1 /**************************************************************************
3 * Copyright 2003 VMware, Inc.
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the
8 * "Software"), to deal in the Software without restriction, including
9 * without limitation the rights to use, copy, modify, merge, publish,
10 * distribute, sub license, and/or sell copies of the Software, and to
11 * permit persons to whom the Software is furnished to do so, subject to
12 * the following conditions:
14 * The above copyright notice and this permission notice (including the
15 * next paragraph) shall be included in all copies or substantial portions
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21 * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
22 * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23 * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24 * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 **************************************************************************/
29 #include "main/mtypes.h"
30 #include "main/context.h"
31 #include "main/enums.h"
32 #include "main/colormac.h"
33 #include "main/fbobject.h"
35 #include "brw_context.h"
36 #include "brw_defines.h"
37 #include "intel_blit.h"
38 #include "intel_buffers.h"
39 #include "intel_fbo.h"
40 #include "intel_reg.h"
41 #include "intel_batchbuffer.h"
42 #include "intel_mipmap_tree.h"
44 #define FILE_DEBUG_FLAG DEBUG_BLIT
47 intel_miptree_set_alpha_to_one(struct brw_context
*brw
,
48 struct intel_mipmap_tree
*mt
,
49 int x
, int y
, int width
, int height
);
51 static GLuint
translate_raster_op(GLenum logicop
)
54 case GL_CLEAR
: return 0x00;
55 case GL_AND
: return 0x88;
56 case GL_AND_REVERSE
: return 0x44;
57 case GL_COPY
: return 0xCC;
58 case GL_AND_INVERTED
: return 0x22;
59 case GL_NOOP
: return 0xAA;
60 case GL_XOR
: return 0x66;
61 case GL_OR
: return 0xEE;
62 case GL_NOR
: return 0x11;
63 case GL_EQUIV
: return 0x99;
64 case GL_INVERT
: return 0x55;
65 case GL_OR_REVERSE
: return 0xDD;
66 case GL_COPY_INVERTED
: return 0x33;
67 case GL_OR_INVERTED
: return 0xBB;
68 case GL_NAND
: return 0x77;
69 case GL_SET
: return 0xFF;
85 unreachable("not reached");
90 * Emits the packet for switching the blitter from X to Y tiled or back.
92 * This has to be called in a single BEGIN_BATCH_BLT_TILED() /
93 * ADVANCE_BATCH_TILED(). This is because BCS_SWCTRL is saved and restored as
94 * part of the power context, not a render context, and if the batchbuffer was
95 * to get flushed between setting and blitting, or blitting and restoring, our
96 * tiling state would leak into other unsuspecting applications (like the X
100 set_blitter_tiling(struct brw_context
*brw
,
101 bool dst_y_tiled
, bool src_y_tiled
)
103 assert(brw
->gen
>= 6);
105 /* Idle the blitter before we update how tiling is interpreted. */
106 OUT_BATCH(MI_FLUSH_DW
);
111 OUT_BATCH(MI_LOAD_REGISTER_IMM
| (3 - 2));
112 OUT_BATCH(BCS_SWCTRL
);
113 OUT_BATCH((BCS_SWCTRL_DST_Y
| BCS_SWCTRL_SRC_Y
) << 16 |
114 (dst_y_tiled
? BCS_SWCTRL_DST_Y
: 0) |
115 (src_y_tiled
? BCS_SWCTRL_SRC_Y
: 0));
118 #define BEGIN_BATCH_BLT_TILED(n, dst_y_tiled, src_y_tiled) do { \
119 BEGIN_BATCH_BLT(n + ((dst_y_tiled || src_y_tiled) ? 14 : 0)); \
120 if (dst_y_tiled || src_y_tiled) \
121 set_blitter_tiling(brw, dst_y_tiled, src_y_tiled); \
124 #define ADVANCE_BATCH_TILED(dst_y_tiled, src_y_tiled) do { \
125 if (dst_y_tiled || src_y_tiled) \
126 set_blitter_tiling(brw, false, false); \
131 blt_pitch(struct intel_mipmap_tree
