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"
29 #include "radeon/r600_cs.h"
32 #include "util/u_format.h"
33 #include "util/u_index_modify.h"
34 #include "util/u_memory.h"
35 #include "util/u_prim.h"
36 #include "util/u_upload_mgr.h"
42 static void si_shader_es(struct si_shader
*shader
)
44 struct si_pm4_state
*pm4
;
45 unsigned num_sgprs
, num_user_sgprs
;
46 unsigned vgpr_comp_cnt
;
49 pm4
= shader
->pm4
= CALLOC_STRUCT(si_pm4_state
);
54 va
= shader
->bo
->gpu_address
;
55 si_pm4_add_bo(pm4
, shader
->bo
, RADEON_USAGE_READ
, RADEON_PRIO_SHADER_DATA
);
57 vgpr_comp_cnt
= shader
->uses_instanceid
? 3 : 0;
59 num_user_sgprs
= SI_VS_NUM_USER_SGPR
;
60 num_sgprs
= shader
->num_sgprs
;
61 /* One SGPR after user SGPRs is pre-loaded with es2gs_offset */
62 if ((num_user_sgprs
+ 1) > num_sgprs
) {
63 /* Last 2 reserved SGPRs are used for VCC */
64 num_sgprs
= num_user_sgprs
+ 1 + 2;
66 assert(num_sgprs
<= 104);
68 si_pm4_set_reg(pm4
, R_00B320_SPI_SHADER_PGM_LO_ES
, va
>> 8);
69 si_pm4_set_reg(pm4
, R_00B324_SPI_SHADER_PGM_HI_ES
, va
>> 40);
70 si_pm4_set_reg(pm4
, R_00B328_SPI_SHADER_PGM_RSRC1_ES
,
71 S_00B328_VGPRS((shader
->num_vgprs
- 1) / 4) |
72 S_00B328_SGPRS((num_sgprs
- 1) / 8) |
73 S_00B328_VGPR_COMP_CNT(vgpr_comp_cnt
));
74 si_pm4_set_reg(pm4
, R_00B32C_SPI_SHADER_PGM_RSRC2_ES
,
75 S_00B32C_USER_SGPR(num_user_sgprs
));
78 static void si_shader_gs(struct si_shader
*shader
)
80 unsigned gs_vert_itemsize
= shader
->selector
->info
.num_outputs
* (16 >> 2);
81 unsigned gs_max_vert_out
= shader
->selector
->gs_max_out_vertices
;
82 unsigned gsvs_itemsize
= gs_vert_itemsize
* gs_max_vert_out
;
84 struct si_pm4_state
*pm4
;
85 unsigned num_sgprs
, num_user_sgprs
;
88 /* The GSVS_RING_ITEMSIZE register takes 15 bits */
89 assert(gsvs_itemsize
< (1 << 15));
91 pm4
= shader
->pm4
= CALLOC_STRUCT(si_pm4_state
);
96 if (gs_max_vert_out
<= 128) {
97 cut_mode
= V_028A40_GS_CUT_128
;
98 } else if (gs_max_vert_out
<= 256) {
99 cut_mode
= V_028A40_GS_CUT_256
;
100 } else if (gs_max_vert_out
<= 512) {
101 cut_mode
= V_028A40_GS_CUT_512
;
103 assert(gs_max_vert_out
<= 1024);
104 cut_mode
= V_028A40_GS_CUT_1024
;
107 si_pm4_set_reg(pm4
, R_028A40_VGT_GS_MODE
,
108 S_028A40_MODE(V_028A40_GS_SCENARIO_G
) |
109 S_028A40_CUT_MODE(cut_mode
)|
110 S_028A40_ES_WRITE_OPTIMIZE(1) |
111 S_028A40_GS_WRITE_OPTIMIZE(1));
113 si_pm4_set_reg(pm4
, R_028A60_VGT_GSVS_RING_OFFSET_1
, gsvs_itemsize
);
114 si_pm4_set_reg(pm4
, R_028A64_VGT_GSVS_RING_OFFSET_2
, gsvs_itemsize
);
115 si_pm4_set_reg(pm4
, R_028A68_VGT_GSVS_RING_OFFSET_3
, gsvs_itemsize
);
117 si_pm4_set_reg(pm4
, R_028AAC_VGT_ESGS_RING_ITEMSIZE
,
118 util_bitcount64(shader
->selector
->gs_used_inputs
) * (16 >> 2));
119 si_pm4_set_reg(pm4
, R_028AB0_VGT_GSVS_RING_ITEMSIZE
, gsvs_itemsize
);
121 si_pm4_set_reg(pm4
, R_028B38_VGT_GS_MAX_VERT_OUT
, gs_max_vert_out
);
123 si_pm4_set_reg(pm4
, R_028B5C_VGT_GS_VERT_ITEMSIZE
, gs_vert_itemsize
);
125 va
= shader
->bo
->gpu_address
;
126 si_pm4_add_bo(pm4
, shader
->bo
, RADEON_USAGE_READ
, RADEON_PRIO_SHADER_DATA
);
127 si_pm4_set_reg(pm4
, R_00B220_SPI_SHADER_PGM_LO_GS
, va
>> 8);
128 si_pm4_set_reg(pm4
, R_00B224_SPI_SHADER_PGM_HI_GS
, va
>> 40);
130 num_user_sgprs
= SI_GS_NUM_USER_SGPR
;
131 num_sgprs
= shader
->num_sgprs
;
132 /* Two SGPRs after user SGPRs are pre-loaded with gs2vs_offset, gs_wave_id */
133 if ((num_user_sgprs
+ 2) > num_sgprs
) {
134 /* Last 2 reserved SGPRs are used for VCC */
135 num_sgprs
= num_user_sgprs
+ 2 + 2;
137 assert(num_sgprs
<= 104);
139 si_pm4_set_reg(pm4
, R_00B228_SPI_SHADER_PGM_RSRC1_GS
,
140 S_00B228_VGPRS((shader
->num_vgprs
- 1) / 4) |
141 S_00B228_SGPRS((num_sgprs
- 1) / 8));
142 si_pm4_set_reg(pm4
, R_00B22C_SPI_SHADER_PGM_RSRC2_GS
,
143 S_00B22C_USER_SGPR(num_user_sgprs
));
146 static void si_shader_vs(struct si_shader
*shader
)
148 struct tgsi_shader_info
*info
= &shader
->selector
->info
;
149 struct si_pm4_state
*pm4
;
150 unsigned num_sgprs
, num_user_sgprs
;
151 unsigned nparams
, i
, vgpr_comp_cnt
;
154 pm4
= shader
->pm4
= CALLOC_STRUCT(si_pm4_state
);
159 va
= shader
->bo
->gpu_address
;
160 si_pm4_add_bo(pm4
, shader
->bo
, RADEON_USAGE_READ
, RADEON_PRIO_SHADER_DATA
);
162 vgpr_comp_cnt
= shader
->uses_instanceid
? 3 : 0;
164 if (shader
->is_gs_copy_shader
)
165 num_user_sgprs
= SI_GSCOPY_NUM_USER_SGPR
;
167 num_user_sgprs
= SI_VS_NUM_USER_SGPR
;
169 num_sgprs
= shader
->num_sgprs
;
170 if (num_user_sgprs
> num_sgprs
) {
171 /* Last 2 reserved SGPRs are used for VCC */
172 num_sgprs
= num_user_sgprs
+ 2;
174 assert(num_sgprs
<= 104);
176 /* Certain attributes (position, psize, etc.) don't count as params.
