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

   1
   2 /**************************************************************************
   3  *
   4  * Copyright 2003 VMware, Inc.
   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 VMWARE 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 "main/bufferobj.h"
  30 #include "main/image.h"
  31 #include "main/macros.h"
  32 #include "main/mtypes.h"
  33 #include "main/pbo.h"
  34 #include "main/texobj.h"
  35 #include "main/texstore.h"
  36 #include "main/texcompress.h"
  37 #include "main/enums.h"
  38 #include "drivers/common/meta.h"
  39
  40 #include "brw_context.h"
  41 #include "intel_batchbuffer.h"
  42 #include "intel_tex.h"
  43 #include "intel_mipmap_tree.h"
  44 #include "intel_blit.h"
  45 #include "intel_tiled_memcpy.h"
  46
  47 #define FILE_DEBUG_FLAG DEBUG_TEXTURE
  48
  49 /**
  50  * \brief A fast path for glTexImage and glTexSubImage.
  51  *
  52  * \param for_glTexImage Was this called from glTexImage or glTexSubImage?
  53  *
  54  * This fast path is taken when the texture format is BGRA, RGBA,
  55  * A or L and when the texture memory is X- or Y-tiled.  It uploads
  56  * the texture data by mapping the texture memory without a GTT fence, thus
  57  * acquiring a tiled view of the memory, and then copying sucessive
  58  * spans within each tile.
  59  *
  60  * This is a performance win over the conventional texture upload path because
  61  * it avoids the performance penalty of writing through the write-combine
  62  * buffer. In the conventional texture upload path,
  63  * texstore.c:store_texsubimage(), the texture memory is mapped through a GTT
  64  * fence, thus acquiring a linear view of the memory, then each row in the
  65  * image is memcpy'd. In this fast path, we replace each row's copy with
  66  * a sequence of copies over each linear span in tile.
  67  *
  68  * One use case is Google Chrome's paint rectangles.  Chrome (as
  69  * of version 21) renders each page as a tiling of 256x256 GL_BGRA textures.
  70  * Each page's content is initially uploaded with glTexImage2D and damaged
  71  * regions are updated with glTexSubImage2D. On some workloads, the
  72  * performance gain of this fastpath on Sandybridge is over 5x.
  73  */
  74 bool
  75 intel_texsubimage_tiled_memcpy(struct gl_context * ctx,
  76                                GLuint dims,
  77                                struct gl_texture_image *texImage,
  78                                GLint xoffset, GLint yoffset, GLint zoffset,
  79                                GLsizei width, GLsizei height, GLsizei depth,
  80                                GLenum format, GLenum type,
  81                                const GLvoid *pixels,
  82                                const struct gl_pixelstore_attrib *packing,
  83                                bool for_glTexImage)
  84 {
  85    struct brw_context *brw = brw_context(ctx);
  86    struct intel_texture_image *image = intel_texture_image(texImage);
  87    int src_pitch;
  88
  89    /* The miptree's buffer. */
  90    drm_intel_bo *bo;
  91
  92    int error = 0;
  93
  94    uint32_t cpp;
  95    mem_copy_fn mem_copy = NULL;
  96
  97    /* This fastpath is restricted to specific texture types:
  98     * a 2D BGRA, RGBA, L8 or A8 texture. It could be generalized to support
  99     * more types.
 100     *
 101     * FINISHME: The restrictions below on packing alignment and packing row
 102     * length are likely unneeded now because we calculate the source stride
 103     * with _mesa_image_row_stride. However, before removing the restrictions
 104     * we need tests.
 105     */
 106    if (!brw->has_llc ||
 107        !(type == GL_UNSIGNED_BYTE || type == GL_UNSIGNED_INT_8_8_8_8_REV) ||
 108        !(texImage->TexObject->Target == GL_TEXTURE_2D ||
 109          texImage->TexObject->Target == GL_TEXTURE_RECTANGLE) ||
 110        pixels == NULL ||
 111        _mesa_is_bufferobj(packing->BufferObj) ||
 112        packing->Alignment > 4 ||
 113        packing->SkipPixels > 0 ||
 114        packing->SkipRows > 0 ||
 115        (packing->RowLength != 0 && packing->RowLength != width) ||
 116        packing->SwapBytes ||
 117        packing->LsbFirst ||
 118        packing->Invert)
 119       return false;
 120
 121    if (!intel_get_memcpy(texImage->TexFormat, format, type, &mem_copy, &cpp,
 122                          INTEL_UPLOAD))
 123       return false;
 124
 125    /* If this is a nontrivial texture view, let another path handle it instead. */
 126    if (texImage->TexObject->MinLayer)
 127       return false;
 128
 129    if (for_glTexImage)
 130       ctx->Driver.AllocTextureImageBuffer(ctx, texImage);
 131
 132    if (!image->mt ||
 133        (image->mt->tiling != I915_TILING_X &&
 134        image->mt->tiling != I915_TILING_Y)) {
 135       /* The algorithm is written only for X- or Y-tiled memory. */
 136       return false;
 137    }
 138
 139    /* Since we are going to write raw data to the miptree, we need to resolve
 140     * any pending fast color clears before we start.
