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>
27 #include "util/u_memory.h"
28 #include "util/u_framebuffer.h"
29 #include "util/u_blitter.h"
30 #include "util/u_index_modify.h"
31 #include "util/u_upload_mgr.h"
32 #include "tgsi/tgsi_parse.h"
34 #include "si_shader.h"
36 #include "../radeon/r600_cs.h"
43 static void si_pipe_shader_vs(struct pipe_context
*ctx
, struct si_pipe_shader
*shader
)
45 struct si_context
*sctx
= (struct si_context
*)ctx
;
46 struct si_pm4_state
*pm4
;
47 unsigned num_sgprs
, num_user_sgprs
;
48 unsigned nparams
, i
, vgpr_comp_cnt
;
51 si_pm4_delete_state(sctx
, vs
, shader
->pm4
);
52 pm4
= shader
->pm4
= si_pm4_alloc_state(sctx
);
57 /* Certain attributes (position, psize, etc.) don't count as params.
58 * VS is required to export at least one param and r600_shader_from_tgsi()
59 * takes care of adding a dummy export.
61 for (nparams
= 0, i
= 0 ; i
< shader
->shader
.noutput
; i
++) {
62 switch (shader
->shader
.output
[i
].name
) {
63 case TGSI_SEMANTIC_CLIPVERTEX
:
64 case TGSI_SEMANTIC_POSITION
:
65 case TGSI_SEMANTIC_PSIZE
:
74 si_pm4_set_reg(pm4
, R_0286C4_SPI_VS_OUT_CONFIG
,
75 S_0286C4_VS_EXPORT_COUNT(nparams
- 1));
77 si_pm4_set_reg(pm4
, R_02870C_SPI_SHADER_POS_FORMAT
,
78 S_02870C_POS0_EXPORT_FORMAT(V_02870C_SPI_SHADER_4COMP
) |
79 S_02870C_POS1_EXPORT_FORMAT(shader
->shader
.nr_pos_exports
> 1 ?
80 V_02870C_SPI_SHADER_4COMP
:
81 V_02870C_SPI_SHADER_NONE
) |
82 S_02870C_POS2_EXPORT_FORMAT(shader
->shader
.nr_pos_exports
> 2 ?
83 V_02870C_SPI_SHADER_4COMP
:
84 V_02870C_SPI_SHADER_NONE
) |
85 S_02870C_POS3_EXPORT_FORMAT(shader
->shader
.nr_pos_exports
> 3 ?
86 V_02870C_SPI_SHADER_4COMP
:
87 V_02870C_SPI_SHADER_NONE
));
89 va
= r600_resource_va(ctx
->screen
, (void *)shader
->bo
);
90 si_pm4_add_bo(pm4
, shader
->bo
, RADEON_USAGE_READ
);
91 si_pm4_set_reg(pm4
, R_00B120_SPI_SHADER_PGM_LO_VS
, va
>> 8);
92 si_pm4_set_reg(pm4
, R_00B124_SPI_SHADER_PGM_HI_VS
, va
>> 40);
94 num_user_sgprs
= SI_VS_NUM_USER_SGPR
;
95 num_sgprs
= shader
->num_sgprs
;
96 if (num_user_sgprs
> num_sgprs
) {
97 /* Last 2 reserved SGPRs are used for VCC */
98 num_sgprs
= num_user_sgprs
+ 2;
100 assert(num_sgprs
<= 104);
102 vgpr_comp_cnt
= shader
->shader
.uses_instanceid
? 3 : 0;
104 si_pm4_set_reg(pm4
, R_00B128_SPI_SHADER_PGM_RSRC1_VS
,
105 S_00B128_VGPRS((shader
->num_vgprs
- 1) / 4) |
106 S_00B128_SGPRS((num_sgprs
- 1) / 8) |
107 S_00B128_VGPR_COMP_CNT(vgpr_comp_cnt
));
108 si_pm4_set_reg(pm4
, R_00B12C_SPI_SHADER_PGM_RSRC2_VS
,
109 S_00B12C_USER_SGPR(num_user_sgprs
) |
110 S_00B12C_SO_BASE0_EN(!!shader
->selector
->so
.stride
[0]) |
111 S_00B12C_SO_BASE1_EN(!!shader
->selector
->so
.stride
[1]) |
112 S_00B12C_SO_BASE2_EN(!!shader
->selector
->so
.stride
[2]) |
113 S_00B12C_SO_BASE3_EN(!!shader
->selector
->so
.stride
[3]) |
114 S_00B12C_SO_EN(!!shader
->selector
->so
.num_outputs
));
116 si_pm4_bind_state(sctx
, vs
, shader
->pm4
);
117 sctx
->b
.flags
|= R600_CONTEXT_INV_SHADER_CACHE
;
120 static void si_pipe_shader_ps(struct pipe_context
*ctx
, struct si_pipe_shader
*shader
)
122 struct si_context
*sctx
= (struct si_context
*)ctx
;
123 struct si_pm4_state
*pm4
;
124 unsigned i
, exports_ps
, spi_ps_in_control
, db_shader_control
;
125 unsigned num_sgprs
, num_user_sgprs
;
126 unsigned spi_baryc_cntl
= 0, spi_ps_input_ena
, spi_shader_z_format
;
129 si_pm4_delete_state(sctx
, ps
, shader
->pm4
);
130 pm4
= shader
->pm4
= si_pm4_alloc_state(sctx
);
135 db_shader_control
= S_02880C_Z_ORDER(V_02880C_EARLY_Z_THEN_LATE_Z
) |
136 S_02880C_ALPHA_TO_MASK_DISABLE(sctx
->fb_cb0_is_integer
);
138 for (i
= 0; i
< shader
->shader
.ninput
; i
++) {
139 switch (shader
->shader
.input
[i
].name
) {
140 case TGSI_SEMANTIC_POSITION
:
141 if (shader
->shader
.input
[i
].centroid
) {
142 /* SPI_BARYC_CNTL.POS_FLOAT_LOCATION
144 * 0 -> Position = pixel center (default)
145 * 1 -> Position = pixel centroid
146 * 2 -> Position = iterated sample number XXX:
147 * What does this mean?
