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 "tgsi/tgsi_parse.h"
33 #include "tgsi/tgsi_scan.h"
34 #include "util/u_format.h"
35 #include "util/u_format_s3tc.h"
36 #include "util/u_framebuffer.h"
37 #include "util/u_helpers.h"
38 #include "util/u_memory.h"
40 static uint32_t cik_num_banks(struct si_screen
*sscreen
, unsigned bpe
, unsigned tile_split
)
42 if (sscreen
->b
.info
.cik_macrotile_mode_array_valid
) {
43 unsigned index
, tileb
;
46 tileb
= MIN2(tile_split
, tileb
);
48 for (index
= 0; tileb
> 64; index
++) {
54 return (sscreen
->b
.info
.cik_macrotile_mode_array
[index
] >> 6) & 0x3;
58 switch (sscreen
->b
.tiling_info
.num_banks
) {
60 return V_02803C_ADDR_SURF_2_BANK
;
62 return V_02803C_ADDR_SURF_4_BANK
;
65 return V_02803C_ADDR_SURF_8_BANK
;
67 return V_02803C_ADDR_SURF_16_BANK
;
71 static unsigned cik_tile_split(unsigned tile_split
)
75 tile_split
= V_028040_ADDR_SURF_TILE_SPLIT_64B
;
78 tile_split
= V_028040_ADDR_SURF_TILE_SPLIT_128B
;
81 tile_split
= V_028040_ADDR_SURF_TILE_SPLIT_256B
;
84 tile_split
= V_028040_ADDR_SURF_TILE_SPLIT_512B
;
88 tile_split
= V_028040_ADDR_SURF_TILE_SPLIT_1KB
;
91 tile_split
= V_028040_ADDR_SURF_TILE_SPLIT_2KB
;
94 tile_split
= V_028040_ADDR_SURF_TILE_SPLIT_4KB
;
100 static unsigned cik_macro_tile_aspect(unsigned macro_tile_aspect
)
102 switch (macro_tile_aspect
) {
105 macro_tile_aspect
= V_02803C_ADDR_SURF_MACRO_ASPECT_1
;
108 macro_tile_aspect
= V_02803C_ADDR_SURF_MACRO_ASPECT_2
;
111 macro_tile_aspect
= V_02803C_ADDR_SURF_MACRO_ASPECT_4
;
114 macro_tile_aspect
= V_02803C_ADDR_SURF_MACRO_ASPECT_8
;
117 return macro_tile_aspect
;
120 static unsigned cik_bank_wh(unsigned bankwh
)
125 bankwh
= V_02803C_ADDR_SURF_BANK_WIDTH_1
;
128 bankwh
= V_02803C_ADDR_SURF_BANK_WIDTH_2
;
131 bankwh
= V_02803C_ADDR_SURF_BANK_WIDTH_4
;
134 bankwh
= V_02803C_ADDR_SURF_BANK_WIDTH_8
;
140 static unsigned cik_db_pipe_config(struct si_screen
*sscreen
, unsigned tile_mode
)
142 if (sscreen
->b
.info
.si_tile_mode_array_valid
) {
143 uint32_t gb_tile_mode
= sscreen
->b
.info
.si_tile_mode_array
[tile_mode
];
145 return G_009910_PIPE_CONFIG(gb_tile_mode
);
148 /* This is probably broken for a lot of chips, but it's only used
149 * if the kernel cannot return the tile mode array for CIK. */
150 switch (sscreen
->b
.info
.r600_num_tile_pipes
) {
152 return V_02803C_X_ADDR_SURF_P16_32X32_16X16
;
154 return V_02803C_X_ADDR_SURF_P8_32X32_16X16
;
157 if (sscreen
->b
.info
.r600_num_backends
== 4)
158 return V_02803C_X_ADDR_SURF_P4_16X16
;
160 return V_02803C_X_ADDR_SURF_P4_8X16
;
162 return V_02803C_ADDR_SURF_P2
;
166 static unsigned si_map_swizzle(unsigned swizzle
)
169 case UTIL_FORMAT_SWIZZLE_Y
:
170 return V_008F0C_SQ_SEL_Y
;
171 case UTIL_FORMAT_SWIZZLE_Z
:
172 return V_008F0C_SQ_SEL_Z
;
173 case UTIL_FORMAT_SWIZZLE_W
:
174 return V_008F0C_SQ_SEL_W
;
175 case UTIL_FORMAT_SWIZZLE_0
:
176 return V_008F0C_SQ_SEL_0
;
177 case UTIL_FORMAT_SWIZZLE_1
:
178 return V_008F0C_SQ_SEL_1
;
179 default: /* UTIL_FORMAT_SWIZZLE_X */
180 return V_008F0C_SQ_SEL_X
;
184 static uint32_t S_FIXED(float value
, uint32_t frac_bits
)
186 return value
* (1 << frac_bits
);
189 /* 12.4 fixed-point */
190 static unsigned si_pack_float_12p4(float x
)
193 x
>= 4096 ? 0xffff : x
* 16;
197 * inferred framebuffer and blender state
199 static void si_update_fb_blend_state(struct si_context
*sctx
)
201 struct si_pm4_state
*pm4
;
202 struct si_state_blend
*blend
= sctx
->queued
.named
.blend
;
208 pm4
= si_pm4_alloc_state(sctx
);
212 mask
= (1ULL << ((unsigned)sctx
->framebuffer
.nr_cbufs
* 4)) - 1;
213 mask
&= blend
->cb_target_mask
;
214 si_pm4_set_reg(pm4
, R_028238_CB_TARGET_MASK
, mask
);
216 si_pm4_set_state(sctx
, fb_blend
, pm4
);
223 static uint32_t si_translate_blend_function(int blend_func
)
225 switch (blend_func
) {
227 return V_028780_COMB_DST_PLUS_SRC
;
228 case PIPE_BLEND_SUBTRACT
:
229 return V_028780_COMB_SRC_MINUS_DST
;
230 case PIPE_BLEND_REVERSE_SUBTRACT
:
231 return V_028780_COMB_DST_MINUS_SRC
;
233 return V_028780_COMB_MIN_DST_SRC
;
235 return V_028780_COMB_MAX_DST_SRC
;
237 R600_ERR("Unknown blend function %d\n", blend_func
);
244 static uint32_t si_translate_blend_factor(int blend_fact
)
246 switch (blend_fact
) {
247 case PIPE_BLENDFACTOR_ONE
:
248 return V_028780_BLEND_ONE
;
249 case PIPE_BLENDFACTOR_SRC_COLOR
:
250 return V_028780_BLEND_SRC_COLOR
;
251 case PIPE_BLENDFACTOR_SRC_ALPHA
:
252 return V_028780_BLEND_SRC_ALPHA
;
253 case PIPE_BLENDFACTOR_DST_ALPHA
:
254 return V_028780_BLEND_DST_ALPHA
;
255 case PIPE_BLENDFACTOR_DST_COLOR
:
256 return V_028780_BLEND_DST_COLOR
;
257 case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
:
258 return V_028780_BLEND_SRC_ALPHA_SATURATE
;
259 case PIPE_BLENDFACTOR_CONST_COLOR
:
260 return V_028780_BLEND_CONSTANT_COLOR
;
261 case PIPE_BLENDFACTOR_CONST_ALPHA
:
262 return V_028780_BLEND_CONSTANT_ALPHA
;
263 case PIPE_BLENDFACTOR_ZERO
:
264 return V_028780_BLEND_ZERO
;
265 case PIPE_BLENDFACTOR_INV_SRC_COLOR
:
266 return V_028780_BLEND_ONE_MINUS_SRC_COLOR
;
267 case PIPE_BLENDFACTOR_INV_SRC_ALPHA
:
268 return V_028780_BLEND_ONE_MINUS_SRC_ALPHA
;
269 case PIPE_BLENDFACTOR_INV_DST_ALPHA
:
270 return V_028780_BLEND_ONE_MINUS_DST_ALPHA
;
271 case PIPE_BLENDFACTOR_INV_DST_COLOR
:
272 return V_028780_BLEND_ONE_MINUS_DST_COLOR
;
273 case PIPE_BLENDFACTOR_INV_CONST_COLOR
:
274 return V_028780_BLEND_ONE_MINUS_CONSTANT_COLOR
;
275 case PIPE_BLENDFACTOR_INV_CONST_ALPHA
:
276 return V_028780_BLEND_ONE_MINUS_CONSTANT_ALPHA
;
277 case PIPE_BLENDFACTOR_SRC1_COLOR
:
278 return V_028780_BLEND_SRC1_COLOR
;
279 case PIPE_BLENDFACTOR_SRC1_ALPHA
:
280 return V_028780_BLEND_SRC1_ALPHA
;
281 case PIPE_BLENDFACTOR_INV_SRC1_COLOR
:
282 return V_028780_BLEND_INV_SRC1_COLOR
;
283 case PIPE_BLENDFACTOR_INV_SRC1_ALPHA
:
284 return V_028780_BLEND_INV_SRC1_ALPHA
;
286 R600_ERR("Bad blend factor %d not supported!\n", blend_fact
);
293 static void *si_create_blend_state_mode(struct pipe_context
*ctx
,
294 const struct pipe_blend_state
*state
,
297 struct si_state_blend
*blend
= CALLOC_STRUCT(si_state_blend
);
298 struct si_pm4_state
*pm4
= &blend
->pm4
;
300 uint32_t color_control
= 0;
305 blend
->alpha_to_one
= state
->alpha_to_one
;
307 if (state
->logicop_enable
) {
308 color_control
|= S_028808_ROP3(state
->logicop_func
| (state
->logicop_func
<< 4));
310 color_control
|= S_028808_ROP3(0xcc);
313 si_pm4_set_reg(pm4
, R_028B70_DB_ALPHA_TO_MASK
,
314 S_028B70_ALPHA_TO_MASK_ENABLE(state
->alpha_to_coverage
) |
315 S_028B70_ALPHA_TO_MASK_OFFSET0(2) |
316 S_028B70_ALPHA_TO_MASK_OFFSET1(2) |
317 S_028B70_ALPHA_TO_MASK_OFFSET2(2) |
318 S_028B70_ALPHA_TO_MASK_OFFSET3(2));
320 blend
->cb_target_mask
= 0;
321 for (int i
= 0; i
< 8; i
++) {
322 /* state->rt entries > 0 only written if independent blending */
323 const int j
= state
->independent_blend_enable
? i
: 0;
325 unsigned eqRGB
= state
->rt
[j
].rgb_func
;
326 unsigned srcRGB
= state
->rt
[j
].rgb_src_factor
;
327 unsigned dstRGB
= state
->rt
[j
].rgb_dst_factor
;
328 unsigned eqA
= state
->rt
[j
].alpha_func
;
329 unsigned srcA
= state
->rt
[j
].alpha_src_factor
;
330 unsigned dstA
= state
->rt
[j
].alpha_dst_factor
;
332 unsigned blend_cntl
= 0;
334 /* we pretend 8 buffer are used, CB_SHADER_MASK will disable unused one */
335 blend
->cb_target_mask
|= state
->rt
[j
].colormask
<< (4 * i
);
337 if (!state
->rt
[j
].blend_enable
) {
338 si_pm4_set_reg(pm4
, R_028780_CB_BLEND0_CONTROL
+ i
* 4, blend_cntl
);
342 blend_cntl
|= S_028780_ENABLE(1);
343 blend_cntl
|= S_028780_COLOR_COMB_FCN(si_translate_blend_function(eqRGB
));
344 blend_cntl
|= S_028780_COLOR_SRCBLEND(si_translate_blend_factor(srcRGB
));
345 blend_cntl
|= S_028780_COLOR_DESTBLEND(si_translate_blend_factor(dstRGB
));
347 if (srcA
!= srcRGB
|| dstA
!= dstRGB
|| eqA
!= eqRGB
) {
348 blend_cntl
|= S_028780_SEPARATE_ALPHA_BLEND(1);
349 blend_cntl
|= S_028780_ALPHA_COMB_FCN(si_translate_blend_function(eqA
));
350 blend_cntl
|= S_028780_ALPHA_SRCBLEND(si_translate_blend_factor(srcA
));
351 blend_cntl
|= S_028780_ALPHA_DESTBLEND(si_translate_blend_factor(dstA
));
353 si_pm4_set_reg(pm4
, R_028780_CB_BLEND0_CONTROL
+ i
* 4, blend_cntl
);
356 if (blend
->cb_target_mask
) {
357 color_control
|= S_028808_MODE(mode
);
359 color_control
|= S_028808_MODE(V_028808_CB_DISABLE
);
361 si_pm4_set_reg(pm4
, R_028808_CB_COLOR_CONTROL
, color_control
);
366 static void *si_create_blend_state(struct pipe_context
*ctx
,
367 const struct pipe_blend_state
*state
)
369 return si_create_blend_state_mode(ctx
, state
, V_028808_CB_NORMAL
);
372 static void si_bind_blend_state(struct pipe_context
*ctx
, void *state
)
374 struct si_context
*sctx
= (struct si_context
*)ctx
;
375 si_pm4_bind_state(sctx
, blend
, (struct si_state_blend
*)state
);
376 si_update_fb_blend_state(sctx
);
379 static void si_delete_blend_state(struct pipe_context
*ctx
, void *state
)
381 struct si_context
*sctx
= (struct si_context
*)ctx
;
382 si_pm4_delete_state(sctx
, blend
, (struct si_state_blend
*)state
);
385 static void si_set_blend_color(struct pipe_context
*ctx
,
386 const struct pipe_blend_color
*state
)
388 struct si_context
*sctx
= (struct si_context
*)ctx
;
389 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
394 si_pm4_set_reg(pm4
, R_028414_CB_BLEND_RED
, fui(state
->color
[0]));
395 si_pm4_set_reg(pm4
, R_028418_CB_BLEND_GREEN
, fui(state
->color
[1]));
396 si_pm4_set_reg(pm4
, R_02841C_CB_BLEND_BLUE
, fui(state
->color
[2]));
397 si_pm4_set_reg(pm4
, R_028420_CB_BLEND_ALPHA
, fui(state
->color
[3]));
399 si_pm4_set_state(sctx
, blend_color
, pm4
);
403 * Clipping, scissors and viewport
406 static void si_set_clip_state(struct pipe_context
*ctx
,
407 const struct pipe_clip_state
*state
)
409 struct si_context
*sctx
= (struct si_context
*)ctx
;
410 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
411 struct pipe_constant_buffer cb
;
416 for (int i
= 0; i
< 6; i
++) {
417 si_pm4_set_reg(pm4
, R_0285BC_PA_CL_UCP_0_X
+ i
* 16,
418 fui(state
->ucp
[i
][0]));
419 si_pm4_set_reg(pm4
, R_0285C0_PA_CL_UCP_0_Y
+ i
* 16,
420 fui(state
->ucp
[i
][1]));
421 si_pm4_set_reg(pm4
, R_0285C4_PA_CL_UCP_0_Z
+ i
* 16,
422 fui(state
->ucp
[i
][2]));
423 si_pm4_set_reg(pm4
, R_0285C8_PA_CL_UCP_0_W
+ i
* 16,
424 fui(state
->ucp
[i
][3]));
428 cb
.user_buffer
= state
->ucp
;
429 cb
.buffer_offset
= 0;
430 cb
.buffer_size
= 4*4*8;
431 ctx
->set_constant_buffer(ctx
, PIPE_SHADER_VERTEX
, NUM_PIPE_CONST_BUFFERS
, &cb
);
432 pipe_resource_reference(&cb
.buffer
, NULL
);
434 si_pm4_set_state(sctx
, clip
, pm4
);
437 static void si_set_scissor_states(struct pipe_context
*ctx
,
439 unsigned num_scissors
,
440 const struct pipe_scissor_state
*state
)
442 struct si_context
*sctx
= (struct si_context
*)ctx
;
443 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
448 si_pm4_set_reg(pm4
, R_028250_PA_SC_VPORT_SCISSOR_0_TL
,
449 S_028250_TL_X(state
->minx
) | S_028250_TL_Y(state
->miny
) |
450 S_028250_WINDOW_OFFSET_DISABLE(1));
451 si_pm4_set_reg(pm4
, R_028254_PA_SC_VPORT_SCISSOR_0_BR
,
452 S_028254_BR_X(state
->maxx
) | S_028254_BR_Y(state
->maxy
));
454 si_pm4_set_state(sctx
, scissor
, pm4
);
457 static void si_set_viewport_states(struct pipe_context
*ctx
,
459 unsigned num_viewports
,
460 const struct pipe_viewport_state
*state
)
462 struct si_context
*sctx
= (struct si_context
*)ctx
;
463 struct si_state_viewport
*viewport
= CALLOC_STRUCT(si_state_viewport
);
464 struct si_pm4_state
*pm4
= &viewport
->pm4
;
466 if (viewport
== NULL
)
469 viewport
->viewport
= *state
;
470 si_pm4_set_reg(pm4
, R_02843C_PA_CL_VPORT_XSCALE_0
, fui(state
->scale
[0]));
471 si_pm4_set_reg(pm4
, R_028440_PA_CL_VPORT_XOFFSET_0
, fui(state
->translate
[0]));
472 si_pm4_set_reg(pm4
, R_028444_PA_CL_VPORT_YSCALE_0
, fui(state
->scale
[1]));
473 si_pm4_set_reg(pm4
, R_028448_PA_CL_VPORT_YOFFSET_0
, fui(state
->translate
[1]));
474 si_pm4_set_reg(pm4
, R_02844C_PA_CL_VPORT_ZSCALE_0
, fui(state
->scale
[2]));
475 si_pm4_set_reg(pm4
, R_028450_PA_CL_VPORT_ZOFFSET_0
, fui(state
->translate
[2]));
477 si_pm4_set_state(sctx
, viewport
, viewport
);
481 * inferred state between framebuffer and rasterizer
483 static void si_update_fb_rs_state(struct si_context
*sctx
)
485 struct si_state_rasterizer
*rs
= sctx
->queued
.named
.rasterizer
;
486 struct si_pm4_state
*pm4
;
489 if (!rs
|| !sctx
->framebuffer
.zsbuf
)
492 offset_units
= sctx
->queued
.named
.rasterizer
->offset_units
;
493 switch (sctx
->framebuffer
.zsbuf
->texture
->format
) {
494 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
495 case PIPE_FORMAT_X8Z24_UNORM
:
496 case PIPE_FORMAT_Z24X8_UNORM
:
497 case PIPE_FORMAT_Z24_UNORM_S8_UINT
:
498 offset_units
*= 2.0f
;
500 case PIPE_FORMAT_Z32_FLOAT
:
501 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
502 offset_units
*= 1.0f
;
504 case PIPE_FORMAT_Z16_UNORM
:
505 offset_units
*= 4.0f
;
511 pm4
= si_pm4_alloc_state(sctx
);
516 /* FIXME some of those reg can be computed with cso */
517 si_pm4_set_reg(pm4
, R_028B80_PA_SU_POLY_OFFSET_FRONT_SCALE
,
518 fui(sctx
->queued
.named
.rasterizer
->offset_scale
));
519 si_pm4_set_reg(pm4
, R_028B84_PA_SU_POLY_OFFSET_FRONT_OFFSET
, fui(offset_units
));
520 si_pm4_set_reg(pm4
, R_028B88_PA_SU_POLY_OFFSET_BACK_SCALE
,
521 fui(sctx
->queued
.named
.rasterizer
->offset_scale
));
522 si_pm4_set_reg(pm4
, R_028B8C_PA_SU_POLY_OFFSET_BACK_OFFSET
, fui(offset_units
));
524 si_pm4_set_state(sctx
, fb_rs
, pm4
);
531 static uint32_t si_translate_fill(uint32_t func
)
534 case PIPE_POLYGON_MODE_FILL
:
535 return V_028814_X_DRAW_TRIANGLES
;
536 case PIPE_POLYGON_MODE_LINE
:
537 return V_028814_X_DRAW_LINES
;
538 case PIPE_POLYGON_MODE_POINT
:
539 return V_028814_X_DRAW_POINTS
;
542 return V_028814_X_DRAW_POINTS
;
546 static void *si_create_rs_state(struct pipe_context
*ctx
,
547 const struct pipe_rasterizer_state
*state
)
549 struct si_state_rasterizer
*rs
= CALLOC_STRUCT(si_state_rasterizer
);
550 struct si_pm4_state
*pm4
= &rs
->pm4
;
552 unsigned prov_vtx
= 1, polygon_dual_mode
;
553 float psize_min
, psize_max
;
559 rs
->two_side
= state
->light_twoside
;
560 rs
->multisample_enable
= state
->multisample
;
561 rs
->clip_plane_enable
= state
->clip_plane_enable
;
562 rs
->line_stipple_enable
= state
->line_stipple_enable
;
564 polygon_dual_mode
= (state
->fill_front
!= PIPE_POLYGON_MODE_FILL
||
565 state
->fill_back
!= PIPE_POLYGON_MODE_FILL
);
567 if (state
->flatshade_first
)
570 rs
->flatshade
= state
->flatshade
;
571 rs
->sprite_coord_enable
= state
->sprite_coord_enable
;
572 rs
->pa_sc_line_stipple
= state
->line_stipple_enable
?
