2 * Copyright 2012 Advanced Micro Devices, Inc.
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
24 * Christian König <christian.koenig@amd.com>
28 #include "si_shader.h"
30 #include "radeon/r600_cs.h"
32 #include "util/u_dual_blend.h"
33 #include "util/u_format.h"
34 #include "util/u_format_s3tc.h"
35 #include "util/u_memory.h"
36 #include "util/u_pstipple.h"
37 #include "util/u_resource.h"
39 /* Initialize an external atom (owned by ../radeon). */
41 si_init_external_atom(struct si_context
*sctx
, struct r600_atom
*atom
,
42 struct r600_atom
**list_elem
)
44 atom
->id
= list_elem
- sctx
->atoms
.array
+ 1;
48 /* Initialize an atom owned by radeonsi. */
49 void si_init_atom(struct si_context
*sctx
, struct r600_atom
*atom
,
50 struct r600_atom
**list_elem
,
51 void (*emit_func
)(struct si_context
*ctx
, struct r600_atom
*state
))
53 atom
->emit
= (void*)emit_func
;
54 atom
->id
= list_elem
- sctx
->atoms
.array
+ 1; /* index+1 in the atom array */
58 unsigned si_array_mode(unsigned mode
)
61 case RADEON_SURF_MODE_LINEAR_ALIGNED
:
62 return V_009910_ARRAY_LINEAR_ALIGNED
;
63 case RADEON_SURF_MODE_1D
:
64 return V_009910_ARRAY_1D_TILED_THIN1
;
65 case RADEON_SURF_MODE_2D
:
66 return V_009910_ARRAY_2D_TILED_THIN1
;
68 case RADEON_SURF_MODE_LINEAR
:
69 return V_009910_ARRAY_LINEAR_GENERAL
;
73 uint32_t si_num_banks(struct si_screen
*sscreen
, struct r600_texture
*tex
)
75 if (sscreen
->b
.chip_class
>= CIK
&&
76 sscreen
->b
.info
.cik_macrotile_mode_array_valid
) {
77 unsigned index
, tileb
;
79 tileb
= 8 * 8 * tex
->surface
.bpe
;
80 tileb
= MIN2(tex
->surface
.tile_split
, tileb
);
82 for (index
= 0; tileb
> 64; index
++) {
87 return (sscreen
->b
.info
.cik_macrotile_mode_array
[index
] >> 6) & 0x3;
90 if (sscreen
->b
.chip_class
== SI
&&
91 sscreen
->b
.info
.si_tile_mode_array_valid
) {
92 /* Don't use stencil_tiling_index, because num_banks is always
93 * read from the depth mode. */
94 unsigned tile_mode_index
= tex
->surface
.tiling_index
[0];
95 assert(tile_mode_index
< 32);
97 return G_009910_NUM_BANKS(sscreen
->b
.info
.si_tile_mode_array
[tile_mode_index
]);
101 switch (sscreen
->b
.info
.r600_num_banks
) {
103 return V_02803C_ADDR_SURF_2_BANK
;
105 return V_02803C_ADDR_SURF_4_BANK
;
108 return V_02803C_ADDR_SURF_8_BANK
;
110 return V_02803C_ADDR_SURF_16_BANK
;
114 unsigned cik_tile_split(unsigned tile_split
)
116 switch (tile_split
) {
118 tile_split
= V_028040_ADDR_SURF_TILE_SPLIT_64B
;
121 tile_split
= V_028040_ADDR_SURF_TILE_SPLIT_128B
;
124 tile_split
= V_028040_ADDR_SURF_TILE_SPLIT_256B
;
127 tile_split
= V_028040_ADDR_SURF_TILE_SPLIT_512B
;
131 tile_split
= V_028040_ADDR_SURF_TILE_SPLIT_1KB
;
134 tile_split
= V_028040_ADDR_SURF_TILE_SPLIT_2KB
;
137 tile_split
= V_028040_ADDR_SURF_TILE_SPLIT_4KB
;
143 unsigned cik_macro_tile_aspect(unsigned macro_tile_aspect
)
145 switch (macro_tile_aspect
) {
148 macro_tile_aspect
= V_02803C_ADDR_SURF_MACRO_ASPECT_1
;
151 macro_tile_aspect
= V_02803C_ADDR_SURF_MACRO_ASPECT_2
;
154 macro_tile_aspect
= V_02803C_ADDR_SURF_MACRO_ASPECT_4
;
157 macro_tile_aspect
= V_02803C_ADDR_SURF_MACRO_ASPECT_8
;
160 return macro_tile_aspect
;
163 unsigned cik_bank_wh(unsigned bankwh
)
168 bankwh
= V_02803C_ADDR_SURF_BANK_WIDTH_1
;
171 bankwh
= V_02803C_ADDR_SURF_BANK_WIDTH_2
;
174 bankwh
= V_02803C_ADDR_SURF_BANK_WIDTH_4
;
177 bankwh
= V_02803C_ADDR_SURF_BANK_WIDTH_8
;
183 unsigned cik_db_pipe_config(struct si_screen
*sscreen
, unsigned tile_mode
)
185 if (sscreen
->b
.info
.si_tile_mode_array_valid
) {
186 uint32_t gb_tile_mode
= sscreen
->b
.info
.si_tile_mode_array
[tile_mode
];
188 return G_009910_PIPE_CONFIG(gb_tile_mode
);
191 /* This is probably broken for a lot of chips, but it's only used
192 * if the kernel cannot return the tile mode array for CIK. */
193 switch (sscreen
->b
.info
.num_tile_pipes
) {
195 return V_02803C_X_ADDR_SURF_P16_32X32_16X16
;
197 return V_02803C_X_ADDR_SURF_P8_32X32_16X16
;
200 if (sscreen
->b
.info
.num_render_backends
== 4)
201 return V_02803C_X_ADDR_SURF_P4_16X16
;
203 return V_02803C_X_ADDR_SURF_P4_8X16
;
205 return V_02803C_ADDR_SURF_P2
;
209 static unsigned si_map_swizzle(unsigned swizzle
)
212 case UTIL_FORMAT_SWIZZLE_Y
:
213 return V_008F0C_SQ_SEL_Y
;
214 case UTIL_FORMAT_SWIZZLE_Z
:
215 return V_008F0C_SQ_SEL_Z
;
216 case UTIL_FORMAT_SWIZZLE_W
:
217 return V_008F0C_SQ_SEL_W
;
218 case UTIL_FORMAT_SWIZZLE_0
:
219 return V_008F0C_SQ_SEL_0
;
220 case UTIL_FORMAT_SWIZZLE_1
:
221 return V_008F0C_SQ_SEL_1
;
222 default: /* UTIL_FORMAT_SWIZZLE_X */
223 return V_008F0C_SQ_SEL_X
;
227 static uint32_t S_FIXED(float value
, uint32_t frac_bits
)
229 return value
* (1 << frac_bits
);
232 /* 12.4 fixed-point */
233 static unsigned si_pack_float_12p4(float x
)
236 x
>= 4096 ? 0xffff : x
* 16;
240 * Inferred framebuffer and blender state.
242 * One of the reasons CB_TARGET_MASK must be derived from the framebuffer state
244 * - The blend state mask is 0xf most of the time.
245 * - The COLOR1 format isn't INVALID because of possible dual-source blending,
246 * so COLOR1 is enabled pretty much all the time.
247 * So CB_TARGET_MASK is the only register that can disable COLOR1.
249 * Another reason is to avoid a hang with dual source blending.
251 static void si_emit_cb_render_state(struct si_context
*sctx
, struct r600_atom
*atom
)
253 struct radeon_winsys_cs
*cs
= sctx
->b
.gfx
.cs
;
254 struct si_state_blend
*blend
= sctx
->queued
.named
.blend
;
255 uint32_t cb_target_mask
= 0, i
;
257 for (i
= 0; i
< sctx
->framebuffer
.state
.nr_cbufs
; i
++)
258 if (sctx
->framebuffer
.state
.cbufs
[i
])
259 cb_target_mask
|= 0xf << (4*i
);
262 cb_target_mask
&= blend
->cb_target_mask
;
264 /* Avoid a hang that happens when dual source blending is enabled
265 * but there is not enough color outputs. This is undefined behavior,
266 * so disable color writes completely.
268 * Reproducible with Unigine Heaven 4.0 and drirc missing.
270 if (blend
&& blend
->dual_src_blend
&&
271 sctx
->ps_shader
.cso
&&
272 (sctx
->ps_shader
.cso
->info
.colors_written
& 0x3) != 0x3)
275 radeon_set_context_reg(cs
, R_028238_CB_TARGET_MASK
, cb_target_mask
);
277 /* STONEY-specific register settings. */
278 if (sctx
->b
.family
== CHIP_STONEY
) {
279 unsigned spi_shader_col_format
=
280 sctx
->ps_shader
.cso
?
281 sctx
->ps_shader
.current
->key
.ps
.epilog
.spi_shader_col_format
: 0;
282 unsigned sx_ps_downconvert
= 0;
283 unsigned sx_blend_opt_epsilon
= 0;
284 unsigned sx_blend_opt_control
= 0;
286 for (i
= 0; i
< sctx
->framebuffer
.state
.nr_cbufs
; i
++) {
287 struct r600_surface
*surf
=
288 (struct r600_surface
*)sctx
->framebuffer
.state
.cbufs
[i
];
289 unsigned format
, swap
, spi_format
, colormask
;
290 bool has_alpha
, has_rgb
;
295 format
= G_028C70_FORMAT(surf
->cb_color_info
);
296 swap
= G_028C70_COMP_SWAP(surf
->cb_color_info
);
297 spi_format
= (spi_shader_col_format
>> (i
* 4)) & 0xf;
298 colormask
= (cb_target_mask
>> (i
* 4)) & 0xf;
300 /* Set if RGB and A are present. */
301 has_alpha
= !G_028C74_FORCE_DST_ALPHA_1(surf
->cb_color_attrib
);
303 if (format
== V_028C70_COLOR_8
||
304 format
== V_028C70_COLOR_16
||
305 format
== V_028C70_COLOR_32
)
306 has_rgb
= !has_alpha
;
310 /* Check the colormask and export format. */
311 if (!(colormask
& (PIPE_MASK_RGBA
& ~PIPE_MASK_A
)))
313 if (!(colormask
& PIPE_MASK_A
))
316 if (spi_format
== V_028714_SPI_SHADER_ZERO
) {
321 /* Disable value checking for disabled channels. */
323 sx_blend_opt_control
|= S_02875C_MRT0_COLOR_OPT_DISABLE(1) << (i
* 4);
325 sx_blend_opt_control
|= S_02875C_MRT0_ALPHA_OPT_DISABLE(1) << (i
* 4);
327 /* Enable down-conversion for 32bpp and smaller formats. */
329 case V_028C70_COLOR_8
:
330 case V_028C70_COLOR_8_8
:
331 case V_028C70_COLOR_8_8_8_8
:
332 /* For 1 and 2-channel formats, use the superset thereof. */
333 if (spi_format
== V_028714_SPI_SHADER_FP16_ABGR
||
334 spi_format
== V_028714_SPI_SHADER_UINT16_ABGR
||
335 spi_format
== V_028714_SPI_SHADER_SINT16_ABGR
) {
336 sx_ps_downconvert
|= V_028754_SX_RT_EXPORT_8_8_8_8
<< (i
* 4);
337 sx_blend_opt_epsilon
|= V_028758_8BIT_FORMAT
<< (i
* 4);
341 case V_028C70_COLOR_5_6_5
:
342 if (spi_format
== V_028714_SPI_SHADER_FP16_ABGR
) {
343 sx_ps_downconvert
|= V_028754_SX_RT_EXPORT_5_6_5
<< (i
* 4);
344 sx_blend_opt_epsilon
|= V_028758_6BIT_FORMAT
<< (i
* 4);
348 case V_028C70_COLOR_1_5_5_5
:
349 if (spi_format
== V_028714_SPI_SHADER_FP16_ABGR
) {
350 sx_ps_downconvert
|= V_028754_SX_RT_EXPORT_1_5_5_5
<< (i
* 4);
351 sx_blend_opt_epsilon
|= V_028758_5BIT_FORMAT
<< (i
* 4);
355 case V_028C70_COLOR_4_4_4_4
:
356 if (spi_format
== V_028714_SPI_SHADER_FP16_ABGR
) {
357 sx_ps_downconvert
|= V_028754_SX_RT_EXPORT_4_4_4_4
<< (i
* 4);
358 sx_blend_opt_epsilon
|= V_028758_4BIT_FORMAT
<< (i
* 4);
362 case V_028C70_COLOR_32
:
363 if (swap
== V_0280A0_SWAP_STD
&&
364 spi_format
== V_028714_SPI_SHADER_32_R
)
365 sx_ps_downconvert
|= V_028754_SX_RT_EXPORT_32_R
<< (i
* 4);
366 else if (swap
== V_0280A0_SWAP_ALT_REV
&&
367 spi_format
== V_028714_SPI_SHADER_32_AR
)
368 sx_ps_downconvert
|= V_028754_SX_RT_EXPORT_32_A
<< (i
* 4);
371 case V_028C70_COLOR_16
:
372 case V_028C70_COLOR_16_16
:
373 /* For 1-channel formats, use the superset thereof. */
374 if (spi_format
== V_028714_SPI_SHADER_UNORM16_ABGR
||
375 spi_format
== V_028714_SPI_SHADER_SNORM16_ABGR
||
376 spi_format
== V_028714_SPI_SHADER_UINT16_ABGR
||
377 spi_format
== V_028714_SPI_SHADER_SINT16_ABGR
) {
378 if (swap
== V_0280A0_SWAP_STD
||
379 swap
== V_0280A0_SWAP_STD_REV
)
380 sx_ps_downconvert
|= V_028754_SX_RT_EXPORT_16_16_GR
<< (i
* 4);
382 sx_ps_downconvert
|= V_028754_SX_RT_EXPORT_16_16_AR
<< (i
* 4);
386 case V_028C70_COLOR_10_11_11
:
387 if (spi_format
== V_028714_SPI_SHADER_FP16_ABGR
) {
388 sx_ps_downconvert
|= V_028754_SX_RT_EXPORT_10_11_11
<< (i
* 4);
389 sx_blend_opt_epsilon
|= V_028758_11BIT_FORMAT
<< (i
* 4);
393 case V_028C70_COLOR_2_10_10_10
:
394 if (spi_format
== V_028714_SPI_SHADER_FP16_ABGR
) {
395 sx_ps_downconvert
|= V_028754_SX_RT_EXPORT_2_10_10_10
<< (i
* 4);
396 sx_blend_opt_epsilon
|= V_028758_10BIT_FORMAT
<< (i
* 4);
402 if (sctx
->screen
->b
.debug_flags
& DBG_NO_RB_PLUS
) {
403 sx_ps_downconvert
= 0;
404 sx_blend_opt_epsilon
= 0;
405 sx_blend_opt_control
= 0;
408 radeon_set_context_reg_seq(cs
, R_028754_SX_PS_DOWNCONVERT
, 3);
409 radeon_emit(cs
, sx_ps_downconvert
); /* R_028754_SX_PS_DOWNCONVERT */
410 radeon_emit(cs
, sx_blend_opt_epsilon
); /* R_028758_SX_BLEND_OPT_EPSILON */
411 radeon_emit(cs
, sx_blend_opt_control
); /* R_02875C_SX_BLEND_OPT_CONTROL */
419 static uint32_t si_translate_blend_function(int blend_func
)
421 switch (blend_func
) {
423 return V_028780_COMB_DST_PLUS_SRC
;
424 case PIPE_BLEND_SUBTRACT
:
425 return V_028780_COMB_SRC_MINUS_DST
;
426 case PIPE_BLEND_REVERSE_SUBTRACT
:
427 return V_028780_COMB_DST_MINUS_SRC
;
429 return V_028780_COMB_MIN_DST_SRC
;
431 return V_028780_COMB_MAX_DST_SRC
;
433 R600_ERR("Unknown blend function %d\n", blend_func
);
440 static uint32_t si_translate_blend_factor(int blend_fact
)
442 switch (blend_fact
) {
443 case PIPE_BLENDFACTOR_ONE
:
444 return V_028780_BLEND_ONE
;
445 case PIPE_BLENDFACTOR_SRC_COLOR
:
446 return V_028780_BLEND_SRC_COLOR
;
447 case PIPE_BLENDFACTOR_SRC_ALPHA
:
448 return V_028780_BLEND_SRC_ALPHA
;
449 case PIPE_BLENDFACTOR_DST_ALPHA
:
450 return V_028780_BLEND_DST_ALPHA
;
451 case PIPE_BLENDFACTOR_DST_COLOR
:
452 return V_028780_BLEND_DST_COLOR
;
453 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
:
454 return V_028780_BLEND_SRC_ALPHA_SATURATE
;
455 case PIPE_BLENDFACTOR_CONST_COLOR
:
456 return V_028780_BLEND_CONSTANT_COLOR
;
457 case PIPE_BLENDFACTOR_CONST_ALPHA
:
458 return V_028780_BLEND_CONSTANT_ALPHA
;
459 case PIPE_BLENDFACTOR_ZERO
:
460 return V_028780_BLEND_ZERO
;
461 case PIPE_BLENDFACTOR_INV_SRC_COLOR
:
462 return V_028780_BLEND_ONE_MINUS_SRC_COLOR
;
463 case PIPE_BLENDFACTOR_INV_SRC_ALPHA
:
464 return V_028780_BLEND_ONE_MINUS_SRC_ALPHA
;
465 case PIPE_BLENDFACTOR_INV_DST_ALPHA
:
466 return V_028780_BLEND_ONE_MINUS_DST_ALPHA
;
467 case PIPE_BLENDFACTOR_INV_DST_COLOR
:
468 return V_028780_BLEND_ONE_MINUS_DST_COLOR
;
469 case PIPE_BLENDFACTOR_INV_CONST_COLOR
:
470 return V_028780_BLEND_ONE_MINUS_CONSTANT_COLOR
;
471 case PIPE_BLENDFACTOR_INV_CONST_ALPHA
:
472 return V_028780_BLEND_ONE_MINUS_CONSTANT_ALPHA
;
473 case PIPE_BLENDFACTOR_SRC1_COLOR
:
474 return V_028780_BLEND_SRC1_COLOR
;
475 case PIPE_BLENDFACTOR_SRC1_ALPHA
:
476 return V_028780_BLEND_SRC1_ALPHA
;
477 case PIPE_BLENDFACTOR_INV_SRC1_COLOR
:
478 return V_028780_BLEND_INV_SRC1_COLOR
;
479 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA
:
480 return V_028780_BLEND_INV_SRC1_ALPHA
;
482 R600_ERR("Bad blend factor %d not supported!\n", blend_fact
);
489 static uint32_t si_translate_blend_opt_function(int blend_func
)
491 switch (blend_func
) {
493 return V_028760_OPT_COMB_ADD
;
494 case PIPE_BLEND_SUBTRACT
:
495 return V_028760_OPT_COMB_SUBTRACT
;
496 case PIPE_BLEND_REVERSE_SUBTRACT
:
497 return V_028760_OPT_COMB_REVSUBTRACT
;
499 return V_028760_OPT_COMB_MIN
;
501 return V_028760_OPT_COMB_MAX
;
503 return V_028760_OPT_COMB_BLEND_DISABLED
;
507 static uint32_t si_translate_blend_opt_factor(int blend_fact
, bool is_alpha
)
509 switch (blend_fact
) {
510 case PIPE_BLENDFACTOR_ZERO
:
511 return V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_ALL
;
512 case PIPE_BLENDFACTOR_ONE
:
513 return V_028760_BLEND_OPT_PRESERVE_ALL_IGNORE_NONE
;
514 case PIPE_BLENDFACTOR_SRC_COLOR
:
515 return is_alpha
? V_028760_BLEND_OPT_PRESERVE_A1_IGNORE_A0
516 : V_028760_BLEND_OPT_PRESERVE_C1_IGNORE_C0
;
517 case PIPE_BLENDFACTOR_INV_SRC_COLOR
:
518 return is_alpha
? V_028760_BLEND_OPT_PRESERVE_A0_IGNORE_A1
519 : V_028760_BLEND_OPT_PRESERVE_C0_IGNORE_C1
;
520 case PIPE_BLENDFACTOR_SRC_ALPHA
:
521 return V_028760_BLEND_OPT_PRESERVE_A1_IGNORE_A0
;
522 case PIPE_BLENDFACTOR_INV_SRC_ALPHA
:
523 return V_028760_BLEND_OPT_PRESERVE_A0_IGNORE_A1
;
524 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
:
525 return is_alpha
? V_028760_BLEND_OPT_PRESERVE_ALL_IGNORE_NONE
526 : V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_A0
;
528 return V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_NONE
;
533 * Get rid of DST in the blend factors by commuting the operands:
534 * func(src * DST, dst * 0) ---> func(src * 0, dst * SRC)
536 static void si_blend_remove_dst(unsigned *func
, unsigned *src_factor
,
537 unsigned *dst_factor
, unsigned expected_dst
,
538 unsigned replacement_src
)
540 if (*src_factor
== expected_dst
&&
541 *dst_factor
== PIPE_BLENDFACTOR_ZERO
) {
542 *src_factor
= PIPE_BLENDFACTOR_ZERO
;
543 *dst_factor
= replacement_src
;
545 /* Commuting the operands requires reversing subtractions. */
546 if (*func
== PIPE_BLEND_SUBTRACT
)
547 *func
= PIPE_BLEND_REVERSE_SUBTRACT
;
548 else if (*func
== PIPE_BLEND_REVERSE_SUBTRACT
)
549 *func
= PIPE_BLEND_SUBTRACT
;
553 static bool si_blend_factor_uses_dst(unsigned factor
)
555 return factor
== PIPE_BLENDFACTOR_DST_COLOR
||
556 factor
== PIPE_BLENDFACTOR_DST_ALPHA
||
557 factor
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
558 factor
== PIPE_BLENDFACTOR_INV_DST_ALPHA
||
559 factor
== PIPE_BLENDFACTOR_INV_DST_COLOR
;
562 static void *si_create_blend_state_mode(struct pipe_context
*ctx
,
563 const struct pipe_blend_state
*state
,
566 struct si_context
*sctx
= (struct si_context
*)ctx
;
567 struct si_state_blend
*blend
= CALLOC_STRUCT(si_state_blend
);
568 struct si_pm4_state
*pm4
= &blend
->pm4
;
569 uint32_t sx_mrt_blend_opt
[8] = {0};
570 uint32_t color_control
= 0;
575 blend
->alpha_to_coverage
= state
->alpha_to_coverage
;
576 blend
->alpha_to_one
= state
->alpha_to_one
;
577 blend
->dual_src_blend
= util_blend_state_is_dual(state
, 0);
579 if (state
->logicop_enable
) {
580 color_control
|= S_028808_ROP3(state
->logicop_func
| (state
->logicop_func
<< 4));
582 color_control
|= S_028808_ROP3(0xcc);
585 si_pm4_set_reg(pm4
, R_028B70_DB_ALPHA_TO_MASK
,
586 S_028B70_ALPHA_TO_MASK_ENABLE(state
->alpha_to_coverage
) |
587 S_028B70_ALPHA_TO_MASK_OFFSET0(2) |
588 S_028B70_ALPHA_TO_MASK_OFFSET1(2) |
589 S_028B70_ALPHA_TO_MASK_OFFSET2(2) |
590 S_028B70_ALPHA_TO_MASK_OFFSET3(2));
592 if (state
->alpha_to_coverage
)
593 blend
->need_src_alpha_4bit
|= 0xf;
595 blend
->cb_target_mask
= 0;
596 for (int i
= 0; i
< 8; i
++) {
597 /* state->rt entries > 0 only written if independent blending */
598 const int j
= state
->independent_blend_enable
? i
: 0;
600 unsigned eqRGB
= state
->rt
[j
].rgb_func
;
601 unsigned srcRGB
= state
->rt
[j
].rgb_src_factor
;
602 unsigned dstRGB
= state
->rt
[j
].rgb_dst_factor
;
603 unsigned eqA
= state
->rt
[j
].alpha_func
;
604 unsigned srcA
= state
->rt
[j
].alpha_src_factor
;
605 unsigned dstA
= state
->rt
[j
].alpha_dst_factor
;
607 unsigned srcRGB_opt
, dstRGB_opt
, srcA_opt
, dstA_opt
;
608 unsigned blend_cntl
= 0;
610 sx_mrt_blend_opt
[i
] =
611 S_028760_COLOR_COMB_FCN(V_028760_OPT_COMB_BLEND_DISABLED
) |
612 S_028760_ALPHA_COMB_FCN(V_028760_OPT_COMB_BLEND_DISABLED
);
614 if (!state
->rt
[j
].colormask
)
617 /* cb_render_state will disable unused ones */
618 blend
->cb_target_mask
|= state
->rt
[j
].colormask
<< (4 * i
);
620 if (!state
->rt
[j
].blend_enable
) {
621 si_pm4_set_reg(pm4
, R_028780_CB_BLEND0_CONTROL
+ i
* 4, blend_cntl
);
625 /* Blending optimizations for Stoney.
