src/mesa/drivers/dri/i965/intel_copy_image.c

   1 /*
   2  * Mesa 3-D graphics library
   3  *
   4  * Copyright (C) 2014 Intel Corporation All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included
  14  * in all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  17  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
  20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  22  * OTHER DEALINGS IN THE SOFTWARE.
  23  *
  24  * Authors:
  25  *    Jason Ekstrand <jason.ekstrand@intel.com>
  26  */
  27
  28 #include "intel_fbo.h"
  29 #include "intel_tex.h"
  30 #include "intel_blit.h"
  31 #include "intel_mipmap_tree.h"
  32 #include "main/formats.h"
  33 #include "main/teximage.h"
  34 #include "drivers/common/meta.h"
  35
  36 static bool
  37 copy_image_with_blitter(struct brw_context *brw,
  38                         struct intel_mipmap_tree *src_mt, int src_level,
  39                         int src_x, int src_y, int src_z,
  40                         struct intel_mipmap_tree *dst_mt, int dst_level,
  41                         int dst_x, int dst_y, int dst_z,
  42                         int src_width, int src_height)
  43 {
  44    GLuint bw, bh;
  45    uint32_t src_image_x, src_image_y, dst_image_x, dst_image_y;
  46
  47    /* The blitter doesn't understand multisampling at all. */
  48    if (src_mt->num_samples > 0 || dst_mt->num_samples > 0)
  49       return false;
  50
  51    if (src_mt->format == MESA_FORMAT_S_UINT8)
  52       return false;
  53
  54    /* According to the Ivy Bridge PRM, Vol1 Part4, section 1.2.1.2 (Graphics
  55     * Data Size Limitations):
  56     *
  57     *    The BLT engine is capable of transferring very large quantities of
  58     *    graphics data. Any graphics data read from and written to the
  59     *    destination is permitted to represent a number of pixels that
  60     *    occupies up to 65,536 scan lines and up to 32,768 bytes per scan line
  61     *    at the destination. The maximum number of pixels that may be
  62     *    represented per scan line’s worth of graphics data depends on the
  63     *    color depth.
  64     *
  65     * Furthermore, intelEmitCopyBlit (which is called below) uses a signed
  66     * 16-bit integer to represent buffer pitch, so it can only handle buffer
  67     * pitches < 32k.
  68     *
  69     * As a result of these two limitations, we can only use the blitter to do
  70     * this copy when the miptree's pitch is less than 32k.
  71     */
  72    if (src_mt->pitch >= 32768 ||
  73        dst_mt->pitch >= 32768) {
  74       perf_debug("Falling back due to >=32k pitch\n");
  75       return false;
  76    }
  77
  78    intel_miptree_get_image_offset(src_mt, src_level, src_z,
  79                                   &src_image_x, &src_image_y);
  80
  81    if (_mesa_is_format_compressed(src_mt->format)) {
  82       _mesa_get_format_block_size(src_mt->format, &bw, &bh);
  83
  84       assert(src_x % bw == 0);
  85       assert(src_y % bh == 0);
  86       assert(src_width % bw == 0);
  87       assert(src_height % bh == 0);
  88
  89       src_x /= (int)bw;
  90       src_y /= (int)bh;
  91       src_width /= (int)bw;
  92       src_height /= (int)bh;
  93    }
  94    src_x += src_image_x;
  95    src_y += src_image_y;
  96
  97    intel_miptree_get_image_offset(dst_mt, dst_level, dst_z,
  98                                   &dst_image_x, &dst_image_y);
  99
 100    if (_mesa_is_format_compressed(dst_mt->format)) {
 101       _mesa_get_format_block_size(dst_mt->format, &bw, &bh);
 102
 103       assert(dst_x % bw == 0);
 104       assert(dst_y % bh == 0);
 105
 106       dst_x /= (int)bw;
 107       dst_y /= (int)bh;
 108    }
 109    dst_x += dst_image_x;
 110    dst_y += dst_image_y;
 111
 112    return intelEmitCopyBlit(brw,
 113                             src_mt->cpp,
 114                             src_mt->pitch,
 115                             src_mt->bo, src_mt->offset,
 116                             src_mt->tiling,
 117                             src_mt->tr_mode,
 118                             dst_mt->pitch,
 119                             dst_mt->bo, dst_mt->offset,
 120                             dst_mt->tiling,
 121                             dst_mt->tr_mode,
 122                             src_x, src_y,
 123                             dst_x, dst_y,
 124                             src_width, src_height,
 125                             GL_COPY);
 126 }
 127
 128 static void
 129 copy_image_with_memcpy(struct brw_context *brw,
 130                        struct intel_mipmap_tree *src_mt, int src_level,
 131                        int src_x, int src_y, int src_z,
 132                        struct intel_mipmap_tree *dst_mt, int dst_level,
 133                        int dst_x, int dst_y, int dst_z,
 134                        int src_width, int src_height)
 135 {
 136    bool same_slice;
 137    void *mapped, *src_mapped, *dst_mapped;
 138    ptrdiff_t src_stride, dst_stride, cpp;
 139    int map_x1, map_y1, map_x2, map_y2;
 140    GLuint src_bw, src_bh;
 141
 142    cpp = _mesa_get_format_bytes(src_mt->format);
 143    _mesa_get_format_block_size(src_mt->format, &src_bw, &src_bh);
 144
 145    assert(src_width % src_bw == 0);
 146    assert(src_height % src_bh == 0);
 147    assert(src_x % src_bw == 0);
 148    assert(src_y % src_bh == 0);
 149
 150    /* If we are on the same miptree, same level, and same slice, then
 151     * intel_miptree_map won't let us map it twice.  We have to do things a
 152     * bit differently.  In particular, we do a single map large enough for
 153     * both portions and in read-write mode.