177 * VS is required to export at least one param and r600_shader_from_tgsi()
178 * takes care of adding a dummy export.
180 for (nparams
= 0, i
= 0 ; i
< info
->num_outputs
; i
++) {
181 switch (info
->output_semantic_name
[i
]) {
182 case TGSI_SEMANTIC_CLIPVERTEX
:
183 case TGSI_SEMANTIC_POSITION
:
184 case TGSI_SEMANTIC_PSIZE
:
193 si_pm4_set_reg(pm4
, R_0286C4_SPI_VS_OUT_CONFIG
,
194 S_0286C4_VS_EXPORT_COUNT(nparams
- 1));
196 si_pm4_set_reg(pm4
, R_02870C_SPI_SHADER_POS_FORMAT
,
197 S_02870C_POS0_EXPORT_FORMAT(V_02870C_SPI_SHADER_4COMP
) |
198 S_02870C_POS1_EXPORT_FORMAT(shader
->nr_pos_exports
> 1 ?
199 V_02870C_SPI_SHADER_4COMP
:
200 V_02870C_SPI_SHADER_NONE
) |
201 S_02870C_POS2_EXPORT_FORMAT(shader
->nr_pos_exports
> 2 ?
202 V_02870C_SPI_SHADER_4COMP
:
203 V_02870C_SPI_SHADER_NONE
) |
204 S_02870C_POS3_EXPORT_FORMAT(shader
->nr_pos_exports
> 3 ?
205 V_02870C_SPI_SHADER_4COMP
:
206 V_02870C_SPI_SHADER_NONE
));
208 si_pm4_set_reg(pm4
, R_00B120_SPI_SHADER_PGM_LO_VS
, va
>> 8);
209 si_pm4_set_reg(pm4
, R_00B124_SPI_SHADER_PGM_HI_VS
, va
>> 40);
210 si_pm4_set_reg(pm4
, R_00B128_SPI_SHADER_PGM_RSRC1_VS
,
211 S_00B128_VGPRS((shader
->num_vgprs
- 1) / 4) |
212 S_00B128_SGPRS((num_sgprs
- 1) / 8) |
213 S_00B128_VGPR_COMP_CNT(vgpr_comp_cnt
));
214 si_pm4_set_reg(pm4
, R_00B12C_SPI_SHADER_PGM_RSRC2_VS
,
215 S_00B12C_USER_SGPR(num_user_sgprs
) |
216 S_00B12C_SO_BASE0_EN(!!shader
->selector
->so
.stride
[0]) |
217 S_00B12C_SO_BASE1_EN(!!shader
->selector
->so
.stride
[1]) |
218 S_00B12C_SO_BASE2_EN(!!shader
->selector
->so
.stride
[2]) |
219 S_00B12C_SO_BASE3_EN(!!shader
->selector
->so
.stride
[3]) |
220 S_00B12C_SO_EN(!!shader
->selector
->so
.num_outputs
));
223 static void si_shader_ps(struct si_shader
*shader
)
225 struct tgsi_shader_info
*info
= &shader
->selector
->info
;
226 struct si_pm4_state
*pm4
;
227 unsigned i
, spi_ps_in_control
;
228 unsigned num_sgprs
, num_user_sgprs
;
229 unsigned spi_baryc_cntl
= 0, spi_ps_input_ena
;
232 pm4
= shader
->pm4
= CALLOC_STRUCT(si_pm4_state
);
237 for (i
= 0; i
< info
->num_inputs
; i
++) {
238 switch (info
->input_semantic_name
[i
]) {
239 case TGSI_SEMANTIC_POSITION
:
240 if (info
->input_interpolate_loc
[i
] ==
241 TGSI_INTERPOLATE_LOC_CENTROID
) {
242 /* SPI_BARYC_CNTL.POS_FLOAT_LOCATION
244 * 0 -> Position = pixel center (default)
245 * 1 -> Position = pixel centroid
246 * 2 -> Position = iterated sample number XXX:
247 * What does this mean?