626 * These transformations don't change the behavior.
628 * First, get rid of DST in the blend factors:
629 * func(src * DST, dst * 0) ---> func(src * 0, dst * SRC)
631 si_blend_remove_dst(&eqRGB
, &srcRGB
, &dstRGB
,
632 PIPE_BLENDFACTOR_DST_COLOR
,
633 PIPE_BLENDFACTOR_SRC_COLOR
);
634 si_blend_remove_dst(&eqA
, &srcA
, &dstA
,
635 PIPE_BLENDFACTOR_DST_COLOR
,
636 PIPE_BLENDFACTOR_SRC_COLOR
);
637 si_blend_remove_dst(&eqA
, &srcA
, &dstA
,
638 PIPE_BLENDFACTOR_DST_ALPHA
,
639 PIPE_BLENDFACTOR_SRC_ALPHA
);
641 /* Look up the ideal settings from tables. */
642 srcRGB_opt
= si_translate_blend_opt_factor(srcRGB
, false);
643 dstRGB_opt
= si_translate_blend_opt_factor(dstRGB
, false);
644 srcA_opt
= si_translate_blend_opt_factor(srcA
, true);
645 dstA_opt
= si_translate_blend_opt_factor(dstA
, true);
647 /* Handle interdependencies. */
648 if (si_blend_factor_uses_dst(srcRGB
))
649 dstRGB_opt
= V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_NONE
;
650 if (si_blend_factor_uses_dst(srcA
))
651 dstA_opt
= V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_NONE
;
653 if (srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
&&
654 (dstRGB
== PIPE_BLENDFACTOR_ZERO
||
655 dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
656 dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
))
657 dstRGB_opt
= V_028760_BLEND_OPT_PRESERVE_NONE_IGNORE_A0
;
659 /* Set the final value. */
660 sx_mrt_blend_opt
[i
] =
661 S_028760_COLOR_SRC_OPT(srcRGB_opt
) |
662 S_028760_COLOR_DST_OPT(dstRGB_opt
) |
663 S_028760_COLOR_COMB_FCN(si_translate_blend_opt_function(eqRGB
)) |
664 S_028760_ALPHA_SRC_OPT(srcA_opt
) |
665 S_028760_ALPHA_DST_OPT(dstA_opt
) |
666 S_028760_ALPHA_COMB_FCN(si_translate_blend_opt_function(eqA
));
668 /* Set blend state. */
669 blend_cntl
|= S_028780_ENABLE(1);
670 blend_cntl
|= S_028780_COLOR_COMB_FCN(si_translate_blend_function(eqRGB
));
671 blend_cntl
|= S_028780_COLOR_SRCBLEND(si_translate_blend_factor(srcRGB
));
672 blend_cntl
|= S_028780_COLOR_DESTBLEND(si_translate_blend_factor(dstRGB
));
674 if (srcA
!= srcRGB
|| dstA
!= dstRGB
|| eqA
!= eqRGB
) {
675 blend_cntl
|= S_028780_SEPARATE_ALPHA_BLEND(1);
676 blend_cntl
|= S_028780_ALPHA_COMB_FCN(si_translate_blend_function(eqA
));
677 blend_cntl
|= S_028780_ALPHA_SRCBLEND(si_translate_blend_factor(srcA
));
678 blend_cntl
|= S_028780_ALPHA_DESTBLEND(si_translate_blend_factor(dstA
));
680 si_pm4_set_reg(pm4
, R_028780_CB_BLEND0_CONTROL
+ i
* 4, blend_cntl
);
682 blend
->blend_enable_4bit
|= 0xf << (i
* 4);
684 /* This is only important for formats without alpha. */
685 if (srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
686 dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
687 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
688 dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
689 srcRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
690 dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
)
691 blend
->need_src_alpha_4bit
|= 0xf << (i
* 4);
694 if (blend
->cb_target_mask
) {
695 color_control
|= S_028808_MODE(mode
);
697 color_control
|= S_028808_MODE(V_028808_CB_DISABLE
);
700 if (sctx
->b
.family
== CHIP_STONEY
) {
701 for (int i
= 0; i
< 8; i
++)
702 si_pm4_set_reg(pm4
, R_028760_SX_MRT0_BLEND_OPT
+ i
* 4,
703 sx_mrt_blend_opt
[i
]);
705 /* RB+ doesn't work with dual source blending, logic op, and RESOLVE. */
706 if (blend
->dual_src_blend
|| state
->logicop_enable
||
707 mode
== V_028808_CB_RESOLVE
)
708 color_control
|= S_028808_DISABLE_DUAL_QUAD(1);
711 si_pm4_set_reg(pm4
, R_028808_CB_COLOR_CONTROL
, color_control
);
715 static void *si_create_blend_state(struct pipe_context
*ctx
,
716 const struct pipe_blend_state
*state
)
718 return si_create_blend_state_mode(ctx
, state
, V_028808_CB_NORMAL
);
721 static void si_bind_blend_state(struct pipe_context
*ctx
, void *state
)
723 struct si_context
*sctx
= (struct si_context
*)ctx
;
724 si_pm4_bind_state(sctx
, blend
, (struct si_state_blend
*)state
);
725 si_mark_atom_dirty(sctx
, &sctx
->cb_render_state
);
728 static void si_delete_blend_state(struct pipe_context
*ctx
, void *state
)
730 struct si_context
*sctx
= (struct si_context
*)ctx
;
731 si_pm4_delete_state(sctx
, blend
, (struct si_state_blend
*)state
);
734 static void si_set_blend_color(struct pipe_context
*ctx
,
735 const struct pipe_blend_color
*state
)
737 struct si_context
*sctx
= (struct si_context
*)ctx
;
739 if (memcmp(&sctx
->blend_color
.state
, state
, sizeof(*state
)) == 0)
742 sctx
->blend_color
.state
= *state
;
743 si_mark_atom_dirty(sctx
, &sctx
->blend_color
.atom
);
746 static void si_emit_blend_color(struct si_context
*sctx
, struct r600_atom
*atom
)
748 struct radeon_winsys_cs
*cs
= sctx
->b
.gfx
.cs
;
750 radeon_set_context_reg_seq(cs
, R_028414_CB_BLEND_RED
, 4);
751 radeon_emit_array(cs
, (uint32_t*)sctx
->blend_color
.state
.color
, 4);
755 * Clipping, scissors and viewport
758 static void si_set_clip_state(struct pipe_context
*ctx
,
759 const struct pipe_clip_state
*state
)
761 struct si_context
*sctx
= (struct si_context
*)ctx
;
762 struct pipe_constant_buffer cb
;
764 if (memcmp(&sctx
->clip_state
.state
, state
, sizeof(*state
)) == 0)
767 sctx
->clip_state
.state
= *state
;
768 si_mark_atom_dirty(sctx
, &sctx
->clip_state
.atom
);
771 cb
.user_buffer
= state
->ucp
;
772 cb
.buffer_offset
= 0;
773 cb
.buffer_size
= 4*4*8;
774 ctx
->set_constant_buffer(ctx
, PIPE_SHADER_VERTEX
, SI_DRIVER_STATE_CONST_BUF
, &cb
);
775 pipe_resource_reference(&cb
.buffer
, NULL
);
778 static void si_emit_clip_state(struct si_context
*sctx
, struct r600_atom
*atom
)
780 struct radeon_winsys_cs
*cs
= sctx
->b
.gfx
.cs
;
782 radeon_set_context_reg_seq(cs
, R_0285BC_PA_CL_UCP_0_X
, 6*4);
783 radeon_emit_array(cs
, (uint32_t*)sctx
->clip_state
.state
.ucp
, 6*4);
786 #define SIX_BITS 0x3F
788 static void si_emit_clip_regs(struct si_context
*sctx
, struct r600_atom
*atom
)
790 struct radeon_winsys_cs
*cs
= sctx
->b
.gfx
.cs
;
791 struct tgsi_shader_info
*info
= si_get_vs_info(sctx
);
792 unsigned window_space
=
793 info
->properties
[TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION
];
794 unsigned clipdist_mask
=
795 info
->writes_clipvertex
? SIX_BITS
: info
->clipdist_writemask
;
797 radeon_set_context_reg(cs
, R_02881C_PA_CL_VS_OUT_CNTL
,
798 S_02881C_USE_VTX_POINT_SIZE(info
->writes_psize
) |
799 S_02881C_USE_VTX_EDGE_FLAG(info
->writes_edgeflag
) |
800 S_02881C_USE_VTX_RENDER_TARGET_INDX(info
->writes_layer
) |
801 S_02881C_USE_VTX_VIEWPORT_INDX(info
->writes_viewport_index
) |
802 S_02881C_VS_OUT_CCDIST0_VEC_ENA((clipdist_mask
& 0x0F) != 0) |
803 S_02881C_VS_OUT_CCDIST1_VEC_ENA((clipdist_mask
& 0xF0) != 0) |
804 S_02881C_VS_OUT_MISC_VEC_ENA(info
->writes_psize
||
805 info
->writes_edgeflag
||
806 info
->writes_layer
||
807 info
->writes_viewport_index
) |
808 S_02881C_VS_OUT_MISC_SIDE_BUS_ENA(1) |
809 (sctx
->queued
.named
.rasterizer
->clip_plane_enable
&
811 radeon_set_context_reg(cs
, R_028810_PA_CL_CLIP_CNTL
,
812 sctx
->queued
.named
.rasterizer
->pa_cl_clip_cntl
|
814 sctx
->queued
.named
.rasterizer
->clip_plane_enable
& SIX_BITS
) |
815 S_028810_CLIP_DISABLE(window_space
));
817 /* reuse needs to be set off if we write oViewport */
818 radeon_set_context_reg(cs
, R_028AB4_VGT_REUSE_OFF
,
819 S_028AB4_REUSE_OFF(info
->writes_viewport_index
));
822 static void si_set_scissor_states(struct pipe_context
*ctx
,
824 unsigned num_scissors
,
825 const struct pipe_scissor_state
*state
)
827 struct si_context
*sctx
= (struct si_context
*)ctx
;
830 for (i
= 0; i
< num_scissors
; i
++)
831 sctx
->scissors
.states
[start_slot
+ i
] = state
[i
];
833 if (!sctx
->queued
.named
.rasterizer
||
834 !sctx
->queued
.named
.rasterizer
->scissor_enable
)
837 sctx
->scissors
.dirty_mask
|= ((1 << num_scissors
) - 1) << start_slot
;
838 si_mark_atom_dirty(sctx
, &sctx
->scissors
.atom
);
841 static void si_get_scissor_from_viewport(struct pipe_viewport_state
*vp
,
842 struct pipe_scissor_state
*scissor
)
844 /* These must be signed, unlike pipe_scissor_state. */
845 int minx
, miny
, maxx
, maxy
, tmp
;
847 /* Convert (-1, -1) and (1, 1) from clip space into window space. */
848 minx
= -vp
->scale
[0] + vp
->translate
[0];
849 miny
= -vp
->scale
[1] + vp
->translate
[1];
850 maxx
= vp
->scale
[0] + vp
->translate
[0];
851 maxy
= vp
->scale
[1] + vp
->translate
[1];
853 /* r600_draw_rectangle sets this. Disable the scissor. */
854 if (minx
== -1 && miny
== -1 && maxx
== 1 && maxy
== 1) {
859 /* Handle inverted viewports. */
871 scissor
->minx
= CLAMP(minx
, 0, 16384);
872 scissor
->miny
= CLAMP(miny
, 0, 16384);
873 scissor
->maxx
= CLAMP(maxx
, 0, 16384);
874 scissor
->maxy
= CLAMP(maxy
, 0, 16384);
877 static void si_clip_scissor(struct pipe_scissor_state
*out
,
878 struct pipe_scissor_state
*clip
)
880 out
->minx
= MAX2(out
->minx
, clip
->minx
);
881 out
->miny
= MAX2(out
->miny
, clip
->miny
);
882 out
->maxx
= MIN2(out
->maxx
, clip
->maxx
);
883 out
->maxy
= MIN2(out
->maxy
, clip
->maxy
);
886 static void si_emit_one_scissor(struct radeon_winsys_cs
*cs
,
887 struct pipe_viewport_state
*vp
,
888 struct pipe_scissor_state
*scissor
)
890 struct pipe_scissor_state final
;
892 /* Since the guard band disables clipping, we have to clip per-pixel
895 si_get_scissor_from_viewport(vp
, &final
);
898 si_clip_scissor(&final
, scissor
);
900 radeon_emit(cs
, S_028250_TL_X(final
.minx
) |
901 S_028250_TL_Y(final
.miny
) |
902 S_028250_WINDOW_OFFSET_DISABLE(1));
903 radeon_emit(cs
, S_028254_BR_X(final
.maxx
) |
904 S_028254_BR_Y(final
.maxy
));
907 static void si_emit_scissors(struct si_context
*sctx
, struct r600_atom
*atom
)
909 struct radeon_winsys_cs
*cs
= sctx
->b
.gfx
.cs
;
910 struct pipe_scissor_state
*states
= sctx
->scissors
.states
;
911 unsigned mask
= sctx
->scissors
.dirty_mask
;
912 bool scissor_enable
= sctx
->queued
.named
.rasterizer
->scissor_enable
;
914 /* The simple case: Only 1 viewport is active. */
915 if (!si_get_vs_info(sctx
)->writes_viewport_index
) {
919 radeon_set_context_reg_seq(cs
, R_028250_PA_SC_VPORT_SCISSOR_0_TL
, 2);
920 si_emit_one_scissor(cs
, &sctx
->viewports
.states
[0],
921 scissor_enable
? &states
[0] : NULL
);
922 sctx
->scissors
.dirty_mask
&= ~1; /* clear one bit */
929 u_bit_scan_consecutive_range(&mask
, &start
, &count
);
931 radeon_set_context_reg_seq(cs
, R_028250_PA_SC_VPORT_SCISSOR_0_TL
+
932 start
* 4 * 2, count
* 2);
933 for (i
= start
; i
< start
+count
; i
++) {
934 si_emit_one_scissor(cs
, &sctx
->viewports
.states
[i
],
935 scissor_enable
? &states
[i
] : NULL
);
938 sctx
->scissors
.dirty_mask
= 0;
941 static void si_set_viewport_states(struct pipe_context
*ctx
,
943 unsigned num_viewports
,
944 const struct pipe_viewport_state
*state
)
946 struct si_context
*sctx
= (struct si_context
*)ctx
;
949 for (i
= 0; i
< num_viewports
; i
++)
950 sctx
->viewports
.states
[start_slot
+ i
] = state
[i
];
952 sctx
->viewports
.dirty_mask
|= ((1 << num_viewports
) - 1) << start_slot
;
953 sctx
->scissors
.dirty_mask
|= ((1 << num_viewports
) - 1) << start_slot
;
954 si_mark_atom_dirty(sctx
, &sctx
->viewports
.atom
);
955 si_mark_atom_dirty(sctx
, &sctx
->scissors
.atom
);
958 static void si_emit_viewports(struct si_context
*sctx
, struct r600_atom
*atom
)
960 struct radeon_winsys_cs
*cs
= sctx
->b
.gfx
.cs
;
961 struct pipe_viewport_state
*states
= sctx
->viewports
.states
;
962 unsigned mask
= sctx
->viewports
.dirty_mask
;
964 /* The simple case: Only 1 viewport is active. */
965 if (!si_get_vs_info(sctx
)->writes_viewport_index
) {
969 radeon_set_context_reg_seq(cs
, R_02843C_PA_CL_VPORT_XSCALE
, 6);
970 radeon_emit(cs
, fui(states
[0].scale
[0]));
971 radeon_emit(cs
, fui(states
[0].translate
[0]));
972 radeon_emit(cs
, fui(states
[0].scale
[1]));
973 radeon_emit(cs
, fui(states
[0].translate
[1]));
974 radeon_emit(cs
, fui(states
[0].scale
[2]));
975 radeon_emit(cs
, fui(states
[0].translate
[2]));
976 sctx
->viewports
.dirty_mask
&= ~1; /* clear one bit */
983 u_bit_scan_consecutive_range(&mask
, &start
, &count
);
985 radeon_set_context_reg_seq(cs
, R_02843C_PA_CL_VPORT_XSCALE
+
986 start
* 4 * 6, count
* 6);
987 for (i
= start
; i
< start
+count
; i
++) {
988 radeon_emit(cs
, fui(states
[i
].scale
[0]));
989 radeon_emit(cs
, fui(states
[i
].translate
[0]));
990 radeon_emit(cs
, fui(states
[i
].scale
[1]));
991 radeon_emit(cs
, fui(states
[i
].translate
[1]));
992 radeon_emit(cs
, fui(states
[i
].scale
[2]));
993 radeon_emit(cs
, fui(states
[i
].translate
[2]));
996 sctx
->viewports
.dirty_mask
= 0;
1000 * inferred state between framebuffer and rasterizer
1002 static void si_update_poly_offset_state(struct si_context
*sctx
)
1004 struct si_state_rasterizer
*rs
= sctx
->queued
.named
.rasterizer
;
1006 if (!rs
|| !rs
->uses_poly_offset
|| !sctx
->framebuffer
.state
.zsbuf
)
1009 switch (sctx
->framebuffer
.state
.zsbuf
->texture
->format
) {
1010 case PIPE_FORMAT_Z16_UNORM
:
1011 si_pm4_bind_state(sctx
, poly_offset
, &rs
->pm4_poly_offset
[0]);
1013 default: /* 24-bit */
1014 si_pm4_bind_state(sctx
, poly_offset
, &rs
->pm4_poly_offset
[1]);
1016 case PIPE_FORMAT_Z32_FLOAT
:
1017 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
1018 si_pm4_bind_state(sctx
, poly_offset
, &rs
->pm4_poly_offset
[2]);
1027 static uint32_t si_translate_fill(uint32_t func
)
1030 case PIPE_POLYGON_MODE_FILL
:
1031 return V_028814_X_DRAW_TRIANGLES
;
1032 case PIPE_POLYGON_MODE_LINE
:
1033 return V_028814_X_DRAW_LINES
;
1034 case PIPE_POLYGON_MODE_POINT
:
1035 return V_028814_X_DRAW_POINTS
;
1038 return V_028814_X_DRAW_POINTS
;
1042 static void *si_create_rs_state(struct pipe_context
*ctx
,
1043 const struct pipe_rasterizer_state
*state
)
1045 struct si_state_rasterizer
*rs
= CALLOC_STRUCT(si_state_rasterizer
);
1046 struct si_pm4_state
*pm4
= &rs
->pm4
;
1048 float psize_min
, psize_max
;
1054 rs
->scissor_enable
= state
->scissor
;
1055 rs
->two_side
= state
->light_twoside
;
1056 rs
->multisample_enable
= state
->multisample
;
1057 rs
->force_persample_interp
= state
->force_persample_interp
;
1058 rs
->clip_plane_enable
= state
->clip_plane_enable
;
1059 rs
->line_stipple_enable
= state
->line_stipple_enable
;
1060 rs
->poly_stipple_enable
= state
->poly_stipple_enable
;
1061 rs
->line_smooth
= state
->line_smooth
;
1062 rs
->poly_smooth
= state
->poly_smooth
;
1063 rs
->uses_poly_offset
= state
->offset_point
|| state
->offset_line
||
1065 rs
->clamp_fragment_color
= state
->clamp_fragment_color
;
1066 rs
->flatshade
= state
->flatshade
;
1067 rs
->sprite_coord_enable
= state
->sprite_coord_enable
;
1068 rs
->rasterizer_discard
= state
->rasterizer_discard
;
1069 rs
->pa_sc_line_stipple
= state
->line_stipple_enable
?