573 S_028A0C_LINE_PATTERN(state
->line_stipple_pattern
) |
574 S_028A0C_REPEAT_COUNT(state
->line_stipple_factor
) : 0;
575 rs
->pa_su_sc_mode_cntl
=
576 S_028814_PROVOKING_VTX_LAST(prov_vtx
) |
577 S_028814_CULL_FRONT(state
->rasterizer_discard
|| (state
->cull_face
& PIPE_FACE_FRONT
) ? 1 : 0) |
578 S_028814_CULL_BACK(state
->rasterizer_discard
|| (state
->cull_face
& PIPE_FACE_BACK
) ? 1 : 0) |
579 S_028814_FACE(!state
->front_ccw
) |
580 S_028814_POLY_OFFSET_FRONT_ENABLE(state
->offset_tri
) |
581 S_028814_POLY_OFFSET_BACK_ENABLE(state
->offset_tri
) |
582 S_028814_POLY_OFFSET_PARA_ENABLE(state
->offset_tri
) |
583 S_028814_POLY_MODE(polygon_dual_mode
) |
584 S_028814_POLYMODE_FRONT_PTYPE(si_translate_fill(state
->fill_front
)) |
585 S_028814_POLYMODE_BACK_PTYPE(si_translate_fill(state
->fill_back
));
586 rs
->pa_cl_clip_cntl
=
587 S_028810_PS_UCP_MODE(3) |
588 S_028810_ZCLIP_NEAR_DISABLE(!state
->depth_clip
) |
589 S_028810_ZCLIP_FAR_DISABLE(!state
->depth_clip
) |
590 S_028810_DX_RASTERIZATION_KILL(state
->rasterizer_discard
) |
591 S_028810_DX_LINEAR_ATTR_CLIP_ENA(1);
594 rs
->offset_units
= state
->offset_units
;
595 rs
->offset_scale
= state
->offset_scale
* 12.0f
;
597 tmp
= S_0286D4_FLAT_SHADE_ENA(1);
598 if (state
->sprite_coord_enable
) {
599 tmp
|= S_0286D4_PNT_SPRITE_ENA(1) |
600 S_0286D4_PNT_SPRITE_OVRD_X(V_0286D4_SPI_PNT_SPRITE_SEL_S
) |
601 S_0286D4_PNT_SPRITE_OVRD_Y(V_0286D4_SPI_PNT_SPRITE_SEL_T
) |
602 S_0286D4_PNT_SPRITE_OVRD_Z(V_0286D4_SPI_PNT_SPRITE_SEL_0
) |
603 S_0286D4_PNT_SPRITE_OVRD_W(V_0286D4_SPI_PNT_SPRITE_SEL_1
);
604 if (state
->sprite_coord_mode
!= PIPE_SPRITE_COORD_UPPER_LEFT
) {
605 tmp
|= S_0286D4_PNT_SPRITE_TOP_1(1);
608 si_pm4_set_reg(pm4
, R_0286D4_SPI_INTERP_CONTROL_0
, tmp
);
610 /* point size 12.4 fixed point */
611 tmp
= (unsigned)(state
->point_size
* 8.0);
612 si_pm4_set_reg(pm4
, R_028A00_PA_SU_POINT_SIZE
, S_028A00_HEIGHT(tmp
) | S_028A00_WIDTH(tmp
));
614 if (state
->point_size_per_vertex
) {
615 psize_min
= util_get_min_point_size(state
);
618 /* Force the point size to be as if the vertex output was disabled. */
619 psize_min
= state
->point_size
;
620 psize_max
= state
->point_size
;
622 /* Divide by two, because 0.5 = 1 pixel. */
623 si_pm4_set_reg(pm4
, R_028A04_PA_SU_POINT_MINMAX
,
624 S_028A04_MIN_SIZE(si_pack_float_12p4(psize_min
/2)) |
625 S_028A04_MAX_SIZE(si_pack_float_12p4(psize_max
/2)));
627 tmp
= (unsigned)state
->line_width
* 8;
628 si_pm4_set_reg(pm4
, R_028A08_PA_SU_LINE_CNTL
, S_028A08_WIDTH(tmp
));
629 si_pm4_set_reg(pm4
, R_028A48_PA_SC_MODE_CNTL_0
,
630 S_028A48_LINE_STIPPLE_ENABLE(state
->line_stipple_enable
) |
631 S_028A48_MSAA_ENABLE(state
->multisample
) |
632 S_028A48_VPORT_SCISSOR_ENABLE(state
->scissor
));
634 si_pm4_set_reg(pm4
, R_028BE4_PA_SU_VTX_CNTL
,
635 S_028BE4_PIX_CENTER(state
->half_pixel_center
) |
636 S_028BE4_QUANT_MODE(V_028BE4_X_16_8_FIXED_POINT_1_256TH
));
638 si_pm4_set_reg(pm4
, R_028B7C_PA_SU_POLY_OFFSET_CLAMP
, fui(state
->offset_clamp
));
643 static void si_bind_rs_state(struct pipe_context
*ctx
, void *state
)
645 struct si_context
*sctx
= (struct si_context
*)ctx
;
646 struct si_state_rasterizer
*rs
= (struct si_state_rasterizer
*)state
;
652 sctx
->sprite_coord_enable
= rs
->sprite_coord_enable
;
653 sctx
->pa_sc_line_stipple
= rs
->pa_sc_line_stipple
;
654 sctx
->pa_su_sc_mode_cntl
= rs
->pa_su_sc_mode_cntl
;
656 si_pm4_bind_state(sctx
, rasterizer
, rs
);
657 si_update_fb_rs_state(sctx
);
660 static void si_delete_rs_state(struct pipe_context
*ctx
, void *state
)
662 struct si_context
*sctx
= (struct si_context
*)ctx
;
663 si_pm4_delete_state(sctx
, rasterizer
, (struct si_state_rasterizer
*)state
);
667 * infeered state between dsa and stencil ref
669 static void si_update_dsa_stencil_ref(struct si_context
*sctx
)
671 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
672 struct pipe_stencil_ref
*ref
= &sctx
->stencil_ref
;
673 struct si_state_dsa
*dsa
= sctx
->queued
.named
.dsa
;
678 si_pm4_set_reg(pm4
, R_028430_DB_STENCILREFMASK
,
679 S_028430_STENCILTESTVAL(ref
->ref_value
[0]) |
680 S_028430_STENCILMASK(dsa
->valuemask
[0]) |
681 S_028430_STENCILWRITEMASK(dsa
->writemask
[0]) |
682 S_028430_STENCILOPVAL(1));
683 si_pm4_set_reg(pm4
, R_028434_DB_STENCILREFMASK_BF
,
684 S_028434_STENCILTESTVAL_BF(ref
->ref_value
[1]) |
685 S_028434_STENCILMASK_BF(dsa
->valuemask
[1]) |
686 S_028434_STENCILWRITEMASK_BF(dsa
->writemask
[1]) |
687 S_028434_STENCILOPVAL_BF(1));
689 si_pm4_set_state(sctx
, dsa_stencil_ref
, pm4
);
692 static void si_set_pipe_stencil_ref(struct pipe_context
*ctx
,
693 const struct pipe_stencil_ref
*state
)
695 struct si_context
*sctx
= (struct si_context
*)ctx
;
696 sctx
->stencil_ref
= *state
;
697 si_update_dsa_stencil_ref(sctx
);
705 static uint32_t si_translate_stencil_op(int s_op
)
708 case PIPE_STENCIL_OP_KEEP
:
709 return V_02842C_STENCIL_KEEP
;
710 case PIPE_STENCIL_OP_ZERO
:
711 return V_02842C_STENCIL_ZERO
;
712 case PIPE_STENCIL_OP_REPLACE
:
713 return V_02842C_STENCIL_REPLACE_TEST
;
714 case PIPE_STENCIL_OP_INCR
:
715 return V_02842C_STENCIL_ADD_CLAMP
;
716 case PIPE_STENCIL_OP_DECR
:
717 return V_02842C_STENCIL_SUB_CLAMP
;
718 case PIPE_STENCIL_OP_INCR_WRAP
:
719 return V_02842C_STENCIL_ADD_WRAP
;
720 case PIPE_STENCIL_OP_DECR_WRAP
:
721 return V_02842C_STENCIL_SUB_WRAP
;
722 case PIPE_STENCIL_OP_INVERT
:
723 return V_02842C_STENCIL_INVERT
;
725 R600_ERR("Unknown stencil op %d", s_op
);
732 static void *si_create_dsa_state(struct pipe_context
*ctx
,
733 const struct pipe_depth_stencil_alpha_state
*state
)
735 struct si_state_dsa
*dsa
= CALLOC_STRUCT(si_state_dsa
);
736 struct si_pm4_state
*pm4
= &dsa
->pm4
;
737 unsigned db_depth_control
;
738 unsigned db_render_control
;
739 uint32_t db_stencil_control
= 0;
745 dsa
->valuemask
[0] = state
->stencil
[0].valuemask
;
746 dsa
->valuemask
[1] = state
->stencil
[1].valuemask
;
747 dsa
->writemask
[0] = state
->stencil
[0].writemask
;
748 dsa
->writemask
[1] = state
->stencil
[1].writemask
;
750 db_depth_control
= S_028800_Z_ENABLE(state
->depth
.enabled
) |
751 S_028800_Z_WRITE_ENABLE(state
->depth
.writemask
) |
752 S_028800_ZFUNC(state
->depth
.func
);
755 if (state
->stencil
[0].enabled
) {
756 db_depth_control
|= S_028800_STENCIL_ENABLE(1);
757 db_depth_control
|= S_028800_STENCILFUNC(state
->stencil
[0].func
);
758 db_stencil_control
|= S_02842C_STENCILFAIL(si_translate_stencil_op(state
->stencil
[0].fail_op
));
759 db_stencil_control
|= S_02842C_STENCILZPASS(si_translate_stencil_op(state
->stencil
[0].zpass_op
));
760 db_stencil_control
|= S_02842C_STENCILZFAIL(si_translate_stencil_op(state
->stencil
[0].zfail_op
));
762 if (state
->stencil
[1].enabled
) {
763 db_depth_control
|= S_028800_BACKFACE_ENABLE(1);
764 db_depth_control
|= S_028800_STENCILFUNC_BF(state
->stencil
[1].func
);
765 db_stencil_control
|= S_02842C_STENCILFAIL_BF(si_translate_stencil_op(state
->stencil
[1].fail_op
));
766 db_stencil_control
|= S_02842C_STENCILZPASS_BF(si_translate_stencil_op(state
->stencil
[1].zpass_op
));
767 db_stencil_control
|= S_02842C_STENCILZFAIL_BF(si_translate_stencil_op(state
->stencil
[1].zfail_op
));
772 if (state
->alpha
.enabled
) {
773 dsa
->alpha_func
= state
->alpha
.func
;
774 dsa
->alpha_ref
= state
->alpha
.ref_value
;
776 si_pm4_set_reg(pm4
, R_00B030_SPI_SHADER_USER_DATA_PS_0
+
777 SI_SGPR_ALPHA_REF
* 4, fui(dsa
->alpha_ref
));
779 dsa
->alpha_func
= PIPE_FUNC_ALWAYS
;
783 db_render_control
= 0;
784 si_pm4_set_reg(pm4
, R_028800_DB_DEPTH_CONTROL
, db_depth_control
);
785 si_pm4_set_reg(pm4
, R_028000_DB_RENDER_CONTROL
, db_render_control
);
786 si_pm4_set_reg(pm4
, R_02842C_DB_STENCIL_CONTROL
, db_stencil_control
);
791 static void si_bind_dsa_state(struct pipe_context
*ctx
, void *state
)
793 struct si_context
*sctx
= (struct si_context
*)ctx
;
794 struct si_state_dsa
*dsa
= state
;
799 si_pm4_bind_state(sctx
, dsa
, dsa
);
800 si_update_dsa_stencil_ref(sctx
);
803 static void si_delete_dsa_state(struct pipe_context
*ctx
, void *state
)
805 struct si_context
*sctx
= (struct si_context
*)ctx
;
806 si_pm4_delete_state(sctx
, dsa
, (struct si_state_dsa
*)state
);
809 static void *si_create_db_flush_dsa(struct si_context
*sctx
, bool copy_depth
,
810 bool copy_stencil
, int sample
)
812 struct pipe_depth_stencil_alpha_state dsa
;
813 struct si_state_dsa
*state
;
815 memset(&dsa
, 0, sizeof(dsa
));
817 state
= sctx
->b
.b
.create_depth_stencil_alpha_state(&sctx
->b
.b
, &dsa
);
818 if (copy_depth
|| copy_stencil
) {
819 si_pm4_set_reg(&state
->pm4
, R_028000_DB_RENDER_CONTROL
,
820 S_028000_DEPTH_COPY(copy_depth
) |
821 S_028000_STENCIL_COPY(copy_stencil
) |
822 S_028000_COPY_CENTROID(1) |
823 S_028000_COPY_SAMPLE(sample
));
825 si_pm4_set_reg(&state
->pm4
, R_028000_DB_RENDER_CONTROL
,
826 S_028000_DEPTH_COMPRESS_DISABLE(1) |
827 S_028000_STENCIL_COMPRESS_DISABLE(1));
836 static uint32_t si_translate_colorformat(enum pipe_format format
)
838 const struct util_format_description
*desc
= util_format_description(format
);
840 #define HAS_SIZE(x,y,z,w) \
841 (desc->channel[0].size == (x) && desc->channel[1].size == (y) && \
842 desc->channel[2].size == (z) && desc->channel[3].size == (w))
844 if (format
== PIPE_FORMAT_R11G11B10_FLOAT
) /* isn't plain */
845 return V_028C70_COLOR_10_11_11
;
847 if (desc
->layout
!= UTIL_FORMAT_LAYOUT_PLAIN
)
848 return V_028C70_COLOR_INVALID
;
850 switch (desc
->nr_channels
) {
852 switch (desc
->channel
[0].size
) {
854 return V_028C70_COLOR_8
;
856 return V_028C70_COLOR_16
;
858 return V_028C70_COLOR_32
;
862 if (desc
->channel
[0].size
== desc
->channel
[1].size
) {
863 switch (desc
->channel
[0].size
) {
865 return V_028C70_COLOR_8_8
;
867 return V_028C70_COLOR_16_16
;
869 return V_028C70_COLOR_32_32
;
871 } else if (HAS_SIZE(8,24,0,0)) {
872 return V_028C70_COLOR_24_8
;
873 } else if (HAS_SIZE(24,8,0,0)) {
874 return V_028C70_COLOR_8_24
;
878 if (HAS_SIZE(5,6,5,0)) {
879 return V_028C70_COLOR_5_6_5
;
880 } else if (HAS_SIZE(32,8,24,0)) {
881 return V_028C70_COLOR_X24_8_32_FLOAT
;
885 if (desc
->channel
[0].size
== desc
->channel
[1].size
&&
886 desc
->channel
[0].size
== desc
->channel
[2].size
&&
887 desc
->channel
[0].size
== desc
->channel
[3].size
) {
888 switch (desc
->channel
[0].size
) {
890 return V_028C70_COLOR_4_4_4_4
;
892 return V_028C70_COLOR_8_8_8_8
;
894 return V_028C70_COLOR_16_16_16_16
;
896 return V_028C70_COLOR_32_32_32_32
;
898 } else if (HAS_SIZE(5,5,5,1)) {
899 return V_028C70_COLOR_1_5_5_5
;
900 } else if (HAS_SIZE(10,10,10,2)) {
901 return V_028C70_COLOR_2_10_10_10
;
905 return V_028C70_COLOR_INVALID
;
908 static uint32_t si_translate_colorswap(enum pipe_format format
)
910 const struct util_format_description
*desc
= util_format_description(format
);
912 #define HAS_SWIZZLE(chan,swz) (desc->swizzle[chan] == UTIL_FORMAT_SWIZZLE_##swz)
914 if (format
== PIPE_FORMAT_R11G11B10_FLOAT
) /* isn't plain */
915 return V_028C70_SWAP_STD
;
917 if (desc
->layout
!= UTIL_FORMAT_LAYOUT_PLAIN
)
920 switch (desc
->nr_channels
) {
922 if (HAS_SWIZZLE(0,X
))
923 return V_028C70_SWAP_STD
; /* X___ */
924 else if (HAS_SWIZZLE(3,X
))
925 return V_028C70_SWAP_ALT_REV
; /* ___X */
928 if ((HAS_SWIZZLE(0,X
) && HAS_SWIZZLE(1,Y
)) ||
929 (HAS_SWIZZLE(0,X
) && HAS_SWIZZLE(1,NONE
)) ||
930 (HAS_SWIZZLE(0,NONE
) && HAS_SWIZZLE(1,Y
)))
931 return V_028C70_SWAP_STD
; /* XY__ */
932 else if ((HAS_SWIZZLE(0,Y
) && HAS_SWIZZLE(1,X
)) ||
933 (HAS_SWIZZLE(0,Y
) && HAS_SWIZZLE(1,NONE
)) ||
934 (HAS_SWIZZLE(0,NONE
) && HAS_SWIZZLE(1,X
)))
935 return V_028C70_SWAP_STD_REV
; /* YX__ */
936 else if (HAS_SWIZZLE(0,X
) && HAS_SWIZZLE(3,Y
))
937 return V_028C70_SWAP_ALT
; /* X__Y */
940 if (HAS_SWIZZLE(0,X
))
941 return V_028C70_SWAP_STD
