2 * Copyright 2003 VMware, Inc.
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:
13 * The above copyright notice and this permission notice (including the
14 * next paragraph) shall be included in all copies or substantial portions
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.
26 #include "main/mtypes.h"
27 #include "main/blit.h"
28 #include "main/context.h"
29 #include "main/enums.h"
30 #include "main/fbobject.h"
32 #include "brw_context.h"
33 #include "brw_defines.h"
34 #include "intel_blit.h"
35 #include "intel_buffers.h"
36 #include "intel_fbo.h"
37 #include "intel_batchbuffer.h"
38 #include "intel_mipmap_tree.h"
40 #define FILE_DEBUG_FLAG DEBUG_BLIT
42 #define SET_TILING_XY_FAST_COPY_BLT(tiling, tr_mode, type) \
46 CMD |= type ## _TILED_X; \
49 if (tr_mode == INTEL_MIPTREE_TRMODE_YS) \
50 CMD |= type ## _TILED_64K; \
52 CMD |= type ## _TILED_Y; \
55 unreachable("not reached"); \
60 intel_miptree_set_alpha_to_one(struct brw_context
*brw
,
61 struct intel_mipmap_tree
*mt
,
62 int x
, int y
, int width
, int height
);
64 static GLuint
translate_raster_op(GLenum logicop
)
67 case GL_CLEAR
: return 0x00;
68 case GL_AND
: return 0x88;
69 case GL_AND_REVERSE
: return 0x44;
70 case GL_COPY
: return 0xCC;
71 case GL_AND_INVERTED
: return 0x22;
72 case GL_NOOP
: return 0xAA;
73 case GL_XOR
: return 0x66;
74 case GL_OR
: return 0xEE;
75 case GL_NOR
: return 0x11;
76 case GL_EQUIV
: return 0x99;
77 case GL_INVERT
: return 0x55;
78 case GL_OR_REVERSE
: return 0xDD;
79 case GL_COPY_INVERTED
: return 0x33;
80 case GL_OR_INVERTED
: return 0xBB;
81 case GL_NAND
: return 0x77;
82 case GL_SET
: return 0xFF;
102 unreachable("not reached");
107 * Emits the packet for switching the blitter from X to Y tiled or back.
109 * This has to be called in a single BEGIN_BATCH_BLT_TILED() /
110 * ADVANCE_BATCH_TILED(). This is because BCS_SWCTRL is saved and restored as
111 * part of the power context, not a render context, and if the batchbuffer was
112 * to get flushed between setting and blitting, or blitting and restoring, our
113 * tiling state would leak into other unsuspecting applications (like the X
117 set_blitter_tiling(struct brw_context
*brw
,
118 bool dst_y_tiled
, bool src_y_tiled
,
121 assert(brw
->gen
>= 6);
123 /* Idle the blitter before we update how tiling is interpreted. */
124 OUT_BATCH(MI_FLUSH_DW
);
129 OUT_BATCH(MI_LOAD_REGISTER_IMM
| (3 - 2));
130 OUT_BATCH(BCS_SWCTRL
);
131 OUT_BATCH((BCS_SWCTRL_DST_Y
| BCS_SWCTRL_SRC_Y
) << 16 |
132 (dst_y_tiled
? BCS_SWCTRL_DST_Y
: 0) |
133 (src_y_tiled
? BCS_SWCTRL_SRC_Y
: 0));
136 #define SET_BLITTER_TILING(...) __map = set_blitter_tiling(__VA_ARGS__, __map)
138 #define BEGIN_BATCH_BLT_TILED(n, dst_y_tiled, src_y_tiled) \
139 BEGIN_BATCH_BLT(n + ((dst_y_tiled || src_y_tiled) ? 14 : 0)); \
140 if (dst_y_tiled || src_y_tiled) \
141 SET_BLITTER_TILING(brw, dst_y_tiled, src_y_tiled)
143 #define ADVANCE_BATCH_TILED(dst_y_tiled, src_y_tiled) \
144 if (dst_y_tiled || src_y_tiled) \
145 SET_BLITTER_TILING(brw, false, false); \
149 blt_pitch(struct intel_mipmap_tree
*mt
)
151 int pitch
= mt
->pitch
;
158 intel_miptree_blit_compatible_formats(mesa_format src
, mesa_format dst
)
160 /* The BLT doesn't handle sRGB conversion */
161 assert(src
== _mesa_get_srgb_format_linear(src
));
162 assert(dst
