1 /**************************************************************************
3 * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
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 TUNGSTEN GRAPHICS 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 "intel_blit.h"
36 #include "intel_buffers.h"
37 #include "intel_context.h"
38 #include "intel_fbo.h"
39 #include "intel_reg.h"
40 #include "intel_regions.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 intel_context
*intel
,
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;
94 * Emits the packet for switching the blitter from X to Y tiled or back.
96 * This has to be called in a single BEGIN_BATCH_BLT_TILED() /
97 * ADVANCE_BATCH_TILED(). This is because BCS_SWCTRL is saved and restored as
98 * part of the power context, not a render context, and if the batchbuffer was
99 * to get flushed between setting and blitting, or blitting and restoring, our
100 * tiling state would leak into other unsuspecting applications (like the X
104 set_blitter_tiling(struct intel_context
*intel
,
105 bool dst_y_tiled
, bool src_y_tiled
)
107 assert(intel
->gen
>= 6);
109 /* Idle the blitter before we update how tiling is interpreted. */
110 OUT_BATCH(MI_FLUSH_DW
);
115 OUT_BATCH(MI_LOAD_REGISTER_IMM
| (3 - 2));
116 OUT_BATCH(BCS_SWCTRL
);
117 OUT_BATCH((BCS_SWCTRL_DST_Y
| BCS_SWCTRL_SRC_Y
) << 16 |
118 (dst_y_tiled
? BCS_SWCTRL_DST_Y
: 0) |
119 (src_y_tiled
? BCS_SWCTRL_SRC_Y
: 0));
122 #define BEGIN_BATCH_BLT_TILED(n, dst_y_tiled, src_y_tiled) do { \
123 BEGIN_BATCH_BLT(n + ((dst_y_tiled || src_y_tiled) ? 14 : 0)); \
124 if (dst_y_tiled || src_y_tiled) \
125 set_blitter_tiling(intel, dst_y_tiled, src_y_tiled); \
128 #define ADVANCE_BATCH_TILED(dst_y_tiled, src_y_tiled) do { \
129 if (dst_y_tiled || src_y_tiled) \
130 set_blitter_tiling(intel, false, false); \
135 * Implements a rectangular block transfer (blit) of pixels between two
138 * Our blitter can operate on 1, 2, or 4-byte-per-pixel data, with generous,
139 * but limited, pitches and sizes allowed.
141 * The src/dst coordinates are relative to the given level/slice of the
144 * If @src_flip or @dst_flip is set, then the rectangle within that miptree
145 * will be inverted (including scanline order) when copying. This is common
146 * in GL when copying between window system and user-created
147 * renderbuffers/textures.
150 intel_miptree_blit(struct intel_context
*intel
,
151 struct intel_mipmap_tree
*src_mt
,
152 int src_level
, int src_slice
,
153 uint32_t src_x
, uint32_t src_y
, bool src_flip
,
154 struct intel_mipmap_tree
*dst_mt
,
155 int dst_level
, int dst_slice
,
156 uint32_t dst_x
, uint32_t dst_y
, bool dst_flip
,
157 uint32_t width
, uint32_t height
,
160 /* No sRGB decode or encode is done by the hardware blitter, which is
161 * consistent with what we want in the callers (glCopyTexSubImage(),
162 * glBlitFramebuffer(), texture validation, etc.).
164 gl_format src_format
= _mesa_get_srgb_format_linear(src_mt
->format
);
165 gl_format dst_format
= _mesa_get_srgb_format_linear(dst_mt
->format
);
167 /* The blitter doesn't support doing any format conversions. We do also
168 * support blitting ARGB8888 to XRGB8888 (trivial, the values dropped into
169 * the X channel don't matter), and XRGB8888 to ARGB8888 by setting the A
170 * channel to 1.0 at the end.