149 spi_baryc_cntl
|= S_0286E0_POS_FLOAT_LOCATION(1);
152 case TGSI_SEMANTIC_FACE
:
157 for (i
= 0; i
< shader
->shader
.noutput
; i
++) {
158 if (shader
->shader
.output
[i
].name
== TGSI_SEMANTIC_POSITION
)
159 db_shader_control
|= S_02880C_Z_EXPORT_ENABLE(1);
160 if (shader
->shader
.output
[i
].name
== TGSI_SEMANTIC_STENCIL
)
161 db_shader_control
|= S_02880C_STENCIL_TEST_VAL_EXPORT_ENABLE(1);
163 if (shader
->shader
.uses_kill
|| shader
->key
.ps
.alpha_func
!= PIPE_FUNC_ALWAYS
)
164 db_shader_control
|= S_02880C_KILL_ENABLE(1);
167 for (i
= 0; i
< shader
->shader
.noutput
; i
++) {
168 if (shader
->shader
.output
[i
].name
== TGSI_SEMANTIC_POSITION
||
169 shader
->shader
.output
[i
].name
== TGSI_SEMANTIC_STENCIL
)
173 /* always at least export 1 component per pixel */
177 spi_ps_in_control
= S_0286D8_NUM_INTERP(shader
->shader
.ninterp
) |
178 S_0286D8_BC_OPTIMIZE_DISABLE(1);
180 si_pm4_set_reg(pm4
, R_0286E0_SPI_BARYC_CNTL
, spi_baryc_cntl
);
181 spi_ps_input_ena
= shader
->spi_ps_input_ena
;
182 /* we need to enable at least one of them, otherwise we hang the GPU */
183 assert(G_0286CC_PERSP_SAMPLE_ENA(spi_ps_input_ena
) ||
184 G_0286CC_PERSP_CENTER_ENA(spi_ps_input_ena
) ||
185 G_0286CC_PERSP_CENTROID_ENA(spi_ps_input_ena
) ||
186 G_0286CC_PERSP_PULL_MODEL_ENA(spi_ps_input_ena
) ||
187 G_0286CC_LINEAR_SAMPLE_ENA(spi_ps_input_ena
) ||
188 G_0286CC_LINEAR_CENTER_ENA(spi_ps_input_ena
) ||
189 G_0286CC_LINEAR_CENTROID_ENA(spi_ps_input_ena
) ||
190 G_0286CC_LINE_STIPPLE_TEX_ENA(spi_ps_input_ena
));
192 si_pm4_set_reg(pm4
, R_0286CC_SPI_PS_INPUT_ENA
, spi_ps_input_ena
);
193 si_pm4_set_reg(pm4
, R_0286D0_SPI_PS_INPUT_ADDR
, spi_ps_input_ena
);
194 si_pm4_set_reg(pm4
, R_0286D8_SPI_PS_IN_CONTROL
, spi_ps_in_control
);
196 if (G_02880C_STENCIL_TEST_VAL_EXPORT_ENABLE(db_shader_control
))
197 spi_shader_z_format
= V_028710_SPI_SHADER_32_GR
;
198 else if (G_02880C_Z_EXPORT_ENABLE(db_shader_control
))
199 spi_shader_z_format
= V_028710_SPI_SHADER_32_R
;
201 spi_shader_z_format
= 0;
202 si_pm4_set_reg(pm4
, R_028710_SPI_SHADER_Z_FORMAT
, spi_shader_z_format
);
203 si_pm4_set_reg(pm4
, R_028714_SPI_SHADER_COL_FORMAT
,
204 shader
->spi_shader_col_format
);
205 si_pm4_set_reg(pm4
, R_02823C_CB_SHADER_MASK
, shader
->cb_shader_mask
);
207 va
= r600_resource_va(ctx
->screen
, (void *)shader
->bo
);
208 si_pm4_add_bo(pm4
, shader
->bo
, RADEON_USAGE_READ
);
209 si_pm4_set_reg(pm4
, R_00B020_SPI_SHADER_PGM_LO_PS
, va
>> 8);
210 si_pm4_set_reg(pm4
, R_00B024_SPI_SHADER_PGM_HI_PS
, va
>> 40);
212 num_user_sgprs
= SI_PS_NUM_USER_SGPR
;
213 num_sgprs
= shader
->num_sgprs
;
214 /* One SGPR after user SGPRs is pre-loaded with {prim_mask, lds_offset} */
215 if ((num_user_sgprs
+ 1) > num_sgprs
) {
216 /* Last 2 reserved SGPRs are used for VCC */
217 num_sgprs
= num_user_sgprs
+ 1 + 2;
219 assert(num_sgprs
<= 104);
221 si_pm4_set_reg(pm4
, R_00B028_SPI_SHADER_PGM_RSRC1_PS
