-
-/**************************************************************************
- *
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+/*
+ * Copyright 2003 VMware, Inc.
* All Rights Reserved.
- *
+ *
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the
* "Software"), to deal in the Software without restriction, including
* without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
+ * distribute, sublicense, and/or sell copies of the Software, and to
* permit persons to whom the Software is furnished to do so, subject to
* the following conditions:
- *
+ *
* The above copyright notice and this permission notice (including the
* next paragraph) shall be included in all copies or substantial portions
* of the Software.
- *
+ *
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
* ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- **************************************************************************/
+ */
#include "main/bufferobj.h"
+#include "main/image.h"
#include "main/macros.h"
#include "main/mtypes.h"
#include "main/pbo.h"
#include "main/texstore.h"
#include "main/texcompress.h"
#include "main/enums.h"
+#include "drivers/common/meta.h"
#include "brw_context.h"
#include "intel_batchbuffer.h"
#include "intel_tex.h"
#include "intel_mipmap_tree.h"
#include "intel_blit.h"
+#include "intel_tiled_memcpy.h"
#define FILE_DEBUG_FLAG DEBUG_TEXTURE
-static bool
-intel_blit_texsubimage(struct gl_context * ctx,
- struct gl_texture_image *texImage,
- GLint xoffset, GLint yoffset,
- GLint width, GLint height,
- GLenum format, GLenum type, const void *pixels,
- const struct gl_pixelstore_attrib *packing)
-{
- struct brw_context *brw = brw_context(ctx);
- struct intel_context *intel = intel_context(ctx);
- struct intel_texture_image *intelImage = intel_texture_image(texImage);
-
- /* Try to do a blit upload of the subimage if the texture is
- * currently busy.
- */
- if (!intelImage->mt)
- return false;
-
- /* The blitter can't handle Y tiling */
- if (intelImage->mt->region->tiling == I915_TILING_Y)
- return false;
-
- if (texImage->TexObject->Target != GL_TEXTURE_2D)
- return false;
-
- /* On gen6, it's probably not worth swapping to the blit ring to do
- * this because of all the overhead involved.
- */
- if (intel->gen >= 6)
- return false;
-
- if (!drm_intel_bo_busy(intelImage->mt->region->bo))
- return false;
-
- DBG("BLT subimage %s target %s level %d offset %d,%d %dx%d\n",
- __FUNCTION__,
- _mesa_lookup_enum_by_nr(texImage->TexObject->Target),
- texImage->Level, xoffset, yoffset, width, height);
-
- pixels = _mesa_validate_pbo_teximage(ctx, 2, width, height, 1,
- format, type, pixels, packing,
- "glTexSubImage");
- if (!pixels)
- return false;
-
- struct intel_mipmap_tree *temp_mt =
- intel_miptree_create(brw, GL_TEXTURE_2D, texImage->TexFormat,
- 0, 0,
- width, height, 1,
- false, 0, INTEL_MIPTREE_TILING_NONE);
- if (!temp_mt)
- goto err;
-
- GLubyte *dst = intel_miptree_map_raw(brw, temp_mt);
- if (!dst)
- goto err;
-
- if (!_mesa_texstore(ctx, 2, texImage->_BaseFormat,
- texImage->TexFormat,
- temp_mt->region->pitch,
- &dst,
- width, height, 1,
- format, type, pixels, packing)) {
- _mesa_error(ctx, GL_OUT_OF_MEMORY, "intelTexSubImage");
- }
-
- intel_miptree_unmap_raw(brw, temp_mt);
-
- bool ret;
-
- ret = intel_miptree_blit(brw,
- temp_mt, 0, 0,
- 0, 0, false,
- intelImage->mt, texImage->Level, texImage->Face,
- xoffset, yoffset, false,
- width, height, GL_COPY);
- assert(ret);
-
- intel_miptree_release(&temp_mt);
- _mesa_unmap_teximage_pbo(ctx, packing);
-
- return ret;
-
-err:
- _mesa_error(ctx, GL_OUT_OF_MEMORY, "intelTexSubImage");
- intel_miptree_release(&temp_mt);
- _mesa_unmap_teximage_pbo(ctx, packing);
- return false;
-}
-
/**
* \brief A fast path for glTexImage and glTexSubImage.
*
* \param for_glTexImage Was this called from glTexImage or glTexSubImage?
*
- * This fast path is taken when the hardware natively supports the texture
- * format (such as GL_BGRA) and when the texture memory is X-tiled. It uploads
+ * This fast path is taken when the texture format is BGRA, RGBA,
+ * A or L and when the texture memory is X- or Y-tiled. It uploads
* the texture data by mapping the texture memory without a GTT fence, thus
- * acquiring a tiled view of the memory, and then memcpy'ing sucessive
- * subspans within each tile.
+ * acquiring a tiled view of the memory, and then copying sucessive
+ * spans within each tile.
*
* This is a performance win over the conventional texture upload path because
* it avoids the performance penalty of writing through the write-combine
* buffer. In the conventional texture upload path,
* texstore.c:store_texsubimage(), the texture memory is mapped through a GTT
* fence, thus acquiring a linear view of the memory, then each row in the
- * image is memcpy'd. In this fast path, we replace each row's memcpy with
- * a sequence of memcpy's over each bit6 swizzle span in the row.
