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 pipe_context
*ctx
, struct si_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
->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
));
80 static void si_shader_gs(struct pipe_context
*ctx
, struct si_shader
*shader
)
82 struct si_context
*sctx
= (struct si_context
*)ctx
;
83 unsigned gs_vert_itemsize
= shader
->noutput
* (16 >> 2);
84 unsigned gs_max_vert_out
= shader
->selector
->gs_max_out_vertices
;
85 unsigned gsvs_itemsize
= gs_vert_itemsize
* gs_max_vert_out
;
87 struct si_pm4_state
*pm4
;
88 unsigned num_sgprs
, num_user_sgprs
;
91 /* The GSVS_RING_ITEMSIZE register takes 15 bits */
92 assert(gsvs_itemsize
< (1 << 15));
94 si_pm4_delete_state(sctx
, gs
, shader
->pm4
);
95 pm4
= shader
->pm4
= si_pm4_alloc_state(sctx
);
100 if (gs_max_vert_out
<= 128) {
101 cut_mode
= V_028A40_GS_CUT_128
;
102 } else if (gs_max_vert_out
<= 256) {
103 cut_mode
= V_028A40_GS_CUT_256
;
104 } else if (gs_max_vert_out
<= 512) {
105 cut_mode
= V_028A40_GS_CUT_512
;
107 assert(gs_max_vert_out
<= 1024);
108 cut_mode
= V_028A40_GS_CUT_1024
;
111 si_pm4_set_reg(pm4
, R_028A40_VGT_GS_MODE
,
112 S_028A40_MODE(V_028A40_GS_SCENARIO_G
) |
113 S_028A40_CUT_MODE(cut_mode
)|
114 S_028A40_ES_WRITE_OPTIMIZE(1) |
115 S_028A40_GS_WRITE_OPTIMIZE(1));
117 si_pm4_set_reg(pm4
, R_028A60_VGT_GSVS_RING_OFFSET_1
, gsvs_itemsize
);
118 si_pm4_set_reg(pm4
, R_028A64_VGT_GSVS_RING_OFFSET_2
, gsvs_itemsize
);
119 si_pm4_set_reg(pm4
, R_028A68_VGT_GSVS_RING_OFFSET_3
, gsvs_itemsize
);
121 si_pm4_set_reg(pm4
, R_028AAC_VGT_ESGS_RING_ITEMSIZE
,
122 util_bitcount64(shader
->selector
->gs_used_inputs
) * (16 >> 2));
123 si_pm4_set_reg(pm4
, R_028AB0_VGT_GSVS_RING_ITEMSIZE
, gsvs_itemsize
);
125 si_pm4_set_reg(pm4
, R_028B38_VGT_GS_MAX_VERT_OUT
, gs_max_vert_out
);
127 si_pm4_set_reg(pm4
, R_028B5C_VGT_GS_VERT_ITEMSIZE
, gs_vert_itemsize
);
129 va
= shader
->bo
->gpu_address
;
130 si_pm4_add_bo(pm4
, shader
->bo
, RADEON_USAGE_READ
, RADEON_PRIO_SHADER_DATA
);
131 si_pm4_set_reg(pm4
, R_00B220_SPI_SHADER_PGM_LO_GS
, va
>> 8);
132 si_pm4_set_reg(pm4
, R_00B224_SPI_SHADER_PGM_HI_GS
, va
>> 40);
134 num_user_sgprs
= SI_GS_NUM_USER_SGPR
;
135 num_sgprs
= shader
->num_sgprs
;
136 /* Two SGPRs after user SGPRs are pre-loaded with gs2vs_offset, gs_wave_id */
137 if ((num_user_sgprs
+ 2) > num_sgprs
) {
138 /* Last 2 reserved SGPRs are used for VCC */
139 num_sgprs
= num_user_sgprs
+ 2 + 2;
141 assert(num_sgprs
<= 104);
143 si_pm4_set_reg(pm4
, R_00B228_SPI_SHADER_PGM_RSRC1_GS
,
144 S_00B228_VGPRS((shader
->num_vgprs
- 1) / 4) |
145 S_00B228_SGPRS((num_sgprs
- 1) / 8));
146 si_pm4_set_reg(pm4
, R_00B22C_SPI_SHADER_PGM_RSRC2_GS
,
147 S_00B22C_USER_SGPR(num_user_sgprs
));
150 static void si_shader_vs(struct pipe_context
*ctx
, struct si_shader
*shader
)
152 struct si_context
*sctx
= (struct si_context
*)ctx
;
153 struct tgsi_shader_info
*info
= &shader
->selector
->info
;
154 struct si_pm4_state
*pm4
;
155 unsigned num_sgprs
, num_user_sgprs
;
156 unsigned nparams
, i
, vgpr_comp_cnt
;
159 si_pm4_delete_state(sctx
, vs
, shader
->pm4
);
160 pm4
= shader
->pm4
= si_pm4_alloc_state(sctx
);
165 va
= shader
->bo
->gpu_address
;
166 si_pm4_add_bo(pm4
, shader
->bo
, RADEON_USAGE_READ
, RADEON_PRIO_SHADER_DATA
);
168 vgpr_comp_cnt
= shader
->uses_instanceid
? 3 : 0;
170 if (shader
->is_gs_copy_shader
)
171 num_user_sgprs
= SI_GSCOPY_NUM_USER_SGPR
;
173 num_user_sgprs
= SI_VS_NUM_USER_SGPR
;
175 num_sgprs
= shader
->num_sgprs
;
176 if (num_user_sgprs
> num_sgprs
) {
177 /* Last 2 reserved SGPRs are used for VCC */
178 num_sgprs
= num_user_sgprs
+ 2;
180 assert(num_sgprs
<= 104);
182 /* Certain attributes (position, psize, etc.) don't count as params.
183 * VS is required to export at least one param and r600_shader_from_tgsi()
184 * takes care of adding a dummy export.
