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_blitter.h"
33 #include "util/u_format.h"
34 #include "util/u_index_modify.h"
35 #include "util/u_memory.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
= r600_resource_va(ctx
->screen
, (void *)shader
->bo
);
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
= r600_resource_va(ctx
->screen
, (void *)shader
->bo
);
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
= r600_resource_va(ctx
->screen
, (void *)shader
->bo
);
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
= r600_resource_va(ctx
->screen
, (void *)shader
->bo
);
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 shader
->sprite_coord_enable
= sctx
->sprite_coord_enable
;
325 sctx
->b
.flags
|= R600_CONTEXT_INV_SHADER_CACHE
;
332 static unsigned si_conv_pipe_prim(unsigned pprim
)
334 static const unsigned prim_conv
[] = {
335 [PIPE_PRIM_POINTS
] = V_008958_DI_PT_POINTLIST
,
336 [PIPE_PRIM_LINES
] = V_008958_DI_PT_LINELIST
,
337 [PIPE_PRIM_LINE_LOOP
] = V_008958_DI_PT_LINELOOP
,
338 [PIPE_PRIM_LINE_STRIP
] = V_008958_DI_PT_LINESTRIP
,
339 [PIPE_PRIM_TRIANGLES
] = V_008958_DI_PT_TRILIST
,
340 [PIPE_PRIM_TRIANGLE_STRIP
] = V_008958_DI_PT_TRISTRIP
,
341 [PIPE_PRIM_TRIANGLE_FAN
] = V_008958_DI_PT_TRIFAN
,
342 [PIPE_PRIM_QUADS
] = V_008958_DI_PT_QUADLIST
,
343 [PIPE_PRIM_QUAD_STRIP
] = V_008958_DI_PT_QUADSTRIP
,
344 [PIPE_PRIM_POLYGON
] = V_008958_DI_PT_POLYGON
,
345 [PIPE_PRIM_LINES_ADJACENCY
] = V_008958_DI_PT_LINELIST_ADJ
,
346 [PIPE_PRIM_LINE_STRIP_ADJACENCY
] = V_008958_DI_PT_LINESTRIP_ADJ
,
347 [PIPE_PRIM_TRIANGLES_ADJACENCY
] = V_008958_DI_PT_TRILIST_ADJ
,
348 [PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY
] = V_008958_DI_PT_TRISTRIP_ADJ
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
375 assert(mode
< Elements(prim_conv
));
377 return prim_conv
[mode
];
380 static bool si_update_draw_info_state(struct si_context
*sctx
,
381 const struct pipe_draw_info
*info
,
382 const struct pipe_index_buffer
*ib
)
384 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
385 struct si_shader
*vs
= si_get_vs_state(sctx
);
386 unsigned prim
= si_conv_pipe_prim(info
->mode
);
387 unsigned gs_out_prim
=
388 si_conv_prim_to_gs_out(sctx
->gs_shader
?
389 sctx
->gs_shader
->current
->shader
.gs_output_prim
:
391 unsigned ls_mask
= 0;
401 if (sctx
->b
.chip_class
>= CIK
) {
402 struct si_state_rasterizer
*rs
= sctx
->queued
.named
.rasterizer
;
403 bool wd_switch_on_eop
= prim
== V_008958_DI_PT_POLYGON
||
404 prim
== V_008958_DI_PT_LINELOOP
||
405 prim
== V_008958_DI_PT_TRIFAN
||
406 prim
== V_008958_DI_PT_TRISTRIP_ADJ
||
407 info
->primitive_restart
||
408 (rs
? rs
->line_stipple_enable
: false);
409 /* If the WD switch is false, the IA switch must be false too. */
