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
43 intel_miptree_set_alpha_to_one(struct brw_context
*brw
,
44 struct intel_mipmap_tree
*mt
,
45 int x
, int y
, int width
, int height
);
47 static GLuint
translate_raster_op(enum gl_logicop_mode logicop
)
49 return logicop
| (logicop
<< 4);
67 unreachable("not reached");
72 * Emits the packet for switching the blitter from X to Y tiled or back.
74 * This has to be called in a single BEGIN_BATCH_BLT_TILED() /
75 * ADVANCE_BATCH_TILED(). This is because BCS_SWCTRL is saved and restored as
76 * part of the power context, not a render context, and if the batchbuffer was
77 * to get flushed between setting and blitting, or blitting and restoring, our
78 * tiling state would leak into other unsuspecting applications (like the X
82 set_blitter_tiling(struct brw_context
*brw
,
83 bool dst_y_tiled
, bool src_y_tiled
,
86 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
87 const unsigned n_dwords
= devinfo
->gen
>= 8 ? 5 : 4;
88 assert(devinfo
->gen
>= 6);
90 /* Idle the blitter before we update how tiling is interpreted. */
91 OUT_BATCH(MI_FLUSH_DW
| (n_dwords
- 2));
98 OUT_BATCH(MI_LOAD_REGISTER_IMM
| (3 - 2));
99 OUT_BATCH(BCS_SWCTRL
);
100 OUT_BATCH((BCS_SWCTRL_DST_Y
| BCS_SWCTRL_SRC_Y
) << 16 |
101 (dst_y_tiled
? BCS_SWCTRL_DST_Y
: 0) |
102 (src_y_tiled
? BCS_SWCTRL_SRC_Y
: 0));
105 #define SET_BLITTER_TILING(...) __map = set_blitter_tiling(__VA_ARGS__, __map)
107 #define BEGIN_BATCH_BLT_TILED(n, dst_y_tiled, src_y_tiled) \
108 unsigned set_tiling_batch_size = 0; \
109 if (dst_y_tiled || src_y_tiled) { \
110 if (devinfo->gen >= 8) \
111 set_tiling_batch_size = 16; \
113 set_tiling_batch_size = 14; \
115 BEGIN_BATCH_BLT(n + set_tiling_batch_size); \
116 if (dst_y_tiled || src_y_tiled) \
117 SET_BLITTER_TILING(brw, dst_y_tiled, src_y_tiled)
119 #define ADVANCE_BATCH_TILED(dst_y_tiled, src_y_tiled) \
120 if (dst_y_tiled || src_y_tiled) \
121 SET_BLITTER_TILING(brw, false, false); \
125 intel_miptree_blit_compatible_formats(mesa_format src
, mesa_format dst
)
127 /* The BLT doesn't handle sRGB conversion */
128 assert(src
== _mesa_get_srgb_format_linear(src
));
129 assert(dst
== _mesa_get_srgb_format_linear(dst
));
131 /* No swizzle or format conversions possible, except... */
135 /* ...we can either discard the alpha channel when going from A->X,
136 * or we can fill the alpha channel with 0xff when going from X->A
138 if (src
== MESA_FORMAT_B8G8R8A8_UNORM
|| src
== MESA_FORMAT_B8G8R8X8_UNORM
)
139 return (dst
== MESA_FORMAT_B8G8R8A8_UNORM
||
140 dst
== MESA_FORMAT_B8G8R8X8_UNORM
);
142 if (src
== MESA_FORMAT_R8G8B8A8_UNORM
|| src
== MESA_FORMAT_R8G8B8X8_UNORM
)
143 return (dst
== MESA_FORMAT_R8G8B8A8_UNORM
||
144 dst
== MESA_FORMAT_R8G8B8X8_UNORM
);
146 /* We can also discard alpha when going from A2->X2 for 2 bit alpha,
147 * however we can't fill the alpha channel with two 1 bits when going
148 * from X2->A2, because intel_miptree_set_alpha_to_one() is not yet
149 * ready for this / can only handle 8 bit alpha.