,
222 S_00B028_VGPRS((shader
->num_vgprs
- 1) / 4) |
223 S_00B028_SGPRS((num_sgprs
- 1) / 8));
224 si_pm4_set_reg(pm4
, R_00B02C_SPI_SHADER_PGM_RSRC2_PS
,
225 S_00B02C_EXTRA_LDS_SIZE(shader
->lds_size
) |
226 S_00B02C_USER_SGPR(num_user_sgprs
));
228 si_pm4_set_reg(pm4
, R_02880C_DB_SHADER_CONTROL
, db_shader_control
);
230 shader
->cb0_is_integer
= sctx
->fb_cb0_is_integer
;
231 shader
->sprite_coord_enable
= sctx
->sprite_coord_enable
;
232 si_pm4_bind_state(sctx
, ps
, shader
->pm4
);
233 sctx
->b
.flags
|= R600_CONTEXT_INV_SHADER_CACHE
;
240 static unsigned si_conv_pipe_prim(unsigned pprim
)
242 static const unsigned prim_conv
[] = {
243 [PIPE_PRIM_POINTS
] = V_008958_DI_PT_POINTLIST
,
244 [PIPE_PRIM_LINES
] = V_008958_DI_PT_LINELIST
,
245 [PIPE_PRIM_LINE_LOOP
] = V_008958_DI_PT_LINELOOP
,
246 [PIPE_PRIM_LINE_STRIP
] = V_008958_DI_PT_LINESTRIP
,
247 [PIPE_PRIM_TRIANGLES
] = V_008958_DI_PT_TRILIST
,
248 [PIPE_PRIM_TRIANGLE_STRIP
] = V_008958_DI_PT_TRISTRIP
,
249 [PIPE_PRIM_TRIANGLE_FAN
] = V_008958_DI_PT_TRIFAN
,
250 [PIPE_PRIM_QUADS
] = V_008958_DI_PT_QUADLIST
,
251 [PIPE_PRIM_QUAD_STRIP
] = V_008958_DI_PT_QUADSTRIP
,
252 [PIPE_PRIM_POLYGON
] = V_008958_DI_PT_POLYGON
,
253 [PIPE_PRIM_LINES_ADJACENCY
] = ~0,
254 [PIPE_PRIM_LINE_STRIP_ADJACENCY
] = ~0,
255 [PIPE_PRIM_TRIANGLES_ADJACENCY
] = ~0,
256 [PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY
] = ~0
258 unsigned result
= prim_conv
[pprim
];
260 R600_ERR("unsupported primitive type %d\n", pprim
);
265 static unsigned si_conv_prim_to_gs_out(unsigned mode
)
267 static const int prim_conv
[] = {
268 [PIPE_PRIM_POINTS
] = V_028A6C_OUTPRIM_TYPE_POINTLIST
,
269 [PIPE_PRIM_LINES
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
270 [PIPE_PRIM_LINE_LOOP
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
271 [PIPE_PRIM_LINE_STRIP
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
272 [PIPE_PRIM_TRIANGLES
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
273 [PIPE_PRIM_TRIANGLE_STRIP
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
274 [PIPE_PRIM_TRIANGLE_FAN
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
275 [PIPE_PRIM_QUADS
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
276 [PIPE_PRIM_QUAD_STRIP
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
277 [PIPE_PRIM_POLYGON
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
278 [PIPE_PRIM_LINES_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
279 [PIPE_PRIM_LINE_STRIP_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
280 [PIPE_PRIM_TRIANGLES_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
281 [PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY
] = V_028A6C_OUTPRIM_TYPE_TRISTRIP
283 assert(mode
< Elements(prim_conv
));
285 return prim_conv
[mode
];
288 static bool si_update_draw_info_state(struct si_context
*sctx
,
289 const struct pipe_draw_info
*info
,
290 const struct pipe_index_buffer
