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
)
155 #define DEBUG_BLIT_FALLBACK 0
157 #define fail_if(cond) \
160 if (DEBUG_BLIT_FALLBACK) { \
161 fprintf(stderr, "falling back: %s for blit:\n", #cond); \
162 util_dump_blit_info(stderr, info); \
163 fprintf(stderr, "\nsrc: "); \
164 util_dump_resource(stderr, info->src.resource); \
165 fprintf(stderr, "\ndst: "); \
166 util_dump_resource(stderr, info->dst.resource); \
167 fprintf(stderr, "\n"); \
174 can_do_blit(const struct pipe_blit_info
*info
)
176 /* I think we can do scaling, but not in z dimension since that would
179 fail_if(info
->dst
.box
.depth
!= info
->src
.box
.depth
);
181 /* Fail if unsupported format: */
182 fail_if(!ok_format(info
->src
.format
));
183 fail_if(!ok_format(info
->dst
.format
));
185 debug_assert(!util_format_is_compressed(info
->src
.format
));
186 debug_assert(!util_format_is_compressed(info
->dst
.format
));
188 fail_if(!ok_dims(info
->src
.resource
, &info
->src
.box
, info
->src
.level
));
190 fail_if(!ok_dims(info
->dst
.resource
, &info
->dst
.box
, info
->dst
.level
));
192 debug_assert(info
->dst
.box
.width
>= 0);
193 debug_assert(info
->dst
.box
.height
>= 0);
194 debug_assert(info
->dst
.box
.depth
>= 0);
196 fail_if(info
->dst
.resource
->nr_samples
> 1);
198 fail_if(info
->window_rectangle_include
);
200 const struct util_format_description
*src_desc
=
201 util_format_description(info
->src
.format
);
202 const struct util_format_description
*dst_desc
=
203 util_format_description(info
->dst
.format
);
204 const int common_channels
= MIN2(src_desc
->nr_channels
, dst_desc
->nr_channels
);
206 if (info
->mask
& PIPE_MASK_RGBA
) {
207 for (int i
= 0; i
< common_channels
; i
++) {
208 fail_if(memcmp(&src_desc
->channel
[i
],
209 &dst_desc
->channel
[i
],
210 sizeof(src_desc
->channel
[0])));
214 fail_if(info
->alpha_blend
);
220 emit_setup(struct fd_batch
*batch
)
222 struct fd_ringbuffer
*ring
= batch
->draw
;
224 fd6_event_write(batch
, ring
, PC_CCU_FLUSH_COLOR_TS
, true);
225 fd6_event_write(batch
, ring
, PC_CCU_FLUSH_DEPTH_TS
, true);
226 fd6_event_write(batch
, ring
, PC_CCU_INVALIDATE_COLOR
, false);
227 fd6_event_write(batch
, ring
, PC_CCU_INVALIDATE_DEPTH
, false);
229 /* normal BLIT_OP_SCALE operation needs bypass RB_CCU_CNTL */
231 OUT_PKT4(ring
, REG_A6XX_RB_CCU_CNTL
, 1);
232 OUT_RING(ring
, fd6_context(batch
->ctx
)->magic
.RB_CCU_CNTL_bypass
);
236 emit_blit_setup(struct fd_ringbuffer
*ring
,
237 enum pipe_format pfmt
, bool scissor_enable
, union pipe_color_union
*color
)
239 enum a6xx_format fmt
= fd6_pipe2color(pfmt
);
240 bool is_srgb
= util_format_is_srgb(pfmt
);
241 enum a6xx_2d_ifmt ifmt
= fd6_ifmt(fmt
);
243 OUT_PKT7(ring
, CP_SET_MARKER
, 1);
244 OUT_RING(ring
, A6XX_CP_SET_MARKER_0_MODE(RM6_BLIT2DSCALE
));
247 assert(ifmt
== R2D_UNORM8
);
248 ifmt
= R2D_UNORM8_SRGB
;
251 uint32_t blit_cntl
= A6XX_RB_2D_BLIT_CNTL_MASK(0xf) |
252 A6XX_RB_2D_BLIT_CNTL_COLOR_FORMAT(fmt
) |
253 A6XX_RB_2D_BLIT_CNTL_IFMT(ifmt
) |
254 COND(color
, A6XX_RB_2D_BLIT_CNTL_SOLID_COLOR
) |
255 COND(scissor_enable
, A6XX_RB_2D_BLIT_CNTL_SCISSOR
);
257 OUT_PKT4(ring
, REG_A6XX_RB_2D_BLIT_CNTL
, 1);
258 OUT_RING(ring
, blit_cntl
);
260 OUT_PKT4(ring
, REG_A6XX_GRAS_2D_BLIT_CNTL
, 1);
261 OUT_RING(ring
, blit_cntl
);
263 if (fmt
== FMT6_10_10_10_2_UNORM_DEST
)
264 fmt
= FMT6_16_16_16_16_FLOAT
;
266 /* This register is probably badly named... it seems that it's
267 * controlling the internal/accumulator format or something like
268 * that. It's certainly not tied to only the src format.
