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_util.h"
  40 #include "nouveau/nouveau_screen.h"
  41 #include "nvfx_context.h"
  42 #include "nvfx_screen.h"
  43 #include "nvfx_resource.h"
  44 #include "nv04_2d.h"
  45
  46 #include <nouveau/nouveau_bo.h>
  47
  48 static INLINE void
  49 nvfx_region_set_format(struct nv04_region* rgn, enum pipe_format format)
  50 {
  51         unsigned bits = util_format_get_blocksizebits(format);
  52         switch(bits)
  53         {
  54         case 8:
  55                 rgn->bpps = 0;
  56                 break;
  57         case 16:
  58                 rgn->bpps = 1;
  59                 break;
  60         case 32:
  61                 rgn->bpps = 2;
  62                 break;
  63         default:
  64                 assert(util_is_pot(bits));
  65                 int shift = log2i(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 static INLINE void
  76 nvfx_region_fixup_swizzled(struct nv04_region* rgn, unsigned zslice, unsigned width, unsigned height, unsigned depth)
  77 {
  78         // TODO: move this code to surface creation?
  79         if((depth <= 1) && (height <= 1 || width <= 2))
  80                 rgn->pitch = width << rgn->bpps;
  81         else if(depth > 1 && height <= 2 && width <= 2)
  82         {
  83                 rgn->pitch = width << rgn->bpps;
  84                 rgn->offset += (zslice * width * height) << rgn->bpps;
  85         }
  86         else
  87         {
  88                 rgn->pitch = 0;
  89                 rgn->z = zslice;
  90                 rgn->w = width;
  91                 rgn->h = height;
  92                 rgn->d = depth;
  93         }
  94 }
  95
  96 static INLINE void
  97 nvfx_region_init_for_surface(struct nv04_region* rgn, struct nvfx_surface* surf, unsigned x, unsigned y, bool for_write)
  98 {
  99         rgn->x = x;
 100         rgn->y = y;
 101         rgn->z = 0;
 102         nvfx_region_set_format(rgn, surf->base.base.format);
 103
 104         if(surf->temp)
 105         {
 106                 rgn->bo = surf->temp->base.bo;
 107                 rgn->offset = 0;
 108                 rgn->pitch = surf->temp->linear_pitch;
 109
 110                 if(for_write)
 111                         util_dirty_surface_set_dirty(nvfx_surface_get_dirty_surfaces(&surf->base.base), &surf->base);
 112         } else {
 113                 rgn->bo = ((struct nvfx_resource*)surf->base.base.texture)->bo;
 114                 rgn->offset = surf->base.base.offset;
 115                 rgn->pitch = surf->pitch;
 116
 117                 if(!(surf->base.base.texture->flags & NVFX_RESOURCE_FLAG_LINEAR))
 118                         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));
 119         }
 120 }
 121
 122 static INLINE void
 123 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)
 124 {
 125         if(pt->target != PIPE_BUFFER)
 126         {
 127                 struct nvfx_surface* ns = (struct nvfx_surface*)util_surfaces_peek(&((struct nvfx_miptree*)pt)->surfaces, pt, sub.face, sub.level, z);
 128                 if(ns && util_dirty_surface_is_dirty(&ns->base))
 129                 {
 130                         nvfx_region_init_for_surface(rgn, ns, x, y, for_write);
 131                         return;
 132                 }
 133         }
 134
 135         rgn->bo = ((struct nvfx_resource*)pt)->bo;
 136         rgn->offset = nvfx_subresource_offset(pt, sub.face, sub.level, z);
 137         rgn->pitch = nvfx_subresource_pitch(pt, sub.level);
 138         rgn->x = x;
 139         rgn->y = y;
 140         rgn->z = 0;
 141
 142         nvfx_region_set_format(rgn, pt->format);
 143         if(!(pt->flags & NVFX_RESOURCE_FLAG_LINEAR))
 144                 nvfx_region_fixup_swizzled(rgn, z, u_minify(pt->width0, sub.level), u_minify(pt->height0, sub.level), u_minify(pt->depth0, sub.level));
 145 }
 146
 147 // TODO: actually test this for all formats, it's probably wrong for some...
