2 * Copyright (C) 2017 Rob Clark <robclark@freedesktop.org>
3 * Copyright © 2018 Google, Inc.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
25 * Rob Clark <robclark@freedesktop.org>
28 #include "util/u_dump.h"
29 #include "util/half_float.h"
31 #include "freedreno_blitter.h"
32 #include "freedreno_fence.h"
33 #include "freedreno_log.h"
34 #include "freedreno_resource.h"
36 #include "fd6_blitter.h"
37 #include "fd6_format.h"
39 #include "fd6_resource.h"
42 static inline enum a6xx_2d_ifmt
43 fd6_ifmt(enum a6xx_format fmt
)
51 case FMT6_8_8_8_8_UNORM
:
52 case FMT6_8_8_8_X8_UNORM
:
53 case FMT6_8_8_8_8_SNORM
:
54 case FMT6_4_4_4_4_UNORM
:
55 case FMT6_5_5_5_1_UNORM
:
56 case FMT6_5_6_5_UNORM
:
63 case FMT6_32_32_32_32_UINT
:
64 case FMT6_32_32_32_32_SINT
:
71 case FMT6_16_16_16_16_UINT
:
72 case FMT6_16_16_16_16_SINT
:
73 case FMT6_10_10_10_2_UINT
:
80 case FMT6_8_8_8_8_UINT
:
81 case FMT6_8_8_8_8_SINT
:
82 case FMT6_Z24_UNORM_S8_UINT
:
83 case FMT6_Z24_UNORM_S8_UINT_AS_R8G8B8A8
:
88 case FMT6_16_16_UNORM
:
89 case FMT6_16_16_SNORM
:
90 case FMT6_16_16_16_16_UNORM
:
91 case FMT6_16_16_16_16_SNORM
:
93 case FMT6_32_32_FLOAT
:
94 case FMT6_32_32_32_32_FLOAT
:
98 case FMT6_16_16_FLOAT
:
99 case FMT6_16_16_16_16_FLOAT
:
100 case FMT6_11_11_10_FLOAT
:
101 case FMT6_10_10_10_2_UNORM_DEST
:
105 unreachable("bad format");
110 /* Make sure none of the requested dimensions extend beyond the size of the
111 * resource. Not entirely sure why this happens, but sometimes it does, and
112 * w/ 2d blt doesn't have wrap modes like a sampler, so force those cases
116 ok_dims(const struct pipe_resource
*r
, const struct pipe_box
*b
, int lvl
)
119 r
->target
== PIPE_TEXTURE_3D
? u_minify(r
->depth0
, lvl
)
122 return (b
->x
>= 0) && (b
->x
+ b
->width
<= u_minify(r
->width0
, lvl
)) &&
123 (b
->y
>= 0) && (b
->y
+ b
->height
<= u_minify(r
->height0
, lvl
)) &&
124 (b
->z
>= 0) && (b
->z
+ b
->depth
<= last_layer
);
128 ok_format(enum pipe_format pfmt
)
130 enum a6xx_format fmt
= fd6_pipe2color(pfmt
);
132 if (util_format_is_compressed(pfmt
))
136 case PIPE_FORMAT_Z24_UNORM_S8_UINT
:
137 case PIPE_FORMAT_Z24X8_UNORM
:
138 case PIPE_FORMAT_Z16_UNORM
:
139 case PIPE_FORMAT_Z32_UNORM
:
140 case PIPE_FORMAT_Z32_FLOAT
:
141 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
142 case PIPE_FORMAT_S8_UINT
:
148 if (fmt
== FMT6_NONE
)
151 if (fmt
== FMT6_10_10_10_2_UNORM_DEST
)
158 #define DEBUG_BLIT_FALLBACK 0
160 #define fail_if(cond) \
163 if (DEBUG_BLIT_FALLBACK) { \
164 fprintf(stderr, "falling back: %s for blit:\n", #cond); \
165 util_dump_blit_info(stderr, info); \
166 fprintf(stderr, "\nsrc: "); \
167 util_dump_resource(stderr, info->src.resource); \
168 fprintf(stderr, "\ndst: "); \
169 util_dump_resource(stderr, info->dst.resource); \
170 fprintf(stderr, "\n"); \
177 can_do_blit(const struct pipe_blit_info
*info
)
179 /* I think we can do scaling, but not in z dimension since that would
182 fail_if(info
->dst
.box
.depth
!= info
->src
.box
.depth
);
184 /* Fail if unsupported format: */
185 fail_if(!ok_format(info
->src
.format
));
186 fail_if(!ok_format(info
->dst
.format
));
188 debug_assert(!util_format_is_compressed(info
->src
.format
));
189 debug_assert(!util_format_is_compressed(info
->dst
.format
));
191 fail_if(!ok_dims(info
->src
.resource
, &info
->src
.box
, info
->src
.level
));
193 fail_if(!ok_dims(info
->dst
.resource
, &info
->dst
.box
, info
->dst
.level
));
195 debug_assert(info
->dst
.box
.width
>= 0);
196 debug_assert(info
->dst
.box
.height
>= 0);
