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_pipe_shader_es(struct pipe_context
*ctx
, struct si_pipe_shader
*shader
)
44 struct si_context
*sctx
= (struct si_context
*)ctx
;
45 struct si_pm4_state
*pm4
;
46 unsigned num_sgprs
, num_user_sgprs
;
47 unsigned vgpr_comp_cnt
;
50 si_pm4_delete_state(sctx
, es
, shader
->pm4
);
51 pm4
= shader
->pm4
= si_pm4_alloc_state(sctx
);
56 va
= shader
->bo
->gpu_address
;
57 si_pm4_add_bo(pm4
, shader
->bo
, RADEON_USAGE_READ
, RADEON_PRIO_SHADER_DATA
);
59 vgpr_comp_cnt
= shader
->shader
.uses_instanceid
? 3 : 0;
61 num_user_sgprs
= SI_VS_NUM_USER_SGPR
;
62 num_sgprs
= shader
->num_sgprs
;
63 /* One SGPR after user SGPRs is pre-loaded with es2gs_offset */
64 if ((num_user_sgprs
+ 1) > num_sgprs
) {
65 /* Last 2 reserved SGPRs are used for VCC */
66 num_sgprs
= num_user_sgprs
+ 1 + 2;
68 assert(num_sgprs
<= 104);
70 si_pm4_set_reg(pm4
, R_00B320_SPI_SHADER_PGM_LO_ES
, va
>> 8);
71 si_pm4_set_reg(pm4
, R_00B324_SPI_SHADER_PGM_HI_ES
, va
>> 40);
72 si_pm4_set_reg(pm4
, R_00B328_SPI_SHADER_PGM_RSRC1_ES
,
73 S_00B328_VGPRS((shader
->num_vgprs
- 1) / 4) |
74 S_00B328_SGPRS((num_sgprs
- 1) / 8) |
75 S_00B328_VGPR_COMP_CNT(vgpr_comp_cnt
));
76 si_pm4_set_reg(pm4
, R_00B32C_SPI_SHADER_PGM_RSRC2_ES
,
77 S_00B32C_USER_SGPR(num_user_sgprs
));
79 sctx
->b
.flags
|= R600_CONTEXT_INV_SHADER_CACHE
;
82 static void si_pipe_shader_gs(struct pipe_context
*ctx
, struct si_pipe_shader
*shader
)
84 struct si_context
*sctx
= (struct si_context
*)ctx
;
85 unsigned gs_vert_itemsize
= shader
->shader
.noutput
* (16 >> 2);
86 unsigned gs_max_vert_out
= shader
->shader
.gs_max_out_vertices
;
87 unsigned gsvs_itemsize
= gs_vert_itemsize
* gs_max_vert_out
;
89 struct si_pm4_state
*pm4
;
90 unsigned num_sgprs
, num_user_sgprs
;
93 /* The GSVS_RING_ITEMSIZE register takes 15 bits */
94 assert(gsvs_itemsize
< (1 << 15));
96 si_pm4_delete_state(sctx
, gs
, shader
->pm4
);
97 pm4
= shader
->pm4
= si_pm4_alloc_state(sctx
);
102 if (gs_max_vert_out
<= 128) {
103 cut_mode
= V_028A40_GS_CUT_128
;
104 } else if (gs_max_vert_out
<= 256) {
105 cut_mode
= V_028A40_GS_CUT_256
;
106 } else if (gs_max_vert_out
<= 512) {
107 cut_mode
= V_028A40_GS_CUT_512
;
109 assert(gs_max_vert_out
<= 1024);
110 cut_mode
= V_028A40_GS_CUT_1024
;
113 si_pm4_set_reg(pm4
, R_028A40_VGT_GS_MODE
,
114 S_028A40_MODE(V_028A40_GS_SCENARIO_G
) |
115 S_028A40_CUT_MODE(cut_mode
)|
116 S_028A40_ES_WRITE_OPTIMIZE(1) |
117 S_028A40_GS_WRITE_OPTIMIZE(1));
119 si_pm4_set_reg(pm4
, R_028A60_VGT_GSVS_RING_OFFSET_1
, gsvs_itemsize
);
120 si_pm4_set_reg(pm4
, R_028A64_VGT_GSVS_RING_OFFSET_2
, gsvs_itemsize
);
121 si_pm4_set_reg(pm4
, R_028A68_VGT_GSVS_RING_OFFSET_3
, gsvs_itemsize
);
123 si_pm4_set_reg(pm4
, R_028AAC_VGT_ESGS_RING_ITEMSIZE
,
124 shader
->shader
.nparam
* (16 >> 2));
125 si_pm4_set_reg(pm4
, R_028AB0_VGT_GSVS_RING_ITEMSIZE
, gsvs_itemsize
);
127 si_pm4_set_reg(pm4
, R_028B38_VGT_GS_MAX_VERT_OUT
, gs_max_vert_out
);
129 si_pm4_set_reg(pm4
, R_028B5C_VGT_GS_VERT_ITEMSIZE
, gs_vert_itemsize
);
131 va
= shader
->bo
->gpu_address
;
132 si_pm4_add_bo(pm4
, shader
->bo
, RADEON_USAGE_READ
, RADEON_PRIO_SHADER_DATA
);
133 si_pm4_set_reg(pm4
, R_00B220_SPI_SHADER_PGM_LO_GS
, va
>> 8);
134 si_pm4_set_reg(pm4
, R_00B224_SPI_SHADER_PGM_HI_GS
, va
>> 40);
136 num_user_sgprs
= SI_GS_NUM_USER_SGPR
;
137 num_sgprs
= shader
->num_sgprs
;
138 /* Two SGPRs after user SGPRs are pre-loaded with gs2vs_offset, gs_wave_id */
139 if ((num_user_sgprs
+ 2) > num_sgprs
) {
140 /* Last 2 reserved SGPRs are used for VCC */
141 num_sgprs
= num_user_sgprs
+ 2 + 2;
143 assert(num_sgprs
<= 104);
145 si_pm4_set_reg(pm4
, R_00B228_SPI_SHADER_PGM_RSRC1_GS
,
146 S_00B228_VGPRS((shader
->num_vgprs
- 1) / 4) |
147 S_00B228_SGPRS((num_sgprs
- 1) / 8));
148 si_pm4_set_reg(pm4
, R_00B22C_SPI_SHADER_PGM_RSRC2_GS
,
149 S_00B22C_USER_SGPR(num_user_sgprs
));
151 sctx
->b
.flags
|= R600_CONTEXT_INV_SHADER_CACHE
;
154 static void si_pipe_shader_vs(struct pipe_context
*ctx
, struct si_pipe_shader
*shader
)
156 struct si_context
*sctx
= (struct si_context
*)ctx
;
157 struct si_pm4_state
*pm4
;
158 unsigned num_sgprs
, num_user_sgprs
;
159 unsigned nparams
, i
, vgpr_comp_cnt
;
162 si_pm4_delete_state(sctx
, vs
, shader
->pm4
);
163 pm4
= shader
->pm4
= si_pm4_alloc_state(sctx
);
168 va
= shader
->bo
->gpu_address
;
169 si_pm4_add_bo(pm4
, shader
->bo
, RADEON_USAGE_READ
, RADEON_PRIO_SHADER_DATA
);
171 vgpr_comp_cnt
= shader
->shader
.uses_instanceid
? 3 : 0;
173 num_user_sgprs
= SI_VS_NUM_USER_SGPR
;
174 num_sgprs
= shader
->num_sgprs
;
175 if (num_user_sgprs
> num_sgprs
) {
176 /* Last 2 reserved SGPRs are used for VCC */
177 num_sgprs
= num_user_sgprs
+ 2;
179 assert(num_sgprs
<= 104);
181 /* Certain attributes (position, psize, etc.) don't count as params.