1070 S_028A0C_LINE_PATTERN(state
->line_stipple_pattern
) |
1071 S_028A0C_REPEAT_COUNT(state
->line_stipple_factor
) : 0;
1072 rs
->pa_cl_clip_cntl
=
1073 S_028810_PS_UCP_MODE(3) |
1074 S_028810_DX_CLIP_SPACE_DEF(state
->clip_halfz
) |
1075 S_028810_ZCLIP_NEAR_DISABLE(!state
->depth_clip
) |
1076 S_028810_ZCLIP_FAR_DISABLE(!state
->depth_clip
) |
1077 S_028810_DX_RASTERIZATION_KILL(state
->rasterizer_discard
) |
1078 S_028810_DX_LINEAR_ATTR_CLIP_ENA(1);
1080 si_pm4_set_reg(pm4
, R_0286D4_SPI_INTERP_CONTROL_0
,
1081 S_0286D4_FLAT_SHADE_ENA(1) |
1082 S_0286D4_PNT_SPRITE_ENA(1) |
1083 S_0286D4_PNT_SPRITE_OVRD_X(V_0286D4_SPI_PNT_SPRITE_SEL_S
) |
1084 S_0286D4_PNT_SPRITE_OVRD_Y(V_0286D4_SPI_PNT_SPRITE_SEL_T
) |
1085 S_0286D4_PNT_SPRITE_OVRD_Z(V_0286D4_SPI_PNT_SPRITE_SEL_0
) |
1086 S_0286D4_PNT_SPRITE_OVRD_W(V_0286D4_SPI_PNT_SPRITE_SEL_1
) |
1087 S_0286D4_PNT_SPRITE_TOP_1(state
->sprite_coord_mode
!= PIPE_SPRITE_COORD_UPPER_LEFT
));
1089 /* point size 12.4 fixed point */
1090 tmp
= (unsigned)(state
->point_size
* 8.0);
1091 si_pm4_set_reg(pm4
, R_028A00_PA_SU_POINT_SIZE
, S_028A00_HEIGHT(tmp
) | S_028A00_WIDTH(tmp
));
1093 if (state
->point_size_per_vertex
) {
1094 psize_min
= util_get_min_point_size(state
);
1097 /* Force the point size to be as if the vertex output was disabled. */
1098 psize_min
= state
->point_size
;
1099 psize_max
= state
->point_size
;
1101 /* Divide by two, because 0.5 = 1 pixel. */
1102 si_pm4_set_reg(pm4
, R_028A04_PA_SU_POINT_MINMAX
,
1103 S_028A04_MIN_SIZE(si_pack_float_12p4(psize_min
/2)) |
1104 S_028A04_MAX_SIZE(si_pack_float_12p4(psize_max
/2)));
1106 tmp
= (unsigned)state
->line_width
* 8;
1107 si_pm4_set_reg(pm4
, R_028A08_PA_SU_LINE_CNTL
, S_028A08_WIDTH(tmp
));
1108 si_pm4_set_reg(pm4
, R_028A48_PA_SC_MODE_CNTL_0
,
1109 S_028A48_LINE_STIPPLE_ENABLE(state
->line_stipple_enable
) |
1110 S_028A48_MSAA_ENABLE(state
->multisample
||
1111 state
->poly_smooth
||
1112 state
->line_smooth
) |
1113 S_028A48_VPORT_SCISSOR_ENABLE(1));
1115 si_pm4_set_reg(pm4
, R_028BE4_PA_SU_VTX_CNTL
,
1116 S_028BE4_PIX_CENTER(state
->half_pixel_center
) |
1117 S_028BE4_QUANT_MODE(V_028BE4_X_16_8_FIXED_POINT_1_256TH
));
1119 si_pm4_set_reg(pm4
, R_028B7C_PA_SU_POLY_OFFSET_CLAMP
, fui(state
->offset_clamp
));
1120 si_pm4_set_reg(pm4
, R_028814_PA_SU_SC_MODE_CNTL
,
1121 S_028814_PROVOKING_VTX_LAST(!state
->flatshade_first
) |
1122 S_028814_CULL_FRONT((state
->cull_face
& PIPE_FACE_FRONT
) ? 1 : 0) |
1123 S_028814_CULL_BACK((state
->cull_face
& PIPE_FACE_BACK
) ? 1 : 0) |
1124 S_028814_FACE(!state
->front_ccw
) |
1125 S_028814_POLY_OFFSET_FRONT_ENABLE(util_get_offset(state
, state
->fill_front
)) |
1126 S_028814_POLY_OFFSET_BACK_ENABLE(util_get_offset(state
, state
->fill_back
)) |
1127 S_028814_POLY_OFFSET_PARA_ENABLE(state
->offset_point
|| state
->offset_line
) |
1128 S_028814_POLY_MODE(state
->fill_front
!= PIPE_POLYGON_MODE_FILL
||
1129 state
->fill_back
!= PIPE_POLYGON_MODE_FILL
) |
1130 S_028814_POLYMODE_FRONT_PTYPE(si_translate_fill(state
->fill_front
)) |
1131 S_028814_POLYMODE_BACK_PTYPE(si_translate_fill(state
->fill_back
)));
1132 si_pm4_set_reg(pm4
, R_00B130_SPI_SHADER_USER_DATA_VS_0
+
1133 SI_SGPR_VS_STATE_BITS
* 4, state
->clamp_vertex_color
);
1135 /* Precalculate polygon offset states for 16-bit, 24-bit, and 32-bit zbuffers. */
1136 for (i
= 0; i
< 3; i
++) {
1137 struct si_pm4_state
*pm4
= &rs
->pm4_poly_offset
[i
];
1138 float offset_units
= state
->offset_units
;
1139 float offset_scale
= state
->offset_scale
* 16.0f
;
1142 case 0: /* 16-bit zbuffer */
1143 offset_units
*= 4.0f
;
1145 case 1: /* 24-bit zbuffer */
1146 offset_units
*= 2.0f
;
1148 case 2: /* 32-bit zbuffer */
1149 offset_units
*= 1.0f
;
1153 si_pm4_set_reg(pm4
, R_028B80_PA_SU_POLY_OFFSET_FRONT_SCALE
,
1155 si_pm4_set_reg(pm4
, R_028B84_PA_SU_POLY_OFFSET_FRONT_OFFSET
,
1157 si_pm4_set_reg(pm4
, R_028B88_PA_SU_POLY_OFFSET_BACK_SCALE
,
1159 si_pm4_set_reg(pm4
, R_028B8C_PA_SU_POLY_OFFSET_BACK_OFFSET
,
1166 static void si_bind_rs_state(struct pipe_context
*ctx
, void *state
)
1168 struct si_context
*sctx
= (struct si_context
*)ctx
;
1169 struct si_state_rasterizer
*old_rs
=
1170 (struct si_state_rasterizer
*)sctx
->queued
.named
.rasterizer
;
1171 struct si_state_rasterizer
*rs
= (struct si_state_rasterizer
*)state
;
1176 if (sctx
->framebuffer
.nr_samples
> 1 &&
1177 (!old_rs
|| old_rs
->multisample_enable
!= rs
->multisample_enable
))
1178 si_mark_atom_dirty(sctx
, &sctx
->db_render_state
);
1180 if (!old_rs
|| old_rs
->scissor_enable
!= rs
->scissor_enable
) {
1181 sctx
->scissors
.dirty_mask
= (1 << SI_MAX_VIEWPORTS
) - 1;
1182 si_mark_atom_dirty(sctx
, &sctx
->scissors
.atom
);
1185 si_pm4_bind_state(sctx
, rasterizer
, rs
);
1186 si_update_poly_offset_state(sctx
);
1188 si_mark_atom_dirty(sctx
, &sctx
->clip_regs
);
1191 static void si_delete_rs_state(struct pipe_context
*ctx
, void *state
)
1193 struct si_context
*sctx
= (struct si_context
*)ctx
;
1195 if (sctx
->queued
.named
.rasterizer
== state
)
1196 si_pm4_bind_state(sctx
, poly_offset
, NULL
);
1197 si_pm4_delete_state(sctx
, rasterizer
, (struct si_state_rasterizer
*)state
);
1201 * infeered state between dsa and stencil ref
1203 static void si_emit_stencil_ref(struct si_context
*sctx
, struct r600_atom
*atom
)
1205 struct radeon_winsys_cs
*cs
= sctx
->b
.gfx
.cs
;
1206 struct pipe_stencil_ref
*ref
= &sctx
->stencil_ref
.state
;
1207 struct si_dsa_stencil_ref_part
*dsa
= &sctx
->stencil_ref
.dsa_part
;
1209 radeon_set_context_reg_seq(cs
, R_028430_DB_STENCILREFMASK
, 2);
1210 radeon_emit(cs
, S_028430_STENCILTESTVAL(ref
->ref_value
[0]) |
1211 S_028430_STENCILMASK(dsa
->valuemask
[0]) |
1212 S_028430_STENCILWRITEMASK(dsa
->writemask
[0]) |
1213 S_028430_STENCILOPVAL(1));
1214 radeon_emit(cs
, S_028434_STENCILTESTVAL_BF(ref
->ref_value
[1]) |
1215 S_028434_STENCILMASK_BF(dsa
->valuemask
[1]) |
1216 S_028434_STENCILWRITEMASK_BF(dsa
->writemask
[1]) |
1217 S_028434_STENCILOPVAL_BF(1));
1220 static void si_set_stencil_ref(struct pipe_context
*ctx
,
1221 const struct pipe_stencil_ref
*state
)
1223 struct si_context
*sctx
= (struct si_context
*)ctx
;
1225 if (memcmp(&sctx
->stencil_ref
.state
, state
, sizeof(*state
)) == 0)
1228 sctx
->stencil_ref
.state
= *state
;
1229 si_mark_atom_dirty(sctx
, &sctx
->stencil_ref
.atom
);
1237 static uint32_t si_translate_stencil_op(int s_op
)
1240 case PIPE_STENCIL_OP_KEEP
:
1241 return V_02842C_STENCIL_KEEP
;
1242 case PIPE_STENCIL_OP_ZERO
:
1243 return V_02842C_STENCIL_ZERO
;
1244 case PIPE_STENCIL_OP_REPLACE
:
1245 return V_02842C_STENCIL_REPLACE_TEST
;
1246 case PIPE_STENCIL_OP_INCR
:
1247 return V_02842C_STENCIL_ADD_CLAMP
;
1248 case PIPE_STENCIL_OP_DECR
:
1249 return V_02842C_STENCIL_SUB_CLAMP
;
1250 case PIPE_STENCIL_OP_INCR_WRAP
:
1251 return V_02842C_STENCIL_ADD_WRAP
;
1252 case PIPE_STENCIL_OP_DECR_WRAP
:
1253 return V_02842C_STENCIL_SUB_WRAP
;
1254 case PIPE_STENCIL_OP_INVERT
:
1255 return V_02842C_STENCIL_INVERT
;
1257 R600_ERR("Unknown stencil op %d", s_op
);
1264 static void *si_create_dsa_state(struct pipe_context
*ctx
,
1265 const struct pipe_depth_stencil_alpha_state
*state
)
1267 struct si_state_dsa
*dsa
= CALLOC_STRUCT(si_state_dsa
);
1268 struct si_pm4_state
*pm4
= &dsa
->pm4
;
1269 unsigned db_depth_control
;
1270 uint32_t db_stencil_control
= 0;
1276 dsa
->stencil_ref
.valuemask
[0] = state
->stencil
[0].valuemask
;
1277 dsa
->stencil_ref
.valuemask
[1] = state
->stencil
[1].valuemask
;
1278 dsa
->stencil_ref
.writemask
[0] = state
->stencil
[0].writemask
;
1279 dsa
->stencil_ref
.writemask
[1] = state
->stencil
[1].writemask
;
1281 db_depth_control
= S_028800_Z_ENABLE(state
->depth
.enabled
) |
1282 S_028800_Z_WRITE_ENABLE(state
->depth
.writemask
) |
1283 S_028800_ZFUNC(state
->depth
.func
) |
1284 S_028800_DEPTH_BOUNDS_ENABLE(state
->depth
.bounds_test
);
1287 if (state
->stencil
[0].enabled
) {
1288 db_depth_control
|= S_028800_STENCIL_ENABLE(1);
1289 db_depth_control
|= S_028800_STENCILFUNC(state
->stencil
[0].func
);
1290 db_stencil_control
|= S_02842C_STENCILFAIL(si_translate_stencil_op(state
->stencil
[0].fail_op
));
1291 db_stencil_control
|= S_02842C_STENCILZPASS(si_translate_stencil_op(state
->stencil
[0].zpass_op
));
1292 db_stencil_control
|= S_02842C_STENCILZFAIL(si_translate_stencil_op(state
->stencil
[0].zfail_op
));
1294 if (state
->stencil
[1].enabled
) {
1295 db_depth_control
|= S_028800_BACKFACE_ENABLE(1);
1296 db_depth_control
|= S_028800_STENCILFUNC_BF(state
->stencil
[1].func
);
1297 db_stencil_control
|= S_02842C_STENCILFAIL_BF(si_translate_stencil_op(state
->stencil
[1].fail_op
));
1298 db_stencil_control
|= S_02842C_STENCILZPASS_BF(si_translate_stencil_op(state
->stencil
[1].zpass_op
));
1299 db_stencil_control
|= S_02842C_STENCILZFAIL_BF(si_translate_stencil_op(state
->stencil
[1].zfail_op
));
1304 if (state
->alpha
.enabled
) {
1305 dsa
->alpha_func
= state
->alpha
.func
;
1307 si_pm4_set_reg(pm4
, R_00B030_SPI_SHADER_USER_DATA_PS_0
+
1308 SI_SGPR_ALPHA_REF
* 4, fui(state
->alpha
.ref_value
));
1310 dsa
->alpha_func
= PIPE_FUNC_ALWAYS
;
1313 si_pm4_set_reg(pm4
, R_028800_DB_DEPTH_CONTROL
, db_depth_control
);
1314 si_pm4_set_reg(pm4
, R_02842C_DB_STENCIL_CONTROL
, db_stencil_control
);
1315 if (state
->depth
.bounds_test
) {
1316 si_pm4_set_reg(pm4
, R_028020_DB_DEPTH_BOUNDS_MIN
, fui(state
->depth
.bounds_min
));
1317 si_pm4_set_reg(pm4
, R_028024_DB_DEPTH_BOUNDS_MAX
, fui(state
->depth
.bounds_max
));
1323 static void si_bind_dsa_state(struct pipe_context
*ctx
, void *state
)
1325 struct si_context
*sctx
= (struct si_context
*)ctx
;
1326 struct si_state_dsa
*dsa
= state
;
1331 si_pm4_bind_state(sctx
, dsa
, dsa
);
1333 if (memcmp(&dsa
->stencil_ref
, &sctx
->stencil_ref
.dsa_part
,
1334 sizeof(struct si_dsa_stencil_ref_part
)) != 0) {
1335 sctx
->stencil_ref
.dsa_part
= dsa
->stencil_ref
;
1336 si_mark_atom_dirty(sctx
, &sctx
->stencil_ref
.atom
);
1340 static void si_delete_dsa_state(struct pipe_context
*ctx
, void *state
)
1342 struct si_context
*sctx
= (struct si_context
*)ctx
;
1343 si_pm4_delete_state(sctx
, dsa
, (struct si_state_dsa
*)state
);
1346 static void *si_create_db_flush_dsa(struct si_context
*sctx
)
1348 struct pipe_depth_stencil_alpha_state dsa
= {};
1350 return sctx
->b
.b
.create_depth_stencil_alpha_state(&sctx
->b
.b
, &dsa
);
1353 /* DB RENDER STATE */
1355 static void si_set_active_query_state(struct pipe_context
*ctx
, boolean enable
)
1357 struct si_context
*sctx
= (struct si_context
*)ctx
;
1359 /* Pipeline stat & streamout queries. */
1361 sctx
->b
.flags
&= ~SI_CONTEXT_STOP_PIPELINE_STATS
;
1362 sctx
->b
.flags
|= SI_CONTEXT_START_PIPELINE_STATS
;
1364 sctx
->b
.flags
&= ~SI_CONTEXT_START_PIPELINE_STATS
;
1365 sctx
->b
.flags
|= SI_CONTEXT_STOP_PIPELINE_STATS
;
1368 /* Occlusion queries. */
1369 if (sctx
->occlusion_queries_disabled
!= !enable
) {
1370 sctx
->occlusion_queries_disabled
= !enable
;
1371 si_mark_atom_dirty(sctx
, &sctx
->db_render_state
);
1375 static void si_set_occlusion_query_state(struct pipe_context
*ctx
, bool enable
)
1377 struct si_context
*sctx
= (struct si_context
*)ctx
;
1379 si_mark_atom_dirty(sctx
, &sctx
->db_render_state
);
1382 static void si_emit_db_render_state(struct si_context
*sctx
, struct r600_atom
*state
)
1384 struct radeon_winsys_cs
*cs
= sctx
->b
.gfx
.cs
;
1385 struct si_state_rasterizer
*rs
= sctx
->queued
.named
.rasterizer
;
1386 unsigned db_shader_control
;
1388 radeon_set_context_reg_seq(cs
, R_028000_DB_RENDER_CONTROL
, 2);
1390 /* DB_RENDER_CONTROL */
1391 if (sctx
->dbcb_depth_copy_enabled
||
1392 sctx
->dbcb_stencil_copy_enabled
) {
1394 S_028000_DEPTH_COPY(sctx
->dbcb_depth_copy_enabled
) |
1395 S_028000_STENCIL_COPY(sctx
->dbcb_stencil_copy_enabled
) |
1396 S_028000_COPY_CENTROID(1) |
1397 S_028000_COPY_SAMPLE(sctx
->dbcb_copy_sample
));
1398 } else if (sctx
->db_flush_depth_inplace
|| sctx
->db_flush_stencil_inplace
) {
1400 S_028000_DEPTH_COMPRESS_DISABLE(sctx
->db_flush_depth_inplace
) |
1401 S_028000_STENCIL_COMPRESS_DISABLE(sctx
->db_flush_stencil_inplace
));
1404 S_028000_DEPTH_CLEAR_ENABLE(sctx
->db_depth_clear
) |
1405 S_028000_STENCIL_CLEAR_ENABLE(sctx
->db_stencil_clear
));
1408 /* DB_COUNT_CONTROL (occlusion queries) */
1409 if (sctx
->b
.num_occlusion_queries
> 0 &&
1410 !sctx
->occlusion_queries_disabled
) {
1411 bool perfect
= sctx
->b
.num_perfect_occlusion_queries
> 0;
1413 if (sctx
->b
.chip_class
>= CIK
) {
1415 S_028004_PERFECT_ZPASS_COUNTS(perfect
) |
1416 S_028004_SAMPLE_RATE(sctx
->framebuffer
.log_samples
) |
1417 S_028004_ZPASS_ENABLE(1) |
1418 S_028004_SLICE_EVEN_ENABLE(1) |
1419 S_028004_SLICE_ODD_ENABLE(1));
1422 S_028004_PERFECT_ZPASS_COUNTS(perfect
) |
1423 S_028004_SAMPLE_RATE(sctx
->framebuffer
.log_samples
));
1426 /* Disable occlusion queries. */
1427 if (sctx
->b
.chip_class
>= CIK
) {
1430 radeon_emit(cs
, S_028004_ZPASS_INCREMENT_DISABLE(1));
1434 /* DB_RENDER_OVERRIDE2 */
1435 radeon_set_context_reg(cs
, R_028010_DB_RENDER_OVERRIDE2
,
1436 S_028010_DISABLE_ZMASK_EXPCLEAR_OPTIMIZATION(sctx
->db_depth_disable_expclear
) |
1437 S_028010_DISABLE_SMEM_EXPCLEAR_OPTIMIZATION(sctx
->db_stencil_disable_expclear
));
1439 db_shader_control
= S_02880C_ALPHA_TO_MASK_DISABLE(sctx
->framebuffer
.cb0_is_integer
) |
1440 sctx
->ps_db_shader_control
;
1442 /* Bug workaround for smoothing (overrasterization) on SI. */
1443 if (sctx
->b
.chip_class
== SI
&& sctx
->smoothing_enabled
) {
1444 db_shader_control
&= C_02880C_Z_ORDER
;
1445 db_shader_control
|= S_02880C_Z_ORDER(V_02880C_LATE_Z
);
1448 /* Disable the gl_SampleMask fragment shader output if MSAA is disabled. */
1449 if (sctx
->framebuffer
.nr_samples
<= 1 || (rs
&& !rs
->multisample_enable
))
1450 db_shader_control
&= C_02880C_MASK_EXPORT_ENABLE
;
1452 if (sctx
->b
.family
== CHIP_STONEY
&&
1453 sctx
->screen
->b
.debug_flags
& DBG_NO_RB_PLUS
)
1454 db_shader_control
|= S_02880C_DUAL_QUAD_DISABLE(1);
1456 radeon_set_context_reg(cs
, R_02880C_DB_SHADER_CONTROL
,
1461 * format translation
1463 static uint32_t si_translate_colorformat(enum pipe_format format
)
1465 const struct util_format_description
*desc
= util_format_description(format
);
1467 #define HAS_SIZE(x,y,z,w) \
1468 (desc->channel[0].size == (x) && desc->channel[1].size == (y) && \
1469 desc->channel[2].size == (z) && desc->channel[3].size == (w))
1471 if (format
== PIPE_FORMAT_R11G11B10_FLOAT
) /* isn't plain */
1472 return V_028C70_COLOR_10_11_11
;
1474 if (desc
->layout
!= UTIL_FORMAT_LAYOUT_PLAIN
)
1475 return V_028C70_COLOR_INVALID
;
1477 switch (desc
->nr_channels
) {
1479 switch (desc
->channel
[0].size
) {
1481 return V_028C70_COLOR_8
;
1483 return V_028C70_COLOR_16
;
1485 return V_028C70_COLOR_32
;
1489 if (desc
->channel
[0].size
== desc
->channel
[1].size
) {
1490 switch (desc
->channel
[0].size
) {
1492 return V_028C70_COLOR_8_8
;
1494 return V_028C70_COLOR_16_16
;
1496 return V_028C70_COLOR_32_32
;
1498 } else if (HAS_SIZE(8,24,0,0)) {
1499 return V_028C70_COLOR_24_8
;
1500 } else if (HAS_SIZE(24,8,0,0)) {
1501 return V_028C70_COLOR_8_24
;
1505 if (HAS_SIZE(5,6,5,0)) {
1506 return V_028C70_COLOR_5_6_5
;
1507 } else if (HAS_SIZE(32,8,24,0)) {
1508 return V_028C70_COLOR_X24_8_32_FLOAT
;
1512 if (desc
->channel
[0].size
== desc
->channel
[1].size
&&
1513 desc
->channel
[0].size
== desc
->channel
[2].size
&&
1514 desc
->channel
[0].size
== desc
->channel
[3].size
) {
1515 switch (desc
->channel
[0].size
) {
1517 return V_028C70_COLOR_4_4_4_4
;
1519 return V_028C70_COLOR_8_8_8_8
;
1521 return V_028C70_COLOR_16_16_16_16
;
1523 return V_028C70_COLOR_32_32_32_32
;
1525 } else if (HAS_SIZE(5,5,5,1)) {
1526 return V_028C70_COLOR_1_5_5_5
;
1527 } else if (HAS_SIZE(10,10,10,2)) {
1528 return V_028C70_COLOR_2_10_10_10
;
1532 return V_028C70_COLOR_INVALID
;
1535 static uint32_t si_colorformat_endian_swap(uint32_t colorformat
)
1537 if (SI_BIG_ENDIAN
) {
1538 switch(colorformat
) {
1539 /* 8-bit buffers. */
1540 case V_028C70_COLOR_8
:
1541 return V_028C70_ENDIAN_NONE
;
1543 /* 16-bit buffers. */
1544 case V_028C70_COLOR_5_6_5
:
1545 case V_028C70_COLOR_1_5_5_5
:
1546 case V_028C70_COLOR_4_4_4_4
:
1547 case V_028C70_COLOR_16
:
1548 case V_028C70_COLOR_8_8
:
1549 return V_028C70_ENDIAN_8IN16
;
1551 /* 32-bit buffers. */
1552 case V_028C70_COLOR_8_8_8_8
:
1553 case V_028C70_COLOR_2_10_10_10
:
1554 case V_028C70_COLOR_8_24
:
1555 case V_028C70_COLOR_24_8
:
1556 case V_028C70_COLOR_16_16
:
1557 return V_028C70_ENDIAN_8IN32
;
1559 /* 64-bit buffers. */
1560 case V_028C70_COLOR_16_16_16_16
:
1561 return V_028C70_ENDIAN_8IN16
;
1563 case V_028C70_COLOR_32_32
:
1564 return V_028C70_ENDIAN_8IN32
;
1566 /* 128-bit buffers. */
1567 case V_028C70_COLOR_32_32_32_32
:
1568 return V_028C70_ENDIAN_8IN32
;
1570 return V_028C70_ENDIAN_NONE
; /* Unsupported. */
1573 return V_028C70_ENDIAN_NONE
;
1577 static uint32_t si_translate_dbformat(enum pipe_format format
)
1580 case PIPE_FORMAT_Z16_UNORM
:
1581 return V_028040_Z_16
;
1582 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
1583 case PIPE_FORMAT_X8Z24_UNORM
:
1584 case PIPE_FORMAT_Z24X8_UNORM
:
1585 case PIPE_FORMAT_Z24_UNORM_S8_UINT
:
1586 return V_028040_Z_24
; /* deprecated on SI */
1587 case PIPE_FORMAT_Z32_FLOAT
:
1588 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
1589 return V_028040_Z_32_FLOAT
;
1591 return V_028040_Z_INVALID
;
1596 * Texture translation
1599 static uint32_t si_translate_texformat(struct pipe_screen
*screen
,
1600 enum pipe_format format
,
1601 const struct util_format_description
*desc
,
1604 struct si_screen
*sscreen
= (struct si_screen
*)screen
;
1605 bool enable_compressed_formats
= (sscreen
->b
.info
.drm_major
== 2 &&
1606 sscreen
->b
.info
.drm_minor
>= 31) ||
1607 sscreen
->b
.info
.drm_major
== 3;
1608 boolean uniform
= TRUE
;
1611 /* Colorspace (return non-RGB formats directly). */
1612 switch (desc
->colorspace
) {
1613 /* Depth stencil formats */
1614 case UTIL_FORMAT_COLORSPACE_ZS
:
1616 case PIPE_FORMAT_Z16_UNORM
:
1617 return V_008F14_IMG_DATA_FORMAT_16
;
1618 case PIPE_FORMAT_X24S8_UINT
:
1619 case PIPE_FORMAT_Z24X8_UNORM
:
1620 case PIPE_FORMAT_Z24_UNORM_S8_UINT
:
1621 return V_008F14_IMG_DATA_FORMAT_8_24
;
1622 case PIPE_FORMAT_X8Z24_UNORM
:
1623 case PIPE_FORMAT_S8X24_UINT
:
1624 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
1625 return V_008F14_IMG_DATA_FORMAT_24_8
;
1626 case PIPE_FORMAT_S8_UINT
:
1627 return V_008F14_IMG_DATA_FORMAT_8
;
1628 case PIPE_FORMAT_Z32_FLOAT
:
1629 return V_008F14_IMG_DATA_FORMAT_32