197 debug_assert(info
->dst
.box
.depth
>= 0);
199 fail_if(info
->dst
.resource
->nr_samples
> 1);
201 fail_if(info
->window_rectangle_include
);
203 const struct util_format_description
*src_desc
=
204 util_format_description(info
->src
.format
);
205 const struct util_format_description
*dst_desc
=
206 util_format_description(info
->dst
.format
);
207 const int common_channels
= MIN2(src_desc
->nr_channels
, dst_desc
->nr_channels
);
209 if (info
->mask
& PIPE_MASK_RGBA
) {
210 for (int i
= 0; i
< common_channels
; i
++) {
211 fail_if(memcmp(&src_desc
->channel
[i
],
212 &dst_desc
->channel
[i
],
213 sizeof(src_desc
->channel
[0])));
217 fail_if(info
->alpha_blend
);
223 emit_setup(struct fd_batch
*batch
)
225 struct fd_ringbuffer
*ring
= batch
->draw
;
227 fd6_event_write(batch
, ring
, PC_CCU_FLUSH_COLOR_TS
, true);
228 fd6_event_write(batch
, ring
, PC_CCU_FLUSH_DEPTH_TS
, true);
229 fd6_event_write(batch
, ring
, PC_CCU_INVALIDATE_COLOR
, false);
230 fd6_event_write(batch
, ring
, PC_CCU_INVALIDATE_DEPTH
, false);
232 /* normal BLIT_OP_SCALE operation needs bypass RB_CCU_CNTL */
234 OUT_PKT4(ring
, REG_A6XX_RB_CCU_CNTL
, 1);
235 OUT_RING(ring
, fd6_context(batch
->ctx
)->magic
.RB_CCU_CNTL_bypass
);
239 blit_control(enum a6xx_format fmt
, bool is_srgb
)
241 enum a6xx_2d_ifmt ifmt
= fd6_ifmt(fmt
);
244 assert(ifmt
== R2D_UNORM8
);
245 ifmt
= R2D_UNORM8_SRGB
;
248 return A6XX_RB_2D_BLIT_CNTL_MASK(0xf) |
249 A6XX_RB_2D_BLIT_CNTL_COLOR_FORMAT(fmt
) |
250 A6XX_RB_2D_BLIT_CNTL_IFMT(ifmt
);
253 /* buffers need to be handled specially since x/width can exceed the bounds
254 * supported by hw.. if necessary decompose into (potentially) two 2D blits
257 emit_blit_buffer(struct fd_context
*ctx
, struct fd_ringbuffer
*ring
,
258 const struct pipe_blit_info
*info
)
260 const struct pipe_box
*sbox
= &info
->src
.box
;
261 const struct pipe_box
*dbox
= &info
->dst
.box
;
262 struct fd_resource
*src
, *dst
;
263 unsigned sshift
, dshift
;
266 fprintf(stderr
, "buffer blit: ");
267 util_dump_blit_info(stderr
, info
);
268 fprintf(stderr
, "\ndst resource: ");
269 util_dump_resource(stderr
, info
->dst
.resource
);
270 fprintf(stderr
, "\nsrc resource: ");
271 util_dump_resource(stderr
, info
->src
.resource
);
272 fprintf(stderr
, "\n");
275 src
= fd_resource(info
->src
.resource
);
276 dst
= fd_resource(info
->dst
.resource
);
278 debug_assert(src
->layout
.cpp
== 1);
279 debug_assert(dst
->layout
.cpp
== 1);
280 debug_assert(info
->src
.resource
->format
== info
->dst
.resource
->format
);
281 debug_assert((sbox
->y
== 0) && (sbox
->height
== 1));
282 debug_assert((dbox
->y
== 0) && (dbox
->height
== 1));
283 debug_assert((sbox
->z
== 0) && (sbox
->depth
== 1));
284 debug_assert((dbox
->z
== 0) && (dbox
->depth
== 1));
285 debug_assert(sbox
->width
== dbox
->width
);
286 debug_assert(info
->src
.level
== 0);
287 debug_assert(info
->dst
.level
== 0);
290 * Buffers can have dimensions bigger than max width, remap into
291 * multiple 1d blits to fit within max dimension
293 * Note that blob uses .ARRAY_PITCH=128 for blitting buffers, which
294 * seems to prevent overfetch related faults. Not quite sure what
297 * Low 6 bits of SRC/DST addresses need to be zero (ie. address
298 * aligned to 64) so we need to shift src/dst x1/x2 to make up the
299 * difference. On top of already splitting up the blit so width
302 * We perhaps could do a bit better, if src and dst are aligned but
303 * in the worst case this means we have to split the copy up into
304 * 16k (0x4000) minus 64 (0x40).