270 OUT_PKT4(ring
, REG_A6XX_SP_2D_SRC_FORMAT
, 1);
271 OUT_RING(ring
, A6XX_SP_2D_SRC_FORMAT_COLOR_FORMAT(fmt
) |
272 COND(util_format_is_pure_sint(pfmt
),
273 A6XX_SP_2D_SRC_FORMAT_SINT
) |
274 COND(util_format_is_pure_uint(pfmt
),
275 A6XX_SP_2D_SRC_FORMAT_UINT
) |
276 COND(util_format_is_snorm(pfmt
),
277 A6XX_SP_2D_SRC_FORMAT_SINT
|
278 A6XX_SP_2D_SRC_FORMAT_NORM
) |
279 COND(util_format_is_unorm(pfmt
),
280 // TODO sometimes blob uses UINT+NORM but dEQP seems unhappy about that
281 // A6XX_SP_2D_SRC_FORMAT_UINT |
282 A6XX_SP_2D_SRC_FORMAT_NORM
) |
283 COND(is_srgb
, A6XX_SP_2D_SRC_FORMAT_SRGB
) |
284 A6XX_SP_2D_SRC_FORMAT_MASK(0xf));
286 OUT_PKT4(ring
, REG_A6XX_RB_UNKNOWN_8C01
, 1);
290 /* buffers need to be handled specially since x/width can exceed the bounds
291 * supported by hw.. if necessary decompose into (potentially) two 2D blits
294 emit_blit_buffer(struct fd_context
*ctx
, struct fd_ringbuffer
*ring
,
295 const struct pipe_blit_info
*info
)
297 const struct pipe_box
*sbox
= &info
->src
.box
;
298 const struct pipe_box
*dbox
= &info
->dst
.box
;
299 struct fd_resource
*src
, *dst
;
300 unsigned sshift
, dshift
;
303 fprintf(stderr
, "buffer blit: ");
304 util_dump_blit_info(stderr
, info
);
305 fprintf(stderr
, "\ndst resource: ");
306 util_dump_resource(stderr
, info
->dst
.resource
);
307 fprintf(stderr
, "\nsrc resource: ");
308 util_dump_resource(stderr
, info
->src
.resource
);
309 fprintf(stderr
, "\n");
312 src
= fd_resource(info
->src
.resource
);
313 dst
= fd_resource(info
->dst
.resource
);
315 debug_assert(src
->layout
.cpp
== 1);
316 debug_assert(dst
->layout
.cpp
== 1);
317 debug_assert(info
->src
.resource
->format
== info
->dst
.resource
->format
);
318 debug_assert((sbox
->y
== 0) && (sbox
->height
== 1));
319 debug_assert((dbox
->y
== 0) && (dbox
->height
== 1));
320 debug_assert((sbox
->z
== 0) && (sbox
->depth
== 1));
321 debug_assert((dbox
->z
== 0) && (dbox
->depth
== 1));
322 debug_assert(sbox
->width
== dbox
->width
);
323 debug_assert(info
->src
.level
== 0);
324 debug_assert(info
->dst
.level
== 0);
327 * Buffers can have dimensions bigger than max width, remap into
328 * multiple 1d blits to fit within max dimension
330 * Note that blob uses .ARRAY_PITCH=128 for blitting buffers, which
331 * seems to prevent overfetch related faults. Not quite sure what
334 * Low 6 bits of SRC/DST addresses need to be zero (ie. address
335 * aligned to 64) so we need to shift src/dst x1/x2 to make up the
336 * difference. On top of already splitting up the blit so width
339 * We perhaps could do a bit better, if src and dst are aligned but
340 * in the worst case this means we have to split the copy up into
341 * 16k (0x4000) minus 64 (0x40).