249 spi_baryc_cntl
|= S_0286E0_POS_FLOAT_LOCATION(1);
252 case TGSI_SEMANTIC_FACE
:
257 spi_ps_in_control
= S_0286D8_NUM_INTERP(shader
->nparam
) |
258 S_0286D8_BC_OPTIMIZE_DISABLE(1);
260 si_pm4_set_reg(pm4
, R_0286E0_SPI_BARYC_CNTL
, spi_baryc_cntl
);
261 spi_ps_input_ena
= shader
->spi_ps_input_ena
;
262 /* we need to enable at least one of them, otherwise we hang the GPU */
263 assert(G_0286CC_PERSP_SAMPLE_ENA(spi_ps_input_ena
) ||
264 G_0286CC_PERSP_CENTER_ENA(spi_ps_input_ena
) ||
265 G_0286CC_PERSP_CENTROID_ENA(spi_ps_input_ena
) ||
266 G_0286CC_PERSP_PULL_MODEL_ENA(spi_ps_input_ena
) ||
267 G_0286CC_LINEAR_SAMPLE_ENA(spi_ps_input_ena
) ||
268 G_0286CC_LINEAR_CENTER_ENA(spi_ps_input_ena
) ||
269 G_0286CC_LINEAR_CENTROID_ENA(spi_ps_input_ena
) ||
270 G_0286CC_LINE_STIPPLE_TEX_ENA(spi_ps_input_ena
));
272 si_pm4_set_reg(pm4
, R_0286CC_SPI_PS_INPUT_ENA
, spi_ps_input_ena
);
273 si_pm4_set_reg(pm4
, R_0286D0_SPI_PS_INPUT_ADDR
, spi_ps_input_ena
);
274 si_pm4_set_reg(pm4
, R_0286D8_SPI_PS_IN_CONTROL
, spi_ps_in_control
);
276 si_pm4_set_reg(pm4
, R_028710_SPI_SHADER_Z_FORMAT
, shader
->spi_shader_z_format
);
277 si_pm4_set_reg(pm4
, R_028714_SPI_SHADER_COL_FORMAT
,
278 shader
->spi_shader_col_format
);
279 si_pm4_set_reg(pm4
, R_02823C_CB_SHADER_MASK
, shader
->cb_shader_mask
);
281 va
= shader
->bo
->gpu_address
;
282 si_pm4_add_bo(pm4
, shader
->bo
, RADEON_USAGE_READ
, RADEON_PRIO_SHADER_DATA
);
283 si_pm4_set_reg(pm4
, R_00B020_SPI_SHADER_PGM_LO_PS
, va
>> 8);
284 si_pm4_set_reg(pm4
, R_00B024_SPI_SHADER_PGM_HI_PS
, va
>> 40);
286 num_user_sgprs
= SI_PS_NUM_USER_SGPR
;
287 num_sgprs
= shader
->num_sgprs
;
288 /* One SGPR after user SGPRs is pre-loaded with {prim_mask, lds_offset} */
289 if ((num_user_sgprs
+ 1) > num_sgprs
) {
290 /* Last 2 reserved SGPRs are used for VCC */
291 num_sgprs
= num_user_sgprs
+ 1 + 2;
293 assert(num_sgprs
<= 104);
295 si_pm4_set_reg(pm4
, R_00B028_SPI_SHADER_PGM_RSRC1_PS
,
296 S_00B028_VGPRS((shader
->num_vgprs
- 1) / 4) |
297 S_00B028_SGPRS((num_sgprs
- 1) / 8));
298 si_pm4_set_reg(pm4
, R_00B02C_SPI_SHADER_PGM_RSRC2_PS
,
299 S_00B02C_EXTRA_LDS_SIZE(shader
->lds_size
) |
300 S_00B02C_USER_SGPR(num_user_sgprs
));
303 void si_shader_init_pm4_state(struct si_shader
*shader
)
305 switch (shader
->selector
->type
) {
306 case PIPE_SHADER_VERTEX
:
307 if (shader
->key
.vs
.as_es
)
308 si_shader_es(shader
);
310 si_shader_vs(shader
);
312 case PIPE_SHADER_GEOMETRY
:
313 si_shader_gs(shader
);
314 si_shader_vs(shader
->gs_copy_shader
);
316 case PIPE_SHADER_FRAGMENT
:
317 si_shader_ps(shader
);
328 static unsigned si_conv_pipe_prim(unsigned pprim
)
330 static const unsigned prim_conv
[] = {
331 [PIPE_PRIM_POINTS
] = V_008958_DI_PT_POINTLIST
,
332 [PIPE_PRIM_LINES
] = V_008958_DI_PT_LINELIST
,
333 [PIPE_PRIM_LINE_LOOP
] = V_008958_DI_PT_LINELOOP
,
334 [PIPE_PRIM_LINE_STRIP
] = V_008958_DI_PT_LINESTRIP
,
335 [PIPE_PRIM_TRIANGLES
] = V_008958_DI_PT_TRILIST
,
336 [PIPE_PRIM_TRIANGLE_STRIP
] = V_008958_DI_PT_TRISTRIP
,
337 [PIPE_PRIM_TRIANGLE_FAN
] = V_008958_DI_PT_TRIFAN
,
338 [PIPE_PRIM_QUADS
] = V_008958_DI_PT_QUADLIST
,
339 [PIPE_PRIM_QUAD_STRIP
] = V_008958_DI_PT_QUADSTRIP
,
340 [PIPE_PRIM_POLYGON
] = V_008958_DI_PT_POLYGON
,
341 [PIPE_PRIM_LINES_ADJACENCY
] = V_008958_DI_PT_LINELIST_ADJ
,
342 [PIPE_PRIM_LINE_STRIP_ADJACENCY
] = V_008958_DI_PT_LINESTRIP_ADJ
,
343 [PIPE_PRIM_TRIANGLES_ADJACENCY
] = V_008958_DI_PT_TRILIST_ADJ
,
344 [PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY
] = V_008958_DI_PT_TRISTRIP_ADJ
,
345 [R600_PRIM_RECTANGLE_LIST
] = V_008958_DI_PT_RECTLIST
347 unsigned result
= prim_conv
[pprim
];
349 R600_ERR("unsupported primitive type %d\n", pprim
);
354 static unsigned si_conv_prim_to_gs_out(unsigned mode
)
356 static const int prim_conv
[] = {
357 [PIPE_PRIM_POINTS
] = V_028A6C_OUTPRIM_TYPE_POINTLIST
,
358 [PIPE_PRIM_LINES
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
359 [PIPE_PRIM_LINE_LOOP
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
360 [PIPE_PRIM_LINE_STRIP
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
361 [PIPE_PRIM_TRIANGLES
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
362 [PIPE_PRIM_TRIANGLE_STRIP
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
363 [PIPE_PRIM_TRIANGLE_FAN
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
364 [PIPE_PRIM_QUADS
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
365 [PIPE_PRIM_QUAD_STRIP
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
366 [PIPE_PRIM_POLYGON
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
367 [PIPE_PRIM_LINES_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
368 [PIPE_PRIM_LINE_STRIP_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
369 [PIPE_PRIM_TRIANGLES_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
370 [PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
371 [R600_PRIM_RECTANGLE_LIST
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
373 assert(mode
< Elements(prim_conv
));
375 return prim_conv
[mode
];
378 static unsigned si_get_ia_multi_vgt_param(struct si_context
*sctx
,
379 const struct pipe_draw_info
*info
)
381 struct si_state_rasterizer
*rs
= sctx
->queued
.named
.rasterizer
;
382 unsigned prim
= info
->mode
;
383 unsigned primgroup_size
= 128; /* recommended without a GS */
385 /* SWITCH_ON_EOP(0) is always preferable. */
386 bool wd_switch_on_eop
= false;
387 bool ia_switch_on_eop
= false;
388 bool partial_vs_wave
= false;
391 primgroup_size
= 64; /* recommended with a GS */
393 /* This is a hardware requirement. */
394 if ((rs
&& rs
->line_stipple_enable
) ||
395 (sctx
->b
.screen
->debug_flags
& DBG_SWITCH_ON_EOP
)) {
396 ia_switch_on_eop
= true;
397 wd_switch_on_eop
= true;
400 if (sctx
->b
.streamout
.streamout_enabled
||
401 sctx
->b
.streamout
.prims_gen_query_enabled
)
402 partial_vs_wave
= true;
404 if (sctx
->b
.chip_class
>= CIK
) {
405 /* WD_SWITCH_ON_EOP has no effect on GPUs with less than
406 * 4 shader engines. Set 1 to pass the assertion below.