307 sshift
= sbox
->x
& 0x3f;
308 dshift
= dbox
->x
& 0x3f;
310 OUT_PKT7(ring
, CP_SET_MARKER
, 1);
311 OUT_RING(ring
, A6XX_CP_SET_MARKER_0_MODE(RM6_BLIT2DSCALE
));
313 uint32_t blit_cntl
= blit_control(FMT6_8_UNORM
, false) | 0x20000000;
314 OUT_PKT4(ring
, REG_A6XX_RB_2D_BLIT_CNTL
, 1);
315 OUT_RING(ring
, blit_cntl
);
317 OUT_PKT4(ring
, REG_A6XX_GRAS_2D_BLIT_CNTL
, 1);
318 OUT_RING(ring
, blit_cntl
);
320 for (unsigned off
= 0; off
< sbox
->width
; off
+= (0x4000 - 0x40)) {
321 unsigned soff
, doff
, w
, p
;
323 soff
= (sbox
->x
+ off
) & ~0x3f;
324 doff
= (dbox
->x
+ off
) & ~0x3f;
326 w
= MIN2(sbox
->width
- off
, (0x4000 - 0x40));
329 debug_assert((soff
+ w
) <= fd_bo_size(src
->bo
));
330 debug_assert((doff
+ w
) <= fd_bo_size(dst
->bo
));
335 OUT_PKT4(ring
, REG_A6XX_SP_PS_2D_SRC_INFO
, 10);
336 OUT_RING(ring
, A6XX_SP_PS_2D_SRC_INFO_COLOR_FORMAT(FMT6_8_UNORM
) |
337 A6XX_SP_PS_2D_SRC_INFO_TILE_MODE(TILE6_LINEAR
) |
338 A6XX_SP_PS_2D_SRC_INFO_COLOR_SWAP(WZYX
) |
340 OUT_RING(ring
, A6XX_SP_PS_2D_SRC_SIZE_WIDTH(sshift
+ w
) |
341 A6XX_SP_PS_2D_SRC_SIZE_HEIGHT(1)); /* SP_PS_2D_SRC_SIZE */
342 OUT_RELOC(ring
, src
->bo
, soff
, 0, 0); /* SP_PS_2D_SRC_LO/HI */
343 OUT_RING(ring
, A6XX_SP_PS_2D_SRC_PITCH_PITCH(p
));
345 OUT_RING(ring
, 0x00000000);
346 OUT_RING(ring
, 0x00000000);
347 OUT_RING(ring
, 0x00000000);
348 OUT_RING(ring
, 0x00000000);
349 OUT_RING(ring
, 0x00000000);
354 OUT_PKT4(ring
, REG_A6XX_RB_2D_DST_INFO
, 9);
355 OUT_RING(ring
, A6XX_RB_2D_DST_INFO_COLOR_FORMAT(FMT6_8_UNORM
) |
356 A6XX_RB_2D_DST_INFO_TILE_MODE(TILE6_LINEAR
) |
357 A6XX_RB_2D_DST_INFO_COLOR_SWAP(WZYX
));
358 OUT_RELOC(ring
, dst
->bo
, doff
, 0, 0); /* RB_2D_DST_LO/HI */
359 OUT_RING(ring
, A6XX_RB_2D_DST_SIZE_PITCH(p
));
360 OUT_RING(ring
, 0x00000000);
361 OUT_RING(ring
, 0x00000000);
362 OUT_RING(ring
, 0x00000000);
363 OUT_RING(ring
, 0x00000000);
364 OUT_RING(ring
, 0x00000000);
369 OUT_PKT4(ring
, REG_A6XX_GRAS_2D_SRC_TL_X
, 4);
370 OUT_RING(ring
, A6XX_GRAS_2D_SRC_TL_X_X(sshift
));
371 OUT_RING(ring
, A6XX_GRAS_2D_SRC_BR_X_X(sshift
+ w
- 1));
372 OUT_RING(ring
, A6XX_GRAS_2D_SRC_TL_Y_Y(0));
373 OUT_RING(ring
, A6XX_GRAS_2D_SRC_BR_Y_Y(0));
375 OUT_PKT4(ring
, REG_A6XX_GRAS_2D_DST_TL
, 2);
376 OUT_RING(ring
, A6XX_GRAS_2D_DST_TL_X(dshift
) | A6XX_GRAS_2D_DST_TL_Y(0));
377 OUT_RING(ring
, A6XX_GRAS_2D_DST_BR_X(dshift
+ w
- 1) | A6XX_GRAS_2D_DST_BR_Y(0));
379 OUT_PKT7(ring
, CP_EVENT_WRITE
, 1);
380 OUT_RING(ring
, 0x3f);
383 OUT_PKT4(ring
, REG_A6XX_RB_UNKNOWN_8C01
, 1);
386 OUT_PKT4(ring
, REG_A6XX_SP_2D_SRC_FORMAT
, 1);
387 OUT_RING(ring
, 0xf180);
389 OUT_PKT4(ring
, REG_A6XX_RB_UNKNOWN_8E04
, 1);
390 OUT_RING(ring
, fd6_context(ctx
)->magic
.RB_UNKNOWN_8E04_blit
);
392 OUT_PKT7(ring
, CP_BLIT
, 1);
393 OUT_RING(ring
, CP_BLIT_0_OP(BLIT_OP_SCALE
));
397 OUT_PKT4(ring
, REG_A6XX_RB_UNKNOWN_8E04
, 1);
398 OUT_RING(ring
, 0); /* RB_UNKNOWN_8E04 */
403 emit_blit_or_clear_texture(struct fd_context
*ctx
, struct fd_ringbuffer
*ring
,
404 const struct pipe_blit_info
*info
, union pipe_color_union
*color
)
406 const struct pipe_box
*sbox
= &info
->src
.box
;
407 const struct pipe_box
*dbox
= &info
->dst
.box
;
408 struct fd_resource
*src
, *dst
;
409 uint32_t spitch
, dpitch
;
410 enum a6xx_format sfmt
, dfmt
;
411 enum a6xx_tile_mode stile