;
1630 case PIPE_FORMAT_X32_S8X24_UINT
:
1631 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
1632 return V_008F14_IMG_DATA_FORMAT_X24_8_32
;
1637 case UTIL_FORMAT_COLORSPACE_YUV
:
1638 goto out_unknown
; /* TODO */
1640 case UTIL_FORMAT_COLORSPACE_SRGB
:
1641 if (desc
->nr_channels
!= 4 && desc
->nr_channels
!= 1)
1649 if (desc
->layout
== UTIL_FORMAT_LAYOUT_RGTC
) {
1650 if (!enable_compressed_formats
)
1654 case PIPE_FORMAT_RGTC1_SNORM
:
1655 case PIPE_FORMAT_LATC1_SNORM
:
1656 case PIPE_FORMAT_RGTC1_UNORM
:
1657 case PIPE_FORMAT_LATC1_UNORM
:
1658 return V_008F14_IMG_DATA_FORMAT_BC4
;
1659 case PIPE_FORMAT_RGTC2_SNORM
:
1660 case PIPE_FORMAT_LATC2_SNORM
:
1661 case PIPE_FORMAT_RGTC2_UNORM
:
1662 case PIPE_FORMAT_LATC2_UNORM
:
1663 return V_008F14_IMG_DATA_FORMAT_BC5
;
1669 if (desc
->layout
== UTIL_FORMAT_LAYOUT_ETC
&&
1670 sscreen
->b
.family
>= CHIP_STONEY
) {
1672 case PIPE_FORMAT_ETC1_RGB8
:
1673 case PIPE_FORMAT_ETC2_RGB8
:
1674 case PIPE_FORMAT_ETC2_SRGB8
:
1675 return V_008F14_IMG_DATA_FORMAT_ETC2_RGB
;
1676 case PIPE_FORMAT_ETC2_RGB8A1
:
1677 case PIPE_FORMAT_ETC2_SRGB8A1
:
1678 return V_008F14_IMG_DATA_FORMAT_ETC2_RGBA1
;
1679 case PIPE_FORMAT_ETC2_RGBA8
:
1680 case PIPE_FORMAT_ETC2_SRGBA8
:
1681 return V_008F14_IMG_DATA_FORMAT_ETC2_RGBA
;
1682 case PIPE_FORMAT_ETC2_R11_UNORM
:
1683 case PIPE_FORMAT_ETC2_R11_SNORM
:
1684 return V_008F14_IMG_DATA_FORMAT_ETC2_R
;
1685 case PIPE_FORMAT_ETC2_RG11_UNORM
:
1686 case PIPE_FORMAT_ETC2_RG11_SNORM
:
1687 return V_008F14_IMG_DATA_FORMAT_ETC2_RG
;
1693 if (desc
->layout
== UTIL_FORMAT_LAYOUT_BPTC
) {
1694 if (!enable_compressed_formats
)
1698 case PIPE_FORMAT_BPTC_RGBA_UNORM
:
1699 case PIPE_FORMAT_BPTC_SRGBA
:
1700 return V_008F14_IMG_DATA_FORMAT_BC7
;
1701 case PIPE_FORMAT_BPTC_RGB_FLOAT
:
1702 case PIPE_FORMAT_BPTC_RGB_UFLOAT
:
1703 return V_008F14_IMG_DATA_FORMAT_BC6
;
1709 if (desc
->layout
== UTIL_FORMAT_LAYOUT_SUBSAMPLED
) {
1711 case PIPE_FORMAT_R8G8_B8G8_UNORM
:
1712 case PIPE_FORMAT_G8R8_B8R8_UNORM
:
1713 return V_008F14_IMG_DATA_FORMAT_GB_GR
;
1714 case PIPE_FORMAT_G8R8_G8B8_UNORM
:
1715 case PIPE_FORMAT_R8G8_R8B8_UNORM
:
1716 return V_008F14_IMG_DATA_FORMAT_BG_RG
;
1722 if (desc
->layout
== UTIL_FORMAT_LAYOUT_S3TC
) {
1723 if (!enable_compressed_formats
)
1726 if (!util_format_s3tc_enabled
) {
1731 case PIPE_FORMAT_DXT1_RGB
:
1732 case PIPE_FORMAT_DXT1_RGBA
:
1733 case PIPE_FORMAT_DXT1_SRGB
:
1734 case PIPE_FORMAT_DXT1_SRGBA
:
1735 return V_008F14_IMG_DATA_FORMAT_BC1
;
1736 case PIPE_FORMAT_DXT3_RGBA
:
1737 case PIPE_FORMAT_DXT3_SRGBA
:
1738 return V_008F14_IMG_DATA_FORMAT_BC2
;
1739 case PIPE_FORMAT_DXT5_RGBA
:
1740 case PIPE_FORMAT_DXT5_SRGBA
:
1741 return V_008F14_IMG_DATA_FORMAT_BC3
;
1747 if (format
== PIPE_FORMAT_R9G9B9E5_FLOAT
) {
1748 return V_008F14_IMG_DATA_FORMAT_5_9_9_9
;
1749 } else if (format
== PIPE_FORMAT_R11G11B10_FLOAT
) {
1750 return V_008F14_IMG_DATA_FORMAT_10_11_11
;
1753 /* R8G8Bx_SNORM - TODO CxV8U8 */
1755 /* See whether the components are of the same size. */
1756 for (i
= 1; i
< desc
->nr_channels
; i
++) {
1757 uniform
= uniform
&& desc
->channel
[0].size
== desc
->channel
[i
].size
;
1760 /* Non-uniform formats. */
1762 switch(desc
->nr_channels
) {
1764 if (desc
->channel
[0].size
== 5 &&
1765 desc
->channel
[1].size
== 6 &&
1766 desc
->channel
[2].size
== 5) {
1767 return V_008F14_IMG_DATA_FORMAT_5_6_5
;
1771 if (desc
->channel
[0].size
== 5 &&
1772 desc
->channel
[1].size
== 5 &&
1773 desc
->channel
[2].size
== 5 &&
1774 desc
->channel
[3].size
== 1) {
1775 return V_008F14_IMG_DATA_FORMAT_1_5_5_5
;
1777 if (desc
->channel
[0].size
== 10 &&
1778 desc
->channel
[1].size
== 10 &&
1779 desc
->channel
[2].size
== 10 &&
1780 desc
->channel
[3].size
== 2) {
1781 return V_008F14_IMG_DATA_FORMAT_2_10_10_10
;
1788 if (first_non_void
< 0 || first_non_void
> 3)
1791 /* uniform formats */
1792 switch (desc
->channel
[first_non_void
].size
) {
1794 switch (desc
->nr_channels
) {
1795 #if 0 /* Not supported for render targets */
1797 return V_008F14_IMG_DATA_FORMAT_4_4
;
1800 return V_008F14_IMG_DATA_FORMAT_4_4_4_4
;
1804 switch (desc
->nr_channels
) {
1806 return V_008F14_IMG_DATA_FORMAT_8
;
1808 return V_008F14_IMG_DATA_FORMAT_8_8
;
1810 return V_008F14_IMG_DATA_FORMAT_8_8_8_8
;
1814 switch (desc
->nr_channels
) {
1816 return V_008F14_IMG_DATA_FORMAT_16
;
1818 return V_008F14_IMG_DATA_FORMAT_16_16
;
1820 return V_008F14_IMG_DATA_FORMAT_16_16_16_16
;
1824 switch (desc
->nr_channels
) {
1826 return V_008F14_IMG_DATA_FORMAT_32
;
1828 return V_008F14_IMG_DATA_FORMAT_32_32
;
1829 #if 0 /* Not supported for render targets */
1831 return V_008F14_IMG_DATA_FORMAT_32_32_32
;
1834 return V_008F14_IMG_DATA_FORMAT_32_32_32_32
;
1839 /* R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format)); */
1843 static unsigned si_tex_wrap(unsigned wrap
)
1847 case PIPE_TEX_WRAP_REPEAT
:
1848 return V_008F30_SQ_TEX_WRAP
;
1849 case PIPE_TEX_WRAP_CLAMP
:
1850 return V_008F30_SQ_TEX_CLAMP_HALF_BORDER
;
1851 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
1852 return V_008F30_SQ_TEX_CLAMP_LAST_TEXEL
;
1853 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
1854 return V_008F30_SQ_TEX_CLAMP_BORDER
;
1855 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
1856 return V_008F30_SQ_TEX_MIRROR
;
1857 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
1858 return V_008F30_SQ_TEX_MIRROR_ONCE_HALF_BORDER
;
1859 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
1860 return V_008F30_SQ_TEX_MIRROR_ONCE_LAST_TEXEL
;
1861 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
1862 return V_008F30_SQ_TEX_MIRROR_ONCE_BORDER
;
1866 static unsigned si_tex_filter(unsigned filter
)
1870 case PIPE_TEX_FILTER_NEAREST
:
1871 return V_008F38_SQ_TEX_XY_FILTER_POINT
;
1872 case PIPE_TEX_FILTER_LINEAR
:
1873 return V_008F38_SQ_TEX_XY_FILTER_BILINEAR
;
1877 static unsigned si_tex_mipfilter(unsigned filter
)
1880 case PIPE_TEX_MIPFILTER_NEAREST
:
1881 return V_008F38_SQ_TEX_Z_FILTER_POINT
;
1882 case PIPE_TEX_MIPFILTER_LINEAR
:
1883 return V_008F38_SQ_TEX_Z_FILTER_LINEAR
;
1885 case PIPE_TEX_MIPFILTER_NONE
:
1886 return V_008F38_SQ_TEX_Z_FILTER_NONE
;
1890 static unsigned si_tex_compare(unsigned compare
)
1894 case PIPE_FUNC_NEVER
:
1895 return V_008F30_SQ_TEX_DEPTH_COMPARE_NEVER
;
1896 case PIPE_FUNC_LESS
:
1897 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESS
;
1898 case PIPE_FUNC_EQUAL
:
1899 return V_008F30_SQ_TEX_DEPTH_COMPARE_EQUAL
;
1900 case PIPE_FUNC_LEQUAL
:
1901 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESSEQUAL
;
1902 case PIPE_FUNC_GREATER
:
1903 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATER
;
1904 case PIPE_FUNC_NOTEQUAL
:
1905 return V_008F30_SQ_TEX_DEPTH_COMPARE_NOTEQUAL
;
1906 case PIPE_FUNC_GEQUAL
:
1907 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATEREQUAL
;
1908 case PIPE_FUNC_ALWAYS
:
1909 return V_008F30_SQ_TEX_DEPTH_COMPARE_ALWAYS
;
1913 static unsigned si_tex_dim(unsigned res_target
, unsigned view_target
,
1914 unsigned nr_samples
)
1916 if (view_target
== PIPE_TEXTURE_CUBE
||
1917 view_target
== PIPE_TEXTURE_CUBE_ARRAY
)
1918 res_target
= view_target
;
1920 switch (res_target
) {
1922 case PIPE_TEXTURE_1D
:
1923 return V_008F1C_SQ_RSRC_IMG_1D
;
1924 case PIPE_TEXTURE_1D_ARRAY
:
1925 return V_008F1C_SQ_RSRC_IMG_1D_ARRAY
;
1926 case PIPE_TEXTURE_2D
:
1927 case PIPE_TEXTURE_RECT
:
1928 return nr_samples
> 1 ? V_008F1C_SQ_RSRC_IMG_2D_MSAA
:
1929 V_008F1C_SQ_RSRC_IMG_2D
;
1930 case PIPE_TEXTURE_2D_ARRAY
:
1931 return nr_samples
> 1 ? V_008F1C_SQ_RSRC_IMG_2D_MSAA_ARRAY
:
1932 V_008F1C_SQ_RSRC_IMG_2D_ARRAY
;
1933 case PIPE_TEXTURE_3D
:
1934 return V_008F1C_SQ_RSRC_IMG_3D
;
1935 case PIPE_TEXTURE_CUBE
:
1936 case PIPE_TEXTURE_CUBE_ARRAY
:
1937 return V_008F1C_SQ_RSRC_IMG_CUBE
;
1942 * Format support testing
1945 static bool si_is_sampler_format_supported(struct pipe_screen
*screen
, enum pipe_format format
)
1947 return si_translate_texformat(screen
, format
, util_format_description(format
),
1948 util_format_get_first_non_void_channel(format
)) != ~0U;
1951 static uint32_t si_translate_buffer_dataformat(struct pipe_screen
*screen
,
1952 const struct util_format_description
*desc
,
1955 unsigned type
= desc
->channel
[first_non_void
].type
;
1958 if (type
== UTIL_FORMAT_TYPE_FIXED
)
1959 return V_008F0C_BUF_DATA_FORMAT_INVALID
;
1961 if (desc
->format
== PIPE_FORMAT_R11G11B10_FLOAT
)
1962 return V_008F0C_BUF_DATA_FORMAT_10_11_11
;
1964 if (desc
->nr_channels
== 4 &&
1965 desc
->channel
[0].size
== 10 &&
1966 desc
->channel
[1].size
== 10 &&
1967 desc
->channel
[2].size
== 10 &&
1968 desc
->channel
[3].size
== 2)
1969 return V_008F0C_BUF_DATA_FORMAT_2_10_10_10
;
1971 /* See whether the components are of the same size. */
1972 for (i
= 0; i
< desc
->nr_channels
; i
++) {
1973 if (desc
->channel
[first_non_void
].size
!= desc
->channel
[i
].size
)
1974 return V_008F0C_BUF_DATA_FORMAT_INVALID
;
1977 switch (desc
->channel
[first_non_void
].size
) {
1979 switch (desc
->nr_channels
) {
1981 return V_008F0C_BUF_DATA_FORMAT_8
;
1983 return V_008F0C_BUF_DATA_FORMAT_8_8
;
1986 return V_008F0C_BUF_DATA_FORMAT_8_8_8_8
;
1990 switch (desc
->nr_channels
) {
1992 return V_008F0C_BUF_DATA_FORMAT_16
;
1994 return V_008F0C_BUF_DATA_FORMAT_16_16
;
1997 return V_008F0C_BUF_DATA_FORMAT_16_16_16_16
;
2001 /* From the Southern Islands ISA documentation about MTBUF:
2002 * 'Memory reads of data in memory that is 32 or 64 bits do not
2003 * undergo any format conversion.'
2005 if (type
!= UTIL_FORMAT_TYPE_FLOAT
&&
2006 !desc
->channel
[first_non_void
].pure_integer
)
2007 return V_008F0C_BUF_DATA_FORMAT_INVALID
;
2009 switch (desc
->nr_channels
) {
2011 return V_008F0C_BUF_DATA_FORMAT_32
;
2013 return V_008F0C_BUF_DATA_FORMAT_32_32
;
2015 return V_008F0C_BUF_DATA_FORMAT_32_32_32
;
2017 return V_008F0C_BUF_DATA_FORMAT_32_32_32_32
;
2022 return V_008F0C_BUF_DATA_FORMAT_INVALID
;
2025 static uint32_t si_translate_buffer_numformat(struct pipe_screen
*screen
,
2026 const struct util_format_description
*desc
,
2029 if (desc
->format
== PIPE_FORMAT_R11G11B10_FLOAT
)
2030 return V_008F0C_BUF_NUM_FORMAT_FLOAT
;
2032 switch (desc
->channel
[first_non_void
].type
) {
2033 case UTIL_FORMAT_TYPE_SIGNED
:
2034 if (desc
->channel
[first_non_void
].normalized
)
2035 return V_008F0C_BUF_NUM_FORMAT_SNORM
;
2036 else if (desc
->channel
[first_non_void
].pure_integer
)
2037 return V_008F0C_BUF_NUM_FORMAT_SINT
;
2039 return V_008F0C_BUF_NUM_FORMAT_SSCALED
;
2041 case UTIL_FORMAT_TYPE_UNSIGNED
:
2042 if (desc
->channel
[first_non_void
].normalized
)
2043 return V_008F0C_BUF_NUM_FORMAT_UNORM
;
2044 else if (desc
->channel
[first_non_void
].pure_integer
)
2045 return V_008F0C_BUF_NUM_FORMAT_UINT
;
2047 return V_008F0C_BUF_NUM_FORMAT_USCALED
;
2049 case UTIL_FORMAT_TYPE_FLOAT
:
2051 return V_008F0C_BUF_NUM_FORMAT_FLOAT
;
2055 static bool si_is_vertex_format_supported(struct pipe_screen
*screen
, enum pipe_format format
)
2057 const struct util_format_description
*desc
;
2059 unsigned data_format
;
2061 desc
= util_format_description(format
);
2062 first_non_void
= util_format_get_first_non_void_channel(format
);
2063 data_format
= si_translate_buffer_dataformat(screen
, desc
, first_non_void
);
2064 return data_format
!= V_008F0C_BUF_DATA_FORMAT_INVALID
;
2067 static bool si_is_colorbuffer_format_supported(enum pipe_format format
)
2069 return si_translate_colorformat(format
) != V_028C70_COLOR_INVALID
&&
2070 r600_translate_colorswap(format
) != ~0U;
2073 static bool si_is_zs_format_supported(enum pipe_format format
)
2075 return si_translate_dbformat(format
) != V_028040_Z_INVALID
;
2078 boolean
si_is_format_supported(struct pipe_screen
*screen
,
2079 enum pipe_format format
,
2080 enum pipe_texture_target target
,
2081 unsigned sample_count
,
2084 unsigned retval
= 0;
2086 if (target
>= PIPE_MAX_TEXTURE_TYPES
) {
2087 R600_ERR("r600: unsupported texture type %d\n", target
);
2091 if (!util_format_is_supported(format
, usage
))
2094 if (sample_count
> 1) {
2095 if (!screen
->get_param(screen
, PIPE_CAP_TEXTURE_MULTISAMPLE
))
2098 switch (sample_count
) {
2104 if (format
== PIPE_FORMAT_NONE
)
2113 if (usage
& PIPE_BIND_SAMPLER_VIEW
) {
2114 if (target
== PIPE_BUFFER
) {
2115 if (si_is_vertex_format_supported(screen
, format
))
2116 retval
|= PIPE_BIND_SAMPLER_VIEW
;
2118 if (si_is_sampler_format_supported(screen
, format
))
2119 retval
|= PIPE_BIND_SAMPLER_VIEW
;
2123 if ((usage
& (PIPE_BIND_RENDER_TARGET
|
2124 PIPE_BIND_DISPLAY_TARGET
|
2127 PIPE_BIND_BLENDABLE
)) &&
2128 si_is_colorbuffer_format_supported(format
)) {
2130 (PIPE_BIND_RENDER_TARGET
|
2131 PIPE_BIND_DISPLAY_TARGET
|
2134 if (!util_format_is_pure_integer(format
) &&
2135 !util_format_is_depth_or_stencil(format
))
2136 retval
|= usage
& PIPE_BIND_BLENDABLE
;
2139 if ((usage
& PIPE_BIND_DEPTH_STENCIL
) &&
2140 si_is_zs_format_supported(format
)) {
2141 retval
|= PIPE_BIND_DEPTH_STENCIL
;
2144 if ((usage
& PIPE_BIND_VERTEX_BUFFER
) &&
2145 si_is_vertex_format_supported(screen
, format
)) {
2146 retval
|= PIPE_BIND_VERTEX_BUFFER
;
2149 if (usage
& PIPE_BIND_TRANSFER_READ
)
2150 retval
|= PIPE_BIND_TRANSFER_READ
;
2151 if (usage
& PIPE_BIND_TRANSFER_WRITE
)
2152 retval
|= PIPE_BIND_TRANSFER_WRITE
;
2154 if ((usage
& PIPE_BIND_LINEAR
) &&
2155 !util_format_is_compressed(format
) &&
2156 !(usage
& PIPE_BIND_DEPTH_STENCIL
))
2157 retval
|= PIPE_BIND_LINEAR
;
2159 return retval
== usage
;
2162 unsigned si_tile_mode_index(struct r600_texture
*rtex
, unsigned level
, bool stencil
)
2164 unsigned tile_mode_index
= 0;
2167 tile_mode_index
= rtex
->surface
.stencil_tiling_index
[level
];
2169 tile_mode_index
= rtex
->surface
.tiling_index
[level
];
2171 return tile_mode_index
;
2175 * framebuffer handling
2178 static void si_choose_spi_color_formats(struct r600_surface
*surf
,
2179 unsigned format
, unsigned swap
,
2180 unsigned ntype
, bool is_depth
)
2182 /* Alpha is needed for alpha-to-coverage.
2183 * Blending may be with or without alpha.
2185 unsigned normal
= 0; /* most optimal, may not support blending or export alpha */
2186 unsigned alpha
= 0; /* exports alpha, but may not support blending */
2187 unsigned blend
= 0; /* supports blending, but may not export alpha */
2188 unsigned blend_alpha
= 0; /* least optimal, supports blending and exports alpha */
2190 /* Choose the SPI color formats. These are required values for Stoney/RB+.
2191 * Other chips have multiple choices, though they are not necessarily better.
2194 case V_028C70_COLOR_5_6_5
:
2195 case V_028C70_COLOR_1_5_5_5
:
2196 case V_028C70_COLOR_5_5_5_1
:
2197 case V_028C70_COLOR_4_4_4_4
:
2198 case V_028C70_COLOR_10_11_11
:
2199 case V_028C70_COLOR_11_11_10
:
2200 case V_028C70_COLOR_8
:
2201 case V_028C70_COLOR_8_8
:
2202 case V_028C70_COLOR_8_8_8_8
:
2203 case V_028C70_COLOR_10_10_10_2
:
2204 case V_028C70_COLOR_2_10_10_10
:
2205 if (ntype
== V_028C70_NUMBER_UINT
)
2206 alpha
= blend
= blend_alpha
= normal
= V_028714_SPI_SHADER_UINT16_ABGR
;
2207 else if (ntype
== V_028C70_NUMBER_SINT
)
2208 alpha
= blend
= blend_alpha
= normal
= V_028714_SPI_SHADER_SINT16_ABGR
;
2210 alpha
= blend
= blend_alpha
= normal
= V_028714_SPI_SHADER_FP16_ABGR
;
2213 case V_028C70_COLOR_16
:
2214 case V_028C70_COLOR_16_16
:
2215 case V_028C70_COLOR_16_16_16_16
:
2216 if (ntype
== V_028C70_NUMBER_UNORM
||
2217 ntype
== V_028C70_NUMBER_SNORM
) {
2218 /* UNORM16 and SNORM16 don't support blending */
2219 if (ntype
== V_028C70_NUMBER_UNORM
)
2220 normal
= alpha
= V_028714_SPI_SHADER_UNORM16_ABGR
;
2222 normal
= alpha
= V_028714_SPI_SHADER_SNORM16_ABGR
;
2224 /* Use 32 bits per channel for blending. */
2225 if (format
== V_028C70_COLOR_16
) {
2226 if (swap
== V_028C70_SWAP_STD
) { /* R */
2227 blend
= V_028714_SPI_SHADER_32_R
;
2228 blend_alpha
= V_028714_SPI_SHADER_32_AR
;
2229 } else if (swap
== V_028C70_SWAP_ALT_REV
) /* A */
2230 blend
= blend_alpha
= V_028714_SPI_SHADER_32_AR
;
2233 } else if (format
== V_028C70_COLOR_16_16
) {
2234 if (swap
== V_028C70_SWAP_STD
) { /* RG */
2235 blend
= V_028714_SPI_SHADER_32_GR
;
2236 blend_alpha
= V_028714_SPI_SHADER_32_ABGR
;
2237 } else if (swap
== V_028C70_SWAP_ALT
) /* RA */
2238 blend
= blend_alpha
= V_028714_SPI_SHADER_32_AR
;
2241 } else /* 16_16_16_16 */
2242 blend
= blend_alpha
= V_028714_SPI_SHADER_32_ABGR
;
2243 } else if (ntype
== V_028C70_NUMBER_UINT
)
2244 alpha
= blend
= blend_alpha
= normal
= V_028714_SPI_SHADER_UINT16_ABGR
;
2245 else if (ntype
== V_028C70_NUMBER_SINT
)
2246 alpha
= blend
= blend_alpha
= normal
= V_028714_SPI_SHADER_SINT16_ABGR
;
2247 else if (ntype
== V_028C70_NUMBER_FLOAT
)
2248 alpha
= blend
= blend_alpha
= normal
= V_028714_SPI_SHADER_FP16_ABGR
;
2253 case V_028C70_COLOR_32
:
2254 if (swap
== V_028C70_SWAP_STD
) { /* R */
2255 blend
= normal
= V_028714_SPI_SHADER_32_R
;
2256 alpha
= blend_alpha
= V_028714_SPI_SHADER_32_AR
;
2257 } else if (swap
== V_028C70_SWAP_ALT_REV
) /* A */
2258 alpha
= blend
= blend_alpha
= normal
= V_028714_SPI_SHADER_32_AR
;
2263 case V_028C70_COLOR_32_32
:
2264 if (swap
== V_028C70_SWAP_STD
) { /* RG */
2265 blend
= normal
= V_028714_SPI_SHADER_32_GR
;
2266 alpha
= blend_alpha
= V_028714_SPI_SHADER_32_ABGR
;
2267 } else if (swap
== V_028C70_SWAP_ALT
) /* RA */
2268 alpha
= blend
= blend_alpha
= normal
= V_028714_SPI_SHADER_32_AR
;
2273 case V_028C70_COLOR_32_32_32_32
:
2274 case V_028C70_COLOR_8_24
:
2275 case V_028C70_COLOR_24_8
:
2276 case V_028C70_COLOR_X24_8_32_FLOAT
:
2277 alpha
= blend
= blend_alpha
= normal
= V_028714_SPI_SHADER_32_ABGR
;
2285 /* The DB->CB copy needs 32_ABGR. */
2287 alpha
= blend
= blend_alpha
= normal
= V_028714_SPI_SHADER_32_ABGR
;
2289 surf
->spi_shader_col_format
= normal
;
2290 surf
->spi_shader_col_format_alpha
= alpha
;
2291 surf
->spi_shader_col_format_blend
= blend
;
2292 surf
->spi_shader_col_format_blend_alpha
= blend_alpha
;
2295 static void si_initialize_color_surface(struct si_context
*sctx
,
2296 struct r600_surface
*surf
)
2298 struct r600_texture
*rtex
= (struct r600_texture
*)surf
->base
.texture
;
2299 unsigned level
= surf
->base
.u
.tex
.level
;
2300 uint64_t offset
= rtex
->surface
.level
[level
].offset
;
2301 unsigned pitch
, slice
;
2302 unsigned color_info
, color_attrib
, color_pitch
, color_view
;
2303 unsigned tile_mode_index
;
2304 unsigned format
, swap
, ntype
, endian
;
2305 const struct util_format_description
*desc
;
2307 unsigned blend_clamp
= 0, blend_bypass
= 0;
2309 /* Layered rendering doesn't work with LINEAR_GENERAL.