 154     */
 155    same_slice = src_mt == dst_mt && src_level == dst_level && src_z == dst_z;
 156
 157    if (same_slice) {
 158       assert(dst_x % src_bw == 0);
 159       assert(dst_y % src_bh == 0);
 160
 161       map_x1 = MIN2(src_x, dst_x);
 162       map_y1 = MIN2(src_y, dst_y);
 163       map_x2 = MAX2(src_x, dst_x) + src_width;
 164       map_y2 = MAX2(src_y, dst_y) + src_height;
 165
 166       intel_miptree_map(brw, src_mt, src_level, src_z,
 167                         map_x1, map_y1, map_x2 - map_x1, map_y2 - map_y1,
 168                         GL_MAP_READ_BIT | GL_MAP_WRITE_BIT,
 169                         &mapped, &src_stride);
 170
 171       dst_stride = src_stride;
 172
 173       /* Set the offsets here so we don't have to think about while looping */
 174       src_mapped = mapped + ((src_y - map_y1) / src_bh) * src_stride +
 175                             ((src_x - map_x1) / src_bw) * cpp;
 176       dst_mapped = mapped + ((dst_y - map_y1) / src_bh) * dst_stride +
 177                             ((dst_x - map_x1) / src_bw) * cpp;
 178    } else {
 179       intel_miptree_map(brw, src_mt, src_level, src_z,
 180                         src_x, src_y, src_width, src_height,
 181                         GL_MAP_READ_BIT, &src_mapped, &src_stride);
 182       intel_miptree_map(brw, dst_mt, dst_level, dst_z,
 183                         dst_x, dst_y, src_width, src_height,
 184                         GL_MAP_WRITE_BIT, &dst_mapped, &dst_stride);
 185    }
 186
 187    src_width /= (int)src_bw;
 188    src_height /= (int)src_bh;
 189
 190    for (int i = 0; i < src_height; ++i) {
 191       memcpy(dst_mapped, src_mapped, src_width * cpp);
 192       src_mapped += src_stride;
 193       dst_mapped += dst_stride;
 194    }
 195
 196    if (same_slice) {
 197       intel_miptree_unmap(brw, src_mt, src_level, src_z);
 198    } else {
 199       intel_miptree_unmap(brw, dst_mt, dst_level, dst_z);
 200       intel_miptree_unmap(brw, src_mt, src_level, src_z);
 201    }
 202 }
 203
 204 static void
 205 copy_miptrees(struct brw_context *brw,
 206               struct intel_mipmap_tree *src_mt,
 207               int src_x, int src_y, int src_z, unsigned src_level,
 208               struct intel_mipmap_tree *dst_mt,
 209               int dst_x, int dst_y, int dst_z, unsigned dst_level,
 210               int src_width, int src_height)
 211 {
 212    unsigned bw, bh;
 213
 214    /* We are now going to try and copy the texture using the blitter.  If
 215     * that fails, we will fall back mapping the texture and using memcpy.
 216     * In either case, we need to do a full resolve.
 217     */
 218    intel_miptree_all_slices_resolve_hiz(brw, src_mt);
 219    intel_miptree_all_slices_resolve_depth(brw, src_mt);
 220    intel_miptree_resolve_color(brw, src_mt, 0);
 221
 222    intel_miptree_all_slices_resolve_hiz(brw, dst_mt);
 223    intel_miptree_all_slices_resolve_depth(brw, dst_mt);
 224    intel_miptree_resolve_color(brw, dst_mt, 0);
 225
 226    _mesa_get_format_block_size(src_mt->format, &bw, &bh);
 227
 228    /* It's legal to have a WxH that's smaller than a compressed block. This
 229     * happens for example when you are using a higher level LOD. For this case,
 230     * we still want to copy the entire block, or else the decompression will be
 231     * incorrect.