*ib
)
292 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
293 struct si_shader
*vs
= &sctx
->vs_shader
->current
->shader
;
294 unsigned prim
= si_conv_pipe_prim(info
->mode
);
295 unsigned gs_out_prim
= si_conv_prim_to_gs_out(info
->mode
);
296 unsigned ls_mask
= 0;
306 if (sctx
->b
.chip_class
>= CIK
) {
307 struct si_state_rasterizer
*rs
= sctx
->queued
.named
.rasterizer
;
308 bool wd_switch_on_eop
= prim
== V_008958_DI_PT_POLYGON
||
309 prim
== V_008958_DI_PT_LINELOOP
||
310 prim
== V_008958_DI_PT_TRIFAN
||
311 prim
== V_008958_DI_PT_TRISTRIP_ADJ
||
312 info
->primitive_restart
||
313 (rs
? rs
->line_stipple_enable
: false);
314 /* If the WD switch is false, the IA switch must be false too. */
315 bool ia_switch_on_eop
= wd_switch_on_eop
;
317 si_pm4_set_reg(pm4
, R_028AA8_IA_MULTI_VGT_PARAM
,
318 S_028AA8_SWITCH_ON_EOP(ia_switch_on_eop
) |
319 S_028AA8_PARTIAL_VS_WAVE_ON(1) |
320 S_028AA8_PRIMGROUP_SIZE(63) |
321 S_028AA8_WD_SWITCH_ON_EOP(wd_switch_on_eop
));
322 si_pm4_set_reg(pm4
, R_028B74_VGT_DISPATCH_DRAW_INDEX
,
323 ib
->index_size
== 4 ? 0xFC000000 : 0xFC00);
325 si_pm4_set_reg(pm4
, R_030908_VGT_PRIMITIVE_TYPE
, prim
);
327 si_pm4_set_reg(pm4
, R_008958_VGT_PRIMITIVE_TYPE
, prim
);
330 si_pm4_set_reg(pm4
, R_028A6C_VGT_GS_OUT_PRIM_TYPE
, gs_out_prim
);
331 si_pm4_set_reg(pm4
, R_028408_VGT_INDX_OFFSET
,
332 info
->indexed
? info
->index_bias
: info
->start
);
333 si_pm4_set_reg(pm4
, R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX
, info
->restart_index
);
334 si_pm4_set_reg(pm4
, R_028A94_VGT_MULTI_PRIM_IB_RESET_EN
, info
->primitive_restart
);
335 si_pm4_set_reg(pm4
, R_00B130_SPI_SHADER_USER_DATA_VS_0
+ SI_SGPR_START_INSTANCE
* 4,
336 info
->start_instance
);
338 if (prim
== V_008958_DI_PT_LINELIST
)
340 else if (prim
== V_008958_DI_PT_LINESTRIP
)
342 si_pm4_set_reg(pm4
, R_028A0C_PA_SC_LINE_STIPPLE
,
343 S_028A0C_AUTO_RESET_CNTL(ls_mask
) |
344 sctx
->pa_sc_line_stipple
);
346 if (info
->mode
== PIPE_PRIM_QUADS
|| info
->mode
== PIPE_PRIM_QUAD_STRIP
|| info
->mode
== PIPE_PRIM_POLYGON
) {
347 si_pm4_set_reg(pm4
, R_028814_PA_SU_SC_MODE_CNTL
,
348 S_028814_PROVOKING_VTX_LAST(1) | sctx
->pa_su_sc_mode_cntl
);
350 si_pm4_set_reg(pm4
, R_028814_PA_SU_SC_MODE_CNTL
, sctx
->pa_su_sc_mode_cntl
);
352 si_pm4_set_reg(pm4
, R_02881C_PA_CL_VS_OUT_CNTL
,
353 S_02881C_USE_VTX_POINT_SIZE(vs
->vs_out_point_size
) |
354 S_02881C_USE_VTX_EDGE_FLAG(vs
->vs_out_edgeflag
) |
355 S_02881C_USE_VTX_RENDER_TARGET_INDX(vs
->vs_out_layer
) |
356 S_02881C_VS_OUT_CCDIST0_VEC_ENA((vs
->clip_dist_write
& 0x0F) != 0) |
357 S_02881C_VS_OUT_CCDIST1_VEC_ENA((vs
->clip_dist_write
& 0xF0) != 0) |
358 S_02881C_VS_OUT_MISC_VEC_ENA(vs
->vs_out_misc_write
) |
359 (sctx
->queued
.named
.rasterizer
->clip_plane_enable
&
360 vs
->clip_dist_write
));
361 si_pm4_set_reg(pm4
, R_028810_PA_CL_CLIP_CNTL
,
362 sctx
->queued
.named
.rasterizer
->pa_cl_clip_cntl
|
363 (vs