151 if (src
== MESA_FORMAT_B10G10R10A2_UNORM
)
152 return (dst
== MESA_FORMAT_B10G10R10A2_UNORM
||
153 dst
== MESA_FORMAT_B10G10R10X2_UNORM
);
155 if (src
== MESA_FORMAT_R10G10B10A2_UNORM
)
156 return (dst
== MESA_FORMAT_R10G10B10A2_UNORM
||
157 dst
== MESA_FORMAT_R10G10B10X2_UNORM
);
163 get_blit_intratile_offset_el(const struct brw_context
*brw
,
164 struct intel_mipmap_tree
*mt
,
165 uint32_t total_x_offset_el
,
166 uint32_t total_y_offset_el
,
167 uint32_t *base_address_offset
,
168 uint32_t *x_offset_el
,
169 uint32_t *y_offset_el
)
171 isl_tiling_get_intratile_offset_el(mt
->surf
.tiling
,
172 mt
->cpp
* 8, mt
->surf
.row_pitch_B
,
173 total_x_offset_el
, total_y_offset_el
,
175 x_offset_el
, y_offset_el
);
176 if (mt
->surf
.tiling
== ISL_TILING_LINEAR
) {
177 /* From the Broadwell PRM docs for XY_SRC_COPY_BLT::SourceBaseAddress:
179 * "Base address of the destination surface: X=0, Y=0. Lower 32bits
180 * of the 48bit addressing. When Src Tiling is enabled (Bit_15
181 * enabled), this address must be 4KB-aligned. When Tiling is not
182 * enabled, this address should be CL (64byte) aligned."
184 * The offsets we get from ISL in the tiled case are already aligned.
185 * In the linear case, we need to do some of our own aligning.
187 uint32_t delta
= *base_address_offset
& 63;
188 assert(delta
% mt
->cpp
== 0);
189 *base_address_offset
-= delta
;
190 *x_offset_el
+= delta
/ mt
->cpp
;
192 assert(*base_address_offset
% 4096 == 0);
197 alignment_valid(struct brw_context
*brw
, unsigned offset
,
198 enum isl_tiling tiling
)
200 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
202 /* Tiled buffers must be page-aligned (4K). */
203 if (tiling
!= ISL_TILING_LINEAR
)
204 return (offset
& 4095) == 0;
206 /* On Gen8+, linear buffers must be cacheline-aligned. */
207 if (devinfo
->gen
>= 8)
208 return (offset
& 63) == 0;
214 xy_blit_cmd(enum isl_tiling src_tiling
, enum isl_tiling dst_tiling
,
223 CMD
= XY_SRC_COPY_BLT_CMD
;
226 CMD
= XY_SRC_COPY_BLT_CMD
| XY_BLT_WRITE_ALPHA
| XY_BLT_WRITE_RGB
;
229 unreachable("not reached");
232 if (dst_tiling
!= ISL_TILING_LINEAR
)
235 if (src_tiling
!= ISL_TILING_LINEAR
)
244 emit_copy_blit(struct brw_context
*brw
,
247 struct brw_bo
*src_buffer
,
249 enum isl_tiling src_tiling
,
251 struct brw_bo
*dst_buffer
,
253 enum isl_tiling dst_tiling
,
254 GLshort src_x
, GLshort src_y
,
255 GLshort dst_x
, GLshort dst_y
,
256 GLshort w
, GLshort h
,
257 enum gl_logicop_mode logic_op
)
259 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
261 int dst_y2
= dst_y
+ h
;
262 int dst_x2
= dst_x
+ w
;
263 bool dst_y_tiled
= dst_tiling
== ISL_TILING_Y0
;
264 bool src_y_tiled
= src_tiling
== ISL_TILING_Y0
;
265 uint32_t src_tile_w
, src_tile_h
;
266 uint32_t dst_tile_w
, dst_tile_h
;
268 if ((dst_y_tiled
|| src_y_tiled
) && devinfo
->gen
< 6)
271 const unsigned bo_sizes
= dst_buffer
->size
+ src_buffer
->size
;
273 /* do space check before going any further */
274 if (!brw_batch_has_aperture_space(brw
, bo_sizes
))
275 intel_batchbuffer_flush(brw
);
277 if (!brw_batch_has_aperture_space(brw
, bo_sizes
))
280 unsigned length
= devinfo
->gen
>= 8 ? 10 : 8;
282 intel_batchbuffer_require_space(brw
, length
* 4);
283 DBG("%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
285 src_buffer
, src_pitch
, src_offset
, src_x
, src_y
,
286 dst_buffer
, dst_pitch
, dst_offset
, dst_x
, dst_y
, w
, h
);
288 intel_get_tile_dims(src_tiling
, cpp
, &src_tile_w
, &src_tile_h
);
289 intel_get_tile_dims(dst_tiling
, cpp
, &dst_tile_w
, &dst_tile_h
);
291 /* For Tiled surfaces, the pitch has to be a multiple of the Tile width
292 * (X direction width of the Tile). This is ensured while allocating the
295 assert(src_tiling
== ISL_TILING_LINEAR
|| (src_pitch
% src_tile_w
) == 0);
296 assert(dst_tiling
== ISL_TILING_LINEAR
|| (dst_pitch
% dst_tile_w
) == 0);
298 /* For big formats (such as floating point), do the copy using 16 or
299 * 32bpp and multiply the coordinates.