== _mesa_get_srgb_format_linear(dst
));
164 /* No swizzle or format conversions possible, except... */
168 /* ...we can either discard the alpha channel when going from A->X,
169 * or we can fill the alpha channel with 0xff when going from X->A
171 if (src
== MESA_FORMAT_B8G8R8A8_UNORM
|| src
== MESA_FORMAT_B8G8R8X8_UNORM
)
172 return (dst
== MESA_FORMAT_B8G8R8A8_UNORM
||
173 dst
== MESA_FORMAT_B8G8R8X8_UNORM
);
175 if (src
== MESA_FORMAT_R8G8B8A8_UNORM
|| src
== MESA_FORMAT_R8G8B8X8_UNORM
)
176 return (dst
== MESA_FORMAT_R8G8B8A8_UNORM
||
177 dst
== MESA_FORMAT_R8G8B8X8_UNORM
);
183 get_blit_intratile_offset_el(const struct brw_context
*brw
,
184 struct intel_mipmap_tree
*mt
,
185 uint32_t total_x_offset_el
,
186 uint32_t total_y_offset_el
,
187 uint32_t *base_address_offset
,
188 uint32_t *x_offset_el
,
189 uint32_t *y_offset_el
)
191 enum isl_tiling tiling
= intel_miptree_get_isl_tiling(mt
);
192 isl_tiling_get_intratile_offset_el(&brw
->isl_dev
,
193 tiling
, mt
->cpp
, mt
->pitch
,
194 total_x_offset_el
, total_y_offset_el
,
196 x_offset_el
, y_offset_el
);
197 if (tiling
== ISL_TILING_LINEAR
) {
198 /* From the Broadwell PRM docs for XY_SRC_COPY_BLT::SourceBaseAddress:
200 * "Base address of the destination surface: X=0, Y=0. Lower 32bits
201 * of the 48bit addressing. When Src Tiling is enabled (Bit_15
202 * enabled), this address must be 4KB-aligned. When Tiling is not
203 * enabled, this address should be CL (64byte) aligned."
205 * The offsets we get from ISL in the tiled case are already aligned.
206 * In the linear case, we need to do some of our own aligning.
208 assert(mt
->pitch
% 64 == 0);
209 uint32_t delta
= *base_address_offset
& 63;
210 assert(delta
% mt
->cpp
== 0);
211 *base_address_offset
-= delta
;
212 *x_offset_el
+= delta
/ mt
->cpp
;
214 assert(*base_address_offset
% 4096 == 0);
219 emit_miptree_blit(struct brw_context
*brw
,
220 struct intel_mipmap_tree
*src_mt
,
221 uint32_t src_x
, uint32_t src_y
,
222 struct intel_mipmap_tree
*dst_mt
,
223 uint32_t dst_x
, uint32_t dst_y
,
224 uint32_t width
, uint32_t height
,
225 bool reverse
, GLenum logicop
)
227 /* According to the Ivy Bridge PRM, Vol1 Part4, section 1.2.1.2 (Graphics
228 * Data Size Limitations):
230 * The BLT engine is capable of transferring very large quantities of
231 * graphics data. Any graphics data read from and written to the
232 * destination is permitted to represent a number of pixels that
233 * occupies up to 65,536 scan lines and up to 32,768 bytes per scan line
234 * at the destination. The maximum number of pixels that may be
235 * represented per scan line’s worth of graphics data depends on the
238 * The blitter's pitch is a signed 16-bit integer, but measured in bytes
239 * for linear surfaces and DWords for tiled surfaces. So the maximum
240 * pitch is 32k linear and 128k tiled.
242 if (blt_pitch(src_mt
) >= 32768 || blt_pitch(dst_mt
) >= 32768) {
243 perf_debug("Falling back due to >= 32k/128k pitch\n");
247 /* We need to split the blit into chunks that each fit within the blitter's
248 * restrictions. We can't use a chunk size of 32768 because we need to
249 * ensure that src_tile_x + chunk_size fits. We choose 16384 because it's
250 * a nice round power of two, big enough that performance won't suffer, and
251 * small enough to guarantee everything fits.
