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

   1 /*
   2  * Copyright 2003 VMware, Inc.
   3  * All Rights Reserved.
   4  *
   5  * Permission is hereby granted, free of charge, to any person obtaining a
   6  * copy of this software and associated documentation files (the
   7  * "Software"), to deal in the Software without restriction, including
   8  * without limitation the rights to use, copy, modify, merge, publish,
   9  * distribute, sublicense, and/or sell copies of the Software, and to
  10  * permit persons to whom the Software is furnished to do so, subject to
  11  * the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the
  14  * next paragraph) shall be included in all copies or substantial portions
  15  * of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  18  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  19  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  20  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  21  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  22  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  23  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  24  */
  25
  26 #include "main/enums.h"
  27 #include "main/mtypes.h"
  28 #include "main/macros.h"
  29 #include "main/fbobject.h"
  30 #include "main/image.h"
  31 #include "main/bufferobj.h"
  32 #include "main/readpix.h"
  33 #include "main/state.h"
  34 #include "main/glformats.h"
  35 #include "drivers/common/meta.h"
  36
  37 #include "brw_context.h"
  38 #include "intel_screen.h"
  39 #include "intel_batchbuffer.h"
  40 #include "intel_blit.h"
  41 #include "intel_buffers.h"
  42 #include "intel_fbo.h"
  43 #include "intel_mipmap_tree.h"
  44 #include "intel_pixel.h"
  45 #include "intel_buffer_objects.h"
  46 #include "intel_tiled_memcpy.h"
  47
  48 #define FILE_DEBUG_FLAG DEBUG_PIXEL
  49
  50 /**
  51  * \brief A fast path for glReadPixels
  52  *
  53  * This fast path is taken when the source format is BGRA, RGBA,
  54  * A or L and when the texture memory is X- or Y-tiled.  It downloads
  55  * the source data by directly mapping the memory without a GTT fence.
  56  * This then needs to be de-tiled on the CPU before presenting the data to
  57  * the user in the linear fasion.
  58  *
  59  * This is a performance win over the conventional texture download path.
  60  * In the conventional texture download path, the texture is either mapped
  61  * through the GTT or copied to a linear buffer with the blitter before
  62  * handing off to a software path.  This allows us to avoid round-tripping
  63  * through the GPU (in the case where we would be blitting) and do only a
  64  * single copy operation.
  65  */
  66 static bool
  67 intel_readpixels_tiled_memcpy(struct gl_context * ctx,
  68                               GLint xoffset, GLint yoffset,
  69                               GLsizei width, GLsizei height,
  70                               GLenum format, GLenum type,
  71                               GLvoid * pixels,
  72                               const struct gl_pixelstore_attrib *pack)
  73 {
  74    struct brw_context *brw = brw_context(ctx);
  75    struct gl_renderbuffer *rb = ctx->ReadBuffer->_ColorReadBuffer;
  76
  77    /* This path supports reading from color buffers only */
  78    if (rb == NULL)
  79       return false;
  80
  81    struct intel_renderbuffer *irb = intel_renderbuffer(rb);
  82    int dst_pitch;
  83
  84    /* The miptree's buffer. */
  85    drm_intel_bo *bo;
  86
  87    int error = 0;
  88
  89    uint32_t cpp;
  90    mem_copy_fn mem_copy = NULL;
  91
  92    /* This fastpath is restricted to specific renderbuffer types:
  93     * a 2D BGRA, RGBA, L8 or A8 texture. It could be generalized to support
  94     * more types.