344 sshift
= sbox
->x
& 0x3f;
345 dshift
= dbox
->x
& 0x3f;
347 emit_blit_setup(ring
, PIPE_FORMAT_R8_UNORM
, false, NULL
);
349 for (unsigned off
= 0; off
< sbox
->width
; off
+= (0x4000 - 0x40)) {
350 unsigned soff
, doff
, w
, p
;
352 soff
= (sbox
->x
+ off
) & ~0x3f;
353 doff
= (dbox
->x
+ off
) & ~0x3f;
355 w
= MIN2(sbox
->width
- off
, (0x4000 - 0x40));
358 debug_assert((soff
+ w
) <= fd_bo_size(src
->bo
));
359 debug_assert((doff
+ w
) <= fd_bo_size(dst
->bo
));
364 OUT_PKT4(ring
, REG_A6XX_SP_PS_2D_SRC_INFO
, 10);
365 OUT_RING(ring
, A6XX_SP_PS_2D_SRC_INFO_COLOR_FORMAT(FMT6_8_UNORM
) |
366 A6XX_SP_PS_2D_SRC_INFO_TILE_MODE(TILE6_LINEAR
) |
367 A6XX_SP_PS_2D_SRC_INFO_COLOR_SWAP(WZYX
) |
369 OUT_RING(ring
, A6XX_SP_PS_2D_SRC_SIZE_WIDTH(sshift
+ w
) |
370 A6XX_SP_PS_2D_SRC_SIZE_HEIGHT(1)); /* SP_PS_2D_SRC_SIZE */
371 OUT_RELOC(ring
, src
->bo
, soff
, 0, 0); /* SP_PS_2D_SRC_LO/HI */
372 OUT_RING(ring
, A6XX_SP_PS_2D_SRC_PITCH_PITCH(p
));
374 OUT_RING(ring
, 0x00000000);
375 OUT_RING(ring
, 0x00000000);
376 OUT_RING(ring
, 0x00000000);
377 OUT_RING(ring
, 0x00000000);
378 OUT_RING(ring
, 0x00000000);
383 OUT_PKT4(ring
, REG_A6XX_RB_2D_DST_INFO
, 9);
384 OUT_RING(ring
, A6XX_RB_2D_DST_INFO_COLOR_FORMAT(FMT6_8_UNORM
) |
385 A6XX_RB_2D_DST_INFO_TILE_MODE(TILE6_LINEAR
) |
386 A6XX_RB_2D_DST_INFO_COLOR_SWAP(WZYX
));
387 OUT_RELOC(ring
, dst
->bo
, doff
, 0, 0); /* RB_2D_DST_LO/HI */
388 OUT_RING(ring
, A6XX_RB_2D_DST_SIZE_PITCH(p
));
389 OUT_RING(ring
, 0x00000000);
390 OUT_RING(ring
, 0x00000000);
391 OUT_RING(ring
, 0x00000000);
392 OUT_RING(ring
, 0x00000000);
393 OUT_RING(ring
, 0x00000000);
398 OUT_PKT4(ring
, REG_A6XX_GRAS_2D_SRC_TL_X
, 4);
399 OUT_RING(ring
, A6XX_GRAS_2D_SRC_TL_X_X(sshift
));
400 OUT_RING(ring
, A6XX_GRAS_2D_SRC_BR_X_X(sshift
+ w
- 1));
401 OUT_RING(ring
, A6XX_GRAS_2D_SRC_TL_Y_Y(0));
402 OUT_RING(ring
, A6XX_GRAS_2D_SRC_BR_Y_Y(0));
404 OUT_PKT4(ring
, REG_A6XX_GRAS_2D_DST_TL
, 2);
405 OUT_RING(ring
, A6XX_GRAS_2D_DST_TL_X(dshift
) | A6XX_GRAS_2D_DST_TL_Y(0));
406 OUT_RING(ring
, A6XX_GRAS_2D_DST_BR_X(dshift
+ w
- 1) | A6XX_GRAS_2D_DST_BR_Y(0));
408 OUT_PKT7(ring
, CP_EVENT_WRITE
, 1);
409 OUT_RING(ring
, 0x3f);
412 OUT_PKT4(ring
, REG_A6XX_RB_UNKNOWN_8E04
, 1);
413 OUT_RING(ring
, fd6_context(ctx
)->magic
.RB_UNKNOWN_8E04_blit
);
415 OUT_PKT7(ring
, CP_BLIT
, 1);
416 OUT_RING(ring
, CP_BLIT_0_OP(BLIT_OP_SCALE
));
420 OUT_PKT4(ring
, REG_A6XX_RB_UNKNOWN_8E04
, 1);
421 OUT_RING(ring
, 0); /* RB_UNKNOWN_8E04 */
426 emit_blit_dst(struct fd_ringbuffer
*ring
, const struct pipe_blit_info
*info
, unsigned layer
)
428 struct fd_resource
*dst
= fd_resource(info
->dst
.resource
);
429 enum a6xx_format fmt
= fd6_pipe2color(info
->dst
.format
);
430 enum a6xx_tile_mode tile
= fd_resource_tile_mode(info
->dst
.resource
, info
->dst
.level
);
431 enum a3xx_color_swap swap
= fd6_resource_swap(dst
, info
->dst
.format
);
432 uint32_t pitch
= fd_resource_pitch(dst
, info
->dst
.level
);
433 bool ubwc_enabled
= fd_resource_ubwc_enabled(dst
, info
->dst