186 for (nparams
= 0, i
= 0 ; i
< info
->num_outputs
; i
++) {
187 switch (info
->output_semantic_name
[i
]) {
188 case TGSI_SEMANTIC_CLIPVERTEX
:
189 case TGSI_SEMANTIC_POSITION
:
190 case TGSI_SEMANTIC_PSIZE
:
199 si_pm4_set_reg(pm4
, R_0286C4_SPI_VS_OUT_CONFIG
,
200 S_0286C4_VS_EXPORT_COUNT(nparams
- 1));
202 si_pm4_set_reg(pm4
, R_02870C_SPI_SHADER_POS_FORMAT
,
203 S_02870C_POS0_EXPORT_FORMAT(V_02870C_SPI_SHADER_4COMP
) |
204 S_02870C_POS1_EXPORT_FORMAT(shader
->nr_pos_exports
> 1 ?
205 V_02870C_SPI_SHADER_4COMP
:
206 V_02870C_SPI_SHADER_NONE
) |
207 S_02870C_POS2_EXPORT_FORMAT(shader
->nr_pos_exports
> 2 ?
208 V_02870C_SPI_SHADER_4COMP
:
209 V_02870C_SPI_SHADER_NONE
) |
210 S_02870C_POS3_EXPORT_FORMAT(shader
->nr_pos_exports
> 3 ?
211 V_02870C_SPI_SHADER_4COMP
:
212 V_02870C_SPI_SHADER_NONE
));
214 si_pm4_set_reg(pm4
, R_00B120_SPI_SHADER_PGM_LO_VS
, va
>> 8);
215 si_pm4_set_reg(pm4
, R_00B124_SPI_SHADER_PGM_HI_VS
, va
>> 40);
216 si_pm4_set_reg(pm4
, R_00B128_SPI_SHADER_PGM_RSRC1_VS
,
217 S_00B128_VGPRS((shader
->num_vgprs
- 1) / 4) |
218 S_00B128_SGPRS((num_sgprs
- 1) / 8) |
219 S_00B128_VGPR_COMP_CNT(vgpr_comp_cnt
));
220 si_pm4_set_reg(pm4
, R_00B12C_SPI_SHADER_PGM_RSRC2_VS
,
221 S_00B12C_USER_SGPR(num_user_sgprs
) |
222 S_00B12C_SO_BASE0_EN(!!shader
->selector
->so
.stride
[0]) |
223 S_00B12C_SO_BASE1_EN(!!shader
->selector
->so
.stride
[1]) |
224 S_00B12C_SO_BASE2_EN(!!shader
->selector
->so
.stride
[2]) |
225 S_00B12C_SO_BASE3_EN(!!shader
->selector
->so
.stride
[3]) |
226 S_00B12C_SO_EN(!!shader
->selector
->so
.num_outputs
));
229 static void si_shader_ps(struct pipe_context
*ctx
, struct si_shader
*shader
)
231 struct si_context
*sctx
= (struct si_context
*)ctx
;
232 struct tgsi_shader_info
*info
= &shader
->selector
->info
;
233 struct si_pm4_state
*pm4
;
234 unsigned i
, spi_ps_in_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 for (i
= 0; i
< info
->num_inputs
; i
++) {
246 switch (info
->input_semantic_name
[i
]) {
247 case TGSI_SEMANTIC_POSITION
:
248 if (info
->input_interpolate_loc
[i
] ==
249 TGSI_INTERPOLATE_LOC_CENTROID
) {
250 /* SPI_BARYC_CNTL.POS_FLOAT_LOCATION
252 * 0 -> Position = pixel center (default)
253 * 1 -> Position = pixel centroid
254 * 2 -> Position = iterated sample number XXX:
255 * What does this mean?
257 spi_baryc_cntl
|= S_0286E0_POS_FLOAT_LOCATION(1);
260 case TGSI_SEMANTIC_FACE
:
265 spi_ps_in_control
= S_0286D8_NUM_INTERP(shader
->nparam
) |
266 S_0286D8_BC_OPTIMIZE_DISABLE(1);
268 si_pm4_set_reg(pm4
, R_0286E0_SPI_BARYC_CNTL
, spi_baryc_cntl
);
269 spi_ps_input_ena
= shader
->spi_ps_input_ena
;
270 /* we need to enable at least one of them, otherwise we hang the GPU */
271 assert(G_0286CC_PERSP_SAMPLE_ENA(spi_ps_input_ena
) ||
272 G_0286CC_PERSP_CENTER_ENA(spi_ps_input_ena
) ||
273 G_0286CC_PERSP_CENTROID_ENA(spi_ps_input_ena
) ||
274 G_0286CC_PERSP_PULL_MODEL_ENA(spi_ps_input_ena
) ||
275 G_0286CC_LINEAR_SAMPLE_ENA(spi_ps_input_ena
) ||
276 G_0286CC_LINEAR_CENTER_ENA(spi_ps_input_ena
) ||
277 G_0286CC_LINEAR_CENTROID_ENA(spi_ps_input_ena
) ||
278 G_0286CC_LINE_STIPPLE_TEX_ENA(spi_ps_input_ena
));
280 si_pm4_set_reg(pm4
, R_0286CC_SPI_PS_INPUT_ENA
, spi_ps_input_ena
);
281 si_pm4_set_reg(pm4
, R_0286D0_SPI_PS_INPUT_ADDR
, spi_ps_input_ena
);
282 si_pm4_set_reg(pm4
, R_0286D8_SPI_PS_IN_CONTROL
, spi_ps_in_control
);
284 si_pm4_set_reg(pm4
, R_028710_SPI_SHADER_Z_FORMAT
, shader
->spi_shader_z_format
);
285 si_pm4_set_reg(pm4
, R_028714_SPI_SHADER_COL_FORMAT
,
286 shader
->spi_shader_col_format
);
287 si_pm4_set_reg(pm4
, R_02823C_CB_SHADER_MASK
, shader
->cb_shader_mask
);
289 va
= shader
->bo
->gpu_address
;
290 si_pm4_add_bo(pm4
, shader
->bo
, RADEON_USAGE_READ
, RADEON_PRIO_SHADER_DATA
);
291 si_pm4_set_reg(pm4
, R_00B020_SPI_SHADER_PGM_LO_PS
, va
>> 8);
292 si_pm4_set_reg(pm4
, R_00B024_SPI_SHADER_PGM_HI_PS
, va
>> 40);
294 num_user_sgprs
= SI_PS_NUM_USER_SGPR
;
295 num_sgprs
= shader
->num_sgprs
;
296 /* One SGPR after user SGPRs is pre-loaded with {prim_mask, lds_offset} */
297 if ((num_user_sgprs
+ 1) > num_sgprs
) {
298 /* Last 2 reserved SGPRs are used for VCC */
299 num_sgprs
= num_user_sgprs
+ 1 + 2;