308 assert(cpp
% 4 == 0);
316 if (!alignment_valid(brw
, dst_offset
, dst_tiling
))
318 if (!alignment_valid(brw
, src_offset
, src_tiling
))
321 /* Blit pitch must be dword-aligned. Otherwise, the hardware appears to drop
322 * the low bits. Offsets must be naturally aligned.
324 if (src_pitch
% 4 != 0 || src_offset
% cpp
!= 0 ||
325 dst_pitch
% 4 != 0 || dst_offset
% cpp
!= 0)
329 BR13
= br13_for_cpp(cpp
) | translate_raster_op(logic_op
) << 16;
331 CMD
= xy_blit_cmd(src_tiling
, dst_tiling
, cpp
);
333 /* For tiled source and destination, pitch value should be specified
334 * as a number of Dwords.
336 if (dst_tiling
!= ISL_TILING_LINEAR
)
339 if (src_tiling
!= ISL_TILING_LINEAR
)
342 if (dst_y2
<= dst_y
|| dst_x2
<= dst_x
)
345 assert(dst_x
< dst_x2
);
346 assert(dst_y
< dst_y2
);
348 BEGIN_BATCH_BLT_TILED(length
, dst_y_tiled
, src_y_tiled
);
349 OUT_BATCH(CMD
| (length
- 2));
350 OUT_BATCH(BR13
| (uint16_t)dst_pitch
);
351 OUT_BATCH(SET_FIELD(dst_y
, BLT_Y
) | SET_FIELD(dst_x
, BLT_X
));
352 OUT_BATCH(SET_FIELD(dst_y2
, BLT_Y
) | SET_FIELD(dst_x2
, BLT_X
));
353 if (devinfo
->gen
>= 8) {
354 OUT_RELOC64(dst_buffer
, RELOC_WRITE
, dst_offset
);
356 OUT_RELOC(dst_buffer
, RELOC_WRITE
, dst_offset
);
358 OUT_BATCH(SET_FIELD(src_y
, BLT_Y
) | SET_FIELD(src_x
, BLT_X
));
359 OUT_BATCH((uint16_t)src_pitch
);
360 if (devinfo
->gen
>= 8) {
361 OUT_RELOC64(src_buffer
, 0, src_offset
);
363 OUT_RELOC(src_buffer
, 0, src_offset
);
366 ADVANCE_BATCH_TILED(dst_y_tiled
, src_y_tiled
);
368 brw_emit_mi_flush(brw
);
374 emit_miptree_blit(struct brw_context
*brw
,
375 struct intel_mipmap_tree
*src_mt
,
376 uint32_t src_x
, uint32_t src_y
,
377 struct intel_mipmap_tree
*dst_mt
,
378 uint32_t dst_x
, uint32_t dst_y
,
379 uint32_t width
, uint32_t height
,
380 bool reverse
, enum gl_logicop_mode logicop
)
382 /* According to the Ivy Bridge PRM, Vol1 Part4, section 1.2.1.2 (Graphics
383 * Data Size Limitations):
385 * The BLT engine is capable of transferring very large quantities of
386 * graphics data. Any graphics data read from and written to the
387 * destination is permitted to represent a number of pixels that
388 * occupies up to 65,536 scan lines and up to 32,768 bytes per scan line
389 * at the destination. The maximum number of pixels that may be
390 * represented per scan line’s worth of graphics data depends on the
393 * The blitter's pitch is a signed 16-bit integer, but measured in bytes
394 * for linear surfaces and DWords for tiled surfaces. So the maximum
395 * pitch is 32k linear and 128k tiled.