253 const uint32_t max_chunk_size
= 16384;
255 for (uint32_t chunk_x
= 0; chunk_x
< width
; chunk_x
+= max_chunk_size
) {
256 for (uint32_t chunk_y
= 0; chunk_y
< height
; chunk_y
+= max_chunk_size
) {
257 const uint32_t chunk_w
= MIN2(max_chunk_size
, width
- chunk_x
);
258 const uint32_t chunk_h
= MIN2(max_chunk_size
, height
- chunk_y
);
260 uint32_t src_offset
, src_tile_x
, src_tile_y
;
261 get_blit_intratile_offset_el(brw
, src_mt
,
262 src_x
+ chunk_x
, src_y
+ chunk_y
,
263 &src_offset
, &src_tile_x
, &src_tile_y
);
265 uint32_t dst_offset
, dst_tile_x
, dst_tile_y
;
266 get_blit_intratile_offset_el(brw
, dst_mt
,
267 dst_x
+ chunk_x
, dst_y
+ chunk_y
,
268 &dst_offset
, &dst_tile_x
, &dst_tile_y
);
270 if (!intelEmitCopyBlit(brw
,
272 reverse
? -src_mt
->pitch
: src_mt
->pitch
,
273 src_mt
->bo
, src_mt
->offset
+ src_offset
,
277 dst_mt
->bo
, dst_mt
->offset
+ dst_offset
,
280 src_tile_x
, src_tile_y
,
281 dst_tile_x
, dst_tile_y
,
284 /* If this is ever going to fail, it will fail on the first chunk */
285 assert(chunk_x
== 0 && chunk_y
== 0);
295 * Implements a rectangular block transfer (blit) of pixels between two
298 * Our blitter can operate on 1, 2, or 4-byte-per-pixel data, with generous,
299 * but limited, pitches and sizes allowed.
301 * The src/dst coordinates are relative to the given level/slice of the
304 * If @src_flip or @dst_flip is set, then the rectangle within that miptree
305 * will be inverted (including scanline order) when copying. This is common
306 * in GL when copying between window system and user-created
307 * renderbuffers/textures.
310 intel_miptree_blit(struct brw_context
*brw
,
311 struct intel_mipmap_tree
*src_mt
,
312 int src_level
, int src_slice
,
313 uint32_t src_x
, uint32_t src_y
, bool src_flip
,
314 struct intel_mipmap_tree
*dst_mt
,
315 int dst_level
, int dst_slice
,
316 uint32_t dst_x
, uint32_t dst_y
, bool dst_flip
,
317 uint32_t width
, uint32_t height
,
320 /* The blitter doesn't understand multisampling at all. */
321 if (src_mt
->num_samples
> 0 || dst_mt
->num_samples
> 0)
324 /* No sRGB decode or encode is done by the hardware blitter, which is
325 * consistent with what we want in many callers (glCopyTexSubImage(),
326 * texture validation, etc.).
328 mesa_format src_format
= _mesa_get_srgb_format_linear(src_mt
->format
);
329 mesa_format dst_format
= _mesa_get_srgb_format_linear(dst_mt
->format
);
331 /* The blitter doesn't support doing any format conversions. We do also
332 * support blitting ARGB8888 to XRGB8888 (trivial, the values dropped into
333 * the X channel don't matter), and XRGB8888 to ARGB8888 by setting the A
334 * channel to 1.0 at the end.
336 if (!intel_miptree_blit_compatible_formats(src_format
, dst_format
)) {
337 perf_debug("%s: Can't use hardware blitter from %s to %s, "
338 "falling back.\n", __func__
,
339 _mesa_get_format_name(src_format
),
340 _mesa_get_format_name(dst_format
));
344 /* The blitter has no idea about HiZ or fast color clears, so we need to
345 * resolve the miptrees before we do anything.