182 * VS is required to export at least one param and r600_shader_from_tgsi()
183 * takes care of adding a dummy export.
185 for (nparams
= 0, i
= 0 ; i
< shader
->shader
.noutput
; i
++) {
186 switch (shader
->shader
.output
[i
].name
) {
187 case TGSI_SEMANTIC_CLIPVERTEX
:
188 case TGSI_SEMANTIC_POSITION
:
189 case TGSI_SEMANTIC_PSIZE
:
198 si_pm4_set_reg(pm4
, R_0286C4_SPI_VS_OUT_CONFIG
,
199 S_0286C4_VS_EXPORT_COUNT(nparams
- 1));
201 si_pm4_set_reg(pm4
, R_02870C_SPI_SHADER_POS_FORMAT
,
202 S_02870C_POS0_EXPORT_FORMAT(V_02870C_SPI_SHADER_4COMP
) |
203 S_02870C_POS1_EXPORT_FORMAT(shader
->shader
.nr_pos_exports
> 1 ?
204 V_02870C_SPI_SHADER_4COMP
:
205 V_02870C_SPI_SHADER_NONE
) |
206 S_02870C_POS2_EXPORT_FORMAT(shader
->shader
.nr_pos_exports
> 2 ?
207 V_02870C_SPI_SHADER_4COMP
:
208 V_02870C_SPI_SHADER_NONE
) |
209 S_02870C_POS3_EXPORT_FORMAT(shader
->shader
.nr_pos_exports
> 3 ?
210 V_02870C_SPI_SHADER_4COMP
:
211 V_02870C_SPI_SHADER_NONE
));
213 si_pm4_set_reg(pm4
, R_00B120_SPI_SHADER_PGM_LO_VS
, va
>> 8);
214 si_pm4_set_reg(pm4
, R_00B124_SPI_SHADER_PGM_HI_VS
, va
>> 40);
215 si_pm4_set_reg(pm4
, R_00B128_SPI_SHADER_PGM_RSRC1_VS
,
216 S_00B128_VGPRS((shader
->num_vgprs
- 1) / 4) |
217 S_00B128_SGPRS((num_sgprs
- 1) / 8) |
218 S_00B128_VGPR_COMP_CNT(vgpr_comp_cnt
));
219 si_pm4_set_reg(pm4
, R_00B12C_SPI_SHADER_PGM_RSRC2_VS
,
220 S_00B12C_USER_SGPR(num_user_sgprs
) |
221 S_00B12C_SO_BASE0_EN(!!shader
->selector
->so
.stride
[0]) |
222 S_00B12C_SO_BASE1_EN(!!shader
->selector
->so
.stride
[1]) |
223 S_00B12C_SO_BASE2_EN(!!shader
->selector
->so
.stride
[2]) |
224 S_00B12C_SO_BASE3_EN(!!shader
->selector
->so
.stride
[3]) |
225 S_00B12C_SO_EN(!!shader
->selector
->so
.num_outputs
));
227 sctx
->b
.flags
|= R600_CONTEXT_INV_SHADER_CACHE
;
230 static void si_pipe_shader_ps(struct pipe_context
*ctx
, struct si_pipe_shader
*shader
)
232 struct si_context
*sctx
= (struct si_context
*)ctx
;
233 struct si_pm4_state
*pm4
;
234 unsigned i
, spi_ps_in_control
, db_shader_control
;
235 unsigned num_sgprs
, num_user_sgprs
;
236 unsigned spi_baryc_cntl
= 0, spi_ps_input_ena
;
239 si_pm4_delete_state(sctx
, ps
, shader
->pm4
);
240 pm4
= shader
->pm4
= si_pm4_alloc_state(sctx
);
245 db_shader_control
= S_02880C_Z_ORDER(V_02880C_EARLY_Z_THEN_LATE_Z
) |
246 S_02880C_ALPHA_TO_MASK_DISABLE(sctx
->framebuffer
.cb0_is_integer
);
248 for (i
= 0; i
< shader
->shader
.ninput
; i
++) {
249 switch (shader
->shader
.input
[i
].name
) {
250 case TGSI_SEMANTIC_POSITION
:
251 if (shader
->shader
.input
[i
].centroid
) {
252 /* SPI_BARYC_CNTL.POS_FLOAT_LOCATION
254 * 0 -> Position = pixel center (default)
255 * 1 -> Position = pixel centroid
256 * 2 -> Position = iterated sample number XXX:
257 * What does this mean?