301 assert(num_sgprs
<= 104);
303 si_pm4_set_reg(pm4
, R_00B028_SPI_SHADER_PGM_RSRC1_PS
,
304 S_00B028_VGPRS((shader
->num_vgprs
- 1) / 4) |
305 S_00B028_SGPRS((num_sgprs
- 1) / 8));
306 si_pm4_set_reg(pm4
, R_00B02C_SPI_SHADER_PGM_RSRC2_PS
,
307 S_00B02C_EXTRA_LDS_SIZE(shader
->lds_size
) |
308 S_00B02C_USER_SGPR(num_user_sgprs
));
315 static unsigned si_conv_pipe_prim(unsigned pprim
)
317 static const unsigned prim_conv
[] = {
318 [PIPE_PRIM_POINTS
] = V_008958_DI_PT_POINTLIST
,
319 [PIPE_PRIM_LINES
] = V_008958_DI_PT_LINELIST
,
320 [PIPE_PRIM_LINE_LOOP
] = V_008958_DI_PT_LINELOOP
,
321 [PIPE_PRIM_LINE_STRIP
] = V_008958_DI_PT_LINESTRIP
,
322 [PIPE_PRIM_TRIANGLES
] = V_008958_DI_PT_TRILIST
,
323 [PIPE_PRIM_TRIANGLE_STRIP
] = V_008958_DI_PT_TRISTRIP
,
324 [PIPE_PRIM_TRIANGLE_FAN
] = V_008958_DI_PT_TRIFAN
,
325 [PIPE_PRIM_QUADS
] = V_008958_DI_PT_QUADLIST
,
326 [PIPE_PRIM_QUAD_STRIP
] = V_008958_DI_PT_QUADSTRIP
,
327 [PIPE_PRIM_POLYGON
] = V_008958_DI_PT_POLYGON
,
328 [PIPE_PRIM_LINES_ADJACENCY
] = V_008958_DI_PT_LINELIST_ADJ
,
329 [PIPE_PRIM_LINE_STRIP_ADJACENCY
] = V_008958_DI_PT_LINESTRIP_ADJ
,
330 [PIPE_PRIM_TRIANGLES_ADJACENCY
] = V_008958_DI_PT_TRILIST_ADJ
,
331 [PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY
] = V_008958_DI_PT_TRISTRIP_ADJ
,
332 [R600_PRIM_RECTANGLE_LIST
] = V_008958_DI_PT_RECTLIST
334 unsigned result
= prim_conv
[pprim
];
336 R600_ERR("unsupported primitive type %d\n", pprim
);
341 static unsigned si_conv_prim_to_gs_out(unsigned mode
)
343 static const int prim_conv
[] = {
344 [PIPE_PRIM_POINTS
] = V_028A6C_OUTPRIM_TYPE_POINTLIST
,
345 [PIPE_PRIM_LINES
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
346 [PIPE_PRIM_LINE_LOOP
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
347 [PIPE_PRIM_LINE_STRIP
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
348 [PIPE_PRIM_TRIANGLES
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
349 [PIPE_PRIM_TRIANGLE_STRIP
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
350 [PIPE_PRIM_TRIANGLE_FAN
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
351 [PIPE_PRIM_QUADS
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
352 [PIPE_PRIM_QUAD_STRIP
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
353 [PIPE_PRIM_POLYGON
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
354 [PIPE_PRIM_LINES_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
355 [PIPE_PRIM_LINE_STRIP_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
356 [PIPE_PRIM_TRIANGLES_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
357 [PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
358 [R600_PRIM_RECTANGLE_LIST
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
360 assert(mode
< Elements(prim_conv
));
362 return prim_conv
[mode
];
365 static unsigned si_get_ia_multi_vgt_param(struct si_context
*sctx
,
366 const struct pipe_draw_info
*info
)
368 struct si_state_rasterizer
*rs
= sctx
->queued
.named
.rasterizer
;
369 unsigned prim
= info
->mode
;
370 unsigned primgroup_size
= 128; /* recommended without a GS */
372 /* SWITCH_ON_EOP(0) is always preferable. */
373 bool wd_switch_on_eop
= false;
374 bool ia_switch_on_eop
= false;
375 bool partial_vs_wave
= false;
378 primgroup_size
= 64; /* recommended with a GS */
380 /* This is a hardware requirement. */
381 if ((rs
&& rs
->line_stipple_enable
) ||
382 (sctx
->b
.screen
->debug_flags
& DBG_SWITCH_ON_EOP
)) {
383 ia_switch_on_eop
= true;
384 wd_switch_on_eop
= true;
387 if (sctx
->b
.streamout
.streamout_enabled
||
388 sctx
->b
.streamout
.prims_gen_query_enabled
)
389 partial_vs_wave
= true;
391 if (sctx
->b
.chip_class
>= CIK
) {
392 /* WD_SWITCH_ON_EOP has no effect on GPUs with less than
393 * 4 shader engines. Set 1 to pass the assertion below.
394 * The other cases are hardware requirements. */
395 if (sctx
->b
.screen
->info
.max_se
< 4 ||
396 prim
== PIPE_PRIM_POLYGON
||
397 prim
== PIPE_PRIM_LINE_LOOP
||
398 prim
== PIPE_PRIM_TRIANGLE_FAN
||
399 prim
== PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY
||
400 info
->primitive_restart
)
401 wd_switch_on_eop
= true;
403 /* Hawaii hangs if instancing is enabled and WD_SWITCH_ON_EOP is 0.
