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_tex.h"
  29 #include "intel_blit.h"
  30 #include "intel_mipmap_tree.h"
  31 #include "main/formats.h"
  32 #include "drivers/common/meta.h"
  33
  34 static bool
  35 copy_image_with_blitter(struct brw_context *brw,
  36                         struct intel_mipmap_tree *src_mt, int src_level,
  37                         int src_x, int src_y, int src_z,
  38                         struct intel_mipmap_tree *dst_mt, int dst_level,
  39                         int dst_x, int dst_y, int dst_z,
  40                         int src_width, int src_height)
  41 {
  42    GLuint bw, bh;
  43    uint32_t src_image_x, src_image_y, dst_image_x, dst_image_y;
  44    int cpp;
  45
  46    /* The blitter doesn't understand multisampling at all. */
  47    if (src_mt->num_samples > 0 || dst_mt->num_samples > 0)
  48       return false;
  49
  50    /* According to the Ivy Bridge PRM, Vol1 Part4, section 1.2.1.2 (Graphics
  51     * Data Size Limitations):
  52     *
  53     *    The BLT engine is capable of transferring very large quantities of
  54     *    graphics data. Any graphics data read from and written to the
  55     *    destination is permitted to represent a number of pixels that
  56     *    occupies up to 65,536 scan lines and up to 32,768 bytes per scan line
  57     *    at the destination. The maximum number of pixels that may be
  58     *    represented per scan line’s worth of graphics data depends on the
  59     *    color depth.
  60     *
  61     * Furthermore, intelEmitCopyBlit (which is called below) uses a signed
  62     * 16-bit integer to represent buffer pitch, so it can only handle buffer
  63     * pitches < 32k.
  64     *
  65     * As a result of these two limitations, we can only use the blitter to do
  66     * this copy when the miptree's pitch is less than 32k.
  67     */
  68    if (src_mt->pitch >= 32768 ||
  69        dst_mt->pitch >= 32768) {
  70       perf_debug("Falling back due to >=32k pitch\n");
  71       return false;
  72    }
  73
  74    intel_miptree_get_image_offset(src_mt, src_level, src_z,
  75                                   &src_image_x, &src_image_y);
  76
  77    if (_mesa_is_format_compressed(src_mt->format)) {
  78       _mesa_get_format_block_size(src_mt->format, &bw, &bh);
  79
  80       assert(src_x % bw == 0);
  81       assert(src_y % bh == 0);
  82       assert(src_width % bw == 0);
  83       assert(src_height % bh == 0);
  84
  85       src_x /= (int)bw;
  86       src_y /= (int)bh;
  87       src_width /= (int)bw;
  88       src_height /= (int)bh;
  89
  90       /* Inside of the miptree, the x offsets are stored in pixels while
  91        * the y offsets are stored in blocks.  We need to scale just the x
  92        * offset.
  93        */
  94       src_image_x /= bw;
  95
  96       cpp = _mesa_get_format_bytes(src_mt->format);
  97    } else {
  98       cpp = src_mt->cpp;
  99    }
 100    src_x += src_image_x;
 101    src_y += src_image_y;
 102
 103    intel_miptree_get_image_offset(dst_mt, dst_level, dst_z,
 104                                   &dst_image_x, &dst_image_y);
 105
 106    if (_mesa_is_format_compressed(dst_mt->format)) {
 107       _mesa_get_format_block_size(dst_mt->format, &bw, &bh);
 108
 109       assert(dst_x % bw == 0);
 110       assert(dst_y % bh == 0);
 111
 112       dst_x /= (int)bw;
 113       dst_y /= (int)bh;
 114
 115       /* Inside of the miptree, the x offsets are stored in pixels while
 116        * the y offsets are stored in blocks.  We need to scale just the x
 117        * offset.
 118        */
 119       dst_image_x /= bw;
 120    }
 121    dst_x += dst_image_x;
 122    dst_y += dst_image_y;
 123
 124    return intelEmitCopyBlit(brw,
 125                             cpp,
 126                             src_mt->pitch,
 127                             src_mt->bo, src_mt->offset,
 128                             src_mt->tiling,
 129                             dst_mt->pitch,
 130                             dst_mt->bo, dst_mt->offset,
 131                             dst_mt->tiling,
 132                             src_x, src_y,
 133                             dst_x, dst_y,
 134                             src_width, src_height,
 135                             GL_COPY);
 136 }
 137
 138 static void
 139 copy_image_with_memcpy(struct brw_context *brw,
 140                        struct intel_mipmap_tree *src_mt, int src_level,
 141                        int src_x, int src_y, int src_z,
 142                        struct intel_mipmap_tree *dst_mt, int dst_level,
 143                        int dst_x, int dst_y, int dst_z,
 144                        int src_width, int src_height)
 145 {
 146    bool same_slice;
 147    uint8_t *mapped, *src_mapped, *dst_mapped;
 148    int src_stride, dst_stride, i, cpp;
 149    int map_x1, map_y1, map_x2, map_y2;
 150    GLuint src_bw, src_bh;
 151
 152    cpp = _mesa_get_format_bytes(src_mt->format);
 153    _mesa_get_format_block_size(src_mt->format, &src_bw, &src_bh);
 154
 155    assert(src_width % src_bw == 0);
 156    assert(src_height % src_bw == 0);
 157    assert(src_x % src_bw == 0);
 158    assert(src_y % src_bw == 0);
 159
 160    /* If we are on the same miptree, same level, and same slice, then
 161     * intel_miptree_map won't let us map it twice.  We have to do things a
 162     * bit differently.  In particular, we do a single map large enough for
 163     * both portions and in read-write mode.
