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)
  98 {
  99         rgn->bo = ((struct nvfx_resource*)surf->base.texture)->bo;
 100         rgn->offset = surf->base.offset;
 101         rgn->pitch = surf->pitch;
 102         rgn->x = x;
 103         rgn->y = y;
 104         rgn->z = 0;
 105
 106         nvfx_region_set_format(rgn, surf->base.format);
 107         if(!(surf->base.texture->flags & NVFX_RESOURCE_FLAG_LINEAR))
 108                 nvfx_region_fixup_swizzled(rgn, surf->base.zslice, surf->base.width, surf->base.height, u_minify(surf->base.texture->depth0, surf->base.level));
 109 }
 110
 111 static INLINE void
 112 nvfx_region_init_for_subresource(struct nv04_region* rgn, struct pipe_resource* pt, struct pipe_subresource sub, unsigned x, unsigned y, unsigned z)
 113 {
 114         rgn->bo = ((struct nvfx_resource*)pt)->bo;
 115         rgn->offset = nvfx_subresource_offset(pt, sub.face, sub.level, z);
 116         rgn->pitch = nvfx_subresource_pitch(pt, sub.level);
 117         rgn->x = x;
 118         rgn->y = y;
 119         rgn->z = 0;
 120
 121         nvfx_region_set_format(rgn, pt->format);
 122         if(!(pt->flags & NVFX_RESOURCE_FLAG_LINEAR))
 123                 nvfx_region_fixup_swizzled(rgn, z, u_minify(pt->width0, sub.level), u_minify(pt->height0, sub.level), u_minify(pt->depth0, sub.level));
 124 }
 125
 126 // TODO: actually test this for all formats, it's probably wrong for some...
 127
 128 static INLINE int
 129 nvfx_surface_format(enum pipe_format format)
 130 {
 131         switch(util_format_get_blocksize(format)) {
 132         case 1:
 133                 return NV04_CONTEXT_SURFACES_2D_FORMAT_Y8;
 134         case 2:
 135                 //return NV04_CONTEXT_SURFACES_2D_FORMAT_Y16;
 136                 return NV04_CONTEXT_SURFACES_2D_FORMAT_R5G6B5;
 137         case 4:
 138                 //if(format == PIPE_FORMAT_B8G8R8X8_UNORM || format == PIPE_FORMAT_B8G8R8A8_UNORM)
 139                         return NV04_CONTEXT_SURFACES_2D_FORMAT_A8R8G8B8;
 140                 //else
 141                 //      return NV04_CONTEXT_SURFACES_2D_FORMAT_Y32;
 142         default:
 143                 return -1;
 144         }
 145 }
 146
 147 static INLINE int
 148 nv04_scaled_image_format(enum pipe_format format)
 149 {
 150         switch(util_format_get_blocksize(format)) {
 151         case 1:
 152                 return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_Y8;
 153         case 2:
 154                 //if(format == PIPE_FORMAT_B5G5R5A1_UNORM)
 155                 //      return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_A1R5G5B5;
 156                 //else
 157                         return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_R5G6B5;
 158         case 4:
 159                 if(format == PIPE_FORMAT_B8G8R8X8_UNORM)
 160                         return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_X8R8G8B8;
 161                 else
 162                         return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_A8R8G8B8;
 163         default:
 164                 return -1;
 165         }
 166 }
 167
 168 static struct blitter_context*
 169 nvfx_get_blitter(struct pipe_context* pipe, int copy)
 170 {
 171         struct nvfx_context* nvfx = nvfx_context(pipe);
 172
 173         struct blitter_context* blitter = nvfx->blitter;
 174         if(!blitter)
 175                 nvfx->blitter = blitter = util_blitter_create(pipe);
 176
 177         util_blitter_save_blend(blitter, nvfx->blend);
 178         util_blitter_save_depth_stencil_alpha(blitter, nvfx->zsa);
 179         util_blitter_save_stencil_ref(blitter, &nvfx->stencil_ref);
 180         util_blitter_save_rasterizer(blitter, nvfx->rasterizer);
 181         util_blitter_save_fragment_shader(blitter, nvfx->fragprog);
 182         util_blitter_save_vertex_shader(blitter, nvfx->vertprog);
 183         util_blitter_save_viewport(blitter, &nvfx->viewport);
 184         util_blitter_save_framebuffer(blitter, &nvfx->framebuffer);
 185         util_blitter_save_clip(blitter, &nvfx->clip);
 186         util_blitter_save_vertex_elements(blitter, nvfx->vtxelt);
 187         util_blitter_save_vertex_buffers(blitter, nvfx->vtxbuf_nr, nvfx->vtxbuf);