407 * The other cases are hardware requirements. */
408 if (sctx
->b
.screen
->info
.max_se
< 4 ||
409 prim
== PIPE_PRIM_POLYGON
||
410 prim
== PIPE_PRIM_LINE_LOOP
||
411 prim
== PIPE_PRIM_TRIANGLE_FAN
||
412 prim
== PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY
||
413 info
->primitive_restart
)
414 wd_switch_on_eop
= true;
416 /* Hawaii hangs if instancing is enabled and WD_SWITCH_ON_EOP is 0.
417 * We don't know that for indirect drawing, so treat it as
418 * always problematic. */
419 if (sctx
->b
.family
== CHIP_HAWAII
&&
420 (info
->indirect
|| info
->instance_count
> 1))
421 wd_switch_on_eop
= true;
423 /* If the WD switch is false, the IA switch must be false too. */
424 assert(wd_switch_on_eop
|| !ia_switch_on_eop
);
427 return S_028AA8_SWITCH_ON_EOP(ia_switch_on_eop
) |
428 S_028AA8_PARTIAL_VS_WAVE_ON(partial_vs_wave
) |
429 S_028AA8_PRIMGROUP_SIZE(primgroup_size
- 1) |
430 S_028AA8_WD_SWITCH_ON_EOP(sctx
->b
.chip_class
>= CIK
? wd_switch_on_eop
: 0);
433 static bool si_update_draw_info_state(struct si_context
*sctx
,
434 const struct pipe_draw_info
*info
,
435 const struct pipe_index_buffer
*ib
)
437 struct si_pm4_state
*pm4
= CALLOC_STRUCT(si_pm4_state
);
438 struct si_shader
*vs
= si_get_vs_state(sctx
);
439 unsigned prim
= si_conv_pipe_prim(info
->mode
);
440 unsigned gs_out_prim
=
441 si_conv_prim_to_gs_out(sctx
->gs_shader
?
442 sctx
->gs_shader
->gs_output_prim
:
444 unsigned ls_mask
= 0;
445 unsigned ia_multi_vgt_param
= si_get_ia_multi_vgt_param(sctx
, info
);
455 if (sctx
->b
.chip_class
>= CIK
) {
456 si_pm4_set_reg(pm4
, R_028B74_VGT_DISPATCH_DRAW_INDEX
,
457 ib
->index_size
== 4 ? 0xFC000000 : 0xFC00);
459 si_pm4_cmd_begin(pm4
, PKT3_DRAW_PREAMBLE
);
460 si_pm4_cmd_add(pm4
, prim
); /* VGT_PRIMITIVE_TYPE */
461 si_pm4_cmd_add(pm4
, ia_multi_vgt_param
); /* IA_MULTI_VGT_PARAM */
462 si_pm4_cmd_add(pm4
, 0); /* VGT_LS_HS_CONFIG */
463 si_pm4_cmd_end(pm4
, false);
465 si_pm4_set_reg(pm4
, R_008958_VGT_PRIMITIVE_TYPE
, prim
);
466 si_pm4_set_reg(pm4
, R_028AA8_IA_MULTI_VGT_PARAM
, ia_multi_vgt_param
);
469 si_pm4_set_reg(pm4
, R_028A6C_VGT_GS_OUT_PRIM_TYPE
, gs_out_prim
);
470 si_pm4_set_reg(pm4
, R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX
, info
->restart_index
);
471 si_pm4_set_reg(pm4
, R_028A94_VGT_MULTI_PRIM_IB_RESET_EN
, info
->primitive_restart
);
473 if (prim
== V_008958_DI_PT_LINELIST
)
475 else if (prim
== V_008958_DI_PT_LINESTRIP
)
477 si_pm4_set_reg(pm4
, R_028A0C_PA_SC_LINE_STIPPLE
,
478 S_028A0C_AUTO_RESET_CNTL(ls_mask
) |
479 sctx
->pa_sc_line_stipple
);
481 if (info
->mode
== PIPE_PRIM_QUADS
|| info
->mode
== PIPE_PRIM_QUAD_STRIP
|| info
->mode
== PIPE_PRIM_POLYGON
) {
482 si_pm4_set_reg(pm4
, R_028814_PA_SU_SC_MODE_CNTL
,
483 S_028814_PROVOKING_VTX_LAST(1) | sctx
->pa_su_sc_mode_cntl
);
485 si_pm4_set_reg(pm4
, R_028814_PA_SU_SC_MODE_CNTL
, sctx
->pa_su_sc_mode_cntl
);
487 si_pm4_set_reg(pm4
, R_02881C_PA_CL_VS_OUT_CNTL
,
488 S_02881C_USE_VTX_POINT_SIZE(vs
->vs_out_point_size
) |
489 S_02881C_USE_VTX_EDGE_FLAG(vs
->vs_out_edgeflag
) |
490 S_02881C_USE_VTX_RENDER_TARGET_INDX(vs
->vs_out_layer
) |
491 S_02881C_VS_OUT_CCDIST0_VEC_ENA((vs
->clip_dist_write
& 0x0F) != 0) |
492 S_02881C_VS_OUT_CCDIST1_VEC_ENA((vs
->clip_dist_write
& 0xF0) != 0) |
493 S_02881C_VS_OUT_MISC_VEC_ENA(vs
->vs_out_misc_write
) |
494 (sctx
->queued
.named
.rasterizer
->clip_plane_enable
&
495 vs
->clip_dist_write
));
496 si_pm4_set_reg(pm4
, R_028810_PA_CL_CLIP_CNTL
,
497 sctx
->queued
.named
.rasterizer
->pa_cl_clip_cntl
|
498 (vs
->clip_dist_write
? 0 :
499 sctx
->queued
.named
.rasterizer
->clip_plane_enable
& 0x3F));
501 si_pm4_set_state(sctx
, draw_info
, pm4
);
505 static void si_update_spi_map(struct si_context
*sctx
)
507 struct si_shader
*ps
= sctx
->ps_shader
->current
;
508 struct si_shader
*vs
= si_get_vs_state(sctx
);
509 struct tgsi_shader_info
*psinfo
= &ps
->selector
->info
;
510 struct tgsi_shader_info
*vsinfo
= &vs
->selector
->info
;
511 struct si_pm4_state
*pm4
= CALLOC_STRUCT(si_pm4_state
);
514 for (i
= 0; i
< psinfo
->num_inputs
; i
++) {
515 unsigned name
= psinfo
->input_semantic_name
[i
];
516 unsigned index
= psinfo
->input_semantic_index
[i
];
517 unsigned interpolate
= psinfo
->input_interpolate
[i
];
518 unsigned param_offset
= ps
->ps_input_param_offset
[i
];
520 if (name