410 bool ia_switch_on_eop
= wd_switch_on_eop
;
412 si_pm4_set_reg(pm4
, R_028B74_VGT_DISPATCH_DRAW_INDEX
,
413 ib
->index_size
== 4 ? 0xFC000000 : 0xFC00);
415 si_pm4_cmd_begin(pm4
, PKT3_DRAW_PREAMBLE
);
416 si_pm4_cmd_add(pm4
, prim
); /* VGT_PRIMITIVE_TYPE */
417 si_pm4_cmd_add(pm4
, /* IA_MULTI_VGT_PARAM */
418 S_028AA8_SWITCH_ON_EOP(ia_switch_on_eop
) |
419 S_028AA8_PARTIAL_VS_WAVE_ON(1) |
420 S_028AA8_PRIMGROUP_SIZE(63) |
421 S_028AA8_WD_SWITCH_ON_EOP(wd_switch_on_eop
));
422 si_pm4_cmd_add(pm4
, 0); /* VGT_LS_HS_CONFIG */
423 si_pm4_cmd_end(pm4
, false);
425 si_pm4_set_reg(pm4
, R_008958_VGT_PRIMITIVE_TYPE
, prim
);
428 si_pm4_set_reg(pm4
, R_028A6C_VGT_GS_OUT_PRIM_TYPE
, gs_out_prim
);
429 si_pm4_set_reg(pm4
, R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX
, info
->restart_index
);
430 si_pm4_set_reg(pm4
, R_028A94_VGT_MULTI_PRIM_IB_RESET_EN
, info
->primitive_restart
);
432 if (prim
== V_008958_DI_PT_LINELIST
)
434 else if (prim
== V_008958_DI_PT_LINESTRIP
)
436 si_pm4_set_reg(pm4
, R_028A0C_PA_SC_LINE_STIPPLE
,
437 S_028A0C_AUTO_RESET_CNTL(ls_mask
) |
438 sctx
->pa_sc_line_stipple
);
440 if (info
->mode
== PIPE_PRIM_QUADS
|| info
->mode
== PIPE_PRIM_QUAD_STRIP
|| info
->mode
== PIPE_PRIM_POLYGON
) {
441 si_pm4_set_reg(pm4
, R_028814_PA_SU_SC_MODE_CNTL
,
442 S_028814_PROVOKING_VTX_LAST(1) | sctx
->pa_su_sc_mode_cntl
);
444 si_pm4_set_reg(pm4
, R_028814_PA_SU_SC_MODE_CNTL
, sctx
->pa_su_sc_mode_cntl
);
446 si_pm4_set_reg(pm4
, R_02881C_PA_CL_VS_OUT_CNTL
,
447 S_02881C_USE_VTX_POINT_SIZE(vs
->vs_out_point_size
) |
448 S_02881C_USE_VTX_EDGE_FLAG(vs
->vs_out_edgeflag
) |
449 S_02881C_USE_VTX_RENDER_TARGET_INDX(vs
->vs_out_layer
) |
450 S_02881C_VS_OUT_CCDIST0_VEC_ENA((vs
->clip_dist_write
& 0x0F) != 0) |
451 S_02881C_VS_OUT_CCDIST1_VEC_ENA((vs
->clip_dist_write
& 0xF0) != 0) |
452 S_02881C_VS_OUT_MISC_VEC_ENA(vs
->vs_out_misc_write
) |
453 (sctx
->queued
.named
.rasterizer
->clip_plane_enable
&
454 vs
->clip_dist_write
));
455 si_pm4_set_reg(pm4
, R_028810_PA_CL_CLIP_CNTL
,
456 sctx
->queued
.named
.rasterizer
->pa_cl_clip_cntl
|
457 (vs
->clip_dist_write
? 0 :
458 sctx
->queued
.named
.rasterizer
->clip_plane_enable
& 0x3F));
460 si_pm4_set_state(sctx
, draw_info
, pm4
);
464 static void si_update_spi_map(struct si_context
*sctx
)
466 struct si_shader
*ps
= &sctx
->ps_shader
->current
->shader
;
467 struct si_shader
*vs
= si_get_vs_state(sctx
);
468 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
471 for (i
= 0; i
< ps
->ninput
; i
++) {
472 unsigned name
= ps
->input
[i
].name
;
473 unsigned param_offset
= ps
->input
[i
].param_offset
;
475 if (name
== TGSI_SEMANTIC_POSITION
)
476 /* Read from preloaded VGPRs, not parameters */
482 if (ps