404 * We don't know that for indirect drawing, so treat it as
405 * always problematic. */
406 if (sctx
->b
.family
== CHIP_HAWAII
&&
407 (info
->indirect
|| info
->instance_count
> 1))
408 wd_switch_on_eop
= true;
410 /* If the WD switch is false, the IA switch must be false too. */
411 assert(wd_switch_on_eop
|| !ia_switch_on_eop
);
414 return S_028AA8_SWITCH_ON_EOP(ia_switch_on_eop
) |
415 S_028AA8_PARTIAL_VS_WAVE_ON(partial_vs_wave
) |
416 S_028AA8_PRIMGROUP_SIZE(primgroup_size
- 1) |
417 S_028AA8_WD_SWITCH_ON_EOP(sctx
->b
.chip_class
>= CIK
? wd_switch_on_eop
: 0);
420 static bool si_update_draw_info_state(struct si_context
*sctx
,
421 const struct pipe_draw_info
*info
,
422 const struct pipe_index_buffer
*ib
)
424 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
425 struct si_shader
*vs
= si_get_vs_state(sctx
);
426 unsigned prim
= si_conv_pipe_prim(info
->mode
);
427 unsigned gs_out_prim
=
428 si_conv_prim_to_gs_out(sctx
->gs_shader
?
429 sctx
->gs_shader
->gs_output_prim
:
431 unsigned ls_mask
= 0;
432 unsigned ia_multi_vgt_param
= si_get_ia_multi_vgt_param(sctx
, info
);
442 if (sctx
->b
.chip_class
>= CIK
) {
443 si_pm4_set_reg(pm4
, R_028B74_VGT_DISPATCH_DRAW_INDEX
,
444 ib
->index_size
== 4 ? 0xFC000000 : 0xFC00);
446 si_pm4_cmd_begin(pm4
, PKT3_DRAW_PREAMBLE
);
447 si_pm4_cmd_add(pm4
, prim
); /* VGT_PRIMITIVE_TYPE */
448 si_pm4_cmd_add(pm4
, ia_multi_vgt_param
); /* IA_MULTI_VGT_PARAM */
449 si_pm4_cmd_add(pm4
, 0); /* VGT_LS_HS_CONFIG */
450 si_pm4_cmd_end(pm4
, false);
452 si_pm4_set_reg(pm4
, R_008958_VGT_PRIMITIVE_TYPE
, prim
);
453 si_pm4_set_reg(pm4
, R_028AA8_IA_MULTI_VGT_PARAM
, ia_multi_vgt_param
);
456 si_pm4_set_reg(pm4
, R_028A6C_VGT_GS_OUT_PRIM_TYPE
, gs_out_prim
);
457 si_pm4_set_reg(pm4
, R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX
, info
->restart_index
);
458 si_pm4_set_reg(pm4
, R_028A94_VGT_MULTI_PRIM_IB_RESET_EN
, info
->primitive_restart
);
460 if (prim
== V_008958_DI_PT_LINELIST
)
462 else if (prim
== V_008958_DI_PT_LINESTRIP
)
464 si_pm4_set_reg(pm4
, R_028A0C_PA_SC_LINE_STIPPLE
,
465 S_028A0C_AUTO_RESET_CNTL(ls_mask
) |
466 sctx
->pa_sc_line_stipple
);
468 if (info
->mode
== PIPE_PRIM_QUADS
|| info
->mode
== PIPE_PRIM_QUAD_STRIP
|| info
->mode
== PIPE_PRIM_POLYGON
) {
469 si_pm4_set_reg(pm4
, R_028814_PA_SU_SC_MODE_CNTL
,
470 S_028814_PROVOKING_VTX_LAST(1) | sctx
->pa_su_sc_mode_cntl
);
472 si_pm4_set_reg(pm4
, R_028814_PA_SU_SC_MODE_CNTL
, sctx
->pa_su_sc_mode_cntl
);
474 si_pm4_set_reg(pm4
, R_02881C_PA_CL_VS_OUT_CNTL
,
475 S_02881C_USE_VTX_POINT_SIZE(vs
->vs_out_point_size
) |
476 S_02881C_USE_VTX_EDGE_FLAG(vs
->vs_out_edgeflag
) |
477 S_02881C_USE_VTX_RENDER_TARGET_INDX(vs
->vs_out_layer
) |
478 S_02881C_VS_OUT_CCDIST0_VEC_ENA((vs
->clip_dist_write
& 0x0F) != 0) |
479 S_02881C_VS_OUT_CCDIST1_VEC_ENA((vs
->clip_dist_write
& 0xF0) != 0) |
480 S_02881C_VS_OUT_MISC_VEC_ENA(vs
->vs_out_misc_write
) |
481 (sctx
->queued
.named
.rasterizer
->clip_plane_enable
&
482 vs
->clip_dist_write
));
483 si_pm4_set_reg(pm4
, R_028810_PA_CL_CLIP_CNTL
,
484 sctx
->queued
.named
.rasterizer
->pa_cl_clip_cntl
|
485 (vs
->clip_dist_write
? 0 :
486 sctx
->queued
.named
.rasterizer
->clip_plane_enable
& 0x3F));
488 si_pm4_set_state(sctx
, draw_info
, pm4
);
492 static void si_update_spi_map(struct si_context
*sctx
)
494 struct si_shader
*ps
= sctx
->ps_shader
->current
;
495 struct si_shader
*vs
= si_get_vs_state(sctx
);
496 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
499 for (i
= 0; i
< ps
->ninput
; i
++) {
500 unsigned name
= ps
->input
[i
].name
;
501 unsigned param_offset
= ps
->input
[i
].param_offset
;
503 if (name
== TGSI_SEMANTIC_POSITION
)
504 /* Read from preloaded VGPRs, not parameters */
510 if (ps
->input
[i
].interpolate
== TGSI_INTERPOLATE_CONSTANT
||
511 (ps
->input
[i
].interpolate
== TGSI_INTERPOLATE_COLOR
&&
512 ps
->key
.ps
.flatshade
)) {
513 tmp
|= S_028644_FLAT_SHADE(1);
516 if (name
== TGSI_SEMANTIC_GENERIC
&&
517 sctx
->sprite_coord_enable
& (1 << ps
->input
[i
].sid
)) {
518 tmp
|= S_028644_PT_SPRITE_TEX(1);
521 for (j
= 0; j
< vs
->noutput
; j
++) {
522 if (name
== vs
->output
[j
].name
&&
523 ps
->input
[i
].sid
== vs
->output
[j
].sid
) {
524 tmp
|= S_028644_OFFSET(vs
->output
[j
].param_offset
);
529 if (j
== vs
->noutput
) {
530 /* No corresponding output found, load defaults into input */
531 tmp
|= S_028644_OFFSET(0x20);
535 R_028644_SPI_PS_INPUT_CNTL_0
+ param_offset
* 4,
538 if (name
== TGSI_SEMANTIC_COLOR
&&
539 ps
->key
.ps
.color_two_side