; /* XYZ */
942 else if (HAS_SWIZZLE(0,Z
))
943 return V_028C70_SWAP_STD_REV
; /* ZYX */
946 /* check the middle channels, the 1st and 4th channel can be NONE */
947 if (HAS_SWIZZLE(1,Y
) && HAS_SWIZZLE(2,Z
))
948 return V_028C70_SWAP_STD
; /* XYZW */
949 else if (HAS_SWIZZLE(1,Z
) && HAS_SWIZZLE(2,Y
))
950 return V_028C70_SWAP_STD_REV
; /* WZYX */
951 else if (HAS_SWIZZLE(1,Y
) && HAS_SWIZZLE(2,X
))
952 return V_028C70_SWAP_ALT
; /* ZYXW */
953 else if (HAS_SWIZZLE(1,X
) && HAS_SWIZZLE(2,Y
))
954 return V_028C70_SWAP_ALT_REV
; /* WXYZ */
960 static uint32_t si_colorformat_endian_swap(uint32_t colorformat
)
963 switch(colorformat
) {
965 case V_028C70_COLOR_8
:
966 return V_028C70_ENDIAN_NONE
;
968 /* 16-bit buffers. */
969 case V_028C70_COLOR_5_6_5
:
970 case V_028C70_COLOR_1_5_5_5
:
971 case V_028C70_COLOR_4_4_4_4
:
972 case V_028C70_COLOR_16
:
973 case V_028C70_COLOR_8_8
:
974 return V_028C70_ENDIAN_8IN16
;
976 /* 32-bit buffers. */
977 case V_028C70_COLOR_8_8_8_8
:
978 case V_028C70_COLOR_2_10_10_10
:
979 case V_028C70_COLOR_8_24
:
980 case V_028C70_COLOR_24_8
:
981 case V_028C70_COLOR_16_16
:
982 return V_028C70_ENDIAN_8IN32
;
984 /* 64-bit buffers. */
985 case V_028C70_COLOR_16_16_16_16
:
986 return V_028C70_ENDIAN_8IN16
;
988 case V_028C70_COLOR_32_32
:
989 return V_028C70_ENDIAN_8IN32
;
991 /* 128-bit buffers. */
992 case V_028C70_COLOR_32_32_32_32
:
993 return V_028C70_ENDIAN_8IN32
;
995 return V_028C70_ENDIAN_NONE
; /* Unsupported. */
998 return V_028C70_ENDIAN_NONE
;
1002 /* Returns the size in bits of the widest component of a CB format */
1003 static unsigned si_colorformat_max_comp_size(uint32_t colorformat
)
1005 switch(colorformat
) {
1006 case V_028C70_COLOR_4_4_4_4
:
1009 case V_028C70_COLOR_1_5_5_5
:
1010 case V_028C70_COLOR_5_5_5_1
:
1013 case V_028C70_COLOR_5_6_5
:
1016 case V_028C70_COLOR_8
:
1017 case V_028C70_COLOR_8_8
:
1018 case V_028C70_COLOR_8_8_8_8
:
1021 case V_028C70_COLOR_10_10_10_2
:
1022 case V_028C70_COLOR_2_10_10_10
:
1025 case V_028C70_COLOR_10_11_11
:
1026 case V_028C70_COLOR_11_11_10
:
1029 case V_028C70_COLOR_16
:
1030 case V_028C70_COLOR_16_16
:
1031 case V_028C70_COLOR_16_16_16_16
:
1034 case V_028C70_COLOR_8_24
:
1035 case V_028C70_COLOR_24_8
:
1038 case V_028C70_COLOR_32
:
1039 case V_028C70_COLOR_32_32
:
1040 case V_028C70_COLOR_32_32_32_32
:
1041 case V_028C70_COLOR_X24_8_32_FLOAT
:
1045 assert(!"Unknown maximum component size");
1049 static uint32_t si_translate_dbformat(enum pipe_format format
)
1052 case PIPE_FORMAT_Z16_UNORM
:
1053 return V_028040_Z_16
;
1054 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
1055 case PIPE_FORMAT_X8Z24_UNORM
:
1056 case PIPE_FORMAT_Z24X8_UNORM
:
1057 case PIPE_FORMAT_Z24_UNORM_S8_UINT
:
1058 return V_028040_Z_24
; /* deprecated on SI */
1059 case PIPE_FORMAT_Z32_FLOAT
:
1060 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
1061 return V_028040_Z_32_FLOAT
;
1063 return V_028040_Z_INVALID
;
1068 * Texture translation
1071 static uint32_t si_translate_texformat(struct pipe_screen
*screen
,
1072 enum pipe_format format
,
1073 const struct util_format_description
*desc
,
1076 struct si_screen
*sscreen
= (struct si_screen
*)screen
;
1077 bool enable_s3tc
= sscreen
->b
.info
.drm_minor
>= 31;
1078 boolean uniform
= TRUE
;
1081 /* Colorspace (return non-RGB formats directly). */
1082 switch (desc
->colorspace
) {
1083 /* Depth stencil formats */
1084 case UTIL_FORMAT_COLORSPACE_ZS
:
1086 case PIPE_FORMAT_Z16_UNORM
:
1087 return V_008F14_IMG_DATA_FORMAT_16
;
1088 case PIPE_FORMAT_X24S8_UINT
:
1089 case PIPE_FORMAT_Z24X8_UNORM
:
1090 case PIPE_FORMAT_Z24_UNORM_S8_UINT
:
1091 return V_008F14_IMG_DATA_FORMAT_8_24
;
1092 case PIPE_FORMAT_X8Z24_UNORM
:
1093 case PIPE_FORMAT_S8X24_UINT
:
1094 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
1095 return V_008F14_IMG_DATA_FORMAT_24_8
;
1096 case PIPE_FORMAT_S8_UINT
:
1097 return V_008F14_IMG_DATA_FORMAT_8
;
1098 case PIPE_FORMAT_Z32_FLOAT
:
1099 return V_008F14_IMG_DATA_FORMAT_32
;
1100 case PIPE_FORMAT_X32_S8X24_UINT
:
1101 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
1102 return V_008F14_IMG_DATA_FORMAT_X24_8_32
;
1107 case UTIL_FORMAT_COLORSPACE_YUV
:
1108 goto out_unknown
; /* TODO */
1110 case UTIL_FORMAT_COLORSPACE_SRGB
:
1111 if (desc
->nr_channels
!= 4 && desc
->nr_channels
!= 1)
1119 if (desc
->layout
== UTIL_FORMAT_LAYOUT_RGTC
) {
1124 case PIPE_FORMAT_RGTC1_SNORM
:
1125 case PIPE_FORMAT_LATC1_SNORM
:
1126 case PIPE_FORMAT_RGTC1_UNORM
:
1127 case PIPE_FORMAT_LATC1_UNORM
:
1128 return V_008F14_IMG_DATA_FORMAT_BC4
;
1129 case PIPE_FORMAT_RGTC2_SNORM
:
1130 case PIPE_FORMAT_LATC2_SNORM
:
1131 case PIPE_FORMAT_RGTC2_UNORM
:
1132 case PIPE_FORMAT_LATC2_UNORM
:
1133 return V_008F14_IMG_DATA_FORMAT_BC5
;
1139 if (desc
->layout
== UTIL_FORMAT_LAYOUT_S3TC
) {
1144 if (!util_format_s3tc_enabled
) {
1149 case PIPE_FORMAT_DXT1_RGB
:
1150 case PIPE_FORMAT_DXT1_RGBA
:
1151 case PIPE_FORMAT_DXT1_SRGB
:
1152 case PIPE_FORMAT_DXT1_SRGBA
:
1153 return V_008F14_IMG_DATA_FORMAT_BC1
;
1154 case PIPE_FORMAT_DXT3_RGBA
:
1155 case PIPE_FORMAT_DXT3_SRGBA
:
1156 return V_008F14_IMG_DATA_FORMAT_BC2
;
1157 case PIPE_FORMAT_DXT5_RGBA
:
1158 case PIPE_FORMAT_DXT5_SRGBA
:
1159 return V_008F14_IMG_DATA_FORMAT_BC3
;
1165 if (format
== PIPE_FORMAT_R9G9B9E5_FLOAT
) {
1166 return V_008F14_IMG_DATA_FORMAT_5_9_9_9
;
1167 } else if (format
== PIPE_FORMAT_R11G11B10_FLOAT
) {
1168 return V_008F14_IMG_DATA_FORMAT_10_11_11
;
1171 /* R8G8Bx_SNORM - TODO CxV8U8 */
1173 /* See whether the components are of the same size. */
1174 for (i
= 1; i
< desc
->nr_channels
; i
++) {
1175 uniform
= uniform
&& desc
->channel
[0].size
== desc
->channel
[i
].size
;
1178 /* Non-uniform formats. */
1180 switch(desc
->nr_channels
) {
1182 if (desc
->channel
[0].size
== 5 &&
1183 desc
->channel
[1].size
== 6 &&
1184 desc
->channel
[2].size
== 5) {
1185 return V_008F14_IMG_DATA_FORMAT_5_6_5
;
1189 if (desc
->channel
[0].size
== 5 &&
1190 desc
->channel
[1].size
== 5 &&
1191 desc
->channel
[2].size
== 5 &&
1192 desc
->channel
[3].size
== 1) {
1193 return V_008F14_IMG_DATA_FORMAT_1_5_5_5
;
1195 if (desc
->channel
[0].size
== 10 &&
1196 desc
->channel
[1].size
== 10 &&
1197 desc
->channel
[2].size
== 10 &&
1198 desc
->channel
[3].size
== 2) {
1199 return V_008F14_IMG_DATA_FORMAT_2_10_10_10
;
1206 if (first_non_void
< 0 || first_non_void
> 3)
1209 /* uniform formats */
1210 switch (desc
->channel
[first_non_void
].size
) {
1212 switch (desc
->nr_channels
) {
1213 #if 0 /* Not supported for render targets */
1215 return V_008F14_IMG_DATA_FORMAT_4_4
;
1218 return V_008F14_IMG_DATA_FORMAT_4_4_4_4
;
1222 switch (desc
->nr_channels
) {
1224 return V_008F14_IMG_DATA_FORMAT_8
;
1226 return V_008F14_IMG_DATA_FORMAT_8_8
;
1228 return V_008F14_IMG_DATA_FORMAT_8_8_8_8
;
1232 switch (desc
->nr_channels
) {
1234 return V_008F14_IMG_DATA_FORMAT_16
;
1236 return V_008F14_IMG_DATA_FORMAT_16_16
;
1238 return V_008F14_IMG_DATA_FORMAT_16_16_16_16
;
1242 switch (desc
->nr_channels
) {
1244 return V_008F14_IMG_DATA_FORMAT_32
;
1246 return V_008F14_IMG_DATA_FORMAT_32_32
;
1247 #if 0 /* Not supported for render targets */
1249 return V_008F14_IMG_DATA_FORMAT_32_32_32
;
1252 return V_008F14_IMG_DATA_FORMAT_32_32_32_32
;
1257 /* R600_ERR("Unable to handle texformat %d %s\n", format, util_format_name(format)); */
1261 static unsigned si_tex_wrap(unsigned wrap
)
1265 case PIPE_TEX_WRAP_REPEAT
:
1266 return V_008F30_SQ_TEX_WRAP
;
1267 case PIPE_TEX_WRAP_CLAMP
:
1268 return V_008F30_SQ_TEX_CLAMP_HALF_BORDER
;
1269 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
1270 return V_008F30_SQ_TEX_CLAMP_LAST_TEXEL
;
1271 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
1272 return V_008F30_SQ_TEX_CLAMP_BORDER
;
1273 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
1274 return V_008F30_SQ_TEX_MIRROR
;
1275 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
1276 return V_008F30_SQ_TEX_MIRROR_ONCE_HALF_BORDER
;
1277 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
1278 return V_008F30_SQ_TEX_MIRROR_ONCE_LAST_TEXEL
;
1279 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
1280 return V_008F30_SQ_TEX_MIRROR_ONCE_BORDER
;
1284 static unsigned si_tex_filter(unsigned filter
)
1288 case PIPE_TEX_FILTER_NEAREST
:
1289 return V_008F38_SQ_TEX_XY_FILTER_POINT
;
1290 case PIPE_TEX_FILTER_LINEAR
:
1291 return V_008F38_SQ_TEX_XY_FILTER_BILINEAR
;
1295 static unsigned si_tex_mipfilter(unsigned filter
)
1298 case PIPE_TEX_MIPFILTER_NEAREST
:
1299 return V_008F38_SQ_TEX_Z_FILTER_POINT
;
1300 case PIPE_TEX_MIPFILTER_LINEAR
:
1301 return V_008F38_SQ_TEX_Z_FILTER_LINEAR
;
1303 case PIPE_TEX_MIPFILTER_NONE
:
1304 return V_008F38_SQ_TEX_Z_FILTER_NONE
;
1308 static unsigned si_tex_compare(unsigned compare
)
1312 case PIPE_FUNC_NEVER
:
1313 return V_008F30_SQ_TEX_DEPTH_COMPARE_NEVER
;
1314 case PIPE_FUNC_LESS
:
1315 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESS
;
1316 case PIPE_FUNC_EQUAL
:
1317 return V_008F30_SQ_TEX_DEPTH_COMPARE_EQUAL
;
1318 case PIPE_FUNC_LEQUAL
:
1319 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESSEQUAL
;
1320 case PIPE_FUNC_GREATER
:
1321 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATER
;
1322 case PIPE_FUNC_NOTEQUAL
:
1323 return V_008F30_SQ_TEX_DEPTH_COMPARE_NOTEQUAL
;
1324 case PIPE_FUNC_GEQUAL
:
1325 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATEREQUAL
;
1326 case PIPE_FUNC_ALWAYS
:
1327 return V_008F30_SQ_TEX_DEPTH_COMPARE_ALWAYS
;
1331 static unsigned si_tex_dim(unsigned dim
, unsigned nr_samples
)
1335 case PIPE_TEXTURE_1D
:
1336 return V_008F1C_SQ_RSRC_IMG_1D
;
1337 case PIPE_TEXTURE_1D_ARRAY
:
1338 return V_008F1C_SQ_RSRC_IMG_1D_ARRAY
;
1339 case PIPE_TEXTURE_2D
:
1340 case PIPE_TEXTURE_RECT
:
1341 return nr_samples
> 1 ? V_008F1C_SQ_RSRC_IMG_2D_MSAA
:
1342 V_008F1C_SQ_RSRC_IMG_2D
;
1343 case PIPE_TEXTURE_2D_ARRAY
:
1344 return nr_samples
> 1 ? V_008F1C_SQ_RSRC_IMG_2D_MSAA_ARRAY
:
1345 V_008F1C_SQ_RSRC_IMG_2D_ARRAY
;
1346 case PIPE_TEXTURE_3D
:
1347 return V_008F1C_SQ_RSRC_IMG_3D
;
1348 case PIPE_TEXTURE_CUBE
:
1349 return V_008F1C_SQ_RSRC_IMG_CUBE
;
1354 * Format support testing
1357 static bool si_is_sampler_format_supported(struct pipe_screen
*screen
, enum pipe_format format
)
1359 return si_translate_texformat(screen
, format
, util_format_description(format
),
1360 util_format_get_first_non_void_channel(format
)) != ~0U;
1363 static uint32_t si_translate_buffer_dataformat(struct pipe_screen
*screen
,
1364 const struct util_format_description
*desc
,
1367 unsigned type
= desc
->channel
[first_non_void
].type
;
1370 if (type
== UTIL_FORMAT_TYPE_FIXED
)
1371 return V_008F0C_BUF_DATA_FORMAT_INVALID
;
1373 if (desc
->nr_channels
== 4 &&
1374 desc
->channel
[0].size
== 10 &&
1375 desc
->channel
[1].size
== 10 &&
1376 desc
->channel
[2].size
== 10 &&
1377 desc
->channel
[3].size
== 2)
1378 return V_008F0C_BUF_DATA_FORMAT_2_10_10_10
;
1380 /* See whether the components are of the same size. */
1381 for (i
= 0; i
< desc
->nr_channels
; i
++) {
1382 if (desc
->channel
[first_non_void
].size
!= desc
->channel
[i
].size
)
1383 return V_008F0C_BUF_DATA_FORMAT_INVALID
;
1386 switch (desc
->channel
[first_non_void
].size
) {
1388 switch (desc
->nr_channels
) {
1390 return V_008F0C_BUF_DATA_FORMAT_8
;
1392 return V_008F0C_BUF_DATA_FORMAT_8_8
;
1395 return V_008F0C_BUF_DATA_FORMAT_8_8_8_8
;
1399 switch (desc
->nr_channels
) {
1401 return V_008F0C_BUF_DATA_FORMAT_16
;
1403 return V_008F0C_BUF_DATA_FORMAT_16_16
;
1406 return V_008F0C_BUF_DATA_FORMAT_16_16_16_16
;
1410 /* From the Southern Islands ISA documentation about MTBUF:
1411 * 'Memory reads of data in memory that is 32 or 64 bits do not
1412 * undergo any format conversion.'