, dtile
;
412 enum a3xx_color_swap sswap
, dswap
;
413 int sx1
, sy1
, sx2
, sy2
;
414 int dx1
, dy1
, dx2
, dy2
;
417 fprintf(stderr
, "texture blit: ");
418 util_dump_blit_info(stderr
, info
);
419 fprintf(stderr
, "\ndst resource: ");
420 util_dump_resource(stderr
, info
->dst
.resource
);
421 fprintf(stderr
, "\nsrc resource: ");
422 util_dump_resource(stderr
, info
->src
.resource
);
423 fprintf(stderr
, "\n");
426 src
= fd_resource(info
->src
.resource
);
427 dst
= fd_resource(info
->dst
.resource
);
429 spitch
= fd_resource_pitch(src
, info
->src
.level
);
430 dpitch
= fd_resource_pitch(dst
, info
->dst
.level
);
432 sfmt
= fd6_pipe2color(info
->src
.format
);
433 dfmt
= fd6_pipe2color(info
->dst
.format
);
435 stile
= fd_resource_tile_mode(info
->src
.resource
, info
->src
.level
);
436 dtile
= fd_resource_tile_mode(info
->dst
.resource
, info
->dst
.level
);
438 /* Linear levels of a tiled resource are always WZYX, so look at
439 * rsc->tile_mode to determine the swap.
441 sswap
= fd6_resource_swap(src
, info
->src
.format
);
442 dswap
= fd6_resource_swap(dst
, info
->dst
.format
);
444 uint32_t nr_samples
= fd_resource_nr_samples(&dst
->base
);
445 sx1
= sbox
->x
* nr_samples
;
447 sx2
= (sbox
->x
+ sbox
->width
) * nr_samples
- 1;
448 sy2
= sbox
->y
+ sbox
->height
- 1;
450 dx1
= dbox
->x
* nr_samples
;
452 dx2
= (dbox
->x
+ dbox
->width
) * nr_samples
- 1;
453 dy2
= dbox
->y
+ dbox
->height
- 1;
455 uint32_t width
= u_minify(src
->base
.width0
, info
->src
.level
) * nr_samples
;
456 uint32_t height
= u_minify(src
->base
.height0
, info
->src
.level
);
458 OUT_PKT7(ring
, CP_SET_MARKER
, 1);
459 OUT_RING(ring
, A6XX_CP_SET_MARKER_0_MODE(RM6_BLIT2DSCALE
));
461 uint32_t blit_cntl
= blit_control(dfmt
, util_format_is_srgb(info
->dst
.format
));
464 blit_cntl
|= A6XX_RB_2D_BLIT_CNTL_SOLID_COLOR
;
466 switch (info
->dst
.format
) {
467 case PIPE_FORMAT_Z24X8_UNORM
:
468 case PIPE_FORMAT_Z24_UNORM_S8_UINT
:
469 case PIPE_FORMAT_X24S8_UINT
: {
470 uint32_t depth_unorm24
= color
->f
[0] * ((1u << 24) - 1);
471 uint8_t stencil
= color
->ui
[1];
472 color
->ui
[0] = depth_unorm24
& 0xff;
473 color
->ui
[1] = (depth_unorm24
>> 8) & 0xff;
474 color
->ui
[2] = (depth_unorm24
>> 16) & 0xff;
475 color
->ui
[3] = stencil
;
477 dfmt
= FMT6_Z24_UNORM_S8_UINT_AS_R8G8B8A8
;
484 OUT_PKT4(ring
, REG_A6XX_RB_2D_SRC_SOLID_C0
, 4);
486 switch (fd6_ifmt(dfmt
)) {
488 case R2D_UNORM8_SRGB
:
489 OUT_RING(ring
, float_to_ubyte(color
->f
[0]));
490 OUT_RING(ring
, float_to_ubyte(color
->f
[1]));
491 OUT_RING(ring
, float_to_ubyte(color
->f
[2]));
492 OUT_RING(ring
, float_to_ubyte(color
->f
[3]));
495 OUT_RING(ring
, _mesa_float_to_half(color
->f
[0]));
496 OUT_RING(ring
, _mesa_float_to_half(color
->f
[1]));
497 OUT_RING(ring
, _mesa_float_to_half(color
->f
[2]));
498 OUT_RING(ring
, _mesa_float_to_half(color
->f
[3]));
499 sfmt
= FMT6_16_16_16_16_FLOAT
;
506 OUT_RING(ring
, color
->ui
[0]);
507 OUT_RING(ring
, color
->ui
[1]);
508 OUT_RING(ring
, color
->ui
[2]);
509 OUT_RING(ring
, color
->ui
[3]);
515 blit_cntl
|= 0x20000000;
517 if (info
->scissor_enable
) {
518 OUT_PKT4(ring
, REG_A6XX_GRAS_RESOLVE_CNTL_1
, 2);
519 OUT_RING(ring
, A6XX_GRAS_RESOLVE_CNTL_1_X(info
->scissor
.minx
) |
520 A6XX_GRAS_RESOLVE_CNTL_1_Y(info
->scissor
.miny
));