->clip_dist_write
? 0 :
364 sctx
->queued
.named
.rasterizer
->clip_plane_enable
& 0x3F));
366 si_pm4_set_state(sctx
, draw_info
, pm4
);
370 static void si_update_spi_map(struct si_context
*sctx
)
372 struct si_shader
*ps
= &sctx
->ps_shader
->current
->shader
;
373 struct si_shader
*vs
= &sctx
->vs_shader
->current
->shader
;
374 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
377 for (i
= 0; i
< ps
->ninput
; i
++) {
378 unsigned name
= ps
->input
[i
].name
;
379 unsigned param_offset
= ps
->input
[i
].param_offset
;
381 if (name
== TGSI_SEMANTIC_POSITION
)
382 /* Read from preloaded VGPRs, not parameters */
388 if (ps
->input
[i
].interpolate
== TGSI_INTERPOLATE_CONSTANT
||
389 (ps
->input
[i
].interpolate
== TGSI_INTERPOLATE_COLOR
&&
390 sctx
->ps_shader
->current
->key
.ps
.flatshade
)) {
391 tmp
|= S_028644_FLAT_SHADE(1);
394 if (name
== TGSI_SEMANTIC_GENERIC
&&
395 sctx
->sprite_coord_enable
& (1 << ps
->input
[i
].sid
)) {
396 tmp
|= S_028644_PT_SPRITE_TEX(1);
399 for (j
= 0; j
< vs
->noutput
; j
++) {
400 if (name
== vs
->output
[j
].name
&&
401 ps
->input
[i
].sid
== vs
->output
[j
].sid
) {
402 tmp
|= S_028644_OFFSET(vs
->output
[j
].param_offset
);
407 if (j
== vs
->noutput
) {
408 /* No corresponding output found, load defaults into input */
409 tmp
|= S_028644_OFFSET(0x20);
413 R_028644_SPI_PS_INPUT_CNTL_0
+ param_offset
* 4,
416 if (name
== TGSI_SEMANTIC_COLOR
&&
417 sctx
->ps_shader
->current
->key
.ps
.color_two_side
) {
418 name
= TGSI_SEMANTIC_BCOLOR
;
424 si_pm4_set_state(sctx
, spi
, pm4
);
427 static void si_update_derived_state(struct si_context
*sctx
)
429 struct pipe_context
* ctx
= (struct pipe_context
*)sctx
;
430 unsigned vs_dirty
= 0, ps_dirty
= 0;
432 if (!sctx
->blitter
->running
) {
433 /* Flush depth textures which need to be flushed. */
434 for (int i
= 0; i
< SI_NUM_SHADERS
; i
++) {
435 if (sctx
->samplers
[i
].depth_texture_mask
) {
436 si_flush_depth_textures(sctx
, &sctx
->samplers
[i
]);
438 if (sctx
->samplers
[i
].compressed_colortex_mask
) {
439 si_decompress_color_textures(sctx
, &sctx
->samplers
[i
]);
444 si_shader_select(ctx
, sctx
->vs_shader
, &vs_dirty
);
446 if (!sctx
->vs_shader
->current
->pm4
) {
447 si_pipe_shader_vs(ctx
, sctx
->vs_shader
->current
);
452 si_pm4_bind_state(sctx
, vs
, sctx
->vs_shader
->current
->pm4
);
456 si_shader_select(ctx
, sctx
->ps_shader
, &ps_dirty
);
458 if (!sctx
->ps_shader
->current
->pm4
) {
459 si_pipe_shader_ps(ctx
, sctx
->ps_shader
->current
);
462 if (sctx
->ps_shader
->current
->cb0_is_integer
!= sctx
->fb_cb0_is_integer
) {
463 si_pipe_shader_ps(ctx
, sctx
->ps_shader
->current
);
468 si_pm4_bind_state(sctx
, ps
, sctx
->ps_shader
->current
->pm4
);
471 if (si_pm4_state_changed(sctx
, ps
) || si_pm4_state_changed(sctx
, vs
)) {
472 /* XXX: Emitting the PS state even when only the VS changed
473 * fixes random failures with piglit glsl-max-varyings.