172 if (src_format
!= dst_format
&&
173 ((src_format
!= MESA_FORMAT_ARGB8888
&&
174 src_format
!= MESA_FORMAT_XRGB8888
) ||
175 (dst_format
!= MESA_FORMAT_ARGB8888
&&
176 dst_format
!= MESA_FORMAT_XRGB8888
))) {
177 perf_debug("%s: Can't use hardware blitter from %s to %s, "
178 "falling back.\n", __FUNCTION__
,
179 _mesa_get_format_name(src_format
),
180 _mesa_get_format_name(dst_format
));
184 /* According to the Ivy Bridge PRM, Vol1 Part4, section 1.2.1.2 (Graphics
185 * Data Size Limitations):
187 * The BLT engine is capable of transferring very large quantities of
188 * graphics data. Any graphics data read from and written to the
189 * destination is permitted to represent a number of pixels that
190 * occupies up to 65,536 scan lines and up to 32,768 bytes per scan line
191 * at the destination. The maximum number of pixels that may be
192 * represented per scan line’s worth of graphics data depends on the
195 * Furthermore, intelEmitCopyBlit (which is called below) uses a signed
196 * 16-bit integer to represent buffer pitch, so it can only handle buffer
199 * As a result of these two limitations, we can only use the blitter to do
200 * this copy when the region's pitch is less than 32k.
202 if (src_mt
->region
->pitch
> 32768 ||
203 dst_mt
->region
->pitch
> 32768) {
204 perf_debug("Falling back due to >32k pitch\n");
209 src_y
= src_mt
->level
[src_level
].height
- src_y
- height
;
212 dst_y
= dst_mt
->level
[dst_level
].height
- dst_y
- height
;
214 int src_pitch
= src_mt
->region
->pitch
;
215 if (src_flip
!= dst_flip
)
216 src_pitch
= -src_pitch
;
218 uint32_t src_image_x
, src_image_y
;
219 intel_miptree_get_image_offset(src_mt
, src_level
, src_slice
,
220 &src_image_x
, &src_image_y
);
221 src_x
+= src_image_x
;
222 src_y
+= src_image_y
;
224 uint32_t dst_image_x
, dst_image_y
;
225 intel_miptree_get_image_offset(dst_mt
, dst_level
, dst_slice
,
226 &dst_image_x
, &dst_image_y
);
227 dst_x
+= dst_image_x
;
228 dst_y
+= dst_image_y
;
230 if (!intelEmitCopyBlit(intel
,
233 src_mt
->region
->bo
, src_mt
->offset
,
234 src_mt
->region
->tiling
,
235 dst_mt
->region
->pitch
,
236 dst_mt
->region
->bo
, dst_mt
->offset
,
237 dst_mt
->region
->tiling
,
245 if (src_mt
->format
== MESA_FORMAT_XRGB8888
&&
246 dst_mt
->format
== MESA_FORMAT_ARGB8888
) {
247 intel_miptree_set_alpha_to_one(intel
, dst_mt
,
258 intelEmitCopyBlit(struct intel_context
*intel
,
261 drm_intel_bo
*src_buffer
,
265 drm_intel_bo
*dst_buffer
,
268 GLshort src_x
, GLshort src_y
,
269 GLshort dst_x
, GLshort dst_y
,
270 GLshort w
, GLshort h
,
273 GLuint CMD
, BR13
, pass
= 0;
274 int dst_y2
= dst_y
+ h
;
275 int dst_x2
= dst_x
+ w
;
276 drm_intel_bo
*aper_array
[3];
277 bool dst_y_tiled
= dst_tiling
== I915_TILING_Y
;
278 bool src_y_tiled
= src_tiling
== I915_TILING_Y
;
281 if (dst_tiling
!= I915_TILING_NONE
) {
282 if (dst_offset
& 4095)
285 if (src_tiling
!= I915_TILING_NONE
) {
286 if (src_offset
& 4095)
289 if ((dst_y_tiled
|| src_y_tiled
) && intel
->gen
< 6)
292 /* do space check before going any further */
294 aper_array
[0] = intel
->batch
.bo
;
295 aper_array
[1] = dst_buffer
;
296 aper_array
[2] = src_buffer
;
298 if (dri_bufmgr_check_aperture_space(aper_array
, 3) != 0) {
299 intel_batchbuffer_flush(intel
);
308 intel_batchbuffer_require_space(intel
, 8 * 4, true);
309 DBG("%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
311 src_buffer
, src_pitch
, src_offset
, src_x
, src_y
,
312 dst_buffer
, dst_pitch
, dst_offset
, dst_x
, dst_y
, w
, h
);
314 /* Blit pitch must be dword-aligned. Otherwise, the hardware appears to drop
317 if (src_pitch
% 4 != 0 || dst_pitch
% 4 != 0)
320 /* For big formats (such as floating point), do the copy using 16 or 32bpp
321 * and multiply the coordinates.