.level
);
434 unsigned off
= fd_resource_offset(dst
, info
->dst
.level
, layer
);
436 if (fmt
== FMT6_Z24_UNORM_S8_UINT
)
437 fmt
= FMT6_Z24_UNORM_S8_UINT_AS_R8G8B8A8
;
439 OUT_PKT4(ring
, REG_A6XX_RB_2D_DST_INFO
, 9);
440 OUT_RING(ring
, A6XX_RB_2D_DST_INFO_COLOR_FORMAT(fmt
) |
441 A6XX_RB_2D_DST_INFO_TILE_MODE(tile
) |
442 A6XX_RB_2D_DST_INFO_COLOR_SWAP(swap
) |
443 COND(util_format_is_srgb(info
->dst
.format
), A6XX_RB_2D_DST_INFO_SRGB
) |
444 COND(ubwc_enabled
, A6XX_RB_2D_DST_INFO_FLAGS
));
445 OUT_RELOC(ring
, dst
->bo
, off
, 0, 0); /* RB_2D_DST_LO/HI */
446 OUT_RING(ring
, A6XX_RB_2D_DST_SIZE_PITCH(pitch
));
447 OUT_RING(ring
, 0x00000000);
448 OUT_RING(ring
, 0x00000000);
449 OUT_RING(ring
, 0x00000000);
450 OUT_RING(ring
, 0x00000000);
451 OUT_RING(ring
, 0x00000000);
454 OUT_PKT4(ring
, REG_A6XX_RB_2D_DST_FLAGS_LO
, 6);
455 fd6_emit_flag_reference(ring
, dst
, info
->dst
.level
, layer
);
456 OUT_RING(ring
, 0x00000000);
457 OUT_RING(ring
, 0x00000000);
458 OUT_RING(ring
, 0x00000000);
463 emit_blit_src(struct fd_ringbuffer
*ring
, const struct pipe_blit_info
*info
, unsigned layer
, unsigned nr_samples
)
465 struct fd_resource
*src
= fd_resource(info
->src
.resource
);
466 enum a6xx_format sfmt
= fd6_pipe2color(info
->src
.format
);
467 enum a6xx_tile_mode stile
= fd_resource_tile_mode(info
->src
.resource
, info
->src
.level
);
468 enum a3xx_color_swap sswap
= fd6_resource_swap(src
, info
->src
.format
);
469 uint32_t pitch
= fd_resource_pitch(src
, info
->src
.level
);
470 bool subwc_enabled
= fd_resource_ubwc_enabled(src
, info
->src
.level
);
471 unsigned soff
= fd_resource_offset(src
, info
->src
.level
, layer
);
472 uint32_t width
= u_minify(src
->base
.width0
, info
->src
.level
) * nr_samples
;
473 uint32_t height
= u_minify(src
->base
.height0
, info
->src
.level
);
476 if (info
->filter
== PIPE_TEX_FILTER_LINEAR
)
477 filter
= A6XX_SP_PS_2D_SRC_INFO_FILTER
;
479 enum a3xx_msaa_samples samples
= fd_msaa_samples(src
->base
.nr_samples
);
481 if (sfmt
== FMT6_10_10_10_2_UNORM_DEST
)
482 sfmt
= FMT6_10_10_10_2_UNORM
;
484 OUT_PKT4(ring
, REG_A6XX_SP_PS_2D_SRC_INFO
, 10);
485 OUT_RING(ring
, A6XX_SP_PS_2D_SRC_INFO_COLOR_FORMAT(sfmt
) |
486 A6XX_SP_PS_2D_SRC_INFO_TILE_MODE(stile
) |
487 A6XX_SP_PS_2D_SRC_INFO_COLOR_SWAP(sswap
) |
488 A6XX_SP_PS_2D_SRC_INFO_SAMPLES(samples
) |
489 COND(samples
> MSAA_ONE
&& (info
->mask
& PIPE_MASK_RGBA
),
490 A6XX_SP_PS_2D_SRC_INFO_SAMPLES_AVERAGE
) |
491 COND(subwc_enabled
, A6XX_SP_PS_2D_SRC_INFO_FLAGS
) |
492 COND(util_format_is_srgb(info
->src
.format
), A6XX_SP_PS_2D_SRC_INFO_SRGB
) |
494 OUT_RING(ring
, A6XX_SP_PS_2D_SRC_SIZE_WIDTH(width
) |
495 A6XX_SP_PS_2D_SRC_SIZE_HEIGHT(height
)); /* SP_PS_2D_SRC_SIZE */
496 OUT_RELOC(ring
, src
->bo
, soff
, 0, 0); /* SP_PS_2D_SRC_LO/HI */
497 OUT_RING(ring
, A6XX_SP_PS_2D_SRC_PITCH_PITCH(pitch
));
499 OUT_RING(ring
, 0x00000000);
500 OUT_RING(ring
, 0x00000000);
501 OUT_RING(ring
, 0x00000000);
502 OUT_RING(ring
, 0x00000000);
503 OUT_RING(ring
, 0x00000000);