 188
 189         if(copy)
 190         {
 191                 util_blitter_save_fragment_sampler_states(blitter, nvfx->nr_samplers, (void**)nvfx->tex_sampler);
 192                 util_blitter_save_fragment_sampler_views(blitter, nvfx->nr_textures, nvfx->fragment_sampler_views);
 193         }
 194
 195         return blitter;
 196 }
 197
 198 static unsigned
 199 nvfx_region_clone(struct nv04_2d_context* ctx, struct nv04_region* rgn, unsigned w, unsigned h, boolean for_read)
 200 {
 201         unsigned begin = nv04_region_begin(rgn, w, h);
 202         unsigned end = nv04_region_end(rgn, w, h);
 203         unsigned size = end - begin;
 204         struct nouveau_bo* bo = 0;
 205         nouveau_bo_new(rgn->bo->device, NOUVEAU_BO_MAP | NOUVEAU_BO_GART, 256, size, &bo);
 206
 207         if(for_read || (size > ((w * h) << rgn->bpps)))
 208                 nv04_memcpy(ctx, bo, 0, rgn->bo, rgn->offset + begin, size);
 209
 210         rgn->bo = bo;
 211         rgn->offset = -begin;
 212         return begin;
 213 }
 214
 215 static void
 216 nvfx_resource_copy_region(struct pipe_context *pipe,
 217                   struct pipe_resource *dstr, struct pipe_subresource subdst,
 218                   unsigned dstx, unsigned dsty, unsigned dstz,
 219                   struct pipe_resource *srcr, struct pipe_subresource subsrc,
 220                   unsigned srcx, unsigned srcy, unsigned srcz,
 221                   unsigned w, unsigned h)
 222 {
 223         struct nv04_2d_context *ctx = nvfx_screen(pipe->screen)->eng2d;
 224         struct nv04_region dst, src;
 225
 226         if(!w || !h)
 227                 return;
 228
 229         static int copy_threshold = -1;
 230         if(copy_threshold < 0)
 231         {
 232                 copy_threshold = debug_get_num_option("NOUVEAU_COPY_THRESHOLD", 0);
 233                 if(copy_threshold < 0)
 234                         copy_threshold = 0;
 235         }
 236
 237         int dst_to_gpu = dstr->usage != PIPE_USAGE_DYNAMIC && dstr->usage != PIPE_USAGE_STAGING;
 238         int src_on_gpu = nvfx_resource_on_gpu(srcr);
 239
 240         nvfx_region_init_for_subresource(&dst, dstr, subdst, dstx, dsty, dstz);
 241         nvfx_region_init_for_subresource(&src, srcr, subsrc, srcx, srcy, srcz);
 242         w = util_format_get_stride(dstr->format, w) >> dst.bpps;
 243         h = util_format_get_nblocksy(dstr->format, h);
 244
 245         int ret;
 246         boolean small = (w * h <= copy_threshold);
 247         if((!dst_to_gpu || !src_on_gpu) && small)
 248                 ret = -1; /* use the CPU */
 249         else
 250                 ret = nv04_region_copy_2d(ctx, &dst, &src, w, h,
 251                         dstr->target == PIPE_BUFFER ? -1 : nvfx_surface_format(dstr->format),
 252                         dstr->target == PIPE_BUFFER ? -1 : nv04_scaled_image_format(dstr->format),
 253                         dst_to_gpu, src_on_gpu);
 254         if(!ret)
 255         {}
 256         else if(ret > 0 && dstr->bind & PIPE_BIND_RENDER_TARGET && srcr->bind & PIPE_BIND_SAMPLER_VIEW)
 257         {
 258                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 1);
 259                 util_blitter_copy_region(blitter, dstr, subdst, dstx, dsty, dstz, srcr, subsrc, srcx, srcy, srcz, w, h, TRUE);
 260         }
 261         else
 262         {
 263                 struct nv04_region dstt = dst;
 264                 struct nv04_region srct = src;
 265                 unsigned dstbegin = 0;
 266
 267                 if(!small)
 268                 {
 269                         if(src_on_gpu)
 270                                 nvfx_region_clone(ctx, &srct, w, h, TRUE);
 271
 272                         if(dst_to_gpu)
 273                                 dstbegin = nvfx_region_clone(ctx, &dstt, w, h, FALSE);
 274                 }
 275
 276                 nv04_region_copy_cpu(&dstt, &srct, w, h);
 277
 278                 if(srct.bo != src.bo)
 279                         nouveau_screen_bo_release(pipe->screen, srct.bo);
 280
 281                 if(dstt.bo != dst.bo)
 282                 {
 283                         nv04_memcpy(ctx, dst.bo, dst.offset + dstbegin, dstt.bo, 0, dstt.bo->size);