*mt
)
133 int pitch
= mt
->pitch
;
140 intel_miptree_blit_compatible_formats(mesa_format src
, mesa_format dst
)
142 /* The BLT doesn't handle sRGB conversion */
143 assert(src
== _mesa_get_srgb_format_linear(src
));
144 assert(dst
== _mesa_get_srgb_format_linear(dst
));
146 /* No swizzle or format conversions possible, except... */
150 /* ...we can either discard the alpha channel when going from A->X,
151 * or we can fill the alpha channel with 0xff when going from X->A
153 if (src
== MESA_FORMAT_B8G8R8A8_UNORM
|| src
== MESA_FORMAT_B8G8R8X8_UNORM
)
154 return (dst
== MESA_FORMAT_B8G8R8A8_UNORM
||
155 dst
== MESA_FORMAT_B8G8R8X8_UNORM
);
157 if (src
== MESA_FORMAT_R8G8B8A8_UNORM
|| src
== MESA_FORMAT_R8G8B8X8_UNORM
)
158 return (dst
== MESA_FORMAT_R8G8B8A8_UNORM
||
159 dst
== MESA_FORMAT_R8G8B8X8_UNORM
);
165 * Implements a rectangular block transfer (blit) of pixels between two
168 * Our blitter can operate on 1, 2, or 4-byte-per-pixel data, with generous,
169 * but limited, pitches and sizes allowed.
171 * The src/dst coordinates are relative to the given level/slice of the
174 * If @src_flip or @dst_flip is set, then the rectangle within that miptree
175 * will be inverted (including scanline order) when copying. This is common
176 * in GL when copying between window system and user-created
177 * renderbuffers/textures.
180 intel_miptree_blit(struct brw_context
*brw
,
181 struct intel_mipmap_tree
*src_mt
,
182 int src_level
, int src_slice
,
183 uint32_t src_x
, uint32_t src_y
, bool src_flip
,
184 struct intel_mipmap_tree
*dst_mt
,
185 int dst_level
, int dst_slice
,
186 uint32_t dst_x
, uint32_t dst_y
, bool dst_flip
,
187 uint32_t width
, uint32_t height
,
190 /* The blitter doesn't understand multisampling at all. */
191 if (src_mt
->num_samples
> 0 || dst_mt
->num_samples
> 0)
194 /* No sRGB decode or encode is done by the hardware blitter, which is
195 * consistent with what we want in the callers (glCopyTexSubImage(),
196 * glBlitFramebuffer(), texture validation, etc.).
198 mesa_format src_format
= _mesa_get_srgb_format_linear(src_mt
->format
);
199 mesa_format dst_format
= _mesa_get_srgb_format_linear(dst_mt
->format
);
201 /* The blitter doesn't support doing any format conversions. We do also
202 * support blitting ARGB8888 to XRGB8888 (trivial, the values dropped into
203 * the X channel don't matter), and XRGB8888 to ARGB8888 by setting the A
204 * channel to 1.0 at the end.
206 if (!intel_miptree_blit_compatible_formats(src_format
, dst_format
)) {
207 perf_debug("%s: Can't use hardware blitter from %s to %s, "
208 "falling back.\n", __func__
,
209 _mesa_get_format_name(src_format
),
210 _mesa_get_format_name(dst_format
));
214 /* According to the Ivy Bridge PRM, Vol1 Part4, section 1.2.1.2 (Graphics
215 * Data Size Limitations):
217 * The BLT engine is capable of transferring very large quantities of
218 * graphics data. Any graphics data read from and written to the
219 * destination is permitted to represent a number of pixels that
220 * occupies up to 65,536 scan lines and up to 32,768 bytes per scan line
221 * at the destination. The maximum number of pixels that may be
222 * represented per scan line’s worth of graphics data depends on the
225 * Furthermore, intelEmitCopyBlit (which is called below) uses a signed
226 * 16-bit integer to represent buffer pitch, so it can only handle buffer
227 * pitches < 32k. However, the pitch is measured in bytes for linear buffers
228 * and dwords for tiled buffers.