506 OUT_PKT4(ring
, REG_A6XX_SP_PS_2D_SRC_FLAGS_LO
, 6);
507 fd6_emit_flag_reference(ring
, src
, info
->src
.level
, layer
);
508 OUT_RING(ring
, 0x00000000);
509 OUT_RING(ring
, 0x00000000);
510 OUT_RING(ring
, 0x00000000);
515 emit_blit_or_clear_texture(struct fd_context
*ctx
, struct fd_ringbuffer
*ring
,
516 const struct pipe_blit_info
*info
, union pipe_color_union
*color
)
518 const struct pipe_box
*sbox
= &info
->src
.box
;
519 const struct pipe_box
*dbox
= &info
->dst
.box
;
520 struct fd_resource
*dst
;
521 enum a6xx_format dfmt
;
522 int sx1
, sy1
, sx2
, sy2
;
523 int dx1
, dy1
, dx2
, dy2
;
526 fprintf(stderr
, "texture blit: ");
527 util_dump_blit_info(stderr
, info
);
528 fprintf(stderr
, "\ndst resource: ");
529 util_dump_resource(stderr
, info
->dst
.resource
);
530 fprintf(stderr
, "\nsrc resource: ");
531 util_dump_resource(stderr
, info
->src
.resource
);
532 fprintf(stderr
, "\n");
535 dst
= fd_resource(info
->dst
.resource
);
537 dfmt
= fd6_pipe2color(info
->dst
.format
);
539 uint32_t nr_samples
= fd_resource_nr_samples(&dst
->base
);
542 sx1
= sbox
->x
* nr_samples
;
544 sx2
= (sbox
->x
+ sbox
->width
) * nr_samples
- 1;
545 sy2
= sbox
->y
+ sbox
->height
- 1;
547 OUT_PKT4(ring
, REG_A6XX_GRAS_2D_SRC_TL_X
, 4);
548 OUT_RING(ring
, A6XX_GRAS_2D_SRC_TL_X_X(sx1
));
549 OUT_RING(ring
, A6XX_GRAS_2D_SRC_BR_X_X(sx2
));
550 OUT_RING(ring
, A6XX_GRAS_2D_SRC_TL_Y_Y(sy1
));
551 OUT_RING(ring
, A6XX_GRAS_2D_SRC_BR_Y_Y(sy2
));
554 dx1
= dbox
->x
* nr_samples
;
556 dx2
= (dbox
->x
+ dbox
->width
) * nr_samples
- 1;
557 dy2
= dbox
->y
+ dbox
->height
- 1;
559 OUT_PKT4(ring
, REG_A6XX_GRAS_2D_DST_TL
, 2);
560 OUT_RING(ring
, A6XX_GRAS_2D_DST_TL_X(dx1
) | A6XX_GRAS_2D_DST_TL_Y(dy1
));
561 OUT_RING(ring
, A6XX_GRAS_2D_DST_BR_X(dx2
) | A6XX_GRAS_2D_DST_BR_Y(dy2
));
564 switch (info
->dst
.format
) {
565 case PIPE_FORMAT_Z24X8_UNORM
:
566 case PIPE_FORMAT_Z24_UNORM_S8_UINT
:
567 case PIPE_FORMAT_X24S8_UINT
: {
568 uint32_t depth_unorm24
= color
->f
[0] * ((1u << 24) - 1);
569 uint8_t stencil
= color
->ui
[1];
570 color
->ui
[0] = depth_unorm24
& 0xff;
571 color
->ui
[1] = (depth_unorm24
>> 8) & 0xff;
572 color
->ui
[2] = (depth_unorm24
>> 16) & 0xff;
573 color
->ui
[3] = stencil
;
580 OUT_PKT4(ring
, REG_A6XX_RB_2D_SRC_SOLID_C0
, 4);
582 switch (fd6_ifmt(dfmt
)) {
584 case R2D_UNORM8_SRGB
:
585 OUT_RING(ring
, float_to_ubyte(color
->f
[0]));
586 OUT_RING(ring
, float_to_ubyte(color
->f
[1]));
587 OUT_RING(ring
, float_to_ubyte(color
->f
[2]));
588 OUT_RING(ring
, float_to_ubyte(color
->f
[3]));
591 OUT_RING(ring
, _mesa_float_to_half(color
->f
[0]));
592 OUT_RING(ring
, _mesa_float_to_half(color
->f
[1]));
593 OUT_RING(ring
, _mesa_float_to_half(color
->f
[2]));
594 OUT_RING(ring
, _mesa_float_to_half(color
->f
[3]));
601 OUT_RING(ring
, color
->ui
[0]);
602 OUT_RING(ring
, color
->ui
[1]);
603 OUT_RING(ring
, color
->ui
[2]);
604 OUT_RING(ring
, color
->ui
[3]);
609 if (info
->scissor_enable
) {
610 OUT_PKT4(ring
, REG_A6XX_GRAS_RESOLVE_CNTL_1
, 2);