) {
540 name
= TGSI_SEMANTIC_BCOLOR
;
546 si_pm4_set_state(sctx
, spi
, pm4
);
549 /* Initialize state related to ESGS / GSVS ring buffers */
550 static void si_init_gs_rings(struct si_context
*sctx
)
552 unsigned esgs_ring_size
= 128 * 1024;
553 unsigned gsvs_ring_size
= 64 * 1024 * 1024;
555 assert(!sctx
->gs_rings
);
556 sctx
->gs_rings
= si_pm4_alloc_state(sctx
);
558 sctx
->esgs_ring
= pipe_buffer_create(sctx
->b
.b
.screen
, PIPE_BIND_CUSTOM
,
559 PIPE_USAGE_DEFAULT
, esgs_ring_size
);
561 sctx
->gsvs_ring
= pipe_buffer_create(sctx
->b
.b
.screen
, PIPE_BIND_CUSTOM
,
562 PIPE_USAGE_DEFAULT
, gsvs_ring_size
);
564 if (sctx
->b
.chip_class
>= CIK
) {
565 si_pm4_set_reg(sctx
->gs_rings
, R_030900_VGT_ESGS_RING_SIZE
,
566 esgs_ring_size
/ 256);
567 si_pm4_set_reg(sctx
->gs_rings
, R_030904_VGT_GSVS_RING_SIZE
,
568 gsvs_ring_size
/ 256);
570 si_pm4_set_reg(sctx
->gs_rings
, R_0088C8_VGT_ESGS_RING_SIZE
,
571 esgs_ring_size
/ 256);
572 si_pm4_set_reg(sctx
->gs_rings
, R_0088CC_VGT_GSVS_RING_SIZE
,
573 gsvs_ring_size
/ 256);
576 si_set_ring_buffer(&sctx
->b
.b
, PIPE_SHADER_VERTEX
, SI_RING_ESGS
,
577 sctx
->esgs_ring
, 0, esgs_ring_size
,
579 si_set_ring_buffer(&sctx
->b
.b
, PIPE_SHADER_GEOMETRY
, SI_RING_ESGS
,
580 sctx
->esgs_ring
, 0, esgs_ring_size
,
582 si_set_ring_buffer(&sctx
->b
.b
, PIPE_SHADER_VERTEX
, SI_RING_GSVS
,
583 sctx
->gsvs_ring
, 0, gsvs_ring_size
,
587 static void si_update_derived_state(struct si_context
*sctx
)
589 struct pipe_context
* ctx
= (struct pipe_context
*)sctx
;
591 if (!sctx
->blitter
->running
) {
592 /* Flush depth textures which need to be flushed. */
593 for (int i
= 0; i
< SI_NUM_SHADERS
; i
++) {
594 if (sctx
->samplers
[i
].depth_texture_mask
) {
595 si_flush_depth_textures(sctx
, &sctx
->samplers
[i
]);
597 if (sctx
->samplers
[i
].compressed_colortex_mask
) {
598 si_decompress_color_textures(sctx
, &sctx
->samplers
[i
]);
603 if (sctx
->gs_shader
) {
604 si_shader_select(ctx
, sctx
->gs_shader
);
606 if (!sctx
->gs_shader
->current
->pm4
) {
607 si_shader_gs(ctx
, sctx
->gs_shader
->current
);
608 si_shader_vs(ctx
, sctx
->gs_shader
->current
->gs_copy_shader
);
611 si_pm4_bind_state(sctx
, gs
, sctx
->gs_shader
->current
->pm4
);
612 si_pm4_bind_state(sctx
, vs
, sctx
->gs_shader
->current
->gs_copy_shader
->pm4
);
614 sctx
->b
.streamout
.stride_in_dw
= sctx
->gs_shader
->so
.stride
;
616 si_shader_select(ctx
, sctx
->vs_shader
);
618 if (!sctx
->vs_shader
->current
->pm4
)
619 si_shader_es(ctx
, sctx
->vs_shader
->current
);
621 si_pm4_bind_state(sctx
, es
, sctx
->vs_shader
->current
->pm4
);
624 si_init_gs_rings(sctx
);
625 if (sctx
->emitted
.named
.gs_rings
!= sctx
->gs_rings
)
626 sctx
->b
.flags
|= R600_CONTEXT_VGT_FLUSH
;
627 si_pm4_bind_state(sctx
, gs_rings
, sctx
->gs_rings
);
629 si_set_ring_buffer(ctx
, PIPE_SHADER_GEOMETRY
, SI_RING_GSVS
,
631 sctx
->gs_shader
->gs_max_out_vertices
*
632 sctx
->gs_shader
->current
->noutput
* 16,
633 64, true, true, 4, 16);
636 sctx
->gs_on
= si_pm4_alloc_state(sctx
);
638 si_pm4_set_reg(sctx
->gs_on
, R_028B54_VGT_SHADER_STAGES_EN
,
639 S_028B54_ES_EN(V_028B54_ES_STAGE_REAL
) |
641 S_028B54_VS_EN(V_028B54_VS_STAGE_COPY_SHADER
));
643 si_pm4_bind_state(sctx
, gs_onoff
, sctx
->gs_on
);
645 si_shader_select(ctx
, sctx
->vs_shader
);
647 if (!sctx
->vs_shader
->current
->pm4
)
648 si_shader_vs(ctx
, sctx
->vs_shader
->current
);
650 si_pm4_bind_state(sctx
, vs
, sctx
->vs_shader
->current
->pm4
);
652 sctx
->b
.streamout
.stride_in_dw
= sctx
->vs_shader
->so
.stride
;
655 sctx
->gs_off
= si_pm4_alloc_state(sctx
);
657 si_pm4_set_reg(sctx
->gs_off
, R_028A40_VGT_GS_MODE
, 0);
658 si_pm4_set_reg(sctx
->gs_off
, R_028B54_VGT_SHADER_STAGES_EN
, 0);
660 si_pm4_bind_state(sctx
, gs_onoff
, sctx
->gs_off
);
661 si_pm4_bind_state(sctx
, gs_rings
, NULL
);
662 si_pm4_bind_state(sctx
, gs
, NULL
);
663 si_pm4_bind_state(sctx
, es
, NULL
);
666 si_shader_select(ctx
, sctx
->ps_shader
);
668 if (!sctx
->ps_shader
->current
) {
669 struct si_shader_selector
*sel
;
671 /* use a dummy shader if compiling the shader (variant) failed */
672 si_make_dummy_ps(sctx
);
673 sel
= sctx
->dummy_pixel_shader
;
674 si_shader_select(ctx
, sel
);
675 sctx
->ps_shader
->current
= sel
->current
;
678 if (!sctx
->ps_shader
->current
->pm4
)
679 si_shader_ps(ctx
, sctx
->ps_shader
->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 /* XXX: Emitting the PS state even when only the VS changed
685 * fixes random failures with piglit glsl-max-varyings.