347 intel_miptree_slice_resolve_depth(brw
, src_mt
, src_level
, src_slice
);
348 intel_miptree_slice_resolve_depth(brw
, dst_mt
, dst_level
, dst_slice
);
349 intel_miptree_resolve_color(brw
, src_mt
, src_level
, src_slice
, 1, 0);
350 intel_miptree_resolve_color(brw
, dst_mt
, dst_level
, dst_slice
, 1, 0);
353 src_y
= minify(src_mt
->physical_height0
, src_level
- src_mt
->first_level
) - src_y
- height
;
356 dst_y
= minify(dst_mt
->physical_height0
, dst_level
- dst_mt
->first_level
) - dst_y
- height
;
358 uint32_t src_image_x
, src_image_y
, dst_image_x
, dst_image_y
;
359 intel_miptree_get_image_offset(src_mt
, src_level
, src_slice
,
360 &src_image_x
, &src_image_y
);
361 intel_miptree_get_image_offset(dst_mt
, dst_level
, dst_slice
,
362 &dst_image_x
, &dst_image_y
);
363 src_x
+= src_image_x
;
364 src_y
+= src_image_y
;
365 dst_x
+= dst_image_x
;
366 dst_y
+= dst_image_y
;
368 if (!emit_miptree_blit(brw
, src_mt
, src_x
, src_y
,
369 dst_mt
, dst_x
, dst_y
, width
, height
,
370 src_flip
!= dst_flip
, logicop
)) {
374 /* XXX This could be done in a single pass using XY_FULL_MONO_PATTERN_BLT */
375 if (_mesa_get_format_bits(src_format
, GL_ALPHA_BITS
) == 0 &&
376 _mesa_get_format_bits(dst_format
, GL_ALPHA_BITS
) > 0) {
377 intel_miptree_set_alpha_to_one(brw
, dst_mt
,
386 intel_miptree_copy(struct brw_context
*brw
,
387 struct intel_mipmap_tree
*src_mt
,
388 int src_level
, int src_slice
,
389 uint32_t src_x
, uint32_t src_y
,
390 struct intel_mipmap_tree
*dst_mt
,
391 int dst_level
, int dst_slice
,
392 uint32_t dst_x
, uint32_t dst_y
,
393 uint32_t src_width
, uint32_t src_height
)
395 /* The blitter doesn't understand multisampling at all. */
396 if (src_mt
->num_samples
> 0 || dst_mt
->num_samples
> 0)
399 if (src_mt
->format
== MESA_FORMAT_S_UINT8
)
402 /* The blitter has no idea about HiZ or fast color clears, so we need to
403 * resolve the miptrees before we do anything.
405 intel_miptree_slice_resolve_depth(brw
, src_mt
, src_level
, src_slice
);
406 intel_miptree_slice_resolve_depth(brw
, dst_mt
, dst_level
, dst_slice
);
407 intel_miptree_resolve_color(brw
, src_mt
, src_level
, src_slice
, 1, 0);
408 intel_miptree_resolve_color(brw
, dst_mt
, dst_level
, dst_slice
, 1, 0);
410 uint32_t src_image_x
, src_image_y
;
411 intel_miptree_get_image_offset(src_mt
, src_level
, src_slice
,
412 &src_image_x
, &src_image_y
);
414 if (_mesa_is_format_compressed(src_mt
->format
)) {
416 _mesa_get_format_block_size(src_mt
->format
, &bw
, &bh
);
418 /* Compressed textures need not have dimensions that are a multiple of
419 * the block size. Rectangles in compressed textures do need to be a
420 * multiple of the block size. The one exception is that the right and
421 * bottom edges may be at the right or bottom edge of the miplevel even
422 * if it's not aligned.
424 assert(src_x
% bw
== 0);
425 assert(src_y
% bh
== 0);
426 assert(src_width
% bw
== 0 ||
427 src_x
+ src_width
== minify(src_mt
->logical_width0
, src_level
));
428 assert(src_height
% bh
== 0 ||
429 src_y
+ src_height
== minify(src_mt
->logical_height0
, src_level
));
433 src_width
/= (int)bw
;
434 src_height
/= (int)bh
;
436 src_x
+= src_image_x
;
437 src_y
+= src_image_y
;
439 uint32_t dst_image_x
, dst_image_y
;
440 intel_miptree_get_image_offset(dst_mt
, dst_level
, dst_slice
,
441 &dst_image_x
, &dst_image_y
);
443 if (_mesa_is_format_compressed(dst_mt
->format
)) {
445 _mesa_get_format_block_size(dst_mt
->format
, &bw
, &bh
);
447 assert(dst_x
% bw
== 0);
448 assert(dst_y
% bh
== 0);
453 dst_x
+= dst_image_x
;
454 dst_y
+= dst_image_y
;
456 return emit_miptree_blit(brw
, src_mt
, src_x
, src_y
,
457 dst_mt
, dst_x
, dst_y
,
458 src_width
, src_height
, false, GL_COPY
);
462 alignment_valid(struct brw_context
*brw
, unsigned offset
, uint32_t tiling
)
464 /* Tiled buffers must be page-aligned (4K). */
465 if (tiling
!= I915_TILING_NONE
)
466 return (offset
& 4095) == 0;
468 /* On Gen8+, linear buffers must be cacheline-aligned. */
470 return (offset
& 63) == 0;
476 can_fast_copy_blit(struct brw_context
*brw
,
477 drm_intel_bo
*src_buffer
,
478 int16_t src_x
, int16_t src_y
,
479 uintptr_t src_offset
, uint32_t src_pitch
,
480 uint32_t src_tiling
, uint32_t src_tr_mode
,
481 drm_intel_bo
*dst_buffer
,
482 int16_t dst_x
, int16_t dst_y
,
483 uintptr_t dst_offset
, uint32_t dst_pitch
,
484 uint32_t dst_tiling
, uint32_t dst_tr_mode
,
485 int16_t w
, int16_t h
, uint32_t cpp
,
488 const bool dst_tiling_none
= dst_tiling
== I915_TILING_NONE
;
489 const bool src_tiling_none
= src_tiling
== I915_TILING_NONE
;
494 /* Enable fast copy blit only if the surfaces are Yf/Ys tiled.
