2 * Copyright 2010 Red Hat Inc.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE.
24 * Authors: Dave Airlie <airlied@redhat.com>
25 * Jerome Glisse <jglisse@redhat.com>
27 #include "r600_formats.h"
28 #include "r600_shader.h"
31 #include "util/u_draw_quad.h"
32 #include "util/u_index_modify.h"
33 #include "util/u_upload_mgr.h"
34 #include "tgsi/tgsi_parse.h"
37 #define R600_PRIM_RECTANGLE_LIST PIPE_PRIM_MAX
39 void r600_init_command_buffer(struct r600_command_buffer
*cb
, unsigned num_dw
)
41 cb
->buf
= CALLOC(1, 4 * num_dw
);
42 cb
->max_num_dw
= num_dw
;
45 void r600_release_command_buffer(struct r600_command_buffer
*cb
)
50 void r600_init_atom(struct r600_context
*rctx
,
51 struct r600_atom
*atom
,
53 void (*emit
)(struct r600_context
*ctx
, struct r600_atom
*state
),
56 assert(id
< R600_NUM_ATOMS
);
57 assert(rctx
->atoms
[id
] == NULL
);
58 rctx
->atoms
[id
] = atom
;
61 atom
->num_dw
= num_dw
;
65 void r600_emit_cso_state(struct r600_context
*rctx
, struct r600_atom
*atom
)
67 r600_emit_command_buffer(rctx
->cs
, ((struct r600_cso_state
*)atom
)->cb
);
70 void r600_emit_alphatest_state(struct r600_context
*rctx
, struct r600_atom
*atom
)
72 struct radeon_winsys_cs
*cs
= rctx
->cs
;
73 struct r600_alphatest_state
*a
= (struct r600_alphatest_state
*)atom
;
74 unsigned alpha_ref
= a
->sx_alpha_ref
;
76 if (rctx
->chip_class
>= EVERGREEN
&& a
->cb0_export_16bpc
) {
80 r600_write_context_reg(cs
, R_028410_SX_ALPHA_TEST_CONTROL
,
81 a
->sx_alpha_test_control
|
82 S_028410_ALPHA_TEST_BYPASS(a
->bypass
));
83 r600_write_context_reg(cs
, R_028438_SX_ALPHA_REF
, alpha_ref
);
86 static void r600_texture_barrier(struct pipe_context
*ctx
)
88 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
90 rctx
->flags
|= R600_CONTEXT_CB_FLUSH
| R600_CONTEXT_TEX_FLUSH
;
93 if (rctx
->chip_class
== R600
) {
94 rctx
->flags
|= R600_CONTEXT_FLUSH_AND_INV
;
98 static unsigned r600_conv_pipe_prim(unsigned prim
)
100 static const unsigned prim_conv
[] = {
101 V_008958_DI_PT_POINTLIST
,
102 V_008958_DI_PT_LINELIST
,
103 V_008958_DI_PT_LINELOOP
,
104 V_008958_DI_PT_LINESTRIP
,
105 V_008958_DI_PT_TRILIST
,
106 V_008958_DI_PT_TRISTRIP
,
107 V_008958_DI_PT_TRIFAN
,
108 V_008958_DI_PT_QUADLIST
,
109 V_008958_DI_PT_QUADSTRIP
,
110 V_008958_DI_PT_POLYGON
,
111 V_008958_DI_PT_LINELIST_ADJ
,
112 V_008958_DI_PT_LINESTRIP_ADJ
,
113 V_008958_DI_PT_TRILIST_ADJ
,
114 V_008958_DI_PT_TRISTRIP_ADJ
,
115 V_008958_DI_PT_RECTLIST
117 return prim_conv
[prim
];
120 /* common state between evergreen and r600 */
122 static void r600_bind_blend_state_internal(struct r600_context
*rctx
,
123 struct r600_blend_state
*blend
, bool blend_disable
)
125 unsigned color_control
;
126 bool update_cb
= false;
128 rctx
->alpha_to_one
= blend
->alpha_to_one
;
129 rctx
->dual_src_blend
= blend
->dual_src_blend
;
131 if (!blend_disable
) {
132 r600_set_cso_state_with_cb(&rctx
->blend_state
, blend
, &blend
->buffer
);
133 color_control
= blend
->cb_color_control
;
135 /* Blending is disabled. */
136 r600_set_cso_state_with_cb(&rctx
->blend_state
, blend
, &blend
->buffer_no_blend
);
137 color_control
= blend
->cb_color_control_no_blend
;
140 /* Update derived states. */