->input
[i
].interpolate
== TGSI_INTERPOLATE_CONSTANT
||
483 (ps
->input
[i
].interpolate
== TGSI_INTERPOLATE_COLOR
&&
484 sctx
->ps_shader
->current
->key
.ps
.flatshade
)) {
485 tmp
|= S_028644_FLAT_SHADE(1);
488 if (name
== TGSI_SEMANTIC_GENERIC
&&
489 sctx
->sprite_coord_enable
& (1 << ps
->input
[i
].sid
)) {
490 tmp
|= S_028644_PT_SPRITE_TEX(1);
493 for (j
= 0; j
< vs
->noutput
; j
++) {
494 if (name
== vs
->output
[j
].name
&&
495 ps
->input
[i
].sid
== vs
->output
[j
].sid
) {
496 tmp
|= S_028644_OFFSET(vs
->output
[j
].param_offset
);
501 if (j
== vs
->noutput
) {
502 /* No corresponding output found, load defaults into input */
503 tmp
|= S_028644_OFFSET(0x20);
507 R_028644_SPI_PS_INPUT_CNTL_0
+ param_offset
* 4,
510 if (name
== TGSI_SEMANTIC_COLOR
&&
511 sctx
->ps_shader
->current
->key
.ps
.color_two_side
) {
512 name
= TGSI_SEMANTIC_BCOLOR
;
518 si_pm4_set_state(sctx
, spi
, pm4
);
521 /* Initialize state related to ESGS / GSVS ring buffers */
522 static void si_init_gs_rings(struct si_context
*sctx
)
524 unsigned size
= 128 * 1024;
526 assert(!sctx
->gs_rings
);
527 sctx
->gs_rings
= si_pm4_alloc_state(sctx
);
529 sctx
->esgs_ring
.buffer
=
530 pipe_buffer_create(sctx
->b
.b
.screen
, PIPE_BIND_CUSTOM
,
531 PIPE_USAGE_DEFAULT
, size
);
532 sctx
->esgs_ring
.buffer_size
= size
;
534 size
= 64 * 1024 * 1024;
535 sctx
->gsvs_ring
.buffer
=
536 pipe_buffer_create(sctx
->b
.b
.screen
, PIPE_BIND_CUSTOM
,
537 PIPE_USAGE_DEFAULT
, size
);
538 sctx
->gsvs_ring
.buffer_size
= size
;
540 if (sctx
->b
.chip_class
>= CIK
) {
541 si_pm4_set_reg(sctx
->gs_rings
, R_030900_VGT_ESGS_RING_SIZE
,
542 sctx
->esgs_ring
.buffer_size
/ 256);
543 si_pm4_set_reg(sctx
->gs_rings
, R_030904_VGT_GSVS_RING_SIZE
,
544 sctx
->gsvs_ring
.buffer_size
/ 256);
546 si_pm4_set_reg(sctx
->gs_rings
, R_0088C8_VGT_ESGS_RING_SIZE
,
547 sctx
->esgs_ring
.buffer_size
/ 256);
548 si_pm4_set_reg(sctx
->gs_rings
, R_0088CC_VGT_GSVS_RING_SIZE
,
549 sctx
->gsvs_ring
.buffer_size
/ 256);
552 si_set_ring_buffer(&sctx
->b
.b
, PIPE_SHADER_VERTEX
, SI_RING_ESGS
,
553 &sctx
->esgs_ring
, 0, sctx
->esgs_ring
.buffer_size
,
555 si_set_ring_buffer(&sctx
->b
.b
, PIPE_SHADER_GEOMETRY
, SI_RING_ESGS
,
556 &sctx
->esgs_ring
, 0, sctx
->esgs_ring
.buffer_size
,
558 si_set_ring_buffer(&sctx
->b
.b
, PIPE_SHADER_VERTEX
, SI_RING_GSVS
,
559 &sctx
->gsvs_ring
, 0, sctx
->gsvs_ring
.buffer_size
,
563 static void si_update_derived_state(struct si_context
*sctx
)
565 struct pipe_context
* ctx
= (struct pipe_context
*)sctx
;
567 if (!sctx
->blitter
->running
) {
568 /* Flush depth textures which need to be flushed. */
569 for (int i
= 0; i
< SI_NUM_SHADERS
; i
++) {
570 if (sctx
->samplers
[i
].depth_texture_mask
) {