495 * FIXME: Based on performance data, remove this condition later to
496 * enable for all types of surfaces.
498 if (src_tr_mode
== INTEL_MIPTREE_TRMODE_NONE
&&
499 dst_tr_mode
== INTEL_MIPTREE_TRMODE_NONE
)
502 if (logic_op
!= GL_COPY
)
505 /* The start pixel for Fast Copy blit should be on an OWord boundary. */
506 if ((dst_x
* cpp
| src_x
* cpp
) & 15)
509 /* For all surface types buffers must be cacheline-aligned. */
510 if ((dst_offset
| src_offset
) & 63)
513 /* Color depths which are not power of 2 or greater than 128 bits are
516 if (!_mesa_is_pow_two(cpp
) || cpp
> 16)
519 /* For Fast Copy Blits the pitch cannot be a negative number. So, bit 15
520 * of the destination pitch must be zero.
522 if ((src_pitch
>> 15 & 1) != 0 || (dst_pitch
>> 15 & 1) != 0)
525 /* For Linear surfaces, the pitch has to be an OWord (16byte) multiple. */
526 if ((src_tiling_none
&& src_pitch
% 16 != 0) ||
527 (dst_tiling_none
&& dst_pitch
% 16 != 0))
534 xy_blit_cmd(uint32_t src_tiling
, uint32_t src_tr_mode
,
535 uint32_t dst_tiling
, uint32_t dst_tr_mode
,
536 uint32_t cpp
, bool use_fast_copy_blit
)
540 if (use_fast_copy_blit
) {
541 CMD
= XY_FAST_COPY_BLT_CMD
;
543 if (dst_tiling
!= I915_TILING_NONE
)
544 SET_TILING_XY_FAST_COPY_BLT(dst_tiling
, dst_tr_mode
, XY_FAST_DST
);
546 if (src_tiling
!= I915_TILING_NONE
)
547 SET_TILING_XY_FAST_COPY_BLT(src_tiling
, src_tr_mode
, XY_FAST_SRC
);
553 CMD
= XY_SRC_COPY_BLT_CMD
;
556 CMD
= XY_SRC_COPY_BLT_CMD
| XY_BLT_WRITE_ALPHA
| XY_BLT_WRITE_RGB
;
559 unreachable("not reached");
562 if (dst_tiling
!= I915_TILING_NONE
)
565 if (src_tiling
!= I915_TILING_NONE
)
574 intelEmitCopyBlit(struct brw_context
*brw
,
577 drm_intel_bo
*src_buffer
,
580 uint32_t src_tr_mode
,
582 drm_intel_bo
*dst_buffer
,
585 uint32_t dst_tr_mode
,
586 GLshort src_x
, GLshort src_y
,
587 GLshort dst_x
, GLshort dst_y
,
588 GLshort w
, GLshort h
,
591 GLuint CMD
, BR13
, pass
= 0;
592 int dst_y2
= dst_y
+ h
;
593 int dst_x2
= dst_x
+ w
;
594 drm_intel_bo
*aper_array
[3];
595 bool dst_y_tiled
= dst_tiling
== I915_TILING_Y
;
596 bool src_y_tiled
= src_tiling
== I915_TILING_Y
;
597 bool use_fast_copy_blit
= false;
598 uint32_t src_tile_w
, src_tile_h
;
599 uint32_t dst_tile_w
, dst_tile_h
;
601 if ((dst_y_tiled
|| src_y_tiled
) && brw
->gen
< 6)
604 /* do space check before going any further */
606 aper_array
[0] = brw
->batch
.bo
;
607 aper_array
[1] = dst_buffer
;
608 aper_array
[2] = src_buffer
;
610 if (dri_bufmgr_check_aperture_space(aper_array
, 3) != 0) {
611 intel_batchbuffer_flush(brw
);
620 unsigned length
= brw
->gen