259 spi_baryc_cntl
|= S_0286E0_POS_FLOAT_LOCATION(1);
262 case TGSI_SEMANTIC_FACE
:
267 db_shader_control
|= shader
->db_shader_control
;
269 if (shader
->shader
.uses_kill
|| shader
->key
.ps
.alpha_func
!= PIPE_FUNC_ALWAYS
)
270 db_shader_control
|= S_02880C_KILL_ENABLE(1);
272 if (sctx
->b
.chip_class
>= CIK
)
274 S_02880C_CONSERVATIVE_Z_EXPORT(shader
->shader
.ps_conservative_z
);
276 spi_ps_in_control
= S_0286D8_NUM_INTERP(shader
->shader
.nparam
) |
277 S_0286D8_BC_OPTIMIZE_DISABLE(1);
279 si_pm4_set_reg(pm4
, R_0286E0_SPI_BARYC_CNTL
, spi_baryc_cntl
);
280 spi_ps_input_ena
= shader
->spi_ps_input_ena
;
281 /* we need to enable at least one of them, otherwise we hang the GPU */
282 assert(G_0286CC_PERSP_SAMPLE_ENA(spi_ps_input_ena
) ||
283 G_0286CC_PERSP_CENTER_ENA(spi_ps_input_ena
) ||
284 G_0286CC_PERSP_CENTROID_ENA(spi_ps_input_ena
) ||
285 G_0286CC_PERSP_PULL_MODEL_ENA(spi_ps_input_ena
) ||
286 G_0286CC_LINEAR_SAMPLE_ENA(spi_ps_input_ena
) ||
287 G_0286CC_LINEAR_CENTER_ENA(spi_ps_input_ena
) ||
288 G_0286CC_LINEAR_CENTROID_ENA(spi_ps_input_ena
) ||
289 G_0286CC_LINE_STIPPLE_TEX_ENA(spi_ps_input_ena
));
291 si_pm4_set_reg(pm4
, R_0286CC_SPI_PS_INPUT_ENA
, spi_ps_input_ena
);
292 si_pm4_set_reg(pm4
, R_0286D0_SPI_PS_INPUT_ADDR
, spi_ps_input_ena
);
293 si_pm4_set_reg(pm4
, R_0286D8_SPI_PS_IN_CONTROL
, spi_ps_in_control
);
295 si_pm4_set_reg(pm4
, R_028710_SPI_SHADER_Z_FORMAT
, shader
->spi_shader_z_format
);
296 si_pm4_set_reg(pm4
, R_028714_SPI_SHADER_COL_FORMAT
,
297 shader
->spi_shader_col_format
);
298 si_pm4_set_reg(pm4
, R_02823C_CB_SHADER_MASK
, shader
->cb_shader_mask
);
300 va
= shader
->bo
->gpu_address
;
301 si_pm4_add_bo(pm4
, shader
->bo
, RADEON_USAGE_READ
, RADEON_PRIO_SHADER_DATA
);
302 si_pm4_set_reg(pm4
, R_00B020_SPI_SHADER_PGM_LO_PS
, va
>> 8);
303 si_pm4_set_reg(pm4
, R_00B024_SPI_SHADER_PGM_HI_PS
, va
>> 40);
305 num_user_sgprs
= SI_PS_NUM_USER_SGPR
;
306 num_sgprs
= shader
->num_sgprs
;
307 /* One SGPR after user SGPRs is pre-loaded with {prim_mask, lds_offset} */
308 if ((num_user_sgprs
+ 1) > num_sgprs
) {
309 /* Last 2 reserved SGPRs are used for VCC */
310 num_sgprs
= num_user_sgprs
+ 1 + 2;
312 assert(num_sgprs
<= 104);
314 si_pm4_set_reg(pm4
, R_00B028_SPI_SHADER_PGM_RSRC1_PS
,
315 S_00B028_VGPRS((shader
->num_vgprs
- 1) / 4) |
316 S_00B028_SGPRS((num_sgprs
- 1) / 8));
317 si_pm4_set_reg(pm4
, R_00B02C_SPI_SHADER_PGM_RSRC2_PS
,
318 S_00B02C_EXTRA_LDS_SIZE(shader
->lds_size
) |
319 S_00B02C_USER_SGPR(num_user_sgprs
));
321 si_pm4_set_reg(pm4
, R_02880C_DB_SHADER_CONTROL
, db_shader_control
);
323 shader
->cb0_is_integer
= sctx
->framebuffer
.cb0_is_integer
;
324 sctx
->b
.flags
|= R600_CONTEXT_INV_SHADER_CACHE
;
331 static unsigned si_conv_pipe_prim(unsigned pprim
)
333 static const unsigned prim_conv
[] = {
334 [PIPE_PRIM_POINTS
] = V_008958_DI_PT_POINTLIST
,
335 [PIPE_PRIM_LINES
] = V_008958_DI_PT_LINELIST
,
336 [PIPE_PRIM_LINE_LOOP
] = V_008958_DI_PT_LINELOOP
,
337 [PIPE_PRIM_LINE_STRIP
] = V_008958_DI_PT_LINESTRIP
,
338 [PIPE_PRIM_TRIANGLES
] = V_008958_DI_PT_TRILIST
,
339 [PIPE_PRIM_TRIANGLE_STRIP
] = V_008958_DI_PT_TRISTRIP
,
340 [PIPE_PRIM_TRIANGLE_FAN
] = V_008958_DI_PT_TRIFAN
,
341 [PIPE_PRIM_QUADS
] = V_008958_DI_PT_QUADLIST
,
342 [PIPE_PRIM_QUAD_STRIP
] = V_008958_DI_PT_QUADSTRIP
,
343 [PIPE_PRIM_POLYGON
] = V_008958_DI_PT_POLYGON
,
344 [PIPE_PRIM_LINES_ADJACENCY
] = V_008958_DI_PT_LINELIST_ADJ
,
345 [PIPE_PRIM_LINE_STRIP_ADJACENCY
] = V_008958_DI_PT_LINESTRIP_ADJ
,
346 [PIPE_PRIM_TRIANGLES_ADJACENCY
] = V_008958_DI_PT_TRILIST_ADJ
,
347 [PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY
] = V_008958_DI_PT_TRISTRIP_ADJ
,
348 [R600_PRIM_RECTANGLE_LIST
] = V_008958_DI_PT_RECTLIST
350 unsigned result
= prim_conv
[pprim
];
352 R600_ERR("unsupported primitive type %d\n", pprim
);
357 static unsigned si_conv_prim_to_gs_out(unsigned mode
)
359 static const int prim_conv
[] = {
360 [PIPE_PRIM_POINTS
] = V_028A6C_OUTPRIM_TYPE_POINTLIST
,
361 [PIPE_PRIM_LINES
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
362 [PIPE_PRIM_LINE_LOOP
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
363 [PIPE_PRIM_LINE_STRIP
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
364 [PIPE_PRIM_TRIANGLES
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
365 [PIPE_PRIM_TRIANGLE_STRIP
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
366 [PIPE_PRIM_TRIANGLE_FAN
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
367 [PIPE_PRIM_QUADS
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
368 [PIPE_PRIM_QUAD_STRIP
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
369 [PIPE_PRIM_POLYGON
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
370 [PIPE_PRIM_LINES_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
371 [PIPE_PRIM_LINE_STRIP_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
372 [PIPE_PRIM_TRIANGLES_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
373 [PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
374 [R600_PRIM_RECTANGLE_LIST
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
376 assert(mode
< Elements(prim_conv
));
378 return prim_conv
[mode
];
381 static unsigned si_get_ia_multi_vgt_param(struct si_context
*sctx
,
382 const struct pipe_draw_info
*info
)
384 struct si_state_rasterizer
*rs
= sctx
->queued
.named
.rasterizer
;
385 unsigned prim
= info
->mode
;
386 unsigned primgroup_size
= 128; /* recommended without a GS */
388 /* SWITCH_ON_EOP(0) is always preferable. */
389 bool wd_switch_on_eop
= false;
390 bool ia_switch_on_eop
= false;
391 bool partial_vs_wave
= false;
394 primgroup_size
= 64; /* recommended with a GS */
396 /* This is a hardware requirement. */
397 if ((rs
&& rs
->line_stipple_enable
) ||
398 (sctx
->b
.screen
->debug_flags
& DBG_SWITCH_ON_EOP
)) {
399 ia_switch_on_eop
= true;
400 wd_switch_on_eop
= true;
403 if (sctx
->b
.streamout
.streamout_enabled
||
404 sctx
->b
.streamout
.prims_gen_query_enabled
)
405 partial_vs_wave
= true;
407 if (sctx
->b
.chip_class
>= CIK
) {
408 /* WD_SWITCH_ON_EOP has no effect on GPUs with less than
409 * 4 shader engines. Set 1 to pass the assertion below.
