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_rect.h"
  36 #include "util/u_blitter.h"
  37
  38 #include "nouveau/nouveau_winsys.h"
  39 #include "nouveau/nouveau_screen.h"
  40 #include "nvfx_context.h"
  41 #include "nvfx_screen.h"
  42 #include "nvfx_resource.h"
  43 #include "nv04_2d.h"
  44
  45 #include <nouveau/nouveau_bo.h>
  46
  47 static INLINE void
  48 nvfx_region_set_format(struct nv04_region* rgn, enum pipe_format format)
  49 {
  50         unsigned bits = util_format_get_blocksizebits(format);
  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         default:
  63                 assert(util_is_pot(bits));
  64                 int shift = util_logbase2(bits) - 3;
  65                 assert(shift >= 2);
  66                 rgn->bpps = 2;
  67                 shift -= 2;
  68
  69                 rgn->x = util_format_get_nblocksx(format, rgn->x) << shift;
  70                 rgn->y = util_format_get_nblocksy(format, rgn->y);
  71         }
  72 }
  73
  74 static INLINE void
  75 nvfx_region_fixup_swizzled(struct nv04_region* rgn, unsigned zslice, unsigned width, unsigned height, unsigned depth)
  76 {
  77         // TODO: move this code to surface creation?
  78         if((depth <= 1) && (height <= 1 || width <= 2))
  79                 rgn->pitch = width << rgn->bpps;
  80         else if(depth > 1 && height <= 2 && width <= 2)
  81         {
  82                 rgn->pitch = width << rgn->bpps;
  83                 rgn->offset += (zslice * width * height) << rgn->bpps;
  84         }
  85         else
  86         {
  87                 rgn->pitch = 0;
  88                 rgn->z = zslice;
  89                 rgn->w = width;
  90                 rgn->h = height;
  91                 rgn->d = depth;
  92         }
  93 }
  94
  95 static INLINE void
  96 nvfx_region_init_for_surface(struct nv04_region* rgn, struct nvfx_surface* surf, unsigned x, unsigned y, bool for_write)
  97 {
  98         rgn->x = x;
  99         rgn->y = y;
 100         rgn->z = 0;
 101         nvfx_region_set_format(rgn, surf->base.base.format);
 102
 103         if(surf->temp)
 104         {
 105                 rgn->bo = surf->temp->base.bo;
 106                 rgn->offset = 0;
 107                 rgn->pitch = surf->temp->linear_pitch;
 108
 109                 if(for_write)
 110                         util_dirty_surface_set_dirty(nvfx_surface_get_dirty_surfaces(&surf->base.base), &surf->base);
 111         } else {
 112                 rgn->bo = ((struct nvfx_resource*)surf->base.base.texture)->bo;
 113                 rgn->offset = surf->base.base.offset;
 114                 rgn->pitch = surf->pitch;
 115
 116                 if(!(surf->base.base.texture->flags & NVFX_RESOURCE_FLAG_LINEAR))
 117                         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));
 118         }
 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->pitch = nvfx_subresource_pitch(pt, sub.level);
 137         rgn->x = x;
 138         rgn->y = y;
 139         rgn->z = 0;
 140
 141         nvfx_region_set_format(rgn, pt->format);
 142         if(!(pt->flags & NVFX_RESOURCE_FLAG_LINEAR))
 143                 nvfx_region_fixup_swizzled(rgn, z, u_minify(pt->width0, sub.level), u_minify(pt->height0, sub.level), u_minify(pt->depth0, sub.level));
 144 }
 145
 146 // TODO: actually test this for all formats, it's probably wrong for some...
 147
 148 static INLINE int
 149 nvfx_surface_format(enum pipe_format format)
 150 {
 151         switch(util_format_get_blocksize(format)) {
 152         case 1:
 153                 return NV04_CONTEXT_SURFACES_2D_FORMAT_Y8;
 154         case 2:
 155                 //return NV04_CONTEXT_SURFACES_2D_FORMAT_Y16;
 156                 return NV04_CONTEXT_SURFACES_2D_FORMAT_R5G6B5;
 157         case 4:
 158                 //if(format == PIPE_FORMAT_B8G8R8X8_UNORM || format == PIPE_FORMAT_B8G8R8A8_UNORM)
 159                         return NV04_CONTEXT_SURFACES_2D_FORMAT_A8R8G8B8;
 160                 //else
 161                 //      return NV04_CONTEXT_SURFACES_2D_FORMAT_Y32;
 162         default:
 163                 return -1;
 164         }
 165 }
 166
 167 static INLINE int
 168 nv04_scaled_image_format(enum pipe_format format)
 169 {
 170         switch(util_format_get_blocksize(format)) {
 171         case 1:
 172                 return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_Y8;
 173         case 2:
 174                 //if(format == PIPE_FORMAT_B5G5R5A1_UNORM)
 175                 //      return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_A1R5G5B5;
 176                 //else
 177                         return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_R5G6B5;
 178         case 4:
 179                 if(format == PIPE_FORMAT_B8G8R8X8_UNORM)
 180                         return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_X8R8G8B8;
 181                 else
 182                         return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_A8R8G8B8;
 183         default:
 184                 return -1;
 185         }
 186 }
 187
 188 // XXX: must save index buffer too!