688 sctx
->emitted
.named
.ps
= NULL
;
689 si_update_spi_map(sctx
);
692 if (sctx
->ps_db_shader_control
!= sctx
->ps_shader
->current
->db_shader_control
) {
693 sctx
->ps_db_shader_control
= sctx
->ps_shader
->current
->db_shader_control
;
694 sctx
->db_render_state
.dirty
= true;
698 static void si_state_draw(struct si_context
*sctx
,
699 const struct pipe_draw_info
*info
,
700 const struct pipe_index_buffer
*ib
)
702 unsigned sh_base_reg
= (sctx
->gs_shader
? R_00B330_SPI_SHADER_USER_DATA_ES_0
:
703 R_00B130_SPI_SHADER_USER_DATA_VS_0
);
704 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
709 if (info
->count_from_stream_output
) {
710 struct r600_so_target
*t
=
711 (struct r600_so_target
*)info
->count_from_stream_output
;
712 uint64_t va
= t
->buf_filled_size
->gpu_address
+
713 t
->buf_filled_size_offset
;
715 si_pm4_set_reg(pm4
, R_028B30_VGT_STRMOUT_DRAW_OPAQUE_VERTEX_STRIDE
,
718 si_pm4_cmd_begin(pm4
, PKT3_COPY_DATA
);
720 COPY_DATA_SRC_SEL(COPY_DATA_MEM
) |
721 COPY_DATA_DST_SEL(COPY_DATA_REG
) |
722 COPY_DATA_WR_CONFIRM
);
723 si_pm4_cmd_add(pm4
, va
); /* src address lo */
724 si_pm4_cmd_add(pm4
, va
>> 32UL); /* src address hi */
725 si_pm4_cmd_add(pm4
, R_028B2C_VGT_STRMOUT_DRAW_OPAQUE_BUFFER_FILLED_SIZE
>> 2);
726 si_pm4_cmd_add(pm4
, 0); /* unused */
727 si_pm4_add_bo(pm4
, t
->buf_filled_size
, RADEON_USAGE_READ
,
729 si_pm4_cmd_end(pm4
, true);
733 si_pm4_cmd_begin(pm4
, PKT3_INDEX_TYPE
);
734 if (ib
->index_size
== 4) {
735 si_pm4_cmd_add(pm4
, V_028A7C_VGT_INDEX_32
| (SI_BIG_ENDIAN
?
736 V_028A7C_VGT_DMA_SWAP_32_BIT
: 0));
738 si_pm4_cmd_add(pm4
, V_028A7C_VGT_INDEX_16
| (SI_BIG_ENDIAN
?
739 V_028A7C_VGT_DMA_SWAP_16_BIT
: 0));
741 si_pm4_cmd_end(pm4
, sctx
->b
.predicate_drawing
);
743 if (!info
->indirect
) {
744 si_pm4_cmd_begin(pm4
, PKT3_NUM_INSTANCES
);
745 si_pm4_cmd_add(pm4
, info
->instance_count
);
746 si_pm4_cmd_end(pm4
, sctx
->b
.predicate_drawing
);
748 si_pm4_set_reg(pm4
, sh_base_reg
+ SI_SGPR_BASE_VERTEX
* 4,
749 info
->indexed
? info
->index_bias
: info
->start
);
750 si_pm4_set_reg(pm4
, sh_base_reg
+ SI_SGPR_START_INSTANCE
* 4,
751 info
->start_instance
);
753 si_pm4_add_bo(pm4
, (struct r600_resource
*)info
->indirect
,
754 RADEON_USAGE_READ
, RADEON_PRIO_MIN
);
758 uint32_t max_size
= (ib
->buffer
->width0
- ib
->offset
) /
759 sctx
->index_buffer
.index_size
;
760 uint64_t va
= r600_resource(ib
->buffer
)->gpu_address
+ ib
->offset
;
762 si_pm4_add_bo(pm4
, (struct r600_resource
*)ib
->buffer
, RADEON_USAGE_READ
,
765 if (info
->indirect
) {
766 uint64_t indirect_va
= r600_resource(info
->indirect
)->gpu_address
;
767 si_cmd_draw_index_indirect(pm4
, indirect_va
, va
, max_size
,
768 info
->indirect_offset
,
769 sh_base_reg
+ SI_SGPR_BASE_VERTEX
* 4,
770 sh_base_reg
+ SI_SGPR_START_INSTANCE
* 4,
771 sctx
->b
.predicate_drawing
);
773 va
+= info
->start
* ib
->index_size
;
774 si_cmd_draw_index_2(pm4
, max_size
, va
, info
->count
,
775 V_0287F0_DI_SRC_SEL_DMA
,
776 sctx
->b
.predicate_drawing
);
779 if (info
->indirect
) {
780 uint64_t indirect_va
= r600_resource(info
->indirect
)->gpu_address
;
781 si_cmd_draw_indirect(pm4
, indirect_va
, info
->indirect_offset
,
782 sh_base_reg
+ SI_SGPR_BASE_VERTEX
* 4,
783 sh_base_reg
+ SI_SGPR_START_INSTANCE
* 4,
784 sctx
->b
.predicate_drawing
);
786 si_cmd_draw_index_auto(pm4
, info
->count
,
787 V_0287F0_DI_SRC_SEL_AUTO_INDEX
|
788 S_0287F0_USE_OPAQUE(!!info
->count_from_stream_output
),
789 sctx
->b
.predicate_drawing
);
793 si_pm4_set_state(sctx
, draw
, pm4
);
796 void si_emit_cache_flush(struct r600_common_context
*sctx
, struct r600_atom
*atom
)
798 struct radeon_winsys_cs
*cs
= sctx
->rings
.gfx
.cs
;
799 uint32_t cp_coher_cntl
= 0;
801 PKT3_SHADER_TYPE_S(!!(sctx
->flags
& R600_CONTEXT_FLAG_COMPUTE
));
803 /* XXX SI flushes both ICACHE and KCACHE if either flag is set.