330 assert(cpp
% 4 == 0);
338 BR13
= br13_for_cpp(cpp
) | translate_raster_op(logic_op
) << 16;
343 CMD
= XY_SRC_COPY_BLT_CMD
;
346 CMD
= XY_SRC_COPY_BLT_CMD
| XY_BLT_WRITE_ALPHA
| XY_BLT_WRITE_RGB
;
352 if (dst_y2
<= dst_y
|| dst_x2
<= dst_x
) {
356 assert(dst_x
< dst_x2
);
357 assert(dst_y
< dst_y2
);
359 BEGIN_BATCH_BLT_TILED(8, dst_y_tiled
, src_y_tiled
);
361 OUT_BATCH(CMD
| (8 - 2));
362 OUT_BATCH(BR13
| (uint16_t)dst_pitch
);
363 OUT_BATCH((dst_y
<< 16) | dst_x
);
364 OUT_BATCH((dst_y2
<< 16) | dst_x2
);
365 OUT_RELOC_FENCED(dst_buffer
,
366 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
368 OUT_BATCH((src_y
<< 16) | src_x
);
369 OUT_BATCH((uint16_t)src_pitch
);
370 OUT_RELOC_FENCED(src_buffer
,
371 I915_GEM_DOMAIN_RENDER
, 0,
374 ADVANCE_BATCH_TILED(dst_y_tiled
, src_y_tiled
);
376 intel_batchbuffer_emit_mi_flush(intel
);
383 * Use blitting to clear the renderbuffers named by 'flags'.
384 * Note: we can't use the ctx->DrawBuffer->_ColorDrawBufferIndexes field
385 * since that might include software renderbuffers or renderbuffers
386 * which we're clearing with triangles.
387 * \param mask bitmask of BUFFER_BIT_* values indicating buffers to clear
390 intelClearWithBlit(struct gl_context
*ctx
, GLbitfield mask
)
392 struct intel_context
*intel
= intel_context(ctx
);
393 struct gl_framebuffer
*fb
= ctx
->DrawBuffer
;
394 GLuint clear_depth_value
, clear_depth_mask
;
395 GLint cx
, cy
, cw
, ch
;
396 GLbitfield fail_mask
= 0;
399 /* Note: we don't use this function on Gen7+ hardware, so we can safely
400 * ignore fast color clear issues.
402 assert(intel
->gen
< 7);
405 * Compute values for clearing the buffers.
407 clear_depth_value
= 0;
408 clear_depth_mask
= 0;
409 if (mask
& BUFFER_BIT_DEPTH
) {
410 clear_depth_value
= (GLuint
) (fb
->_DepthMax
* ctx
->Depth
.Clear
);
411 clear_depth_mask
= XY_BLT_WRITE_RGB
;
413 if (mask
& BUFFER_BIT_STENCIL
) {
414 clear_depth_value
|= (ctx
->Stencil
.Clear
& 0xff) << 24;
415 clear_depth_mask
|= XY_BLT_WRITE_ALPHA
;
419 if (_mesa_is_winsys_fbo(fb
))
420 cy
= ctx
->DrawBuffer
->Height
- fb
->_Ymax
;
423 cw
= fb
->_Xmax
- fb
->_Xmin
;
424 ch
= fb
->_Ymax
- fb
->_Ymin
;
426 if (cw
== 0 || ch
== 0)
429 /* Loop over all renderbuffers */
430 mask
&= (1 << BUFFER_COUNT
) - 1;
432 GLuint buf
= ffs(mask
) - 1;
433 bool is_depth_stencil
= buf
== BUFFER_DEPTH
|| buf
== BUFFER_STENCIL
;
434 struct intel_renderbuffer
*irb
;
438 struct intel_region
*region
;
440 drm_intel_bo
*aper_array
[2];
444 irb
= intel_get_renderbuffer(fb
, buf
);
445 if (irb
&& irb
->mt
) {
446 region
= irb
->mt
->region
;
450 fail_mask
|= 1 << buf
;
454 /* OK, clear this renderbuffer */
455 x1
= cx
+ irb
->draw_x
;
456 y1
= cy
+ irb
->draw_y
;
457 x2
= cx
+ cw
+ irb
->draw_x
;
458 y2
= cy
+ ch
+ irb
->draw_y
;
460 pitch
= region
->pitch
;
463 DBG("%s dst:buf(%p)/%d %d,%d sz:%dx%d\n",
466 x1
, y1
, x2
- x1
, y2
- y1
);
469 CMD
= XY_COLOR_BLT_CMD
;
471 /* Setup the blit command */
473 if (is_depth_stencil
) {
474 CMD
|= clear_depth_mask
;
477 CMD
|= XY_BLT_WRITE_ALPHA
| XY_BLT_WRITE_RGB
;
481 assert(region
->tiling
!= I915_TILING_Y
);
485 if (is_depth_stencil
) {
486 clear_val
= clear_depth_value