== TGSI_SEMANTIC_POSITION
)
521 /* Read from preloaded VGPRs, not parameters */
527 if (interpolate
== TGSI_INTERPOLATE_CONSTANT
||
528 (interpolate
== TGSI_INTERPOLATE_COLOR
&&
529 ps
->key
.ps
.flatshade
)) {
530 tmp
|= S_028644_FLAT_SHADE(1);
533 if (name
== TGSI_SEMANTIC_GENERIC
&&
534 sctx
->sprite_coord_enable
& (1 << index
)) {
535 tmp
|= S_028644_PT_SPRITE_TEX(1);
538 for (j
= 0; j
< vsinfo
->num_outputs
; j
++) {
539 if (name
== vsinfo
->output_semantic_name
[j
] &&
540 index
== vsinfo
->output_semantic_index
[j
]) {
541 tmp
|= S_028644_OFFSET(vs
->vs_output_param_offset
[j
]);
546 if (j
== vsinfo
->num_outputs
) {
547 /* No corresponding output found, load defaults into input */
548 tmp
|= S_028644_OFFSET(0x20);
552 R_028644_SPI_PS_INPUT_CNTL_0
+ param_offset
* 4,
555 if (name
== TGSI_SEMANTIC_COLOR
&&
556 ps
->key
.ps
.color_two_side
) {
557 name
= TGSI_SEMANTIC_BCOLOR
;
563 si_pm4_set_state(sctx
, spi
, pm4
);
566 /* Initialize state related to ESGS / GSVS ring buffers */
567 static void si_init_gs_rings(struct si_context
*sctx
)
569 unsigned esgs_ring_size
= 128 * 1024;
570 unsigned gsvs_ring_size
= 64 * 1024 * 1024;
572 assert(!sctx
->gs_rings
);
573 sctx
->gs_rings
= CALLOC_STRUCT(si_pm4_state
);
575 sctx
->esgs_ring
= pipe_buffer_create(sctx
->b
.b
.screen
, PIPE_BIND_CUSTOM
,
576 PIPE_USAGE_DEFAULT
, esgs_ring_size
);
578 sctx
->gsvs_ring
= pipe_buffer_create(sctx
->b
.b
.screen
, PIPE_BIND_CUSTOM
,
579 PIPE_USAGE_DEFAULT
, gsvs_ring_size
);
581 if (sctx
->b
.chip_class
>= CIK
) {
582 si_pm4_set_reg(sctx
->gs_rings
, R_030900_VGT_ESGS_RING_SIZE
,
583 esgs_ring_size
/ 256);
584 si_pm4_set_reg(sctx
->gs_rings
, R_030904_VGT_GSVS_RING_SIZE
,
585 gsvs_ring_size
/ 256);
587 si_pm4_set_reg(sctx
->gs_rings
, R_0088C8_VGT_ESGS_RING_SIZE
,
588 esgs_ring_size
/ 256);
589 si_pm4_set_reg(sctx
->gs_rings
, R_0088CC_VGT_GSVS_RING_SIZE
,
590 gsvs_ring_size
/ 256);
593 si_set_ring_buffer(&sctx
->b
.b
, PIPE_SHADER_VERTEX
, SI_RING_ESGS
,
594 sctx
->esgs_ring
, 0, esgs_ring_size
,
596 si_set_ring_buffer(&sctx
->b
.b
, PIPE_SHADER_GEOMETRY
, SI_RING_ESGS
,
597 sctx
->esgs_ring
, 0, esgs_ring_size
,
599 si_set_ring_buffer(&sctx
->b
.b
, PIPE_SHADER_VERTEX
, SI_RING_GSVS
,
600 sctx
->gsvs_ring
, 0, gsvs_ring_size
,
604 static void si_update_derived_state(struct si_context
*sctx
)
606 struct pipe_context
* ctx
= (struct pipe_context
*)sctx
;
608 if (!sctx
->blitter
->running
) {
609 /* Flush depth textures which need to be flushed. */
610 for (int i
= 0; i
< SI_NUM_SHADERS
; i
++) {
611 if (sctx
->samplers
[i
].depth_texture_mask
) {
612 si_flush_depth_textures(sctx
, &sctx
->samplers
[i
]);
614 if (sctx
->samplers
[i
].compressed_colortex_mask
) {
615 si_decompress_color_textures(sctx
, &sctx
->samplers
[i
]);
620 if (sctx
->gs_shader
) {
621 si_shader_select(ctx
, sctx
->gs_shader
);
622 si_pm4_bind_state(sctx
, gs
, sctx
->gs_shader
->current
->pm4
);
623 si_pm4_bind_state(sctx
, vs
, sctx
->gs_shader
->current
->gs_copy_shader
->pm4
);
625 sctx
->b
.streamout
.stride_in_dw
= sctx
->gs_shader
->so
.stride
;
627 si_shader_select(ctx
, sctx
->vs_shader
);
628 si_pm4_bind_state(sctx
, es
, sctx
->vs_shader
->current
->pm4
);
631 si_init_gs_rings(sctx
);
632 if (sctx
->emitted
.named
.gs_rings
!= sctx
->gs_rings
)
633 sctx
->b
.flags
|= R600_CONTEXT_VGT_FLUSH
;
634 si_pm4_bind_state(sctx
, gs_rings
, sctx
->gs_rings
);
636 si_set_ring_buffer(ctx
, PIPE_SHADER_GEOMETRY
, SI_RING_GSVS
,
638 sctx
->gs_shader
->gs_max_out_vertices
*
639 sctx
->gs_shader
->info
.num_outputs
* 16,
640 64, true, true, 4, 16);
643 sctx
->gs_on
= CALLOC_STRUCT(si_pm4_state
);
645 si_pm4_set_reg(sctx
->gs_on
, R_028B54_VGT_SHADER_STAGES_EN
,
646 S_028B54_ES_EN(V_028B54_ES_STAGE_REAL
) |
648 S_028B54_VS_EN(V_028B54_VS_STAGE_COPY_SHADER
));
650 si_pm4_bind_state(sctx
, gs_onoff
, sctx
->gs_on
);
652 si_shader_select(ctx
, sctx
->vs_shader
);
653 si_pm4_bind_state(sctx
, vs
, sctx
->vs_shader
->current
->pm4
);
655 sctx
->b
.streamout
.stride_in_dw
= sctx
->vs_shader
->so
.stride
;
658 sctx
->gs_off
= CALLOC_STRUCT(si_pm4_state
);
660 si_pm4_set_reg(sctx
->gs_off
, R_028A40_VGT_GS_MODE
, 0);
661 si_pm4_set_reg(sctx
->gs_off
, R_028B54_VGT_SHADER_STAGES_EN
, 0);
663 si_pm4_bind_state(sctx
, gs_onoff
, sctx
->gs_off
);
664 si_pm4_bind_state(sctx
, gs_rings
, NULL
);
665 si_pm4_bind_state(sctx
