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