521 OUT_RING(ring
, A6XX_GRAS_RESOLVE_CNTL_1_X(info
->scissor
.maxx
- 1) |
522 A6XX_GRAS_RESOLVE_CNTL_1_Y(info
->scissor
.maxy
- 1));
523 blit_cntl
|= A6XX_RB_2D_BLIT_CNTL_SCISSOR
;
526 OUT_PKT4(ring
, REG_A6XX_RB_2D_BLIT_CNTL
, 1);
527 OUT_RING(ring
, blit_cntl
);
529 OUT_PKT4(ring
, REG_A6XX_GRAS_2D_BLIT_CNTL
, 1);
530 OUT_RING(ring
, blit_cntl
);
532 for (unsigned i
= 0; i
< info
->dst
.box
.depth
; i
++) {
533 unsigned soff
= fd_resource_offset(src
, info
->src
.level
, sbox
->z
+ i
);
534 unsigned doff
= fd_resource_offset(dst
, info
->dst
.level
, dbox
->z
+ i
);
535 bool subwc_enabled
= fd_resource_ubwc_enabled(src
, info
->src
.level
);
536 bool dubwc_enabled
= fd_resource_ubwc_enabled(dst
, info
->dst
.level
);
542 if (info
->filter
== PIPE_TEX_FILTER_LINEAR
)
543 filter
= A6XX_SP_PS_2D_SRC_INFO_FILTER
;
545 enum a3xx_msaa_samples samples
= fd_msaa_samples(src
->base
.nr_samples
);
547 if (sfmt
== FMT6_10_10_10_2_UNORM_DEST
)
548 sfmt
= FMT6_10_10_10_2_UNORM
;
550 OUT_PKT4(ring
, REG_A6XX_SP_PS_2D_SRC_INFO
, 10);
551 OUT_RING(ring
, A6XX_SP_PS_2D_SRC_INFO_COLOR_FORMAT(sfmt
) |
552 A6XX_SP_PS_2D_SRC_INFO_TILE_MODE(stile
) |
553 A6XX_SP_PS_2D_SRC_INFO_COLOR_SWAP(sswap
) |
554 A6XX_SP_PS_2D_SRC_INFO_SAMPLES(samples
) |
555 COND(samples
> MSAA_ONE
&& (info
->mask
& PIPE_MASK_RGBA
),
556 A6XX_SP_PS_2D_SRC_INFO_SAMPLES_AVERAGE
) |
557 COND(subwc_enabled
, A6XX_SP_PS_2D_SRC_INFO_FLAGS
) |
558 COND(util_format_is_srgb(info
->src
.format
), A6XX_SP_PS_2D_SRC_INFO_SRGB
) |
560 OUT_RING(ring
, A6XX_SP_PS_2D_SRC_SIZE_WIDTH(width
) |
561 A6XX_SP_PS_2D_SRC_SIZE_HEIGHT(height
)); /* SP_PS_2D_SRC_SIZE */
562 OUT_RELOC(ring
, src
->bo
, soff
, 0, 0); /* SP_PS_2D_SRC_LO/HI */
563 OUT_RING(ring
, A6XX_SP_PS_2D_SRC_PITCH_PITCH(spitch
));
565 OUT_RING(ring
, 0x00000000);
566 OUT_RING(ring
, 0x00000000);
567 OUT_RING(ring
, 0x00000000);
568 OUT_RING(ring
, 0x00000000);
569 OUT_RING(ring
, 0x00000000);
572 OUT_PKT4(ring
, REG_A6XX_SP_PS_2D_SRC_FLAGS_LO
, 6);
573 fd6_emit_flag_reference(ring
, src
, info
->src
.level
, sbox
->z
+ i
);
574 OUT_RING(ring
, 0x00000000);
575 OUT_RING(ring
, 0x00000000);
576 OUT_RING(ring
, 0x00000000);
582 OUT_PKT4(ring
, REG_A6XX_RB_2D_DST_INFO
, 9);
583 OUT_RING(ring
, A6XX_RB_2D_DST_INFO_COLOR_FORMAT(dfmt
) |
584 A6XX_RB_2D_DST_INFO_TILE_MODE(dtile
) |
585 A6XX_RB_2D_DST_INFO_COLOR_SWAP(dswap
) |
586 COND(util_format_is_srgb(info
->dst
.format
), A6XX_RB_2D_DST_INFO_SRGB
) |
587 COND(dubwc_enabled
, A6XX_RB_2D_DST_INFO_FLAGS
));
588 OUT_RELOC(ring
, dst
->bo
, doff
, 0, 0); /* RB_2D_DST_LO/HI */
589 OUT_RING(ring
, A6XX_RB_2D_DST_SIZE_PITCH(dpitch
));
590 OUT_RING(ring
, 0x00000000);
591 OUT_RING(ring
, 0x00000000);
592 OUT_RING(ring
, 0x00000000);
593 OUT_RING(ring
, 0x00000000);
594 OUT_RING(ring
, 0x00000000);
597 OUT_PKT4(ring
, REG_A6XX_RB_2D_DST_FLAGS_LO
, 6);
598 fd6_emit_flag_reference(ring
, dst
, info
->dst
.level
, dbox
->z
+ i
);
599 OUT_RING(ring
, 0x00000000);
600 OUT_RING(ring
, 0x00000000);
601 OUT_RING(ring
, 0x00000000);
607 OUT_PKT4(ring
, REG_A6XX_GRAS_2D_SRC_TL_X
, 4);
608 OUT_RING(ring
, A6XX_GRAS_2D_SRC_TL_X_X(sx1
));
609 OUT_RING(ring
, A6XX_GRAS_2D_SRC_BR_X_X(sx2
));
610 OUT_RING(ring
, A6XX_GRAS_2D_SRC_TL_Y_Y(sy1
));
611 OUT_RING(ring
, A6XX_GRAS_2D_SRC_BR_Y_Y(sy2
));
613 OUT_PKT4(ring