  95     */
  96    if (!brw->has_llc ||
  97        !(type == GL_UNSIGNED_BYTE || type == GL_UNSIGNED_INT_8_8_8_8_REV) ||
  98        pixels == NULL ||
  99        _mesa_is_bufferobj(pack->BufferObj) ||
 100        pack->Alignment > 4 ||
 101        pack->SkipPixels > 0 ||
 102        pack->SkipRows > 0 ||
 103        (pack->RowLength != 0 && pack->RowLength != width) ||
 104        pack->SwapBytes ||
 105        pack->LsbFirst ||
 106        pack->Invert)
 107       return false;
 108
 109    /* Only a simple blit, no scale, bias or other mapping. */
 110    if (ctx->_ImageTransferState)
 111       return false;
 112
 113    /* This renderbuffer can come from a texture.  In this case, we impose
 114     * some of the same restrictions we have for textures and adjust for
 115     * miplevels.
 116     */
 117    if (rb->TexImage) {
 118       if (rb->TexImage->TexObject->Target != GL_TEXTURE_2D &&
 119           rb->TexImage->TexObject->Target != GL_TEXTURE_RECTANGLE)
 120          return false;
 121
 122       int level = rb->TexImage->Level + rb->TexImage->TexObject->MinLevel;
 123
 124       /* Adjust x and y offset based on miplevel */
 125       xoffset += irb->mt->level[level].level_x;
 126       yoffset += irb->mt->level[level].level_y;
 127    }
 128
 129    /* It is possible that the renderbuffer (or underlying texture) is
 130     * multisampled.  Since ReadPixels from a multisampled buffer requires a
 131     * multisample resolve, we can't handle this here
 132     */
 133    if (rb->NumSamples > 1)
 134       return false;
 135
 136    /* We can't handle copying from RGBX or BGRX because the tiled_memcpy
 137     * function doesn't set the last channel to 1. Note this checks BaseFormat
 138     * rather than TexFormat in case the RGBX format is being simulated with an
 139     * RGBA format.
 140     */
 141    if (rb->_BaseFormat == GL_RGB)
 142       return false;
 143
 144    if (!intel_get_memcpy(rb->Format, format, type, &mem_copy, &cpp,
 145                          INTEL_DOWNLOAD))
 146       return false;
 147
 148    if (!irb->mt ||
 149        (irb->mt->tiling != I915_TILING_X &&
 150        irb->mt->tiling != I915_TILING_Y)) {
 151       /* The algorithm is written only for X- or Y-tiled memory. */
 152       return false;
 153    }
 154
 155    /* Since we are going to read raw data to the miptree, we need to resolve
 156     * any pending fast color clears before we start.
 157     */
 158    intel_miptree_resolve_color(brw, irb->mt, 0);
 159
 160    bo = irb->mt->bo;
 161
 162    if (drm_intel_bo_references(brw->batch.bo, bo)) {
 163       perf_debug("Flushing before mapping a referenced bo.\n");
 164       intel_batchbuffer_flush(brw);
 165    }
 166
 167    error = brw_bo_map(brw, bo, false /* write enable */, "miptree");
 168    if (error) {
 169       DBG("%s: failed to map bo\n", __func__);
 170       return false;
 171    }
 172
 173    dst_pitch = _mesa_image_row_stride(pack, width, format, type);
 174
 175    /* For a window-system renderbuffer, the buffer is actually flipped
 176     * vertically, so we need to handle that.  Since the detiling function
 177     * can only really work in the forwards direction, we have to be a
 178     * little creative.  First, we compute the Y-offset of the first row of
 179     * the renderbuffer (in renderbuffer coordinates).  We then match that
 180     * with the last row of the client's data.  Finally, we give
 181     * tiled_to_linear a negative pitch so that it walks through the
 182     * client's data backwards as it walks through the renderbufer forwards.
 183     */
 184    if (rb->Name == 0) {
 185       yoffset = rb->Height - yoffset - height;
 186       pixels += (ptrdiff_t) (height - 1) * dst_pitch;
 187       dst_pitch = -dst_pitch;
 188    }
 189
 190    /* We postponed printing this message until having committed to executing
 191     * the function.