>= 8 ? 10 : 8;
622 intel_batchbuffer_require_space(brw
, length
* 4, BLT_RING
);
623 DBG("%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
625 src_buffer
, src_pitch
, src_offset
, src_x
, src_y
,
626 dst_buffer
, dst_pitch
, dst_offset
, dst_x
, dst_y
, w
, h
);
628 intel_get_tile_dims(src_tiling
, src_tr_mode
, cpp
, &src_tile_w
, &src_tile_h
);
629 intel_get_tile_dims(dst_tiling
, dst_tr_mode
, cpp
, &dst_tile_w
, &dst_tile_h
);
631 /* For Tiled surfaces, the pitch has to be a multiple of the Tile width
632 * (X direction width of the Tile). This is ensured while allocating the
635 assert(src_tiling
== I915_TILING_NONE
|| (src_pitch
% src_tile_w
) == 0);
636 assert(dst_tiling
== I915_TILING_NONE
|| (dst_pitch
% dst_tile_w
) == 0);
638 use_fast_copy_blit
= can_fast_copy_blit(brw
,
641 src_offset
, src_pitch
,
642 src_tiling
, src_tr_mode
,
645 dst_offset
, dst_pitch
,
646 dst_tiling
, dst_tr_mode
,
647 w
, h
, cpp
, logic_op
);
648 if (!use_fast_copy_blit
&&
649 (src_tr_mode
!= INTEL_MIPTREE_TRMODE_NONE
||
650 dst_tr_mode
!= INTEL_MIPTREE_TRMODE_NONE
))
653 if (use_fast_copy_blit
) {
654 assert(logic_op
== GL_COPY
);
656 /* When two sequential fast copy blits have different source surfaces,
657 * but their destinations refer to the same destination surfaces and
658 * therefore destinations overlap it is imperative that a flush be
659 * inserted between the two blits.
661 * FIXME: Figure out a way to avoid flushing when not required.
663 brw_emit_mi_flush(brw
);
666 BR13
= br13_for_cpp(cpp
);
668 if (src_tr_mode
== INTEL_MIPTREE_TRMODE_YF
)
669 BR13
|= XY_FAST_SRC_TRMODE_YF
;
671 if (dst_tr_mode
== INTEL_MIPTREE_TRMODE_YF
)
672 BR13
|= XY_FAST_DST_TRMODE_YF
;
674 CMD
= xy_blit_cmd(src_tiling
, src_tr_mode
,
675 dst_tiling
, dst_tr_mode
,
676 cpp
, use_fast_copy_blit
);
679 /* For big formats (such as floating point), do the copy using 16 or
680 * 32bpp and multiply the coordinates.
689 assert(cpp
% 4 == 0);
697 if (!alignment_valid(brw
, dst_offset
, dst_tiling
))
699 if (!alignment_valid(brw
, src_offset
, src_tiling
))
702 /* Blit pitch must be dword-aligned. Otherwise, the hardware appears to drop
703 * the low bits. Offsets must be naturally aligned.
705 if (src_pitch
% 4 != 0 || src_offset
% cpp
!= 0 ||
706 dst_pitch
% 4 != 0 || dst_offset
% cpp
!= 0)
710 BR13
= br13_for_cpp(cpp
) | translate_raster_op(logic_op
) << 16;
712 CMD
= xy_blit_cmd(src_tiling
, src_tr_mode
,
713 dst_tiling
, dst_tr_mode
,
714 cpp
, use_fast_copy_blit
);
717 /* For tiled source and destination, pitch value should be specified
718 * as a number of Dwords.