476 sctx
->emitted
.named
.ps
= NULL
;
477 si_update_spi_map(sctx
);
481 static void si_vertex_buffer_update(struct si_context
*sctx
)
483 struct pipe_context
*ctx
= &sctx
->b
.b
;
484 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
485 bool bound
[PIPE_MAX_ATTRIBS
] = {};
489 sctx
->b
.flags
|= R600_CONTEXT_INV_TEX_CACHE
;
491 count
= sctx
->vertex_elements
->count
;
492 assert(count
<= 256 / 4);
494 si_pm4_sh_data_begin(pm4
);
495 for (i
= 0 ; i
< count
; i
++) {
496 struct pipe_vertex_element
*ve
= &sctx
->vertex_elements
->elements
[i
];
497 struct pipe_vertex_buffer
*vb
;
498 struct r600_resource
*rbuffer
;
501 if (ve
->vertex_buffer_index
>= sctx
->nr_vertex_buffers
)
504 vb
= &sctx
->vertex_buffer
[ve
->vertex_buffer_index
];
505 rbuffer
= (struct r600_resource
*)vb
->buffer
;
510 offset
+= vb
->buffer_offset
;
511 offset
+= ve
->src_offset
;
513 va
= r600_resource_va(ctx
->screen
, (void*)rbuffer
);
516 /* Fill in T# buffer resource description */
517 si_pm4_sh_data_add(pm4
, va
& 0xFFFFFFFF);
518 si_pm4_sh_data_add(pm4
, (S_008F04_BASE_ADDRESS_HI(va
>> 32) |
519 S_008F04_STRIDE(vb
->stride
)));
521 /* Round up by rounding down and adding 1 */
522 si_pm4_sh_data_add(pm4
,
523 (vb
->buffer
->width0
- offset
-
524 util_format_get_blocksize(ve
->src_format
)) /
527 si_pm4_sh_data_add(pm4
, vb
->buffer
->width0
- offset
);
528 si_pm4_sh_data_add(pm4
, sctx
->vertex_elements
->rsrc_word3
[i
]);
530 if (!bound
[ve
->vertex_buffer_index
]) {
531 si_pm4_add_bo(pm4
, rbuffer
, RADEON_USAGE_READ
);
532 bound
[ve
->vertex_buffer_index
] = true;
535 si_pm4_sh_data_end(pm4
, R_00B130_SPI_SHADER_USER_DATA_VS_0
, SI_SGPR_VERTEX_BUFFER
);
536 si_pm4_set_state(sctx
, vertex_buffers
, pm4
);
539 static void si_state_draw(struct si_context
*sctx
,
540 const struct pipe_draw_info
*info
,
541 const struct pipe_index_buffer
*ib
)
543 struct si_pm4_state
*pm4
= si_pm4_alloc_state(sctx
);
548 /* queries need some special values
549 * (this is non-zero if any query is active) */
550 if (sctx
->b
.num_occlusion_queries
> 0) {
551 if (sctx
->b
.chip_class
>= CIK
) {
552 si_pm4_set_reg(pm4
, R_028004_DB_COUNT_CONTROL
,
553 S_028004_PERFECT_ZPASS_COUNTS(1) |
554 S_028004_SAMPLE_RATE(sctx
->fb_log_samples
) |
555 S_028004_ZPASS_ENABLE(1) |
556 S_028004_SLICE_EVEN_ENABLE(1) |
557 S_028004_SLICE_ODD_ENABLE(1));
559 si_pm4_set_reg(pm4
, R_028004_DB_COUNT_CONTROL
,
560 S_028004_PERFECT_ZPASS_COUNTS(1) |
561 S_028004_SAMPLE_RATE(sctx
->fb_log_samples
));
565 if (info
->count_from_stream_output
) {
566 struct r600_so_target
*t
=
567 (struct r600_so_target
*)info
->count_from_stream_output
;
568 uint64_t va
= r600_resource_va(&sctx
->screen
->b
.b
,
569 &t
->buf_filled_size
->b
.b
);
570 va
+= t
->buf_filled_size_offset
;
572 si_pm4_set_reg(pm4
, R_028B30_VGT_STRMOUT_DRAW_OPAQUE_VERTEX_STRIDE
,
575 si_pm4_cmd_begin(pm4
, PKT3_COPY_DATA
);
577 COPY_DATA_SRC_SEL(COPY_DATA_MEM
) |
578 COPY_DATA_DST_SEL(COPY_DATA_REG
) |
579 COPY_DATA_WR_CONFIRM
);
580 si_pm4_cmd_add(pm4
, va
); /* src address lo */
581 si_pm4_cmd_add(pm4
, va
>> 32UL); /* src address hi */
582 si_pm4_cmd_add(pm4
, R_028B2C_VGT_STRMOUT_DRAW_OPAQUE_BUFFER_FILLED_SIZE
>> 2);
583 si_pm4_cmd_add(pm4
, 0); /* unused */
584 si_pm4_add_bo(pm4
, t
->buf_filled_size
, RADEON_USAGE_READ
);
585 si_pm4_cmd_end(pm4
, true);
589 si_pm4_cmd_begin(pm4
, PKT3_INDEX_TYPE
);
590 if (ib
->index_size
== 4) {
591 si_pm4_cmd_add(pm4
, V_028A7C_VGT_INDEX_32
| (SI_BIG_ENDIAN
?
592 V_028A7C_VGT_DMA_SWAP_32_BIT
: 0));
594 si_pm4_cmd_add(pm4
, V_028A7C_VGT_INDEX_16
| (SI_BIG_ENDIAN
?