397 if (intel_miptree_blt_pitch(src_mt
) >= 32768 ||
398 intel_miptree_blt_pitch(dst_mt
) >= 32768) {
399 perf_debug("Falling back due to >= 32k/128k pitch\n");
403 /* We need to split the blit into chunks that each fit within the blitter's
404 * restrictions. We can't use a chunk size of 32768 because we need to
405 * ensure that src_tile_x + chunk_size fits. We choose 16384 because it's
406 * a nice round power of two, big enough that performance won't suffer, and
407 * small enough to guarantee everything fits.
409 const uint32_t max_chunk_size
= 16384;
411 for (uint32_t chunk_x
= 0; chunk_x
< width
; chunk_x
+= max_chunk_size
) {
412 for (uint32_t chunk_y
= 0; chunk_y
< height
; chunk_y
+= max_chunk_size
) {
413 const uint32_t chunk_w
= MIN2(max_chunk_size
, width
- chunk_x
);
414 const uint32_t chunk_h
= MIN2(max_chunk_size
, height
- chunk_y
);
416 uint32_t src_offset
, src_tile_x
, src_tile_y
;
417 get_blit_intratile_offset_el(brw
, src_mt
,
418 src_x
+ chunk_x
, src_y
+ chunk_y
,
419 &src_offset
, &src_tile_x
, &src_tile_y
);
421 uint32_t dst_offset
, dst_tile_x
, dst_tile_y
;
422 get_blit_intratile_offset_el(brw
, dst_mt
,
423 dst_x
+ chunk_x
, dst_y
+ chunk_y
,
424 &dst_offset
, &dst_tile_x
, &dst_tile_y
);
426 if (!emit_copy_blit(brw
,
428 reverse
? -src_mt
->surf
.row_pitch_B
:
429 src_mt
->surf
.row_pitch_B
,
430 src_mt
->bo
, src_mt
->offset
+ src_offset
,
432 dst_mt
->surf
.row_pitch_B
,
433 dst_mt
->bo
, dst_mt
->offset
+ dst_offset
,
435 src_tile_x
, src_tile_y
,
436 dst_tile_x
, dst_tile_y
,
439 /* If this is ever going to fail, it will fail on the first chunk */
440 assert(chunk_x
== 0 && chunk_y
== 0);
450 * Implements a rectangular block transfer (blit) of pixels between two
453 * Our blitter can operate on 1, 2, or 4-byte-per-pixel data, with generous,
454 * but limited, pitches and sizes allowed.
456 * The src/dst coordinates are relative to the given level/slice of the
459 * If @src_flip or @dst_flip is set, then the rectangle within that miptree
460 * will be inverted (including scanline order) when copying. This is common
461 * in GL when copying between window system and user-created
462 * renderbuffers/textures.
465 intel_miptree_blit(struct brw_context
*brw
,
466 struct intel_mipmap_tree
*src_mt
,
467 int src_level
, int src_slice
,
468 uint32_t src_x
, uint32_t src_y
, bool src_flip
,
469 struct intel_mipmap_tree
*dst_mt
,
470 int dst_level
, int dst_slice
,
471 uint32_t dst_x
, uint32_t dst_y
, bool dst_flip
,
472 uint32_t width
, uint32_t height
,
473 enum gl_logicop_mode logicop
)
475 /* The blitter doesn't understand multisampling at all. */
476 if (src_mt
->surf
.samples
> 1 || dst_mt
->surf
.samples
> 1)
479 /* No sRGB decode or encode is done by the hardware blitter, which is
480 * consistent with what we want in many callers (glCopyTexSubImage(),
481 * texture validation, etc.).