571 si_flush_depth_textures(sctx
, &sctx
->samplers
[i
]);
573 if (sctx
->samplers
[i
].compressed_colortex_mask
) {
574 si_decompress_color_textures(sctx
, &sctx
->samplers
[i
]);
579 if (sctx
->gs_shader
) {
580 si_shader_select(ctx
, sctx
->gs_shader
);
582 if (!sctx
->gs_shader
->current
->pm4
) {
583 si_pipe_shader_gs(ctx
, sctx
->gs_shader
->current
);
584 si_pipe_shader_vs(ctx
,
585 sctx
->gs_shader
->current
->gs_copy_shader
);
588 si_pm4_bind_state(sctx
, gs
, sctx
->gs_shader
->current
->pm4
);
589 si_pm4_bind_state(sctx
, vs
, sctx
->gs_shader
->current
->gs_copy_shader
->pm4
);
591 sctx
->b
.streamout
.stride_in_dw
= sctx
->gs_shader
->so
.stride
;
593 si_shader_select(ctx
, sctx
->vs_shader
);
595 if (!sctx
->vs_shader
->current
->pm4
)
596 si_pipe_shader_es(ctx
, sctx
->vs_shader
->current
);
598 si_pm4_bind_state(sctx
, es
, sctx
->vs_shader
->current
->pm4
);
601 si_init_gs_rings(sctx
);
602 if (sctx
->emitted
.named
.gs_rings
!= sctx
->gs_rings
)
603 sctx
->b
.flags
|= R600_CONTEXT_VGT_FLUSH
;
604 si_pm4_bind_state(sctx
, gs_rings
, sctx
->gs_rings
);
606 si_set_ring_buffer(ctx
, PIPE_SHADER_GEOMETRY
, SI_RING_GSVS
,
608 sctx
->gs_shader
->current
->shader
.gs_max_out_vertices
*
609 sctx
->gs_shader
->current
->shader
.noutput
* 16,
610 64, true, true, 4, 16);
613 sctx
->gs_on
= si_pm4_alloc_state(sctx
);
615 si_pm4_set_reg(sctx
->gs_on
, R_028B54_VGT_SHADER_STAGES_EN
,
616 S_028B54_ES_EN(V_028B54_ES_STAGE_REAL
) |
618 S_028B54_VS_EN(V_028B54_VS_STAGE_COPY_SHADER
));
620 si_pm4_bind_state(sctx
, gs_onoff
, sctx
->gs_on
);
622 si_shader_select(ctx
, sctx
->vs_shader
);
624 if (!sctx
->vs_shader
->current
->pm4
)
625 si_pipe_shader_vs(ctx
, sctx
->vs_shader
->current
);
627 si_pm4_bind_state(sctx
, vs
, sctx
->vs_shader
->current
->pm4
);
629 sctx
->b
.streamout
.stride_in_dw
= sctx
->vs_shader
->so
.stride
;
632 sctx
->gs_off
= si_pm4_alloc_state(sctx
);
634 si_pm4_set_reg(sctx
->gs_off
, R_028A40_VGT_GS_MODE
, 0);
635 si_pm4_set_reg(sctx
->gs_off
, R_028B54_VGT_SHADER_STAGES_EN
, 0);
637 si_pm4_bind_state(sctx
, gs_onoff
, sctx
->gs_off
);
638 si_pm4_bind_state(sctx
, gs_rings
, NULL
);
639 si_pm4_bind_state(sctx
, gs
, NULL
);
640 si_pm4_bind_state(sctx
, es
, NULL
);
643 si_shader_select(ctx
, sctx
->ps_shader
);
645 if (!sctx
->ps_shader
->current
->pm4
||
646 sctx
->ps_shader
->current
->cb0_is_integer
!= sctx
->framebuffer
.cb0_is_integer
)
647 si_pipe_shader_ps(ctx
, sctx
->ps_shader
->current
);
649 si_pm4_bind_state(sctx
, ps
, sctx
->ps_shader
->current
->pm4
);
651 if (si_pm4_state_changed(sctx
, ps
) || si_pm4_state_changed(sctx
, vs
)) {
652 /* XXX: Emitting the PS state even when only the VS changed
653 * fixes random failures with piglit glsl-max-varyings.