 189 static struct blitter_context*
 190 nvfx_get_blitter(struct pipe_context* pipe, int copy)
 191 {
 192         struct nvfx_context* nvfx = nvfx_context(pipe);
 193
 194         struct blitter_context* blitter = nvfx->blitter;
 195         if(!blitter)
 196                 nvfx->blitter = blitter = util_blitter_create(pipe);
 197
 198         util_blitter_save_blend(blitter, nvfx->blend);
 199         util_blitter_save_depth_stencil_alpha(blitter, nvfx->zsa);
 200         util_blitter_save_stencil_ref(blitter, &nvfx->stencil_ref);
 201         util_blitter_save_rasterizer(blitter, nvfx->rasterizer);
 202         util_blitter_save_fragment_shader(blitter, nvfx->fragprog);
 203         util_blitter_save_vertex_shader(blitter, nvfx->vertprog);
 204         util_blitter_save_viewport(blitter, &nvfx->viewport);
 205         util_blitter_save_framebuffer(blitter, &nvfx->framebuffer);
 206         util_blitter_save_clip(blitter, &nvfx->clip);
 207         util_blitter_save_vertex_elements(blitter, nvfx->vtxelt);
 208         util_blitter_save_vertex_buffers(blitter, nvfx->vtxbuf_nr, nvfx->vtxbuf);
 209
 210         if(copy)
 211         {
 212                 util_blitter_save_fragment_sampler_states(blitter, nvfx->nr_samplers, (void**)nvfx->tex_sampler);
 213                 util_blitter_save_fragment_sampler_views(blitter, nvfx->nr_textures, nvfx->fragment_sampler_views);
 214         }
 215
 216         return blitter;
 217 }
 218
 219 static unsigned
 220 nvfx_region_clone(struct nv04_2d_context* ctx, struct nv04_region* rgn, unsigned w, unsigned h, boolean for_read)
 221 {
 222         unsigned begin = nv04_region_begin(rgn, w, h);
 223         unsigned end = nv04_region_end(rgn, w, h);
 224         unsigned size = end - begin;
 225         struct nouveau_bo* bo = 0;
 226         nouveau_bo_new(rgn->bo->device, NOUVEAU_BO_MAP | NOUVEAU_BO_GART, 256, size, &bo);
 227
 228         if(for_read || (size > ((w * h) << rgn->bpps)))
 229                 nv04_memcpy(ctx, bo, 0, rgn->bo, rgn->offset + begin, size);
 230
 231         rgn->bo = bo;
 232         rgn->offset = -begin;
 233         return begin;
 234 }
 235
 236 static void
 237 nvfx_resource_copy_region(struct pipe_context *pipe,
 238                   struct pipe_resource *dstr, struct pipe_subresource subdst,
 239                   unsigned dstx, unsigned dsty, unsigned dstz,
 240                   struct pipe_resource *srcr, struct pipe_subresource subsrc,
 241                   unsigned srcx, unsigned srcy, unsigned srcz,
 242                   unsigned w, unsigned h)
 243 {
 244         struct nv04_2d_context *ctx = nvfx_screen(pipe->screen)->eng2d;
 245         struct nv04_region dst, src;
 246
 247         if(!w || !h)
 248                 return;
 249
 250         static int copy_threshold = -1;
 251         if(copy_threshold < 0)
 252                 copy_threshold = debug_get_num_option("NOUVEAU_COPY_THRESHOLD", 4);
 253
 254         int dst_to_gpu = dstr->usage != PIPE_USAGE_DYNAMIC && dstr->usage != PIPE_USAGE_STAGING;
 255         int src_on_gpu = nvfx_resource_on_gpu(srcr);
 256
 257         nvfx_region_init_for_subresource(&dst, dstr, subdst, dstx, dsty, dstz, TRUE);
 258         nvfx_region_init_for_subresource(&src, srcr, subsrc, srcx, srcy, srcz, FALSE);
 259         w = util_format_get_stride(dstr->format, w) >> dst.bpps;
 260         h = util_format_get_nblocksy(dstr->format, h);
 261
 262         int ret;
 263         boolean small = (w * h <= copy_threshold);
 264         if((!dst_to_gpu || !src_on_gpu) && small)
 265                 ret = -1; /* use the CPU */
 266         else
 267                 ret = nv04_region_copy_2d(ctx, &dst, &src, w, h,
 268                         dstr->target == PIPE_BUFFER ? -1 : nvfx_surface_format(dstr->format),
 269                         dstr->target == PIPE_BUFFER ? -1 : nv04_scaled_image_format(dstr->format),
 270                         dst_to_gpu, src_on_gpu);
 271         if(!ret)
 272         {}
 273         else if(ret > 0 && dstr->bind & PIPE_BIND_RENDER_TARGET && srcr->bind & PIPE_BIND_SAMPLER_VIEW)
 274         {
 275                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 1);
 276                 util_blitter_copy_region(blitter, dstr, subdst, dstx, dsty, dstz, srcr, subsrc, srcx, srcy, srcz, w, h, TRUE);