 192     */
 193    DBG("%s: x,y=(%d,%d) (w,h)=(%d,%d) format=0x%x type=0x%x "
 194        "mesa_format=0x%x tiling=%d "
 195        "pack=(alignment=%d row_length=%d skip_pixels=%d skip_rows=%d)\n",
 196        __func__, xoffset, yoffset, width, height,
 197        format, type, rb->Format, irb->mt->tiling,
 198        pack->Alignment, pack->RowLength, pack->SkipPixels,
 199        pack->SkipRows);
 200
 201    tiled_to_linear(
 202       xoffset * cpp, (xoffset + width) * cpp,
 203       yoffset, yoffset + height,
 204       pixels - (ptrdiff_t) yoffset * dst_pitch - (ptrdiff_t) xoffset * cpp,
 205       bo->virtual,
 206       dst_pitch, irb->mt->pitch,
 207       brw->has_swizzling,
 208       irb->mt->tiling,
 209       mem_copy
 210    );
 211
 212    drm_intel_bo_unmap(bo);
 213    return true;
 214 }
 215
 216 void
 217 intelReadPixels(struct gl_context * ctx,
 218                 GLint x, GLint y, GLsizei width, GLsizei height,
 219                 GLenum format, GLenum type,
 220                 const struct gl_pixelstore_attrib *pack, GLvoid * pixels)
 221 {
 222    bool ok;
 223
 224    struct brw_context *brw = brw_context(ctx);
 225    bool dirty;
 226
 227    DBG("%s\n", __func__);
 228
 229    if (_mesa_is_bufferobj(pack->BufferObj)) {
 230       if (_mesa_meta_pbo_GetTexSubImage(ctx, 2, NULL, x, y, 0, width, height, 1,
 231                                         format, type, pixels, pack)) {
 232          /* _mesa_meta_pbo_GetTexSubImage() implements PBO transfers by
 233           * binding the user-provided BO as a fake framebuffer and rendering
 234           * to it.  This breaks the invariant of the GL that nothing is able
 235           * to render to a BO, causing nondeterministic corruption issues
 236           * because the render cache is not coherent with a number of other
 237           * caches that the BO could potentially be bound to afterwards.
 238           *
 239           * This could be solved in the same way that we guarantee texture
 240           * coherency after a texture is attached to a framebuffer and
 241           * rendered to, but that would involve checking *all* BOs bound to
 242           * the pipeline for the case we need to emit a cache flush due to
 243           * previous rendering to any of them -- Including vertex, index,
 244           * uniform, atomic counter, shader image, transform feedback,
 245           * indirect draw buffers, etc.
 246           *
 247           * That would increase the per-draw call overhead even though it's
 248           * very unlikely that any of the BOs bound to the pipeline has been
 249           * rendered to via a PBO at any point, so it seems better to just
 250           * flush here unconditionally.
 251           */
 252          brw_emit_mi_flush(brw);
 253          return;
 254       }
 255
 256       perf_debug("%s: fallback to CPU mapping in PBO case\n", __func__);
 257    }
 258
 259    ok = intel_readpixels_tiled_memcpy(ctx, x, y, width, height,
 260                                       format, type, pixels, pack);
 261    if(ok)
 262       return;
 263
 264    /* glReadPixels() wont dirty the front buffer, so reset the dirty
 265     * flag after calling intel_prepare_render(). */
 266    dirty = brw->front_buffer_dirty;
 267    intel_prepare_render(brw);
 268    brw->front_buffer_dirty = dirty;
 269
 270    /* Update Mesa state before calling _mesa_readpixels().
 271     * XXX this may not be needed since ReadPixels no longer uses the
 272     * span code.
 273     */
 274
 275    if (ctx->NewState)
 276       _mesa_update_state(ctx);
 277
 278    _mesa_readpixels(ctx, x, y, width, height, format, type, pack, pixels);
 279
 280    /* There's an intel_prepare_render() call in intelSpanRenderStart(). */
 281    brw->front_buffer_dirty = dirty;
 282 }