230 * As a result of these two limitations, we can only use the blitter to do
231 * this copy when the miptree's pitch is less than 32k linear or 128k tiled.
233 if (blt_pitch(src_mt
) >= 32768 || blt_pitch(dst_mt
) >= 32768) {
234 perf_debug("Falling back due to >= 32k/128k pitch\n");
238 /* The blitter has no idea about HiZ or fast color clears, so we need to
239 * resolve the miptrees before we do anything.
241 intel_miptree_slice_resolve_depth(brw
, src_mt
, src_level
, src_slice
);
242 intel_miptree_slice_resolve_depth(brw
, dst_mt
, dst_level
, dst_slice
);
243 intel_miptree_resolve_color(brw
, src_mt
);
244 intel_miptree_resolve_color(brw
, dst_mt
);
247 src_y
= minify(src_mt
->physical_height0
, src_level
- src_mt
->first_level
) - src_y
- height
;
250 dst_y
= minify(dst_mt
->physical_height0
, dst_level
- dst_mt
->first_level
) - dst_y
- height
;
252 int src_pitch
= src_mt
->pitch
;
253 if (src_flip
!= dst_flip
)
254 src_pitch
= -src_pitch
;
256 uint32_t src_image_x
, src_image_y
, dst_image_x
, dst_image_y
;
257 intel_miptree_get_image_offset(src_mt
, src_level
, src_slice
,
258 &src_image_x
, &src_image_y
);
259 intel_miptree_get_image_offset(dst_mt
, dst_level
, dst_slice
,
260 &dst_image_x
, &dst_image_y
);
261 src_x
+= src_image_x
;
262 src_y
+= src_image_y
;
263 dst_x
+= dst_image_x
;
264 dst_y
+= dst_image_y
;
266 /* The blitter interprets the 16-bit destination x/y as a signed 16-bit
267 * value. The values we're working with are unsigned, so make sure we don't
270 if (src_x
>= 32768 || src_y
>= 32768 || dst_x
>= 32768 || dst_y
>= 32768) {
271 perf_debug("Falling back due to >=32k offset [src(%d, %d) dst(%d, %d)]\n",
272 src_x
, src_y
, dst_x
, dst_y
);
276 if (!intelEmitCopyBlit(brw
,
279 src_mt
->bo
, src_mt
->offset
,
282 dst_mt
->bo
, dst_mt
->offset
,
291 /* XXX This could be done in a single pass using XY_FULL_MONO_PATTERN_BLT */
292 if (_mesa_get_format_bits(src_format
, GL_ALPHA_BITS
) == 0 &&
293 _mesa_get_format_bits(dst_format
, GL_ALPHA_BITS
) > 0) {
294 intel_miptree_set_alpha_to_one(brw
, dst_mt
,
303 alignment_valid(struct brw_context
*brw
, unsigned offset
, uint32_t tiling
)
305 /* Tiled buffers must be page-aligned (4K). */
306 if (tiling
!= I915_TILING_NONE
)
307 return (offset
& 4095) == 0;
309 /* On Gen8+, linear buffers must be cacheline-aligned. */
311 return (offset
& 63) == 0;
319 intelEmitCopyBlit(struct brw_context
*brw
,
322 drm_intel_bo
*src_buffer
,
326 drm_intel_bo
*dst_buffer
,
329 GLshort src_x
, GLshort src_y
,
330 GLshort dst_x