410 * The other cases are hardware requirements. */
411 if (sctx
->b
.screen
->info
.max_se
< 4 ||
412 prim
== PIPE_PRIM_POLYGON
||
413 prim
== PIPE_PRIM_LINE_LOOP
||
414 prim
== PIPE_PRIM_TRIANGLE_FAN
||
415 prim
== PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY
||
416 info
->primitive_restart
)
417 wd_switch_on_eop
= true;
419 /* Hawaii hangs if instancing is enabled and WD_SWITCH_ON_EOP is 0.
420 * We don't know that for indirect drawing, so treat it as
421 * always problematic. */
422 if (sctx
->b
.family
== CHIP_HAWAII
&&
423 (info
->indirect
|| info
->instance_count
> 1))
424 wd_switch_on_eop
= true;
426 /* If the WD switch is false, the IA switch must be false too. */
427 assert(wd_switch_on_eop
|| !ia_switch_on_eop
);
430 return S_028AA8_SWITCH_ON_EOP(ia_switch_on_eop
) |
431 S_028AA8_PARTIAL_VS_WAVE_ON(partial_vs_wave
) |
432 S_028AA8_PRIMGROUP_SIZE(primgroup_size
- 1) |
433 S_028AA8_WD_SWITCH_ON_EOP(sctx
->b
.chip_class
>= CIK
? wd_switch_on_eop
: 0);
436 static bool si_update_draw_info_state(struct si_context
*sctx
,
437 const struct pipe_draw_info
*info
,
438 const struct pipe_index_buffer
*ib
)
440 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
441 struct si_shader
*vs
= si_get_vs_state(sctx
);
442 unsigned prim
= si_conv_pipe_prim(info
->mode
);
443 unsigned gs_out_prim
=
444 si_conv_prim_to_gs_out(sctx
->gs_shader
?
445 sctx
->gs_shader
->current
->shader
.gs_output_prim
:
447 unsigned ls_mask
= 0;
448 unsigned ia_multi_vgt_param
= si_get_ia_multi_vgt_param(sctx
, info
);
458 if (sctx
->b
.chip_class
>= CIK
) {
459 si_pm4_set_reg(pm4
, R_028B74_VGT_DISPATCH_DRAW_INDEX
,
460 ib
->index_size
== 4 ? 0xFC000000 : 0xFC00);
462 si_pm4_cmd_begin(pm4
, PKT3_DRAW_PREAMBLE
);
463 si_pm4_cmd_add(pm4
, prim
); /* VGT_PRIMITIVE_TYPE */
464 si_pm4_cmd_add(pm4
, ia_multi_vgt_param
); /* IA_MULTI_VGT_PARAM */
465 si_pm4_cmd_add(pm4
, 0); /* VGT_LS_HS_CONFIG */
466 si_pm4_cmd_end(pm4
, false);
468 si_pm4_set_reg(pm4
, R_008958_VGT_PRIMITIVE_TYPE
, prim
);
469 si_pm4_set_reg(pm4
, R_028AA8_IA_MULTI_VGT_PARAM
, ia_multi_vgt_param
);
472 si_pm4_set_reg(pm4
, R_028A6C_VGT_GS_OUT_PRIM_TYPE
, gs_out_prim
);
473 si_pm4_set_reg(pm4
, R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX
, info
->restart_index
);
474 si_pm4_set_reg(pm4
, R_028A94_VGT_MULTI_PRIM_IB_RESET_EN
, info
->primitive_restart
);
476 if (prim
== V_008958_DI_PT_LINELIST
)
478 else if (prim
== V_008958_DI_PT_LINESTRIP
)
480 si_pm4_set_reg(pm4
, R_028A0C_PA_SC_LINE_STIPPLE
,
481 S_028A0C_AUTO_RESET_CNTL(ls_mask
) |
482 sctx
->pa_sc_line_stipple
);
484 if (info
->mode
== PIPE_PRIM_QUADS
|| info
->mode
== PIPE_PRIM_QUAD_STRIP
|| info
->mode
== PIPE_PRIM_POLYGON
) {
485 si_pm4_set_reg(pm4
, R_028814_PA_SU_SC_MODE_CNTL
,
486 S_028814_PROVOKING_VTX_LAST(1) | sctx
->pa_su_sc_mode_cntl
);
488 si_pm4_set_reg(pm4
, R_028814_PA_SU_SC_MODE_CNTL
, sctx
->pa_su_sc_mode_cntl
);
490 si_pm4_set_reg(pm4
, R_02881C_PA_CL_VS_OUT_CNTL
,
491 S_02881C_USE_VTX_POINT_SIZE(vs
->vs_out_point_size
) |
492 S_02881C_USE_VTX_EDGE_FLAG(vs
->vs_out_edgeflag
) |
493 S_02881C_USE_VTX_RENDER_TARGET_INDX(vs
->vs_out_layer
) |
494 S_02881C_VS_OUT_CCDIST0_VEC_ENA((vs
->clip_dist_write
& 0x0F) != 0) |
495 S_02881C_VS_OUT_CCDIST1_VEC_ENA((vs
->clip_dist_write
& 0xF0) != 0) |
496 S_02881C_VS_OUT_MISC_VEC_ENA(vs
->vs_out_misc_write
) |
497 (sctx
->queued
.named
.rasterizer
->clip_plane_enable
&
498 vs
->clip_dist_write
));
499 si_pm4_set_reg(pm4
, R_028810_PA_CL_CLIP_CNTL
,
500 sctx
->queued
.named
.rasterizer
->pa_cl_clip_cntl
|
501 (vs
->clip_dist_write
? 0 :
502 sctx
->queued
.named
.rasterizer
->clip_plane_enable
& 0x3F));
504 si_pm4_set_state(sctx
, draw_info
, pm4
);
508 static void si_update_spi_map(struct si_context
*sctx
)
510 struct si_shader
*ps
= &sctx
->ps_shader
->current
->shader
;
511 struct si_shader
*vs
= si_get_vs_state(sctx
);
512 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
515 for (i
= 0; i
< ps
->ninput
; i
++) {
516 unsigned name
= ps
->input
[i
].name
;
517 unsigned param_offset
= ps
->input
[i
].param_offset
;
519 if (name
== TGSI_SEMANTIC_POSITION
)
520 /* Read from preloaded VGPRs, not parameters */
526 if (ps
->input
[i
].interpolate
== TGSI_INTERPOLATE_CONSTANT
||
527 (ps
->input
[i
].interpolate
== TGSI_INTERPOLATE_COLOR
&&
528 sctx
->ps_shader
->current
->key
.ps
.flatshade
)) {
529 tmp
|= S_028644_FLAT_SHADE(1);
532 if (name
== TGSI_SEMANTIC_GENERIC
&&
533 sctx
->sprite_coord_enable
& (1 << ps
->input
[i
].sid
)) {
534 tmp
|= S_028644_PT_SPRITE_TEX(1);
537 for (j
= 0; j
< vs
->noutput
; j
++) {
538 if (name
== vs
->output
[j
].name
&&
539 ps
->input
[i
].sid
== vs
->output
[j
].sid
) {
540 tmp
|= S_028644_OFFSET(vs
->output
[j
].param_offset
);
545 if (j
== vs
->noutput
) {
546 /* No corresponding output found, load defaults into input */