483 mesa_format src_format
= _mesa_get_srgb_format_linear(src_mt
->format
);
484 mesa_format dst_format
= _mesa_get_srgb_format_linear(dst_mt
->format
);
486 /* The blitter doesn't support doing any format conversions. We do also
487 * support blitting ARGB8888 to XRGB8888 (trivial, the values dropped into
488 * the X channel don't matter), and XRGB8888 to ARGB8888 by setting the A
489 * channel to 1.0 at the end. Also trivially ARGB2101010 to XRGB2101010,
490 * but not XRGB2101010 to ARGB2101010 yet.
492 if (!intel_miptree_blit_compatible_formats(src_format
, dst_format
)) {
493 perf_debug("%s: Can't use hardware blitter from %s to %s, "
494 "falling back.\n", __func__
,
495 _mesa_get_format_name(src_format
),
496 _mesa_get_format_name(dst_format
));
500 /* The blitter has no idea about HiZ or fast color clears, so we need to
501 * resolve the miptrees before we do anything.
503 intel_miptree_access_raw(brw
, src_mt
, src_level
, src_slice
, false);
504 intel_miptree_access_raw(brw
, dst_mt
, dst_level
, dst_slice
, true);
507 const unsigned h0
= src_mt
->surf
.phys_level0_sa
.height
;
508 src_y
= minify(h0
, src_level
- src_mt
->first_level
) - src_y
- height
;
512 const unsigned h0
= dst_mt
->surf
.phys_level0_sa
.height
;
513 dst_y
= minify(h0
, dst_level
- dst_mt
->first_level
) - dst_y
- height
;
516 uint32_t src_image_x
, src_image_y
, dst_image_x
, dst_image_y
;
517 intel_miptree_get_image_offset(src_mt
, src_level
, src_slice
,
518 &src_image_x
, &src_image_y
);
519 intel_miptree_get_image_offset(dst_mt
, dst_level
, dst_slice
,
520 &dst_image_x
, &dst_image_y
);
521 src_x
+= src_image_x
;
522 src_y
+= src_image_y
;
523 dst_x
+= dst_image_x
;
524 dst_y
+= dst_image_y
;
526 if (!emit_miptree_blit(brw
, src_mt
, src_x
, src_y
,
527 dst_mt
, dst_x
, dst_y
, width
, height
,
528 src_flip
!= dst_flip
, logicop
)) {
532 /* XXX This could be done in a single pass using XY_FULL_MONO_PATTERN_BLT */
533 if (_mesa_get_format_bits(src_format
, GL_ALPHA_BITS
) == 0 &&
534 _mesa_get_format_bits(dst_format
, GL_ALPHA_BITS
) > 0) {
535 intel_miptree_set_alpha_to_one(brw
, dst_mt
,
544 intel_miptree_copy(struct brw_context
*brw
,
545 struct intel_mipmap_tree
*src_mt
,
546 int src_level
, int src_slice
,
547 uint32_t src_x
, uint32_t src_y
,
548 struct intel_mipmap_tree
*dst_mt
,
549 int dst_level
, int dst_slice
,
550 uint32_t dst_x
, uint32_t dst_y
,
551 uint32_t src_width
, uint32_t src_height
)
553 /* The blitter doesn't understand multisampling at all. */
554 if (src_mt
->surf
.samples
> 1 || dst_mt
->surf
.samples
> 1)
557 if (src_mt
->format
== MESA_FORMAT_S_UINT8
)
560 /* The blitter has no idea about HiZ or fast color clears, so we need to
561 * resolve the miptrees before we do anything.
563 intel_miptree_access_raw(brw
, src_mt
, src_level
, src_slice
, false);
564 intel_miptree_access_raw(brw
, dst_mt
, dst_level
, dst_slice
, true);
566 uint32_t src_image_x
, src_image_y
;
567 intel_miptree_get_image_offset(src_mt
, src_level
, src_slice
,
568 &src_image_x
, &src_image_y
);
570 if (_mesa_is_format_compressed(src_mt
->format
)) {
572 _mesa_get_format_block_size(src_mt
->format
, &bw
, &bh
);
574 /* Compressed textures need not have dimensions that are a multiple of
575 * the block size. Rectangles in compressed textures do need to be a
576 * multiple of the block size. The one exception is that the right and
577 * bottom edges may be at the right or bottom edge of the miplevel even
578 * if it's not aligned.