804 * XXX CIK shouldn't have this issue. Test CIK before separating the flags
805 * XXX to ensure there is no regression. Also find out if there is another
806 * XXX way to flush either ICACHE or KCACHE but not both for SI. */
807 if (sctx
->flags
& (R600_CONTEXT_INV_SHADER_CACHE
|
808 R600_CONTEXT_INV_CONST_CACHE
)) {
809 cp_coher_cntl
|= S_0085F0_SH_ICACHE_ACTION_ENA(1) |
810 S_0085F0_SH_KCACHE_ACTION_ENA(1);
812 if (sctx
->flags
& (R600_CONTEXT_INV_TEX_CACHE
|
813 R600_CONTEXT_STREAMOUT_FLUSH
)) {
814 cp_coher_cntl
|= S_0085F0_TC_ACTION_ENA(1) |
815 S_0085F0_TCL1_ACTION_ENA(1);
817 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_CB
) {
818 cp_coher_cntl
|= S_0085F0_CB_ACTION_ENA(1) |
819 S_0085F0_CB0_DEST_BASE_ENA(1) |
820 S_0085F0_CB1_DEST_BASE_ENA(1) |
821 S_0085F0_CB2_DEST_BASE_ENA(1) |
822 S_0085F0_CB3_DEST_BASE_ENA(1) |
823 S_0085F0_CB4_DEST_BASE_ENA(1) |
824 S_0085F0_CB5_DEST_BASE_ENA(1) |
825 S_0085F0_CB6_DEST_BASE_ENA(1) |
826 S_0085F0_CB7_DEST_BASE_ENA(1);
828 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_DB
) {
829 cp_coher_cntl
|= S_0085F0_DB_ACTION_ENA(1) |
830 S_0085F0_DB_DEST_BASE_ENA(1);
834 if (sctx
->chip_class
>= CIK
) {
835 radeon_emit(cs
, PKT3(PKT3_ACQUIRE_MEM
, 5, 0) | compute
);
836 radeon_emit(cs
, cp_coher_cntl
); /* CP_COHER_CNTL */
837 radeon_emit(cs
, 0xffffffff); /* CP_COHER_SIZE */
838 radeon_emit(cs
, 0xff); /* CP_COHER_SIZE_HI */
839 radeon_emit(cs
, 0); /* CP_COHER_BASE */
840 radeon_emit(cs
, 0); /* CP_COHER_BASE_HI */
841 radeon_emit(cs
, 0x0000000A); /* POLL_INTERVAL */
843 radeon_emit(cs
, PKT3(PKT3_SURFACE_SYNC
, 3, 0) | compute
);
844 radeon_emit(cs
, cp_coher_cntl
); /* CP_COHER_CNTL */
845 radeon_emit(cs
, 0xffffffff); /* CP_COHER_SIZE */
846 radeon_emit(cs
, 0); /* CP_COHER_BASE */
847 radeon_emit(cs
, 0x0000000A); /* POLL_INTERVAL */
851 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_CB_META
) {
852 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
853 radeon_emit(cs
, EVENT_TYPE(V_028A90_FLUSH_AND_INV_CB_META
) | EVENT_INDEX(0));
855 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_DB_META
) {
856 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
857 radeon_emit(cs
, EVENT_TYPE(V_028A90_FLUSH_AND_INV_DB_META
) | EVENT_INDEX(0));
859 if (sctx
->flags
& R600_CONTEXT_FLUSH_WITH_INV_L2
) {
860 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
861 radeon_emit(cs
, EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH
) | EVENT_INDEX(7) |
865 if (sctx
->flags
& (R600_CONTEXT_WAIT_3D_IDLE
|
866 R600_CONTEXT_PS_PARTIAL_FLUSH
)) {
867 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
868 radeon_emit(cs
, EVENT_TYPE(V_028A90_PS_PARTIAL_FLUSH
) | EVENT_INDEX(4));
869 } else if (sctx
->flags
& R600_CONTEXT_STREAMOUT_FLUSH
) {
870 /* Needed if streamout buffers are going to be used as a source. */
871 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
872 radeon_emit(cs
, EVENT_TYPE(V_028A90_VS_PARTIAL_FLUSH
) | EVENT_INDEX(4));
875 if (sctx
->flags
& R600_CONTEXT_CS_PARTIAL_FLUSH
) {
876 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
877 radeon_emit(cs
, EVENT_TYPE(V_028A90_CS_PARTIAL_FLUSH
| EVENT_INDEX(4)));
880 if (sctx
->flags
& R600_CONTEXT_VGT_FLUSH
) {
881 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
882 radeon_emit(cs
, EVENT_TYPE(V_028A90_VGT_FLUSH
) | EVENT_INDEX(0));
884 if (sctx
->flags
& R600_CONTEXT_VGT_STREAMOUT_SYNC
) {
885 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0) | compute
);
886 radeon_emit(cs
, EVENT_TYPE(V_028A90_VGT_STREAMOUT_SYNC
) | EVENT_INDEX(0));
892 const struct r600_atom si_atom_cache_flush
= { si_emit_cache_flush
, 21 }; /* number of CS dwords */
894 static void si_get_draw_start_count(struct si_context
*sctx
,
895 const struct pipe_draw_info
*info
,
896 unsigned *start
, unsigned *count
)
898 if (info
->indirect
) {
899 struct r600_resource
*indirect
=
900 (struct r600_resource
*)info
->indirect
;
901 int *data
= r600_buffer_map_sync_with_rings(&sctx
->b
,
902 indirect
, PIPE_TRANSFER_READ
);
903 data
+= info
->indirect_offset
/sizeof(int);
907 *start
= info
->start
;
908 *count
= info
->count
;
912 void si_draw_vbo(struct pipe_context
*ctx
, const struct pipe_draw_info
*info
)
914 struct si_context
*sctx