1414 if (type
!= UTIL_FORMAT_TYPE_FLOAT
&&
1415 !desc
->channel
[first_non_void
].pure_integer
)
1416 return V_008F0C_BUF_DATA_FORMAT_INVALID
;
1418 switch (desc
->nr_channels
) {
1420 return V_008F0C_BUF_DATA_FORMAT_32
;
1422 return V_008F0C_BUF_DATA_FORMAT_32_32
;
1424 return V_008F0C_BUF_DATA_FORMAT_32_32_32
;
1426 return V_008F0C_BUF_DATA_FORMAT_32_32_32_32
;
1431 return V_008F0C_BUF_DATA_FORMAT_INVALID
;
1434 static uint32_t si_translate_buffer_numformat(struct pipe_screen
*screen
,
1435 const struct util_format_description
*desc
,
1438 switch (desc
->channel
[first_non_void
].type
) {
1439 case UTIL_FORMAT_TYPE_SIGNED
:
1440 if (desc
->channel
[first_non_void
].normalized
)
1441 return V_008F0C_BUF_NUM_FORMAT_SNORM
;
1442 else if (desc
->channel
[first_non_void
].pure_integer
)
1443 return V_008F0C_BUF_NUM_FORMAT_SINT
;
1445 return V_008F0C_BUF_NUM_FORMAT_SSCALED
;
1447 case UTIL_FORMAT_TYPE_UNSIGNED
:
1448 if (desc
->channel
[first_non_void
].normalized
)
1449 return V_008F0C_BUF_NUM_FORMAT_UNORM
;
1450 else if (desc
->channel
[first_non_void
].pure_integer
)
1451 return V_008F0C_BUF_NUM_FORMAT_UINT
;
1453 return V_008F0C_BUF_NUM_FORMAT_USCALED
;
1455 case UTIL_FORMAT_TYPE_FLOAT
:
1457 return V_008F0C_BUF_NUM_FORMAT_FLOAT
;
1461 static bool si_is_vertex_format_supported(struct pipe_screen
*screen
, enum pipe_format format
)
1463 const struct util_format_description
*desc
;
1465 unsigned data_format
;
1467 desc
= util_format_description(format
);
1468 first_non_void
= util_format_get_first_non_void_channel(format
);
1469 data_format
= si_translate_buffer_dataformat(screen
, desc
, first_non_void
);
1470 return data_format
!= V_008F0C_BUF_DATA_FORMAT_INVALID
;
1473 static bool si_is_colorbuffer_format_supported(enum pipe_format format
)
1475 return si_translate_colorformat(format
) != V_028C70_COLOR_INVALID
&&
1476 si_translate_colorswap(format
) != ~0U;
1479 static bool si_is_zs_format_supported(enum pipe_format format
)
1481 return si_translate_dbformat(format
) != V_028040_Z_INVALID
;
1484 boolean
si_is_format_supported(struct pipe_screen
*screen
,
1485 enum pipe_format format
,
1486 enum pipe_texture_target target
,
1487 unsigned sample_count
,
1490 struct si_screen
*sscreen
= (struct si_screen
*)screen
;
1491 unsigned retval
= 0;
1493 if (target
>= PIPE_MAX_TEXTURE_TYPES
) {
1494 R600_ERR("r600: unsupported texture type %d\n", target
);
1498 if (!util_format_is_supported(format
, usage
))
1501 if (sample_count
> 1) {
1502 if (HAVE_LLVM
< 0x0304)
1505 /* 2D tiling on CIK is supported since DRM 2.35.0 */
1506 if (sscreen
->b
.chip_class
>= CIK
&& sscreen
->b
.info
.drm_minor
< 35)
1509 switch (sample_count
) {
1519 if (usage
& PIPE_BIND_SAMPLER_VIEW
) {
1520 if (target
== PIPE_BUFFER
) {
1521 if (si_is_vertex_format_supported(screen
, format
))
1522 retval
|= PIPE_BIND_SAMPLER_VIEW
;
1524 if (si_is_sampler_format_supported(screen
, format
))
1525 retval
|= PIPE_BIND_SAMPLER_VIEW
;
1529 if ((usage
& (PIPE_BIND_RENDER_TARGET
|
1530 PIPE_BIND_DISPLAY_TARGET
|
1532 PIPE_BIND_SHARED
)) &&
1533 si_is_colorbuffer_format_supported(format
)) {
1535 (PIPE_BIND_RENDER_TARGET
|
1536 PIPE_BIND_DISPLAY_TARGET
|
1541 if ((usage
& PIPE_BIND_DEPTH_STENCIL
) &&
1542 si_is_zs_format_supported(format
)) {
1543 retval
|= PIPE_BIND_DEPTH_STENCIL
;
1546 if ((usage
& PIPE_BIND_VERTEX_BUFFER
) &&
1547 si_is_vertex_format_supported(screen
, format
)) {
1548 retval
|= PIPE_BIND_VERTEX_BUFFER
;
1551 if (usage
& PIPE_BIND_TRANSFER_READ
)
1552 retval
|= PIPE_BIND_TRANSFER_READ
;
1553 if (usage
& PIPE_BIND_TRANSFER_WRITE
)
1554 retval
|= PIPE_BIND_TRANSFER_WRITE
;
1556 return retval
== usage
;
1559 static unsigned si_tile_mode_index(struct r600_texture
*rtex
, unsigned level
, bool stencil
)
1561 unsigned tile_mode_index
= 0;
1564 tile_mode_index
= rtex
->surface
.stencil_tiling_index
[level
];
1566 tile_mode_index
= rtex
->surface
.tiling_index
[level
];
1568 return tile_mode_index
;
1572 * framebuffer handling
1575 static void si_initialize_color_surface(struct si_context
*sctx
,
1576 struct r600_surface
*surf
)
1578 struct r600_texture
*rtex
= (struct r600_texture
*)surf
->base
.texture
;
1579 unsigned level
= surf
->base
.u
.tex
.level
;
1580 uint64_t offset
= rtex
->surface
.level
[level
].offset
;
1581 unsigned pitch
, slice
;
1582 unsigned color_info
, color_attrib
, color_pitch
, color_view
;
1583 unsigned tile_mode_index
;
1584 unsigned format
, swap
, ntype
, endian
;
1585 const struct util_format_description
*desc
;
1587 unsigned blend_clamp
= 0, blend_bypass
= 0;
1588 unsigned max_comp_size
;
1590 /* Layered rendering doesn't work with LINEAR_GENERAL.
1591 * (LINEAR_ALIGNED and others work) */
1592 if (rtex
->surface
.level
[level
].mode
== RADEON_SURF_MODE_LINEAR
) {
1593 assert(surf
->base
.u
.tex
.first_layer
== surf
->base
.u
.tex
.last_layer
);
1594 offset
+= rtex
->surface
.level
[level
].slice_size
*
1595 surf
->base
.u
.tex
.first_layer
;
1598 color_view
= S_028C6C_SLICE_START(surf
->base
.u
.tex
.first_layer
) |
1599 S_028C6C_SLICE_MAX(surf
->base
.u
.tex
.last_layer
);
1602 pitch
= (rtex
->surface
.level
[level
].nblk_x
) / 8 - 1;
1603 slice
= (rtex
->surface
.level
[level
].nblk_x
* rtex
->surface
.level
[level
].nblk_y
) / 64;
1608 tile_mode_index
= si_tile_mode_index(rtex
, level
, false);
1610 desc
= util_format_description(surf
->base
.format
);
1611 for (i
= 0; i
< 4; i
++) {
1612 if (desc
->channel
[i
].type
!= UTIL_FORMAT_TYPE_VOID
) {
1616 if (i
== 4 || desc
->channel
[i
].type
== UTIL_FORMAT_TYPE_FLOAT
) {
1617 ntype
= V_028C70_NUMBER_FLOAT
;
1619 ntype
= V_028C70_NUMBER_UNORM
;
1620 if (desc
->colorspace
== UTIL_FORMAT_COLORSPACE_SRGB
)
1621 ntype
= V_028C70_NUMBER_SRGB
;
1622 else if (desc
->channel
[i
].type
== UTIL_FORMAT_TYPE_SIGNED
) {
1623 if (desc
->channel
[i
].pure_integer
) {
1624 ntype
= V_028C70_NUMBER_SINT
;
1626 assert(desc
->channel
[i
].normalized
);
1627 ntype
= V_028C70_NUMBER_SNORM
;
1629 } else if (desc
->channel
[i
].type
== UTIL_FORMAT_TYPE_UNSIGNED
) {
1630 if (desc
->channel
[i
].pure_integer
) {
1631 ntype
= V_028C70_NUMBER_UINT
;
1633 assert(desc
->channel
[i
].normalized
);
1634 ntype
= V_028C70_NUMBER_UNORM
;
1639 format
= si_translate_colorformat(surf
->base
.format
);
1640 if (format
== V_028C70_COLOR_INVALID
) {
1641 R600_ERR("Invalid CB format: %d, disabling CB.\n", surf
->base
.format
);
1643 assert(format
!= V_028C70_COLOR_INVALID
);
1644 swap
= si_translate_colorswap(surf
->base
.format
);
1645 if (rtex
->resource
.b
.b
.usage
== PIPE_USAGE_STAGING
) {
1646 endian
= V_028C70_ENDIAN_NONE
;
1648 endian
= si_colorformat_endian_swap(format
);
1651 /* blend clamp should be set for all NORM/SRGB types */
1652 if (ntype
== V_028C70_NUMBER_UNORM
||
1653 ntype
== V_028C70_NUMBER_SNORM
||
1654 ntype
== V_028C70_NUMBER_SRGB
)
1657 /* set blend bypass according to docs if SINT/UINT or
1658 8/24 COLOR variants */
1659 if (ntype
== V_028C70_NUMBER_UINT
|| ntype
== V_028C70_NUMBER_SINT
||
1660 format
== V_028C70_COLOR_8_24
|| format
== V_028C70_COLOR_24_8
||
1661 format
== V_028C70_COLOR_X24_8_32_FLOAT
) {
1666 color_info
= S_028C70_FORMAT(format
) |
1667 S_028C70_COMP_SWAP(swap
) |
1668 S_028C70_BLEND_CLAMP(blend_clamp
) |
1669 S_028C70_BLEND_BYPASS(blend_bypass
) |
1670 S_028C70_NUMBER_TYPE(ntype
) |
1671 S_028C70_ENDIAN(endian
);
1673 color_pitch
= S_028C64_TILE_MAX(pitch
);
1675 color_attrib
= S_028C74_TILE_MODE_INDEX(tile_mode_index
) |
1676 S_028C74_FORCE_DST_ALPHA_1(desc
->swizzle
[3] == UTIL_FORMAT_SWIZZLE_1
);
1678 if (rtex
->resource
.b
.b
.nr_samples
> 1) {
1679 unsigned log_samples
= util_logbase2(rtex
->resource
.b
.b
.nr_samples
);
1681 color_attrib
|= S_028C74_NUM_SAMPLES(log_samples
) |
1682 S_028C74_NUM_FRAGMENTS(log_samples
);
1684 if (rtex
->fmask
.size
) {
1685 color_info
|= S_028C70_COMPRESSION(1);
1686 unsigned fmask_bankh
= util_logbase2(rtex
->fmask
.bank_height
);
1688 color_attrib
|= S_028C74_FMASK_TILE_MODE_INDEX(rtex
->fmask
.tile_mode_index
);
1690 if (sctx
->b
.chip_class
== SI
) {
1691 /* due to a hw bug, FMASK_BANK_HEIGHT must be set on SI too */
1692 color_attrib
|= S_028C74_FMASK_BANK_HEIGHT(fmask_bankh
);
1694 if (sctx
->b
.chip_class
>= CIK
) {
1695 color_pitch
|= S_028C64_FMASK_TILE_MAX(rtex
->fmask
.pitch
/ 8 - 1);
1700 if (rtex
->cmask
.size
) {
1701 color_info
|= S_028C70_FAST_CLEAR(1);
1704 offset
+= r600_resource_va(sctx
->b
.b
.screen
, surf
->base
.texture
);
1706 surf
->cb_color_base
= offset
>> 8;
1707 surf
->cb_color_pitch
= color_pitch
;
1708 surf
->cb_color_slice
= S_028C68_TILE_MAX(slice
);
1709 surf
->cb_color_view
= color_view
;
1710 surf
->cb_color_info
= color_info
;
1711 surf
->cb_color_attrib
= color_attrib
;
1713 if (rtex
->cmask
.size
) {
1714 surf
->cb_color_cmask
= (offset
+ rtex
->cmask
.offset
) >> 8;
1715 surf
->cb_color_cmask_slice
= S_028C80_TILE_MAX(rtex
->cmask
.slice_tile_max
);
1717 if (rtex
->fmask
.size
) {
1718 surf
->cb_color_fmask
= (offset
+ rtex
->fmask
.offset
) >> 8;
1719 surf
->cb_color_fmask_slice
= S_028C88_TILE_MAX(rtex
->fmask
.slice_tile_max
);
1722 /* Determine pixel shader export format */
1723 max_comp_size
= si_colorformat_max_comp_size(format
);
1724 if (ntype
== V_028C70_NUMBER_SRGB
||
1725 ((ntype
== V_028C70_NUMBER_UNORM
|| ntype
== V_028C70_NUMBER_SNORM
) &&
1726 max_comp_size
<= 10) ||
1727 (ntype
== V_028C70_NUMBER_FLOAT
&& max_comp_size
<= 16)) {
1728 surf
->export_16bpc
= true;
1731 surf
->color_initialized
= true;
1734 static void si_db(struct si_context
*sctx
, struct si_pm4_state
*pm4
,
1735 const struct pipe_framebuffer_state
*state
)
1737 struct si_screen
*sscreen
= sctx
->screen
;
1738 struct r600_texture
*rtex
;
1739 struct r600_surface
*surf
;
1740 unsigned level
, pitch
, slice
, format
, tile_mode_index
, array_mode
;
1741 unsigned macro_aspect
, tile_split
, stile_split
, bankh
, bankw
, nbanks
, pipe_config
;
1742 uint32_t z_info
, s_info
, db_depth_info
;
1743 uint64_t z_offs
, s_offs
;
1744 uint32_t db_htile_data_base
, db_htile_surface
, pa_su_poly_offset_db_fmt_cntl
;
1746 if (state
->zsbuf
== NULL
) {
1747 si_pm4_set_reg(pm4
, R_028040_DB_Z_INFO
, S_028040_FORMAT(V_028040_Z_INVALID
));
1748 si_pm4_set_reg(pm4
, R_028044_DB_STENCIL_INFO
, S_028044_FORMAT(V_028044_STENCIL_INVALID
));
1752 surf
= (struct r600_surface
*)state
->zsbuf
;
1753 level
= surf
->base
.u
.tex
.level
;
1754 rtex
= (struct r600_texture
*)surf
->base
.texture
;
1756 format
= si_translate_dbformat(rtex
->resource
.b
.b
.format
);
1758 switch (sctx
->framebuffer
.zsbuf
->texture
->format
) {
1759 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
1760 case PIPE_FORMAT_X8Z24_UNORM
:
1761 case PIPE_FORMAT_Z24X8_UNORM
:
1762 case PIPE_FORMAT_Z24_UNORM_S8_UINT
:
1763 pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-24);
1765 case PIPE_FORMAT_Z32_FLOAT
:
1766 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
1767 pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-23) |
1768 S_028B78_POLY_OFFSET_DB_IS_FLOAT_FMT(1);
1770 case PIPE_FORMAT_Z16_UNORM
:
1771 pa_su_poly_offset_db_fmt_cntl
= S_028B78_POLY_OFFSET_NEG_NUM_DB_BITS(-16);
1777 if (format
== V_028040_Z_INVALID
) {
1778 R600_ERR("Invalid DB format: %d, disabling DB.\n", rtex
->resource
.b
.b
.format
);
1780 assert(format
!= V_028040_Z_INVALID
);
1782 s_offs
= z_offs
= r600_resource_va(sctx
->b
.b
.screen
, surf
->base
.texture
);
1783 z_offs
+= rtex
->surface
.level
[level
].offset
;
1784 s_offs
+= rtex
->surface
.stencil_level
[level
].offset
;
1789 pitch
= (rtex
->surface
.level
[level
].nblk_x
/ 8) - 1;
1790 slice
= (rtex
->surface
.level
[level
].nblk_x
* rtex
->surface
.level
[level
].nblk_y
) / 64;
1795 db_depth_info
= S_02803C_ADDR5_SWIZZLE_MASK(1);
1797 z_info
= S_028040_FORMAT(format
);
1798 if (rtex
->resource
.b
.b
.nr_samples
> 1) {
1799 z_info
|= S_028040_NUM_SAMPLES(util_logbase2(rtex
->resource
.b
.b
.nr_samples
));
1802 if (rtex
->surface
.flags
& RADEON_SURF_SBUFFER
)
1803 s_info
= S_028044_FORMAT(V_028044_STENCIL_8
);
1805 s_info
= S_028044_FORMAT(V_028044_STENCIL_INVALID
);
1807 if (sctx
->b
.chip_class
>= CIK
) {
1808 switch (rtex
->surface
.level
[level
].mode
) {
1809 case RADEON_SURF_MODE_2D
:
1810 array_mode
= V_02803C_ARRAY_2D_TILED_THIN1
;
1812 case RADEON_SURF_MODE_1D
:
1813 case RADEON_SURF_MODE_LINEAR_ALIGNED
:
1814 case RADEON_SURF_MODE_LINEAR
:
1816 array_mode
= V_02803C_ARRAY_1D_TILED_THIN1
;
1819 tile_split
= rtex
->surface
.tile_split
;
1820 stile_split
= rtex
->surface
.stencil_tile_split
;
1821 macro_aspect
= rtex
->surface
.mtilea
;
1822 bankw
= rtex
->surface
.bankw
;
1823 bankh
= rtex
->surface
.bankh
;
1824 tile_split
= cik_tile_split(tile_split
);
1825 stile_split
= cik_tile_split(stile_split
);
1826 macro_aspect
= cik_macro_tile_aspect(macro_aspect
);
1827 bankw
= cik_bank_wh(bankw
);
1828 bankh
= cik_bank_wh(bankh
);
1829 nbanks
= cik_num_banks(sscreen
, rtex
->surface
.bpe
, rtex
->surface
.tile_split
);
1830 tile_mode_index
= si_tile_mode_index(rtex
, level
, false);
1831 pipe_config
= cik_db_pipe_config(sscreen
, tile_mode_index
);
1833 db_depth_info
|= S_02803C_ARRAY_MODE(array_mode
) |
1834 S_02803C_PIPE_CONFIG(pipe_config
) |
1835 S_02803C_BANK_WIDTH(bankw
) |
1836 S_02803C_BANK_HEIGHT(bankh
) |
1837 S_02803C_MACRO_TILE_ASPECT(macro_aspect
) |
1838 S_02803C_NUM_BANKS(nbanks
);
1839 z_info
|= S_028040_TILE_SPLIT(tile_split
);
1840 s_info
|= S_028044_TILE_SPLIT(stile_split
);
1842 tile_mode_index
= si_tile_mode_index(rtex
, level
, false);
1843 z_info
|= S_028040_TILE_MODE_INDEX(tile_mode_index
);
1844 tile_mode_index
= si_tile_mode_index(rtex
, level
, true);
1845 s_info
|= S_028044_TILE_MODE_INDEX(tile_mode_index
);
1848 /* HiZ aka depth buffer htile */
1849 /* use htile only for first level */
1850 if (rtex
->htile_buffer
&& !level
) {
1851 const struct util_format_description
*fmt_desc
;
1853 z_info
|= S_028040_TILE_SURFACE_ENABLE(1);
1855 /* This is optimal for the clear value of 1.0 and using
1856 * the LESS and LEQUAL test functions. Set this to 0
1857 * for the opposite case. This can only be changed when
1859 z_info
|= S_028040_ZRANGE_PRECISION(1);
1861 fmt_desc
= util_format_description(rtex
->resource
.b
.b
.format
);
1862 if (!util_format_has_stencil(fmt_desc
)) {
1863 /* Use all of the htile_buffer for depth */
1864 s_info
|= S_028044_TILE_STENCIL_DISABLE(1);
1867 uint64_t va
= r600_resource_va(&sctx
->screen
->b
.b
, &rtex
->htile_buffer
->b
.b
);
1868 db_htile_data_base
= va
>> 8;
1869 db_htile_surface
= S_028ABC_FULL_CACHE(1);
1871 si_pm4_add_bo(pm4
, rtex
->htile_buffer
, RADEON_USAGE_READWRITE
);
1873 db_htile_data_base
= 0;
1874 db_htile_surface
= 0;
1877 si_pm4_set_reg(pm4
, R_028008_DB_DEPTH_VIEW
,
1878 S_028008_SLICE_START(state
->zsbuf
->u
.tex
.first_layer
) |
1879 S_028008_SLICE_MAX(state
->zsbuf
->u
.tex
.last_layer
));
1880 si_pm4_set_reg(pm4
, R_028014_DB_HTILE_DATA_BASE
, db_htile_data_base
);
1882 si_pm4_set_reg(pm4
, R_02803C_DB_DEPTH_INFO
, db_depth_info
);
1883 si_pm4_set_reg(pm4
, R_028040_DB_Z_INFO
, z_info
);
1884 si_pm4_set_reg(pm4
, R_028044_DB_STENCIL_INFO
, s_info
);
1886 si_pm4_add_bo(pm4
, &rtex
->resource
, RADEON_USAGE_READWRITE
);
1887 si_pm4_set_reg(pm4
, R_028048_DB_Z_READ_BASE
, z_offs
);
1888 si_pm4_set_reg(pm4
, R_02804C_DB_STENCIL_READ_BASE
, s_offs
);
1889 si_pm4_set_reg(pm4