656 sctx
->emitted
.named
.ps
= NULL
;
657 si_update_spi_map(sctx
);
661 static void si_vertex_buffer_update(struct si_context
*sctx
)
663 struct pipe_context
*ctx
= &sctx
->b
.b
;
664 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
665 bool bound
[PIPE_MAX_ATTRIBS
] = {};
669 sctx
->b
.flags
|= R600_CONTEXT_INV_TEX_CACHE
;
671 count
= sctx
->vertex_elements
->count
;
672 assert(count
<= 256 / 4);
674 si_pm4_sh_data_begin(pm4
);
675 for (i
= 0 ; i
< count
; i
++) {
676 struct pipe_vertex_element
*ve
= &sctx
->vertex_elements
->elements
[i
];
677 struct pipe_vertex_buffer
*vb
;
678 struct r600_resource
*rbuffer
;
681 if (ve
->vertex_buffer_index
>= sctx
->nr_vertex_buffers
)
684 vb
= &sctx
->vertex_buffer
[ve
->vertex_buffer_index
];
685 rbuffer
= (struct r600_resource
*)vb
->buffer
;
690 offset
+= vb
->buffer_offset
;
691 offset
+= ve
->src_offset
;
693 va
= r600_resource_va(ctx
->screen
, (void*)rbuffer
);
696 /* Fill in T# buffer resource description */
697 si_pm4_sh_data_add(pm4
, va
& 0xFFFFFFFF);
698 si_pm4_sh_data_add(pm4
, (S_008F04_BASE_ADDRESS_HI(va
>> 32) |
699 S_008F04_STRIDE(vb
->stride
)));
701 /* Round up by rounding down and adding 1 */
702 si_pm4_sh_data_add(pm4
,
703 (vb
->buffer
->width0
- offset
-
704 util_format_get_blocksize(ve
->src_format
)) /
707 si_pm4_sh_data_add(pm4
, vb
->buffer
->width0
- offset
);
708 si_pm4_sh_data_add(pm4
, sctx
->vertex_elements
->rsrc_word3
[i
]);
710 if (!bound
[ve
->vertex_buffer_index
]) {
711 si_pm4_add_bo(pm4
, rbuffer
, RADEON_USAGE_READ
,
712 RADEON_PRIO_SHADER_BUFFER_RO
);
713 bound
[ve
->vertex_buffer_index
] = true;
716 si_pm4_sh_data_end(pm4
, sctx
->gs_shader
?
717 R_00B330_SPI_SHADER_USER_DATA_ES_0
:
718 R_00B130_SPI_SHADER_USER_DATA_VS_0
,
719 SI_SGPR_VERTEX_BUFFER
);
720 si_pm4_set_state(sctx
, vertex_buffers
, pm4
);
723 static void si_state_draw(struct si_context
*sctx
,
724 const struct pipe_draw_info
*info
,
725 const struct pipe_index_buffer
*ib
)
727 unsigned sh_base_reg
= (sctx
->gs_shader
? R_00B330_SPI_SHADER_USER_DATA_ES_0
:
728 R_00B130_SPI_SHADER_USER_DATA_VS_0
);
729 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
734 /* queries need some special values
735 * (this is non-zero if any query is active) */
736 if (sctx
->b
.num_occlusion_queries
> 0) {
737 if (sctx
->b
.chip_class
>= CIK
) {
738 si_pm4_set_reg(pm4
, R_028004_DB_COUNT_CONTROL
,
739 S_028004_PERFECT_ZPASS_COUNTS(1) |
740 S_028004_SAMPLE_RATE(sctx
->framebuffer
.log_samples
) |
741 S_028004_ZPASS_ENABLE(1) |
742 S_028004_SLICE_EVEN_ENABLE(1) |
743 S_028004_SLICE_ODD_ENABLE(1));
745 si_pm4_set_reg(pm4
, R_028004_DB_COUNT_CONTROL
,
746 S_028004_PERFECT_ZPASS_COUNTS(1) |
747 S_028004_SAMPLE_RATE(sctx
->framebuffer
.log_samples
));
751 if (info
->count_from_stream_output
) {
752 struct r600_so_target
*t
=
753 (struct r600_so_target
*)info
->count_from_stream_output
;
754 uint64_t va
= r600_resource_va(&sctx
->screen
->b
.b
,
755 &t
->buf_filled_size
->b
.b
);
756 va
+= t
->buf_filled_size_offset
;
758 si_pm4_set_reg(pm4
, R_028B30_VGT_STRMOUT_DRAW_OPAQUE_VERTEX_STRIDE
,
761 si_pm4_cmd_begin(pm4
, PKT3_COPY_DATA
);
763 COPY_DATA_SRC_SEL(COPY_DATA_MEM
) |
764 COPY_DATA_DST_SEL(COPY_DATA_REG
) |
765 COPY_DATA_WR_CONFIRM
);
766 si_pm4_cmd_add(pm4
, va
); /* src address lo */
767 si_pm4_cmd_add(pm4
, va
>> 32UL); /* src address hi */
768 si_pm4_cmd_add(pm4
, R_028B2C_VGT_STRMOUT_DRAW_OPAQUE_BUFFER_FILLED_SIZE
>> 2);
769 si_pm4_cmd_add(pm4
, 0); /* unused */
770 si_pm4_add_bo(pm4
, t
->buf_filled_size
, RADEON_USAGE_READ
,
772 si_pm4_cmd_end(pm4
, true);
776 si_pm4_cmd_begin(pm4
, PKT3_INDEX_TYPE
);
777 if (ib
->index_size
== 4) {
778 si_pm4_cmd_add(pm4
, V_028A7C_VGT_INDEX_32
| (SI_BIG_ENDIAN
?