2310 * (LINEAR_ALIGNED and others work) */
2311 if (rtex
->surface
.level
[level
].mode
== RADEON_SURF_MODE_LINEAR
) {
2312 assert(surf
->base
.u
.tex
.first_layer
== surf
->base
.u
.tex
.last_layer
);
2313 offset
+= rtex
->surface
.level
[level
].slice_size
*
2314 surf
->base
.u
.tex
.first_layer
;
2317 color_view
= S_028C6C_SLICE_START(surf
->base
.u
.tex
.first_layer
) |
2318 S_028C6C_SLICE_MAX(surf
->base
.u
.tex
.last_layer
);
2321 pitch
= (rtex
->surface
.level
[level
].nblk_x
) / 8 - 1;
2322 slice
= (rtex
->surface
.level
[level
].nblk_x
* rtex
->surface
.level
[level
].nblk_y
) / 64;
2327 tile_mode_index
= si_tile_mode_index(rtex
, level
, false);
2329 desc
= util_format_description(surf
->base
.format
);
2330 for (i
= 0; i
< 4; i
++) {
2331 if (desc
->channel
[i
].type
!= UTIL_FORMAT_TYPE_VOID
) {
2335 if (i
== 4 || desc
->channel
[i
].type
== UTIL_FORMAT_TYPE_FLOAT
) {
2336 ntype
= V_028C70_NUMBER_FLOAT
;
2338 ntype
= V_028C70_NUMBER_UNORM
;
2339 if (desc
->colorspace
== UTIL_FORMAT_COLORSPACE_SRGB
)
2340 ntype
= V_028C70_NUMBER_SRGB
;
2341 else if (desc
->channel
[i
].type
== UTIL_FORMAT_TYPE_SIGNED
) {
2342 if (desc
->channel
[i
].pure_integer
) {
2343 ntype
= V_028C70_NUMBER_SINT
;
2345 assert(desc
->channel
[i
].normalized
);
2346 ntype
= V_028C70_NUMBER_SNORM
;
2348 } else if (desc
->channel
[i
].type
== UTIL_FORMAT_TYPE_UNSIGNED
) {
2349 if (desc
->channel
[i
].pure_integer
) {
2350 ntype
= V_028C70_NUMBER_UINT
;
2352 assert(desc
->channel
[i
].normalized
);
2353 ntype
= V_028C70_NUMBER_UNORM
;
2358 format
= si_translate_colorformat(surf
->base
.format
);
2359 if (format
== V_028C70_COLOR_INVALID
) {
2360 R600_ERR("Invalid CB format: %d, disabling CB.\n", surf
->base
.format
);
2362 assert(format
!= V_028C70_COLOR_INVALID
);
2363 swap
= r600_translate_colorswap(surf
->base
.format
);
2364 endian
= si_colorformat_endian_swap(format
);
2366 /* blend clamp should be set for all NORM/SRGB types */
2367 if (ntype
== V_028C70_NUMBER_UNORM
||
2368 ntype
== V_028C70_NUMBER_SNORM
||
2369 ntype
== V_028C70_NUMBER_SRGB
)
2372 /* set blend bypass according to docs if SINT/UINT or
2373 8/24 COLOR variants */
2374 if (ntype
== V_028C70_NUMBER_UINT
|| ntype
== V_028C70_NUMBER_SINT
||
2375 format
== V_028C70_COLOR_8_24
|| format
== V_028C70_COLOR_24_8
||
2376 format
== V_028C70_COLOR_X24_8_32_FLOAT
) {
2381 if ((ntype
== V_028C70_NUMBER_UINT
|| ntype
== V_028C70_NUMBER_SINT
) &&
2382 (format
== V_028C70_COLOR_8
||
2383 format
== V_028C70_COLOR_8_8
||
2384 format
== V_028C70_COLOR_8_8_8_8
))
2385 surf
->color_is_int8
= true;
2387 color_info
= S_028C70_FORMAT(format
) |
2388 S_028C70_COMP_SWAP(swap
) |
2389 S_028C70_BLEND_CLAMP(blend_clamp
) |
2390 S_028C70_BLEND_BYPASS(blend_bypass
) |
2391 S_028C70_NUMBER_TYPE(ntype
) |
2392 S_028C70_ENDIAN(endian
);
2394 color_pitch
= S_028C64_TILE_MAX(pitch
);
2396 /* Intensity is implemented as Red, so treat it that way. */
2397 color_attrib
= S_028C74_TILE_MODE_INDEX(tile_mode_index
) |
2398 S_028C74_FORCE_DST_ALPHA_1(desc
->swizzle
[3] == UTIL_FORMAT_SWIZZLE_1
||
2399 util_format_is_intensity(surf
->base
.format
));
2401 if (rtex
->resource
.b
.b
.nr_samples
> 1) {
2402 unsigned log_samples
= util_logbase2(rtex
->resource
.b
.b
.nr_samples
);
2404 color_attrib
|= S_028C74_NUM_SAMPLES(log_samples
) |
2405 S_028C74_NUM_FRAGMENTS(log_samples
);
2407 if (rtex
->fmask
.size
) {
2408 color_info
|= S_028C70_COMPRESSION(1);
2409 unsigned fmask_bankh
= util_logbase2(rtex
->fmask
.bank_height
);
2411 color_attrib
|= S_028C74_FMASK_TILE_MODE_INDEX(rtex
->fmask
.tile_mode_index
);
2413 if (sctx
->b
.chip_class
== SI
) {
2414 /* due to a hw bug, FMASK_BANK_HEIGHT must be set on SI too */
2415 color_attrib
|= S_028C74_FMASK_BANK_HEIGHT(fmask_bankh
);
2417 if (sctx
->b
.chip_class
>= CIK
) {
2418 color_pitch
|= S_028C64_FMASK_TILE_MAX(rtex
->fmask
.pitch_in_pixels
/ 8 - 1);
2423 offset
+= rtex
->resource
.gpu_address
;
2425 surf
->cb_color_base
= offset
>> 8;
2426 surf
->cb_color_pitch
= color_pitch
;
2427 surf
->cb_color_slice
= S_028C68_TILE_MAX(slice
);
2428 surf
->cb_color_view
= color_view
;
2429 surf
->cb_color_info
= color_info
;
2430 surf
->cb_color_attrib
= color_attrib
;
2432 if (sctx
->b
.chip_class
>= VI
&& rtex
->dcc_offset
) {
2433 unsigned max_uncompressed_block_size
= 2;
2435 if (rtex
->surface
.nsamples
> 1) {
2436 if (rtex
->surface
.bpe
== 1)
2437 max_uncompressed_block_size
= 0;
2438 else if (rtex
->surface
.bpe
== 2)
2439 max_uncompressed_block_size
= 1;
2442 surf
->cb_dcc_control
= S_028C78_MAX_UNCOMPRESSED_BLOCK_SIZE(max_uncompressed_block_size
) |
2443 S_028C78_INDEPENDENT_64B_BLOCKS(1);
2444 surf
->cb_dcc_base
= (rtex
->resource
.gpu_address
+
2446 rtex
->surface
.level
[level
].dcc_offset
) >> 8;
2449 if (rtex
->fmask
.size
) {
2450 surf
->cb_color_fmask
= (offset
+ rtex
->fmask
.offset
) >> 8;
2451 surf
->cb_color_fmask_slice
= S_028C88_TILE_MAX(rtex
->fmask
.slice_tile_max
);
2453 /* This must be set for fast clear to work without FMASK. */
2454 surf
->cb_color_fmask
= surf
->cb_color_base
;
2455 surf
->cb_color_fmask_slice
= surf
->cb_color_slice
;
2456 surf
->cb_color_attrib
|= S_028C74_FMASK_TILE_MODE_INDEX(tile_mode_index
);
2458 if (sctx
->b
.chip_class
== SI
) {
2459 unsigned bankh
= util_logbase2(rtex
->surface
.bankh
);
2460 surf
->cb_color_attrib
|= S_028C74_FMASK_BANK_HEIGHT(bankh
);
2463 if (sctx
->b
.chip_class
>= CIK
) {
2464 surf
->cb_color_pitch
|= S_028C64_FMASK_TILE_MAX(pitch
);
2468 /* Determine pixel shader export format */
2469 si_choose_spi_color_formats(surf
, format
, swap
, ntype
, rtex
->is_depth
);
2471 surf
->color_initialized
= true;
2474 static void si_init_depth_surface(struct si_context
*sctx
,
2475 struct r600_surface
*surf
)
2477 struct si_screen
*sscreen
= sctx
->screen
;
2478 struct r600_texture
*rtex
= (struct r600_texture
*)surf
->base
.texture
;
2479 unsigned level
= surf
->base
.u
.tex
.level
;
2480 struct radeon_surf_level
*levelinfo
= &rtex
->surface
.level
[level
];
2481 unsigned format
, tile_mode_index
, array_mode
;
2482 unsigned macro_aspect
, tile_split
, stile_split
, bankh
, bankw
, nbanks
, pipe_config
;
2483 uint32_t z_info
, s_info
, db_depth_info
;
2484 uint64_t z_offs
, s_offs
;
2485 uint32_t db_htile_data_base
, db_htile_surface
, pa_su_poly_offset_db_fmt_cntl
= 0;
2487 switch (sctx
->framebuffer
.state
.zsbuf
->texture
->format
) {
2488 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
2489 case PIPE_FORMAT_X8Z24_UNORM
:
2490 case PIPE_FORMAT_Z24X8_UNORM
:
2491 case PIPE_FORMAT_Z24_UNORM_S8_UINT
:
2492 pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-24);
2494 case PIPE_FORMAT_Z32_FLOAT
:
2495 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
2496 pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-23) |
2497 S_028B78_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
2499 case PIPE_FORMAT_Z16_UNORM
:
2500 pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-16);
2506 format
= si_translate_dbformat(rtex
->resource
.b
.b
.format
);
2508 if (format
== V_028040_Z_INVALID
) {
2509 R600_ERR("Invalid DB format: %d, disabling DB.\n", rtex
->resource
.b
.b
.format
);
2511 assert(format
!= V_028040_Z_INVALID
);
2513 s_offs
= z_offs
= rtex
->resource
.gpu_address
;
2514 z_offs
+= rtex
->surface
.level
[level
].offset
;
2515 s_offs
+= rtex
->surface
.stencil_level
[level
].offset
;
2517 db_depth_info
= S_02803C_ADDR5_SWIZZLE_MASK(1);
2519 z_info
= S_028040_FORMAT(format
);
2520 if (rtex
->resource
.b
.b
.nr_samples
> 1) {
2521 z_info
|= S_028040_NUM_SAMPLES(util_logbase2(rtex
->resource
.b
.b
.nr_samples
));
2524 if (rtex
->surface
.flags
& RADEON_SURF_SBUFFER
)
2525 s_info
= S_028044_FORMAT(V_028044_STENCIL_8
);
2527 s_info
= S_028044_FORMAT(V_028044_STENCIL_INVALID
);
2529 if (sctx
->b
.chip_class
>= CIK
) {
2530 switch (rtex
->surface
.level
[level
].mode
) {
2531 case RADEON_SURF_MODE_2D
:
2532 array_mode
= V_02803C_ARRAY_2D_TILED_THIN1
;
2534 case RADEON_SURF_MODE_1D
:
2535 case RADEON_SURF_MODE_LINEAR_ALIGNED
:
2536 case RADEON_SURF_MODE_LINEAR
:
2538 array_mode
= V_02803C_ARRAY_1D_TILED_THIN1
;
2541 tile_split
= rtex
->surface
.tile_split
;
2542 stile_split
= rtex
->surface
.stencil_tile_split
;
2543 macro_aspect
= rtex
->surface
.mtilea
;
2544 bankw
= rtex
->surface
.bankw
;
2545 bankh
= rtex
->surface
.bankh
;
2546 tile_split
= cik_tile_split(tile_split
);
2547 stile_split
= cik_tile_split(stile_split
);
2548 macro_aspect
= cik_macro_tile_aspect(macro_aspect
);
2549 bankw
= cik_bank_wh(bankw
);
2550 bankh
= cik_bank_wh(bankh
);
2551 nbanks
= si_num_banks(sscreen
, rtex
);
2552 tile_mode_index
= si_tile_mode_index(rtex
, level
, false);
2553 pipe_config
= cik_db_pipe_config(sscreen
, tile_mode_index
);
2555 db_depth_info
|= S_02803C_ARRAY_MODE(array_mode
) |
2556 S_02803C_PIPE_CONFIG(pipe_config
) |
2557 S_02803C_BANK_WIDTH(bankw
) |
2558 S_02803C_BANK_HEIGHT(bankh
) |
2559 S_02803C_MACRO_TILE_ASPECT(macro_aspect
) |
2560 S_02803C_NUM_BANKS(nbanks
);
2561 z_info
|= S_028040_TILE_SPLIT(tile_split
);
2562 s_info
|= S_028044_TILE_SPLIT(stile_split
);
2564 tile_mode_index
= si_tile_mode_index(rtex
, level
, false);
2565 z_info
|= S_028040_TILE_MODE_INDEX(tile_mode_index
);
2566 tile_mode_index
= si_tile_mode_index(rtex
, level
, true);
2567 s_info
|= S_028044_TILE_MODE_INDEX(tile_mode_index
);
2570 /* HiZ aka depth buffer htile */
2571 /* use htile only for first level */
2572 if (rtex
->htile_buffer
&& !level
) {
2573 z_info
|= S_028040_TILE_SURFACE_ENABLE(1) |
2574 S_028040_ALLOW_EXPCLEAR(1);
2576 if (rtex
->surface
.flags
& RADEON_SURF_SBUFFER
)
2577 s_info
|= S_028044_ALLOW_EXPCLEAR(1);
2579 /* Use all of the htile_buffer for depth if there's no stencil. */
2580 s_info
|= S_028044_TILE_STENCIL_DISABLE(1);
2582 uint64_t va
= rtex
->htile_buffer
->gpu_address
;
2583 db_htile_data_base
= va
>> 8;
2584 db_htile_surface
= S_028ABC_FULL_CACHE(1);
2586 db_htile_data_base
= 0;
2587 db_htile_surface
= 0;
2590 assert(levelinfo
->nblk_x
% 8 == 0 && levelinfo
->nblk_y
% 8 == 0);
2592 surf
->db_depth_view
= S_028008_SLICE_START(surf
->base
.u
.tex
.first_layer
) |
2593 S_028008_SLICE_MAX(surf
->base
.u
.tex
.last_layer
);
2594 surf
->db_htile_data_base
= db_htile_data_base
;
2595 surf
->db_depth_info
= db_depth_info
;
2596 surf
->db_z_info
= z_info
;
2597 surf
->db_stencil_info
= s_info
;
2598 surf
->db_depth_base
= z_offs
>> 8;
2599 surf
->db_stencil_base
= s_offs
>> 8;
2600 surf
->db_depth_size
= S_028058_PITCH_TILE_MAX((levelinfo
->nblk_x
/ 8) - 1) |
2601 S_028058_HEIGHT_TILE_MAX((levelinfo
->nblk_y
/ 8) - 1);
2602 surf
->db_depth_slice
= S_02805C_SLICE_TILE_MAX((levelinfo
->nblk_x
*
2603 levelinfo
->nblk_y
) / 64 - 1);
2604 surf
->db_htile_surface
= db_htile_surface
;
2605 surf
->pa_su_poly_offset_db_fmt_cntl
= pa_su_poly_offset_db_fmt_cntl
;
2607 surf
->depth_initialized
= true;
2610 static void si_set_framebuffer_state(struct pipe_context
*ctx
,
2611 const struct pipe_framebuffer_state
*state
)
2613 struct si_context
*sctx
= (struct si_context
*)ctx
;
2614 struct pipe_constant_buffer constbuf
= {0};
2615 struct r600_surface
*surf
= NULL
;
2616 struct r600_texture
*rtex
;
2617 bool old_cb0_is_integer
= sctx
->framebuffer
.cb0_is_integer
;
2618 unsigned old_nr_samples
= sctx
->framebuffer
.nr_samples
;
2621 /* Only flush TC when changing the framebuffer state, because
2622 * the only client not using TC that can change textures is
2625 * Flush all CB and DB caches here because all buffers can be used
2626 * for write by both TC (with shader image stores) and CB/DB.
2628 sctx
->b
.flags
|= SI_CONTEXT_INV_VMEM_L1
|
2629 SI_CONTEXT_INV_GLOBAL_L2
|
2630 SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER
;
2632 /* Take the maximum of the old and new count. If the new count is lower,
2633 * dirtying is needed to disable the unbound colorbuffers.
2635 sctx
->framebuffer
.dirty_cbufs
|=
2636 (1 << MAX2(sctx
->framebuffer
.state
.nr_cbufs
, state
->nr_cbufs
)) - 1;
2637 sctx
->framebuffer
.dirty_zsbuf
|= sctx
->framebuffer
.state
.zsbuf
!= state
->zsbuf
;
2639 util_copy_framebuffer_state(&sctx
->framebuffer
.state
, state
);
2641 sctx
->framebuffer
.spi_shader_col_format
= 0;
2642 sctx
->framebuffer
.spi_shader_col_format_alpha
= 0;
2643 sctx
->framebuffer
.spi_shader_col_format_blend
= 0;
2644 sctx
->framebuffer
.spi_shader_col_format_blend_alpha
= 0;
2645 sctx
->framebuffer
.color_is_int8
= 0;
2647 sctx
->framebuffer
.compressed_cb_mask
= 0;
2648 sctx
->framebuffer
.nr_samples
= util_framebuffer_get_num_samples(state
);
2649 sctx
->framebuffer
.log_samples
= util_logbase2(sctx
->framebuffer
.nr_samples
);
2650 sctx
->framebuffer
.cb0_is_integer
= state
->nr_cbufs
&& state
->cbufs
[0] &&
2651 util_format_is_pure_integer(state
->cbufs
[0]->format
);
2653 if (sctx
->framebuffer
.cb0_is_integer
!= old_cb0_is_integer
)
2654 si_mark_atom_dirty(sctx
, &sctx
->db_render_state
);
2656 for (i
= 0; i
< state
->nr_cbufs
; i
++) {
2657 if (!state
->cbufs
[i
])
2660 surf
= (struct r600_surface
*)state
->cbufs
[i
];
2661 rtex
= (struct r600_texture
*)surf
->base
.texture
;
2663 if (!surf
->color_initialized
) {
2664 si_initialize_color_surface(sctx
, surf
);
2667 sctx
->framebuffer
.spi_shader_col_format
|=
2668 surf
->spi_shader_col_format
<< (i
* 4);
2669 sctx
->framebuffer
.spi_shader_col_format_alpha
|=
2670 surf
->spi_shader_col_format_alpha
<< (i
* 4);
2671 sctx
->framebuffer
.spi_shader_col_format_blend
|=
2672 surf
->spi_shader_col_format_blend
<< (i
* 4);
2673 sctx
->framebuffer
.spi_shader_col_format_blend_alpha
|=
2674 surf
->spi_shader_col_format_blend_alpha
<< (i
* 4);
2676 if (surf
->color_is_int8
)
2677 sctx
->framebuffer
.color_is_int8
|= 1 << i
;
2679 if (rtex
->fmask
.size
&& rtex
->cmask
.size
) {
2680 sctx
->framebuffer
.compressed_cb_mask
|= 1 << i
;
2682 r600_context_add_resource_size(ctx
, surf
->base
.texture
);
2684 /* Set the second SPI format for possible dual-src blending. */
2685 if (i
== 1 && surf
) {
2686 sctx
->framebuffer
.spi_shader_col_format
|=
2687 surf
->spi_shader_col_format
<< (i
* 4);
2688 sctx
->framebuffer
.spi_shader_col_format_alpha
|=
2689 surf
->spi_shader_col_format_alpha
<< (i
* 4);
2690 sctx
->framebuffer
.spi_shader_col_format_blend
|=
2691 surf
->spi_shader_col_format_blend
<< (i
* 4);
2692 sctx
->framebuffer
.spi_shader_col_format_blend_alpha
|=
2693 surf
->spi_shader_col_format_blend_alpha
<< (i
* 4);
2697 surf
= (struct r600_surface
*)state
->zsbuf
;
2699 if (!surf
->depth_initialized
) {
2700 si_init_depth_surface(sctx
, surf
);
2702 r600_context_add_resource_size(ctx
, surf
->base
.texture
);
2705 si_update_poly_offset_state(sctx
);
2706 si_mark_atom_dirty(sctx
, &sctx
->cb_render_state
);
2707 si_mark_atom_dirty(sctx
, &sctx
->framebuffer
.atom
);
2709 if (sctx
->framebuffer
.nr_samples
!= old_nr_samples
) {
2710 si_mark_atom_dirty(sctx
, &sctx
->msaa_config
);
2711 si_mark_atom_dirty(sctx
, &sctx
->db_render_state
);
2713 /* Set sample locations as fragment shader constants. */
2714 switch (sctx
->framebuffer
.nr_samples
) {
2716 constbuf
.user_buffer
= sctx
->b
.sample_locations_1x
;
2719 constbuf
.user_buffer
= sctx
->b
.sample_locations_2x
;
2722 constbuf
.user_buffer
= sctx
->b
.sample_locations_4x
;
2725 constbuf
.user_buffer
= sctx
->b
.sample_locations_8x
;
2728 constbuf
.user_buffer
= sctx
->b
.sample_locations_16x
;
2731 R600_ERR("Requested an invalid number of samples %i.\n",
2732 sctx
->framebuffer
.nr_samples
);
2735 constbuf
.buffer_size
= sctx
->framebuffer
.nr_samples
* 2 * 4;
2736 ctx
->set_constant_buffer(ctx
, PIPE_SHADER_FRAGMENT
,
2737 SI_DRIVER_STATE_CONST_BUF
, &constbuf
);
2739 /* Smoothing (only possible with nr_samples == 1) uses the same
2740 * sample locations as the MSAA it simulates.
2742 * Therefore, don't update the sample locations when
2743 * transitioning from no AA to smoothing-equivalent AA, and
2746 if ((sctx
->framebuffer
.nr_samples
!= 1 ||
2747 old_nr_samples
!= SI_NUM_SMOOTH_AA_SAMPLES
) &&
2748 (sctx
->framebuffer
.nr_samples
!= SI_NUM_SMOOTH_AA_SAMPLES
||
2749 old_nr_samples
!= 1))
2750 si_mark_atom_dirty(sctx
, &sctx
->msaa_sample_locs
);
2754 static void si_emit_framebuffer_state(struct si_context
*sctx
, struct r600_atom
*atom
)
2756 struct radeon_winsys_cs
*cs
= sctx
->b
.gfx
.cs
;
2757 struct pipe_framebuffer_state
*state
= &sctx
->framebuffer
.state
;
2758 unsigned i
, nr_cbufs
= state
->nr_cbufs
;
2759 struct r600_texture
*tex
= NULL
;
2760 struct r600_surface
*cb
= NULL
;
2763 for (i
= 0; i
< nr_cbufs
; i
++) {
2764 if (!(sctx
->framebuffer
.dirty_cbufs
& (1 << i
)))
2767 cb
= (struct r600_surface
*)state
->cbufs
[i
];
2769 radeon_set_context_reg(cs
, R_028C70_CB_COLOR0_INFO
+ i
* 0x3C,
2770 S_028C70_FORMAT(V_028C70_COLOR_INVALID
));
2774 tex
= (struct r600_texture
*)cb
->base
.texture
;
2775 radeon_add_to_buffer_list(&sctx
->b
, &sctx
->b
.gfx
,
2776 &tex
->resource
, RADEON_USAGE_READWRITE
,
2777 tex
->surface
.nsamples
> 1 ?