547 tmp
|= S_028644_OFFSET(0x20);
551 R_028644_SPI_PS_INPUT_CNTL_0
+ param_offset
* 4,
554 if (name
== TGSI_SEMANTIC_COLOR
&&
555 sctx
->ps_shader
->current
->key
.ps
.color_two_side
) {
556 name
= TGSI_SEMANTIC_BCOLOR
;
562 si_pm4_set_state(sctx
, spi
, pm4
);
565 /* Initialize state related to ESGS / GSVS ring buffers */
566 static void si_init_gs_rings(struct si_context
*sctx
)
568 unsigned size
= 128 * 1024;
570 assert(!sctx
->gs_rings
);
571 sctx
->gs_rings
= si_pm4_alloc_state(sctx
);
573 sctx
->esgs_ring
.buffer
=
574 pipe_buffer_create(sctx
->b
.b
.screen
, PIPE_BIND_CUSTOM
,
575 PIPE_USAGE_DEFAULT
, size
);
576 sctx
->esgs_ring
.buffer_size
= size
;
578 size
= 64 * 1024 * 1024;
579 sctx
->gsvs_ring
.buffer
=
580 pipe_buffer_create(sctx
->b
.b
.screen
, PIPE_BIND_CUSTOM
,
581 PIPE_USAGE_DEFAULT
, size
);
582 sctx
->gsvs_ring
.buffer_size
= size
;
584 if (sctx
->b
.chip_class
>= CIK
) {
585 si_pm4_set_reg(sctx
->gs_rings
, R_030900_VGT_ESGS_RING_SIZE
,
586 sctx
->esgs_ring
.buffer_size
/ 256);
587 si_pm4_set_reg(sctx
->gs_rings
, R_030904_VGT_GSVS_RING_SIZE
,
588 sctx
->gsvs_ring
.buffer_size
/ 256);
590 si_pm4_set_reg(sctx
->gs_rings
, R_0088C8_VGT_ESGS_RING_SIZE
,
591 sctx
->esgs_ring
.buffer_size
/ 256);
592 si_pm4_set_reg(sctx
->gs_rings
, R_0088CC_VGT_GSVS_RING_SIZE
,
593 sctx
->gsvs_ring
.buffer_size
/ 256);
596 si_set_ring_buffer(&sctx
->b
.b
, PIPE_SHADER_VERTEX
, SI_RING_ESGS
,
597 &sctx
->esgs_ring
, 0, sctx
->esgs_ring
.buffer_size
,
599 si_set_ring_buffer(&sctx
->b
.b
, PIPE_SHADER_GEOMETRY
, SI_RING_ESGS
,
600 &sctx
->esgs_ring
, 0, sctx
->esgs_ring
.buffer_size
,
602 si_set_ring_buffer(&sctx
->b
.b
, PIPE_SHADER_VERTEX
, SI_RING_GSVS
,
603 &sctx
->gsvs_ring
, 0, sctx
->gsvs_ring
.buffer_size
,
607 static void si_update_derived_state(struct si_context
*sctx
)
609 struct pipe_context
* ctx
= (struct pipe_context
*)sctx
;
611 if (!sctx
->blitter
->running
) {
612 /* Flush depth textures which need to be flushed. */
613 for (int i
= 0; i
< SI_NUM_SHADERS
; i
++) {
614 if (sctx
->samplers
[i
].depth_texture_mask
) {
615 si_flush_depth_textures(sctx
, &sctx
->samplers
[i
]);
617 if (sctx
->samplers
[i
].compressed_colortex_mask
) {
618 si_decompress_color_textures(sctx
, &sctx
->samplers
[i
]);
623 if (sctx
->gs_shader
) {
624 si_shader_select(ctx
, sctx
->gs_shader
);
626 if (!sctx
->gs_shader
->current
->pm4
) {
627 si_pipe_shader_gs(ctx
, sctx
->gs_shader
->current
);
628 si_pipe_shader_vs(ctx
,
629 sctx
->gs_shader
->current
->gs_copy_shader
);
632 si_pm4_bind_state(sctx
, gs
, sctx
->gs_shader
->current
->pm4
);
633 si_pm4_bind_state(sctx
, vs
, sctx
->gs_shader
->current
->gs_copy_shader
->pm4
);
635 sctx
->b
.streamout
.stride_in_dw
= sctx
->gs_shader
->so
.stride
;
637 si_shader_select(ctx
, sctx
->vs_shader
);
639 if (!sctx
->vs_shader
->current
->pm4
)
640 si_pipe_shader_es(ctx
, sctx
->vs_shader
->current
);
642 si_pm4_bind_state(sctx
, es
, sctx
->vs_shader
->current
->pm4
);
645 si_init_gs_rings(sctx
);
646 if (sctx
->emitted
.named
.gs_rings
!= sctx
->gs_rings
)
647 sctx
->b
.flags
|= R600_CONTEXT_VGT_FLUSH
;
648 si_pm4_bind_state(sctx
, gs_rings
, sctx
->gs_rings
);
650 si_set_ring_buffer(ctx
, PIPE_SHADER_GEOMETRY
, SI_RING_GSVS
,
652 sctx
->gs_shader
->current
->shader
.gs_max_out_vertices
*
653 sctx
->gs_shader
->current
->shader
.noutput
* 16,
654 64, true, true, 4, 16);
657 sctx
->gs_on
= si_pm4_alloc_state(sctx
);
659 si_pm4_set_reg(sctx
->gs_on
, R_028B54_VGT_SHADER_STAGES_EN
,
660 S_028B54_ES_EN(V_028B54_ES_STAGE_REAL
) |
662 S_028B54_VS_EN(V_028B54_VS_STAGE_COPY_SHADER
));
664 si_pm4_bind_state(sctx
, gs_onoff
, sctx
->gs_on
);
666 si_shader_select(ctx
, sctx
->vs_shader
);
668 if (!sctx
->vs_shader
->current
->pm4
)
669 si_pipe_shader_vs(ctx
, sctx
->vs_shader
->current
);
671 si_pm4_bind_state(sctx
, vs
, sctx
->vs_shader
->current
->pm4
);
673 sctx
->b
.streamout
.stride_in_dw
= sctx
->vs_shader
->so
.stride
;
676 sctx
->gs_off
= si_pm4_alloc_state(sctx
);
678 si_pm4_set_reg(sctx
->gs_off
, R_028A40_VGT_GS_MODE
, 0);
679 si_pm4_set_reg(sctx
->gs_off
, R_028B54_VGT_SHADER_STAGES_EN
, 0);
681 si_pm4_bind_state(sctx
, gs_onoff
, sctx
->gs_off
);
682 si_pm4_bind_state(sctx
, gs_rings
, NULL
);
683 si_pm4_bind_state(sctx
, gs
, NULL
);
684 si_pm4_bind_state(sctx
, es
, NULL
);
687 si_shader_select(ctx
, sctx
->ps_shader
);
689 if (!sctx
->ps_shader
->current
->pm4
||
690 sctx
->ps_shader
->current
->cb0_is_integer
!= sctx
->framebuffer
.cb0_is_integer
)
691 si_pipe_shader_ps(ctx
, sctx
->ps_shader
->current
);
693 si_pm4_bind_state(sctx
, ps
, sctx
->ps_shader
->current
->pm4
);
695 if (si_pm4_state_changed(sctx
, ps
) || si_pm4_state_changed(sctx
, vs
)) {
696 /* XXX: Emitting the PS state even when only the VS changed
697 * fixes random failures with piglit glsl-max-varyings.
