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 // XXX: must save index buffer too!
 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         struct blitter_context* blitter = nvfx->blitter;
 197         if(!blitter)
 198                 nvfx->blitter = blitter = util_blitter_create(pipe);
 199
 200         util_blitter_save_blend(blitter, nvfx->blend);
 201         util_blitter_save_depth_stencil_alpha(blitter, nvfx->zsa);
 202         util_blitter_save_stencil_ref(blitter, &nvfx->stencil_ref);
 203         util_blitter_save_rasterizer(blitter, nvfx->rasterizer);
 204         util_blitter_save_fragment_shader(blitter, nvfx->fragprog);
 205         util_blitter_save_vertex_shader(blitter, nvfx->vertprog);
 206         util_blitter_save_viewport(blitter, &nvfx->viewport);
 207         util_blitter_save_framebuffer(blitter, &nvfx->framebuffer);
 208         util_blitter_save_clip(blitter, &nvfx->clip);
 209         util_blitter_save_vertex_elements(blitter, nvfx->vtxelt);
 210         util_blitter_save_vertex_buffers(blitter, nvfx->vtxbuf_nr, nvfx->vtxbuf);
 211
 212         if(copy)
 213         {
 214                 util_blitter_save_fragment_sampler_states(blitter, nvfx->nr_samplers, (void**)nvfx->tex_sampler);
 215                 util_blitter_save_fragment_sampler_views(blitter, nvfx->nr_textures, nvfx->fragment_sampler_views);
 216         }
 217
 218         return blitter;
 219 }
 220
 221 static unsigned
 222 nvfx_region_clone(struct nv04_2d_context* ctx, struct nv04_region* rgn, unsigned w, unsigned h, boolean for_read)
 223 {
 224         unsigned begin = nv04_region_begin(rgn, w, h);
 225         unsigned end = nv04_region_end(rgn, w, h);
 226         unsigned size = end - begin;
 227         struct nouveau_bo* bo = 0;
 228         nouveau_bo_new(rgn->bo->device, NOUVEAU_BO_MAP | NOUVEAU_BO_GART, 256, size, &bo);
 229
 230         if(for_read || (size > ((w * h) << rgn->bpps)))
 231                 nv04_memcpy(ctx, bo, 0, rgn->bo, rgn->offset + begin, size);
 232
 233         rgn->bo = bo;
 234         rgn->offset = -begin;
 235         return begin;
 236 }
 237
 238 static void
 239 nvfx_resource_copy_region(struct pipe_context *pipe,
 240                   struct pipe_resource *dstr, struct pipe_subresource subdst,
 241                   unsigned dstx, unsigned dsty, unsigned dstz,
 242                   struct pipe_resource *srcr, struct pipe_subresource subsrc,
 243                   unsigned srcx, unsigned srcy, unsigned srcz,
 244                   unsigned w, unsigned h)
 245 {
 246         static int copy_threshold = -1;
 247         struct nv04_2d_context *ctx = nvfx_screen(pipe->screen)->eng2d;
 248         struct nv04_region dst, src;
 249         int dst_to_gpu;
 250         int src_on_gpu;
 251         boolean small;
 252         int ret;
 253
 254         if(!w || !h)
 255                 return;
 256
 257         if(copy_threshold < 0)
 258                 copy_threshold = debug_get_num_option("NOUVEAU_COPY_THRESHOLD", 4);
 259
 260         dst_to_gpu = dstr->usage != PIPE_USAGE_DYNAMIC && dstr->usage != PIPE_USAGE_STAGING;
 261         src_on_gpu = nvfx_resource_on_gpu(srcr);
 262
 263         nvfx_region_init_for_subresource(&dst, dstr, subdst, dstx, dsty, dstz, TRUE);
 264         nvfx_region_init_for_subresource(&src, srcr, subsrc, srcx, srcy, srcz, FALSE);
 265         w = util_format_get_stride(dstr->format, w) >> dst.bpps;
 266         h = util_format_get_nblocksy(dstr->format, h);
 267
 268         small = (w * h <= copy_threshold);
 269         if((!dst_to_gpu || !src_on_gpu) && small)
 270                 ret = -1; /* use the CPU */
 271         else
 272                 ret = nv04_region_copy_2d(ctx, &dst, &src, w, h,
 273                         dstr->target == PIPE_BUFFER ? -1 : nvfx_surface_format(dstr->format),
 274                         dstr->target == PIPE_BUFFER ? -1 : nv04_scaled_image_format(dstr->format),
 275                         dst_to_gpu, src_on_gpu);
 276         if(!ret)
 277         {}
 278         else if(ret > 0 && dstr->bind & PIPE_BIND_RENDER_TARGET && srcr->bind & PIPE_BIND_SAMPLER_VIEW)
 279         {
 280                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 1);
 281                 util_blitter_copy_region(blitter, dstr, subdst, dstx, dsty, dstz, srcr, subsrc, srcx, srcy, srcz, w, h, TRUE);
