }
 321 }
 322
 323 static int
 324 nvfx_surface_fill(struct pipe_context* pipe, struct pipe_surface *dsts,
 325                   unsigned dx, unsigned dy, unsigned w, unsigned h, unsigned value)
 326 {
 327         struct nv04_2d_context *ctx = nvfx_screen(pipe->screen)->eng2d;
 328         struct nv04_region dst;
 329         int ret;
 330         /* Always try to use the GPU right now, if possible
 331          * If the user wanted the surface data on the CPU, he would have cleared with memset (hopefully) */
 332
 333         // we don't care about interior pixel order since we set all them to the same value
 334         nvfx_region_init_for_surface(&dst, (struct nvfx_surface*)dsts, dx, dy, TRUE);
 335
 336         w = util_format_get_stride(dsts->format, w) >> dst.bpps;
 337         h = util_format_get_nblocksy(dsts->format, h);
 338
 339         ret = nv04_region_fill_2d(ctx, &dst, w, h, value);
 340         if(ret > 0 && dsts->texture->bind & PIPE_BIND_RENDER_TARGET)
 341                 return 1;
 342         else if(ret)
 343         {
 344                 struct nv04_region dstt = dst;
 345                 unsigned dstbegin = 0;
 346
 347                 if(nvfx_resource_on_gpu(dsts->texture))
 348                         dstbegin = nvfx_region_clone(ctx, &dstt, w, h, FALSE);
 349
 350                 nv04_region_fill_cpu(&dstt, w, h, value);
 351
 352                 if(dstt.bo != dst.bo)
 353                 {
 354                         nv04_memcpy(ctx, dst.bo, dst.offset + dstbegin, dstt.bo, 0, dstt.bo->size);
 355                         nouveau_screen_bo_release(pipe->screen, dstt.bo);
 356                 }
 357         }
 358
 359         return 0;
 360 }
 361
 362
 363 void
 364 nvfx_screen_surface_takedown(struct pipe_screen *pscreen)
 365 {
 366         nv04_2d_context_takedown(nvfx_screen(pscreen)->eng2d);
 367         nvfx_screen(pscreen)->eng2d = 0;
 368 }
 369
 370 int
 371 nvfx_screen_surface_init(struct pipe_screen *pscreen)
 372 {
 373         struct nv04_2d_context* ctx = nv04_2d_context_init(nouveau_screen(pscreen)->channel);
 374         if(!ctx)
 375                 return -1;
 376         nvfx_screen(pscreen)->eng2d = ctx;
 377         return 0;
 378 }
 379
 380 static void
 381 nvfx_surface_copy_temp(struct pipe_context* pipe, struct pipe_surface* surf, int to_temp)
 382 {
 383         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 384         struct pipe_subresource tempsr, surfsr;
 385         struct nvfx_context* nvfx = nvfx_context(pipe);
 386
 387         /* temporarily detach the temp, so it isn't used in place of the actual resource */
 388         struct nvfx_miptree* temp = ns->temp;
 389         ns->temp = 0;
 390
 391         // TODO: we really should do this validation before setting these variable in draw calls
 392         unsigned use_vertex_buffers = nvfx->use_vertex_buffers;
 393         boolean use_index_buffer = nvfx->use_index_buffer;
 394         unsigned base_vertex = nvfx->base_vertex;
 395
 396         tempsr.face = 0;
 397         tempsr.level = 0;
 398         surfsr.face = surf->face;
 399         surfsr.level = surf->level;
 400
 401         if(to_temp)
 402                 nvfx_resource_copy_region(pipe, &temp->base.base, tempsr, 0, 0, 0, surf->texture, surfsr, 0, 0, surf->zslice, surf->width, surf->height);
 403         else
 404                 nvfx_resource_copy_region(pipe, surf->texture, surfsr, 0, 0, surf->zslice, &temp->base.base, tempsr, 0, 0, 0, surf->width, surf->height);
 405
 406         /* If this triggers, it probably means we attempted to use the blitter
 407          * but failed due to non-renderability of the target.