779 V_028A7C_VGT_DMA_SWAP_32_BIT
: 0));
781 si_pm4_cmd_add(pm4
, V_028A7C_VGT_INDEX_16
| (SI_BIG_ENDIAN
?
782 V_028A7C_VGT_DMA_SWAP_16_BIT
: 0));
784 si_pm4_cmd_end(pm4
, sctx
->b
.predicate_drawing
);
786 if (!info
->indirect
) {
787 si_pm4_cmd_begin(pm4
, PKT3_NUM_INSTANCES
);
788 si_pm4_cmd_add(pm4
, info
->instance_count
);
789 si_pm4_cmd_end(pm4
, sctx
->b
.predicate_drawing
);
791 si_pm4_set_reg(pm4
, sh_base_reg
+ SI_SGPR_BASE_VERTEX
* 4,
792 info
->indexed
? info
->index_bias
: info
->start
);
793 si_pm4_set_reg(pm4
, sh_base_reg
+ SI_SGPR_START_INSTANCE
* 4,
794 info
->start_instance
);
796 si_pm4_add_bo(pm4
, (struct r600_resource
*)info
->indirect
,
797 RADEON_USAGE_READ
, RADEON_PRIO_MIN
);
801 uint32_t max_size
= (ib
->buffer
->width0
- ib
->offset
) /
802 sctx
->index_buffer
.index_size
;
804 va
= r600_resource_va(&sctx
->screen
->b
.b
, ib
->buffer
);
807 si_pm4_add_bo(pm4
, (struct r600_resource
*)ib
->buffer
, RADEON_USAGE_READ
,
810 if (info
->indirect
) {
811 uint64_t indirect_va
= r600_resource_va(&sctx
->screen
->b
.b
,
813 si_cmd_draw_index_indirect(pm4
, indirect_va
, va
, max_size
,
814 info
->indirect_offset
,
815 sh_base_reg
+ SI_SGPR_BASE_VERTEX
* 4,
816 sh_base_reg
+ SI_SGPR_START_INSTANCE
* 4,
817 sctx
->b
.predicate_drawing
);
819 va
+= info
->start
* ib
->index_size
;
820 si_cmd_draw_index_2(pm4
, max_size
, va
, info
->count
,
821 V_0287F0_DI_SRC_SEL_DMA
,
822 sctx
->b
.predicate_drawing
);
825 if (info
->indirect
) {
826 uint64_t indirect_va
= r600_resource_va(&sctx
->screen
->b
.b
,
828 si_cmd_draw_indirect(pm4
, indirect_va
, info
->indirect_offset
,
829 sh_base_reg
+ SI_SGPR_BASE_VERTEX
* 4,
830 sh_base_reg
+ SI_SGPR_START_INSTANCE
* 4,
831 sctx
->b
.predicate_drawing
);
833 si_cmd_draw_index_auto(pm4
, info
->count
,
834 V_0287F0_DI_SRC_SEL_AUTO_INDEX
|
835 S_0287F0_USE_OPAQUE(!!info
->count_from_stream_output
),
836 sctx
->b
.predicate_drawing
);
840 si_pm4_set_state(sctx
, draw
, pm4
);
843 void si_emit_cache_flush(struct r600_common_context
*sctx
, struct r600_atom
*atom
)
845 struct radeon_winsys_cs
*cs
= sctx
->rings
.gfx
.cs
;
846 uint32_t cp_coher_cntl
= 0;
848 /* XXX SI flushes both ICACHE and KCACHE if either flag is set.