700 sctx
->emitted
.named
.ps
= NULL
;
701 si_update_spi_map(sctx
);
705 static void si_state_draw(struct si_context
*sctx
,
706 const struct pipe_draw_info
*info
,
707 const struct pipe_index_buffer
*ib
)
709 unsigned sh_base_reg
= (sctx
->gs_shader
? R_00B330_SPI_SHADER_USER_DATA_ES_0
:
710 R_00B130_SPI_SHADER_USER_DATA_VS_0
);
711 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
716 /* queries need some special values
717 * (this is non-zero if any query is active) */
718 if (sctx
->b
.num_occlusion_queries
> 0) {
719 if (sctx
->b
.chip_class
>= CIK
) {
720 si_pm4_set_reg(pm4
, R_028004_DB_COUNT_CONTROL
,
721 S_028004_PERFECT_ZPASS_COUNTS(1) |
722 S_028004_SAMPLE_RATE(sctx
->framebuffer
.log_samples
) |
723 S_028004_ZPASS_ENABLE(1) |
724 S_028004_SLICE_EVEN_ENABLE(1) |
725 S_028004_SLICE_ODD_ENABLE(1));
727 si_pm4_set_reg(pm4
, R_028004_DB_COUNT_CONTROL
,
728 S_028004_PERFECT_ZPASS_COUNTS(1) |
729 S_028004_SAMPLE_RATE(sctx
->framebuffer
.log_samples
));
732 /* Disable occlusion queries. */
733 if (sctx
->b
.chip_class
>= CIK
) {
734 si_pm4_set_reg(pm4
, R_028004_DB_COUNT_CONTROL
, 0);
736 si_pm4_set_reg(pm4
, R_028004_DB_COUNT_CONTROL
,
737 S_028004_ZPASS_INCREMENT_DISABLE(1));
741 /* DB_RENDER_CONTROL */
742 if (sctx
->dbcb_depth_copy_enabled
||
743 sctx
->dbcb_stencil_copy_enabled
) {
744 si_pm4_set_reg(pm4
, R_028000_DB_RENDER_CONTROL
,
745 S_028000_DEPTH_COPY(sctx
->dbcb_depth_copy_enabled
) |
746 S_028000_STENCIL_COPY(sctx
->dbcb_stencil_copy_enabled
) |
747 S_028000_COPY_CENTROID(1) |
748 S_028000_COPY_SAMPLE(sctx
->dbcb_copy_sample
));
749 } else if (sctx
->db_inplace_flush_enabled
) {
750 si_pm4_set_reg(pm4
, R_028000_DB_RENDER_CONTROL
,
751 S_028000_DEPTH_COMPRESS_DISABLE(1) |
752 S_028000_STENCIL_COMPRESS_DISABLE(1));
753 } else if (sctx
->db_depth_clear
) {
754 si_pm4_set_reg(pm4
, R_028000_DB_RENDER_CONTROL
,
755 S_028000_DEPTH_CLEAR_ENABLE(1));
757 si_pm4_set_reg(pm4
, R_028000_DB_RENDER_CONTROL
, 0);
760 /* DB_RENDER_OVERRIDE2 */
761 if (sctx
->db_depth_disable_expclear
) {
762 si_pm4_set_reg(pm4
, R_028010_DB_RENDER_OVERRIDE2
,
763 S_028010_DISABLE_ZMASK_EXPCLEAR_OPTIMIZATION(1));
765 si_pm4_set_reg(pm4
, R_028010_DB_RENDER_OVERRIDE2
, 0);
768 if (info
->count_from_stream_output
) {
769 struct r600_so_target
*t
=
770 (struct r600_so_target
*)info
->count_from_stream_output
;
771 uint64_t va
= t
->buf_filled_size
->gpu_address
+
772 t
->buf_filled_size_offset
;
774 si_pm4_set_reg(pm4
, R_028B30_VGT_STRMOUT_DRAW_OPAQUE_VERTEX_STRIDE
,
777 si_pm4_cmd_begin(pm4
, PKT3_COPY_DATA
);
779 COPY_DATA_SRC_SEL(COPY_DATA_MEM
) |
780 COPY_DATA_DST_SEL(COPY_DATA_REG
) |
781 COPY_DATA_WR_CONFIRM
);
782 si_pm4_cmd_add(pm4
, va
); /* src address lo */
783 si_pm4_cmd_add(pm4
, va
>> 32UL); /* src address hi */
784 si_pm4_cmd_add(pm4
, R_028B2C_VGT_STRMOUT_DRAW_OPAQUE_BUFFER_FILLED_SIZE
>> 2);
785 si_pm4_cmd_add(pm4
, 0); /* unused */
786 si_pm4_add_bo(pm4
, t
->buf_filled_size
, RADEON_USAGE_READ
,
788 si_pm4_cmd_end(pm4
, true);
792 si_pm4_cmd_begin(pm4
, PKT3_INDEX_TYPE
);
793 if (ib
->index_size
== 4) {
794 si_pm4_cmd_add(pm4
, V_028A7C_VGT_INDEX_32
| (SI_BIG_ENDIAN
?