2778 RADEON_PRIO_COLOR_BUFFER_MSAA
:
2779 RADEON_PRIO_COLOR_BUFFER
);
2781 if (tex
->cmask_buffer
&& tex
->cmask_buffer
!= &tex
->resource
) {
2782 radeon_add_to_buffer_list(&sctx
->b
, &sctx
->b
.gfx
,
2783 tex
->cmask_buffer
, RADEON_USAGE_READWRITE
,
2787 radeon_set_context_reg_seq(cs
, R_028C60_CB_COLOR0_BASE
+ i
* 0x3C,
2788 sctx
->b
.chip_class
>= VI
? 14 : 13);
2789 radeon_emit(cs
, cb
->cb_color_base
); /* R_028C60_CB_COLOR0_BASE */
2790 radeon_emit(cs
, cb
->cb_color_pitch
); /* R_028C64_CB_COLOR0_PITCH */
2791 radeon_emit(cs
, cb
->cb_color_slice
); /* R_028C68_CB_COLOR0_SLICE */
2792 radeon_emit(cs
, cb
->cb_color_view
); /* R_028C6C_CB_COLOR0_VIEW */
2793 radeon_emit(cs
, cb
->cb_color_info
| tex
->cb_color_info
); /* R_028C70_CB_COLOR0_INFO */
2794 radeon_emit(cs
, cb
->cb_color_attrib
); /* R_028C74_CB_COLOR0_ATTRIB */
2795 radeon_emit(cs
, cb
->cb_dcc_control
); /* R_028C78_CB_COLOR0_DCC_CONTROL */
2796 radeon_emit(cs
, tex
->cmask
.base_address_reg
); /* R_028C7C_CB_COLOR0_CMASK */
2797 radeon_emit(cs
, tex
->cmask
.slice_tile_max
); /* R_028C80_CB_COLOR0_CMASK_SLICE */
2798 radeon_emit(cs
, cb
->cb_color_fmask
); /* R_028C84_CB_COLOR0_FMASK */
2799 radeon_emit(cs
, cb
->cb_color_fmask_slice
); /* R_028C88_CB_COLOR0_FMASK_SLICE */
2800 radeon_emit(cs
, tex
->color_clear_value
[0]); /* R_028C8C_CB_COLOR0_CLEAR_WORD0 */
2801 radeon_emit(cs
, tex
->color_clear_value
[1]); /* R_028C90_CB_COLOR0_CLEAR_WORD1 */
2803 if (sctx
->b
.chip_class
>= VI
)
2804 radeon_emit(cs
, cb
->cb_dcc_base
); /* R_028C94_CB_COLOR0_DCC_BASE */
2806 /* set CB_COLOR1_INFO for possible dual-src blending */
2807 if (i
== 1 && state
->cbufs
[0] &&
2808 sctx
->framebuffer
.dirty_cbufs
& (1 << 0)) {
2809 radeon_set_context_reg(cs
, R_028C70_CB_COLOR0_INFO
+ 1 * 0x3C,
2810 cb
->cb_color_info
| tex
->cb_color_info
);
2814 if (sctx
->framebuffer
.dirty_cbufs
& (1 << i
))
2815 radeon_set_context_reg(cs
, R_028C70_CB_COLOR0_INFO
+ i
* 0x3C, 0);
2818 if (state
->zsbuf
&& sctx
->framebuffer
.dirty_zsbuf
) {
2819 struct r600_surface
*zb
= (struct r600_surface
*)state
->zsbuf
;
2820 struct r600_texture
*rtex
= (struct r600_texture
*)zb
->base
.texture
;
2822 radeon_add_to_buffer_list(&sctx
->b
, &sctx
->b
.gfx
,
2823 &rtex
->resource
, RADEON_USAGE_READWRITE
,
2824 zb
->base
.texture
->nr_samples
> 1 ?
2825 RADEON_PRIO_DEPTH_BUFFER_MSAA
:
2826 RADEON_PRIO_DEPTH_BUFFER
);
2828 if (zb
->db_htile_data_base
) {
2829 radeon_add_to_buffer_list(&sctx
->b
, &sctx
->b
.gfx
,
2830 rtex
->htile_buffer
, RADEON_USAGE_READWRITE
,
2834 radeon_set_context_reg(cs
, R_028008_DB_DEPTH_VIEW
, zb
->db_depth_view
);
2835 radeon_set_context_reg(cs
, R_028014_DB_HTILE_DATA_BASE
, zb
->db_htile_data_base
);
2837 radeon_set_context_reg_seq(cs
, R_02803C_DB_DEPTH_INFO
, 9);
2838 radeon_emit(cs
, zb
->db_depth_info
); /* R_02803C_DB_DEPTH_INFO */
2839 radeon_emit(cs
, zb
->db_z_info
| /* R_028040_DB_Z_INFO */
2840 S_028040_ZRANGE_PRECISION(rtex
->depth_clear_value
!= 0));
2841 radeon_emit(cs
, zb
->db_stencil_info
); /* R_028044_DB_STENCIL_INFO */
2842 radeon_emit(cs
, zb
->db_depth_base
); /* R_028048_DB_Z_READ_BASE */
2843 radeon_emit(cs
, zb
->db_stencil_base
); /* R_02804C_DB_STENCIL_READ_BASE */
2844 radeon_emit(cs
, zb
->db_depth_base
); /* R_028050_DB_Z_WRITE_BASE */
2845 radeon_emit(cs
, zb
->db_stencil_base
); /* R_028054_DB_STENCIL_WRITE_BASE */
2846 radeon_emit(cs
, zb
->db_depth_size
); /* R_028058_DB_DEPTH_SIZE */
2847 radeon_emit(cs
, zb
->db_depth_slice
); /* R_02805C_DB_DEPTH_SLICE */
2849 radeon_set_context_reg_seq(cs
, R_028028_DB_STENCIL_CLEAR
, 2);
2850 radeon_emit(cs
, rtex
->stencil_clear_value
); /* R_028028_DB_STENCIL_CLEAR */
2851 radeon_emit(cs
, fui(rtex
->depth_clear_value
)); /* R_02802C_DB_DEPTH_CLEAR */
2853 radeon_set_context_reg(cs
, R_028ABC_DB_HTILE_SURFACE
, zb
->db_htile_surface
);
2854 radeon_set_context_reg(cs
, R_028B78_PA_SU_POLY_OFFSET_DB_FMT_CNTL
,
2855 zb
->pa_su_poly_offset_db_fmt_cntl
);
2856 } else if (sctx
->framebuffer
.dirty_zsbuf
) {
2857 radeon_set_context_reg_seq(cs
, R_028040_DB_Z_INFO
, 2);
2858 radeon_emit(cs
, S_028040_FORMAT(V_028040_Z_INVALID
)); /* R_028040_DB_Z_INFO */
2859 radeon_emit(cs
, S_028044_FORMAT(V_028044_STENCIL_INVALID
)); /* R_028044_DB_STENCIL_INFO */
2862 /* Framebuffer dimensions. */
2863 /* PA_SC_WINDOW_SCISSOR_TL is set in si_init_config() */
2864 radeon_set_context_reg(cs
, R_028208_PA_SC_WINDOW_SCISSOR_BR
,
2865 S_028208_BR_X(state
->width
) | S_028208_BR_Y(state
->height
));
2867 sctx
->framebuffer
.dirty_cbufs
= 0;
2868 sctx
->framebuffer
.dirty_zsbuf
= false;
2871 static void si_emit_msaa_sample_locs(struct si_context
*sctx
,
2872 struct r600_atom
*atom
)
2874 struct radeon_winsys_cs
*cs
= sctx
->b
.gfx
.cs
;
2875 unsigned nr_samples
= sctx
->framebuffer
.nr_samples
;
2877 cayman_emit_msaa_sample_locs(cs
, nr_samples
> 1 ? nr_samples
:
2878 SI_NUM_SMOOTH_AA_SAMPLES
);
2881 static void si_emit_msaa_config(struct si_context
*sctx
, struct r600_atom
*atom
)
2883 struct radeon_winsys_cs
*cs
= sctx
->b
.gfx
.cs
;
2885 cayman_emit_msaa_config(cs
, sctx
->framebuffer
.nr_samples
,
2886 sctx
->ps_iter_samples
,
2887 sctx
->smoothing_enabled
? SI_NUM_SMOOTH_AA_SAMPLES
: 0);
2891 static void si_set_min_samples(struct pipe_context
*ctx
, unsigned min_samples
)
2893 struct si_context
*sctx
= (struct si_context
*)ctx
;
2895 if (sctx
->ps_iter_samples
== min_samples
)
2898 sctx
->ps_iter_samples
= min_samples
;
2900 if (sctx
->framebuffer
.nr_samples
> 1)
2901 si_mark_atom_dirty(sctx
, &sctx
->msaa_config
);
2909 * Build the sampler view descriptor for a buffer texture.
2910 * @param state 256-bit descriptor; only the high 128 bits are filled in
2913 si_make_buffer_descriptor(struct si_screen
*screen
, struct r600_resource
*buf
,
2914 enum pipe_format format
,
2915 unsigned first_element
, unsigned last_element
,
2918 const struct util_format_description
*desc
;
2922 unsigned num_records
;
2923 unsigned num_format
, data_format
;
2925 desc
= util_format_description(format
);
2926 first_non_void
= util_format_get_first_non_void_channel(format
);
2927 stride
= desc
->block
.bits
/ 8;
2928 va
= buf
->gpu_address
+ first_element
* stride
;
2929 num_format
= si_translate_buffer_numformat(&screen
->b
.b
, desc
, first_non_void
);
2930 data_format
= si_translate_buffer_dataformat(&screen
->b
.b
, desc
, first_non_void
);
2932 num_records
= last_element
+ 1 - first_element
;
2933 num_records
= MIN2(num_records
, buf
->b
.b
.width0
/ stride
);
2935 if (screen
->b
.chip_class
>= VI
)
2936 num_records
*= stride
;
2939 state
[5] = S_008F04_BASE_ADDRESS_HI(va
>> 32) |
2940 S_008F04_STRIDE(stride
);
2941 state
[6] = num_records
;
2942 state
[7] = S_008F0C_DST_SEL_X(si_map_swizzle(desc
->swizzle
[0])) |
2943 S_008F0C_DST_SEL_Y(si_map_swizzle(desc
->swizzle
[1])) |
2944 S_008F0C_DST_SEL_Z(si_map_swizzle(desc
->swizzle
[2])) |
2945 S_008F0C_DST_SEL_W(si_map_swizzle(desc
->swizzle
[3])) |
2946 S_008F0C_NUM_FORMAT(num_format
) |
2947 S_008F0C_DATA_FORMAT(data_format
);
2951 * Build the sampler view descriptor for a texture.
2954 si_make_texture_descriptor(struct si_screen
*screen
,
2955 struct r600_texture
*tex
,
2957 enum pipe_texture_target target
,
2958 enum pipe_format pipe_format
,
2959 const unsigned char state_swizzle
[4],
2960 unsigned base_level
, unsigned first_level
, unsigned last_level
,
2961 unsigned first_layer
, unsigned last_layer
,
2962 unsigned width
, unsigned height
, unsigned depth
,
2964 uint32_t *fmask_state
)
2966 struct pipe_resource
*res
= &tex
->resource
.b
.b
;
2967 const struct radeon_surf_level
*surflevel
= tex
->surface
.level
;
2968 const struct util_format_description
*desc
;
2969 unsigned char swizzle
[4];
2971 unsigned num_format
, data_format
, type
;
2975 /* Texturing with separate depth and stencil. */
2976 if (tex
->is_depth
&& !tex
->is_flushing_texture
) {
2977 switch (pipe_format
) {
2978 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
2979 pipe_format
= PIPE_FORMAT_Z32_FLOAT
;
2981 case PIPE_FORMAT_X8Z24_UNORM
:
2982 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
2983 /* Z24 is always stored like this. */
2984 pipe_format
= PIPE_FORMAT_Z24X8_UNORM
;
2986 case PIPE_FORMAT_X24S8_UINT
:
2987 case PIPE_FORMAT_S8X24_UINT
:
2988 case PIPE_FORMAT_X32_S8X24_UINT
:
2989 pipe_format
= PIPE_FORMAT_S8_UINT
;
2990 surflevel
= tex
->surface
.stencil_level
;
2996 desc
= util_format_description(pipe_format
);
2998 if (desc
->colorspace
== UTIL_FORMAT_COLORSPACE_ZS
) {
2999 const unsigned char swizzle_xxxx
[4] = {0, 0, 0, 0};
3000 const unsigned char swizzle_yyyy
[4] = {1, 1, 1, 1};
3002 switch (pipe_format
) {
3003 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
3004 case PIPE_FORMAT_X24S8_UINT
:
3005 case PIPE_FORMAT_X32_S8X24_UINT
:
3006 case PIPE_FORMAT_X8Z24_UNORM
:
3007 util_format_compose_swizzles(swizzle_yyyy
, state_swizzle
, swizzle
);
3010 util_format_compose_swizzles(swizzle_xxxx
, state_swizzle
, swizzle
);
3013 util_format_compose_swizzles(desc
->swizzle
, state_swizzle
, swizzle
);
3016 first_non_void
= util_format_get_first_non_void_channel(pipe_format
);
3018 switch (pipe_format
) {
3019 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
3020 num_format
= V_008F14_IMG_NUM_FORMAT_UNORM
;
3023 if (first_non_void
< 0) {
3024 if (util_format_is_compressed(pipe_format
)) {
3025 switch (pipe_format
) {
3026 case PIPE_FORMAT_DXT1_SRGB
:
3027 case PIPE_FORMAT_DXT1_SRGBA
:
3028 case PIPE_FORMAT_DXT3_SRGBA
:
3029 case PIPE_FORMAT_DXT5_SRGBA
:
3030 case PIPE_FORMAT_BPTC_SRGBA
:
3031 case PIPE_FORMAT_ETC2_SRGB8
:
3032 case PIPE_FORMAT_ETC2_SRGB8A1
:
3033 case PIPE_FORMAT_ETC2_SRGBA8
:
3034 num_format
= V_008F14_IMG_NUM_FORMAT_SRGB
;
3036 case PIPE_FORMAT_RGTC1_SNORM
:
3037 case PIPE_FORMAT_LATC1_SNORM
:
3038 case PIPE_FORMAT_RGTC2_SNORM
:
3039 case PIPE_FORMAT_LATC2_SNORM
:
3040 case PIPE_FORMAT_ETC2_R11_SNORM
:
3041 case PIPE_FORMAT_ETC2_RG11_SNORM
:
3042 /* implies float, so use SNORM/UNORM to determine
3043 whether data is signed or not */
3044 case PIPE_FORMAT_BPTC_RGB_FLOAT
:
3045 num_format
= V_008F14_IMG_NUM_FORMAT_SNORM
;
3048 num_format
= V_008F14_IMG_NUM_FORMAT_UNORM
;
3051 } else if (desc
->layout
== UTIL_FORMAT_LAYOUT_SUBSAMPLED
) {
3052 num_format
= V_008F14_IMG_NUM_FORMAT_UNORM
;
3054 num_format
= V_008F14_IMG_NUM_FORMAT_FLOAT
;
3056 } else if (desc
->colorspace
== UTIL_FORMAT_COLORSPACE_SRGB
) {
3057 num_format
= V_008F14_IMG_NUM_FORMAT_SRGB
;
3059 num_format
= V_008F14_IMG_NUM_FORMAT_UNORM
;
3061 switch (desc
->channel
[first_non_void
].type
) {
3062 case UTIL_FORMAT_TYPE_FLOAT
:
3063 num_format
= V_008F14_IMG_NUM_FORMAT_FLOAT
;
3065 case UTIL_FORMAT_TYPE_SIGNED
:
3066 if (desc
->channel
[first_non_void
].normalized
)
3067 num_format
= V_008F14_IMG_NUM_FORMAT_SNORM
;
3068 else if (desc
->channel
[first_non_void
].pure_integer
)
3069 num_format
= V_008F14_IMG_NUM_FORMAT_SINT
;
3071 num_format
= V_008F14_IMG_NUM_FORMAT_SSCALED
;
3073 case UTIL_FORMAT_TYPE_UNSIGNED
:
3074 if (desc
->channel
[first_non_void
].normalized
)
3075 num_format
= V_008F14_IMG_NUM_FORMAT_UNORM
;
3076 else if (desc
->channel
[first_non_void
].pure_integer
)
3077 num_format
= V_008F14_IMG_NUM_FORMAT_UINT
;
3079 num_format
= V_008F14_IMG_NUM_FORMAT_USCALED
;
3084 data_format
= si_translate_texformat(&screen
->b
.b
, pipe_format
, desc
, first_non_void
);
3085 if (data_format
== ~0) {
3090 (res
->target
== PIPE_TEXTURE_CUBE
||
3091 res
->target
== PIPE_TEXTURE_CUBE_ARRAY
||
3092 res
->target
== PIPE_TEXTURE_3D
)) {
3093 /* For the purpose of shader images, treat cube maps and 3D
3094 * textures as 2D arrays. For 3D textures, the address
3095 * calculations for mipmaps are different, so we rely on the
3096 * caller to effectively disable mipmaps.
3098 type
= V_008F1C_SQ_RSRC_IMG_2D_ARRAY
;
3100 assert(res
->target
!= PIPE_TEXTURE_3D
|| (first_level
== 0 && last_level
== 0));
3102 type
= si_tex_dim(res
->target
, target
, res
->nr_samples
);
3105 if (type
== V_008F1C_SQ_RSRC_IMG_1D_ARRAY
) {
3107 depth
= res
->array_size
;
3108 } else if (type
== V_008F1C_SQ_RSRC_IMG_2D_ARRAY
||
3109 type
== V_008F1C_SQ_RSRC_IMG_2D_MSAA_ARRAY
) {
3110 if (sampler
|| res
->target
!= PIPE_TEXTURE_3D
)
3111 depth
= res
->array_size
;
3112 } else if (type
== V_008F1C_SQ_RSRC_IMG_CUBE
)
3113 depth
= res
->array_size
/ 6;
3115 pitch
= surflevel
[base_level
].nblk_x
* util_format_get_blockwidth(pipe_format
);
3116 va
= tex
->resource
.gpu_address
+ surflevel
[base_level
].offset
;
3119 state
[1] = (S_008F14_BASE_ADDRESS_HI(va
>> 40) |
3120 S_008F14_DATA_FORMAT(data_format
) |
3121 S_008F14_NUM_FORMAT(num_format
));
3122 state
[2] = (S_008F18_WIDTH(width
- 1) |
3123 S_008F18_HEIGHT(height
- 1));
3124 state
[3] = (S_008F1C_DST_SEL_X(si_map_swizzle(swizzle
[0])) |
3125 S_008F1C_DST_SEL_Y(si_map_swizzle(swizzle
[1])) |
3126 S_008F1C_DST_SEL_Z(si_map_swizzle(swizzle
[2])) |
3127 S_008F1C_DST_SEL_W(si_map_swizzle(swizzle
[3])) |
3128 S_008F1C_BASE_LEVEL(res
->nr_samples
> 1 ?
3130 S_008F1C_LAST_LEVEL(res
->nr_samples
> 1 ?
3131 util_logbase2(res
->nr_samples
) :
3133 S_008F1C_TILING_INDEX(si_tile_mode_index(tex
, base_level
, false)) |
3134 S_008F1C_POW2_PAD(res
->last_level
> 0) |
3135 S_008F1C_TYPE(type
));
3136 state
[4] = (S_008F20_DEPTH(depth
- 1) | S_008F20_PITCH(pitch
- 1));
3137 state
[5] = (S_008F24_BASE_ARRAY(first_layer
) |
3138 S_008F24_LAST_ARRAY(last_layer
));
3140 if (tex
->dcc_offset
) {
3141 unsigned swap
= r600_translate_colorswap(pipe_format
);
3143 state
[6] = S_008F28_COMPRESSION_EN(1) | S_008F28_ALPHA_IS_ON_MSB(swap
<= 1);
3144 state
[7] = (tex
->resource
.gpu_address
+
3146 surflevel
[base_level
].dcc_offset
) >> 8;
3152 /* Initialize the sampler view for FMASK. */
3153 if (tex
->fmask
.size
) {
3154 uint32_t fmask_format
;
3156 va
= tex
->resource
.gpu_address
+ tex
->fmask
.offset
;
3158 switch (res
->nr_samples
) {
3160 fmask_format
= V_008F14_IMG_DATA_FORMAT_FMASK8_S2_F2
;
3163 fmask_format
= V_008F14_IMG_DATA_FORMAT_FMASK8_S4_F4
;
3166 fmask_format
= V_008F14_IMG_DATA_FORMAT_FMASK32_S8_F8
;
3170 fmask_format
= V_008F14_IMG_DATA_FORMAT_INVALID
;
3173 fmask_state
[0] = va
>> 8;
3174 fmask_state
[1] = S_008F14_BASE_ADDRESS_HI(va
>> 40) |
3175 S_008F14_DATA_FORMAT(fmask_format
) |
3176 S_008F14_NUM_FORMAT(V_008F14_IMG_NUM_FORMAT_UINT
);
3177 fmask_state
[2] = S_008F18_WIDTH(width
- 1) |
3178 S_008F18_HEIGHT(height
- 1);
3179 fmask_state
[3] = S_008F1C_DST_SEL_X(V_008F1C_SQ_SEL_X
) |
3180 S_008F1C_DST_SEL_Y(V_008F1C_SQ_SEL_X
) |
3181 S_008F1C_DST_SEL_Z(V_008F1C_SQ_SEL_X
) |
3182 S_008F1C_DST_SEL_W(V_008F1C_SQ_SEL_X
) |
3183 S_008F1C_TILING_INDEX(tex
->fmask
.tile_mode_index
) |
3184 S_008F1C_TYPE(si_tex_dim(res
->target
, target
, 0));
3185 fmask_state
[4] = S_008F20_DEPTH(depth
- 1) |
3186 S_008F20_PITCH(tex
->fmask
.pitch_in_pixels
- 1);
3187 fmask_state
[5] = S_008F24_BASE_ARRAY(first_layer
) |
3188 S_008F24_LAST_ARRAY(last_layer
);
3195 * Create a sampler view.