, GLshort dst_y
,
331 GLshort w
, GLshort h
,
334 GLuint CMD
, BR13
, pass
= 0;
335 int dst_y2
= dst_y
+ h
;
336 int dst_x2
= dst_x
+ w
;
337 drm_intel_bo
*aper_array
[3];
338 bool dst_y_tiled
= dst_tiling
== I915_TILING_Y
;
339 bool src_y_tiled
= src_tiling
== I915_TILING_Y
;
341 if (!alignment_valid(brw
, dst_offset
, dst_tiling
))
343 if (!alignment_valid(brw
, src_offset
, src_tiling
))
346 if ((dst_y_tiled
|| src_y_tiled
) && brw
->gen
< 6)
349 assert(!dst_y_tiled
|| (dst_pitch
% 128) == 0);
350 assert(!src_y_tiled
|| (src_pitch
% 128) == 0);
352 /* do space check before going any further */
354 aper_array
[0] = brw
->batch
.bo
;
355 aper_array
[1] = dst_buffer
;
356 aper_array
[2] = src_buffer
;
358 if (dri_bufmgr_check_aperture_space(aper_array
, 3) != 0) {
359 intel_batchbuffer_flush(brw
);
368 unsigned length
= brw
->gen
>= 8 ? 10 : 8;
370 intel_batchbuffer_require_space(brw
, length
* 4, BLT_RING
);
371 DBG("%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
373 src_buffer
, src_pitch
, src_offset
, src_x
, src_y
,
374 dst_buffer
, dst_pitch
, dst_offset
, dst_x
, dst_y
, w
, h
);
376 /* Blit pitch must be dword-aligned. Otherwise, the hardware appears to drop
377 * the low bits. Offsets must be naturally aligned.
379 if (src_pitch
% 4 != 0 || src_offset
% cpp
!= 0 ||
380 dst_pitch
% 4 != 0 || dst_offset
% cpp
!= 0)
383 /* For big formats (such as floating point), do the copy using 16 or 32bpp
384 * and multiply the coordinates.
393 assert(cpp
% 4 == 0);
401 BR13
= br13_for_cpp(cpp
) | translate_raster_op(logic_op
) << 16;
406 CMD
= XY_SRC_COPY_BLT_CMD
;
409 CMD
= XY_SRC_COPY_BLT_CMD
| XY_BLT_WRITE_ALPHA
| XY_BLT_WRITE_RGB
;
415 if (dst_tiling
!= I915_TILING_NONE
) {
419 if (src_tiling
!= I915_TILING_NONE
) {
424 if (dst_y2
<= dst_y
|| dst_x2
<= dst_x
) {
428 assert(dst_x
< dst_x2
);
429 assert(dst_y
< dst_y2
);
430 assert(src_offset
+ (src_y
+ h
- 1) * abs(src_pitch
) +
431 (w
* cpp
) <= src_buffer
->size
);
432 assert(dst_offset
+ (dst_y
+ h
- 1) * abs(dst_pitch
) +
433 (w
* cpp
) <= dst_buffer
->size
);
435 BEGIN_BATCH_BLT_TILED(length
, dst_y_tiled
, src_y_tiled
);
436 OUT_BATCH(CMD
| (length
- 2));
437 OUT_BATCH(BR13
| (uint16_t)dst_pitch
);
438 OUT_BATCH(SET_FIELD(dst_y
, BLT_Y
) | SET_FIELD(dst_x
, BLT_X
));
439 OUT_BATCH(SET_FIELD(dst_y2
, BLT_Y
) | SET_FIELD(dst_x2
, BLT_X
));
441 OUT_RELOC64(dst_buffer
,
442 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
445 OUT_RELOC(dst_buffer
,
446 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