795 V_028A7C_VGT_DMA_SWAP_32_BIT
: 0));
797 si_pm4_cmd_add(pm4
, V_028A7C_VGT_INDEX_16
| (SI_BIG_ENDIAN
?
798 V_028A7C_VGT_DMA_SWAP_16_BIT
: 0));
800 si_pm4_cmd_end(pm4
, sctx
->b
.predicate_drawing
);
802 if (!info
->indirect
) {
803 si_pm4_cmd_begin(pm4
, PKT3_NUM_INSTANCES
);
804 si_pm4_cmd_add(pm4
, info
->instance_count
);
805 si_pm4_cmd_end(pm4
, sctx
->b
.predicate_drawing
);
807 si_pm4_set_reg(pm4
, sh_base_reg
+ SI_SGPR_BASE_VERTEX
* 4,
808 info
->indexed
? info
->index_bias
: info
->start
);
809 si_pm4_set_reg(pm4
, sh_base_reg
+ SI_SGPR_START_INSTANCE
* 4,
810 info
->start_instance
);
812 si_pm4_add_bo(pm4
, (struct r600_resource
*)info
->indirect
,
813 RADEON_USAGE_READ
, RADEON_PRIO_MIN
);
817 uint32_t max_size
= (ib
->buffer
->width0
- ib
->offset
) /
818 sctx
->index_buffer
.index_size
;
819 uint64_t va
= r600_resource(ib
->buffer
)->gpu_address
+ ib
->offset
;
821 si_pm4_add_bo(pm4
, (struct r600_resource
*)ib
->buffer
, RADEON_USAGE_READ
,
824 if (info
->indirect
) {
825 uint64_t indirect_va
= r600_resource(info
->indirect
)->gpu_address
;
826 si_cmd_draw_index_indirect(pm4
, indirect_va
, va
, max_size
,
827 info
->indirect_offset
,
828 sh_base_reg
+ SI_SGPR_BASE_VERTEX
* 4,
829 sh_base_reg
+ SI_SGPR_START_INSTANCE
* 4,
830 sctx
->b
.predicate_drawing
);
832 va
+= info
->start
* ib
->index_size
;
833 si_cmd_draw_index_2(pm4
, max_size
, va
, info
->count
,
834 V_0287F0_DI_SRC_SEL_DMA
,
835 sctx
->b
.predicate_drawing
);
838 if (info
->indirect
) {
839 uint64_t indirect_va
= r600_resource(info
->indirect
)->gpu_address
;
840 si_cmd_draw_indirect(pm4
, indirect_va
, info
->indirect_offset
,
841 sh_base_reg
+ SI_SGPR_BASE_VERTEX
* 4,
842 sh_base_reg
+ SI_SGPR_START_INSTANCE
* 4,
843 sctx
->b
.predicate_drawing
);
845 si_cmd_draw_index_auto(pm4
, info
->count
,
846 V_0287F0_DI_SRC_SEL_AUTO_INDEX
|
847 S_0287F0_USE_OPAQUE(!!info
->count_from_stream_output
),
848 sctx
->b
.predicate_drawing
);
852 si_pm4_set_state(sctx
, draw
, pm4
);
855 void si_emit_cache_flush(struct r600_common_context
*sctx
, struct r600_atom
*atom
)
857 struct radeon_winsys_cs
*cs
= sctx
->rings
.gfx
.cs
;
858 uint32_t cp_coher_cntl
= 0;
860 /* XXX SI flushes both ICACHE and KCACHE if either flag is set.
861 * XXX CIK shouldn't have this issue. Test CIK before separating the flags
862 * XXX to ensure there is no regression. Also find out if there is another
863 * XXX way to flush either ICACHE or KCACHE but not both for SI. */
864 if (sctx
->flags
& (R600_CONTEXT_INV_SHADER_CACHE
|
865 R600_CONTEXT_INV_CONST_CACHE
)) {
866 cp_coher_cntl
|= S_0085F0_SH_ICACHE_ACTION_ENA(1) |
867 S_0085F0_SH_KCACHE_ACTION_ENA(1);
869 if (sctx
->flags
& (R600_CONTEXT_INV_TEX_CACHE
|
870 R600_CONTEXT_STREAMOUT_FLUSH
)) {
871 cp_coher_cntl
|= S_0085F0_TC_ACTION_ENA(1) |
872 S_0085F0_TCL1_ACTION_ENA(1);
874 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_CB
) {
875 cp_coher_cntl
|= S_0085F0_CB_ACTION_ENA(1) |
876 S_0085F0_CB0_DEST_BASE_ENA(1) |
877 S_0085F0_CB1_DEST_BASE_ENA(1) |
878 S_0085F0_CB2_DEST_BASE_ENA(1) |
879 S_0085F0_CB3_DEST_BASE_ENA(1) |
880 S_0085F0_CB4_DEST_BASE_ENA(1) |
881 S_0085F0_CB5_DEST_BASE_ENA(1) |
882 S_0085F0_CB6_DEST_BASE_ENA(1) |
883 S_0085F0_CB7_DEST_BASE_ENA(1);
885 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_DB
) {
886 cp_coher_cntl
|= S_0085F0_DB_ACTION_ENA(1) |
887 S_0085F0_DB_DEST_BASE_ENA(1);
891 if (sctx
->chip_class
>= CIK
) {
892 radeon_emit(cs
, PKT3(PKT3_ACQUIRE_MEM
, 5, 0));
893 radeon_emit(cs
, cp_coher_cntl
); /* CP_COHER_CNTL */
894 radeon_emit(cs
, 0xffffffff); /* CP_COHER_SIZE */
895 radeon_emit(cs
, 0xff); /* CP_COHER_SIZE_HI */
896 radeon_emit(cs
, 0); /* CP_COHER_BASE */
897 radeon_emit(cs
, 0); /* CP_COHER_BASE_HI */
898 radeon_emit(cs
, 0x0000000A); /* POLL_INTERVAL */
900 radeon_emit(cs
, PKT3(PKT3_SURFACE_SYNC
, 3, 0));
901 radeon_emit(cs
, cp_coher_cntl
); /* CP_COHER_CNTL */
902 radeon_emit(cs
, 0xffffffff); /* CP_COHER_SIZE */
903 radeon_emit(cs
, 0); /* CP_COHER_BASE */
904 radeon_emit(cs
, 0x0000000A); /* POLL_INTERVAL */
908 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_CB_META
) {
909 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
910 radeon_emit(cs
, EVENT_TYPE(V_028A90_FLUSH_AND_INV_CB_META
) | EVENT_INDEX(0));
912 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_DB_META
) {
913 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
914 radeon_emit(cs
, EVENT_TYPE(V_028A90_FLUSH_AND_INV_DB_META
) | EVENT_INDEX(0));
917 if (sctx
->flags
& (R600_CONTEXT_WAIT_3D_IDLE
|
918 R600_CONTEXT_PS_PARTIAL_FLUSH
)) {
919 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
920 radeon_emit(cs
, EVENT_TYPE(V_028A90_PS_PARTIAL_FLUSH
) | EVENT_INDEX(4));
921 } else if (sctx
->flags
& R600_CONTEXT_STREAMOUT_FLUSH
) {