, R_028050_DB_Z_WRITE_BASE
, z_offs
);
1890 si_pm4_set_reg(pm4
, R_028054_DB_STENCIL_WRITE_BASE
, s_offs
);
1892 si_pm4_set_reg(pm4
, R_028058_DB_DEPTH_SIZE
, S_028058_PITCH_TILE_MAX(pitch
));
1893 si_pm4_set_reg(pm4
, R_02805C_DB_DEPTH_SLICE
, S_02805C_SLICE_TILE_MAX(slice
));
1895 si_pm4_set_reg(pm4
, R_028ABC_DB_HTILE_SURFACE
, db_htile_surface
);
1896 si_pm4_set_reg(pm4
, R_028B78_PA_SU_POLY_OFFSET_DB_FMT_CNTL
, pa_su_poly_offset_db_fmt_cntl
);
1899 #define FILL_SREG(s0x, s0y, s1x, s1y, s2x, s2y, s3x, s3y) \
1900 (((s0x) & 0xf) | (((s0y) & 0xf) << 4) | \
1901 (((s1x) & 0xf) << 8) | (((s1y) & 0xf) << 12) | \
1902 (((s2x) & 0xf) << 16) | (((s2y) & 0xf) << 20) | \
1903 (((s3x) & 0xf) << 24) | (((s3y) & 0xf) << 28))
1906 * There are two locations (-4, 4), (4, -4). */
1907 static uint32_t sample_locs_2x
[] = {
1908 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1909 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1910 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1911 FILL_SREG(-4, 4, 4, -4, -4, 4, 4, -4),
1913 static unsigned max_dist_2x
= 4;
1915 * There are 4 locations: (-2, -2), (2, 2), (-6, 6), (6, -6). */
1916 static uint32_t sample_locs_4x
[] = {
1917 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1918 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1919 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1920 FILL_SREG(-2, -2, 2, 2, -6, 6, 6, -6),
1922 static unsigned max_dist_4x
= 6;
1923 /* Cayman/SI 8xMSAA */
1924 static uint32_t cm_sample_locs_8x
[] = {
1925 FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
1926 FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
1927 FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
1928 FILL_SREG(-2, -5, 3, -4, -1, 5, -6, -2),
1929 FILL_SREG( 6, 0, 0, 0, -5, 3, 4, 4),
1930 FILL_SREG( 6, 0, 0, 0, -5, 3, 4, 4),
1931 FILL_SREG( 6, 0, 0, 0, -5, 3, 4, 4),
1932 FILL_SREG( 6, 0, 0, 0, -5, 3, 4, 4),
1934 static unsigned cm_max_dist_8x
= 8;
1935 /* Cayman/SI 16xMSAA */
1936 static uint32_t cm_sample_locs_16x
[] = {
1937 FILL_SREG(-7, -3, 7, 3, 1, -5, -5, 5),
1938 FILL_SREG(-7, -3, 7, 3, 1, -5, -5, 5),
1939 FILL_SREG(-7, -3, 7, 3, 1, -5, -5, 5),
1940 FILL_SREG(-7, -3, 7, 3, 1, -5, -5, 5),
1941 FILL_SREG(-3, -7, 3, 7, 5, -1, -1, 1),
1942 FILL_SREG(-3, -7, 3, 7, 5, -1, -1, 1),
1943 FILL_SREG(-3, -7, 3, 7, 5, -1, -1, 1),
1944 FILL_SREG(-3, -7, 3, 7, 5, -1, -1, 1),
1945 FILL_SREG(-8, -6, 4, 2, 2, -8, -2, 6),
1946 FILL_SREG(-8, -6, 4, 2, 2, -8, -2, 6),
1947 FILL_SREG(-8, -6, 4, 2, 2, -8, -2, 6),
1948 FILL_SREG(-8, -6, 4, 2, 2, -8, -2, 6),
1949 FILL_SREG(-4, -2, 0, 4, 6, -4, -6, 0),
1950 FILL_SREG(-4, -2, 0, 4, 6, -4, -6, 0),
1951 FILL_SREG(-4, -2, 0, 4, 6, -4, -6, 0),
1952 FILL_SREG(-4, -2, 0, 4, 6, -4, -6, 0),
1954 static unsigned cm_max_dist_16x
= 8;
1956 static void si_get_sample_position(struct pipe_context
*ctx
,
1957 unsigned sample_count
,
1958 unsigned sample_index
,
1965 switch (sample_count
) {
1968 out_value
[0] = out_value
[1] = 0.5;
1971 offset
= 4 * (sample_index
* 2);
1972 val
.idx
= (sample_locs_2x
[0] >> offset
) & 0xf;
1973 out_value
[0] = (float)(val
.idx
+ 8) / 16.0f
;
1974 val
.idx
= (sample_locs_2x
[0] >> (offset
+ 4)) & 0xf;
1975 out_value
[1] = (float)(val
.idx
+ 8) / 16.0f
;
1978 offset
= 4 * (sample_index
* 2);
1979 val
.idx
= (sample_locs_4x
[0] >> offset
) & 0xf;
1980 out_value
[0] = (float)(val
.idx
+ 8) / 16.0f
;
1981 val
.idx
= (sample_locs_4x
[0] >> (offset
+ 4)) & 0xf;
1982 out_value
[1] = (float)(val
.idx
+ 8) / 16.0f
;
1985 offset
= 4 * (sample_index
% 4 * 2);
1986 index
= (sample_index
/ 4) * 4;
1987 val
.idx
= (cm_sample_locs_8x
[index
] >> offset
) & 0xf;
1988 out_value
[0] = (float)(val
.idx
+ 8) / 16.0f
;
1989 val
.idx
= (cm_sample_locs_8x
[index
] >> (offset
+ 4)) & 0xf;
1990 out_value
[1] = (float)(val
.idx
+ 8) / 16.0f
;
1993 offset
= 4 * (sample_index
% 4 * 2);
1994 index
= (sample_index
/ 4) * 4;
1995 val
.idx
= (cm_sample_locs_16x
[index
] >> offset
) & 0xf;
1996 out_value
[0] = (float)(val
.idx
+ 8) / 16.0f
;
1997 val
.idx
= (cm_sample_locs_16x
[index
] >> (offset
+ 4)) & 0xf;
1998 out_value
[1] = (float)(val
.idx
+ 8) / 16.0f
;
2003 static void si_set_msaa_state(struct si_context
*sctx
, struct si_pm4_state
*pm4
, int nr_samples
)
2005 unsigned max_dist
= 0;
2007 switch (nr_samples
) {
2012 si_pm4_set_reg(pm4
, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0
, sample_locs_2x
[0]);
2013 si_pm4_set_reg(pm4
, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0
, sample_locs_2x
[1]);
2014 si_pm4_set_reg(pm4
, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0
, sample_locs_2x
[2]);
2015 si_pm4_set_reg(pm4
, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0
, sample_locs_2x
[3]);
2016 max_dist
= max_dist_2x
;
2019 si_pm4_set_reg(pm4
, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0
, sample_locs_4x
[0]);
2020 si_pm4_set_reg(pm4
, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0
, sample_locs_4x
[1]);
2021 si_pm4_set_reg(pm4
, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0
, sample_locs_4x
[2]);
2022 si_pm4_set_reg(pm4
, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0
, sample_locs_4x
[3]);
2023 max_dist
= max_dist_4x
;
2026 si_pm4_set_reg(pm4
, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0
, cm_sample_locs_8x
[0]);
2027 si_pm4_set_reg(pm4
, R_028BFC_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_1
, cm_sample_locs_8x
[4]);
2028 si_pm4_set_reg(pm4
, R_028C00_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_2
, 0);
2029 si_pm4_set_reg(pm4
, R_028C04_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_3
, 0);
2030 si_pm4_set_reg(pm4
, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0
, cm_sample_locs_8x
[1]);
2031 si_pm4_set_reg(pm4
, R_028C0C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_1
, cm_sample_locs_8x
[5]);
2032 si_pm4_set_reg(pm4
, R_028C10_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_2
, 0);
2033 si_pm4_set_reg(pm4
, R_028C14_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_3
, 0);
2034 si_pm4_set_reg(pm4
, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0
, cm_sample_locs_8x
[2]);
2035 si_pm4_set_reg(pm4
, R_028C1C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_1
, cm_sample_locs_8x
[6]);
2036 si_pm4_set_reg(pm4
, R_028C20_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_2
, 0);
2037 si_pm4_set_reg(pm4
, R_028C24_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_3
, 0);
2038 si_pm4_set_reg(pm4
, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0
, cm_sample_locs_8x
[3]);
2039 si_pm4_set_reg(pm4
, R_028C2C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_1
, cm_sample_locs_8x
[7]);
2040 max_dist
= cm_max_dist_8x
;
2043 si_pm4_set_reg(pm4
, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0
, cm_sample_locs_16x
[0]);
2044 si_pm4_set_reg(pm4
, R_028BFC_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_1
, cm_sample_locs_16x
[4]);
2045 si_pm4_set_reg(pm4
, R_028C00_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_2
, cm_sample_locs_16x
[8]);
2046 si_pm4_set_reg(pm4
, R_028C04_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_3
, cm_sample_locs_16x
[12]);
2047 si_pm4_set_reg(pm4
, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0
, cm_sample_locs_16x
[1]);
2048 si_pm4_set_reg(pm4
, R_028C0C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_1
, cm_sample_locs_16x
[5]);
2049 si_pm4_set_reg(pm4
, R_028C10_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_2
, cm_sample_locs_16x
[9]);
2050 si_pm4_set_reg(pm4
, R_028C14_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_3
, cm_sample_locs_16x
[13]);
2051 si_pm4_set_reg(pm4
, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0
, cm_sample_locs_16x
[2]);
2052 si_pm4_set_reg(pm4
, R_028C1C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_1
, cm_sample_locs_16x
[6]);
2053 si_pm4_set_reg(pm4
, R_028C20_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_2
, cm_sample_locs_16x
[10]);
2054 si_pm4_set_reg(pm4
, R_028C24_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_3
, cm_sample_locs_16x
[14]);
2055 si_pm4_set_reg(pm4
, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0
, cm_sample_locs_16x
[3]);
2056 si_pm4_set_reg(pm4
, R_028C2C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_1
, cm_sample_locs_16x
[7]);
2057 si_pm4_set_reg(pm4
, R_028C30_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_2
, cm_sample_locs_16x
[11]);
2058 si_pm4_set_reg(pm4
, R_028C34_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_3
, cm_sample_locs_16x
[15]);
2059 max_dist
= cm_max_dist_16x
;
2063 if (nr_samples
> 1) {
2064 unsigned log_samples
= util_logbase2(nr_samples
);
2066 si_pm4_set_reg(pm4
, R_028BDC_PA_SC_LINE_CNTL
,
2067 S_028BDC_LAST_PIXEL(1) |
2068 S_028BDC_EXPAND_LINE_WIDTH(1));
2069 si_pm4_set_reg(pm4
, R_028BE0_PA_SC_AA_CONFIG
,
2070 S_028BE0_MSAA_NUM_SAMPLES(log_samples
) |
2071 S_028BE0_MAX_SAMPLE_DIST(max_dist
) |
2072 S_028BE0_MSAA_EXPOSED_SAMPLES(log_samples
));
2074 si_pm4_set_reg(pm4
, R_028804_DB_EQAA
,
2075 S_028804_MAX_ANCHOR_SAMPLES(log_samples
) |
2076 S_028804_PS_ITER_SAMPLES(log_samples
) |
2077 S_028804_MASK_EXPORT_NUM_SAMPLES(log_samples
) |
2078 S_028804_ALPHA_TO_MASK_NUM_SAMPLES(log_samples
) |
2079 S_028804_HIGH_QUALITY_INTERSECTIONS(1) |
2080 S_028804_STATIC_ANCHOR_ASSOCIATIONS(1));
2082 si_pm4_set_reg(pm4
, R_028BDC_PA_SC_LINE_CNTL
, S_028BDC_LAST_PIXEL(1));
2083 si_pm4_set_reg(pm4
, R_028BE0_PA_SC_AA_CONFIG
, 0);
2085 si_pm4_set_reg(pm4
, R_028804_DB_EQAA
,
2086 S_028804_HIGH_QUALITY_INTERSECTIONS(1) |
2087 S_028804_STATIC_ANCHOR_ASSOCIATIONS(1));
2091 static void si_set_framebuffer_state(struct pipe_context
*ctx
,
2092 const struct pipe_framebuffer_state
*state
)
2094 struct si_context
*sctx
= (struct si_context
*)ctx
;
2095 struct r600_surface
*surf
= NULL
;
2096 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
2102 if (sctx
->framebuffer
.nr_cbufs
) {
2103 sctx
->b
.flags
|= R600_CONTEXT_FLUSH_AND_INV_CB
|
2104 R600_CONTEXT_FLUSH_AND_INV_CB_META
;
2106 if (sctx
->framebuffer
.zsbuf
) {
2107 sctx
->b
.flags
|= R600_CONTEXT_FLUSH_AND_INV_DB
|
2108 R600_CONTEXT_FLUSH_AND_INV_DB_META
;
2111 util_copy_framebuffer_state(&sctx
->framebuffer
, state
);
2114 sctx
->export_16bpc
= 0;
2115 sctx
->fb_compressed_cb_mask
= 0;
2117 for (i
= 0; i
< state
->nr_cbufs
; i
++) {
2118 struct r600_texture
*rtex
;
2120 if (!state
->cbufs
[i
]) {
2121 si_pm4_set_reg(pm4
, R_028C70_CB_COLOR0_INFO
+ i
* 0x3C,
2122 S_028C70_FORMAT(V_028C70_COLOR_INVALID
));
2126 surf
= (struct r600_surface
*)state
->cbufs
[i
];
2127 rtex
= (struct r600_texture
*)surf
->base
.texture
;
2129 if (!surf
->color_initialized
) {
2130 si_initialize_color_surface(sctx
, surf
);
2133 if (surf
->export_16bpc
) {
2134 sctx
->export_16bpc
|= 1 << i
;
2137 if (rtex
->fmask
.size
|| rtex
->cmask
.size
) {
2138 sctx
->fb_compressed_cb_mask
|= 1 << i
;
2141 si_pm4_add_bo(pm4
, &rtex
->resource
, RADEON_USAGE_READWRITE
);
2142 si_pm4_set_reg(pm4
, R_028C60_CB_COLOR0_BASE
+ i
* 0x3C, surf
->cb_color_base
);
2143 si_pm4_set_reg(pm4
, R_028C64_CB_COLOR0_PITCH
+ i
* 0x3C, surf
->cb_color_pitch
);
2144 si_pm4_set_reg(pm4
, R_028C68_CB_COLOR0_SLICE
+ i
* 0x3C, surf
->cb_color_slice
);
2145 si_pm4_set_reg(pm4
, R_028C6C_CB_COLOR0_VIEW
+ i
* 0x3C, surf
->cb_color_view
);
2146 si_pm4_set_reg(pm4
, R_028C70_CB_COLOR0_INFO
+ i
* 0x3C, surf
->cb_color_info
);
2147 si_pm4_set_reg(pm4
, R_028C74_CB_COLOR0_ATTRIB
+ i
* 0x3C, surf
->cb_color_attrib
);
2148 si_pm4_set_reg(pm4
, R_028C7C_CB_COLOR0_CMASK
+ i
* 0x3C, surf
->cb_color_cmask
);
2149 si_pm4_set_reg(pm4
, R_028C80_CB_COLOR0_CMASK_SLICE
+ i
* 0x3C, surf
->cb_color_cmask_slice
);
2150 si_pm4_set_reg(pm4
, R_028C84_CB_COLOR0_FMASK
+ i
* 0x3C, surf
->cb_color_fmask
);
2151 si_pm4_set_reg(pm4
, R_028C88_CB_COLOR0_FMASK_SLICE
+ i
* 0x3C, surf
->cb_color_fmask_slice
);
2153 /* Set CB_COLOR1_INFO for possible dual-src blending. */
2154 if (i
== 1 && surf
) {
2155 si_pm4_set_reg(pm4
, R_028C70_CB_COLOR0_INFO
+ 1 * 0x3C, surf
->cb_color_info
);
2156 /* Also set the 16BPC export. */
2157 if (surf
->export_16bpc
) {
2158 sctx
->export_16bpc
|= 1 << 1;
2162 for (; i
< 8; i
++) {
2163 si_pm4_set_reg(pm4
, R_028C70_CB_COLOR0_INFO
+ i
* 0x3C,
2164 S_028C70_FORMAT(V_028C70_COLOR_INVALID
));
2167 assert(!(sctx
->export_16bpc
& ~0xff));
2168 si_db(sctx
, pm4
, state
);
2170 /* PA_SC_WINDOW_SCISSOR_TL is set in si_init_config() */
2171 si_pm4_set_reg(pm4
, R_028208_PA_SC_WINDOW_SCISSOR_BR
,
2172 S_028208_BR_X(state
->width
) | S_028208_BR_Y(state
->height
));
2174 nr_samples
= util_framebuffer_get_num_samples(state
);
2176 si_set_msaa_state(sctx
, pm4
, nr_samples
);
2177 sctx
->fb_log_samples
= util_logbase2(nr_samples
);
2178 sctx
->fb_cb0_is_integer
= state
->nr_cbufs
&& state
->cbufs
[0] &&
2179 util_format_is_pure_integer(state
->cbufs
[0]->format
);
2181 si_pm4_set_state(sctx
, framebuffer
, pm4
);
2182 si_update_fb_rs_state(sctx
);
2183 si_update_fb_blend_state(sctx
);
2190 /* Compute the key for the hw shader variant */
2191 static INLINE
void si_shader_selector_key(struct pipe_context
*ctx
,
2192 struct si_pipe_shader_selector
*sel
,
2193 union si_shader_key
*key
)
2195 struct si_context
*sctx
= (struct si_context
*)ctx
;
2196 memset(key
, 0, sizeof(*key
));
2198 if ((sel
->type
== PIPE_SHADER_VERTEX
|| sel
->type
== PIPE_SHADER_GEOMETRY
) &&
2199 sctx
->queued
.named
.rasterizer
) {
2200 if (sctx
->queued
.named
.rasterizer
->clip_plane_enable
& 0xf0)
2201 key
->vs
.ucps_enabled
|= 0x2;
2202 if (sctx
->queued
.named
.rasterizer
->clip_plane_enable
& 0xf)
2203 key
->vs
.ucps_enabled
|= 0x1;
2206 if (sel
->type
== PIPE_SHADER_VERTEX
) {
2208 if (!sctx
->vertex_elements
)
2211 for (i
= 0; i
< sctx
->vertex_elements
->count
; ++i
)
2212 key
->vs
.instance_divisors
[i
] = sctx
->vertex_elements
->elements
[i
].instance_divisor
;
2214 key
->vs
.as_es
= sctx
->gs_shader
!= NULL
;
2215 } else if (sel
->type
== PIPE_SHADER_FRAGMENT
) {
2216 if (sel
->fs_write_all
)
2217 key
->ps
.nr_cbufs
= sctx
->framebuffer
.nr_cbufs
;
2218 key
->ps
.export_16bpc
= sctx
->export_16bpc
;
2220 if (sctx
->queued
.named
.rasterizer
) {
2221 key
->ps
.color_two_side
= sctx
->queued
.named
.rasterizer
->two_side
;
2222 key
->ps
.flatshade
= sctx
->queued
.named
.rasterizer
->flatshade
;
2224 if (sctx
->queued
.named
.blend
) {
2225 key
->ps
.alpha_to_one
= sctx
->queued
.named
.blend
->alpha_to_one
&&
2226 sctx
->queued
.named
.rasterizer
->multisample_enable
&&
2227 !sctx
->fb_cb0_is_integer
;
2230 if (sctx
->queued
.named
.dsa
) {
2231 key
->ps
.alpha_func
= sctx
->queued
.named
.dsa
->alpha_func
;
2233 /* Alpha-test should be disabled if colorbuffer 0 is integer. */
2234 if (sctx
->framebuffer
.nr_cbufs
&&
2235 sctx
->framebuffer
.cbufs
[0] &&
2236 util_format_is_pure_integer(sctx
->framebuffer
.cbufs
[0]->texture
->format
))
2237 key
->ps
.alpha_func
= PIPE_FUNC_ALWAYS
;
2239 key
->ps
.alpha_func
= PIPE_FUNC_ALWAYS
;
2244 /* Select the hw shader variant depending on the current state. */
2245 int si_shader_select(struct pipe_context
*ctx
,
2246 struct si_pipe_shader_selector
*sel
)
2248 union si_shader_key key
;
2249 struct si_pipe_shader
* shader
= NULL
;
2252 si_shader_selector_key(ctx
, sel
, &key
);
2254 /* Check if we don't need to change anything.