449 OUT_BATCH(SET_FIELD(src_y
, BLT_Y
) | SET_FIELD(src_x
, BLT_X
));
450 OUT_BATCH((uint16_t)src_pitch
);
452 OUT_RELOC64(src_buffer
,
453 I915_GEM_DOMAIN_RENDER
, 0,
456 OUT_RELOC(src_buffer
,
457 I915_GEM_DOMAIN_RENDER
, 0,
461 ADVANCE_BATCH_TILED(dst_y_tiled
, src_y_tiled
);
463 intel_batchbuffer_emit_mi_flush(brw
);
469 intelEmitImmediateColorExpandBlit(struct brw_context
*brw
,
471 GLubyte
*src_bits
, GLuint src_size
,
474 drm_intel_bo
*dst_buffer
,
477 GLshort x
, GLshort y
,
478 GLshort w
, GLshort h
,
481 int dwords
= ALIGN(src_size
, 8) / 4;
482 uint32_t opcode
, br13
, blit_cmd
;
484 if (dst_tiling
!= I915_TILING_NONE
) {
485 if (dst_offset
& 4095)
487 if (dst_tiling
== I915_TILING_Y
)
491 assert((logic_op
>= GL_CLEAR
) && (logic_op
<= (GL_CLEAR
+ 0x0f)));
492 assert(dst_pitch
> 0);
497 DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d, %d bytes %d dwords\n",
499 dst_buffer
, dst_pitch
, dst_offset
, x
, y
, w
, h
, src_size
, dwords
);
501 unsigned xy_setup_blt_length
= brw
->gen
>= 8 ? 10 : 8;
502 intel_batchbuffer_require_space(brw
, (xy_setup_blt_length
* 4) +
503 (3 * 4) + dwords
* 4, BLT_RING
);
505 opcode
= XY_SETUP_BLT_CMD
;
507 opcode
|= XY_BLT_WRITE_ALPHA
| XY_BLT_WRITE_RGB
;
508 if (dst_tiling
!= I915_TILING_NONE
) {
509 opcode
|= XY_DST_TILED
;
513 br13
= dst_pitch
| (translate_raster_op(logic_op
) << 16) | (1 << 29);
514 br13
|= br13_for_cpp(cpp
);
516 blit_cmd
= XY_TEXT_IMMEDIATE_BLIT_CMD
| XY_TEXT_BYTE_PACKED
; /* packing? */
517 if (dst_tiling
!= I915_TILING_NONE
)
518 blit_cmd
|= XY_DST_TILED
;
520 BEGIN_BATCH_BLT(xy_setup_blt_length
+ 3);
521 OUT_BATCH(opcode
| (xy_setup_blt_length
- 2));
523 OUT_BATCH((0 << 16) | 0); /* clip x1, y1 */
524 OUT_BATCH((100 << 16) | 100); /* clip x2, y2 */
526 OUT_RELOC64(dst_buffer
,
527 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
530 OUT_RELOC(dst_buffer
,
531 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
534 OUT_BATCH(0); /* bg */
535 OUT_BATCH(fg_color
); /* fg */
536 OUT_BATCH(0); /* pattern base addr */
540 OUT_BATCH(blit_cmd
| ((3 - 2) + dwords
));
541 OUT_BATCH(SET_FIELD(y
, BLT_Y
) | SET_FIELD(x
, BLT_X
));
542 OUT_BATCH(SET_FIELD(y
+ h
, BLT_Y
) | SET_FIELD(x
+ w
, BLT_X
));
545 intel_batchbuffer_data(brw
, src_bits
, dwords
* 4, BLT_RING
);
547 intel_batchbuffer_emit_mi_flush(brw
);
552 /* We don't have a memmove-type blit like some other hardware, so we'll do a
553 * rectangular blit covering a large space, then emit 1-scanline blit at the
554 * end to cover the last if we need.