849 * XXX CIK shouldn't have this issue. Test CIK before separating the flags
850 * XXX to ensure there is no regression. Also find out if there is another
851 * XXX way to flush either ICACHE or KCACHE but not both for SI. */
852 if (sctx
->flags
& (R600_CONTEXT_INV_SHADER_CACHE
|
853 R600_CONTEXT_INV_CONST_CACHE
)) {
854 cp_coher_cntl
|= S_0085F0_SH_ICACHE_ACTION_ENA(1) |
855 S_0085F0_SH_KCACHE_ACTION_ENA(1);
857 if (sctx
->flags
& (R600_CONTEXT_INV_TEX_CACHE
|
858 R600_CONTEXT_STREAMOUT_FLUSH
)) {
859 cp_coher_cntl
|= S_0085F0_TC_ACTION_ENA(1) |
860 S_0085F0_TCL1_ACTION_ENA(1);
862 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_CB
) {
863 cp_coher_cntl
|= S_0085F0_CB_ACTION_ENA(1) |
864 S_0085F0_CB0_DEST_BASE_ENA(1) |
865 S_0085F0_CB1_DEST_BASE_ENA(1) |
866 S_0085F0_CB2_DEST_BASE_ENA(1) |
867 S_0085F0_CB3_DEST_BASE_ENA(1) |
868 S_0085F0_CB4_DEST_BASE_ENA(1) |
869 S_0085F0_CB5_DEST_BASE_ENA(1) |
870 S_0085F0_CB6_DEST_BASE_ENA(1) |
871 S_0085F0_CB7_DEST_BASE_ENA(1);
873 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_DB
) {
874 cp_coher_cntl
|= S_0085F0_DB_ACTION_ENA(1) |
875 S_0085F0_DB_DEST_BASE_ENA(1);
879 if (sctx
->chip_class
>= CIK
) {
880 radeon_emit(cs
, PKT3(PKT3_ACQUIRE_MEM
, 5, 0));
881 radeon_emit(cs
, cp_coher_cntl
); /* CP_COHER_CNTL */
882 radeon_emit(cs
, 0xffffffff); /* CP_COHER_SIZE */
883 radeon_emit(cs
, 0xff); /* CP_COHER_SIZE_HI */
884 radeon_emit(cs
, 0); /* CP_COHER_BASE */
885 radeon_emit(cs
, 0); /* CP_COHER_BASE_HI */
886 radeon_emit(cs
, 0x0000000A); /* POLL_INTERVAL */
888 radeon_emit(cs
, PKT3(PKT3_SURFACE_SYNC
, 3, 0));
889 radeon_emit(cs
, cp_coher_cntl
); /* CP_COHER_CNTL */
890 radeon_emit(cs
, 0xffffffff); /* CP_COHER_SIZE */
891 radeon_emit(cs
, 0); /* CP_COHER_BASE */
892 radeon_emit(cs
, 0x0000000A); /* POLL_INTERVAL */
896 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_CB_META
) {
897 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
898 radeon_emit(cs
, EVENT_TYPE(V_028A90_FLUSH_AND_INV_CB_META
) | EVENT_INDEX(0));
900 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_DB_META
) {
901 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
902 radeon_emit(cs
, EVENT_TYPE(V_028A90_FLUSH_AND_INV_DB_META
) | EVENT_INDEX(0));
905 if (sctx
->flags
& (R600_CONTEXT_WAIT_3D_IDLE
|
906 R600_CONTEXT_PS_PARTIAL_FLUSH
)) {
907 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
908 radeon_emit(cs
, EVENT_TYPE(V_028A90_PS_PARTIAL_FLUSH
) | EVENT_INDEX(4));
909 } else if (sctx
->flags
& R600_CONTEXT_STREAMOUT_FLUSH
) {
910 /* Needed if streamout buffers are going to be used as a source. */
911 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
912 radeon_emit(cs
, EVENT_TYPE(V_028A90_VS_PARTIAL_FLUSH
) | EVENT_INDEX(4));
915 if (sctx
->flags
& R600_CONTEXT_VGT_FLUSH
) {
916 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
917 radeon_emit(cs
, EVENT_TYPE(V_028A90_VGT_FLUSH
) | EVENT_INDEX(0));
923 const struct r600_atom si_atom_cache_flush
= { si_emit_cache_flush
, 13 }; /* number of CS dwords */
925 static void si_get_draw_start_count(struct si_context
*sctx
,
926 const struct pipe_draw_info
*info
,
927 unsigned *start
, unsigned *count
)
929 if (info
->indirect
) {
930 struct r600_resource
*indirect
=
931 (struct r600_resource
*)info
->indirect
;
932 int *data
= r600_buffer_map_sync_with_rings(&sctx
->b
,