;
489 GLfloat
*color
= ctx
->Color
.ClearColor
.f
;
491 _mesa_unclamped_float_rgba_to_ubyte(clear
, color
);
493 switch (intel_rb_format(irb
)) {
494 case MESA_FORMAT_ARGB8888
:
495 case MESA_FORMAT_XRGB8888
:
496 clear_val
= PACK_COLOR_8888(clear
[3], clear
[0],
499 case MESA_FORMAT_RGB565
:
500 clear_val
= PACK_COLOR_565(clear
[0], clear
[1], clear
[2]);
502 case MESA_FORMAT_ARGB4444
:
503 clear_val
= PACK_COLOR_4444(clear
[3], clear
[0],
506 case MESA_FORMAT_ARGB1555
:
507 clear_val
= PACK_COLOR_1555(clear
[3], clear
[0],
511 clear_val
= PACK_COLOR_8888(clear
[3], clear
[3],
515 fail_mask
|= 1 << buf
;
520 BR13
|= br13_for_cpp(cpp
);
525 /* do space check before going any further */
526 aper_array
[0] = intel
->batch
.bo
;
527 aper_array
[1] = region
->bo
;
529 if (drm_intel_bufmgr_check_aperture_space(aper_array
,
530 ARRAY_SIZE(aper_array
)) != 0) {
531 intel_batchbuffer_flush(intel
);
535 OUT_BATCH(CMD
| (6 - 2));
537 OUT_BATCH((y1
<< 16) | x1
);
538 OUT_BATCH((y2
<< 16) | x2
);
539 OUT_RELOC_FENCED(region
->bo
,
540 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
542 OUT_BATCH(clear_val
);
545 if (intel
->always_flush_cache
)
546 intel_batchbuffer_emit_mi_flush(intel
);
548 if (buf
== BUFFER_DEPTH
|| buf
== BUFFER_STENCIL
)
549 mask
&= ~(BUFFER_BIT_DEPTH
| BUFFER_BIT_STENCIL
);
556 intelEmitImmediateColorExpandBlit(struct intel_context
*intel
,
558 GLubyte
*src_bits
, GLuint src_size
,
561 drm_intel_bo
*dst_buffer
,
564 GLshort x
, GLshort y
,
565 GLshort w
, GLshort h
,
568 int dwords
= ALIGN(src_size
, 8) / 4;
569 uint32_t opcode
, br13
, blit_cmd
;
571 if (dst_tiling
!= I915_TILING_NONE
) {
572 if (dst_offset
& 4095)
574 if (dst_tiling
== I915_TILING_Y
)
578 assert( logic_op
- GL_CLEAR
>= 0 );
579 assert( logic_op
- GL_CLEAR
< 0x10 );
580 assert(dst_pitch
> 0);
585 DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d, %d bytes %d dwords\n",
587 dst_buffer
, dst_pitch
, dst_offset
, x
, y
, w
, h
, src_size
, dwords
);
589 intel_batchbuffer_require_space(intel
,
594 opcode
= XY_SETUP_BLT_CMD
;
596 opcode
|= XY_BLT_WRITE_ALPHA
| XY_BLT_WRITE_RGB
;
598 br13
= dst_pitch
| (translate_raster_op(logic_op
) << 16) | (1 << 29);
599 br13
|= br13_for_cpp(cpp
);
601 blit_cmd
= XY_TEXT_IMMEDIATE_BLIT_CMD
| XY_TEXT_BYTE_PACKED
; /* packing? */
602 if (dst_tiling
!= I915_TILING_NONE
)
603 blit_cmd
|= XY_DST_TILED
;
605 BEGIN_BATCH_BLT(8 + 3);
606 OUT_BATCH(opcode
| (8 - 2));
608 OUT_BATCH((0 << 16) | 0); /* clip x1, y1 */
609 OUT_BATCH((100 << 16) | 100); /* clip x2, y2 */
610 OUT_RELOC_FENCED(dst_buffer
,
611 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
613 OUT_BATCH(0); /* bg */
614 OUT_BATCH(fg_color
); /* fg */
615 OUT_BATCH(0); /* pattern base addr */
617 OUT_BATCH(blit_cmd
| ((3 - 2) + dwords
));
618 OUT_BATCH((y
<< 16) | x
);
619 OUT_BATCH(((y
+ h
) << 16) | (x
+ w
));
622 intel_batchbuffer_data(intel
, src_bits
, dwords
* 4, true);
624 intel_batchbuffer_emit_mi_flush(intel
);
629 /* We don't have a memmove-type blit like some other hardware, so we'll do a
630 * rectangular blit covering a large space, then emit 1-scanline blit at the
631 * end to cover the last if we need.