580 assert(src_x
% bw
== 0);
581 assert(src_y
% bh
== 0);
583 assert(src_width
% bw
== 0 ||
585 minify(src_mt
->surf
.logical_level0_px
.width
, src_level
));
586 assert(src_height
% bh
== 0 ||
587 src_y
+ src_height
==
588 minify(src_mt
->surf
.logical_level0_px
.height
, src_level
));
592 src_width
= DIV_ROUND_UP(src_width
, (int)bw
);
593 src_height
= DIV_ROUND_UP(src_height
, (int)bh
);
595 src_x
+= src_image_x
;
596 src_y
+= src_image_y
;
598 uint32_t dst_image_x
, dst_image_y
;
599 intel_miptree_get_image_offset(dst_mt
, dst_level
, dst_slice
,
600 &dst_image_x
, &dst_image_y
);
602 if (_mesa_is_format_compressed(dst_mt
->format
)) {
604 _mesa_get_format_block_size(dst_mt
->format
, &bw
, &bh
);
606 assert(dst_x
% bw
== 0);
607 assert(dst_y
% bh
== 0);
612 dst_x
+= dst_image_x
;
613 dst_y
+= dst_image_y
;
615 return emit_miptree_blit(brw
, src_mt
, src_x
, src_y
,
616 dst_mt
, dst_x
, dst_y
,
617 src_width
, src_height
, false, COLOR_LOGICOP_COPY
);
621 intelEmitImmediateColorExpandBlit(struct brw_context
*brw
,
623 GLubyte
*src_bits
, GLuint src_size
,
626 struct brw_bo
*dst_buffer
,
628 enum isl_tiling dst_tiling
,
629 GLshort x
, GLshort y
,
630 GLshort w
, GLshort h
,
631 enum gl_logicop_mode logic_op
)
633 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
634 int dwords
= ALIGN(src_size
, 8) / 4;
635 uint32_t opcode
, br13
, blit_cmd
;
637 if (dst_tiling
!= ISL_TILING_LINEAR
) {
638 if (dst_offset
& 4095)
640 if (dst_tiling
== ISL_TILING_Y0
)
644 assert((unsigned) logic_op
<= 0x0f);
645 assert(dst_pitch
> 0);
650 DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d, %d bytes %d dwords\n",
652 dst_buffer
, dst_pitch
, dst_offset
, x
, y
, w
, h
, src_size
, dwords
);
654 unsigned xy_setup_blt_length
= devinfo
->gen
>= 8 ? 10 : 8;
655 intel_batchbuffer_require_space(brw
, (xy_setup_blt_length
* 4) +
656 (3 * 4) + dwords
* 4);
658 opcode
= XY_SETUP_BLT_CMD
;
660 opcode
|= XY_BLT_WRITE_ALPHA
| XY_BLT_WRITE_RGB
;
661 if (dst_tiling
!= ISL_TILING_LINEAR
) {
662 opcode
|= XY_DST_TILED
;
666 br13
= dst_pitch
| (translate_raster_op(logic_op
) << 16) | (1 << 29);
667 br13
|= br13_for_cpp(cpp
);
669 blit_cmd
= XY_TEXT_IMMEDIATE_BLIT_CMD
| XY_TEXT_BYTE_PACKED
; /* packing? */
670 if (dst_tiling
!= ISL_TILING_LINEAR
)
671 blit_cmd
|= XY_DST_TILED
;
673 BEGIN_BATCH_BLT(xy_setup_blt_length
+ 3);
674 OUT_BATCH(opcode
| (xy_setup_blt_length
- 2));
676 OUT_BATCH((0 << 16) | 0); /* clip x1, y1 */
677 OUT_BATCH((100 << 16) | 100); /* clip x2, y2 */
678 if (devinfo
->gen
>= 8) {
679 OUT_RELOC64(dst_buffer
, RELOC_WRITE
, dst_offset
);
681 OUT_RELOC(dst_buffer
, RELOC_WRITE
, dst_offset
);
683 OUT_BATCH(0); /* bg */
684 OUT_BATCH(fg_color
); /* fg */
685 OUT_BATCH(0); /* pattern base addr */
686 if (devinfo
->gen
>= 8)
689 OUT_BATCH(blit_cmd
| ((3 - 2) + dwords
));
690 OUT_BATCH(SET_FIELD(y
, BLT_Y
) | SET_FIELD(x
, BLT_X
));
691 OUT_BATCH(SET_FIELD(y
+ h
, BLT_Y
) | SET_FIELD(x
+ w
, BLT_X
));
694 intel_batchbuffer_data(brw
, src_bits
, dwords
* 4);
696 brw_emit_mi_flush(brw
);
702 * Used to initialize the alpha value of an ARGB8888 miptree after copying
703 * into it from an XRGB8888 source.