 148
 149 static INLINE int
 150 nvfx_surface_format(enum pipe_format format)
 151 {
 152         switch(util_format_get_blocksize(format)) {
 153         case 1:
 154                 return NV04_CONTEXT_SURFACES_2D_FORMAT_Y8;
 155         case 2:
 156                 //return NV04_CONTEXT_SURFACES_2D_FORMAT_Y16;
 157                 return NV04_CONTEXT_SURFACES_2D_FORMAT_R5G6B5;
 158         case 4:
 159                 //if(format == PIPE_FORMAT_B8G8R8X8_UNORM || format == PIPE_FORMAT_B8G8R8A8_UNORM)
 160                         return NV04_CONTEXT_SURFACES_2D_FORMAT_A8R8G8B8;
 161                 //else
 162                 //      return NV04_CONTEXT_SURFACES_2D_FORMAT_Y32;
 163         default:
 164                 return -1;
 165         }
 166 }
 167
 168 static INLINE int
 169 nv04_scaled_image_format(enum pipe_format format)
 170 {
 171         switch(util_format_get_blocksize(format)) {
 172         case 1:
 173                 return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_Y8;
 174         case 2:
 175                 //if(format == PIPE_FORMAT_B5G5R5A1_UNORM)
 176                 //      return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_A1R5G5B5;
 177                 //else
 178                         return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_R5G6B5;
 179         case 4:
 180                 if(format == PIPE_FORMAT_B8G8R8X8_UNORM)
 181                         return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_X8R8G8B8;
 182                 else
 183                         return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_A8R8G8B8;
 184         default:
 185                 return -1;
 186         }
 187 }
 188
 189 // XXX: must save index buffer too!
 190 static struct blitter_context*
 191 nvfx_get_blitter(struct pipe_context* pipe, int copy)
 192 {
 193         struct nvfx_context* nvfx = nvfx_context(pipe);
 194
 195         struct blitter_context* blitter = nvfx->blitter;
 196         if(!blitter)
 197                 nvfx->blitter = blitter = util_blitter_create(pipe);
 198
 199         util_blitter_save_blend(blitter, nvfx->blend);
 200         util_blitter_save_depth_stencil_alpha(blitter, nvfx->zsa);
 201         util_blitter_save_stencil_ref(blitter, &nvfx->stencil_ref);
 202         util_blitter_save_rasterizer(blitter, nvfx->rasterizer);
 203         util_blitter_save_fragment_shader(blitter, nvfx->fragprog);
 204         util_blitter_save_vertex_shader(blitter, nvfx->vertprog);
 205         util_blitter_save_viewport(blitter, &nvfx->viewport);
 206         util_blitter_save_framebuffer(blitter, &nvfx->framebuffer);
 207         util_blitter_save_clip(blitter, &nvfx->clip);
 208         util_blitter_save_vertex_elements(blitter, nvfx->vtxelt);
 209         util_blitter_save_vertex_buffers(blitter, nvfx->vtxbuf_nr, nvfx->vtxbuf);
 210
 211         if(copy)
 212         {
 213                 util_blitter_save_fragment_sampler_states(blitter, nvfx->nr_samplers, (void**)nvfx->tex_sampler);
 214                 util_blitter_save_fragment_sampler_views(blitter, nvfx->nr_textures, nvfx->fragment_sampler_views);
 215         }
 216
 217         return blitter;
 218 }
 219
 220 static unsigned
 221 nvfx_region_clone(struct nv04_2d_context* ctx, struct nv04_region* rgn, unsigned w, unsigned h, boolean for_read)
 222 {
 223         unsigned begin = nv04_region_begin(rgn, w, h);
 224         unsigned end = nv04_region_end(rgn, w, h);
 225         unsigned size = end - begin;
 226         struct nouveau_bo* bo = 0;
 227         nouveau_bo_new(rgn->bo->device, NOUVEAU_BO_MAP | NOUVEAU_BO_GART, 256, size, &bo);
 228
 229         if(for_read || (size > ((w * h) << rgn->bpps)))
 230                 nv04_memcpy(ctx, bo, 0, rgn->bo, rgn->offset + begin, size);
 231
 232         rgn->bo = bo;
 233         rgn->offset = -begin;
 234         return begin;
 235 }
 236
 237 static void
 238 nvfx_resource_copy_region(struct pipe_context *pipe,
 239                   struct pipe_resource *dstr, struct pipe_subresource subdst,
 240                   unsigned dstx, unsigned dsty, unsigned dstz,
 241                   struct pipe_resource *srcr, struct pipe_subresource subsrc,