611 OUT_RING(ring
, A6XX_GRAS_RESOLVE_CNTL_1_X(info
->scissor
.minx
) |
612 A6XX_GRAS_RESOLVE_CNTL_1_Y(info
->scissor
.miny
));
613 OUT_RING(ring
, A6XX_GRAS_RESOLVE_CNTL_1_X(info
->scissor
.maxx
- 1) |
614 A6XX_GRAS_RESOLVE_CNTL_1_Y(info
->scissor
.maxy
- 1));
617 emit_blit_setup(ring
, info
->dst
.format
, info
->scissor_enable
, color
);
619 for (unsigned i
= 0; i
< info
->dst
.box
.depth
; i
++) {
622 emit_blit_src(ring
, info
, sbox
->z
+ i
, nr_samples
);
624 emit_blit_dst(ring
, info
, dbox
->z
+ i
);
629 OUT_PKT7(ring
, CP_EVENT_WRITE
, 1);
630 OUT_RING(ring
, 0x3f);
633 OUT_PKT4(ring
, REG_A6XX_RB_UNKNOWN_8E04
, 1);
634 OUT_RING(ring
, fd6_context(ctx
)->magic
.RB_UNKNOWN_8E04_blit
);
636 OUT_PKT7(ring
, CP_BLIT
, 1);
637 OUT_RING(ring
, CP_BLIT_0_OP(BLIT_OP_SCALE
));
641 OUT_PKT4(ring
, REG_A6XX_RB_UNKNOWN_8E04
, 1);
642 OUT_RING(ring
, 0); /* RB_UNKNOWN_8E04 */
647 fd6_clear_surface(struct fd_context
*ctx
,
648 struct fd_ringbuffer
*ring
, struct pipe_surface
*psurf
,
649 uint32_t width
, uint32_t height
, union pipe_color_union
*color
)
651 struct pipe_blit_info info
= {};
653 info
.dst
.resource
= psurf
->texture
;
654 info
.dst
.level
= psurf
->u
.tex
.level
;
657 info
.dst
.box
.z
= psurf
->u
.tex
.first_layer
;
658 info
.dst
.box
.width
= width
;
659 info
.dst
.box
.height
= height
;
660 info
.dst
.box
.depth
= psurf
->u
.tex
.last_layer
+ 1 - psurf
->u
.tex
.first_layer
;
661 info
.dst
.format
= psurf
->format
;
663 info
.mask
= util_format_get_mask(psurf
->format
);
664 info
.filter
= PIPE_TEX_FILTER_NEAREST
;
665 info
.scissor_enable
= 0;
667 emit_blit_or_clear_texture(ctx
, ring
, &info
, color
);
671 handle_rgba_blit(struct fd_context
*ctx
, const struct pipe_blit_info
*info
)
673 struct fd_batch
*batch
;
675 debug_assert(!(info
->mask
& PIPE_MASK_ZS
));
677 if (!can_do_blit(info
))
680 batch
= fd_bc_alloc_batch(&ctx
->screen
->batch_cache
, ctx
, true);
682 fd6_emit_restore(batch
, batch
->draw
);
683 fd6_emit_lrz_flush(batch
->draw
);
685 fd_screen_lock(ctx
->screen
);
687 fd_batch_resource_read(batch
, fd_resource(info
->src
.resource
));
688 fd_batch_resource_write(batch
, fd_resource(info
->dst
.resource
));
690 fd_screen_unlock(ctx
->screen
);
692 /* Clearing last_fence must come after the batch dependency tracking
693 * (resource_read()/resource_write()), as that can trigger a flush,
694 * re-populating last_fence
696 fd_fence_ref(&ctx
->last_fence
, NULL
);
698 fd_batch_set_stage(batch
, FD_STAGE_BLIT
);
700 fd_log_stream(batch
, stream
, util_dump_blit_info(stream
, info
));
704 if ((info
->src
.resource
->target
== PIPE_BUFFER
) &&
705 (info
->dst
.resource
->target
== PIPE_BUFFER
)) {
706 assert(fd_resource(info
->src
.resource
)->layout
.tile_mode
== TILE6_LINEAR
);
707 assert(fd_resource(info
->dst
.resource
)->layout
.tile_mode
== TILE6_LINEAR
);
708 fd_log(batch
, "START BLIT (BUFFER)");
709 emit_blit_buffer(ctx
, batch
->draw
, info
);
710 fd_log(batch
, "END BLIT (BUFFER)");
712 /* I don't *think* we need to handle blits between buffer <-> !buffer */