 232     */
 233    if (src_width < bw)
 234       src_width = ALIGN_NPOT(src_width, bw);
 235
 236    if (src_height < bh)
 237       src_height = ALIGN_NPOT(src_height, bh);
 238
 239    if (copy_image_with_blitter(brw, src_mt, src_level,
 240                                src_x, src_y, src_z,
 241                                dst_mt, dst_level,
 242                                dst_x, dst_y, dst_z,
 243                                src_width, src_height))
 244       return;
 245
 246    /* This is a worst-case scenario software fallback that maps the two
 247     * textures and does a memcpy between them.
 248     */
 249    copy_image_with_memcpy(brw, src_mt, src_level,
 250                           src_x, src_y, src_z,
 251                           dst_mt, dst_level,
 252                           dst_x, dst_y, dst_z,
 253                           src_width, src_height);
 254 }
 255
 256 static void
 257 intel_copy_image_sub_data(struct gl_context *ctx,
 258                           struct gl_texture_image *src_image,
 259                           struct gl_renderbuffer *src_renderbuffer,
 260                           int src_x, int src_y, int src_z,
 261                           struct gl_texture_image *dst_image,
 262                           struct gl_renderbuffer *dst_renderbuffer,
 263                           int dst_x, int dst_y, int dst_z,
 264                           int src_width, int src_height)
 265 {
 266    struct brw_context *brw = brw_context(ctx);
 267    struct intel_mipmap_tree *src_mt, *dst_mt;
 268    unsigned src_level, dst_level;
 269
 270    if (_mesa_meta_CopyImageSubData_uncompressed(ctx,
 271                                                 src_image, src_renderbuffer,
 272                                                 src_x, src_y, src_z,
 273                                                 dst_image, dst_renderbuffer,
 274                                                 dst_x, dst_y, dst_z,
 275                                                 src_width, src_height)) {
 276       return;
 277    }
 278
 279    if (src_image) {
 280       src_mt = intel_texture_image(src_image)->mt;
 281       src_level = src_image->Level + src_image->TexObject->MinLevel;
 282
 283       /* Cube maps actually have different images per face */
 284       if (src_image->TexObject->Target == GL_TEXTURE_CUBE_MAP)
 285          src_z = src_image->Face;
 286
 287       src_z += src_image->TexObject->MinLayer;
 288    } else {
 289       assert(src_renderbuffer);
 290       src_mt = intel_renderbuffer(src_renderbuffer)->mt;
 291       src_image = src_renderbuffer->TexImage;
 292       src_level = 0;
 293    }
 294
 295    if (dst_image) {
 296       dst_mt = intel_texture_image(dst_image)->mt;
 297
 298       dst_level = dst_image->Level + dst_image->TexObject->MinLevel;
 299
 300       /* Cube maps actually have different images per face */
 301       if (dst_image->TexObject->Target == GL_TEXTURE_CUBE_MAP)
 302          dst_z = dst_image->Face;
 303
 304       dst_z += dst_image->TexObject->MinLayer;
 305    } else {
 306       assert(dst_renderbuffer);
 307       dst_mt = intel_renderbuffer(dst_renderbuffer)->mt;
 308       dst_image = dst_renderbuffer->TexImage;
 309       dst_level = 0;
 310    }
 311
 312    if (src_mt->num_samples > 0 || dst_mt->num_samples > 0) {
 313       _mesa_problem(ctx, "Failed to copy multisampled texture with BLORP\n");
 314       return;
 315    }
 316
 317    copy_miptrees(brw, src_mt, src_x, src_y, src_z, src_level,
 318                  dst_mt, dst_x, dst_y, dst_z, dst_level,
 319                  src_width, src_height);
 320
 321    /* CopyImage only works for equal formats, texture view equivalence
 322     * classes, and a couple special cases for compressed textures.
 323     *
 324     * Notably, GL_DEPTH_STENCIL does not appear in any equivalence
 325     * classes, so we know the formats must be the same, and thus both
 326     * will either have stencil, or not.  They can't be mismatched.
 327     */
 328    assert((src_mt->stencil_mt != NULL) == (dst_mt->stencil_mt != NULL));
 329
 330    if (dst_mt->stencil_mt) {
 331       copy_miptrees(brw, src_mt->stencil_mt, src_x, src_y, src_z, src_level,
 332                     dst_mt->stencil_mt, dst_x, dst_y, dst_z, dst_level,
 333                     src_width, src_height);
 334    }
 335 }
 336
 337 void
 338 intelInitCopyImageFuncs(struct dd_function_table *functions)
 339 {
 340    functions->CopyImageSubData = intel_copy_image_sub_data;
 341 }