 242                   unsigned srcx, unsigned srcy, unsigned srcz,
 243                   unsigned w, unsigned h)
 244 {
 245         struct nv04_2d_context *ctx = nvfx_screen(pipe->screen)->eng2d;
 246         struct nv04_region dst, src;
 247
 248         if(!w || !h)
 249                 return;
 250
 251         static int copy_threshold = -1;
 252         if(copy_threshold < 0)
 253         {
 254                 copy_threshold = debug_get_num_option("NOUVEAU_COPY_THRESHOLD", 0);
 255                 if(copy_threshold < 0)
 256                         copy_threshold = 0;
 257         }
 258
 259         int dst_to_gpu = dstr->usage != PIPE_USAGE_DYNAMIC && dstr->usage != PIPE_USAGE_STAGING;
 260         int src_on_gpu = nvfx_resource_on_gpu(srcr);
 261
 262         nvfx_region_init_for_subresource(&dst, dstr, subdst, dstx, dsty, dstz, TRUE);
 263         nvfx_region_init_for_subresource(&src, srcr, subsrc, srcx, srcy, srcz, FALSE);
 264         w = util_format_get_stride(dstr->format, w) >> dst.bpps;
 265         h = util_format_get_nblocksy(dstr->format, h);
 266
 267         int ret;
 268         boolean 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         /* Always try to use the GPU right now, if possible
 318          * If the user wanted the surface data on the CPU, he would have cleared with memset (hopefully) */
 319
 320         // we don't care about interior pixel order since we set all them to the same value
 321         nvfx_region_init_for_surface(&dst, (struct nvfx_surface*)dsts, dx, dy, TRUE);
 322
 323         w = util_format_get_stride(dsts->format, w) >> dst.bpps;
 324         h = util_format_get_nblocksy(dsts->format, h);
 325
 326         int ret = nv04_region_fill_2d(ctx, &dst, w, h, value);
 327         if(ret > 0 && dsts->texture->bind & PIPE_BIND_RENDER_TARGET)
 328                 return 1;
 329         else if(ret)
 330         {
 331                 struct nv04_region dstt = dst;
 332                 unsigned dstbegin = 0;
 333
 334                 if(nvfx_resource_on_gpu(dsts->texture))
 335                         dstbegin = nvfx_region_clone(ctx, &dstt, w, h, FALSE);
 336
 337                 nv04_region_fill_cpu(&dstt, w, h, value);
 338
 339                 if(dstt.bo != dst.bo)
 340                 {
 341                         nv04_memcpy(ctx, dst.bo, dst.offset + dstbegin, dstt.bo, 0, dstt.bo->size);
 342                         nouveau_screen_bo_release(pipe->screen, dstt.bo);
 343                 }
 344         }
 345
 346         return 0;
 347 }
 348
 349
 350 void
 351 nvfx_screen_surface_takedown(struct pipe_screen *pscreen)
 352 {
 353         nv04_2d_context_takedown(nvfx_screen(pscreen)->eng2d);
 354         nvfx_screen(pscreen)->eng2d = 0;
 355 }
 356
 357 int
 358 nvfx_screen_surface_init(struct pipe_screen *pscreen)
 359 {
 360         struct nv04_2d_context* ctx = nv04_2d_context_init(nouveau_screen(pscreen)->channel);
 361         if(!ctx)
 362                 return -1;
 363         nvfx_screen(pscreen)->eng2d = ctx;
 364         return 0;
 365 }
 366
 367 static void
 368 nvfx_surface_copy_temp(struct pipe_context* pipe, struct pipe_surface* surf, int to_temp)
 369 {
 370         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 371         struct pipe_subresource tempsr, surfsr;
 372         struct pipe_resource *idxbuf_buffer;
 373         unsigned idxbuf_format;
 374
 375         tempsr.face = 0;
 376         tempsr.level = 0;
 377         surfsr.face = surf->face;
 378         surfsr.level = surf->level;
 379
 380         // TODO: do this properly, in blitter save
 381         idxbuf_buffer = ((struct nvfx_context*)pipe)->idxbuf_buffer;
 382         idxbuf_format = ((struct nvfx_context*)pipe)->idxbuf_format;
 383
 384         if(to_temp)
 385                 nvfx_resource_copy_region(pipe, &ns->temp->base.base, tempsr, 0, 0, 0, surf->texture, surfsr, 0, 0, surf->zslice, surf->width, surf->height);
 386         else
 387                 nvfx_resource_copy_region(pipe, surf->texture, surfsr, 0, 0, surf->zslice, &ns->temp->base.base, tempsr, 0, 0, 0, surf->width, surf->height);
 388