 277         }
 278         else
 279         {
 280                 struct nv04_region dstt = dst;
 281                 struct nv04_region srct = src;
 282                 unsigned dstbegin = 0;
 283
 284                 if(!small)
 285                 {
 286                         if(src_on_gpu)
 287                                 nvfx_region_clone(ctx, &srct, w, h, TRUE);
 288
 289                         if(dst_to_gpu)
 290                                 dstbegin = nvfx_region_clone(ctx, &dstt, w, h, FALSE);
 291                 }
 292
 293                 nv04_region_copy_cpu(&dstt, &srct, w, h);
 294
 295                 if(srct.bo != src.bo)
 296                         nouveau_screen_bo_release(pipe->screen, srct.bo);
 297
 298                 if(dstt.bo != dst.bo)
 299                 {
 300                         nv04_memcpy(ctx, dst.bo, dst.offset + dstbegin, dstt.bo, 0, dstt.bo->size);
 301                         nouveau_screen_bo_release(pipe->screen, dstt.bo);
 302                 }
 303         }
 304 }
 305
 306 static int
 307 nvfx_surface_fill(struct pipe_context* pipe, struct pipe_surface *dsts,
 308                   unsigned dx, unsigned dy, unsigned w, unsigned h, unsigned value)
 309 {
 310         struct nv04_2d_context *ctx = nvfx_screen(pipe->screen)->eng2d;
 311         struct nv04_region dst;
 312         /* Always try to use the GPU right now, if possible
 313          * If the user wanted the surface data on the CPU, he would have cleared with memset (hopefully) */
 314
 315         // we don't care about interior pixel order since we set all them to the same value
 316         nvfx_region_init_for_surface(&dst, (struct nvfx_surface*)dsts, dx, dy, TRUE);
 317
 318         w = util_format_get_stride(dsts->format, w) >> dst.bpps;
 319         h = util_format_get_nblocksy(dsts->format, h);
 320
 321         int ret = nv04_region_fill_2d(ctx, &dst, w, h, value);
 322         if(ret > 0 && dsts->texture->bind & PIPE_BIND_RENDER_TARGET)
 323                 return 1;
 324         else if(ret)
 325         {
 326                 struct nv04_region dstt = dst;
 327                 unsigned dstbegin = 0;
 328
 329                 if(nvfx_resource_on_gpu(dsts->texture))
 330                         dstbegin = nvfx_region_clone(ctx, &dstt, w, h, FALSE);
 331
 332                 nv04_region_fill_cpu(&dstt, w, h, value);
 333
 334                 if(dstt.bo != dst.bo)
 335                 {
 336                         nv04_memcpy(ctx, dst.bo, dst.offset + dstbegin, dstt.bo, 0, dstt.bo->size);
 337                         nouveau_screen_bo_release(pipe->screen, dstt.bo);
 338                 }
 339         }
 340
 341         return 0;
 342 }
 343
 344
 345 void
 346 nvfx_screen_surface_takedown(struct pipe_screen *pscreen)
 347 {
 348         nv04_2d_context_takedown(nvfx_screen(pscreen)->eng2d);
 349         nvfx_screen(pscreen)->eng2d = 0;
 350 }
 351
 352 int
 353 nvfx_screen_surface_init(struct pipe_screen *pscreen)
 354 {
 355         struct nv04_2d_context* ctx = nv04_2d_context_init(nouveau_screen(pscreen)->channel);
 356         if(!ctx)
 357                 return -1;
 358         nvfx_screen(pscreen)->eng2d = ctx;
 359         return 0;
 360 }
 361
 362 static void
 363 nvfx_surface_copy_temp(struct pipe_context* pipe, struct pipe_surface* surf, int to_temp)
 364 {
 365         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 366         struct pipe_subresource tempsr, surfsr;
 367         struct nvfx_context* nvfx = nvfx_context(pipe);
 368
 369         // TODO: we really should do this validation before setting these variable in draw calls
 370         unsigned use_vertex_buffers = nvfx->use_vertex_buffers;
 371         boolean use_index_buffer = nvfx->use_index_buffer;
 372         unsigned base_vertex = nvfx->base_vertex;
 373
 374         tempsr.face = 0;
 375         tempsr.level = 0;
 376         surfsr.face = surf->face;
 377         surfsr.level = surf->level;
 378
 379         if(to_temp)
 380                 nvfx_resource_copy_region(pipe, &ns->temp->base.base, tempsr, 0, 0, 0, surf->texture, surfsr, 0, 0, surf->zslice, surf->width, surf->height);
 381         else
 382                 nvfx_resource_copy_region(pipe, surf->texture, surfsr, 0, 0, surf->zslice, &ns->temp->base.base, tempsr, 0, 0, 0, surf->width, surf->height);