933 indirect
, PIPE_TRANSFER_READ
);
934 data
+= info
->indirect_offset
/sizeof(int);
938 *start
= info
->start
;
939 *count
= info
->count
;
943 void si_draw_vbo(struct pipe_context
*ctx
, const struct pipe_draw_info
*info
)
945 struct si_context
*sctx
= (struct si_context
*)ctx
;
946 struct pipe_index_buffer ib
= {};
949 if (!info
->count
&& !info
->indirect
&&
950 (info
->indexed
|| !info
->count_from_stream_output
))
953 if (!sctx
->ps_shader
|| !sctx
->vs_shader
)
956 si_update_derived_state(sctx
);
957 si_vertex_buffer_update(sctx
);
960 /* Initialize the index buffer struct. */
961 pipe_resource_reference(&ib
.buffer
, sctx
->index_buffer
.buffer
);
962 ib
.user_buffer
= sctx
->index_buffer
.user_buffer
;
963 ib
.index_size
= sctx
->index_buffer
.index_size
;
964 ib
.offset
= sctx
->index_buffer
.offset
;
966 /* Translate or upload, if needed. */
967 if (ib
.index_size
== 1) {
968 struct pipe_resource
*out_buffer
= NULL
;
969 unsigned out_offset
, start
, count
, start_offset
;
972 si_get_draw_start_count(sctx
, info
, &start
, &count
);
973 start_offset
= start
* ib
.index_size
;
975 u_upload_alloc(sctx
->b
.uploader
, start_offset
, count
* 2,
976 &out_offset
, &out_buffer
, &ptr
);
978 util_shorten_ubyte_elts_to_userptr(&sctx
->b
.b
, &ib
, 0,
979 ib
.offset
+ start_offset
,
982 pipe_resource_reference(&ib
.buffer
, NULL
);
983 ib
.user_buffer
= NULL
;
984 ib
.buffer
= out_buffer
;
985 /* info->start will be added by the drawing code */
986 ib
.offset
= out_offset
- start_offset
;
988 } else if (ib
.user_buffer
&& !ib
.buffer
) {
989 unsigned start
, count
, start_offset
;
991 si_get_draw_start_count(sctx
, info
, &start
, &count
);
992 start_offset
= start
* ib
.index_size
;
994 u_upload_data(sctx
->b
.uploader
, start_offset
, count
* ib
.index_size
,
995 (char*)ib
.user_buffer
+ start_offset
,
996 &ib
.offset
, &ib
.buffer
);
997 /* info->start will be added by the drawing code */
998 ib
.offset
-= start_offset
;
1002 if (!si_update_draw_info_state(sctx
, info
, &ib
))
1005 si_state_draw(sctx
, info
, &ib
);
1007 sctx
->pm4_dirty_cdwords
+= si_pm4_dirty_dw(sctx
);
1009 /* Check flush flags. */
1011 sctx
->atoms
.s
.cache_flush
->dirty
= true;
1013 si_need_cs_space(sctx
, 0, TRUE
);
1016 for (i
= 0; i
< SI_NUM_ATOMS(sctx
); i
++) {
1017 if (sctx
->atoms
.array
[i
]->dirty
) {
1018 sctx
->atoms
.array
[i
]->emit(&sctx
->b
, sctx
->atoms
.array
[i
]);
1019 sctx
->atoms
.array
[i
]->dirty
= false;
1023 si_pm4_emit_dirty(sctx
);
1024 sctx
->pm4_dirty_cdwords
= 0;
1027 if (sctx
->screen
->b
.trace_bo
) {
1028 si_trace_emit(sctx
);
1032 /* Set the depth buffer as dirty. */
1033 if (sctx
->framebuffer
.state
.zsbuf
) {
1034 struct pipe_surface
*surf
= sctx
->framebuffer
.state
.zsbuf
;
1035 struct r600_texture
*rtex
= (struct r600_texture
*)surf
->texture
;
1037 rtex
->dirty_level_mask
|= 1 << surf
->u
.tex
.level
;
1039 if (sctx
->framebuffer
.compressed_cb_mask
) {
1040 struct pipe_surface
*surf
;
1041 struct r600_texture
*rtex
;
1042 unsigned mask
= sctx
->framebuffer
.compressed_cb_mask
;
1045 unsigned i
= u_bit_scan(&mask
);
1046 surf
= sctx
->framebuffer
.state
.cbufs
[i
];
1047 rtex
= (struct r600_texture
*)surf
->texture
;
1049 rtex
->dirty_level_mask
|= 1 << surf
->u
.tex
.level
;
1053 pipe_resource_reference(&ib
.buffer
, NULL
);
1054 sctx
->b
.num_draw_calls
++;