 389         ((struct nvfx_context*)pipe)->idxbuf_buffer = idxbuf_buffer;
 390         ((struct nvfx_context*)pipe)->idxbuf_format = idxbuf_format;
 391 }
 392
 393 void
 394 nvfx_surface_create_temp(struct pipe_context* pipe, struct pipe_surface* surf)
 395 {
 396         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 397         struct pipe_resource template;
 398         memset(&template, 0, sizeof(struct pipe_resource));
 399         template.target = PIPE_TEXTURE_2D;
 400         template.format = surf->format;
 401         template.width0 = surf->width;
 402         template.height0 = surf->height;
 403         template.depth0 = 1;
 404         template.nr_samples = surf->texture->nr_samples;
 405         template.flags = NVFX_RESOURCE_FLAG_LINEAR;
 406
 407         ns->temp = (struct nvfx_miptree*)nvfx_miptree_create(pipe->screen, &template);
 408         nvfx_surface_copy_temp(pipe, surf, 1);
 409 }
 410
 411 void
 412 nvfx_surface_flush(struct pipe_context* pipe, struct pipe_surface* surf)
 413 {
 414         struct nvfx_context* nvfx = (struct nvfx_context*)pipe;
 415         struct nvfx_surface* ns = (struct nvfx_surface*)surf;
 416         boolean bound = FALSE;
 417
 418         /* must be done before the copy, otherwise the copy will use the temp as destination */
 419         util_dirty_surface_set_clean(nvfx_surface_get_dirty_surfaces(surf), &ns->base);
 420
 421         nvfx_surface_copy_temp(pipe, surf, 0);
 422
 423         if(nvfx->framebuffer.zsbuf == surf)
 424                 bound = TRUE;
 425         else
 426         {
 427                 for(unsigned i = 0; i < nvfx->framebuffer.nr_cbufs; ++i)
 428                 {
 429                         if(nvfx->framebuffer.cbufs[i] == surf)
 430                         {
 431                                 bound = TRUE;
 432                                 break;
 433                         }
 434                 }
 435         }
 436
 437         if(!bound)
 438                 pipe_resource_reference((struct pipe_resource**)&ns->temp, 0);
 439 }
 440
 441 static void
 442 nvfx_clear_render_target(struct pipe_context *pipe,
 443                          struct pipe_surface *dst,
 444                          const float *rgba,
 445                          unsigned dstx, unsigned dsty,
 446                          unsigned width, unsigned height)
 447 {
 448         union util_color uc;
 449         util_pack_color(rgba, dst->format, &uc);
 450
 451         if(util_format_get_blocksizebits(dst->format) > 32
 452                 || nvfx_surface_fill(pipe, dst, dstx, dsty, width, height, uc.ui))
 453         {
 454                 // TODO: probably should use hardware clear here instead if possible
 455                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 0);
 456                 util_blitter_clear_render_target(blitter, dst, rgba, dstx, dsty, width, height);
 457         }
 458 }
 459
 460 static void
 461 nvfx_clear_depth_stencil(struct pipe_context *pipe,
 462                          struct pipe_surface *dst,
 463                          unsigned clear_flags,
 464                          double depth,
 465                          unsigned stencil,
 466                          unsigned dstx, unsigned dsty,
 467                          unsigned width, unsigned height)
 468 {
 469         if(util_format_get_blocksizebits(dst->format) > 32
 470                 || nvfx_surface_fill(pipe, dst, dstx, dsty, width, height, util_pack_z_stencil(dst->format, depth, stencil)))
 471         {
 472                 // TODO: probably should use hardware clear here instead if possible
 473                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 0);
 474                 util_blitter_clear_depth_stencil(blitter, dst, clear_flags, depth, stencil, dstx, dsty, width, height);
 475         }
 476 }
 477
 478
 479 void
 480 nvfx_init_surface_functions(struct nvfx_context *nvfx)
 481 {
 482         nvfx->pipe.resource_copy_region = nvfx_resource_copy_region;
 483         nvfx->pipe.clear_render_target = nvfx_clear_render_target;
 484         nvfx->pipe.clear_depth_stencil = nvfx_clear_depth_stencil;
 485 }