595 V_028A7C_VGT_DMA_SWAP_16_BIT
: 0));
597 si_pm4_cmd_end(pm4
, sctx
->b
.predicate_drawing
);
599 si_pm4_cmd_begin(pm4
, PKT3_NUM_INSTANCES
);
600 si_pm4_cmd_add(pm4
, info
->instance_count
);
601 si_pm4_cmd_end(pm4
, sctx
->b
.predicate_drawing
);
604 uint32_t max_size
= (ib
->buffer
->width0
- ib
->offset
) /
605 sctx
->index_buffer
.index_size
;
607 va
= r600_resource_va(&sctx
->screen
->b
.b
, ib
->buffer
);
610 si_pm4_add_bo(pm4
, (struct r600_resource
*)ib
->buffer
, RADEON_USAGE_READ
);
611 si_cmd_draw_index_2(pm4
, max_size
, va
, info
->count
,
612 V_0287F0_DI_SRC_SEL_DMA
,
613 sctx
->b
.predicate_drawing
);
615 uint32_t initiator
= V_0287F0_DI_SRC_SEL_AUTO_INDEX
;
616 initiator
|= S_0287F0_USE_OPAQUE(!!info
->count_from_stream_output
);
617 si_cmd_draw_index_auto(pm4
, info
->count
, initiator
, sctx
->b
.predicate_drawing
);
620 si_pm4_set_state(sctx
, draw
, pm4
);
623 void si_emit_cache_flush(struct r600_common_context
*sctx
, struct r600_atom
*atom
)
625 struct radeon_winsys_cs
*cs
= sctx
->rings
.gfx
.cs
;
626 uint32_t cp_coher_cntl
= 0;
628 /* XXX SI flushes both ICACHE and KCACHE if either flag is set.
629 * XXX CIK shouldn't have this issue. Test CIK before separating the flags
630 * XXX to ensure there is no regression. Also find out if there is another
631 * XXX way to flush either ICACHE or KCACHE but not both for SI. */
632 if (sctx
->flags
& (R600_CONTEXT_INV_SHADER_CACHE
|
633 R600_CONTEXT_INV_CONST_CACHE
)) {
634 cp_coher_cntl
|= S_0085F0_SH_ICACHE_ACTION_ENA(1) |
635 S_0085F0_SH_KCACHE_ACTION_ENA(1);
637 if (sctx
->flags
& (R600_CONTEXT_INV_TEX_CACHE
|
638 R600_CONTEXT_STREAMOUT_FLUSH
)) {
639 cp_coher_cntl
|= S_0085F0_TC_ACTION_ENA(1) |
640 S_0085F0_TCL1_ACTION_ENA(1);
642 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_CB
) {
643 cp_coher_cntl
|= S_0085F0_CB_ACTION_ENA(1) |
644 S_0085F0_CB0_DEST_BASE_ENA(1) |
645 S_0085F0_CB1_DEST_BASE_ENA(1) |
646 S_0085F0_CB2_DEST_BASE_ENA(1) |
647 S_0085F0_CB3_DEST_BASE_ENA(1) |
648 S_0085F0_CB4_DEST_BASE_ENA(1) |
649 S_0085F0_CB5_DEST_BASE_ENA(1) |
650 S_0085F0_CB6_DEST_BASE_ENA(1) |
651 S_0085F0_CB7_DEST_BASE_ENA(1);
653 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_DB
) {
654 cp_coher_cntl
|= S_0085F0_DB_ACTION_ENA(1) |
655 S_0085F0_DB_DEST_BASE_ENA(1);
659 if (sctx
->chip_class
>= CIK
) {
660 radeon_emit(cs
, PKT3(PKT3_ACQUIRE_MEM
, 5, 0));
661 radeon_emit(cs
, cp_coher_cntl
); /* CP_COHER_CNTL */
662 radeon_emit(cs
, 0xffffffff); /* CP_COHER_SIZE */
663 radeon_emit(cs
, 0xff); /* CP_COHER_SIZE_HI */
664 radeon_emit(cs
, 0); /* CP_COHER_BASE */
665 radeon_emit(cs
, 0); /* CP_COHER_BASE_HI */
666 radeon_emit(cs
, 0x0000000A); /* POLL_INTERVAL */
668 radeon_emit(cs
, PKT3(PKT3_SURFACE_SYNC
, 3, 0));
669 radeon_emit(cs
, cp_coher_cntl
); /* CP_COHER_CNTL */
670 radeon_emit(cs
, 0xffffffff); /* CP_COHER_SIZE */
671 radeon_emit(cs
, 0); /* CP_COHER_BASE */
672 radeon_emit(cs
, 0x0000000A); /* POLL_INTERVAL */
676 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_CB_META
) {
677 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
678 radeon_emit(cs
, EVENT_TYPE(V_028A90_FLUSH_AND_INV_CB_META
) | EVENT_INDEX(0));
680 if (sctx
->flags
& R600_CONTEXT_FLUSH_AND_INV_DB_META
) {
681 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