2255 * This path is also used for most shaders that don't need multiple
2256 * variants, it will cost just a computation of the key and this
2258 if (likely(sel
->current
&& memcmp(&sel
->current
->key
, &key
, sizeof(key
)) == 0)) {
2262 /* lookup if we have other variants in the list */
2263 if (sel
->num_shaders
> 1) {
2264 struct si_pipe_shader
*p
= sel
->current
, *c
= p
->next_variant
;
2266 while (c
&& memcmp(&c
->key
, &key
, sizeof(key
)) != 0) {
2268 c
= c
->next_variant
;
2272 p
->next_variant
= c
->next_variant
;
2278 shader
->next_variant
= sel
->current
;
2279 sel
->current
= shader
;
2281 shader
= CALLOC(1, sizeof(struct si_pipe_shader
));
2282 shader
->selector
= sel
;
2285 shader
->next_variant
= sel
->current
;
2286 sel
->current
= shader
;
2287 r
= si_pipe_shader_create(ctx
, shader
);
2289 R600_ERR("Failed to build shader variant (type=%u) %d\n",
2291 sel
->current
= NULL
;
2301 static void *si_create_shader_state(struct pipe_context
*ctx
,
2302 const struct pipe_shader_state
*state
,
2303 unsigned pipe_shader_type
)
2305 struct si_pipe_shader_selector
*sel
= CALLOC_STRUCT(si_pipe_shader_selector
);
2308 sel
->type
= pipe_shader_type
;
2309 sel
->tokens
= tgsi_dup_tokens(state
->tokens
);
2310 sel
->so
= state
->stream_output
;
2312 if (pipe_shader_type
== PIPE_SHADER_FRAGMENT
) {
2313 struct tgsi_shader_info info
;
2315 tgsi_scan_shader(state
->tokens
, &info
);
2316 sel
->fs_write_all
= info
.color0_writes_all_cbufs
;
2319 r
= si_shader_select(ctx
, sel
);
2328 static void *si_create_fs_state(struct pipe_context
*ctx
,
2329 const struct pipe_shader_state
*state
)
2331 return si_create_shader_state(ctx
, state
, PIPE_SHADER_FRAGMENT
);
2334 #if HAVE_LLVM >= 0x0305
2336 static void *si_create_gs_state(struct pipe_context
*ctx
,
2337 const struct pipe_shader_state
*state
)
2339 return si_create_shader_state(ctx
, state
, PIPE_SHADER_GEOMETRY
);
2344 static void *si_create_vs_state(struct pipe_context
*ctx
,
2345 const struct pipe_shader_state
*state
)
2347 return si_create_shader_state(ctx
, state
, PIPE_SHADER_VERTEX
);
2350 static void si_bind_vs_shader(struct pipe_context
*ctx
, void *state
)
2352 struct si_context
*sctx
= (struct si_context
*)ctx
;
2353 struct si_pipe_shader_selector
*sel
= state
;
2355 if (sctx
->vs_shader
== sel
)
2358 if (!sel
|| !sel
->current
)
2361 sctx
->vs_shader
= sel
;
2364 #if HAVE_LLVM >= 0x0305
2366 static void si_bind_gs_shader(struct pipe_context
*ctx
, void *state
)
2368 struct si_context
*sctx
= (struct si_context
*)ctx
;
2369 struct si_pipe_shader_selector
*sel
= state
;
2371 if (sctx
->gs_shader
== sel
)
2374 sctx
->gs_shader
= sel
;
2379 static void si_bind_ps_shader(struct pipe_context
*ctx
, void *state
)
2381 struct si_context
*sctx
= (struct si_context
*)ctx
;
2382 struct si_pipe_shader_selector
*sel
= state
;
2384 if (sctx
->ps_shader
== sel
)
2387 if (!sel
|| !sel
->current
)
2388 sel
= sctx
->dummy_pixel_shader
;
2390 sctx
->ps_shader
= sel
;
2393 static void si_delete_shader_selector(struct pipe_context
*ctx
,
2394 struct si_pipe_shader_selector
*sel
)
2396 struct si_context
*sctx
= (struct si_context
*)ctx
;
2397 struct si_pipe_shader
*p
= sel
->current
, *c
;
2400 c
= p
->next_variant
;
2401 si_pm4_delete_state(sctx
, vs
, p
->pm4
);
2402 si_pipe_shader_destroy(ctx
, p
);
2411 static void si_delete_vs_shader(struct pipe_context
*ctx
, void *state
)
2413 struct si_context
*sctx
= (struct si_context
*)ctx
;
2414 struct si_pipe_shader_selector
*sel
= (struct si_pipe_shader_selector
*)state
;
2416 if (sctx
->vs_shader
== sel
) {
2417 sctx
->vs_shader
= NULL
;
2420 si_delete_shader_selector(ctx
, sel
);
2423 #if HAVE_LLVM >= 0x0305
2425 static void si_delete_gs_shader(struct pipe_context
*ctx
, void *state
)
2427 struct si_context
*sctx
= (struct si_context
*)ctx
;
2428 struct si_pipe_shader_selector
*sel
= (struct si_pipe_shader_selector
*)state
;
2430 if (sctx
->gs_shader
== sel
) {
2431 sctx
->gs_shader
= NULL
;
2434 si_delete_shader_selector(ctx
, sel
);
2439 static void si_delete_ps_shader(struct pipe_context
*ctx
, void *state
)
2441 struct si_context
*sctx
= (struct si_context
*)ctx
;
2442 struct si_pipe_shader_selector
*sel
= (struct si_pipe_shader_selector
*)state
;
2444 if (sctx
->ps_shader
== sel
) {
2445 sctx
->ps_shader
= NULL
;
2448 si_delete_shader_selector(ctx
, sel
);
2455 static struct pipe_sampler_view
*si_create_sampler_view(struct pipe_context
*ctx
,
2456 struct pipe_resource
*texture
,
2457 const struct pipe_sampler_view
*state
)
2459 struct si_pipe_sampler_view
*view
= CALLOC_STRUCT(si_pipe_sampler_view
);
2460 struct r600_texture
*tmp
= (struct r600_texture
*)texture
;
2461 const struct util_format_description
*desc
;
2462 unsigned format
, num_format
;
2464 unsigned char state_swizzle
[4], swizzle
[4];
2465 unsigned height
, depth
, width
;
2466 enum pipe_format pipe_format
= state
->format
;
2467 struct radeon_surface_level
*surflevel
;
2474 /* initialize base object */
2475 view
->base
= *state
;
2476 view
->base
.texture
= NULL
;
2477 pipe_resource_reference(&view
->base
.texture
, texture
);
2478 view
->base
.reference
.count
= 1;
2479 view
->base
.context
= ctx
;
2480 view
->resource
= &tmp
->resource
;
2482 /* Buffer resource. */
2483 if (texture
->target
== PIPE_BUFFER
) {
2486 desc
= util_format_description(state
->format
);
2487 first_non_void
= util_format_get_first_non_void_channel(state
->format
);
2488 stride
= desc
->block
.bits
/ 8;
2489 va
= r600_resource_va(ctx
->screen
, texture
) + state
->u
.buf
.first_element
*stride
;
2490 format
= si_translate_buffer_dataformat(ctx
->screen
, desc
, first_non_void
);
2491 num_format
= si_translate_buffer_numformat(ctx
->screen
, desc
, first_non_void
);
2493 view
->state
[0] = va
;
2494 view
->state
[1] = S_008F04_BASE_ADDRESS_HI(va
>> 32) |
2495 S_008F04_STRIDE(stride
);
2496 view
->state
[2] = state
->u
.buf
.last_element
+ 1 - state
->u
.buf
.first_element
;
2497 view
->state
[3] = S_008F0C_DST_SEL_X(si_map_swizzle(desc
->swizzle
[0])) |
2498 S_008F0C_DST_SEL_Y(si_map_swizzle(desc
->swizzle
[1])) |
2499 S_008F0C_DST_SEL_Z(si_map_swizzle(desc
->swizzle
[2])) |
2500 S_008F0C_DST_SEL_W(si_map_swizzle(desc
->swizzle
[3])) |
2501 S_008F0C_NUM_FORMAT(num_format
) |
2502 S_008F0C_DATA_FORMAT(format
);
2506 state_swizzle
[0] = state
->swizzle_r
;
2507 state_swizzle
[1] = state
->swizzle_g
;
2508 state_swizzle
[2] = state
->swizzle_b
;
2509 state_swizzle
[3] = state
->swizzle_a
;
2511 surflevel
= tmp
->surface
.level
;
2513 /* Texturing with separate depth and stencil. */
2514 if (tmp
->is_depth
&& !tmp
->is_flushing_texture
) {
2515 switch (pipe_format
) {
2516 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT
:
2517 pipe_format
= PIPE_FORMAT_Z32_FLOAT
;
2519 case PIPE_FORMAT_X8Z24_UNORM
:
2520 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
2521 /* Z24 is always stored like this. */
2522 pipe_format
= PIPE_FORMAT_Z24X8_UNORM
;
2524 case PIPE_FORMAT_X24S8_UINT
:
2525 case PIPE_FORMAT_S8X24_UINT
:
2526 case PIPE_FORMAT_X32_S8X24_UINT
:
2527 pipe_format
= PIPE_FORMAT_S8_UINT
;
2528 surflevel
= tmp
->surface
.stencil_level
;
2534 desc
= util_format_description(pipe_format
);
2536 if (desc
->colorspace
== UTIL_FORMAT_COLORSPACE_ZS
) {
2537 const unsigned char swizzle_xxxx
[4] = {0, 0, 0, 0};
2538 const unsigned char swizzle_yyyy
[4] = {1, 1, 1, 1};
2540 switch (pipe_format
) {
2541 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
2542 case PIPE_FORMAT_X24S8_UINT
:
2543 case PIPE_FORMAT_X32_S8X24_UINT
:
2544 case PIPE_FORMAT_X8Z24_UNORM
:
2545 util_format_compose_swizzles(swizzle_yyyy
, state_swizzle
, swizzle
);
2548 util_format_compose_swizzles(swizzle_xxxx
, state_swizzle
, swizzle
);
2551 util_format_compose_swizzles(desc
->swizzle
, state_swizzle
, swizzle
);
2554 first_non_void
= util_format_get_first_non_void_channel(pipe_format
);
2556 switch (pipe_format
) {
2557 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
2558 num_format
= V_008F14_IMG_NUM_FORMAT_UNORM
;
2561 if (first_non_void
< 0) {
2562 if (util_format_is_compressed(pipe_format
)) {
2563 switch (pipe_format
) {
2564 case PIPE_FORMAT_DXT1_SRGB
:
2565 case PIPE_FORMAT_DXT1_SRGBA
:
2566 case PIPE_FORMAT_DXT3_SRGBA
:
2567 case PIPE_FORMAT_DXT5_SRGBA
:
2568 num_format
= V_008F14_IMG_NUM_FORMAT_SRGB
;
2570 case PIPE_FORMAT_RGTC1_SNORM
:
2571 case PIPE_FORMAT_LATC1_SNORM
:
2572 case PIPE_FORMAT_RGTC2_SNORM
:
2573 case PIPE_FORMAT_LATC2_SNORM
:
2574 num_format
= V_008F14_IMG_NUM_FORMAT_SNORM
;
2577 num_format
= V_008F14_IMG_NUM_FORMAT_UNORM
;
2581 num_format
= V_008F14_IMG_NUM_FORMAT_FLOAT
;
2583 } else if (desc
->colorspace
== UTIL_FORMAT_COLORSPACE_SRGB
) {
2584 num_format
= V_008F14_IMG_NUM_FORMAT_SRGB
;
2586 num_format
= V_008F14_IMG_NUM_FORMAT_UNORM
;
2588 switch (desc
->channel
[first_non_void
].type
) {
2589 case UTIL_FORMAT_TYPE_FLOAT
:
2590 num_format
= V_008F14_IMG_NUM_FORMAT_FLOAT
;
2592 case UTIL_FORMAT_TYPE_SIGNED
:
2593 if (desc
->channel
[first_non_void
].normalized
)
2594 num_format
= V_008F14_IMG_NUM_FORMAT_SNORM
;
2595 else if (desc
->channel
[first_non_void
].pure_integer
)
2596 num_format
= V_008F14_IMG_NUM_FORMAT_SINT
;
2598 num_format
= V_008F14_IMG_NUM_FORMAT_SSCALED
;
2600 case UTIL_FORMAT_TYPE_UNSIGNED
:
2601 if (desc
->channel
[first_non_void
].normalized
)
2602 num_format
= V_008F14_IMG_NUM_FORMAT_UNORM
;
2603 else if (desc
->channel
[first_non_void
].pure_integer
)
2604 num_format
= V_008F14_IMG_NUM_FORMAT_UINT
;
2606 num_format
= V_008F14_IMG_NUM_FORMAT_USCALED
;
2611 format
= si_translate_texformat(ctx
->screen
, pipe_format
, desc
, first_non_void
);
2616 /* not supported any more */
2617 //endian = si_colorformat_endian_swap(format);
2619 width
= surflevel
[0].npix_x
;
2620 height
= surflevel
[0].npix_y
;
2621 depth
= surflevel
[0].npix_z
;
2622 pitch
= surflevel
[0].nblk_x
* util_format_get_blockwidth(pipe_format
);
2624 if (texture
->target
== PIPE_TEXTURE_1D_ARRAY
) {
2626 depth
= texture
->array_size
;
2627 } else if (texture
->target
== PIPE_TEXTURE_2D_ARRAY
) {
2628 depth
= texture
->array_size
;
2631 va
= r600_resource_va(ctx
->screen
, texture
);
2632 va
+= surflevel
[0].offset
;
2633 va
+= tmp
->mipmap_shift
* surflevel
[texture
->last_level
].slice_size
;
2634 view
->state
[0] = va
>> 8;
2635 view
->state
[1] = (S_008F14_BASE_ADDRESS_HI(va
>> 40) |
2636 S_008F14_DATA_FORMAT(format
) |
2637 S_008F14_NUM_FORMAT(num_format
));
2638 view
->state
[2] = (S_008F18_WIDTH(width
- 1) |
2639 S_008F18_HEIGHT(height
- 1));
2640 view
->state
[3] = (S_008F1C_DST_SEL_X(si_map_swizzle(swizzle
[0])) |
2641 S_008F1C_DST_SEL_Y(si_map_swizzle(swizzle
[1])) |
2642 S_008F1C_DST_SEL_Z(si_map_swizzle(swizzle
[2])) |
2643 S_008F1C_DST_SEL_W(si_map_swizzle(swizzle
[3])) |
2644 S_008F1C_BASE_LEVEL(texture
->nr_samples
> 1 ?
2645 0 : state
->u
.tex
.first_level
- tmp
->mipmap_shift
) |
2646 S_008F1C_LAST_LEVEL(texture
->nr_samples
> 1 ?