, gs
, NULL
);
666 si_pm4_bind_state(sctx
, es
, NULL
);
669 si_shader_select(ctx
, sctx
->ps_shader
);
671 if (!sctx
->ps_shader
->current
) {
672 struct si_shader_selector
*sel
;
674 /* use a dummy shader if compiling the shader (variant) failed */
675 si_make_dummy_ps(sctx
);
676 sel
= sctx
->dummy_pixel_shader
;
677 si_shader_select(ctx
, sel
);
678 sctx
->ps_shader
->current
= sel
->current
;
681 si_pm4_bind_state(sctx
, ps
, sctx
->ps_shader
->current
->pm4
);
683 if (si_pm4_state_changed(sctx
, ps
) || si_pm4_state_changed(sctx
, vs
))
684 si_update_spi_map(sctx
);
686 if (sctx
->ps_db_shader_control
!= sctx
->ps_shader
->current
->db_shader_control
) {
687 sctx
->ps_db_shader_control
= sctx
->ps_shader
->current
->db_shader_control
;
688 sctx
->db_render_state
.dirty
= true;
692 static void si_state_draw(struct si_context
*sctx
,
693 const struct pipe_draw_info
*info
,
694 const struct pipe_index_buffer
*ib
)
696 unsigned sh_base_reg
= (sctx
->gs_shader
? R_00B330_SPI_SHADER_USER_DATA_ES_0
:
697 R_00B130_SPI_SHADER_USER_DATA_VS_0
);
698 struct si_pm4_state
*pm4
= CALLOC_STRUCT(si_pm4_state
);
703 if (info
->count_from_stream_output
) {
704 struct r600_so_target
*t
=
705 (struct r600_so_target
*)info
->count_from_stream_output
;
706 uint64_t va
= t
->buf_filled_size
->gpu_address
+
707 t
->buf_filled_size_offset
;
709 si_pm4_set_reg(pm4
, R_028B30_VGT_STRMOUT_DRAW_OPAQUE_VERTEX_STRIDE
,
712 si_pm4_cmd_begin(pm4
, PKT3_COPY_DATA
);
714 COPY_DATA_SRC_SEL(COPY_DATA_MEM
) |
715 COPY_DATA_DST_SEL(COPY_DATA_REG
) |
716 COPY_DATA_WR_CONFIRM
);
717 si_pm4_cmd_add(pm4
, va
); /* src address lo */
718 si_pm4_cmd_add(pm4
, va
>> 32UL); /* src address hi */
719 si_pm4_cmd_add(pm4
, R_028B2C_VGT_STRMOUT_DRAW_OPAQUE_BUFFER_FILLED_SIZE
>> 2);
720 si_pm4_cmd_add(pm4
, 0); /* unused */
721 si_pm4_add_bo(pm4
, t
->buf_filled_size
, RADEON_USAGE_READ
,
723 si_pm4_cmd_end(pm4
, true);
727 si_pm4_cmd_begin(pm4
, PKT3_INDEX_TYPE
);
728 if (ib
->index_size
== 4) {
729 si_pm4_cmd_add(pm4
, V_028A7C_VGT_INDEX_32
| (SI_BIG_ENDIAN
?
730 V_028A7C_VGT_DMA_SWAP_32_BIT
: 0));
732 si_pm4_cmd_add(pm4
, V_028A7C_VGT_INDEX_16
| (SI_BIG_ENDIAN
?
733 V_028A7C_VGT_DMA_SWAP_16_BIT
: 0));
735 si_pm4_cmd_end(pm4
, sctx
->b
.predicate_drawing
);
737 if (!info
->indirect
) {
738 si_pm4_cmd_begin(pm4
, PKT3_NUM_INSTANCES
);
739 si_pm4_cmd_add(pm4
, info
->instance_count
);
740 si_pm4_cmd_end(pm4
, sctx
->b
.predicate_drawing
);
742 si_pm4_set_reg(pm4
, sh_base_reg
+ SI_SGPR_BASE_VERTEX
* 4,
743 info
->indexed
? info
->index_bias
: info
->start
);
744 si_pm4_set_reg(pm4
, sh_base_reg
+ SI_SGPR_START_INSTANCE
* 4,
745 info
->start_instance
);
747 si_pm4_add_bo(pm4
, (struct r600_resource
*)info
->indirect
,
748 RADEON_USAGE_READ
, RADEON_PRIO_MIN
);
752 uint32_t max_size
= (ib
->buffer
->width0
- ib
->offset
) /
754 uint64_t va
= r600_resource(ib
->buffer
)->gpu_address
+ ib
->offset
;
756 si_pm4_add_bo(pm4
, (struct r600_resource
*)ib
->buffer
, RADEON_USAGE_READ
,
759 if (info
->indirect
) {
760 uint64_t indirect_va
= r600_resource(info
->indirect
)->gpu_address
;
761 si_cmd_draw_index_indirect(pm4
, indirect_va
, va
, max_size
,
762 info
->indirect_offset
,
763 sh_base_reg
+ SI_SGPR_BASE_VERTEX
* 4,
764 sh_base_reg
+ SI_SGPR_START_INSTANCE
* 4,
765 sctx
->b
.predicate_drawing
);
767 va
+= info
->start
* ib
->index_size
;
768 si_cmd_draw_index_2(pm4
, max_size
, va
, info
->count
,
769 V_0287F0_DI_SRC_SEL_DMA
,
770 sctx
->b
.predicate_drawing
);
773 if (info
->indirect
) {
774 uint64_t indirect_va
= r600_resource(info
->indirect
)->gpu_address
;
775 si_cmd_draw_indirect(pm4
, indirect_va
, info
->indirect_offset
,
776 sh_base_reg
+ SI_SGPR_BASE_VERTEX
* 4,
777 sh_base_reg
+ SI_SGPR_START_INSTANCE
* 4,
778 sctx
->b
.predicate_drawing
);
780 si_cmd_draw_index_auto(pm4
, info
->count
,
781 V_0287F0_DI_SRC_SEL_AUTO_INDEX
|
782 S_0287F0_USE_OPAQUE(!!info
->count_from_stream_output
),
783 sctx
->b
.predicate_drawing
);
787 si_pm4_set_state(sctx
, draw
, pm4
);
790 void si_emit_cache_flush(struct r600_common_context
*sctx
, struct r600_atom
*atom
)
792 struct radeon_winsys_cs
*cs
= sctx
->rings
.gfx
.cs
;
793 uint32_t cp_coher_cntl
= 0;
795 PKT3_SHADER_TYPE_S(!!(sctx
->flags
& R600_CONTEXT_FLAG_COMPUTE
));
797 /* XXX SI flushes both ICACHE and KCACHE if either flag is set.