 141     */
 142    intel_miptree_resolve_color(brw, image->mt);
 143
 144    bo = image->mt->bo;
 145
 146    if (drm_intel_bo_references(brw->batch.bo, bo)) {
 147       perf_debug("Flushing before mapping a referenced bo.\n");
 148       intel_batchbuffer_flush(brw);
 149    }
 150
 151    error = brw_bo_map(brw, bo, true /* write enable */, "miptree");
 152    if (error || bo->virtual == NULL) {
 153       DBG("%s: failed to map bo\n", __func__);
 154       return false;
 155    }
 156
 157    src_pitch = _mesa_image_row_stride(packing, width, format, type);
 158
 159    /* We postponed printing this message until having committed to executing
 160     * the function.
 161     */
 162    DBG("%s: level=%d offset=(%d,%d) (w,h)=(%d,%d) format=0x%x type=0x%x "
 163        "mesa_format=0x%x tiling=%d "
 164        "packing=(alignment=%d row_length=%d skip_pixels=%d skip_rows=%d) "
 165        "for_glTexImage=%d\n",
 166        __func__, texImage->Level, xoffset, yoffset, width, height,
 167        format, type, texImage->TexFormat, image->mt->tiling,
 168        packing->Alignment, packing->RowLength, packing->SkipPixels,
 169        packing->SkipRows, for_glTexImage);
 170
 171    int level = texImage->Level + texImage->TexObject->MinLevel;
 172
 173    /* Adjust x and y offset based on miplevel */
 174    xoffset += image->mt->level[level].level_x;
 175    yoffset += image->mt->level[level].level_y;
 176
 177    linear_to_tiled(
 178       xoffset * cpp, (xoffset + width) * cpp,
 179       yoffset, yoffset + height,
 180       bo->virtual,
 181       pixels - (ptrdiff_t) yoffset * src_pitch - (ptrdiff_t) xoffset * cpp,
 182       image->mt->pitch, src_pitch,
 183       brw->has_swizzling,
 184       image->mt->tiling,
 185       mem_copy
 186    );
 187
 188    drm_intel_bo_unmap(bo);
 189    return true;
 190 }
 191
 192 static void
 193 intelTexSubImage(struct gl_context * ctx,
 194                  GLuint dims,
 195                  struct gl_texture_image *texImage,
 196                  GLint xoffset, GLint yoffset, GLint zoffset,
 197                  GLsizei width, GLsizei height, GLsizei depth,
 198                  GLenum format, GLenum type,
 199                  const GLvoid * pixels,
 200                  const struct gl_pixelstore_attrib *packing)
 201 {
 202    struct intel_texture_image *intelImage = intel_texture_image(texImage);
 203    bool ok;
 204
 205    bool tex_busy = intelImage->mt && drm_intel_bo_busy(intelImage->mt->bo);
 206
 207    DBG("%s mesa_format %s target %s format %s type %s level %d %dx%dx%d\n",
 208        __func__, _mesa_get_format_name(texImage->TexFormat),
 209        _mesa_enum_to_string(texImage->TexObject->Target),
 210        _mesa_enum_to_string(format), _mesa_enum_to_string(type),
 211        texImage->Level, texImage->Width, texImage->Height, texImage->Depth);
 212
 213    ok = _mesa_meta_pbo_TexSubImage(ctx, dims, texImage,
 214                                    xoffset, yoffset, zoffset,
 215                                    width, height, depth, format, type,
 216                                    pixels, false, tex_busy, packing);
 217    if (ok)
 218       return;
 219
 220    ok = intel_texsubimage_tiled_memcpy(ctx, dims, texImage,
 221                                        xoffset, yoffset, zoffset,
 222                                        width, height, depth,
 223                                        format, type, pixels, packing,
 224                                        false /*for_glTexImage*/);
 225    if (ok)
 226      return;
 227
 228    _mesa_store_texsubimage(ctx, dims, texImage,
 229                            xoffset, yoffset, zoffset,
 230                            width, height, depth,
 231                            format, type, pixels, packing);
 232 }
 233
 234 void
 235 intelInitTextureSubImageFuncs(struct dd_function_table *functions)
 236 {
 237    functions->TexSubImage = intelTexSubImage;
 238 }