2647 util_logbase2(texture
->nr_samples
) :
2648 state
->u
.tex
.last_level
- tmp
->mipmap_shift
) |
2649 S_008F1C_TILING_INDEX(si_tile_mode_index(tmp
, 0, false)) |
2650 S_008F1C_POW2_PAD(texture
->last_level
> 0) |
2651 S_008F1C_TYPE(si_tex_dim(texture
->target
, texture
->nr_samples
)));
2652 view
->state
[4] = (S_008F20_DEPTH(depth
- 1) | S_008F20_PITCH(pitch
- 1));
2653 view
->state
[5] = (S_008F24_BASE_ARRAY(state
->u
.tex
.first_layer
) |
2654 S_008F24_LAST_ARRAY(state
->u
.tex
.last_layer
));
2658 /* Initialize the sampler view for FMASK. */
2659 if (tmp
->fmask
.size
) {
2660 uint64_t va
= r600_resource_va(ctx
->screen
, texture
) + tmp
->fmask
.offset
;
2661 uint32_t fmask_format
;
2663 switch (texture
->nr_samples
) {
2665 fmask_format
= V_008F14_IMG_DATA_FORMAT_FMASK8_S2_F2
;
2668 fmask_format
= V_008F14_IMG_DATA_FORMAT_FMASK8_S4_F4
;
2671 fmask_format
= V_008F14_IMG_DATA_FORMAT_FMASK32_S8_F8
;
2675 fmask_format
= V_008F14_IMG_DATA_FORMAT_INVALID
;
2678 view
->fmask_state
[0] = va
>> 8;
2679 view
->fmask_state
[1] = S_008F14_BASE_ADDRESS_HI(va
>> 40) |
2680 S_008F14_DATA_FORMAT(fmask_format
) |
2681 S_008F14_NUM_FORMAT(V_008F14_IMG_NUM_FORMAT_UINT
);
2682 view
->fmask_state
[2] = S_008F18_WIDTH(width
- 1) |
2683 S_008F18_HEIGHT(height
- 1);
2684 view
->fmask_state
[3] = S_008F1C_DST_SEL_X(V_008F1C_SQ_SEL_X
) |
2685 S_008F1C_DST_SEL_Y(V_008F1C_SQ_SEL_X
) |
2686 S_008F1C_DST_SEL_Z(V_008F1C_SQ_SEL_X
) |
2687 S_008F1C_DST_SEL_W(V_008F1C_SQ_SEL_X
) |
2688 S_008F1C_TILING_INDEX(tmp
->fmask
.tile_mode_index
) |
2689 S_008F1C_TYPE(si_tex_dim(texture
->target
, 0));
2690 view
->fmask_state
[4] = S_008F20_DEPTH(depth
- 1) |
2691 S_008F20_PITCH(tmp
->fmask
.pitch
- 1);
2692 view
->fmask_state
[5] = S_008F24_BASE_ARRAY(state
->u
.tex
.first_layer
) |
2693 S_008F24_LAST_ARRAY(state
->u
.tex
.last_layer
);
2694 view
->fmask_state
[6] = 0;
2695 view
->fmask_state
[7] = 0;
2701 static void si_sampler_view_destroy(struct pipe_context
*ctx
,
2702 struct pipe_sampler_view
*state
)
2704 struct r600_pipe_sampler_view
*resource
= (struct r600_pipe_sampler_view
*)state
;
2706 pipe_resource_reference(&state
->texture
, NULL
);
2710 static bool wrap_mode_uses_border_color(unsigned wrap
, bool linear_filter
)
2712 return wrap
== PIPE_TEX_WRAP_CLAMP_TO_BORDER
||
2713 wrap
== PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
||
2715 (wrap
== PIPE_TEX_WRAP_CLAMP
||
2716 wrap
== PIPE_TEX_WRAP_MIRROR_CLAMP
));
2719 static bool sampler_state_needs_border_color(const struct pipe_sampler_state
*state
)
2721 bool linear_filter
= state
->min_img_filter
!= PIPE_TEX_FILTER_NEAREST
||
2722 state
->mag_img_filter
!= PIPE_TEX_FILTER_NEAREST
;
2724 return (state
->border_color
.ui
[0] || state
->border_color
.ui
[1] ||
2725 state
->border_color
.ui
[2] || state
->border_color
.ui
[3]) &&
2726 (wrap_mode_uses_border_color(state
->wrap_s
, linear_filter
) ||
2727 wrap_mode_uses_border_color(state
->wrap_t
, linear_filter
) ||
2728 wrap_mode_uses_border_color(state
->wrap_r
, linear_filter
));
2731 static void *si_create_sampler_state(struct pipe_context
*ctx
,
2732 const struct pipe_sampler_state
*state
)
2734 struct si_pipe_sampler_state
*rstate
= CALLOC_STRUCT(si_pipe_sampler_state
);
2735 unsigned aniso_flag_offset
= state
->max_anisotropy
> 1 ? 2 : 0;
2736 unsigned border_color_type
;
2738 if (rstate
== NULL
) {
2742 if (sampler_state_needs_border_color(state
))
2743 border_color_type
= V_008F3C_SQ_TEX_BORDER_COLOR_REGISTER
;
2745 border_color_type
= V_008F3C_SQ_TEX_BORDER_COLOR_TRANS_BLACK
;
2747 rstate
->val
[0] = (S_008F30_CLAMP_X(si_tex_wrap(state
->wrap_s
)) |
2748 S_008F30_CLAMP_Y(si_tex_wrap(state
->wrap_t
)) |
2749 S_008F30_CLAMP_Z(si_tex_wrap(state
->wrap_r
)) |
2750 (state
->max_anisotropy
& 0x7) << 9 | /* XXX */
2751 S_008F30_DEPTH_COMPARE_FUNC(si_tex_compare(state
->compare_func
)) |
2752 S_008F30_FORCE_UNNORMALIZED(!state
->normalized_coords
) |
2753 aniso_flag_offset
<< 16 | /* XXX */
2754 S_008F30_DISABLE_CUBE_WRAP(!state
->seamless_cube_map
));
2755 rstate
->val
[1] = (S_008F34_MIN_LOD(S_FIXED(CLAMP(state
->min_lod
, 0, 15), 8)) |
2756 S_008F34_MAX_LOD(S_FIXED(CLAMP(state
->max_lod
, 0, 15), 8)));
2757 rstate
->val
[2] = (S_008F38_LOD_BIAS(S_FIXED(CLAMP(state
->lod_bias
, -16, 16), 8)) |
2758 S_008F38_XY_MAG_FILTER(si_tex_filter(state
->mag_img_filter
)) |
2759 S_008F38_XY_MIN_FILTER(si_tex_filter(state
->min_img_filter
)) |
2760 S_008F38_MIP_FILTER(si_tex_mipfilter(state
->min_mip_filter
)));
2761 rstate
->val
[3] = S_008F3C_BORDER_COLOR_TYPE(border_color_type
);
2763 if (border_color_type
== V_008F3C_SQ_TEX_BORDER_COLOR_REGISTER
) {
2764 memcpy(rstate
->border_color
, state
->border_color
.ui
,
2765 sizeof(rstate
->border_color
));
2771 /* XXX consider moving this function to si_descriptors.c for gcc to inline
2772 * the si_set_sampler_view calls. LTO might help too. */
2773 static void si_set_sampler_views(struct pipe_context
*ctx
,
2774 unsigned shader
, unsigned start
,
2776 struct pipe_sampler_view
**views
)
2778 struct si_context
*sctx
= (struct si_context
*)ctx
;
2779 struct si_textures_info
*samplers
= &sctx
->samplers
[shader
];
2780 struct si_pipe_sampler_view
**rviews
= (struct si_pipe_sampler_view
**)views
;
2783 if (shader
>= SI_NUM_SHADERS
)
2788 for (i
= 0; i
< count
; i
++) {
2790 samplers
->depth_texture_mask
&= ~(1 << i
);
2791 samplers
->compressed_colortex_mask
&= ~(1 << i
);
2792 si_set_sampler_view(sctx
, shader
, i
, NULL
, NULL
);
2793 si_set_sampler_view(sctx
, shader
, FMASK_TEX_OFFSET
+ i
,
2798 si_set_sampler_view(sctx
, shader
, i
, views
[i
], rviews
[i
]->state
);
2800 if (views
[i
]->texture
->target
!= PIPE_BUFFER
) {
2801 struct r600_texture
*rtex
=
2802 (struct r600_texture
*)views
[i
]->texture
;
2804 if (rtex
->is_depth
&& !rtex
->is_flushing_texture
) {
2805 samplers
->depth_texture_mask
|= 1 << i
;
2807 samplers
->depth_texture_mask
&= ~(1 << i
);
2809 if (rtex
->cmask
.size
|| rtex
->fmask
.size
) {
2810 samplers
->compressed_colortex_mask
|= 1 << i
;
2812 samplers
->compressed_colortex_mask
&= ~(1 << i
);
2815 if (rtex
->fmask
.size
) {
2816 si_set_sampler_view(sctx
, shader
, FMASK_TEX_OFFSET
+ i
,
2817 views
[i
], rviews
[i
]->fmask_state
);
2819 si_set_sampler_view(sctx
, shader
, FMASK_TEX_OFFSET
+ i
,
2824 for (; i
< samplers
->n_views
; i
++) {
2825 samplers
->depth_texture_mask
&= ~(1 << i
);
2826 samplers
->compressed_colortex_mask
&= ~(1 << i
);
2827 si_set_sampler_view(sctx
, shader
, i
, NULL
, NULL
);
2828 si_set_sampler_view(sctx
, shader
, FMASK_TEX_OFFSET
+ i
,
2832 samplers
->n_views
= count
;
2833 sctx
->b
.flags
|= R600_CONTEXT_INV_TEX_CACHE
;
2836 static void si_set_sampler_states(struct si_context
*sctx
,
2837 struct si_pm4_state
*pm4
,
2838 unsigned count
, void **states
,
2839 struct si_textures_info
*samplers
,
2840 unsigned user_data_reg
)
2842 struct si_pipe_sampler_state
**rstates
= (struct si_pipe_sampler_state
**)states
;
2843 uint32_t *border_color_table
= NULL
;
2849 sctx
->b
.flags
|= R600_CONTEXT_INV_TEX_CACHE
;
2851 si_pm4_sh_data_begin(pm4
);
2852 for (i
= 0; i
< count
; i
++) {
2854 G_008F3C_BORDER_COLOR_TYPE(rstates
[i
]->val
[3]) ==
2855 V_008F3C_SQ_TEX_BORDER_COLOR_REGISTER
) {
2856 if (!sctx
->border_color_table
||
2857 ((sctx
->border_color_offset
+ count
- i
) &
2858 C_008F3C_BORDER_COLOR_PTR
)) {
2859 r600_resource_reference(&sctx
->border_color_table
, NULL
);
2860 sctx
->border_color_offset
= 0;
2862 sctx
->border_color_table
=
2863 si_resource_create_custom(&sctx
->screen
->b
.b
,
2868 if (!border_color_table
) {
2869 border_color_table
=
2870 sctx
->b
.ws
->buffer_map(sctx
->border_color_table
->cs_buf
,
2871 sctx
->b
.rings
.gfx
.cs
,
2872 PIPE_TRANSFER_WRITE
|
2873 PIPE_TRANSFER_UNSYNCHRONIZED
);
2876 for (j
= 0; j
< 4; j
++) {
2877 border_color_table
[4 * sctx
->border_color_offset
+ j
] =
2878 util_le32_to_cpu(rstates
[i
]->border_color
[j
]);
2881 rstates
[i
]->val
[3] &= C_008F3C_BORDER_COLOR_PTR
;
2882 rstates
[i
]->val
[3] |= S_008F3C_BORDER_COLOR_PTR(sctx
->border_color_offset
++);
2885 for (j
= 0; j
< Elements(rstates
[i
]->val
); ++j
) {
2886 si_pm4_sh_data_add(pm4
, rstates
[i
] ? rstates
[i
]->val
[j
] : 0);
2889 si_pm4_sh_data_end(pm4
, user_data_reg
, SI_SGPR_SAMPLER
);
2891 if (border_color_table
) {
2892 uint64_t va_offset
=
2893 r600_resource_va(&sctx
->screen
->b
.b
,
2894 (void*)sctx
->border_color_table
);
2896 si_pm4_set_reg(pm4
, R_028080_TA_BC_BASE_ADDR
, va_offset
>> 8);
2897 if (sctx
->b
.chip_class
>= CIK
)
2898 si_pm4_set_reg(pm4
, R_028084_TA_BC_BASE_ADDR_HI
, va_offset
>> 40);
2899 sctx
->b
.ws
->buffer_unmap(sctx
->border_color_table
->cs_buf
);
2900 si_pm4_add_bo(pm4
, sctx
->border_color_table
, RADEON_USAGE_READ
);
2903 memcpy(samplers
->samplers
, states
, sizeof(void*) * count
);
2906 samplers
->n_samplers
= count
;
2909 static void si_bind_vs_sampler_states(struct pipe_context
*ctx
, unsigned count
, void **states
)
2911 struct si_context
*sctx
= (struct si_context
*)ctx
;
2912 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
2914 si_set_sampler_states(sctx
, pm4
, count
, states
,
2915 &sctx
->samplers
[PIPE_SHADER_VERTEX
],
2916 R_00B130_SPI_SHADER_USER_DATA_VS_0
);
2917 #if HAVE_LLVM >= 0x0305
2918 si_set_sampler_states(sctx
, pm4
, count
, states
,
2919 &sctx
->samplers
[PIPE_SHADER_VERTEX
],
2920 R_00B330_SPI_SHADER_USER_DATA_ES_0
);
2922 si_pm4_set_state(sctx
, vs_sampler
, pm4
);
2925 static void si_bind_gs_sampler_states(struct pipe_context
*ctx
, unsigned count
, void **states
)
2927 struct si_context
*sctx
= (struct si_context
*)ctx
;
2928 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
2930 si_set_sampler_states(sctx
, pm4
, count
, states
,
2931 &sctx
->samplers
[PIPE_SHADER_GEOMETRY
],
2932 R_00B230_SPI_SHADER_USER_DATA_GS_0
);
2933 si_pm4_set_state(sctx
, gs_sampler
, pm4
);
2936 static void si_bind_ps_sampler_states(struct pipe_context
*ctx
, unsigned count
, void **states
)
2938 struct si_context
*sctx
= (struct si_context
*)ctx
;
2939 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
2941 si_set_sampler_states(sctx
, pm4
, count
, states
,
2942 &sctx
->samplers
[PIPE_SHADER_FRAGMENT
],
2943 R_00B030_SPI_SHADER_USER_DATA_PS_0
);
2944 si_pm4_set_state(sctx
, ps_sampler
, pm4
);
2948 static void si_bind_sampler_states(struct pipe_context
*ctx
, unsigned shader
,
2949 unsigned start
, unsigned count
,
2955 case PIPE_SHADER_VERTEX
:
2956 si_bind_vs_sampler_states(ctx
, count
, states
);
2958 case PIPE_SHADER_GEOMETRY
:
2959 si_bind_gs_sampler_states(ctx
, count
, states
);
2961 case PIPE_SHADER_FRAGMENT
:
2962 si_bind_ps_sampler_states(ctx
, count
, states
);
2971 static void si_set_sample_mask(struct pipe_context
*ctx
, unsigned sample_mask
)
2973 struct si_context
*sctx
= (struct si_context
*)ctx
;
2974 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
2975 uint16_t mask
= sample_mask
;
2980 si_pm4_set_reg(pm4
, R_028C38_PA_SC_AA_MASK_X0Y0_X1Y0
, mask
| (mask
<< 16));
2981 si_pm4_set_reg(pm4
, R_028C3C_PA_SC_AA_MASK_X0Y1_X1Y1
, mask
| (mask
<< 16));
2983 si_pm4_set_state(sctx
, sample_mask
, pm4
);
2986 static void si_delete_sampler_state(struct pipe_context
*ctx
, void *state
)
2992 * Vertex elements & buffers
2995 static void *si_create_vertex_elements(struct pipe_context
*ctx
,
2997 const struct pipe_vertex_element
*elements
)
2999 struct si_vertex_element
*v
= CALLOC_STRUCT(si_vertex_element
);
3002 assert(count
< PIPE_MAX_ATTRIBS
);
3007 for (i
= 0; i
< count
; ++i
) {
3008 const struct util_format_description
*desc
;
3009 unsigned data_format
, num_format
;
3012 desc
= util_format_description(elements
[i
].src_format
);
3013 first_non_void
= util_format_get_first_non_void_channel(elements
[i
].src_format
);
3014 data_format
= si_translate_buffer_dataformat(ctx
->screen
, desc
, first_non_void
);
3015 num_format
= si_translate_buffer_numformat(ctx
->screen
, desc
, first_non_void
);
3017 v
->rsrc_word3
[i
] = S_008F0C_DST_SEL_X(si_map_swizzle(desc
->swizzle
[0])) |
3018 S_008F0C_DST_SEL_Y(si_map_swizzle(desc
->swizzle
[1])) |
3019 S_008F0C_DST_SEL_Z(si_map_swizzle(desc
->swizzle
[2])) |
3020 S_008F0C_DST_SEL_W(si_map_swizzle(desc
->swizzle
[3])) |
3021 S_008F0C_NUM_FORMAT(num_format
) |
3022 S_008F0C_DATA_FORMAT(data_format
);
3024 memcpy(v
->elements
, elements
, sizeof(struct pipe_vertex_element
) * count
);
3029 static void si_bind_vertex_elements(struct pipe_context
*ctx
, void *state
)
3031 struct si_context
*sctx
= (struct si_context
*)ctx
;
3032 struct si_vertex_element
*v
= (struct si_vertex_element
*)state
;
3034 sctx
->vertex_elements
= v
;
3037 static void si_delete_vertex_element(struct pipe_context
*ctx
, void *state
)
3039 struct si_context
*sctx
= (struct si_context
*)ctx
;