798 * XXX CIK shouldn't have this issue. Test CIK before separating the flags
799 * XXX to ensure there is no regression. Also find out if there is another
800 * XXX way to flush either ICACHE or KCACHE but not both for SI. */
801 if (sctx
->flags
& (R600_CONTEXT_INV_SHADER_CACHE
|
802 R600_CONTEXT_INV_CONST_CACHE
)) {
803 cp_coher_cntl
|= S_0085F0_SH_ICACHE_ACTION_ENA(1) |
804 S_0085F0_SH_KCACHE_ACTION_ENA(1);
806 if (sctx
->flags
& (R600_CONTEXT_INV_TEX_CACHE
|
807 R600_CONTEXT_STREAMOUT_FLUSH
)) {
808 cp_coher_cntl
|= S_0085F0_TC_ACTION_ENA(1) |
809 S_0085F0_TCL1_ACTION_ENA(1);
811 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_CB
) {
812 cp_coher_cntl
|= S_0085F0_CB_ACTION_ENA(1) |
813 S_0085F0_CB0_DEST_BASE_ENA(1) |
814 S_0085F0_CB1_DEST_BASE_ENA(1) |
815 S_0085F0_CB2_DEST_BASE_ENA(1) |
816 S_0085F0_CB3_DEST_BASE_ENA(1) |
817 S_0085F0_CB4_DEST_BASE_ENA(1) |
818 S_0085F0_CB5_DEST_BASE_ENA(1) |
819 S_0085F0_CB6_DEST_BASE_ENA(1) |
820 S_0085F0_CB7_DEST_BASE_ENA(1);
822 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_DB
) {
823 cp_coher_cntl
|= S_0085F0_DB_ACTION_ENA(1) |
824 S_0085F0_DB_DEST_BASE_ENA(1);
828 if (sctx
->chip_class
>= CIK
) {
829 radeon_emit(cs
, PKT3(PKT3_ACQUIRE_MEM
, 5, 0) | compute
);
830 radeon_emit(cs
, cp_coher_cntl
); /* CP_COHER_CNTL */
831 radeon_emit(cs
, 0xffffffff); /* CP_COHER_SIZE */
832 radeon_emit(cs
, 0xff); /* CP_COHER_SIZE_HI */
833 radeon_emit(cs
, 0); /* CP_COHER_BASE */
834 radeon_emit(cs
, 0); /* CP_COHER_BASE_HI */
835 radeon_emit(cs
, 0x0000000A); /* POLL_INTERVAL */
837 radeon_emit(cs
, PKT3(PKT3_SURFACE_SYNC
, 3, 0) | compute
);
838 radeon_emit(cs
, cp_coher_cntl
); /* CP_COHER_CNTL */
839 radeon_emit(cs
, 0xffffffff); /* CP_COHER_SIZE */
840 radeon_emit(cs
, 0); /* CP_COHER_BASE */
841 radeon_emit(cs
, 0x0000000A); /* POLL_INTERVAL */
845 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_CB_META
) {
846 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
847 radeon_emit(cs
, EVENT_TYPE(V_028A90_FLUSH_AND_INV_CB_META
) | EVENT_INDEX(0));
849 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_DB_META
) {
850 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
851 radeon_emit(cs
, EVENT_TYPE(V_028A90_FLUSH_AND_INV_DB_META
) | EVENT_INDEX(0));
853 if (sctx
->flags
& R600_CONTEXT_FLUSH_WITH_INV_L2
) {
854 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
855 radeon_emit(cs
, EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH
) | EVENT_INDEX(7) |
859 if (sctx
->flags
& (R600_CONTEXT_WAIT_3D_IDLE
|
860 R600_CONTEXT_PS_PARTIAL_FLUSH
)) {
861 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
862 radeon_emit(cs
, EVENT_TYPE(V_028A90_PS_PARTIAL_FLUSH
) | EVENT_INDEX(4));
863 } else if (sctx
->flags
& R600_CONTEXT_STREAMOUT_FLUSH
) {
864 /* Needed if streamout buffers are going to be used as a source. */
865 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
866 radeon_emit(cs
, EVENT_TYPE(V_028A90_VS_PARTIAL_FLUSH
) | EVENT_INDEX(4));
869 if (sctx
->flags
& R600_CONTEXT_CS_PARTIAL_FLUSH
) {
870 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
871 radeon_emit(cs
, EVENT_TYPE(V_028A90_CS_PARTIAL_FLUSH
| EVENT_INDEX(4)));
874 if (sctx
->flags
& R600_CONTEXT_VGT_FLUSH
) {
875 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
876 radeon_emit(cs
, EVENT_TYPE(V_028A90_VGT_FLUSH
) | EVENT_INDEX(0));
878 if (sctx
->flags
& R600_CONTEXT_VGT_STREAMOUT_SYNC
) {
879 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
880 radeon_emit(cs
, EVENT_TYPE(V_028A90_VGT_STREAMOUT_SYNC
) | EVENT_INDEX(0));
886 const struct r600_atom si_atom_cache_flush
= { si_emit_cache_flush
, 21 }; /* number of CS dwords */
888 static void si_get_draw_start_count(struct si_context
*sctx
,
889 const struct pipe_draw_info
*info
,
890 unsigned *start
, unsigned *count
)
892 if (info
->indirect
) {
893 struct r600_resource
*indirect
=
894 (struct r600_resource
*)info
->indirect
;
895 int *data
= r600_buffer_map_sync_with_rings(&sctx
->b
,
896 indirect
, PIPE_TRANSFER_READ
);
897 data
+= info
->indirect_offset
/sizeof(int);
901 *start
= info
->start
;
902 *count
= info
->count
;
906 void si_draw_vbo(struct pipe_context
*ctx
, const struct pipe_draw_info
*info
)
908 struct si_context