, REG_A6XX_GRAS_2D_DST_TL
, 2);
614 OUT_RING(ring
, A6XX_GRAS_2D_DST_TL_X(dx1
) | A6XX_GRAS_2D_DST_TL_Y(dy1
));
615 OUT_RING(ring
, A6XX_GRAS_2D_DST_BR_X(dx2
) | A6XX_GRAS_2D_DST_BR_Y(dy2
));
617 OUT_PKT7(ring
, CP_EVENT_WRITE
, 1);
618 OUT_RING(ring
, 0x3f);
621 OUT_PKT4(ring
, REG_A6XX_RB_UNKNOWN_8C01
, 1);
624 if (dfmt
== FMT6_10_10_10_2_UNORM_DEST
)
625 sfmt
= FMT6_16_16_16_16_FLOAT
;
627 /* This register is probably badly named... it seems that it's
628 * controlling the internal/accumulator format or something like
629 * that. It's certainly not tied to only the src format.
631 OUT_PKT4(ring
, REG_A6XX_SP_2D_SRC_FORMAT
, 1);
632 OUT_RING(ring
, A6XX_SP_2D_SRC_FORMAT_COLOR_FORMAT(sfmt
) |
633 COND(util_format_is_pure_sint(info
->src
.format
),
634 A6XX_SP_2D_SRC_FORMAT_SINT
) |
635 COND(util_format_is_pure_uint(info
->src
.format
),
636 A6XX_SP_2D_SRC_FORMAT_UINT
) |
637 COND(util_format_is_snorm(info
->src
.format
),
638 A6XX_SP_2D_SRC_FORMAT_SINT
|
639 A6XX_SP_2D_SRC_FORMAT_NORM
) |
640 COND(util_format_is_unorm(info
->src
.format
),
641 // TODO sometimes blob uses UINT+NORM but dEQP seems unhappy about that
642 // A6XX_SP_2D_SRC_FORMAT_UINT |
643 A6XX_SP_2D_SRC_FORMAT_NORM
) |
644 COND(util_format_is_srgb(info
->dst
.format
), A6XX_SP_2D_SRC_FORMAT_SRGB
) |
645 A6XX_SP_2D_SRC_FORMAT_MASK(0xf));
647 OUT_PKT4(ring
, REG_A6XX_RB_UNKNOWN_8E04
, 1);
648 OUT_RING(ring
, fd6_context(ctx
)->magic
.RB_UNKNOWN_8E04_blit
);
650 OUT_PKT7(ring
, CP_BLIT
, 1);
651 OUT_RING(ring
, CP_BLIT_0_OP(BLIT_OP_SCALE
));
655 OUT_PKT4(ring
, REG_A6XX_RB_UNKNOWN_8E04
, 1);
656 OUT_RING(ring
, 0); /* RB_UNKNOWN_8E04 */
661 fd6_clear_surface(struct fd_context
*ctx
,
662 struct fd_ringbuffer
*ring
, struct pipe_surface
*psurf
,
663 uint32_t width
, uint32_t height
, union pipe_color_union
*color
)
665 struct pipe_blit_info info
= {};
667 info
.dst
.resource
= psurf
->texture
;
668 info
.dst
.level
= psurf
->u
.tex
.level
;
671 info
.dst
.box
.z
= psurf
->u
.tex
.first_layer
;
672 info
.dst
.box
.width
= width
;
673 info
.dst
.box
.height
= height
;
674 info
.dst
.box
.depth
= psurf
->u
.tex
.last_layer
+ 1 - psurf
->u
.tex
.first_layer
;
675 info
.dst
.format
= psurf
->format
;
677 info
.mask
= util_format_get_mask(psurf
->format
);
678 info
.filter
= PIPE_TEX_FILTER_NEAREST
;
679 info
.scissor_enable
= 0;
681 emit_blit_or_clear_texture(ctx
, ring
, &info
, color
);
685 handle_rgba_blit(struct fd_context
*ctx
, const struct pipe_blit_info
*info
)
687 struct fd_batch
*batch
;
689 debug_assert(!(info
->mask
& PIPE_MASK_ZS
));
691 if (!can_do_blit(info
))
694 batch
= fd_bc_alloc_batch(&ctx
->screen
->batch_cache
, ctx
, true);
696 fd6_emit_restore(batch
, batch
->draw
);
697 fd6_emit_lrz_flush(batch
->draw
);
699 fd_screen_lock(ctx
->screen
);
701 fd_batch_resource_read(batch
, fd_resource(info
->src
.resource
));
702 fd_batch_resource_write(batch
, fd_resource(info
->dst
.resource
));
704 fd_screen_unlock(ctx
->screen
);
706 /* Clearing last_fence must come after the batch dependency tracking
707 * (resource_read()/resource_write()), as that can trigger a flush,
708 * re-populating last_fence
710 fd_fence_ref(&ctx
->last_fence
, NULL
);
712 fd_batch_set_stage(batch
, FD_STAGE_BLIT
);
714 fd_log_stream(batch
, stream
, util_dump_blit_info(stream
, info
));