713 debug_assert(info
->src
.resource
->target
!= PIPE_BUFFER
);
714 debug_assert(info
->dst
.resource
->target
!= PIPE_BUFFER
);
715 fd_log(batch
, "START BLIT (TEXTURE)");
716 emit_blit_or_clear_texture(ctx
, batch
->draw
, info
, NULL
);
717 fd_log(batch
, "END BLIT (TEXTURE)");
720 fd6_event_write(batch
, batch
->draw
, PC_CCU_FLUSH_COLOR_TS
, true);
721 fd6_event_write(batch
, batch
->draw
, PC_CCU_FLUSH_DEPTH_TS
, true);
722 fd6_event_write(batch
, batch
->draw
, CACHE_FLUSH_TS
, true);
723 fd6_cache_inv(batch
, batch
->draw
);
725 fd_resource(info
->dst
.resource
)->valid
= true;
726 batch
->needs_flush
= true;
728 fd_batch_flush(batch
);
729 fd_batch_reference(&batch
, NULL
);
735 * Re-written z/s blits can still fail for various reasons (for example MSAA).
736 * But we want to do the fallback blit with the re-written pipe_blit_info,
737 * in particular as u_blitter cannot blit stencil. So handle the fallback
738 * ourself and never "fail".
741 do_rewritten_blit(struct fd_context
*ctx
, const struct pipe_blit_info
*info
)
743 bool success
= handle_rgba_blit(ctx
, info
);
745 success
= fd_blitter_blit(ctx
, info
);
746 debug_assert(success
); /* fallback should never fail! */
751 * Handle depth/stencil blits either via u_blitter and/or re-writing the
752 * blit into an equivilant format that we can handle
755 handle_zs_blit(struct fd_context
*ctx
, const struct pipe_blit_info
*info
)
757 struct pipe_blit_info blit
= *info
;
760 fprintf(stderr
, "---- handle_zs_blit: ");
761 util_dump_blit_info(stderr
, info
);
762 fprintf(stderr
, "\ndst resource: ");
763 util_dump_resource(stderr
, info
->dst
.resource
);
764 fprintf(stderr
, "\nsrc resource: ");
765 util_dump_resource(stderr
, info
->src
.resource
);
766 fprintf(stderr
, "\n");
769 switch (info
->dst
.format
) {
770 case PIPE_FORMAT_S8_UINT
:
771 debug_assert(info
->mask
== PIPE_MASK_S
);
772 blit
.mask
= PIPE_MASK_R
;
773 blit
.src
.format
= PIPE_FORMAT_R8_UINT
;
774 blit
.dst
.format
= PIPE_FORMAT_R8_UINT
;
775 return do_rewritten_blit(ctx
, &blit
);
777 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
778 if (info
->mask
& PIPE_MASK_Z
) {
779 blit
.mask
= PIPE_MASK_R
;
780 blit
.src
.format
= PIPE_FORMAT_R32_FLOAT
;
781 blit
.dst
.format
= PIPE_FORMAT_R32_FLOAT
;
782 do_rewritten_blit(ctx
, &blit
);
785 if (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 blit
.src
.resource
= &fd_resource(info
->src
.resource
)->stencil
->base
;
790 blit
.dst
.resource
= &fd_resource(info
->dst
.resource
)->stencil
->base
;
791 do_rewritten_blit(ctx
, &blit
);
796 case PIPE_FORMAT_Z16_UNORM
:
797 blit
.mask
= PIPE_MASK_R
;
798 blit
.src
.format
= PIPE_FORMAT_R16_UNORM
;
799 blit
.dst
.format
= PIPE_FORMAT_R16_UNORM
;
800 return do_rewritten_blit(ctx
, &blit
);
802 case PIPE_FORMAT_Z32_UNORM
:
803 case PIPE_FORMAT_Z32_FLOAT
:
804 debug_assert(info
->mask
== PIPE_MASK_Z
);
805 blit
.mask
= PIPE_MASK_R
;
806 blit
.src
.format
= PIPE_FORMAT_R32_UINT
;
807 blit
.dst
.format