557 intel_emit_linear_blit(struct brw_context
*brw
,
558 drm_intel_bo
*dst_bo
,
559 unsigned int dst_offset
,
560 drm_intel_bo
*src_bo
,
561 unsigned int src_offset
,
564 struct gl_context
*ctx
= &brw
->ctx
;
565 GLuint pitch
, height
;
566 int16_t src_x
, dst_x
;
569 /* The pitch given to the GPU must be DWORD aligned, and
570 * we want width to match pitch. Max width is (1 << 15 - 1),
571 * rounding that down to the nearest DWORD is 1 << 15 - 4
573 pitch
= ROUND_DOWN_TO(MIN2(size
, (1 << 15) - 1), 4);
574 height
= (pitch
== 0) ? 1 : size
/ pitch
;
575 src_x
= src_offset
% 64;
576 dst_x
= dst_offset
% 64;
577 ok
= intelEmitCopyBlit(brw
, 1,
578 pitch
, src_bo
, src_offset
- src_x
, I915_TILING_NONE
,
579 pitch
, dst_bo
, dst_offset
- dst_x
, I915_TILING_NONE
,
580 src_x
, 0, /* src x/y */
581 dst_x
, 0, /* dst x/y */
582 pitch
, height
, /* w, h */
585 _mesa_problem(ctx
, "Failed to linear blit %dx%d\n", pitch
, height
);
587 src_offset
+= pitch
* height
;
588 dst_offset
+= pitch
* height
;
589 src_x
= src_offset
% 64;
590 dst_x
= dst_offset
% 64;
591 size
-= pitch
* height
;
592 assert (size
< (1 << 15));
593 pitch
= ALIGN(size
, 4);
596 ok
= intelEmitCopyBlit(brw
, 1,
597 pitch
, src_bo
, src_offset
- src_x
, I915_TILING_NONE
,
598 pitch
, dst_bo
, dst_offset
- dst_x
, I915_TILING_NONE
,
599 src_x
, 0, /* src x/y */
600 dst_x
, 0, /* dst x/y */
604 _mesa_problem(ctx
, "Failed to linear blit %dx%d\n", size
, 1);
609 * Used to initialize the alpha value of an ARGB8888 miptree after copying
610 * into it from an XRGB8888 source.
612 * This is very common with glCopyTexImage2D(). Note that the coordinates are
613 * relative to the start of the miptree, not relative to a slice within the
617 intel_miptree_set_alpha_to_one(struct brw_context
*brw
,
618 struct intel_mipmap_tree
*mt
,
619 int x
, int y
, int width
, int height
)
623 drm_intel_bo
*aper_array
[2];
628 DBG("%s dst:buf(%p)/%d %d,%d sz:%dx%d\n",
629 __func__
, mt
->bo
, pitch
, x
, y
, width
, height
);
631 BR13
= br13_for_cpp(cpp
) | 0xf0 << 16;
632 CMD
= XY_COLOR_BLT_CMD
;
633 CMD
|= XY_BLT_WRITE_ALPHA
;
635 if (mt
->tiling
!= I915_TILING_NONE
) {
641 /* do space check before going any further */
642 aper_array
[0] = brw
->batch
.bo
;
643 aper_array
[1] = mt
->bo
;
645 if (drm_intel_bufmgr_check_aperture_space(aper_array
,
646 ARRAY_SIZE(aper_array
)) != 0) {
647 intel_batchbuffer_flush(brw
);
650 unsigned length
= brw
->gen
>= 8 ? 7 : 6;
651 bool dst_y_tiled
= mt
->tiling
== I915_TILING_Y
;
653 BEGIN_BATCH_BLT_TILED(length
, dst_y_tiled
, false);
654 OUT_BATCH(CMD
| (length
- 2));
656 OUT_BATCH(SET_FIELD(y
, BLT_Y
) | SET_FIELD(x
, BLT_X
));
657 OUT_BATCH(SET_FIELD(y
+ height
, BLT_Y
) | SET_FIELD(x
+ width
, BLT_X
));
660 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
664 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
667 OUT_BATCH(0xffffffff); /* white, but only alpha gets written */
668 ADVANCE_BATCH_TILED(dst_y_tiled
, false);
670 intel_batchbuffer_emit_mi_flush(brw
);