718 if ((info
->src
.resource
->target
== PIPE_BUFFER
) &&
719 (info
->dst
.resource
->target
== PIPE_BUFFER
)) {
720 assert(fd_resource(info
->src
.resource
)->layout
.tile_mode
== TILE6_LINEAR
);
721 assert(fd_resource(info
->dst
.resource
)->layout
.tile_mode
== TILE6_LINEAR
);
722 fd_log(batch
, "START BLIT (BUFFER)");
723 emit_blit_buffer(ctx
, batch
->draw
, info
);
724 fd_log(batch
, "END BLIT (BUFFER)");
726 /* I don't *think* we need to handle blits between buffer <-> !buffer */
727 debug_assert(info
->src
.resource
->target
!= PIPE_BUFFER
);
728 debug_assert(info
->dst
.resource
->target
!= PIPE_BUFFER
);
729 fd_log(batch
, "START BLIT (TEXTURE)");
730 emit_blit_or_clear_texture(ctx
, batch
->draw
, info
, NULL
);
731 fd_log(batch
, "END BLIT (TEXTURE)");
734 fd6_event_write(batch
, batch
->draw
, PC_CCU_FLUSH_COLOR_TS
, true);
735 fd6_event_write(batch
, batch
->draw
, PC_CCU_FLUSH_DEPTH_TS
, true);
736 fd6_event_write(batch
, batch
->draw
, CACHE_FLUSH_TS
, true);
737 fd6_cache_inv(batch
, batch
->draw
);
739 fd_resource(info
->dst
.resource
)->valid
= true;
740 batch
->needs_flush
= true;
742 fd_batch_flush(batch
);
743 fd_batch_reference(&batch
, NULL
);
749 * Re-written z/s blits can still fail for various reasons (for example MSAA).
750 * But we want to do the fallback blit with the re-written pipe_blit_info,
751 * in particular as u_blitter cannot blit stencil. So handle the fallback
752 * ourself and never "fail".
755 do_rewritten_blit(struct fd_context
*ctx
, const struct pipe_blit_info
*info
)
757 bool success
= handle_rgba_blit(ctx
, info
);
759 success
= fd_blitter_blit(ctx
, info
);
760 debug_assert(success
); /* fallback should never fail! */
765 * Handle depth/stencil blits either via u_blitter and/or re-writing the
766 * blit into an equivilant format that we can handle
769 handle_zs_blit(struct fd_context
*ctx
, const struct pipe_blit_info
*info
)
771 struct pipe_blit_info blit
= *info
;
774 fprintf(stderr
, "---- handle_zs_blit: ");
775 util_dump_blit_info(stderr
, info
);
776 fprintf(stderr
, "\ndst resource: ");
777 util_dump_resource(stderr
, info
->dst
.resource
);
778 fprintf(stderr
, "\nsrc resource: ");
779 util_dump_resource(stderr
, info
->src
.resource
);
780 fprintf(stderr
, "\n");
783 switch (info
->dst
.format
) {
784 case PIPE_FORMAT_S8_UINT
:
785 debug_assert(info
->mask
== PIPE_MASK_S
);
786 blit
.mask
= PIPE_MASK_R
;
787 blit
.src
.format
= PIPE_FORMAT_R8_UINT
;
788 blit
.dst
.format
= PIPE_FORMAT_R8_UINT
;
789 return do_rewritten_blit(ctx
, &blit
);
791 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
792 if (info
->mask
& PIPE_MASK_Z
) {
793 blit
.mask
= PIPE_MASK_R
;
794 blit
.src
.format
= PIPE_FORMAT_R32_FLOAT
;
795 blit
.dst
.format
= PIPE_FORMAT_R32_FLOAT
;
796 do_rewritten_blit(ctx
, &blit
);
799 if (info
->mask
& PIPE_MASK_S
) {
800 blit
.mask
= PIPE_MASK_R
;
801 blit
.src
.format
= PIPE_FORMAT_R8_UINT
;
802 blit
.dst
.format
= PIPE_FORMAT_R8_UINT
;
803 blit
.src
.resource
= &fd_resource(info
->src
.resource
)->stencil
->base
;
804 blit
.dst
.resource
= &fd_resource(info
->dst
.resource
)->stencil
->base
;
805 do_rewritten_blit(ctx
, &blit
);
810 case PIPE_FORMAT_Z16_UNORM
:
811 blit
.mask
= PIPE_MASK_R
;
812 blit
.src
.format
= PIPE_FORMAT_R16_UNORM
;
813 blit
.dst
.format