705 * This is very common with glCopyTexImage2D(). Note that the coordinates are
706 * relative to the start of the miptree, not relative to a slice within the
710 intel_miptree_set_alpha_to_one(struct brw_context
*brw
,
711 struct intel_mipmap_tree
*mt
,
712 int x
, int y
, int width
, int height
)
714 const struct gen_device_info
*devinfo
= &brw
->screen
->devinfo
;
718 pitch
= mt
->surf
.row_pitch_B
;
721 DBG("%s dst:buf(%p)/%d %d,%d sz:%dx%d\n",
722 __func__
, mt
->bo
, pitch
, x
, y
, width
, height
);
724 /* Note: Currently only handles 8 bit alpha channel. Extension to < 8 Bit
725 * alpha channel would be likely possible via ROP code 0xfa instead of 0xf0
726 * and writing a suitable bit-mask instead of 0xffffffff.
728 BR13
= br13_for_cpp(cpp
) | 0xf0 << 16;
729 CMD
= XY_COLOR_BLT_CMD
;
730 CMD
|= XY_BLT_WRITE_ALPHA
;
732 if (mt
->surf
.tiling
!= ISL_TILING_LINEAR
) {
738 /* do space check before going any further */
739 if (!brw_batch_has_aperture_space(brw
, mt
->bo
->size
))
740 intel_batchbuffer_flush(brw
);
742 unsigned length
= devinfo
->gen
>= 8 ? 7 : 6;
743 const bool dst_y_tiled
= mt
->surf
.tiling
== ISL_TILING_Y0
;
745 /* We need to split the blit into chunks that each fit within the blitter's
746 * restrictions. We can't use a chunk size of 32768 because we need to
747 * ensure that src_tile_x + chunk_size fits. We choose 16384 because it's
748 * a nice round power of two, big enough that performance won't suffer, and
749 * small enough to guarantee everything fits.
751 const uint32_t max_chunk_size
= 16384;
753 for (uint32_t chunk_x
= 0; chunk_x
< width
; chunk_x
+= max_chunk_size
) {
754 for (uint32_t chunk_y
= 0; chunk_y
< height
; chunk_y
+= max_chunk_size
) {
755 const uint32_t chunk_w
= MIN2(max_chunk_size
, width
- chunk_x
);
756 const uint32_t chunk_h
= MIN2(max_chunk_size
, height
- chunk_y
);
758 uint32_t offset
, tile_x
, tile_y
;
759 get_blit_intratile_offset_el(brw
, mt
,
760 x
+ chunk_x
, y
+ chunk_y
,
761 &offset
, &tile_x
, &tile_y
);
763 BEGIN_BATCH_BLT_TILED(length
, dst_y_tiled
, false);
764 OUT_BATCH(CMD
| (length
- 2));
766 OUT_BATCH(SET_FIELD(y
+ chunk_y
, BLT_Y
) |
767 SET_FIELD(x
+ chunk_x
, BLT_X
));
768 OUT_BATCH(SET_FIELD(y
+ chunk_y
+ chunk_h
, BLT_Y
) |
769 SET_FIELD(x
+ chunk_x
+ chunk_w
, BLT_X
));
770 if (devinfo
->gen
>= 8) {
771 OUT_RELOC64(mt
->bo
, RELOC_WRITE
, mt
->offset
+ offset
);
773 OUT_RELOC(mt
->bo
, RELOC_WRITE
, mt
->offset
+ offset
);
775 OUT_BATCH(0xffffffff); /* white, but only alpha gets written */
776 ADVANCE_BATCH_TILED(dst_y_tiled
, false);
780 brw_emit_mi_flush(brw
);