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    /* According to the Ivy Bridge PRM, Vol1 Part4, section 1.2.1.2 (Graphics
  52     * Data Size Limitations):
  53     *
  54     *    The BLT engine is capable of transferring very large quantities of
  55     *    graphics data. Any graphics data read from and written to the
  56     *    destination is permitted to represent a number of pixels that
  57     *    occupies up to 65,536 scan lines and up to 32,768 bytes per scan line
  58     *    at the destination. The maximum number of pixels that may be
  59     *    represented per scan line’s worth of graphics data depends on the
  60     *    color depth.
  61     *
  62     * Furthermore, intelEmitCopyBlit (which is called below) uses a signed
  63     * 16-bit integer to represent buffer pitch, so it can only handle buffer
  64     * pitches < 32k.
  65     *
  66     * As a result of these two limitations, we can only use the blitter to do
  67     * this copy when the miptree's pitch is less than 32k.
  68     */
  69    if (src_mt->pitch >= 32768 ||
  70        dst_mt->pitch >= 32768) {
  71       perf_debug("Falling back due to >=32k pitch\n");
  72       return false;
  73    }
  74
  75    intel_miptree_get_image_offset(src_mt, src_level, src_z,
  76                                   &src_image_x, &src_image_y);
  77
  78    if (_mesa_is_format_compressed(src_mt->format)) {
  79       _mesa_get_format_block_size(src_mt->format, &bw, &bh);
  80
  81       assert(src_x % bw == 0);
  82       assert(src_y % bh == 0);
  83       assert(src_width % bw == 0);
  84       assert(src_height % bh == 0);
  85
  86       src_x /= (int)bw;
  87       src_y /= (int)bh;
  88       src_width /= (int)bw;
  89       src_height /= (int)bh;
  90    }
  91    src_x += src_image_x;
  92    src_y += src_image_y;
  93
  94    intel_miptree_get_image_offset(dst_mt, dst_level, dst_z,
  95                                   &dst_image_x, &dst_image_y);
  96
  97    if (_mesa_is_format_compressed(dst_mt->format)) {
  98       _mesa_get_format_block_size(dst_mt->format, &bw, &bh);
  99
 100       assert(dst_x % bw == 0);
 101       assert(dst_y % bh == 0);
 102
 103       dst_x /= (int)bw;
 104       dst_y /= (int)bh;
 105    }
 106    dst_x += dst_image_x;
 107    dst_y += dst_image_y;
 108
 109    return intelEmitCopyBlit(brw,
 110                             src_mt->cpp,
 111                             src_mt->pitch,
 112                             src_mt->bo, src_mt->offset,
 113                             src_mt->tiling,
 114                             src_mt->tr_mode,
 115                             dst_mt->pitch,
 116                             dst_mt->bo, dst_mt->offset,
 117                             dst_mt->tiling,
 118                             dst_mt->tr_mode,
 119                             src_x, src_y,
 120                             dst_x, dst_y,
 121                             src_width, src_height,
 122                             GL_COPY);
 123 }
 124
 125 static void
 126 copy_image_with_memcpy(struct brw_context *brw,
 127                        struct intel_mipmap_tree *src_mt, int src_level,
 128                        int src_x, int src_y, int src_z,
 129                        struct intel_mipmap_tree *dst_mt, int dst_level,
 130                        int dst_x, int dst_y, int dst_z,
 131                        int src_width, int src_height)
 132 {
 133    bool same_slice;
 134    void *mapped, *src_mapped, *dst_mapped;
 135    ptrdiff_t src_stride, dst_stride, cpp;
 136    int map_x1, map_y1, map_x2, map_y2;
 137    GLuint src_bw, src_bh;
 138
 139    cpp = _mesa_get_format_bytes(src_mt->format);
 140    _mesa_get_format_block_size(src_mt->format, &src_bw, &src_bh);
 141
 142    assert(src_width % src_bw == 0);
 143    assert(src_height % src_bh == 0);
 144    assert(src_x % src_bw == 0);
 145    assert(src_y % src_bh == 0);
 146
 147    /* If we are on the same miptree, same level, and same slice, then
 148     * intel_miptree_map won't let us map it twice.  We have to do things a
 149     * bit differently.  In particular, we do a single map large enough for
 150     * both portions and in read-write mode.