634 intel_emit_linear_blit(struct intel_context
*intel
,
635 drm_intel_bo
*dst_bo
,
636 unsigned int dst_offset
,
637 drm_intel_bo
*src_bo
,
638 unsigned int src_offset
,
641 struct gl_context
*ctx
= &intel
->ctx
;
642 GLuint pitch
, height
;
645 /* The pitch given to the GPU must be DWORD aligned, and
646 * we want width to match pitch. Max width is (1 << 15 - 1),
647 * rounding that down to the nearest DWORD is 1 << 15 - 4
649 pitch
= ROUND_DOWN_TO(MIN2(size
, (1 << 15) - 1), 4);
650 height
= (pitch
== 0) ? 1 : size
/ pitch
;
651 ok
= intelEmitCopyBlit(intel
, 1,
652 pitch
, src_bo
, src_offset
, I915_TILING_NONE
,
653 pitch
, dst_bo
, dst_offset
, I915_TILING_NONE
,
656 pitch
, height
, /* w, h */
659 _mesa_problem(ctx
, "Failed to linear blit %dx%d\n", pitch
, height
);
661 src_offset
+= pitch
* height
;
662 dst_offset
+= pitch
* height
;
663 size
-= pitch
* height
;
664 assert (size
< (1 << 15));
665 pitch
= ALIGN(size
, 4);
667 ok
= intelEmitCopyBlit(intel
, 1,
668 pitch
, src_bo
, src_offset
, I915_TILING_NONE
,
669 pitch
, dst_bo
, dst_offset
, I915_TILING_NONE
,
675 _mesa_problem(ctx
, "Failed to linear blit %dx%d\n", size
, 1);
680 * Used to initialize the alpha value of an ARGB8888 miptree after copying
681 * into it from an XRGB8888 source.
683 * This is very common with glCopyTexImage2D(). Note that the coordinates are
684 * relative to the start of the miptree, not relative to a slice within the
688 intel_miptree_set_alpha_to_one(struct intel_context
*intel
,
689 struct intel_mipmap_tree
*mt
,
690 int x
, int y
, int width
, int height
)
692 struct intel_region
*region
= mt
->region
;
695 drm_intel_bo
*aper_array
[2];
698 pitch
= region
->pitch
;
701 DBG("%s dst:buf(%p)/%d %d,%d sz:%dx%d\n",
702 __FUNCTION__
, region
->bo
, pitch
, x
, y
, width
, height
);
704 BR13
= br13_for_cpp(cpp
) | 0xf0 << 16;
705 CMD
= XY_COLOR_BLT_CMD
;
706 CMD
|= XY_BLT_WRITE_ALPHA
;
710 /* do space check before going any further */
711 aper_array
[0] = intel
->batch
.bo
;
712 aper_array
[1] = region
->bo
;
714 if (drm_intel_bufmgr_check_aperture_space(aper_array
,
715 ARRAY_SIZE(aper_array
)) != 0) {
716 intel_batchbuffer_flush(intel
);
719 bool dst_y_tiled
= region
->tiling
== I915_TILING_Y
;
721 BEGIN_BATCH_BLT_TILED(6, dst_y_tiled
, false);
722 OUT_BATCH(CMD
| (6 - 2));
724 OUT_BATCH((y
<< 16) | x
);
725 OUT_BATCH(((y
+ height
) << 16) | (x
+ width
));
726 OUT_RELOC_FENCED(region
->bo
,
727 I915_GEM_DOMAIN_RENDER
, I915_GEM_DOMAIN_RENDER
,
729 OUT_BATCH(0xffffffff); /* white, but only alpha gets written */
730 ADVANCE_BATCH_TILED(dst_y_tiled
, false);
732 intel_batchbuffer_emit_mi_flush(intel
);