682 radeon_emit(cs
, EVENT_TYPE(V_028A90_FLUSH_AND_INV_DB_META
) | EVENT_INDEX(0));
685 if (sctx
->flags
& (R600_CONTEXT_WAIT_3D_IDLE
|
686 R600_CONTEXT_PS_PARTIAL_FLUSH
)) {
687 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
688 radeon_emit(cs
, EVENT_TYPE(V_028A90_PS_PARTIAL_FLUSH
) | EVENT_INDEX(4));
689 } else if (sctx
->flags
& R600_CONTEXT_STREAMOUT_FLUSH
) {
690 /* Needed if streamout buffers are going to be used as a source. */
691 radeon_emit(cs
, PKT3(PKT3_EVENT_WRITE
, 0, 0));
692 radeon_emit(cs
, EVENT_TYPE(V_028A90_VS_PARTIAL_FLUSH
) | EVENT_INDEX(4));
698 const struct r600_atom si_atom_cache_flush
= { si_emit_cache_flush
, 13 }; /* number of CS dwords */
700 void si_draw_vbo(struct pipe_context
*ctx
, const struct pipe_draw_info
*info
)
702 struct si_context
*sctx
= (struct si_context
*)ctx
;
703 struct pipe_index_buffer ib
= {};
706 if (!info
->count
&& (info
->indexed
|| !info
->count_from_stream_output
))
709 if (!sctx
->ps_shader
|| !sctx
->vs_shader
)
712 si_update_derived_state(sctx
);
713 si_vertex_buffer_update(sctx
);
716 /* Initialize the index buffer struct. */
717 pipe_resource_reference(&ib
.buffer
, sctx
->index_buffer
.buffer
);
718 ib
.user_buffer
= sctx
->index_buffer
.user_buffer
;
719 ib
.index_size
= sctx
->index_buffer
.index_size
;
720 ib
.offset
= sctx
->index_buffer
.offset
+ info
->start
* ib
.index_size
;
722 /* Translate or upload, if needed. */
723 if (ib
.index_size
== 1) {
724 struct pipe_resource
*out_buffer
= NULL
;
728 u_upload_alloc(sctx
->b
.uploader
, 0, info
->count
* 2,
729 &out_offset
, &out_buffer
, &ptr
);
731 util_shorten_ubyte_elts_to_userptr(
732 &sctx
->b
.b
, &ib
, 0, ib
.offset
, info
->count
, ptr
);
734 pipe_resource_reference(&ib
.buffer
, NULL
);
735 ib
.user_buffer
= NULL
;
736 ib
.buffer
= out_buffer
;
737 ib
.offset
= out_offset
;
741 if (ib
.user_buffer
&& !ib
.buffer
) {
742 u_upload_data(sctx
->b
.uploader
, 0, info
->count
* ib
.index_size
,
743 ib
.user_buffer
, &ib
.offset
, &ib
.buffer
);
747 if (!si_update_draw_info_state(sctx
, info
, &ib
))
750 si_state_draw(sctx
, info
, &ib
);
752 sctx
->pm4_dirty_cdwords
+= si_pm4_dirty_dw(sctx
);
754 /* Check flush flags. */
756 sctx
->atoms
.cache_flush
->dirty
= true;
758 si_need_cs_space(sctx
, 0, TRUE
);
761 for (i
= 0; i
< SI_NUM_ATOMS(sctx
); i
++) {
762 if (sctx
->atoms
.array
[i
]->dirty
) {
763 sctx
->atoms
.array
[i
]->emit(&sctx
->b
, sctx
->atoms
.array
[i
]);
764 sctx
->atoms
.array
[i
]->dirty
= false;
768 si_pm4_emit_dirty(sctx
);
769 sctx
->pm4_dirty_cdwords
= 0;
772 if (sctx
->screen
->b
.trace_bo
) {
777 /* Set the depth buffer as dirty. */
778 if (sctx
->framebuffer
.zsbuf
) {
779 struct pipe_surface
*surf
= sctx
->framebuffer
.zsbuf
;
780 struct r600_texture
*rtex
= (struct r600_texture
*)surf
->texture
;
782 rtex
->dirty_level_mask
|= 1 << surf
->u
.tex
.level
;
784 if (sctx
->fb_compressed_cb_mask
) {
785 struct pipe_surface
*surf
;
786 struct r600_texture
*rtex
;
787 unsigned mask
= sctx
->fb_compressed_cb_mask
;
790 unsigned i
= u_bit_scan(&mask
);
791 surf
= sctx
->framebuffer
.cbufs
[i
];
792 rtex
= (struct r600_texture
*)surf
->texture
;
794 rtex
->dirty_level_mask
|= 1 << surf
->u
.tex
.level
;
798 pipe_resource_reference(&ib
.buffer
, NULL
);
799 sctx
->b
.num_draw_calls
++;