 151     */
 152    same_slice = src_mt == dst_mt && src_level == dst_level && src_z == dst_z;
 153
 154    if (same_slice) {
 155       assert(dst_x % src_bw == 0);
 156       assert(dst_y % src_bh == 0);
 157
 158       map_x1 = MIN2(src_x, dst_x);
 159       map_y1 = MIN2(src_y, dst_y);
 160       map_x2 = MAX2(src_x, dst_x) + src_width;
 161       map_y2 = MAX2(src_y, dst_y) + src_height;
 162
 163       intel_miptree_map(brw, src_mt, src_level, src_z,
 164                         map_x1, map_y1, map_x2 - map_x1, map_y2 - map_y1,
 165                         GL_MAP_READ_BIT | GL_MAP_WRITE_BIT,
 166                         &mapped, &src_stride);
 167
 168       dst_stride = src_stride;
 169
 170       /* Set the offsets here so we don't have to think about while looping */
 171       src_mapped = mapped + ((src_y - map_y1) / src_bh) * src_stride +
 172                             ((src_x - map_x1) / src_bw) * cpp;
 173       dst_mapped = mapped + ((dst_y - map_y1) / src_bh) * dst_stride +
 174                             ((dst_x - map_x1) / src_bw) * cpp;
 175    } else {
 176       intel_miptree_map(brw, src_mt, src_level, src_z,
 177                         src_x, src_y, src_width, src_height,
 178                         GL_MAP_READ_BIT, &src_mapped, &src_stride);
 179       intel_miptree_map(brw, dst_mt, dst_level, dst_z,
 180                         dst_x, dst_y, src_width, src_height,
 181                         GL_MAP_WRITE_BIT, &dst_mapped, &dst_stride);
 182    }
 183
 184    src_width /= (int)src_bw;
 185    src_height /= (int)src_bh;
 186
 187    for (int i = 0; i < src_height; ++i) {
 188       memcpy(dst_mapped, src_mapped, src_width * cpp);
 189       src_mapped += src_stride;
 190       dst_mapped += dst_stride;
 191    }
 192
 193    if (same_slice) {
 194       intel_miptree_unmap(brw, src_mt, src_level, src_z);
 195    } else {
 196       intel_miptree_unmap(brw, dst_mt, dst_level, dst_z);
 197       intel_miptree_unmap(brw, src_mt, src_level, src_z);
 198    }
 199 }
 200
 201
 202 static void
 203 intel_copy_image_sub_data(struct gl_context *ctx,
 204                           struct gl_texture_image *src_image,
 205                           struct gl_renderbuffer *src_renderbuffer,
 206                           int src_x, int src_y, int src_z,
 207                           struct gl_texture_image *dst_image,
 208                           struct gl_renderbuffer *dst_renderbuffer,
 209                           int dst_x, int dst_y, int dst_z,
 210                           int src_width, int src_height)
 211 {
 212    struct brw_context *brw = brw_context(ctx);
 213    struct intel_mipmap_tree *src_mt, *dst_mt;
 214    unsigned src_level, dst_level;
 215    GLuint bw, bh;
 216
 217    if (_mesa_meta_CopyImageSubData_uncompressed(ctx,
 218                                                 src_image, src_renderbuffer,
 219                                                 src_x, src_y, src_z,
 220                                                 dst_image, dst_renderbuffer,
 221                                                 dst_x, dst_y, dst_z,
 222                                                 src_width, src_height)) {
 223       return;
 224    }
 225
 226    if (src_image) {
 227       src_mt = intel_texture_image(src_image)->mt;
 228       src_level = src_image->Level + src_image->TexObject->MinLevel;
 229
 230       /* Cube maps actually have different images per face */
 231       if (src_image->TexObject->Target == GL_TEXTURE_CUBE_MAP)
 232          src_z = src_image->Face;
 233
 234       src_z += src_image->TexObject->MinLayer;
 235    } else {
 236       assert(src_renderbuffer);
 237       src_mt = intel_renderbuffer(src_renderbuffer)->mt;
 238       src_image = src_renderbuffer->TexImage;
 239       src_level = 0;
 240    }
 241
 242    if (dst_image) {
 243       dst_mt = intel_texture_image(dst_image)->mt;
 244
 245       dst_level = dst_image->Level + dst_image->TexObject->MinLevel;
 246
 247       /* Cube maps actually have different images per face */
 248       if (dst_image->TexObject->Target == GL_TEXTURE_CUBE_MAP)
 249          dst_z = dst_image->Face;
 250
 251       dst_z += dst_image->TexObject->MinLayer;
 252    } else {
 253       assert(dst_renderbuffer);
 254       dst_mt = intel_renderbuffer(dst_renderbuffer)->mt;
 255       dst_image = dst_renderbuffer->TexImage;
 256       dst_level = 0;
 257    }
 258
 259    if (src_mt->num_samples > 0 || dst_mt->num_samples > 0) {
 260       _mesa_problem(ctx, "Failed to copy multisampled texture with BLORP\n");
 261       return;
 262    }
 263
 264    /* We are now going to try and copy the texture using the blitter.  If
 265     * that fails, we will fall back mapping the texture and using memcpy.
 266     * In either case, we need to do a full resolve.
 267     */
 268    intel_miptree_all_slices_resolve_hiz(brw, src_mt);
 269    intel_miptree_all_slices_resolve_depth(brw, src_mt);
 270    intel_miptree_resolve_color(brw, src_mt, 0);
 271
 272    intel_miptree_all_slices_resolve_hiz(brw, dst_mt);
 273    intel_miptree_all_slices_resolve_depth(brw, dst_mt);
 274    intel_miptree_resolve_color(brw, dst_mt, 0);
 275
 276    _mesa_get_format_block_size(src_mt->format, &bw, &bh);
 277
 278    /* It's legal to have a WxH that's smaller than a compressed block. This
 279     * happens for example when you are using a higher level LOD. For this case,
 280     * we still want to copy the entire block, or else the decompression will be
 281     * incorrect.
 282     */
 283    if (src_width < bw)
 284       src_width = ALIGN_NPOT(src_width, bw);
 285
 286    if (src_height < bh)
 287       src_height = ALIGN_NPOT(src_height, bh);
 288
 289    if (copy_image_with_blitter(brw, src_mt, src_level,
 290                                src_x, src_y, src_z,
 291                                dst_mt, dst_level,
 292                                dst_x, dst_y, dst_z,
 293                                src_width, src_height))
 294       return;
 295
 296    /* This is a worst-case scenario software fallback that maps the two
 297     * textures and does a memcpy between them.
 298     */
 299    copy_image_with_memcpy(brw, src_mt, src_level,
 300                           src_x, src_y, src_z,
 301                           dst_mt, dst_level,
 302                           dst_x, dst_y, dst_z,
 303                           src_width, src_height);
 304 }
 305
 306 void
 307 intelInitCopyImageFuncs(struct dd_function_table *functions)
 308 {
 309    functions->CopyImageSubData = intel_copy_image_sub_data;
 310 }