= PIPE_FORMAT_R32_UINT
;
808 return do_rewritten_blit(ctx
, &blit
);
810 case PIPE_FORMAT_Z24X8_UNORM
:
811 case PIPE_FORMAT_Z24_UNORM_S8_UINT
:
813 if (info
->mask
& PIPE_MASK_Z
)
814 blit
.mask
|= PIPE_MASK_R
| PIPE_MASK_G
| PIPE_MASK_B
;
815 if (info
->mask
& PIPE_MASK_S
)
816 blit
.mask
|= PIPE_MASK_A
;
817 blit
.src
.format
= PIPE_FORMAT_Z24_UNORM_S8_UINT_AS_R8G8B8A8
;
818 blit
.dst
.format
= PIPE_FORMAT_Z24_UNORM_S8_UINT_AS_R8G8B8A8
;
819 return fd_blitter_blit(ctx
, &blit
);
827 handle_compressed_blit(struct fd_context
*ctx
, const struct pipe_blit_info
*info
)
829 struct pipe_blit_info blit
= *info
;
832 fprintf(stderr
, "---- handle_compressed_blit: ");
833 util_dump_blit_info(stderr
, info
);
834 fprintf(stderr
, "\ndst resource: ");
835 util_dump_resource(stderr
, info
->dst
.resource
);
836 fprintf(stderr
, "\nsrc resource: ");
837 util_dump_resource(stderr
, info
->src
.resource
);
838 fprintf(stderr
, "\n");
841 if (info
->src
.format
!= info
->dst
.format
)
842 return fd_blitter_blit(ctx
, info
);
844 if (util_format_get_blocksize(info
->src
.format
) == 8) {
845 blit
.src
.format
= blit
.dst
.format
= PIPE_FORMAT_R16G16B16A16_UINT
;
847 debug_assert(util_format_get_blocksize(info
->src
.format
) == 16);
848 blit
.src
.format
= blit
.dst
.format
= PIPE_FORMAT_R32G32B32A32_UINT
;
851 int bw
= util_format_get_blockwidth(info
->src
.format
);
852 int bh
= util_format_get_blockheight(info
->src
.format
);
854 /* NOTE: x/y *must* be aligned to block boundary (ie. in
855 * glCompressedTexSubImage2D()) but width/height may not
859 debug_assert((blit
.src
.box
.x
% bw
) == 0);
860 debug_assert((blit
.src
.box
.y
% bh
) == 0);
862 blit
.src
.box
.x
/= bw
;
863 blit
.src
.box
.y
/= bh
;
864 blit
.src
.box
.width
= DIV_ROUND_UP(blit
.src
.box
.width
, bw
);
865 blit
.src
.box
.height
= DIV_ROUND_UP(blit
.src
.box
.height
, bh
);
867 debug_assert((blit
.dst
.box
.x
% bw
) == 0);
868 debug_assert((blit
.dst
.box
.y
% bh
) == 0);
870 blit
.dst
.box
.x
/= bw
;
871 blit
.dst
.box
.y
/= bh
;
872 blit
.dst
.box
.width
= DIV_ROUND_UP(blit
.dst
.box
.width
, bw
);
873 blit
.dst
.box
.height
= DIV_ROUND_UP(blit
.dst
.box
.height
, bh
);
875 return do_rewritten_blit(ctx
, &blit
);
879 fd6_blit(struct fd_context
*ctx
, const struct pipe_blit_info
*info
)
881 if (info
->mask
& PIPE_MASK_ZS
)
882 return handle_zs_blit(ctx
, info
);
883 if (util_format_is_compressed(info
->src
.format
) ||
884 util_format_is_compressed(info
->dst
.format
))
885 return handle_compressed_blit(ctx
, info
);
887 return handle_rgba_blit(ctx
, info
);
891 fd6_blitter_init(struct pipe_context
*pctx
)
893 if (fd_mesa_debug
& FD_DBG_NOBLIT
)
896 fd_context(pctx
)->blit
= fd6_blit
;
900 fd6_tile_mode(const struct pipe_resource
*tmpl
)
902 /* if the mipmap level 0 is still too small to be tiled, then don't
905 if (fd_resource_level_linear(tmpl
, 0))
908 /* basically just has to be a format we can blit, so uploads/downloads
909 * via linear staging buffer works:
911 if (ok_format(tmpl
->format
))