3197 * @param ctx context
3198 * @param texture texture
3199 * @param state sampler view template
3200 * @param width0 width0 override (for compressed textures as int)
3201 * @param height0 height0 override (for compressed textures as int)
3202 * @param force_level set the base address to the level (for compressed textures)
3204 struct pipe_sampler_view
*
3205 si_create_sampler_view_custom(struct pipe_context
*ctx
,
3206 struct pipe_resource
*texture
,
3207 const struct pipe_sampler_view
*state
,
3208 unsigned width0
, unsigned height0
,
3209 unsigned force_level
)
3211 struct si_context
*sctx
= (struct si_context
*)ctx
;
3212 struct si_sampler_view
*view
= CALLOC_STRUCT(si_sampler_view
);
3213 struct r600_texture
*tmp
= (struct r600_texture
*)texture
;
3214 unsigned base_level
, first_level
, last_level
;
3215 unsigned char state_swizzle
[4];
3216 unsigned height
, depth
, width
;
3217 unsigned last_layer
= state
->u
.tex
.last_layer
;
3222 /* initialize base object */
3223 view
->base
= *state
;
3224 view
->base
.texture
= NULL
;
3225 view
->base
.reference
.count
= 1;
3226 view
->base
.context
= ctx
;
3228 /* NULL resource, obey swizzle (only ZERO and ONE make sense). */
3230 view
->state
[3] = S_008F1C_DST_SEL_X(si_map_swizzle(state
->swizzle_r
)) |
3231 S_008F1C_DST_SEL_Y(si_map_swizzle(state
->swizzle_g
)) |
3232 S_008F1C_DST_SEL_Z(si_map_swizzle(state
->swizzle_b
)) |
3233 S_008F1C_DST_SEL_W(si_map_swizzle(state
->swizzle_a
)) |
3234 S_008F1C_TYPE(V_008F1C_SQ_RSRC_IMG_1D
);
3238 pipe_resource_reference(&view
->base
.texture
, texture
);
3240 if (state
->format
== PIPE_FORMAT_X24S8_UINT
||
3241 state
->format
== PIPE_FORMAT_S8X24_UINT
||
3242 state
->format
== PIPE_FORMAT_X32_S8X24_UINT
||
3243 state
->format
== PIPE_FORMAT_S8_UINT
)
3244 view
->is_stencil_sampler
= true;
3246 /* Buffer resource. */
3247 if (texture
->target
== PIPE_BUFFER
) {
3248 si_make_buffer_descriptor(sctx
->screen
,
3249 (struct r600_resource
*)texture
,
3251 state
->u
.buf
.first_element
,
3252 state
->u
.buf
.last_element
,
3255 LIST_ADDTAIL(&view
->list
, &sctx
->b
.texture_buffers
);
3259 state_swizzle
[0] = state
->swizzle_r
;
3260 state_swizzle
[1] = state
->swizzle_g
;
3261 state_swizzle
[2] = state
->swizzle_b
;
3262 state_swizzle
[3] = state
->swizzle_a
;
3265 first_level
= state
->u
.tex
.first_level
;
3266 last_level
= state
->u
.tex
.last_level
;
3269 depth
= texture
->depth0
;
3272 assert(force_level
== first_level
&&
3273 force_level
== last_level
);
3274 base_level
= force_level
;
3277 width
= u_minify(width
, force_level
);
3278 height
= u_minify(height
, force_level
);
3279 depth
= u_minify(depth
, force_level
);
3282 /* This is not needed if state trackers set last_layer correctly. */
3283 if (state
->target
== PIPE_TEXTURE_1D
||
3284 state
->target
== PIPE_TEXTURE_2D
||
3285 state
->target
== PIPE_TEXTURE_RECT
||
3286 state
->target
== PIPE_TEXTURE_CUBE
)
3287 last_layer
= state
->u
.tex
.first_layer
;
3289 si_make_texture_descriptor(sctx
->screen
, tmp
, true, state
->target
,
3290 state
->format
, state_swizzle
,
3291 base_level
, first_level
, last_level
,
3292 state
->u
.tex
.first_layer
, last_layer
,
3293 width
, height
, depth
,
3294 view
->state
, view
->fmask_state
);
3299 static struct pipe_sampler_view
*
3300 si_create_sampler_view(struct pipe_context
*ctx
,
3301 struct pipe_resource
*texture
,
3302 const struct pipe_sampler_view
*state
)
3304 return si_create_sampler_view_custom(ctx
, texture
, state
,
3305 texture
? texture
->width0
: 0,
3306 texture
? texture
->height0
: 0, 0);
3309 static void si_sampler_view_destroy(struct pipe_context
*ctx
,
3310 struct pipe_sampler_view
*state
)
3312 struct si_sampler_view
*view
= (struct si_sampler_view
*)state
;
3314 if (state
->texture
&& state
->texture
->target
== PIPE_BUFFER
)
3315 LIST_DELINIT(&view
->list
);
3317 pipe_resource_reference(&state
->texture
, NULL
);
3321 static bool wrap_mode_uses_border_color(unsigned wrap
, bool linear_filter
)
3323 return wrap
== PIPE_TEX_WRAP_CLAMP_TO_BORDER
||
3324 wrap
== PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
||
3326 (wrap
== PIPE_TEX_WRAP_CLAMP
||
3327 wrap
== PIPE_TEX_WRAP_MIRROR_CLAMP
));
3330 static bool sampler_state_needs_border_color(const struct pipe_sampler_state
*state
)
3332 bool linear_filter
= state
->min_img_filter
!= PIPE_TEX_FILTER_NEAREST
||
3333 state
->mag_img_filter
!= PIPE_TEX_FILTER_NEAREST
;
3335 return (state
->border_color
.ui
[0] || state
->border_color
.ui
[1] ||
3336 state
->border_color
.ui
[2] || state
->border_color
.ui
[3]) &&
3337 (wrap_mode_uses_border_color(state
->wrap_s
, linear_filter
) ||
3338 wrap_mode_uses_border_color(state
->wrap_t
, linear_filter
) ||
3339 wrap_mode_uses_border_color(state
->wrap_r
, linear_filter
));
3342 static void *si_create_sampler_state(struct pipe_context
*ctx
,
3343 const struct pipe_sampler_state
*state
)
3345 struct si_context
*sctx
= (struct si_context
*)ctx
;
3346 struct si_sampler_state
*rstate
= CALLOC_STRUCT(si_sampler_state
);
3347 unsigned aniso_flag_offset
= state
->max_anisotropy
> 1 ? 2 : 0;
3348 unsigned border_color_type
, border_color_index
= 0;
3354 if (!sampler_state_needs_border_color(state
))
3355 border_color_type
= V_008F3C_SQ_TEX_BORDER_COLOR_TRANS_BLACK
;
3356 else if (state
->border_color
.f
[0] == 0 &&
3357 state
->border_color
.f
[1] == 0 &&
3358 state
->border_color
.f
[2] == 0 &&
3359 state
->border_color
.f
[3] == 0)
3360 border_color_type
= V_008F3C_SQ_TEX_BORDER_COLOR_TRANS_BLACK
;
3361 else if (state
->border_color
.f
[0] == 0 &&
3362 state
->border_color
.f
[1] == 0 &&
3363 state
->border_color
.f
[2] == 0 &&
3364 state
->border_color
.f
[3] == 1)
3365 border_color_type
= V_008F3C_SQ_TEX_BORDER_COLOR_OPAQUE_BLACK
;
3366 else if (state
->border_color
.f
[0] == 1 &&
3367 state
->border_color
.f
[1] == 1 &&
3368 state
->border_color
.f
[2] == 1 &&
3369 state
->border_color
.f
[3] == 1)
3370 border_color_type
= V_008F3C_SQ_TEX_BORDER_COLOR_OPAQUE_WHITE
;
3374 border_color_type
= V_008F3C_SQ_TEX_BORDER_COLOR_REGISTER
;
3376 /* Check if the border has been uploaded already. */
3377 for (i
= 0; i
< sctx
->border_color_count
; i
++)
3378 if (memcmp(&sctx
->border_color_table
[i
], &state
->border_color
,
3379 sizeof(state
->border_color
)) == 0)
3382 if (i
>= SI_MAX_BORDER_COLORS
) {
3383 /* Getting 4096 unique border colors is very unlikely. */
3384 fprintf(stderr
, "radeonsi: The border color table is full. "
3385 "Any new border colors will be just black. "
3386 "Please file a bug.\n");
3387 border_color_type
= V_008F3C_SQ_TEX_BORDER_COLOR_TRANS_BLACK
;
3389 if (i
== sctx
->border_color_count
) {
3390 /* Upload a new border color. */
3391 memcpy(&sctx
->border_color_table
[i
], &state
->border_color
,
3392 sizeof(state
->border_color
));
3393 util_memcpy_cpu_to_le32(&sctx
->border_color_map
[i
],
3394 &state
->border_color
,
3395 sizeof(state
->border_color
));
3396 sctx
->border_color_count
++;
3399 border_color_index
= i
;
3403 rstate
->val
[0] = (S_008F30_CLAMP_X(si_tex_wrap(state
->wrap_s
)) |
3404 S_008F30_CLAMP_Y(si_tex_wrap(state
->wrap_t
)) |
3405 S_008F30_CLAMP_Z(si_tex_wrap(state
->wrap_r
)) |
3406 r600_tex_aniso_filter(state
->max_anisotropy
) << 9 |
3407 S_008F30_DEPTH_COMPARE_FUNC(si_tex_compare(state
->compare_func
)) |
3408 S_008F30_FORCE_UNNORMALIZED(!state
->normalized_coords
) |
3409 S_008F30_DISABLE_CUBE_WRAP(!state
->seamless_cube_map
));
3410 rstate
->val
[1] = (S_008F34_MIN_LOD(S_FIXED(CLAMP(state
->min_lod
, 0, 15), 8)) |
3411 S_008F34_MAX_LOD(S_FIXED(CLAMP(state
->max_lod
, 0, 15), 8)));
3412 rstate
->val
[2] = (S_008F38_LOD_BIAS(S_FIXED(CLAMP(state
->lod_bias
, -16, 16), 8)) |
3413 S_008F38_XY_MAG_FILTER(si_tex_filter(state
->mag_img_filter
) | aniso_flag_offset
) |
3414 S_008F38_XY_MIN_FILTER(si_tex_filter(state
->min_img_filter
) | aniso_flag_offset
) |
3415 S_008F38_MIP_FILTER(si_tex_mipfilter(state
->min_mip_filter
)));
3416 rstate
->val
[3] = S_008F3C_BORDER_COLOR_PTR(border_color_index
) |
3417 S_008F3C_BORDER_COLOR_TYPE(border_color_type
);
3421 static void si_set_sample_mask(struct pipe_context
*ctx
, unsigned sample_mask
)
3423 struct si_context
*sctx
= (struct si_context
*)ctx
;
3425 if (sctx
->sample_mask
.sample_mask
== (uint16_t)sample_mask
)
3428 sctx
->sample_mask
.sample_mask
= sample_mask
;
3429 si_mark_atom_dirty(sctx
, &sctx
->sample_mask
.atom
);
3432 static void si_emit_sample_mask(struct si_context
*sctx
, struct r600_atom
*atom
)
3434 struct radeon_winsys_cs
*cs
= sctx
->b
.gfx
.cs
;
3435 unsigned mask
= sctx
->sample_mask
.sample_mask
;
3437 radeon_set_context_reg_seq(cs
, R_028C38_PA_SC_AA_MASK_X0Y0_X1Y0
, 2);
3438 radeon_emit(cs
, mask
| (mask
<< 16));
3439 radeon_emit(cs
, mask
| (mask
<< 16));
3442 static void si_delete_sampler_state(struct pipe_context
*ctx
, void *state
)
3448 * Vertex elements & buffers
3451 static void *si_create_vertex_elements(struct pipe_context
*ctx
,
3453 const struct pipe_vertex_element
*elements
)
3455 struct si_vertex_element
*v
= CALLOC_STRUCT(si_vertex_element
);
3458 assert(count
< SI_MAX_ATTRIBS
);
3463 for (i
= 0; i
< count
; ++i
) {
3464 const struct util_format_description
*desc
;
3465 unsigned data_format
, num_format
;
3468 desc
= util_format_description(elements
[i
].src_format
);
3469 first_non_void
= util_format_get_first_non_void_channel(elements
[i
].src_format
);
3470 data_format
= si_translate_buffer_dataformat(ctx
->screen
, desc
, first_non_void
);
3471 num_format
= si_translate_buffer_numformat(ctx
->screen
, desc
, first_non_void
);
3473 v
->rsrc_word3
[i
] = S_008F0C_DST_SEL_X(si_map_swizzle(desc
->swizzle
[0])) |
3474 S_008F0C_DST_SEL_Y(si_map_swizzle(desc
->swizzle
[1])) |
3475 S_008F0C_DST_SEL_Z(si_map_swizzle(desc
->swizzle
[2])) |
3476 S_008F0C_DST_SEL_W(si_map_swizzle(desc
->swizzle
[3])) |
3477 S_008F0C_NUM_FORMAT(num_format
) |
3478 S_008F0C_DATA_FORMAT(data_format
);
3479 v
->format_size
[i
] = desc
->block
.bits
/ 8;
3481 memcpy(v
->elements
, elements
, sizeof(struct pipe_vertex_element
) * count
);
3486 static void si_bind_vertex_elements(struct pipe_context
*ctx
, void *state
)
3488 struct si_context
*sctx
= (struct si_context
*)ctx
;
3489 struct si_vertex_element
*v
= (struct si_vertex_element
*)state
;
3491 sctx
->vertex_elements
= v
;
3492 sctx
->vertex_buffers_dirty
= true;
3495 static void si_delete_vertex_element(struct pipe_context
*ctx
, void *state
)
3497 struct si_context
*sctx
= (struct si_context
*)ctx
;
3499 if (sctx
->vertex_elements
== state
)
3500 sctx
->vertex_elements
= NULL
;
3504 static void si_set_vertex_buffers(struct pipe_context
*ctx
,
3505 unsigned start_slot
, unsigned count
,
3506 const struct pipe_vertex_buffer
*buffers
)
3508 struct si_context
*sctx
= (struct si_context
*)ctx
;
3509 struct pipe_vertex_buffer
*dst
= sctx
->vertex_buffer
+ start_slot
;
3512 assert(start_slot
+ count
<= Elements(sctx
->vertex_buffer
));
3515 for (i
= 0; i
< count
; i
++) {
3516 const struct pipe_vertex_buffer
*src
= buffers
+ i
;
3517 struct pipe_vertex_buffer
*dsti
= dst
+ i
;
3519 pipe_resource_reference(&dsti
->buffer
, src
->buffer
);
3520 dsti
->buffer_offset
= src
->buffer_offset
;
3521 dsti
->stride
= src
->stride
;
3522 r600_context_add_resource_size(ctx
, src
->buffer
);
3525 for (i
= 0; i
< count
; i
++) {
3526 pipe_resource_reference(&dst
[i
].buffer
, NULL
);
3529 sctx
->vertex_buffers_dirty
= true;
3532 static void si_set_index_buffer(struct pipe_context
*ctx
,
3533 const struct pipe_index_buffer
*ib
)
3535 struct si_context
*sctx
= (struct si_context
*)ctx
;
3538 pipe_resource_reference(&sctx
->index_buffer
.buffer
, ib
->buffer
);
3539 memcpy(&sctx
->index_buffer
, ib
, sizeof(*ib
));
3540 r600_context_add_resource_size(ctx
, ib
->buffer
);
3542 pipe_resource_reference(&sctx
->index_buffer
.buffer
, NULL
);
3549 static void si_set_polygon_stipple(struct pipe_context
*ctx
,
3550 const struct pipe_poly_stipple
*state
)
3552 struct si_context
*sctx
= (struct si_context
*)ctx
;
3553 struct pipe_resource
*tex
;
3554 struct pipe_sampler_view
*view
;
3555 bool is_zero
= true;
3559 /* The hardware obeys 0 and 1 swizzles in the descriptor even if
3560 * the resource is NULL/invalid. Take advantage of this fact and skip
3561 * texture allocation if the stipple pattern is constant.
3563 * This is an optimization for the common case when stippling isn't
3564 * used but set_polygon_stipple is still called by st/mesa.
3566 for (i
= 0; i
< Elements(state
->stipple
); i
++) {
3567 is_zero
= is_zero
&& state
->stipple
[i
] == 0;
3568 is_one
= is_one
&& state
->stipple
[i
] == 0xffffffff;
3571 if (is_zero
|| is_one
) {
3572 struct pipe_sampler_view templ
= {{0}};
3574 templ
.swizzle_r
= PIPE_SWIZZLE_ZERO
;
3575 templ
.swizzle_g
= PIPE_SWIZZLE_ZERO
;
3576 templ
.swizzle_b
= PIPE_SWIZZLE_ZERO
;
3577 /* The pattern should be inverted in the texture. */
3578 templ
.swizzle_a
= is_zero
? PIPE_SWIZZLE_ONE
: PIPE_SWIZZLE_ZERO
;
3580 view
= ctx
->create_sampler_view(ctx
, NULL
, &templ
);
3582 /* Create a new texture. */
3583 tex
= util_pstipple_create_stipple_texture(ctx
, state
->stipple
);
3587 view
= util_pstipple_create_sampler_view(ctx
, tex
);
3588 pipe_resource_reference(&tex
, NULL
);
3591 ctx
->set_sampler_views(ctx
, PIPE_SHADER_FRAGMENT
,
3592 SI_POLY_STIPPLE_SAMPLER
, 1, &view
);
3593 pipe_sampler_view_reference(&view
, NULL
);
3595 /* Bind the sampler state if needed. */
3596 if (!sctx
->pstipple_sampler_state
) {
3597 sctx
->pstipple_sampler_state
= util_pstipple_create_sampler(ctx
);
3598 ctx
->bind_sampler_states(ctx
, PIPE_SHADER_FRAGMENT
,
3599 SI_POLY_STIPPLE_SAMPLER
, 1,
3600 &sctx
->pstipple_sampler_state
);
3604 static void si_set_tess_state(struct pipe_context
*ctx
,
3605 const float default_outer_level
[4],
3606 const float default_inner_level
[2])
3608 struct si_context
*sctx
= (struct si_context
*)ctx
;
3609 struct pipe_constant_buffer cb
;
3612 memcpy(array
, default_outer_level
, sizeof(float) * 4);
3613 memcpy(array
+4, default_inner_level
, sizeof(float) * 2);
3616 cb
.user_buffer
= NULL
;
3617 cb
.buffer_size
= sizeof(array
);
3619 si_upload_const_buffer(sctx
, (struct r600_resource
**)&cb
.buffer
,
3620 (void*)array
, sizeof(array
),
3623 ctx
->set_constant_buffer(ctx
, PIPE_SHADER_TESS_CTRL
,
3624 SI_DRIVER_STATE_CONST_BUF
, &cb
);
3625 pipe_resource_reference(&cb
.buffer
, NULL
);
3628 static void si_texture_barrier(struct pipe_context
*ctx
)
3630 struct si_context
*sctx
= (struct si_context
*)ctx
;
3632 sctx
->b
.flags
|= SI_CONTEXT_INV_VMEM_L1
|
3633 SI_CONTEXT_INV_GLOBAL_L2
|
3634 SI_CONTEXT_FLUSH_AND_INV_CB
;
3637 static void si_memory_barrier(struct pipe_context
*ctx
, unsigned flags
)
3639 struct si_context
*sctx
= (struct si_context
*)ctx
;
3641 /* Subsequent commands must wait for all shader invocations to
3643 sctx
->b
.flags
|= SI_CONTEXT_PS_PARTIAL_FLUSH
;
3645 if (flags
& PIPE_BARRIER_CONSTANT_BUFFER
)
3646 sctx
->b
.flags
|= SI_CONTEXT_INV_SMEM_L1
|
3647 SI_CONTEXT_INV_VMEM_L1
;
3649 if (flags
& (PIPE_BARRIER_VERTEX_BUFFER
|
3650 PIPE_BARRIER_SHADER_BUFFER
|
3651 PIPE_BARRIER_TEXTURE
|
3652 PIPE_BARRIER_IMAGE
|
3653 PIPE_BARRIER_STREAMOUT_BUFFER
)) {
3654 /* As far as I can tell, L1 contents are written back to L2
3655 * automatically at end of shader, but the contents of other
3656 * L1 caches might still be stale. */
3657 sctx
->b
.flags
|= SI_CONTEXT_INV_VMEM_L1
;
3660 if (flags
& PIPE_BARRIER_INDEX_BUFFER
) {
3661 sctx
->b
.flags
|= SI_CONTEXT_INV_VMEM_L1
;
3663 /* Indices are read through TC L2 since VI. */
3664 if (sctx
->screen
->b
.chip_class
<= CIK
)
3665 sctx
->b
.flags
|= SI_CONTEXT_INV_GLOBAL_L2
;
3668 if (flags
& PIPE_BARRIER_FRAMEBUFFER
)
3669 sctx
->b
.flags
|= SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER
;
3671 if (flags
& (PIPE_BARRIER_MAPPED_BUFFER
|
3672 PIPE_BARRIER_FRAMEBUFFER
|
3673 PIPE_BARRIER_INDIRECT_BUFFER
)) {
3674 /* Not sure if INV_GLOBAL_L2 is the best thing here.