141 if (rctx
->cb_misc_state
.blend_colormask
!= blend
->cb_target_mask
) {
142 rctx
->cb_misc_state
.blend_colormask
= blend
->cb_target_mask
;
145 if (rctx
->chip_class
<= R700
&&
146 rctx
->cb_misc_state
.cb_color_control
!= color_control
) {
147 rctx
->cb_misc_state
.cb_color_control
= color_control
;
150 if (rctx
->cb_misc_state
.dual_src_blend
!= blend
->dual_src_blend
) {
151 rctx
->cb_misc_state
.dual_src_blend
= blend
->dual_src_blend
;
155 rctx
->cb_misc_state
.atom
.dirty
= true;
159 static void r600_bind_blend_state(struct pipe_context
*ctx
, void *state
)
161 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
162 struct r600_blend_state
*blend
= (struct r600_blend_state
*)state
;
167 r600_bind_blend_state_internal(rctx
, blend
, rctx
->force_blend_disable
);
170 static void r600_set_blend_color(struct pipe_context
*ctx
,
171 const struct pipe_blend_color
*state
)
173 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
175 rctx
->blend_color
.state
= *state
;
176 rctx
->blend_color
.atom
.dirty
= true;
179 void r600_emit_blend_color(struct r600_context
*rctx
, struct r600_atom
*atom
)
181 struct radeon_winsys_cs
*cs
= rctx
->cs
;
182 struct pipe_blend_color
*state
= &rctx
->blend_color
.state
;
184 r600_write_context_reg_seq(cs
, R_028414_CB_BLEND_RED
, 4);
185 r600_write_value(cs
, fui(state
->color
[0])); /* R_028414_CB_BLEND_RED */
186 r600_write_value(cs
, fui(state
->color
[1])); /* R_028418_CB_BLEND_GREEN */
187 r600_write_value(cs
, fui(state
->color
[2])); /* R_02841C_CB_BLEND_BLUE */
188 r600_write_value(cs
, fui(state
->color
[3])); /* R_028420_CB_BLEND_ALPHA */
191 void r600_emit_vgt_state(struct r600_context
*rctx
, struct r600_atom
*atom
)
193 struct radeon_winsys_cs
*cs
= rctx
->cs
;
194 struct r600_vgt_state
*a
= (struct r600_vgt_state
*)atom
;
196 r600_write_context_reg(cs
, R_028A94_VGT_MULTI_PRIM_IB_RESET_EN
, a
->vgt_multi_prim_ib_reset_en
);
197 r600_write_context_reg(cs
, R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX
, a
->vgt_multi_prim_ib_reset_indx
);
200 void r600_emit_vgt2_state(struct r600_context
*rctx
, struct r600_atom
*atom
)
202 struct radeon_winsys_cs
*cs
= rctx
->cs
;
203 struct r600_vgt2_state
*a
= (struct r600_vgt2_state
*)atom
;
205 r600_write_context_reg(cs
, R_028408_VGT_INDX_OFFSET
, a
->vgt_indx_offset
);
208 static void r600_set_clip_state(struct pipe_context
*ctx
,
209 const struct pipe_clip_state
*state
)
211 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
212 struct pipe_constant_buffer cb
;
214 rctx
->clip_state
.state
= *state
;
215 rctx
->clip_state
.atom
.dirty
= true;
218 cb
.user_buffer
= state
->ucp
;
219 cb
.buffer_offset
= 0;
220 cb
.buffer_size
= 4*4*8;
221 ctx
->set_constant_buffer(ctx
, PIPE_SHADER_VERTEX
, R600_UCP_CONST_BUFFER
, &cb
);
222 pipe_resource_reference(&cb
.buffer
, NULL
);
225 static void r600_set_stencil_ref(struct pipe_context
*ctx
,
226 const struct r600_stencil_ref
*state
)
228 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
230 rctx
->stencil_ref
.state
= *state
;
231 rctx
->stencil_ref
.atom
.dirty
= true;
234 void r600_emit_stencil_ref(struct r600_context
*rctx
, struct r600_atom
*atom