= (struct si_context
*)ctx
;
915 struct pipe_index_buffer ib
= {};
918 if (!info
->count
&& !info
->indirect
&&
919 (info
->indexed
|| !info
->count_from_stream_output
))
922 if (!sctx
->ps_shader
|| !sctx
->vs_shader
)
925 si_update_derived_state(sctx
);
927 if (sctx
->vertex_buffers_dirty
) {
928 si_update_vertex_buffers(sctx
);
929 sctx
->vertex_buffers_dirty
= false;
933 /* Initialize the index buffer struct. */
934 pipe_resource_reference(&ib
.buffer
, sctx
->index_buffer
.buffer
);
935 ib
.user_buffer
= sctx
->index_buffer
.user_buffer
;
936 ib
.index_size
= sctx
->index_buffer
.index_size
;
937 ib
.offset
= sctx
->index_buffer
.offset
;
939 /* Translate or upload, if needed. */
940 if (ib
.index_size
== 1) {
941 struct pipe_resource
*out_buffer
= NULL
;
942 unsigned out_offset
, start
, count
, start_offset
;
945 si_get_draw_start_count(sctx
, info
, &start
, &count
);
946 start_offset
= start
* ib
.index_size
;
948 u_upload_alloc(sctx
->b
.uploader
, start_offset
, count
* 2,
949 &out_offset
, &out_buffer
, &ptr
);
951 util_shorten_ubyte_elts_to_userptr(&sctx
->b
.b
, &ib
, 0,
952 ib
.offset
+ start_offset
,
955 pipe_resource_reference(&ib
.buffer
, NULL
);
956 ib
.user_buffer
= NULL
;
957 ib
.buffer
= out_buffer
;
958 /* info->start will be added by the drawing code */
959 ib
.offset
= out_offset
- start_offset
;
961 } else if (ib
.user_buffer
&& !ib
.buffer
) {
962 unsigned start
, count
, start_offset
;
964 si_get_draw_start_count(sctx
, info
, &start
, &count
);
965 start_offset
= start
* ib
.index_size
;
967 u_upload_data(sctx
->b
.uploader
, start_offset
, count
* ib
.index_size
,
968 (char*)ib
.user_buffer
+ start_offset
,
969 &ib
.offset
, &ib
.buffer
);
970 /* info->start will be added by the drawing code */
971 ib
.offset
-= start_offset
;
975 if (!si_update_draw_info_state(sctx
, info
, &ib
))
978 si_state_draw(sctx
, info
, &ib
);
980 sctx
->pm4_dirty_cdwords
+= si_pm4_dirty_dw(sctx
);
982 /* Check flush flags. */
984 sctx
->atoms
.s
.cache_flush
->dirty
= true;
986 si_need_cs_space(sctx
, 0, TRUE
);
989 for (i
= 0; i
< SI_NUM_ATOMS(sctx
); i
++) {
990 if (sctx
->atoms
.array
[i
]->dirty
) {
991 sctx
->atoms
.array
[i
]->emit(&sctx
->b
, sctx
->atoms
.array
[i
]);
992 sctx
->atoms
.array
[i
]->dirty
= false;
996 si_pm4_emit_dirty(sctx
);
997 sctx
->pm4_dirty_cdwords
= 0;
1000 if (sctx
->screen
->b
.trace_bo
) {
1001 si_trace_emit(sctx
);
1005 /* Workaround for a VGT hang when streamout is enabled.
1006 * It must be done after drawing. */
1007 if (sctx
->b
.family
== CHIP_HAWAII
&&
1008 (sctx
->b
.streamout
.streamout_enabled
||
1009 sctx
->b
.streamout
.prims_gen_query_enabled
)) {
1010 sctx
->b
.flags
|= R600_CONTEXT_VGT_STREAMOUT_SYNC
;
1013 /* Set the depth buffer as dirty. */
1014 if (sctx
->framebuffer
.state
.zsbuf
) {
1015 struct pipe_surface
*surf
= sctx
->framebuffer
.state
.zsbuf
;
1016 struct r600_texture
*rtex
= (struct r600_texture
*)surf
->texture
;
1018 rtex
->dirty_level_mask
|= 1 << surf
->u
.tex
.level
;
1020 if (sctx
->framebuffer
.compressed_cb_mask
) {
1021 struct pipe_surface
*surf
;
1022 struct r600_texture
*rtex
;
1023 unsigned mask
= sctx
->framebuffer
.compressed_cb_mask
;
1026 unsigned i
= u_bit_scan(&mask
);
1027 surf
= sctx
->framebuffer
.state
.cbufs
[i
];
1028 rtex
= (struct r600_texture
*)surf
->texture
;
1030 rtex
->dirty_level_mask
|= 1 << surf
->u
.tex
.level
;
1034 pipe_resource_reference(&ib
.buffer
, NULL
);
1035 sctx
->b
.num_draw_calls
++;
1039 void si_trace_emit(struct si_context
*sctx
)
1041 struct si_screen
*sscreen
= sctx
->screen
;
1042 struct radeon_winsys_cs
*cs
= sctx
->b
.rings
.gfx
.cs
;
1045 va
= sscreen
->b
.trace_bo
->gpu_address
;
1046 r600_context_bo_reloc(&sctx
->b
, &sctx
->b
.rings
.gfx
, sscreen
->b
.trace_bo
,
1047 RADEON_USAGE_READWRITE
, RADEON_PRIO_MIN
);
1048 radeon_emit(cs
, PKT3(PKT3_WRITE_DATA
, 4, 0));
1049 radeon_emit(cs
, PKT3_WRITE_DATA_DST_SEL(PKT3_WRITE_DATA_DST_SEL_MEM_SYNC
) |
1050 PKT3_WRITE_DATA_WR_CONFIRM
|
1051 PKT3_WRITE_DATA_ENGINE_SEL(PKT3_WRITE_DATA_ENGINE_SEL_ME
));
1052 radeon_emit(cs
, va
& 0xFFFFFFFFUL
);
1053 radeon_emit(cs
, (va
>> 32UL) & 0xFFFFFFFFUL
);
1054 radeon_emit(cs
, cs
->cdw
);
1055 radeon_emit(cs
, sscreen
->b
.cs_count
);