720 if (dst_tiling
!= I915_TILING_NONE
)
723 if (src_tiling
!= I915_TILING_NONE
)
726 if (dst_y2
<= dst_y
|| dst_x2
<= dst_x
)
729 assert(dst_x
< dst_x2
);
730 assert(dst_y
< dst_y2
);
732 BEGIN_BATCH_BLT_TILED(length
, dst_y_tiled
, src_y_tiled
);
733 OUT_BATCH(CMD
| (length
- 2));
734 OUT_BATCH(BR13
| (uint16_t)dst_pitch
);
735 OUT_BATCH(SET_FIELD(dst_y
, BLT_Y
) | SET_FIELD(dst_x
, BLT_X
));
736 OUT_BATCH(SET_FIELD(dst_y2
, BLT_Y
) | SET_FIELD(dst_x2
, BLT_X
));
738 OUT_RELOC64(dst_buffer
,
739 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
742 OUT_RELOC(dst_buffer
,
743 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
746 OUT_BATCH(SET_FIELD(src_y
, BLT_Y
) | SET_FIELD(src_x
, BLT_X
));
747 OUT_BATCH((uint16_t)src_pitch
);
749 OUT_RELOC64(src_buffer
,
750 I915_GEM_DOMAIN_RENDER
, 0,
753 OUT_RELOC(src_buffer
,
754 I915_GEM_DOMAIN_RENDER
, 0,
758 ADVANCE_BATCH_TILED(dst_y_tiled
, src_y_tiled
);
760 brw_emit_mi_flush(brw
);
766 intelEmitImmediateColorExpandBlit(struct brw_context
*brw
,
768 GLubyte
*src_bits
, GLuint src_size
,
771 drm_intel_bo
*dst_buffer
,
774 GLshort x
, GLshort y
,
775 GLshort w
, GLshort h
,
778 int dwords
= ALIGN(src_size
, 8) / 4;
779 uint32_t opcode
, br13
, blit_cmd
;
781 if (dst_tiling
!= I915_TILING_NONE
) {
782 if (dst_offset
& 4095)
784 if (dst_tiling
== I915_TILING_Y
)
788 assert((logic_op
>= GL_CLEAR
) && (logic_op
<= (GL_CLEAR
+ 0x0f)));
789 assert(dst_pitch
> 0);
794 DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d, %d bytes %d dwords\n",
796 dst_buffer
, dst_pitch
, dst_offset
, x
, y
, w
, h
, src_size
, dwords
);
798 unsigned xy_setup_blt_length
= brw
->gen
>= 8 ? 10 : 8;
799 intel_batchbuffer_require_space(brw
, (xy_setup_blt_length
* 4) +
800 (3 * 4) + dwords
* 4, BLT_RING
);
802 opcode
= XY_SETUP_BLT_CMD
;
804 opcode
|= XY_BLT_WRITE_ALPHA
| XY_BLT_WRITE_RGB
;
805 if (dst_tiling
!= I915_TILING_NONE
) {
806 opcode
|= XY_DST_TILED
;
810 br13
= dst_pitch
| (translate_raster_op(logic_op
) << 16) | (1 << 29);
811 br13
|= br13_for_cpp(cpp
);
813 blit_cmd
= XY_TEXT_IMMEDIATE_BLIT_CMD
| XY_TEXT_BYTE_PACKED
; /* packing? */
814 if (dst_tiling
!= I915_TILING_NONE
)
815 blit_cmd
|= XY_DST_TILED
;
817 BEGIN_BATCH_BLT(xy_setup_blt_length
+ 3);
818 OUT_BATCH(opcode
| (xy_setup_blt_length
- 2));
820 OUT_BATCH((0 << 16) | 0); /* clip x1, y1 */
821 OUT_BATCH((100 << 16) | 100); /* clip x2, y2 */
823 OUT_RELOC64(dst_buffer
,
824 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
827 OUT_RELOC(dst_buffer
,
828 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
831 OUT_BATCH(0); /* bg */
832 OUT_BATCH(fg_color
); /* fg */
833 OUT_BATCH(0); /* pattern base addr */
837 OUT_BATCH(blit_cmd
| ((3 - 2) + dwords
));
838 OUT_BATCH(SET_FIELD(y
, BLT_Y
) | SET_FIELD(x
, BLT_X
));
839 OUT_BATCH(SET_FIELD(y
+ h
, BLT_Y
) | SET_FIELD(x
+ w
, BLT_X
));
842 intel_batchbuffer_data(brw
, src_bits
, dwords
* 4, BLT_RING
);
844 brw_emit_mi_flush(brw
);
849 /* We don't have a memmove-type blit like some other hardware, so we'll do a
850 * rectangular blit covering a large space, then emit 1-scanline blit at the
851 * end to cover the last if we need.
