2 /**************************************************************************
4 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
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:
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial portions
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 TUNGSTEN GRAPHICS 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.
27 **************************************************************************/
29 #include "main/bufferobj.h"
30 #include "main/macros.h"
31 #include "main/mtypes.h"
33 #include "main/texobj.h"
34 #include "main/texstore.h"
35 #include "main/texcompress.h"
36 #include "main/enums.h"
38 #include "brw_context.h"
39 #include "intel_batchbuffer.h"
40 #include "intel_tex.h"
41 #include "intel_mipmap_tree.h"
42 #include "intel_blit.h"
44 #define FILE_DEBUG_FLAG DEBUG_TEXTURE
47 intel_blit_texsubimage(struct gl_context
* ctx
,
48 struct gl_texture_image
*texImage
,
49 GLint xoffset
, GLint yoffset
,
50 GLint width
, GLint height
,
51 GLenum format
, GLenum type
, const void *pixels
,
52 const struct gl_pixelstore_attrib
*packing
)
54 struct brw_context
*brw
= brw_context(ctx
);
55 struct intel_texture_image
*intelImage
= intel_texture_image(texImage
);
57 /* Try to do a blit upload of the subimage if the texture is
63 /* The blitter can't handle Y tiling */
64 if (intelImage
->mt
->region
->tiling
== I915_TILING_Y
)
67 if (texImage
->TexObject
->Target
!= GL_TEXTURE_2D
)
70 /* On gen6, it's probably not worth swapping to the blit ring to do
71 * this because of all the overhead involved.
76 if (!drm_intel_bo_busy(intelImage
->mt
->region
->bo
))
79 DBG("BLT subimage %s target %s level %d offset %d,%d %dx%d\n",
81 _mesa_lookup_enum_by_nr(texImage
->TexObject
->Target
),
82 texImage
->Level
, xoffset
, yoffset
, width
, height
);
84 pixels
= _mesa_validate_pbo_teximage(ctx
, 2, width
, height
, 1,
85 format
, type
, pixels
, packing
,
90 struct intel_mipmap_tree
*temp_mt
=
91 intel_miptree_create(brw
, GL_TEXTURE_2D
, texImage
->TexFormat
,
94 false, 0, INTEL_MIPTREE_TILING_NONE
);
98 GLubyte
*dst
= intel_miptree_map_raw(brw
, temp_mt
);
102 if (!_mesa_texstore(ctx
, 2, texImage
->_BaseFormat
,
104 temp_mt
->region
->pitch
,
107 format
, type
, pixels
, packing
)) {
108 _mesa_error(ctx
, GL_OUT_OF_MEMORY
, "intelTexSubImage");
111 intel_miptree_unmap_raw(brw
, temp_mt
);
115 ret
= intel_miptree_blit(brw
,
118 intelImage
->mt
, texImage
->Level
, texImage
->Face
,
119 xoffset
, yoffset
, false,
120 width
, height
, GL_COPY
);
123 intel_miptree_release(&temp_mt
);
124 _mesa_unmap_teximage_pbo(ctx
, packing
);
129 _mesa_error(ctx
, GL_OUT_OF_MEMORY
, "intelTexSubImage");
130 intel_miptree_release(&temp_mt
);
131 _mesa_unmap_teximage_pbo(ctx
, packing
);
136 * \brief A fast path for glTexImage and glTexSubImage.
138 * \param for_glTexImage Was this called from glTexImage or glTexSubImage?
140 * This fast path is taken when the hardware natively supports the texture
141 * format (such as GL_BGRA) and when the texture memory is X-tiled. It uploads
142 * the texture data by mapping the texture memory without a GTT fence, thus
143 * acquiring a tiled view of the memory, and then memcpy'ing sucessive
144 * subspans within each tile.
146 * This is a performance win over the conventional texture upload path because
147 * it avoids the performance penalty of writing through the write-combine
148 * buffer. In the conventional texture upload path,
149 * texstore.c:store_texsubimage(), the texture memory is mapped through a GTT
150 * fence, thus acquiring a linear view of the memory, then each row in the
151 * image is memcpy'd. In this fast path, we replace each row's memcpy with
152 * a sequence of memcpy's over each bit6 swizzle span in the row.
154 * This fast path's use case is Google Chrome's paint rectangles. Chrome (as
155 * of version 21) renders each page as a tiling of 256x256 GL_BGRA textures.
156 * Each page's content is initially uploaded with glTexImage2D and damaged
157 * regions are updated with glTexSubImage2D. On some workloads, the
158 * performance gain of this fastpath on Sandybridge is over 5x.
161 intel_texsubimage_tiled_memcpy(struct gl_context
* ctx
,
163 struct gl_texture_image
*texImage
,
164 GLint xoffset
, GLint yoffset
, GLint zoffset
,
165 GLsizei width
, GLsizei height
, GLsizei depth
,
166 GLenum format
, GLenum type
,
167 const GLvoid
*pixels
,
168 const struct gl_pixelstore_attrib
*packing
,
171 struct brw_context
*brw
= brw_context(ctx
);
172 struct intel_texture_image
*image
= intel_texture_image(texImage
);
174 /* The miptree's buffer. */
179 /* This fastpath is restricted to a specific texture type: level 0 of
180 * a 2D BGRA texture. It could be generalized to support more types by
181 * varying the arithmetic loop below.