*sctx
= (struct si_context
*)ctx
;
909 struct pipe_index_buffer ib
= {};
912 if (!info
->count
&& !info
->indirect
&&
913 (info
->indexed
|| !info
->count_from_stream_output
))
916 if (!sctx
->ps_shader
|| !sctx
->vs_shader
)
919 si_update_derived_state(sctx
);
921 if (sctx
->vertex_buffers_dirty
) {
922 si_update_vertex_buffers(sctx
);
923 sctx
->vertex_buffers_dirty
= false;
927 /* Initialize the index buffer struct. */
928 pipe_resource_reference(&ib
.buffer
, sctx
->index_buffer
.buffer
);
929 ib
.user_buffer
= sctx
->index_buffer
.user_buffer
;
930 ib
.index_size
= sctx
->index_buffer
.index_size
;
931 ib
.offset
= sctx
->index_buffer
.offset
;
933 /* Translate or upload, if needed. */
934 if (ib
.index_size
== 1) {
935 struct pipe_resource
*out_buffer
= NULL
;
936 unsigned out_offset
, start
, count
, start_offset
;
939 si_get_draw_start_count(sctx
, info
, &start
, &count
);
940 start_offset
= start
* ib
.index_size
;
942 u_upload_alloc(sctx
->b
.uploader
, start_offset
, count
* 2,
943 &out_offset
, &out_buffer
, &ptr
);
945 util_shorten_ubyte_elts_to_userptr(&sctx
->b
.b
, &ib
, 0,
946 ib
.offset
+ start_offset
,
949 pipe_resource_reference(&ib
.buffer
, NULL
);
950 ib
.user_buffer
= NULL
;
951 ib
.buffer
= out_buffer
;
952 /* info->start will be added by the drawing code */
953 ib
.offset
= out_offset
- start_offset
;
955 } else if (ib
.user_buffer
&& !ib
.buffer
) {
956 unsigned start
, count
, start_offset
;
958 si_get_draw_start_count(sctx
, info
, &start
, &count
);
959 start_offset
= start
* ib
.index_size
;
961 u_upload_data(sctx
->b
.uploader
, start_offset
, count
* ib
.index_size
,
962 (char*)ib
.user_buffer
+ start_offset
,
963 &ib
.offset
, &ib
.buffer
);
964 /* info->start will be added by the drawing code */
965 ib
.offset
-= start_offset
;
969 if (!si_update_draw_info_state(sctx
, info
, &ib
))
972 si_state_draw(sctx
, info
, &ib
);
974 sctx
->pm4_dirty_cdwords
+= si_pm4_dirty_dw(sctx
);
976 /* Check flush flags. */
978 sctx
->atoms
.s
.cache_flush
->dirty
= true;
980 si_need_cs_space(sctx
, 0, TRUE
);
983 for (i
= 0; i
< SI_NUM_ATOMS(sctx
); i
++) {
984 if (sctx
->atoms
.array
[i
]->dirty
) {
985 sctx
->atoms
.array
[i
]->emit(&sctx
->b
, sctx
->atoms
.array
[i
]);
986 sctx
->atoms
.array
[i
]->dirty
= false;
990 si_pm4_emit_dirty(sctx
);
991 sctx
->pm4_dirty_cdwords
= 0;
994 if (sctx
->screen
->b
.trace_bo
) {
999 /* Workaround for a VGT hang when streamout is enabled.
1000 * It must be done after drawing. */
1001 if (sctx
->b
.family
== CHIP_HAWAII
&&
1002 (sctx
->b
.streamout
.streamout_enabled
||
1003 sctx
->b
.streamout
.prims_gen_query_enabled
)) {
1004 sctx
->b
.flags
|= R600_CONTEXT_VGT_STREAMOUT_SYNC
;
1007 /* Set the depth buffer as dirty. */
1008 if (sctx
->framebuffer
.state
.zsbuf
) {
1009 struct pipe_surface
*surf
= sctx
->framebuffer
.state
.zsbuf
;
1010 struct r600_texture
*rtex
= (struct r600_texture
*)surf
->texture
;
1012 rtex
->dirty_level_mask
|= 1 << surf
->u
.tex
.level
;
1014 if (sctx
->framebuffer
.compressed_cb_mask
) {
1015 struct pipe_surface
*surf
;
1016 struct r600_texture
*rtex
;
1017 unsigned mask
= sctx
->framebuffer
.compressed_cb_mask
;
1020 unsigned i
= u_bit_scan(&mask
);
1021 surf
= sctx
->framebuffer
.state
.cbufs
[i
];
1022 rtex
= (struct r600_texture
*)surf
->texture
;
1024 rtex
->dirty_level_mask
|= 1 << surf
->u
.tex
.level
;
1028 pipe_resource_reference(&ib
.buffer
, NULL
);
1029 sctx
->b
.num_draw_calls
++;
1033 void si_trace_emit(struct si_context
*sctx
)
1035 struct si_screen
*sscreen
= sctx
->screen
;
1036 struct radeon_winsys_cs
*cs
= sctx
->b
.rings
.gfx
.cs
;
1039 va
= sscreen
->b
.trace_bo
->gpu_address
;
1040 r600_context_bo_reloc(&sctx
->b
, &sctx
->b
.rings
.gfx
, sscreen
->b
.trace_bo
,
1041 RADEON_USAGE_READWRITE
, RADEON_PRIO_MIN
);
1042 radeon_emit(cs
, PKT3(PKT3_WRITE_DATA
, 4, 0));
1043 radeon_emit(cs
, PKT3_WRITE_DATA_DST_SEL(PKT3_WRITE_DATA_DST_SEL_MEM_SYNC
) |
1044 PKT3_WRITE_DATA_WR_CONFIRM
|
1045 PKT3_WRITE_DATA_ENGINE_SEL(PKT3_WRITE_DATA_ENGINE_SEL_ME
));
1046 radeon_emit(cs
, va
& 0xFFFFFFFFUL
);
1047 radeon_emit(cs
, (va
>> 32UL) & 0xFFFFFFFFUL
);
1048 radeon_emit(cs
, cs
->cdw
);
1049 radeon_emit(cs
, sscreen
->b
.cs_count
);