= PIPE_FORMAT_R16_UNORM
;
814 return do_rewritten_blit(ctx
, &blit
);
816 case PIPE_FORMAT_Z32_UNORM
:
817 case PIPE_FORMAT_Z32_FLOAT
:
818 debug_assert(info
->mask
== PIPE_MASK_Z
);
819 blit
.mask
= PIPE_MASK_R
;
820 blit
.src
.format
= PIPE_FORMAT_R32_UINT
;
821 blit
.dst
.format
= PIPE_FORMAT_R32_UINT
;
822 return do_rewritten_blit(ctx
, &blit
);
824 case PIPE_FORMAT_Z24X8_UNORM
:
825 case PIPE_FORMAT_Z24_UNORM_S8_UINT
:
827 if (info
->mask
& PIPE_MASK_Z
)
828 blit
.mask
|= PIPE_MASK_R
| PIPE_MASK_G
| PIPE_MASK_B
;
829 if (info
->mask
& PIPE_MASK_S
)
830 blit
.mask
|= PIPE_MASK_A
;
831 blit
.src
.format
= PIPE_FORMAT_Z24_UNORM_S8_UINT_AS_R8G8B8A8
;
832 blit
.dst
.format
= PIPE_FORMAT_Z24_UNORM_S8_UINT_AS_R8G8B8A8
;
833 return fd_blitter_blit(ctx
, &blit
);
841 handle_compressed_blit(struct fd_context
*ctx
, const struct pipe_blit_info
*info
)
843 struct pipe_blit_info blit
= *info
;
846 fprintf(stderr
, "---- handle_compressed_blit: ");
847 util_dump_blit_info(stderr
, info
);
848 fprintf(stderr
, "\ndst resource: ");
849 util_dump_resource(stderr
, info
->dst
.resource
);
850 fprintf(stderr
, "\nsrc resource: ");
851 util_dump_resource(stderr
, info
->src
.resource
);
852 fprintf(stderr
, "\n");
855 if (info
->src
.format
!= info
->dst
.format
)
856 return fd_blitter_blit(ctx
, info
);
858 if (util_format_get_blocksize(info
->src
.format
) == 8) {
859 blit
.src
.format
= blit
.dst
.format
= PIPE_FORMAT_R16G16B16A16_UINT
;
861 debug_assert(util_format_get_blocksize(info
->src
.format
) == 16);
862 blit
.src
.format
= blit
.dst
.format
= PIPE_FORMAT_R32G32B32A32_UINT
;
865 int bw
= util_format_get_blockwidth(info
->src
.format
);
866 int bh
= util_format_get_blockheight(info
->src
.format
);
868 /* NOTE: x/y *must* be aligned to block boundary (ie. in
869 * glCompressedTexSubImage2D()) but width/height may not
873 debug_assert((blit
.src
.box
.x
% bw
) == 0);
874 debug_assert((blit
.src
.box
.y
% bh
) == 0);
876 blit
.src
.box
.x
/= bw
;
877 blit
.src
.box
.y
/= bh
;
878 blit
.src
.box
.width
= DIV_ROUND_UP(blit
.src
.box
.width
, bw
);
879 blit
.src
.box
.height
= DIV_ROUND_UP(blit
.src
.box
.height
, bh
);
881 debug_assert((blit
.dst
.box
.x
% bw
) == 0);
882 debug_assert((blit
.dst
.box
.y
% bh
) == 0);
884 blit
.dst
.box
.x
/= bw
;
885 blit
.dst
.box
.y
/= bh
;
886 blit
.dst
.box
.width
= DIV_ROUND_UP(blit
.dst
.box
.width
, bw
);
887 blit
.dst
.box
.height
= DIV_ROUND_UP(blit
.dst
.box
.height
, bh
);
889 return do_rewritten_blit(ctx
, &blit
);
893 fd6_blit(struct fd_context
*ctx
, const struct pipe_blit_info
*info
)
895 if (info
->mask
& PIPE_MASK_ZS
)
896 return handle_zs_blit(ctx
, info
);
897 if (util_format_is_compressed(info
->src
.format
) ||
898 util_format_is_compressed(info
->dst
.format
))
899 return handle_compressed_blit(ctx
, info
);
901 return handle_rgba_blit(ctx
, info
);
905 fd6_blitter_init(struct pipe_context
*pctx
)
907 if (fd_mesa_debug
& FD_DBG_NOBLIT
)
910 fd_context(pctx
)->blit
= fd6_blit
;
914 fd6_tile_mode(const struct pipe_resource
*tmpl
)
916 /* if the mipmap level 0 is still too small to be tiled, then don't
919 if (fd_resource_level_linear(tmpl
, 0))
922 /* basically just has to be a format we can blit, so uploads/downloads
923 * via linear staging buffer works:
925 if (ok_format(tmpl
->format
))