3676 * We need to make sure that TC L1 & L2 are written back to
3677 * memory, because neither CPU accesses nor CB fetches consider
3678 * TC, but there's no need to invalidate any TC cache lines. */
3679 sctx
->b
.flags
|= SI_CONTEXT_INV_GLOBAL_L2
;
3683 static void *si_create_blend_custom(struct si_context
*sctx
, unsigned mode
)
3685 struct pipe_blend_state blend
;
3687 memset(&blend
, 0, sizeof(blend
));
3688 blend
.independent_blend_enable
= true;
3689 blend
.rt
[0].colormask
= 0xf;
3690 return si_create_blend_state_mode(&sctx
->b
.b
, &blend
, mode
);
3693 static void si_need_gfx_cs_space(struct pipe_context
*ctx
, unsigned num_dw
,
3694 bool include_draw_vbo
)
3696 si_need_cs_space((struct si_context
*)ctx
);
3699 static void si_init_config(struct si_context
*sctx
);
3701 void si_init_state_functions(struct si_context
*sctx
)
3703 si_init_external_atom(sctx
, &sctx
->b
.render_cond_atom
, &sctx
->atoms
.s
.render_cond
);
3704 si_init_external_atom(sctx
, &sctx
->b
.streamout
.begin_atom
, &sctx
->atoms
.s
.streamout_begin
);
3705 si_init_external_atom(sctx
, &sctx
->b
.streamout
.enable_atom
, &sctx
->atoms
.s
.streamout_enable
);
3707 si_init_atom(sctx
, &sctx
->cache_flush
, &sctx
->atoms
.s
.cache_flush
, si_emit_cache_flush
);
3708 si_init_atom(sctx
, &sctx
->framebuffer
.atom
, &sctx
->atoms
.s
.framebuffer
, si_emit_framebuffer_state
);
3709 si_init_atom(sctx
, &sctx
->msaa_sample_locs
, &sctx
->atoms
.s
.msaa_sample_locs
, si_emit_msaa_sample_locs
);
3710 si_init_atom(sctx
, &sctx
->db_render_state
, &sctx
->atoms
.s
.db_render_state
, si_emit_db_render_state
);
3711 si_init_atom(sctx
, &sctx
->msaa_config
, &sctx
->atoms
.s
.msaa_config
, si_emit_msaa_config
);
3712 si_init_atom(sctx
, &sctx
->sample_mask
.atom
, &sctx
->atoms
.s
.sample_mask
, si_emit_sample_mask
);
3713 si_init_atom(sctx
, &sctx
->cb_render_state
, &sctx
->atoms
.s
.cb_render_state
, si_emit_cb_render_state
);
3714 si_init_atom(sctx
, &sctx
->blend_color
.atom
, &sctx
->atoms
.s
.blend_color
, si_emit_blend_color
);
3715 si_init_atom(sctx
, &sctx
->clip_regs
, &sctx
->atoms
.s
.clip_regs
, si_emit_clip_regs
);
3716 si_init_atom(sctx
, &sctx
->clip_state
.atom
, &sctx
->atoms
.s
.clip_state
, si_emit_clip_state
);
3717 si_init_atom(sctx
, &sctx
->scissors
.atom
, &sctx
->atoms
.s
.scissors
, si_emit_scissors
);
3718 si_init_atom(sctx
, &sctx
->viewports
.atom
, &sctx
->atoms
.s
.viewports
, si_emit_viewports
);
3719 si_init_atom(sctx
, &sctx
->stencil_ref
.atom
, &sctx
->atoms
.s
.stencil_ref
, si_emit_stencil_ref
);
3721 sctx
->b
.b
.create_blend_state
= si_create_blend_state
;
3722 sctx
->b
.b
.bind_blend_state
= si_bind_blend_state
;
3723 sctx
->b
.b
.delete_blend_state
= si_delete_blend_state
;
3724 sctx
->b
.b
.set_blend_color
= si_set_blend_color
;
3726 sctx
->b
.b
.create_rasterizer_state
= si_create_rs_state
;
3727 sctx
->b
.b
.bind_rasterizer_state
= si_bind_rs_state
;
3728 sctx
->b
.b
.delete_rasterizer_state
= si_delete_rs_state
;
3730 sctx
->b
.b
.create_depth_stencil_alpha_state
= si_create_dsa_state
;
3731 sctx
->b
.b
.bind_depth_stencil_alpha_state
= si_bind_dsa_state
;
3732 sctx
->b
.b
.delete_depth_stencil_alpha_state
= si_delete_dsa_state
;
3734 sctx
->custom_dsa_flush
= si_create_db_flush_dsa(sctx
);
3735 sctx
->custom_blend_resolve
= si_create_blend_custom(sctx
, V_028808_CB_RESOLVE
);
3736 sctx
->custom_blend_decompress
= si_create_blend_custom(sctx
, V_028808_CB_FMASK_DECOMPRESS
);
3737 sctx
->custom_blend_fastclear
= si_create_blend_custom(sctx
, V_028808_CB_ELIMINATE_FAST_CLEAR
);
3738 sctx
->custom_blend_dcc_decompress
= si_create_blend_custom(sctx
, V_028808_CB_DCC_DECOMPRESS
);
3740 sctx
->b
.b
.set_clip_state
= si_set_clip_state
;
3741 sctx
->b
.b
.set_scissor_states
= si_set_scissor_states
;
3742 sctx
->b
.b
.set_viewport_states
= si_set_viewport_states
;
3743 sctx
->b
.b
.set_stencil_ref
= si_set_stencil_ref
;
3745 sctx
->b
.b
.set_framebuffer_state
= si_set_framebuffer_state
;
3746 sctx
->b
.b
.get_sample_position
= cayman_get_sample_position
;
3748 sctx
->b
.b
.create_sampler_state
= si_create_sampler_state
;
3749 sctx
->b
.b
.delete_sampler_state
= si_delete_sampler_state
;
3751 sctx
->b
.b
.create_sampler_view
= si_create_sampler_view
;
3752 sctx
->b
.b
.sampler_view_destroy
= si_sampler_view_destroy
;
3754 sctx
->b
.b
.set_sample_mask
= si_set_sample_mask
;
3756 sctx
->b
.b
.create_vertex_elements_state
= si_create_vertex_elements
;
3757 sctx
->b
.b
.bind_vertex_elements_state
= si_bind_vertex_elements
;
3758 sctx
->b
.b
.delete_vertex_elements_state
= si_delete_vertex_element
;
3759 sctx
->b
.b
.set_vertex_buffers
= si_set_vertex_buffers
;
3760 sctx
->b
.b
.set_index_buffer
= si_set_index_buffer
;
3762 sctx
->b
.b
.texture_barrier
= si_texture_barrier
;
3763 sctx
->b
.b
.memory_barrier
= si_memory_barrier
;
3764 sctx
->b
.b
.set_polygon_stipple
= si_set_polygon_stipple
;
3765 sctx
->b
.b
.set_min_samples
= si_set_min_samples
;
3766 sctx
->b
.b
.set_tess_state
= si_set_tess_state
;
3768 sctx
->b
.b
.set_active_query_state
= si_set_active_query_state
;
3769 sctx
->b
.set_occlusion_query_state
= si_set_occlusion_query_state
;
3770 sctx
->b
.need_gfx_cs_space
= si_need_gfx_cs_space
;
3772 sctx
->b
.b
.draw_vbo
= si_draw_vbo
;
3774 if (sctx
->b
.chip_class
>= CIK
) {
3775 sctx
->b
.dma_copy
= cik_sdma_copy
;
3777 sctx
->b
.dma_copy
= si_dma_copy
;
3780 si_init_config(sctx
);
3783 static void si_query_opaque_metadata(struct r600_common_screen
*rscreen
,
3784 struct r600_texture
*rtex
,
3785 struct radeon_bo_metadata
*md
)
3787 struct si_screen
*sscreen
= (struct si_screen
*)rscreen
;
3788 struct pipe_resource
*res
= &rtex
->resource
.b
.b
;
3789 static const unsigned char swizzle
[] = {
3795 uint32_t desc
[8], i
;
3796 bool is_array
= util_resource_is_array_texture(res
);
3798 /* DRM 2.x.x doesn't support this. */
3799 if (rscreen
->info
.drm_major
!= 3)
3802 assert(rtex
->fmask
.size
== 0);
3804 /* Metadata image format format version 1:
3805 * [0] = 1 (metadata format identifier)
3806 * [1] = (VENDOR_ID << 16) | PCI_ID
3807 * [2:9] = image descriptor for the whole resource
3808 * [2] is always 0, because the base address is cleared
3809 * [9] is the DCC offset bits [39:8] from the beginning of
3811 * [10:10+LAST_LEVEL] = mipmap level offset bits [39:8] for each level
3814 md
->metadata
[0] = 1; /* metadata image format version 1 */
3816 /* TILE_MODE_INDEX is ambiguous without a PCI ID. */
3817 md
->metadata
[1] = (ATI_VENDOR_ID
<< 16) | rscreen
->info
.pci_id
;
3819 si_make_texture_descriptor(sscreen
, rtex
, true,
3820 res
->target
, res
->format
,
3821 swizzle
, 0, 0, res
->last_level
, 0,
3822 is_array
? res
->array_size
- 1 : 0,
3823 res
->width0
, res
->height0
, res
->depth0
,
3826 /* Clear the base address and set the relative DCC offset. */
3828 desc
[1] &= C_008F14_BASE_ADDRESS_HI
;
3829 desc
[7] = rtex
->dcc_offset
>> 8;
3831 /* Dwords [2:9] contain the image descriptor. */
3832 memcpy(&md
->metadata
[2], desc
, sizeof(desc
));
3834 /* Dwords [10:..] contain the mipmap level offsets. */
3835 for (i
= 0; i
<= res
->last_level
; i
++)
3836 md
->metadata
[10+i
] = rtex
->surface
.level
[i
].offset
>> 8;
3838 md
->size_metadata
= (11 + res
->last_level
) * 4;
3841 void si_init_screen_state_functions(struct si_screen
*sscreen
)
3843 sscreen
->b
.query_opaque_metadata
= si_query_opaque_metadata
;
3847 si_write_harvested_raster_configs(struct si_context
*sctx
,
3848 struct si_pm4_state
*pm4
,
3849 unsigned raster_config
,
3850 unsigned raster_config_1
)
3852 unsigned sh_per_se
= MAX2(sctx
->screen
->b
.info
.max_sh_per_se
, 1);
3853 unsigned num_se
= MAX2(sctx
->screen
->b
.info
.max_se
, 1);
3854 unsigned rb_mask
= sctx
->screen
->b
.info
.enabled_rb_mask
;
3855 unsigned num_rb
= MIN2(sctx
->screen
->b
.info
.num_render_backends
, 16);
3856 unsigned rb_per_pkr
= MIN2(num_rb
/ num_se
/ sh_per_se
, 2);
3857 unsigned rb_per_se
= num_rb
/ num_se
;
3858 unsigned se_mask
[4];
3861 se_mask
[0] = ((1 << rb_per_se
) - 1) & rb_mask
;
3862 se_mask
[1] = (se_mask
[0] << rb_per_se
) & rb_mask
;
3863 se_mask
[2] = (se_mask
[1] << rb_per_se
) & rb_mask
;
3864 se_mask
[3] = (se_mask
[2] << rb_per_se
) & rb_mask
;
3866 assert(num_se
== 1 || num_se
== 2 || num_se
== 4);
3867 assert(sh_per_se
== 1 || sh_per_se
== 2);
3868 assert(rb_per_pkr
== 1 || rb_per_pkr
== 2);
3870 /* XXX: I can't figure out what the *_XSEL and *_YSEL
3871 * fields are for, so I'm leaving them as their default
3874 if ((num_se
> 2) && ((!se_mask
[0] && !se_mask
[1]) ||
3875 (!se_mask
[2] && !se_mask
[3]))) {
3876 raster_config_1
&= C_028354_SE_PAIR_MAP
;
3878 if (!se_mask
[0] && !se_mask
[1]) {
3880 S_028354_SE_PAIR_MAP(V_028354_RASTER_CONFIG_SE_PAIR_MAP_3
);
3883 S_028354_SE_PAIR_MAP(V_028354_RASTER_CONFIG_SE_PAIR_MAP_0
);
3887 for (se
= 0; se
< num_se
; se
++) {
3888 unsigned raster_config_se
= raster_config
;
3889 unsigned pkr0_mask
= ((1 << rb_per_pkr
) - 1) << (se
* rb_per_se
);
3890 unsigned pkr1_mask
= pkr0_mask
<< rb_per_pkr
;
3891 int idx
= (se
/ 2) * 2;
3893 if ((num_se
> 1) && (!se_mask
[idx
] || !se_mask
[idx
+ 1])) {
3894 raster_config_se
&= C_028350_SE_MAP
;
3896 if (!se_mask
[idx
]) {
3898 S_028350_SE_MAP(V_028350_RASTER_CONFIG_SE_MAP_3
);
3901 S_028350_SE_MAP(V_028350_RASTER_CONFIG_SE_MAP_0
);
3905 pkr0_mask
&= rb_mask
;
3906 pkr1_mask
&= rb_mask
;
3907 if (rb_per_se
> 2 && (!pkr0_mask
|| !pkr1_mask
)) {
3908 raster_config_se
&= C_028350_PKR_MAP
;
3912 S_028350_PKR_MAP(V_028350_RASTER_CONFIG_PKR_MAP_3
);
3915 S_028350_PKR_MAP(V_028350_RASTER_CONFIG_PKR_MAP_0
);
3919 if (rb_per_se
>= 2) {
3920 unsigned rb0_mask
= 1 << (se
* rb_per_se
);
3921 unsigned rb1_mask
= rb0_mask
<< 1;
3923 rb0_mask
&= rb_mask
;
3924 rb1_mask
&= rb_mask
;
3925 if (!rb0_mask
|| !rb1_mask
) {
3926 raster_config_se
&= C_028350_RB_MAP_PKR0
;
3930 S_028350_RB_MAP_PKR0(V_028350_RASTER_CONFIG_RB_MAP_3
);
3933 S_028350_RB_MAP_PKR0(V_028350_RASTER_CONFIG_RB_MAP_0
);
3937 if (rb_per_se
> 2) {
3938 rb0_mask
= 1 << (se
* rb_per_se
+ rb_per_pkr
);
3939 rb1_mask
= rb0_mask
<< 1;
3940 rb0_mask
&= rb_mask
;
3941 rb1_mask
&= rb_mask
;
3942 if (!rb0_mask
|| !rb1_mask
) {
3943 raster_config_se
&= C_028350_RB_MAP_PKR1
;
3947 S_028350_RB_MAP_PKR1(V_028350_RASTER_CONFIG_RB_MAP_3
);
3950 S_028350_RB_MAP_PKR1(V_028350_RASTER_CONFIG_RB_MAP_0
);
3956 /* GRBM_GFX_INDEX has a different offset on SI and CI+ */
3957 if (sctx
->b
.chip_class
< CIK
)
3958 si_pm4_set_reg(pm4
, GRBM_GFX_INDEX
,
3959 SE_INDEX(se
) | SH_BROADCAST_WRITES
|
3960 INSTANCE_BROADCAST_WRITES
);
3962 si_pm4_set_reg(pm4
, R_030800_GRBM_GFX_INDEX
,
3963 S_030800_SE_INDEX(se
) | S_030800_SH_BROADCAST_WRITES(1) |
3964 S_030800_INSTANCE_BROADCAST_WRITES(1));
3965 si_pm4_set_reg(pm4
, R_028350_PA_SC_RASTER_CONFIG
, raster_config_se
);
3966 if (sctx
->b
.chip_class
>= CIK
)
3967 si_pm4_set_reg(pm4
, R_028354_PA_SC_RASTER_CONFIG_1
, raster_config_1
);
3970 /* GRBM_GFX_INDEX has a different offset on SI and CI+ */
3971 if (sctx
->b
.chip_class
< CIK
)
3972 si_pm4_set_reg(pm4
, GRBM_GFX_INDEX
,
3973 SE_BROADCAST_WRITES
| SH_BROADCAST_WRITES
|
3974 INSTANCE_BROADCAST_WRITES
);
3976 si_pm4_set_reg(pm4
, R_030800_GRBM_GFX_INDEX
,
3977 S_030800_SE_BROADCAST_WRITES(1) | S_030800_SH_BROADCAST_WRITES(1) |
3978 S_030800_INSTANCE_BROADCAST_WRITES(1));
3981 static void si_init_config(struct si_context
*sctx
)
3983 struct si_screen
*sscreen
= sctx
->screen
;
3984 unsigned num_rb
= MIN2(sctx
->screen
->b
.info
.num_render_backends
, 16);
3985 unsigned rb_mask
= sctx
->screen
->b
.info
.enabled_rb_mask
;
3986 unsigned raster_config
, raster_config_1
;
3987 uint64_t border_color_va
= sctx
->border_color_buffer
->gpu_address
;
3988 struct si_pm4_state
*pm4
= CALLOC_STRUCT(si_pm4_state
);
3994 si_pm4_cmd_begin(pm4
, PKT3_CONTEXT_CONTROL
);
3995 si_pm4_cmd_add(pm4
, 0x80000000);
3996 si_pm4_cmd_add(pm4
, 0x80000000);
3997 si_pm4_cmd_end(pm4
, false);
3999 /* This enables pipeline stat & streamout queries.
4000 * They are only disabled by blits.
4002 si_pm4_cmd_begin(pm4
, PKT3_EVENT_WRITE
);
4003 si_pm4_cmd_add(pm4
, EVENT_TYPE(V_028A90_PIPELINESTAT_START
) |
4005 si_pm4_cmd_end(pm4
, false);
4007 si_pm4_set_reg(pm4
, R_028A18_VGT_HOS_MAX_TESS_LEVEL
, fui(64));
4008 si_pm4_set_reg(pm4
, R_028A1C_VGT_HOS_MIN_TESS_LEVEL
, fui(0));
4010 /* FIXME calculate these values somehow ??? */
4011 si_pm4_set_reg(pm4
, R_028A54_VGT_GS_PER_ES
, SI_GS_PER_ES
);
4012 si_pm4_set_reg(pm4
, R_028A58_VGT_ES_PER_GS
, 0x40);
4013 si_pm4_set_reg(pm4
, R_028A5C_VGT_GS_PER_VS
, 0x2);
4015 si_pm4_set_reg(pm4
, R_028A8C_VGT_PRIMITIVEID_RESET
, 0x0);
4016 si_pm4_set_reg(pm4
, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET
, 0);
4018 si_pm4_set_reg(pm4
, R_028B98_VGT_STRMOUT_BUFFER_CONFIG
, 0x0);
4019 si_pm4_set_reg(pm4
, R_028AB8_VGT_VTX_CNT_EN
, 0x0);
4020 if (sctx
->b
.chip_class
< CIK
)
4021 si_pm4_set_reg(pm4
, R_008A14_PA_CL_ENHANCE
, S_008A14_NUM_CLIP_SEQ(3) |
4022 S_008A14_CLIP_VTX_REORDER_ENA(1));
4024 si_pm4_set_reg(pm4
, R_028BD4_PA_SC_CENTROID_PRIORITY_0
, 0x76543210);
4025 si_pm4_set_reg(pm4
, R_028BD8_PA_SC_CENTROID_PRIORITY_1
, 0xfedcba98);
4027 si_pm4_set_reg(pm4
, R_02882C_PA_SU_PRIM_FILTER_CNTL
, 0);
4029 for (i
= 0; i
< 16; i
++) {
4030 si_pm4_set_reg(pm4
, R_0282D0_PA_SC_VPORT_ZMIN_0
+ i
*8, 0);
4031 si_pm4_set_reg(pm4
, R_0282D4_PA_SC_VPORT_ZMAX_0
+ i
*8, fui(1.0));
4034 switch (sctx
->screen
->b
.family
) {
4037 raster_config
= 0x2a00126a;
4038 raster_config_1
= 0x00000000;
4041 raster_config
= 0x0000124a;
4042 raster_config_1
= 0x00000000;
4045 raster_config
= 0x00000082;
4046 raster_config_1
= 0x00000000;
4049 raster_config
= 0x00000000;
4050 raster_config_1
= 0x00000000;
4053 raster_config
= 0x16000012;
4054 raster_config_1
= 0x00000000;
4057 raster_config
= 0x3a00161a;
4058 raster_config_1
= 0x0000002e;
4061 if (sscreen
->b
.info
.cik_macrotile_mode_array
[0] == 0x000000e8) {
4062 /* old kernels with old tiling config */
4063 raster_config
= 0x16000012;
4064 raster_config_1
= 0x0000002a;
4066 raster_config
= 0x3a00161a;
4067 raster_config_1
= 0x0000002e;
4070 case CHIP_POLARIS10
:
4071 raster_config
= 0x16000012;
4072 raster_config_1
= 0x0000002a;
4074 case CHIP_POLARIS11
:
4075 raster_config
= 0x16000012;
4076 raster_config_1
= 0x00000000;
4079 raster_config
= 0x16000012;
4080 raster_config_1
= 0x0000002a;
4083 raster_config
= 0x00000002;
4084 raster_config_1
= 0x00000000;
4087 raster_config
= 0x00000002;
4088 raster_config_1
= 0x00000000;
4091 /* KV should be 0x00000002, but that causes problems with radeon */
4092 raster_config
= 0x00000000; /* 0x00000002 */
4093 raster_config_1
= 0x00000000;
4098 raster_config
= 0x00000000;
4099 raster_config_1
= 0x00000000;
4103 "radeonsi: Unknown GPU, using 0 for raster_config\n");
4104 raster_config
= 0x00000000;
4105 raster_config_1
= 0x00000000;
4109 /* Always use the default config when all backends are enabled
4110 * (or when we failed to determine the enabled backends).
4112 if (!rb_mask
|| util_bitcount(rb_mask
) >= num_rb
) {
4113 si_pm4_set_reg(pm4
, R_028350_PA_SC_RASTER_CONFIG
,
4115 if (sctx
->b
.chip_class
>= CIK
)
4116 si_pm4_set_reg(pm4
, R_028354_PA_SC_RASTER_CONFIG_1
,
4119 si_write_harvested_raster_configs(sctx
, pm4
, raster_config
, raster_config_1
);
4122 si_pm4_set_reg(pm4
, R_028204_PA_SC_WINDOW_SCISSOR_TL
, S_028204_WINDOW_OFFSET_DISABLE(1));
4123 si_pm4_set_reg(pm4
, R_028240_PA_SC_GENERIC_SCISSOR_TL
, S_028240_WINDOW_OFFSET_DISABLE(1));
4124 si_pm4_set_reg(pm4
, R_028244_PA_SC_GENERIC_SCISSOR_BR
,
4125 S_028244_BR_X(16384) | S_028244_BR_Y(16384));
4126 si_pm4_set_reg(pm4
, R_028030_PA_SC_SCREEN_SCISSOR_TL
, 0);
4127 si_pm4_set_reg(pm4
, R_028034_PA_SC_SCREEN_SCISSOR_BR
,
4128 S_028034_BR_X(16384) | S_028034_BR_Y(16384));
4130 si_pm4_set_reg(pm4
, R_02820C_PA_SC_CLIPRECT_RULE
, 0xFFFF);
4131 si_pm4_set_reg(pm4
, R_028230_PA_SC_EDGERULE
, 0xAAAAAAAA);
4132 /* PA_SU_HARDWARE_SCREEN_OFFSET must be 0 due to hw bug on SI */
4133 si_pm4_set_reg(pm4
, R_028234_PA_SU_HARDWARE_SCREEN_OFFSET
, 0);
4134 si_pm4_set_reg(pm4
, R_028820_PA_CL_NANINF_CNTL
, 0);
4135 si_pm4_set_reg(pm4
, R_028BE8_PA_CL_GB_VERT_CLIP_ADJ
, fui(1.0));
4136 si_pm4_set_reg(pm4
, R_028BEC_PA_CL_GB_VERT_DISC_ADJ
, fui(1.0));
4137 si_pm4_set_reg(pm4
, R_028BF0_PA_CL_GB_HORZ_CLIP_ADJ
, fui(1.0));
4138 si_pm4_set_reg(pm4
, R_028BF4_PA_CL_GB_HORZ_DISC_ADJ
, fui(1.0));
4139 si_pm4_set_reg(pm4
, R_028AC0_DB_SRESULTS_COMPARE_STATE0
, 0x0);
4140 si_pm4_set_reg(pm4
, R_028AC4_DB_SRESULTS_COMPARE_STATE1
, 0x0);
4141 si_pm4_set_reg(pm4
, R_028AC8_DB_PRELOAD_CONTROL
, 0x0);
4142 si_pm4_set_reg(pm4
, R_02800C_DB_RENDER_OVERRIDE
,
4143 S_02800C_FORCE_HIS_ENABLE0(V_02800C_FORCE_DISABLE
) |
4144 S_02800C_FORCE_HIS_ENABLE1(V_02800C_FORCE_DISABLE
));
4146 si_pm4_set_reg(pm4
, R_028400_VGT_MAX_VTX_INDX
, ~0);
4147 si_pm4_set_reg(pm4
, R_028404_VGT_MIN_VTX_INDX
, 0);
4148 si_pm4_set_reg(pm4
, R_028408_VGT_INDX_OFFSET
, 0);
4150 if (sctx
->b
.chip_class
>= CIK
) {
4151 si_pm4_set_reg(pm4
, R_00B41C_SPI_SHADER_PGM_RSRC3_HS
, 0);
4152 si_pm4_set_reg(pm4
, R_00B31C_SPI_SHADER_PGM_RSRC3_ES
, S_00B31C_CU_EN(0xffff));
4153 si_pm4_set_reg(pm4
, R_00B21C_SPI_SHADER_PGM_RSRC3_GS
, S_00B21C_CU_EN(0xffff));
4155 if (sscreen
->b
.info
.num_good_compute_units
/
4156 (sscreen
->b
.info
.max_se
* sscreen
->b
.info
.max_sh_per_se
) <= 4) {
4157 /* Too few available compute units per SH. Disallowing
4158 * VS to run on CU0 could hurt us more than late VS
4159 * allocation would help.
4161 * LATE_ALLOC_VS = 2 is the highest safe number.
4163 si_pm4_set_reg(pm4
, R_00B51C_SPI_SHADER_PGM_RSRC3_LS
, S_00B51C_CU_EN(0xffff));
4164 si_pm4_set_reg(pm4
, R_00B118_SPI_SHADER_PGM_RSRC3_VS
, S_00B118_CU_EN(0xffff));
4165 si_pm4_set_reg(pm4
, R_00B11C_SPI_SHADER_LATE_ALLOC_VS
, S_00B11C_LIMIT(2));
4167 /* Set LATE_ALLOC_VS == 31. It should be less than
4168 * the number of scratch waves. Limitations:
4169 * - VS can't execute on CU0.
4170 * - If HS writes outputs to LDS, LS can't execute on CU0.
4172 si_pm4_set_reg(pm4
, R_00B51C_SPI_SHADER_PGM_RSRC3_LS
, S_00B51C_CU_EN(0xfffe));
4173 si_pm4_set_reg(pm4
, R_00B118_SPI_SHADER_PGM_RSRC3_VS
, S_00B118_CU_EN(0xfffe));
4174 si_pm4_set_reg(pm4
, R_00B11C_SPI_SHADER_LATE_ALLOC_VS
, S_00B11C_LIMIT(31));
4177 si_pm4_set_reg(pm4
, R_00B01C_SPI_SHADER_PGM_RSRC3_PS
, S_00B01C_CU_EN(0xffff));
4180 if (sctx
->b
.chip_class
>= VI
) {
4181 si_pm4_set_reg(pm4
, R_028424_CB_DCC_CONTROL
,
4182 S_028424_OVERWRITE_COMBINER_MRT_SHARING_DISABLE(1) |
4183 S_028424_OVERWRITE_COMBINER_WATERMARK(4));
4184 si_pm4_set_reg(pm4
, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL
, 30);
4185 si_pm4_set_reg(pm4
, R_028C5C_VGT_OUT_DEALLOC_CNTL
, 32);
4188 if (sctx
->b
.family
== CHIP_STONEY
)
4189 si_pm4_set_reg(pm4
, R_028C40_PA_SC_SHADER_CONTROL
, 0);
4191 si_pm4_set_reg(pm4
, R_028080_TA_BC_BASE_ADDR
, border_color_va
>> 8);
4192 if (sctx
->b
.chip_class
>= CIK
)
4193 si_pm4_set_reg(pm4
, R_028084_TA_BC_BASE_ADDR_HI
, border_color_va
>> 40);
4194 si_pm4_add_bo(pm4
, sctx
->border_color_buffer
, RADEON_USAGE_READ
,
4195 RADEON_PRIO_BORDER_COLORS
);
4197 si_pm4_upload_indirect_buffer(sctx
, pm4
);
4198 sctx
->init_config
= pm4
;