 284                         nouveau_screen_bo_release(pipe->screen, dstt.bo);
 285                 }
 286         }
 287 }
 288
 289 static int
 290 nvfx_surface_fill(struct pipe_context* pipe, struct pipe_surface *dsts,
 291                   unsigned dx, unsigned dy, unsigned w, unsigned h, unsigned value)
 292 {
 293         struct nv04_2d_context *ctx = nvfx_screen(pipe->screen)->eng2d;
 294         struct nv04_region dst;
 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 */
 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);
 300         w = util_format_get_stride(dsts->format, w) >> dst.bpps;
 301         h = util_format_get_nblocksy(dsts->format, h);
 302
 303         int ret = nv04_region_fill_2d(ctx, &dst, w, h, value);
 304         if(ret > 0 && dsts->texture->bind & PIPE_BIND_RENDER_TARGET)
 305                 return 1;
 306         else if(ret)
 307         {
 308                 struct nv04_region dstt = dst;
 309                 unsigned dstbegin = 0;
 310
 311                 if(nvfx_resource_on_gpu(dsts->texture))
 312                         dstbegin = nvfx_region_clone(ctx, &dstt, w, h, FALSE);
 313
 314                 nv04_region_fill_cpu(&dstt, w, h, value);
 315
 316                 if(dstt.bo != dst.bo)
 317                 {
 318                         nv04_memcpy(ctx, dst.bo, dst.offset + dstbegin, dstt.bo, 0, dstt.bo->size);
 319                         nouveau_screen_bo_release(pipe->screen, dstt.bo);
 320                 }
 321         }
 322
 323         return 0;
 324 }
 325
 326
 327 void
 328 nvfx_screen_surface_takedown(struct pipe_screen *pscreen)
 329 {
 330         nv04_2d_context_takedown(nvfx_screen(pscreen)->eng2d);
 331         nvfx_screen(pscreen)->eng2d = 0;
 332 }
 333
 334 int
 335 nvfx_screen_surface_init(struct pipe_screen *pscreen)
 336 {
 337         struct nv04_2d_context* ctx = nv04_2d_context_init(nouveau_screen(pscreen)->channel);
 338         if(!ctx)
 339                 return -1;
 340         nvfx_screen(pscreen)->eng2d = ctx;
 341         return 0;
 342 }
 343
 344 static void
 345 nvfx_clear_render_target(struct pipe_context *pipe,
 346                          struct pipe_surface *dst,
 347                          const float *rgba,
 348                          unsigned dstx, unsigned dsty,
 349                          unsigned width, unsigned height)
 350 {
 351         union util_color uc;
 352         util_pack_color(rgba, dst->format, &uc);
 353
 354         if(util_format_get_blocksizebits(dst->format) > 32
 355                 || nvfx_surface_fill(pipe, dst, dstx, dsty, width, height, uc.ui))
 356         {
 357                 // TODO: probably should use hardware clear here instead if possible
 358                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 0);
 359                 util_blitter_clear_render_target(blitter, dst, rgba, dstx, dsty, width, height);
 360         }
 361 }
 362
 363 static void
 364 nvfx_clear_depth_stencil(struct pipe_context *pipe,
 365                          struct pipe_surface *dst,
 366                          unsigned clear_flags,
 367                          double depth,
 368                          unsigned stencil,
 369                          unsigned dstx, unsigned dsty,
 370                          unsigned width, unsigned height)
 371 {
 372         if(util_format_get_blocksizebits(dst->format) > 32
 373                 || nvfx_surface_fill(pipe, dst, dstx, dsty, width, height, util_pack_z_stencil(dst->format, depth, stencil)))
 374         {
 375                 // TODO: probably should use hardware clear here instead if possible
 376                 struct blitter_context* blitter = nvfx_get_blitter(pipe, 0);
 377                 util_blitter_clear_depth_stencil(blitter, dst, clear_flags, depth, stencil, dstx, dsty, width, height);
 378         }
 379 }
 380
 381
 382 void
 383 nvfx_init_surface_functions(struct nvfx_context *nvfx)
 384 {
 385         nvfx->pipe.resource_copy_region = nvfx_resource_copy_region;
 386         nvfx->pipe.clear_render_target = nvfx_clear_render_target;
 387         nvfx->pipe.clear_depth_stencil = nvfx_clear_depth_stencil;
 388 }