+ * image is memcpy'd. In this fast path, we replace each row's copy with
+ * a sequence of copies over each linear span in tile.
*
- * This fast path's use case is Google Chrome's paint rectangles. Chrome (as
+ * One use case is Google Chrome's paint rectangles. Chrome (as
* of version 21) renders each page as a tiling of 256x256 GL_BGRA textures.
* Each page's content is initially uploaded with glTexImage2D and damaged
* regions are updated with glTexSubImage2D. On some workloads, the
{
struct brw_context *brw = brw_context(ctx);
struct intel_texture_image *image = intel_texture_image(texImage);
+ int src_pitch;
/* The miptree's buffer. */
- drm_intel_bo *bo;
+ struct brw_bo *bo;
int error = 0;
- /* This fastpath is restricted to a specific texture type: level 0 of
- * a 2D BGRA texture. It could be generalized to support more types by
- * varying the arithmetic loop below.
+ uint32_t cpp;
+ mem_copy_fn mem_copy = NULL;
+
+ /* This fastpath is restricted to specific texture types:
+ * a 2D BGRA, RGBA, L8 or A8 texture. It could be generalized to support
+ * more types.
+ *
+ * FINISHME: The restrictions below on packing alignment and packing row
+ * length are likely unneeded now because we calculate the source stride
+ * with _mesa_image_row_stride. However, before removing the restrictions
+ * we need tests.
*/
if (!brw->has_llc ||
- format != GL_BGRA ||
- type != GL_UNSIGNED_BYTE ||
- texImage->TexFormat != MESA_FORMAT_ARGB8888 ||
- texImage->TexObject->Target != GL_TEXTURE_2D ||
- texImage->Level != 0 ||
+ !(type == GL_UNSIGNED_BYTE || type == GL_UNSIGNED_INT_8_8_8_8_REV) ||
+ !(texImage->TexObject->Target == GL_TEXTURE_2D ||
+ texImage->TexObject->Target == GL_TEXTURE_RECTANGLE) ||
pixels == NULL ||
_mesa_is_bufferobj(packing->BufferObj) ||
packing->Alignment > 4 ||
packing->Invert)
return false;
+ /* Only a simple blit, no scale, bias or other mapping. */
+ if (ctx->_ImageTransferState)
+ return false;
+
+ if (!intel_get_memcpy(texImage->TexFormat, format, type, &mem_copy, &cpp))
+ return false;
+
+ /* If this is a nontrivial texture view, let another path handle it instead. */
+ if (texImage->TexObject->MinLayer)
+ return false;
+
if (for_glTexImage)
ctx->Driver.AllocTextureImageBuffer(ctx, texImage);
if (!image->mt ||
- image->mt->region->tiling != I915_TILING_X) {
- /* The algorithm below is written only for X-tiled memory. */
+ (image->mt->tiling != I915_TILING_X &&
+ image->mt->tiling != I915_TILING_Y)) {
+ /* The algorithm is written only for X- or Y-tiled memory. */
return false;
}
/* Since we are going to write raw data to the miptree, we need to resolve
* any pending fast color clears before we start.
*/
- intel_miptree_resolve_color(brw, image->mt);
+ intel_miptree_all_slices_resolve_color(brw, image->mt, 0);
- bo = image->mt->region->bo;
+ bo = image->mt->bo;
- if (drm_intel_bo_references(brw->batch.bo, bo)) {
+ if (brw_batch_references(&brw->batch, bo)) {
perf_debug("Flushing before mapping a referenced bo.\n");
intel_batchbuffer_flush(brw);
}
- if (unlikely(brw->perf_debug)) {
- if (drm_intel_bo_busy(bo)) {
- perf_debug("Mapping a busy BO, causing a stall on the GPU.\n");
- }
- }
-
- error = drm_intel_bo_map(bo, true /*write_enable*/);
+ error = brw_bo_map(brw, bo, true /* write enable */);
if (error || bo->virtual == NULL) {
- DBG("%s: failed to map bo\n", __FUNCTION__);
+ DBG("%s: failed to map bo\n", __func__);
return false;
}
+ src_pitch = _mesa_image_row_stride(packing, width, format, type);
+
/* We postponed printing this message until having committed to executing
* the function.
*/
- DBG("%s: level=%d offset=(%d,%d) (w,h)=(%d,%d)\n",
- __FUNCTION__, texImage->Level, xoffset, yoffset, width, height);
-
- /* In the tiling algorithm below, some variables are in units of pixels,
- * others are in units of bytes, and others (such as height) are unitless.
- * Each variable name is suffixed with its units.