3041 if (sctx
->vertex_elements
== state
)
3042 sctx
->vertex_elements
= NULL
;
3046 static void si_set_vertex_buffers(struct pipe_context
*ctx
, unsigned start_slot
, unsigned count
,
3047 const struct pipe_vertex_buffer
*buffers
)
3049 struct si_context
*sctx
= (struct si_context
*)ctx
;
3051 util_set_vertex_buffers_count(sctx
->vertex_buffer
, &sctx
->nr_vertex_buffers
, buffers
, start_slot
, count
);
3054 static void si_set_index_buffer(struct pipe_context
*ctx
,
3055 const struct pipe_index_buffer
*ib
)
3057 struct si_context
*sctx
= (struct si_context
*)ctx
;
3060 pipe_resource_reference(&sctx
->index_buffer
.buffer
, ib
->buffer
);
3061 memcpy(&sctx
->index_buffer
, ib
, sizeof(*ib
));
3063 pipe_resource_reference(&sctx
->index_buffer
.buffer
, NULL
);
3070 static void si_set_polygon_stipple(struct pipe_context
*ctx
,
3071 const struct pipe_poly_stipple
*state
)
3075 static void si_texture_barrier(struct pipe_context
*ctx
)
3077 struct si_context
*sctx
= (struct si_context
*)ctx
;
3079 sctx
->b
.flags
|= R600_CONTEXT_INV_TEX_CACHE
|
3080 R600_CONTEXT_FLUSH_AND_INV_CB
;
3083 static void *si_create_blend_custom(struct si_context
*sctx
, unsigned mode
)
3085 struct pipe_blend_state blend
;
3087 memset(&blend
, 0, sizeof(blend
));
3088 blend
.independent_blend_enable
= true;
3089 blend
.rt
[0].colormask
= 0xf;
3090 return si_create_blend_state_mode(&sctx
->b
.b
, &blend
, mode
);
3093 static struct pipe_surface
*r600_create_surface(struct pipe_context
*pipe
,
3094 struct pipe_resource
*texture
,
3095 const struct pipe_surface
*surf_tmpl
)
3097 struct r600_texture
*rtex
= (struct r600_texture
*)texture
;
3098 struct r600_surface
*surface
= CALLOC_STRUCT(r600_surface
);
3099 unsigned level
= surf_tmpl
->u
.tex
.level
;
3101 if (surface
== NULL
)
3104 assert(surf_tmpl
->u
.tex
.first_layer
<= util_max_layer(texture
, surf_tmpl
->u
.tex
.level
));
3105 assert(surf_tmpl
->u
.tex
.last_layer
<= util_max_layer(texture
, surf_tmpl
->u
.tex
.level
));
3107 pipe_reference_init(&surface
->base
.reference
, 1);
3108 pipe_resource_reference(&surface
->base
.texture
, texture
);
3109 surface
->base
.context
= pipe
;
3110 surface
->base
.format
= surf_tmpl
->format
;
3111 surface
->base
.width
= rtex
->surface
.level
[level
].npix_x
;
3112 surface
->base
.height
= rtex
->surface
.level
[level
].npix_y
;
3113 surface
->base
.texture
= texture
;
3114 surface
->base
.u
.tex
.first_layer
= surf_tmpl
->u
.tex
.first_layer
;
3115 surface
->base
.u
.tex
.last_layer
= surf_tmpl
->u
.tex
.last_layer
;
3116 surface
->base
.u
.tex
.level
= level
;
3118 return &surface
->base
;
3121 static void r600_surface_destroy(struct pipe_context
*pipe
,
3122 struct pipe_surface
*surface
)
3124 pipe_resource_reference(&surface
->texture
, NULL
);
3128 static boolean
si_dma_copy(struct pipe_context
*ctx
,
3129 struct pipe_resource
*dst
,
3131 unsigned dst_x
, unsigned dst_y
, unsigned dst_z
,
3132 struct pipe_resource
*src
,
3134 const struct pipe_box
*src_box
)
3136 /* XXX implement this or share evergreen_dma_blit with r600g */
3140 static void si_set_occlusion_query_state(struct pipe_context
*ctx
, bool enable
)
3142 /* XXX Turn this into a proper state. Right now the queries are
3143 * enabled in draw_vbo, which snoops r600_common_context to see
3144 * if any occlusion queries are active. */
3147 static void si_need_gfx_cs_space(struct pipe_context
*ctx
, unsigned num_dw
,
3148 bool include_draw_vbo
)
3150 si_need_cs_space((struct si_context
*)ctx
, num_dw
, include_draw_vbo
);
3153 void si_init_state_functions(struct si_context
*sctx
)
3157 sctx
->b
.b
.create_blend_state
= si_create_blend_state
;
3158 sctx
->b
.b
.bind_blend_state
= si_bind_blend_state
;
3159 sctx
->b
.b
.delete_blend_state
= si_delete_blend_state
;
3160 sctx
->b
.b
.set_blend_color
= si_set_blend_color
;
3162 sctx
->b
.b
.create_rasterizer_state
= si_create_rs_state
;
3163 sctx
->b
.b
.bind_rasterizer_state
= si_bind_rs_state
;
3164 sctx
->b
.b
.delete_rasterizer_state
= si_delete_rs_state
;
3166 sctx
->b
.b
.create_depth_stencil_alpha_state
= si_create_dsa_state
;
3167 sctx
->b
.b
.bind_depth_stencil_alpha_state
= si_bind_dsa_state
;
3168 sctx
->b
.b
.delete_depth_stencil_alpha_state
= si_delete_dsa_state
;
3170 for (i
= 0; i
< 8; i
++) {
3171 sctx
->custom_dsa_flush_depth_stencil
[i
] = si_create_db_flush_dsa(sctx
, true, true, i
);
3172 sctx
->custom_dsa_flush_depth
[i
] = si_create_db_flush_dsa(sctx
, true, false, i
);
3173 sctx
->custom_dsa_flush_stencil
[i
] = si_create_db_flush_dsa(sctx
, false, true, i
);
3175 sctx
->custom_dsa_flush_inplace
= si_create_db_flush_dsa(sctx
, false, false, 0);
3176 sctx
->custom_blend_resolve
= si_create_blend_custom(sctx
, V_028808_CB_RESOLVE
);
3177 sctx
->custom_blend_decompress
= si_create_blend_custom(sctx
, V_028808_CB_FMASK_DECOMPRESS
);
3179 sctx
->b
.b
.set_clip_state
= si_set_clip_state
;
3180 sctx
->b
.b
.set_scissor_states
= si_set_scissor_states
;
3181 sctx
->b
.b
.set_viewport_states
= si_set_viewport_states
;
3182 sctx
->b
.b
.set_stencil_ref
= si_set_pipe_stencil_ref
;
3184 sctx
->b
.b
.set_framebuffer_state
= si_set_framebuffer_state
;
3185 sctx
->b
.b
.get_sample_position
= si_get_sample_position
;
3187 sctx
->b
.b
.create_vs_state
= si_create_vs_state
;
3188 sctx
->b
.b
.create_fs_state
= si_create_fs_state
;
3189 sctx
->b
.b
.bind_vs_state
= si_bind_vs_shader
;
3190 sctx
->b
.b
.bind_fs_state
= si_bind_ps_shader
;
3191 sctx
->b
.b
.delete_vs_state
= si_delete_vs_shader
;
3192 sctx
->b
.b
.delete_fs_state
= si_delete_ps_shader
;
3193 #if HAVE_LLVM >= 0x0305
3194 sctx
->b
.b
.create_gs_state
= si_create_gs_state
;
3195 sctx
->b
.b
.bind_gs_state
= si_bind_gs_shader
;
3196 sctx
->b
.b
.delete_gs_state
= si_delete_gs_shader
;
3199 sctx
->b
.b
.create_sampler_state
= si_create_sampler_state
;
3200 sctx
->b
.b
.bind_sampler_states
= si_bind_sampler_states
;
3201 sctx
->b
.b
.delete_sampler_state
= si_delete_sampler_state
;
3203 sctx
->b
.b
.create_sampler_view
= si_create_sampler_view
;
3204 sctx
->b
.b
.set_sampler_views
= si_set_sampler_views
;
3205 sctx
->b
.b
.sampler_view_destroy
= si_sampler_view_destroy
;
3207 sctx
->b
.b
.set_sample_mask
= si_set_sample_mask
;
3209 sctx
->b
.b
.create_vertex_elements_state
= si_create_vertex_elements
;
3210 sctx
->b
.b
.bind_vertex_elements_state
= si_bind_vertex_elements
;
3211 sctx
->b
.b
.delete_vertex_elements_state
= si_delete_vertex_element
;
3212 sctx
->b
.b
.set_vertex_buffers
= si_set_vertex_buffers
;
3213 sctx
->b
.b
.set_index_buffer
= si_set_index_buffer
;
3215 sctx
->b
.b
.texture_barrier
= si_texture_barrier
;
3216 sctx
->b
.b
.set_polygon_stipple
= si_set_polygon_stipple
;
3217 sctx
->b
.b
.create_surface
= r600_create_surface
;
3218 sctx
->b
.b
.surface_destroy
= r600_surface_destroy
;
3219 sctx
->b
.dma_copy
= si_dma_copy
;
3220 sctx
->b
.set_occlusion_query_state
= si_set_occlusion_query_state
;
3221 sctx
->b
.need_gfx_cs_space
= si_need_gfx_cs_space
;
3223 sctx
->b
.b
.draw_vbo
= si_draw_vbo
;
3226 void si_init_config(struct si_context
*sctx
)
3228 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
3233 si_cmd_context_control(pm4
);
3235 si_pm4_set_reg(pm4
, R_028A4C_PA_SC_MODE_CNTL_1
, 0x0);
3237 si_pm4_set_reg(pm4
, R_028A10_VGT_OUTPUT_PATH_CNTL
, 0x0);
3238 si_pm4_set_reg(pm4
, R_028A14_VGT_HOS_CNTL
, 0x0);
3239 si_pm4_set_reg(pm4
, R_028A18_VGT_HOS_MAX_TESS_LEVEL
, 0x0);
3240 si_pm4_set_reg(pm4
, R_028A1C_VGT_HOS_MIN_TESS_LEVEL
, 0x0);
3241 si_pm4_set_reg(pm4
, R_028A20_VGT_HOS_REUSE_DEPTH
, 0x0);
3242 si_pm4_set_reg(pm4
, R_028A24_VGT_GROUP_PRIM_TYPE
, 0x0);
3243 si_pm4_set_reg(pm4
, R_028A28_VGT_GROUP_FIRST_DECR
, 0x0);
3244 si_pm4_set_reg(pm4
, R_028A2C_VGT_GROUP_DECR
, 0x0);
3245 si_pm4_set_reg(pm4
, R_028A30_VGT_GROUP_VECT_0_CNTL
, 0x0);
3246 si_pm4_set_reg(pm4
, R_028A34_VGT_GROUP_VECT_1_CNTL
, 0x0);
3247 si_pm4_set_reg(pm4
, R_028A38_VGT_GROUP_VECT_0_FMT_CNTL
, 0x0);
3248 si_pm4_set_reg(pm4
, R_028A3C_VGT_GROUP_VECT_1_FMT_CNTL
, 0x0);
3250 /* FIXME calculate these values somehow ??? */
3251 si_pm4_set_reg(pm4
, R_028A54_VGT_GS_PER_ES
, 0x80);
3252 si_pm4_set_reg(pm4
, R_028A58_VGT_ES_PER_GS
, 0x40);
3253 si_pm4_set_reg(pm4
, R_028A5C_VGT_GS_PER_VS
, 0x2);
3255 si_pm4_set_reg(pm4
, R_028A84_VGT_PRIMITIVEID_EN
, 0x0);
3256 si_pm4_set_reg(pm4
, R_028A8C_VGT_PRIMITIVEID_RESET
, 0x0);
3257 si_pm4_set_reg(pm4
, R_028AB8_VGT_VTX_CNT_EN
, 0);
3258 si_pm4_set_reg(pm4
, R_028B28_VGT_STRMOUT_DRAW_OPAQUE_OFFSET
, 0);
3260 si_pm4_set_reg(pm4
, R_028B60_VGT_GS_VERT_ITEMSIZE_1
, 0);
3261 si_pm4_set_reg(pm4
, R_028B64_VGT_GS_VERT_ITEMSIZE_2
, 0);
3262 si_pm4_set_reg(pm4
, R_028B68_VGT_GS_VERT_ITEMSIZE_3
, 0);
3263 si_pm4_set_reg(pm4
, R_028B90_VGT_GS_INSTANCE_CNT
, 0);
3265 si_pm4_set_reg(pm4
, R_028B94_VGT_STRMOUT_CONFIG
, 0x0);
3266 si_pm4_set_reg(pm4
, R_028B98_VGT_STRMOUT_BUFFER_CONFIG
, 0x0);
3267 if (sctx
->b
.chip_class
== SI
) {
3268 si_pm4_set_reg(pm4
, R_028AA8_IA_MULTI_VGT_PARAM
,
3269 S_028AA8_SWITCH_ON_EOP(1) |
3270 S_028AA8_PARTIAL_VS_WAVE_ON(1) |
3271 S_028AA8_PRIMGROUP_SIZE(63));
3273 si_pm4_set_reg(pm4
, R_028AB4_VGT_REUSE_OFF
, 0x00000000);
3274 si_pm4_set_reg(pm4
, R_028AB8_VGT_VTX_CNT_EN
, 0x0);
3275 if (sctx
->b
.chip_class
< CIK
)
3276 si_pm4_set_reg(pm4
, R_008A14_PA_CL_ENHANCE
, S_008A14_NUM_CLIP_SEQ(3) |
3277 S_008A14_CLIP_VTX_REORDER_ENA(1));
3279 si_pm4_set_reg(pm4
, R_028BD4_PA_SC_CENTROID_PRIORITY_0
, 0x76543210);
3280 si_pm4_set_reg(pm4
, R_028BD8_PA_SC_CENTROID_PRIORITY_1
, 0xfedcba98);
3282 si_pm4_set_reg(pm4
, R_02882C_PA_SU_PRIM_FILTER_CNTL
, 0);
3284 if (sctx
->b
.chip_class
>= CIK
) {
3285 switch (sctx
->screen
->b
.family
) {
3287 si_pm4_set_reg(pm4
, R_028350_PA_SC_RASTER_CONFIG
, 0x16000012);
3288 si_pm4_set_reg(pm4
, R_028354_PA_SC_RASTER_CONFIG_1
, 0x00000000);
3291 si_pm4_set_reg(pm4
, R_028350_PA_SC_RASTER_CONFIG
, 0x3a00161a);
3292 si_pm4_set_reg(pm4
, R_028354_PA_SC_RASTER_CONFIG_1
, 0x0000002e);
3299 si_pm4_set_reg(pm4
, R_028350_PA_SC_RASTER_CONFIG
, 0x00000000);
3300 si_pm4_set_reg(pm4
, R_028354_PA_SC_RASTER_CONFIG_1
, 0x00000000);
3304 switch (sctx
->screen
->b
.family
) {
3307 si_pm4_set_reg(pm4
, R_028350_PA_SC_RASTER_CONFIG
, 0x2a00126a);
3310 si_pm4_set_reg(pm4
, R_028350_PA_SC_RASTER_CONFIG
, 0x0000124a);
3313 si_pm4_set_reg(pm4
, R_028350_PA_SC_RASTER_CONFIG
, 0x00000082);
3316 si_pm4_set_reg(pm4
, R_028350_PA_SC_RASTER_CONFIG
, 0x00000000);
3319 si_pm4_set_reg(pm4
, R_028350_PA_SC_RASTER_CONFIG
, 0x00000000);
3324 si_pm4_set_reg(pm4
, R_028204_PA_SC_WINDOW_SCISSOR_TL
, S_028204_WINDOW_OFFSET_DISABLE(1));
3325 si_pm4_set_reg(pm4
, R_028240_PA_SC_GENERIC_SCISSOR_TL
, S_028240_WINDOW_OFFSET_DISABLE(1));
3326 si_pm4_set_reg(pm4
, R_028244_PA_SC_GENERIC_SCISSOR_BR
,
3327 S_028244_BR_X(16384) | S_028244_BR_Y(16384));
3328 si_pm4_set_reg(pm4
, R_028030_PA_SC_SCREEN_SCISSOR_TL
, 0);
3329 si_pm4_set_reg(pm4
, R_028034_PA_SC_SCREEN_SCISSOR_BR
,
3330 S_028034_BR_X(16384) | S_028034_BR_Y(16384));
3332 si_pm4_set_reg(pm4
, R_02820C_PA_SC_CLIPRECT_RULE
, 0xFFFF);
3333 si_pm4_set_reg(pm4
, R_028230_PA_SC_EDGERULE
, 0xAAAAAAAA);
3334 si_pm4_set_reg(pm4
, R_0282D0_PA_SC_VPORT_ZMIN_0
, 0x00000000);
3335 si_pm4_set_reg(pm4
, R_0282D4_PA_SC_VPORT_ZMAX_0
, 0x3F800000);
3336 si_pm4_set_reg(pm4
, R_028818_PA_CL_VTE_CNTL
, 0x0000043F);
3337 si_pm4_set_reg(pm4
, R_028820_PA_CL_NANINF_CNTL
, 0x00000000);
3338 si_pm4_set_reg(pm4
, R_028BE8_PA_CL_GB_VERT_CLIP_ADJ
, 0x3F800000);
3339 si_pm4_set_reg(pm4
, R_028BEC_PA_CL_GB_VERT_DISC_ADJ
, 0x3F800000);
3340 si_pm4_set_reg(pm4
, R_028BF0_PA_CL_GB_HORZ_CLIP_ADJ
, 0x3F800000);
3341 si_pm4_set_reg(pm4
, R_028BF4_PA_CL_GB_HORZ_DISC_ADJ
, 0x3F800000);
3342 si_pm4_set_reg(pm4
, R_028020_DB_DEPTH_BOUNDS_MIN
, 0x00000000);
3343 si_pm4_set_reg(pm4
, R_028024_DB_DEPTH_BOUNDS_MAX
, 0x00000000);
3344 si_pm4_set_reg(pm4
, R_028028_DB_STENCIL_CLEAR
, 0x00000000);
3345 si_pm4_set_reg(pm4
, R_02802C_DB_DEPTH_CLEAR
, 0x3F800000);
3346 si_pm4_set_reg(pm4
, R_028AC0_DB_SRESULTS_COMPARE_STATE0
, 0x0);
3347 si_pm4_set_reg(pm4
, R_028AC4_DB_SRESULTS_COMPARE_STATE1
, 0x0);
3348 si_pm4_set_reg(pm4
, R_028AC8_DB_PRELOAD_CONTROL
, 0x0);
3349 si_pm4_set_reg(pm4
, R_02800C_DB_RENDER_OVERRIDE
,
3350 S_02800C_FORCE_HIS_ENABLE0(V_02800C_FORCE_DISABLE
) |
3351 S_02800C_FORCE_HIS_ENABLE1(V_02800C_FORCE_DISABLE
));
3352 si_pm4_set_reg(pm4
, R_028400_VGT_MAX_VTX_INDX
, ~0);
3353 si_pm4_set_reg(pm4
, R_028404_VGT_MIN_VTX_INDX
, 0);
3355 if (sctx
->b
.chip_class
>= CIK
) {
3356 si_pm4_set_reg(pm4
, R_00B118_SPI_SHADER_PGM_RSRC3_VS
, S_00B118_CU_EN(0xffff));
3357 si_pm4_set_reg(pm4
, R_00B11C_SPI_SHADER_LATE_ALLOC_VS
, S_00B11C_LIMIT(0));
3358 si_pm4_set_reg(pm4
, R_00B01C_SPI_SHADER_PGM_RSRC3_PS
, S_00B01C_CU_EN(0xffff));
3361 si_pm4_set_state(sctx
, init
, pm4
);