185 type
!= GL_UNSIGNED_BYTE
||
186 texImage
->TexFormat
!= MESA_FORMAT_ARGB8888
||
187 texImage
->TexObject
->Target
!= GL_TEXTURE_2D
||
188 texImage
->Level
!= 0 ||
190 _mesa_is_bufferobj(packing
->BufferObj
) ||
191 packing
->Alignment
> 4 ||
192 packing
->SkipPixels
> 0 ||
193 packing
->SkipRows
> 0 ||
194 (packing
->RowLength
!= 0 && packing
->RowLength
!= width
) ||
195 packing
->SwapBytes
||
201 ctx
->Driver
.AllocTextureImageBuffer(ctx
, texImage
);
204 image
->mt
->region
->tiling
!= I915_TILING_X
) {
205 /* The algorithm below is written only for X-tiled memory. */
209 /* Since we are going to write raw data to the miptree, we need to resolve
210 * any pending fast color clears before we start.
212 intel_miptree_resolve_color(brw
, image
->mt
);
214 bo
= image
->mt
->region
->bo
;
216 if (drm_intel_bo_references(brw
->batch
.bo
, bo
)) {
217 perf_debug("Flushing before mapping a referenced bo.\n");
218 intel_batchbuffer_flush(brw
);
221 if (unlikely(brw
->perf_debug
)) {
222 if (drm_intel_bo_busy(bo
)) {
223 perf_debug("Mapping a busy BO, causing a stall on the GPU.\n");
227 error
= drm_intel_bo_map(bo
, true /*write_enable*/);
228 if (error
|| bo
->virtual == NULL
) {
229 DBG("%s: failed to map bo\n", __FUNCTION__
);
233 /* We postponed printing this message until having committed to executing
236 DBG("%s: level=%d offset=(%d,%d) (w,h)=(%d,%d)\n",
237 __FUNCTION__
, texImage
->Level
, xoffset
, yoffset
, width
, height
);
239 /* In the tiling algorithm below, some variables are in units of pixels,
240 * others are in units of bytes, and others (such as height) are unitless.
241 * Each variable name is suffixed with its units.
244 const uint32_t x_max_pixels
= xoffset
+ width
;
245 const uint32_t y_max_pixels
= yoffset
+ height
;
247 const uint32_t tile_size_bytes
= 4096;
249 const uint32_t tile_width_bytes
= 512;
250 const uint32_t tile_width_pixels
= 128;
252 const uint32_t tile_height
= 8;
254 const uint32_t cpp
= 4; /* chars per pixel of GL_BGRA */
255 const uint32_t swizzle_width_pixels
= 16;
257 const uint32_t stride_bytes
= image
->mt
->region
->pitch
;
258 const uint32_t width_tiles
= stride_bytes
/ tile_width_bytes
;
260 for (uint32_t y_pixels
= yoffset
; y_pixels
< y_max_pixels
; ++y_pixels
) {
261 const uint32_t y_offset_bytes
= (y_pixels
/ tile_height
) * width_tiles
* tile_size_bytes
262 + (y_pixels
% tile_height
) * tile_width_bytes
;
264 for (uint32_t x_pixels
= xoffset
; x_pixels
< x_max_pixels
; x_pixels
+= swizzle_width_pixels
) {
265 const uint32_t x_offset_bytes
= (x_pixels
/ tile_width_pixels
) * tile_size_bytes
266 + (x_pixels
% tile_width_pixels
) * cpp
;
268 intptr_t offset_bytes
= y_offset_bytes
+ x_offset_bytes
;
269 if (brw
->has_swizzling
) {
271 /* Clear, unoptimized version. */
272 bool bit6
= (offset_bytes
>> 6) & 1;
273 bool bit9
= (offset_bytes
>> 9) & 1;
274 bool bit10
= (offset_bytes
>> 10) & 1;
277 offset_bytes
^= (1 << 6);
279 /* Optimized, obfuscated version. */
280 offset_bytes
^= ((offset_bytes
>> 3) ^ (offset_bytes
>> 4))
285 const uint32_t swizzle_bound_pixels
= ALIGN(x_pixels
+ 1, swizzle_width_pixels
);
286 const uint32_t memcpy_bound_pixels
= MIN2(x_max_pixels
, swizzle_bound_pixels
);
287 const uint32_t copy_size
= cpp
* (memcpy_bound_pixels
- x_pixels
);
289 memcpy(bo
->virtual + offset_bytes
, pixels
, copy_size
);
291 x_pixels
-= (x_pixels
% swizzle_width_pixels
);
295 drm_intel_bo_unmap(bo
);
300 intelTexSubImage(struct gl_context
* ctx
,
302 struct gl_texture_image
*texImage
,
303 GLint xoffset
, GLint yoffset
, GLint zoffset
,
304 GLsizei width
, GLsizei height
, GLsizei depth
,
305 GLenum format
, GLenum type
,
306 const GLvoid
* pixels
,
307 const struct gl_pixelstore_attrib
*packing
)
311 ok
= intel_texsubimage_tiled_memcpy(ctx
, dims
, texImage
,
312 xoffset
, yoffset
, zoffset
,
313 width
, height
, depth
,
314 format
, type
, pixels
, packing
,
315 false /*for_glTexImage*/);
319 /* The intel_blit_texsubimage() function only handles 2D images */
320 if (dims
!= 2 || !intel_blit_texsubimage(ctx
, texImage
,
323 format
, type
, pixels
, packing
)) {
324 _mesa_store_texsubimage(ctx
, dims
, texImage
,
325 xoffset
, yoffset
, zoffset
,
326 width
, height
, depth
,
327 format
, type
, pixels
, packing
);
332 intelInitTextureSubImageFuncs(struct dd_function_table
*functions
)
334 functions
->TexSubImage
= intelTexSubImage
;