 282         }
 283         else
 284         {
 285                 struct nv04_region dstt = dst;
 286                 struct nv04_region srct = src;
 287                 unsigned dstbegin = 0;
 288
 289                 if(!small)
 290                 {
 291                         if(src_on_gpu)
 292                                 nvfx_region_clone(ctx, &srct, w, h, TRUE);
 293
 294                         if(dst_to_gpu)
 295                                 dstbegin = nvfx_region_clone(ctx, &dstt, w, h, FALSE);
 296                 }
 297
 298                 nv04_region_copy_cpu(&dstt, &srct, w, h);
 299
 300                 if(srct.bo != src.bo)
 301                         nouveau_screen_bo_release(pipe->screen, srct.bo);
 302
 303                 if(dstt.bo != dst.bo)
 304                 {
 305                         nv04_memcpy(ctx, dst.bo, dst.offset + dstbegin, dstt.bo, 0, dstt.bo->size);
 306                         nouveau_screen_bo_release(pipe->screen, dstt.bo);
 307                 }
 308         }
 309 }
 310
 311 static int
 312 nvfx_surface_fill(struct pipe_context* pipe, struct pipe_surface *dsts,
 313                   unsigned dx, unsigned dy, unsigned w, unsigned h, unsigned value)
 314 {
 315         struct nv04_2d_context *ctx = nvfx_screen(pipe->screen)->eng2d;
 316         struct nv04_region dst;
 317         int ret;
 318         /* Always try to use the GPU right now, if possible
 319          * If the user wanted the surface data on the CPU, he would have cleared with memset (hopefully) */
 320
 321         // we don't care about interior pixel order since we set all them to the same value
 322         nvfx_region_init_for_surface(&dst, (struct nvfx_surface*)dsts, dx, dy, TRUE);
 323
 324         w = util_format_get_stride(dsts->format, w) >> dst.bpps;
 325         h = util_format_get_nblocksy(dsts->format, h);
 326
 327         ret = nv04_region_fill_2d(ctx, &dst, w, h, value);
 328         if(ret > 0 && dsts->texture->bind & PIPE_BIND_RENDER_TARGET)
 329                 return 1;
 330         else if(ret)
 331         {
 332                 struct nv04_region dstt = dst;
 333                 unsigned dstbegin = 0;
 334
 335                 if(nvfx_resource_on_gpu(dsts->texture))
 336                         dstbegin = nvfx_region_clone(ctx, &dstt, w, h, FALSE);
 337
 338                 nv04_region_fill_cpu(&dstt, w, h, value);
 339
 340                 if(dstt.bo != dst.bo)
 341                 {
 342                         nv04_memcpy(ctx, dst.bo, dst.offset + dstbegin, dstt.bo, 0, dstt.bo->size);
 343                         nouveau_screen_bo_release(pipe->screen, dstt.bo);
 344                 }
 345         }
 346
 347         return 0;
 348 }
 349
 350
 351 void
 352 nvfx_screen_surface_takedown(struct pipe_screen *pscreen)
 353 {
 354         nv04_2d_context_takedown(nvfx_screen(pscreen)->eng2d);
 355         nvfx_screen(pscreen)->eng2d = 0;
 356 }
 357
 358 int
 359 nvfx_screen_surface_init(struct pipe_screen *pscreen)
 360 {
 361         struct nv04_2d_context* ctx = nv04_2d_context_init(nouveau_screen(pscreen)->channel);
 362         if(!ctx)
 363                 return -1;
 364         nvfx_screen(pscreen)->eng2d = ctx;
 365         return 0;
 366 }
 367
 368 static void
 369 nvfx_surface_copy_temp(struct pipe_context* pipe, struct pipe_surface* surf, int to_temp)
 370 {
 371         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 372         struct pipe_subresource tempsr, surfsr;
 373         struct nvfx_context* nvfx = nvfx_context(pipe);
 374
 375         // TODO: we really should do this validation before setting these variable in draw calls
 376         unsigned use_vertex_buffers = nvfx->use_vertex_buffers;
 377         boolean use_index_buffer = nvfx->use_index_buffer;
 378         unsigned base_vertex = nvfx->base_vertex;
 379
 380         tempsr.face = 0;
 381         tempsr.level = 0;
 382         surfsr.face = surf->face;
 383         surfsr.level = surf->level;
 384
 385         if(to_temp)
 386                 nvfx_resource_copy_region(pipe, &ns->temp->base.base, tempsr, 0, 0, 0, surf->texture, surfsr, 0, 0, surf->zslice, surf->width, surf->height);
 387         else
 388                 nvfx_resource_copy_region(pipe, surf->texture, surfsr, 0, 0, surf->zslice, &ns->temp->base.base, tempsr, 0, 0, 0, surf->width, surf->height);