854 intel_emit_linear_blit(struct brw_context
*brw
,
855 drm_intel_bo
*dst_bo
,
856 unsigned int dst_offset
,
857 drm_intel_bo
*src_bo
,
858 unsigned int src_offset
,
861 struct gl_context
*ctx
= &brw
->ctx
;
862 GLuint pitch
, height
;
863 int16_t src_x
, dst_x
;
867 /* The pitch given to the GPU must be DWORD aligned, and
868 * we want width to match pitch. Max width is (1 << 15 - 1),
869 * rounding that down to the nearest DWORD is 1 << 15 - 4
871 pitch
= ROUND_DOWN_TO(MIN2(size
, (1 << 15) - 64), 4);
872 height
= (size
< pitch
|| pitch
== 0) ? 1 : size
/ pitch
;
874 src_x
= src_offset
% 64;
875 dst_x
= dst_offset
% 64;
876 pitch
= ALIGN(MIN2(size
, (1 << 15) - 64), 4);
877 assert(src_x
+ pitch
< 1 << 15);
878 assert(dst_x
+ pitch
< 1 << 15);
880 ok
= intelEmitCopyBlit(brw
, 1,
881 pitch
, src_bo
, src_offset
- src_x
, I915_TILING_NONE
,
882 INTEL_MIPTREE_TRMODE_NONE
,
883 pitch
, dst_bo
, dst_offset
- dst_x
, I915_TILING_NONE
,
884 INTEL_MIPTREE_TRMODE_NONE
,
885 src_x
, 0, /* src x/y */
886 dst_x
, 0, /* dst x/y */
887 MIN2(size
, pitch
), height
, /* w, h */
890 _mesa_problem(ctx
, "Failed to linear blit %dx%d\n",
891 MIN2(size
, pitch
), height
);
906 * Used to initialize the alpha value of an ARGB8888 miptree after copying
907 * into it from an XRGB8888 source.
909 * This is very common with glCopyTexImage2D(). Note that the coordinates are
910 * relative to the start of the miptree, not relative to a slice within the
914 intel_miptree_set_alpha_to_one(struct brw_context
*brw
,
915 struct intel_mipmap_tree
*mt
,
916 int x
, int y
, int width
, int height
)
920 drm_intel_bo
*aper_array
[2];
925 DBG("%s dst:buf(%p)/%d %d,%d sz:%dx%d\n",
926 __func__
, mt
->bo
, pitch
, x
, y
, width
, height
);
928 BR13
= br13_for_cpp(cpp
) | 0xf0 << 16;
929 CMD
= XY_COLOR_BLT_CMD
;
930 CMD
|= XY_BLT_WRITE_ALPHA
;
932 if (mt
->tiling
!= I915_TILING_NONE
) {
938 /* do space check before going any further */
939 aper_array
[0] = brw
->batch
.bo
;
940 aper_array
[1] = mt
->bo
;
942 if (drm_intel_bufmgr_check_aperture_space(aper_array
,
943 ARRAY_SIZE(aper_array
)) != 0) {
944 intel_batchbuffer_flush(brw
);
947 unsigned length
= brw
->gen
>= 8 ? 7 : 6;
948 bool dst_y_tiled
= mt
->tiling
== I915_TILING_Y
;
950 /* We need to split the blit into chunks that each fit within the blitter's
951 * restrictions. We can't use a chunk size of 32768 because we need to
952 * ensure that src_tile_x + chunk_size fits. We choose 16384 because it's
953 * a nice round power of two, big enough that performance won't suffer, and
954 * small enough to guarantee everything fits.
956 const uint32_t max_chunk_size
= 16384;
958 for (uint32_t chunk_x
= 0; chunk_x
< width
; chunk_x
+= max_chunk_size
) {
959 for (uint32_t chunk_y
= 0; chunk_y
< height
; chunk_y
+= max_chunk_size
) {
960 const uint32_t chunk_w
= MIN2(max_chunk_size
, width
- chunk_x
);
961 const uint32_t chunk_h
= MIN2(max_chunk_size
, height
- chunk_y
);
963 uint32_t offset
, tile_x
, tile_y
;
964 get_blit_intratile_offset_el(brw
, mt
,
965 x
+ chunk_x
, y
+ chunk_y
,
966 &offset
, &tile_x
, &tile_y
);
968 BEGIN_BATCH_BLT_TILED(length
, dst_y_tiled
, false);
969 OUT_BATCH(CMD
| (length
- 2));
971 OUT_BATCH(SET_FIELD(y
+ chunk_y
, BLT_Y
) |
972 SET_FIELD(x
+ chunk_x
, BLT_X
));
973 OUT_BATCH(SET_FIELD(y
+ chunk_y
+ chunk_h
, BLT_Y
) |
974 SET_FIELD(x
+ chunk_x
+ chunk_w
, BLT_X
));
977 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
981 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
984 OUT_BATCH(0xffffffff); /* white, but only alpha gets written */
985 ADVANCE_BATCH_TILED(dst_y_tiled
, false);
989 brw_emit_mi_flush(brw
);