 383
 384         nvfx->use_vertex_buffers = use_vertex_buffers;
 385         nvfx->use_index_buffer = use_index_buffer;
 386         nvfx->base_vertex = base_vertex;
 387
 388         nvfx->dirty |= NVFX_NEW_ARRAYS;
 389         nvfx->draw_dirty |= NVFX_NEW_ARRAYS;
 390 }
 391
 392 void
 393 nvfx_surface_create_temp(struct pipe_context* pipe, struct pipe_surface* surf)
 394 {
 395         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 396         struct pipe_resource template;
 397         memset(&template, 0, sizeof(struct pipe_resource));
 398         template.target = PIPE_TEXTURE_2D;
 399         template.format = surf->format;
 400         template.width0 = surf->width;
 401         template.height0 = surf->height;
 402         template.depth0 = 1;
 403         template.nr_samples = surf->texture->nr_samples;
 404         template.flags = NVFX_RESOURCE_FLAG_LINEAR;
 405
 406         ns->temp = (struct nvfx_miptree*)nvfx_miptree_create(pipe->screen, &template);
 407         nvfx_surface_copy_temp(pipe, surf, 1);
 408 }
 409
 410 void
 411 nvfx_surface_flush(struct pipe_context* pipe, struct pipe_surface* surf)
 412 {
 413         struct nvfx_context* nvfx = (struct nvfx_context*)pipe;
 414         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 415         boolean bound = FALSE;
 416
 417         /* must be done before the copy, otherwise the copy will use the temp as destination */
 418         util_dirty_surface_set_clean(nvfx_surface_get_dirty_surfaces(surf), &ns->base);
 419
 420         nvfx_surface_copy_temp(pipe, surf, 0);
 421
 422         if(nvfx->framebuffer.zsbuf == surf)
 423                 bound = TRUE;
 424         else
 425         {
 426                 for(unsigned i = 0; i < nvfx->framebuffer.nr_cbufs; ++i)
 427                 {
 428                         if(nvfx->framebuffer.cbufs[i] == surf)
 429                         {
 430                                 bound = TRUE;
 431                                 break;
 432                         }
 433                 }
 434         }
 435
 436         if(!bound)
 437                 pipe_resource_reference((struct pipe_resource**)&ns->temp, 0);
 438 }
 439
 440 static void
 441 nvfx_clear_render_target(struct pipe_context *pipe,
 442                          struct pipe_surface *dst,
 443                          const float *rgba,
 444                          unsigned dstx, unsigned dsty,
 445                          unsigned width, unsigned height)
 446 {
 447         union util_color uc;
 448         util_pack_color(rgba, dst->format, &uc);
 449
 450         if(util_format_get_blocksizebits(dst->format) > 32
 451                 || nvfx_surface_fill(pipe, dst, dstx, dsty, width, height, uc.ui))
 452         {
 453                 // TODO: probably should use hardware clear here instead if possible
 454                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 0);
 455                 util_blitter_clear_render_target(blitter, dst, rgba, dstx, dsty, width, height);
 456         }
 457 }
 458
 459 static void
 460 nvfx_clear_depth_stencil(struct pipe_context *pipe,
 461                          struct pipe_surface *dst,
 462                          unsigned clear_flags,
 463                          double depth,
 464                          unsigned stencil,
 465                          unsigned dstx, unsigned dsty,
 466                          unsigned width, unsigned height)
 467 {
 468         if(util_format_get_blocksizebits(dst->format) > 32
 469                 || nvfx_surface_fill(pipe, dst, dstx, dsty, width, height, util_pack_z_stencil(dst->format, depth, stencil)))
 470         {
 471                 // TODO: probably should use hardware clear here instead if possible
 472                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 0);
 473                 util_blitter_clear_depth_stencil(blitter, dst, clear_flags, depth, stencil, dstx, dsty, width, height);
 474         }
 475 }
 476
 477
 478 void
 479 nvfx_init_surface_functions(struct nvfx_context *nvfx)
 480 {
 481         nvfx->pipe.resource_copy_region = nvfx_resource_copy_region;
 482         nvfx->pipe.clear_render_target = nvfx_clear_render_target;
 483         nvfx->pipe.clear_depth_stencil = nvfx_clear_depth_stencil;
 484 }