 389
 390         nvfx->use_vertex_buffers = use_vertex_buffers;
 391         nvfx->use_index_buffer = use_index_buffer;
 392         nvfx->base_vertex = base_vertex;
 393
 394         nvfx->dirty |= NVFX_NEW_ARRAYS;
 395         nvfx->draw_dirty |= NVFX_NEW_ARRAYS;
 396 }
 397
 398 void
 399 nvfx_surface_create_temp(struct pipe_context* pipe, struct pipe_surface* surf)
 400 {
 401         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 402         struct pipe_resource template;
 403         memset(&template, 0, sizeof(struct pipe_resource));
 404         template.target = PIPE_TEXTURE_2D;
 405         template.format = surf->format;
 406         template.width0 = surf->width;
 407         template.height0 = surf->height;
 408         template.depth0 = 1;
 409         template.nr_samples = surf->texture->nr_samples;
 410         template.flags = NVFX_RESOURCE_FLAG_LINEAR;
 411
 412         ns->temp = (struct nvfx_miptree*)nvfx_miptree_create(pipe->screen, &template);
 413         nvfx_surface_copy_temp(pipe, surf, 1);
 414 }
 415
 416 void
 417 nvfx_surface_flush(struct pipe_context* pipe, struct pipe_surface* surf)
 418 {
 419         struct nvfx_context* nvfx = (struct nvfx_context*)pipe;
 420         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 421         boolean bound = FALSE;
 422
 423         /* must be done before the copy, otherwise the copy will use the temp as destination */
 424         util_dirty_surface_set_clean(nvfx_surface_get_dirty_surfaces(surf), &ns->base);
 425
 426         nvfx_surface_copy_temp(pipe, surf, 0);
 427
 428         if(nvfx->framebuffer.zsbuf == surf)
 429                 bound = TRUE;
 430         else
 431         {
 432                 for(unsigned i = 0; i < nvfx->framebuffer.nr_cbufs; ++i)
 433                 {
 434                         if(nvfx->framebuffer.cbufs[i] == surf)
 435                         {
 436                                 bound = TRUE;
 437                                 break;
 438                         }
 439                 }
 440         }
 441
 442         if(!bound)
 443                 pipe_resource_reference((struct pipe_resource**)&ns->temp, 0);
 444 }
 445
 446 static void
 447 nvfx_clear_render_target(struct pipe_context *pipe,
 448                          struct pipe_surface *dst,
 449                          const float *rgba,
 450                          unsigned dstx, unsigned dsty,
 451                          unsigned width, unsigned height)
 452 {
 453         union util_color uc;
 454         util_pack_color(rgba, dst->format, &uc);
 455
 456         if(util_format_get_blocksizebits(dst->format) > 32
 457                 || nvfx_surface_fill(pipe, dst, dstx, dsty, width, height, uc.ui))
 458         {
 459                 // TODO: probably should use hardware clear here instead if possible
 460                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 0);
 461                 util_blitter_clear_render_target(blitter, dst, rgba, dstx, dsty, width, height);
 462         }
 463 }
 464
 465 static void
 466 nvfx_clear_depth_stencil(struct pipe_context *pipe,
 467                          struct pipe_surface *dst,
 468                          unsigned clear_flags,
 469                          double depth,
 470                          unsigned stencil,
 471                          unsigned dstx, unsigned dsty,
 472                          unsigned width, unsigned height)
 473 {
 474         if(util_format_get_blocksizebits(dst->format) > 32
 475                 || nvfx_surface_fill(pipe, dst, dstx, dsty, width, height, util_pack_z_stencil(dst->format, depth, stencil)))
 476         {
 477                 // TODO: probably should use hardware clear here instead if possible
 478                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 0);
 479                 util_blitter_clear_depth_stencil(blitter, dst, clear_flags, depth, stencil, dstx, dsty, width, height);
 480         }
 481 }
 482
 483
 484 void
 485 nvfx_init_surface_functions(struct nvfx_context *nvfx)
 486 {
 487         nvfx->pipe.resource_copy_region = nvfx_resource_copy_region;
 488         nvfx->pipe.clear_render_target = nvfx_clear_render_target;
 489         nvfx->pipe.clear_depth_stencil = nvfx_clear_depth_stencil;
 490 }