 164     */
 165    same_slice = src_mt == dst_mt && src_level == dst_level && src_z == dst_z;
 166
 167    if (same_slice) {
 168       assert(dst_x % src_bw == 0);
 169       assert(dst_y % src_bw == 0);
 170
 171       map_x1 = MIN2(src_x, dst_x);
 172       map_y1 = MIN2(src_y, dst_y);
 173       map_x2 = MAX2(src_x, dst_x) + src_width;
 174       map_y2 = MAX2(src_y, dst_y) + src_height;
 175
 176       intel_miptree_map(brw, src_mt, src_level, src_z,
 177                         map_x1, map_y1, map_x2 - map_x1, map_y2 - map_y1,
 178                         GL_MAP_READ_BIT | GL_MAP_WRITE_BIT,
 179                         (void **)&mapped, &src_stride);
 180
 181       dst_stride = src_stride;
 182
 183       /* Set the offsets here so we don't have to think about while looping */
 184       src_mapped = mapped + ((src_y - map_y1) / src_bh) * src_stride +
 185                             ((src_x - map_x1) / src_bw) * cpp;
 186       dst_mapped = mapped + ((dst_y - map_y1) / src_bh) * dst_stride +
 187                             ((dst_x - map_x1) / src_bw) * cpp;
 188    } else {
 189       intel_miptree_map(brw, src_mt, src_level, src_z,
 190                         src_x, src_y, src_width, src_height,
 191                         GL_MAP_READ_BIT, (void **)&src_mapped, &src_stride);
 192       intel_miptree_map(brw, dst_mt, dst_level, dst_z,
 193                         dst_x, dst_y, src_width, src_height,
 194                         GL_MAP_WRITE_BIT, (void **)&dst_mapped, &dst_stride);
 195    }
 196
 197    src_width /= (int)src_bw;
 198    src_height /= (int)src_bh;
 199
 200    for (i = 0; i < src_height; ++i) {
 201       memcpy(dst_mapped, src_mapped, src_width * cpp);
 202       src_mapped += src_stride;
 203       dst_mapped += dst_stride;
 204    }
 205
 206    if (same_slice) {
 207       intel_miptree_unmap(brw, src_mt, src_level, src_z);
 208    } else {
 209       intel_miptree_unmap(brw, dst_mt, dst_level, dst_z);
 210       intel_miptree_unmap(brw, src_mt, src_level, src_z);
 211    }
 212 }
 213
 214 static void
 215 intel_copy_image_sub_data(struct gl_context *ctx,
 216                           struct gl_texture_image *src_image,
 217                           int src_x, int src_y, int src_z,
 218                           struct gl_texture_image *dst_image,
 219                           int dst_x, int dst_y, int dst_z,
 220                           int src_width, int src_height)
 221 {
 222    struct brw_context *brw = brw_context(ctx);
 223    struct intel_texture_image *intel_src_image = intel_texture_image(src_image);
 224    struct intel_texture_image *intel_dst_image = intel_texture_image(dst_image);
 225
 226    if (_mesa_meta_CopyImageSubData_uncompressed(ctx,
 227                                                 src_image, src_x, src_y, src_z,
 228                                                 dst_image, dst_x, dst_y, dst_z,
 229                                                 src_width, src_height)) {
 230       return;
 231    }
 232
 233    if (intel_src_image->mt->num_samples > 0 ||
 234        intel_dst_image->mt->num_samples > 0) {
 235       _mesa_problem(ctx, "Failed to copy multisampled texture with meta path\n");
 236       return;
 237    }
 238
 239    /* Cube maps actually have different images per face */
 240    if (src_image->TexObject->Target == GL_TEXTURE_CUBE_MAP)
 241       src_z = src_image->Face;
 242    if (dst_image->TexObject->Target == GL_TEXTURE_CUBE_MAP)
 243       dst_z = dst_image->Face;
 244
 245    /* We are now going to try and copy the texture using the blitter.  If
 246     * that fails, we will fall back mapping the texture and using memcpy.
 247     * In either case, we need to do a full resolve.
 248     */
 249    intel_miptree_all_slices_resolve_hiz(brw, intel_src_image->mt);
 250    intel_miptree_all_slices_resolve_depth(brw, intel_src_image->mt);
 251    intel_miptree_resolve_color(brw, intel_src_image->mt);
 252
 253    intel_miptree_all_slices_resolve_hiz(brw, intel_dst_image->mt);
 254    intel_miptree_all_slices_resolve_depth(brw, intel_dst_image->mt);
 255    intel_miptree_resolve_color(brw, intel_dst_image->mt);
 256
 257    unsigned src_level = src_image->Level + src_image->TexObject->MinLevel;
 258    unsigned dst_level = dst_image->Level + dst_image->TexObject->MinLevel;
 259    if (copy_image_with_blitter(brw, intel_src_image->mt, src_level,
 260                                src_x, src_y, src_z,
 261                                intel_dst_image->mt, dst_level,
 262                                dst_x, dst_y, dst_z,
 263                                src_width, src_height))
 264       return;
 265
 266    /* This is a worst-case scenario software fallback that maps the two
 267     * textures and does a memcpy between them.
 268     */
 269    copy_image_with_memcpy(brw, intel_src_image->mt, src_level,
 270                           src_x, src_y, src_z,
 271                           intel_dst_image->mt, dst_level,
 272                           dst_x, dst_y, dst_z,
 273                           src_width, src_height);
 274 }
 275
 276 void
 277 intelInitCopyImageFuncs(struct dd_function_table *functions)
 278 {
 279    functions->CopyImageSubData = intel_copy_image_sub_data;
 280 }