- */
-
- const uint32_t x_max_pixels = xoffset + width;
- const uint32_t y_max_pixels = yoffset + height;
-
- const uint32_t tile_size_bytes = 4096;
-
- const uint32_t tile_width_bytes = 512;
- const uint32_t tile_width_pixels = 128;
-
- const uint32_t tile_height = 8;
-
- const uint32_t cpp = 4; /* chars per pixel of GL_BGRA */
- const uint32_t swizzle_width_pixels = 16;
-
- const uint32_t stride_bytes = image->mt->region->pitch;
- const uint32_t width_tiles = stride_bytes / tile_width_bytes;
-
- for (uint32_t y_pixels = yoffset; y_pixels < y_max_pixels; ++y_pixels) {
- const uint32_t y_offset_bytes = (y_pixels / tile_height) * width_tiles * tile_size_bytes
- + (y_pixels % tile_height) * tile_width_bytes;
-
- for (uint32_t x_pixels = xoffset; x_pixels < x_max_pixels; x_pixels += swizzle_width_pixels) {
- const uint32_t x_offset_bytes = (x_pixels / tile_width_pixels) * tile_size_bytes
- + (x_pixels % tile_width_pixels) * cpp;
-
- intptr_t offset_bytes = y_offset_bytes + x_offset_bytes;
- if (brw->has_swizzling) {
-#if 0
- /* Clear, unoptimized version. */
- bool bit6 = (offset_bytes >> 6) & 1;
- bool bit9 = (offset_bytes >> 9) & 1;
- bool bit10 = (offset_bytes >> 10) & 1;
-
- if (bit9 ^ bit10)
- offset_bytes ^= (1 << 6);
-#else
- /* Optimized, obfuscated version. */
- offset_bytes ^= ((offset_bytes >> 3) ^ (offset_bytes >> 4))
- & (1 << 6);
-#endif
- }
-
- const uint32_t swizzle_bound_pixels = ALIGN(x_pixels + 1, swizzle_width_pixels);
- const uint32_t memcpy_bound_pixels = MIN2(x_max_pixels, swizzle_bound_pixels);
- const uint32_t copy_size = cpp * (memcpy_bound_pixels - x_pixels);
-
- memcpy(bo->virtual + offset_bytes, pixels, copy_size);
- pixels += copy_size;
- x_pixels -= (x_pixels % swizzle_width_pixels);
- }
- }
-
- drm_intel_bo_unmap(bo);
+ DBG("%s: level=%d offset=(%d,%d) (w,h)=(%d,%d) format=0x%x type=0x%x "
+ "mesa_format=0x%x tiling=%d "
+ "packing=(alignment=%d row_length=%d skip_pixels=%d skip_rows=%d) "
+ "for_glTexImage=%d\n",
+ __func__, texImage->Level, xoffset, yoffset, width, height,
+ format, type, texImage->TexFormat, image->mt->tiling,
+ packing->Alignment, packing->RowLength, packing->SkipPixels,
+ packing->SkipRows, for_glTexImage);
+
+ int level = texImage->Level + texImage->TexObject->MinLevel;
+
+ /* Adjust x and y offset based on miplevel */
+ xoffset += image->mt->level[level].level_x;
+ yoffset += image->mt->level[level].level_y;
+
+ linear_to_tiled(
+ xoffset * cpp, (xoffset + width) * cpp,
+ yoffset, yoffset + height,
+ bo->virtual,
+ pixels - (ptrdiff_t) yoffset * src_pitch - (ptrdiff_t) xoffset * cpp,
+ image->mt->pitch, src_pitch,
+ brw->has_swizzling,
+ image->mt->tiling,
+ mem_copy
+ );
+
+ brw_bo_unmap(bo);
return true;
}
const GLvoid * pixels,
const struct gl_pixelstore_attrib *packing)
{
+ struct intel_mipmap_tree *mt = intel_texture_image(texImage)->mt;
bool ok;
+ bool tex_busy = mt && brw_bo_busy(mt->bo);
+
+ if (mt && mt->format == MESA_FORMAT_S_UINT8)
+ mt->r8stencil_needs_update = true;
+
+ DBG("%s mesa_format %s target %s format %s type %s level %d %dx%dx%d\n",
+ __func__, _mesa_get_format_name(texImage->TexFormat),
+ _mesa_enum_to_string(texImage->TexObject->Target),
+ _mesa_enum_to_string(format), _mesa_enum_to_string(type),
+ texImage->Level, texImage->Width, texImage->Height, texImage->Depth);
+
+ ok = _mesa_meta_pbo_TexSubImage(ctx, dims, texImage,
+ xoffset, yoffset, zoffset,
+ width, height, depth, format, type,
+ pixels, tex_busy, packing);
+ if (ok)
+ return;
+
ok = intel_texsubimage_tiled_memcpy(ctx, dims, texImage,
xoffset, yoffset, zoffset,
width, height, depth,
if (ok)
return;
- /* The intel_blit_texsubimage() function only handles 2D images */
- if (dims != 2 || !intel_blit_texsubimage(ctx, texImage,
- xoffset, yoffset,
- width, height,
- format, type, pixels, packing)) {
- _mesa_store_texsubimage(ctx, dims, texImage,
- xoffset, yoffset, zoffset,
- width, height, depth,
- format, type, pixels, packing);
- }
+ _mesa_store_texsubimage(ctx, dims, texImage,
+ xoffset, yoffset, zoffset,
+ width, height, depth,
+ format, type, pixels, packing);
}
void