922 /* Needed if streamout buffers are going to be used as a source. */
923 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
924 radeon_emit(cs
, EVENT_TYPE(V_028A90_VS_PARTIAL_FLUSH
) | EVENT_INDEX(4));
927 if (sctx
->flags
& R600_CONTEXT_VGT_FLUSH
) {
928 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
929 radeon_emit(cs
, EVENT_TYPE(V_028A90_VGT_FLUSH
) | EVENT_INDEX(0));
931 if (sctx
->flags
& R600_CONTEXT_VGT_STREAMOUT_SYNC
) {
932 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
933 radeon_emit(cs
, EVENT_TYPE(V_028A90_VGT_STREAMOUT_SYNC
) | EVENT_INDEX(0));
939 const struct r600_atom si_atom_cache_flush
= { si_emit_cache_flush
, 17 }; /* number of CS dwords */
941 static void si_get_draw_start_count(struct si_context
*sctx
,
942 const struct pipe_draw_info
*info
,
943 unsigned *start
, unsigned *count
)
945 if (info
->indirect
) {
946 struct r600_resource
*indirect
=
947 (struct r600_resource
*)info
->indirect
;
948 int *data
= r600_buffer_map_sync_with_rings(&sctx
->b
,
949 indirect
, PIPE_TRANSFER_READ
);
950 data
+= info
->indirect_offset
/sizeof(int);
954 *start
= info
->start
;
955 *count
= info
->count
;
959 void si_draw_vbo(struct pipe_context
*ctx
, const struct pipe_draw_info
*info
)
961 struct si_context
*sctx
= (struct si_context
*)ctx
;
962 struct pipe_index_buffer ib
= {};
965 if (!info
->count
&& !info
->indirect
&&
966 (info
->indexed
|| !info
->count_from_stream_output
))
969 if (!sctx
->ps_shader
|| !sctx
->vs_shader
)
972 si_update_derived_state(sctx
);
974 if (sctx
->vertex_buffers_dirty
) {
975 si_update_vertex_buffers(sctx
);
976 sctx
->vertex_buffers_dirty
= false;
980 /* Initialize the index buffer struct. */
981 pipe_resource_reference(&ib
.buffer
, sctx
->index_buffer
.buffer
);
982 ib
.user_buffer
= sctx
->index_buffer
.user_buffer
;
983 ib
.index_size
= sctx
->index_buffer
.index_size
;
984 ib
.offset
= sctx
->index_buffer
.offset
;
986 /* Translate or upload, if needed. */
987 if (ib
.index_size
== 1) {
988 struct pipe_resource
*out_buffer
= NULL
;
989 unsigned out_offset
, start
, count
, start_offset
;
992 si_get_draw_start_count(sctx
, info
, &start
, &count
);
993 start_offset
= start
* ib
.index_size
;
995 u_upload_alloc(sctx
->b
.uploader
, start_offset
, count
* 2,
996 &out_offset
, &out_buffer
, &ptr
);
998 util_shorten_ubyte_elts_to_userptr(&sctx
->b
.b
, &ib
, 0,
999 ib
.offset
+ start_offset
,
1002 pipe_resource_reference(&ib
.buffer
, NULL
);
1003 ib
.user_buffer
= NULL
;
1004 ib
.buffer
= out_buffer
;
1005 /* info->start will be added by the drawing code */
1006 ib
.offset
= out_offset
- start_offset
;
1008 } else if (ib
.user_buffer
&& !ib
.buffer
) {
1009 unsigned start
, count
, start_offset
;
1011 si_get_draw_start_count(sctx
, info
, &start
, &count
);
1012 start_offset
= start
* ib
.index_size
;
1014 u_upload_data(sctx
->b
.uploader
, start_offset
, count
* ib
.index_size
,
1015 (char*)ib
.user_buffer
+ start_offset
,
1016 &ib
.offset
, &ib
.buffer
);
1017 /* info->start will be added by the drawing code */
1018 ib
.offset
-= start_offset
;
1022 if (!si_update_draw_info_state(sctx
, info
, &ib
))
1025 si_state_draw(sctx
, info
, &ib
);
1027 sctx
->pm4_dirty_cdwords
+= si_pm4_dirty_dw(sctx
);
1029 /* Check flush flags. */
1031 sctx
->atoms
.s
.cache_flush
->dirty
= true;
1033 si_need_cs_space(sctx
, 0, TRUE
);
1036 for (i
= 0; i
< SI_NUM_ATOMS(sctx
); i
++) {
1037 if (sctx
->atoms
.array
[i
]->dirty
) {
1038 sctx
->atoms
.array
[i
]->emit(&sctx
->b
, sctx
->atoms
.array
[i
]);
1039 sctx
->atoms
.array
[i
]->dirty
= false;
1043 si_pm4_emit_dirty(sctx
);
1044 sctx
->pm4_dirty_cdwords
= 0;
1047 if (sctx
->screen
->b
.trace_bo
) {
1048 si_trace_emit(sctx
);
1052 /* Workaround for a VGT hang when streamout is enabled.
1053 * It must be done after drawing. */
1054 if (sctx
->b
.family
== CHIP_HAWAII
&&
1055 (sctx
->b
.streamout
.streamout_enabled
||
1056 sctx
->b
.streamout
.prims_gen_query_enabled
)) {
1057 sctx
->b
.flags
|= R600_CONTEXT_VGT_STREAMOUT_SYNC
;
1060 /* Set the depth buffer as dirty. */
1061 if (sctx
->framebuffer
.state
.zsbuf
) {
1062 struct pipe_surface
*surf
= sctx
->framebuffer
.state
.zsbuf
;
1063 struct r600_texture
*rtex
= (struct r600_texture
*)surf
->texture
;
1065 rtex
->dirty_level_mask
|= 1 << surf
->u
.tex
.level
;
1067 if (sctx
->framebuffer
.compressed_cb_mask
) {
1068 struct pipe_surface
*surf
;
1069 struct r600_texture
*rtex
;
1070 unsigned mask
= sctx
->framebuffer
.compressed_cb_mask
;
1073 unsigned i
= u_bit_scan(&mask
);
1074 surf
= sctx
->framebuffer
.state
.cbufs
[i
];
1075 rtex
= (struct r600_texture
*)surf
->texture
;
1077 rtex
->dirty_level_mask
|= 1 << surf
->u
.tex
.level
;
1081 pipe_resource_reference(&ib
.buffer
, NULL
);
1082 sctx
->b
.num_draw_calls
++;