 408          * Obviously, this would lead to infinite recursion if supported. */
 409         assert(!ns->temp);
 410
 411         ns->temp = temp;
 412
 413         nvfx->use_vertex_buffers = use_vertex_buffers;
 414         nvfx->use_index_buffer = use_index_buffer;
 415         nvfx->base_vertex = base_vertex;
 416
 417         nvfx->dirty |= NVFX_NEW_ARRAYS;
 418         nvfx->draw_dirty |= NVFX_NEW_ARRAYS;
 419 }
 420
 421 void
 422 nvfx_surface_create_temp(struct pipe_context* pipe, struct pipe_surface* surf)
 423 {
 424         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 425         struct pipe_resource template;
 426         memset(&template, 0, sizeof(struct pipe_resource));
 427         template.target = PIPE_TEXTURE_2D;
 428         template.format = surf->format;
 429         template.width0 = surf->width;
 430         template.height0 = surf->height;
 431         template.depth0 = 1;
 432         template.nr_samples = surf->texture->nr_samples;
 433         template.flags = NVFX_RESOURCE_FLAG_LINEAR;
 434
 435         assert(!ns->temp && !util_dirty_surface_is_dirty(&ns->base));
 436
 437         ns->temp = (struct nvfx_miptree*)nvfx_miptree_create(pipe->screen, &template);
 438         nvfx_surface_copy_temp(pipe, surf, 1);
 439 }
 440
 441 void
 442 nvfx_surface_flush(struct pipe_context* pipe, struct pipe_surface* surf)
 443 {
 444         struct nvfx_context* nvfx = (struct nvfx_context*)pipe;
 445         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 446         boolean bound = FALSE;
 447
 448         nvfx_surface_copy_temp(pipe, surf, 0);
 449
 450         util_dirty_surface_set_clean(nvfx_surface_get_dirty_surfaces(surf), &ns->base);
 451
 452         if(nvfx->framebuffer.zsbuf == surf)
 453                 bound = TRUE;
 454         else
 455         {
 456                 for(unsigned i = 0; i < nvfx->framebuffer.nr_cbufs; ++i)
 457                 {
 458                         if(nvfx->framebuffer.cbufs[i] == surf)
 459                         {
 460                                 bound = TRUE;
 461                                 break;
 462                         }
 463                 }
 464         }
 465
 466         if(!bound)
 467                 pipe_resource_reference((struct pipe_resource**)&ns->temp, 0);
 468 }
 469
 470 static void
 471 nvfx_clear_render_target(struct pipe_context *pipe,
 472                          struct pipe_surface *dst,
 473                          const float *rgba,
 474                          unsigned dstx, unsigned dsty,
 475                          unsigned width, unsigned height)
 476 {
 477         union util_color uc;
 478         util_pack_color(rgba, dst->format, &uc);
 479
 480         if(util_format_get_blocksizebits(dst->format) > 32
 481                 || nvfx_surface_fill(pipe, dst, dstx, dsty, width, height, uc.ui))
 482         {
 483                 // TODO: probably should use hardware clear here instead if possible
 484                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 0);
 485                 util_blitter_clear_render_target(blitter, dst, rgba, dstx, dsty, width, height);
 486                 nvfx_put_blitter(pipe, blitter);
 487         }
 488 }
 489
 490 static void
 491 nvfx_clear_depth_stencil(struct pipe_context *pipe,
 492                          struct pipe_surface *dst,
 493                          unsigned clear_flags,
 494                          double depth,
 495                          unsigned stencil,
 496                          unsigned dstx, unsigned dsty,
 497                          unsigned width, unsigned height)
 498 {
 499         if(util_format_get_blocksizebits(dst->format) > 32
 500                 || nvfx_surface_fill(pipe, dst, dstx, dsty, width, height, util_pack_z_stencil(dst->format, depth, stencil)))
 501         {
 502                 // TODO: probably should use hardware clear here instead if possible
 503                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 0);
 504                 util_blitter_clear_depth_stencil(blitter, dst, clear_flags, depth, stencil, dstx, dsty, width, height);
 505                 nvfx_put_blitter(pipe, blitter);
 506         }
 507 }
 508
 509
 510 void
 511 nvfx_init_surface_functions(struct nvfx_context *nvfx)
 512 {
 513         nvfx->pipe.resource_copy_region = nvfx_resource_copy_region;
 514         nvfx->pipe.clear_render_target = nvfx_clear_render_target;
 515         nvfx->pipe.clear_depth_stencil = nvfx_clear_depth_stencil;
 516 }