)
236 struct radeon_winsys_cs
*cs
= rctx
->cs
;
237 struct r600_stencil_ref_state
*a
= (struct r600_stencil_ref_state
*)atom
;
239 r600_write_context_reg_seq(cs
, R_028430_DB_STENCILREFMASK
, 2);
240 r600_write_value(cs
, /* R_028430_DB_STENCILREFMASK */
241 S_028430_STENCILREF(a
->state
.ref_value
[0]) |
242 S_028430_STENCILMASK(a
->state
.valuemask
[0]) |
243 S_028430_STENCILWRITEMASK(a
->state
.writemask
[0]));
244 r600_write_value(cs
, /* R_028434_DB_STENCILREFMASK_BF */
245 S_028434_STENCILREF_BF(a
->state
.ref_value
[1]) |
246 S_028434_STENCILMASK_BF(a
->state
.valuemask
[1]) |
247 S_028434_STENCILWRITEMASK_BF(a
->state
.writemask
[1]));
250 static void r600_set_pipe_stencil_ref(struct pipe_context
*ctx
,
251 const struct pipe_stencil_ref
*state
)
253 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
254 struct r600_dsa_state
*dsa
= (struct r600_dsa_state
*)rctx
->dsa_state
.cso
;
255 struct r600_stencil_ref ref
;
257 rctx
->stencil_ref
.pipe_state
= *state
;
262 ref
.ref_value
[0] = state
->ref_value
[0];
263 ref
.ref_value
[1] = state
->ref_value
[1];
264 ref
.valuemask
[0] = dsa
->valuemask
[0];
265 ref
.valuemask
[1] = dsa
->valuemask
[1];
266 ref
.writemask
[0] = dsa
->writemask
[0];
267 ref
.writemask
[1] = dsa
->writemask
[1];
269 r600_set_stencil_ref(ctx
, &ref
);
272 static void r600_bind_dsa_state(struct pipe_context
*ctx
, void *state
)
274 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
275 struct r600_dsa_state
*dsa
= state
;
276 struct r600_stencil_ref ref
;
281 r600_set_cso_state_with_cb(&rctx
->dsa_state
, dsa
, &dsa
->buffer
);
283 ref
.ref_value
[0] = rctx
->stencil_ref
.pipe_state
.ref_value
[0];
284 ref
.ref_value
[1] = rctx
->stencil_ref
.pipe_state
.ref_value
[1];
285 ref
.valuemask
[0] = dsa
->valuemask
[0];
286 ref
.valuemask
[1] = dsa
->valuemask
[1];
287 ref
.writemask
[0] = dsa
->writemask
[0];
288 ref
.writemask
[1] = dsa
->writemask
[1];
290 r600_set_stencil_ref(ctx
, &ref
);
292 /* Update alphatest state. */
293 if (rctx
->alphatest_state
.sx_alpha_test_control
!= dsa
->sx_alpha_test_control
||
294 rctx
->alphatest_state
.sx_alpha_ref
!= dsa
->alpha_ref
) {
295 rctx
->alphatest_state
.sx_alpha_test_control
= dsa
->sx_alpha_test_control
;
296 rctx
->alphatest_state
.sx_alpha_ref
= dsa
->alpha_ref
;
297 rctx
->alphatest_state
.atom
.dirty
= true;
301 static void r600_bind_rs_state(struct pipe_context
*ctx
, void *state
)
303 struct r600_rasterizer_state
*rs
= (struct r600_rasterizer_state
*)state
;
304 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
309 rctx
->rasterizer
= rs
;
311 r600_set_cso_state_with_cb(&rctx
->rasterizer_state
, rs
, &rs
->buffer
);
313 if (rs
->offset_enable
&&
314 (rs
->offset_units
!= rctx
->poly_offset_state
.offset_units
||
315 rs
->offset_scale
!= rctx
->poly_offset_state
.offset_scale
)) {
316 rctx
->poly_offset_state
.offset_units
= rs
->offset_units
;
317 rctx
->poly_offset_state
.offset_scale
= rs
->offset_scale
;
318 rctx
->poly_offset_state
.atom
.dirty
= true;
321 /* Update clip_misc_state. */
322 if (rctx
->clip_misc_state
.pa_cl_clip_cntl
!= rs
->pa_cl_clip_cntl
||
323 rctx
->clip_misc_state
.clip_plane_enable
!= rs
->clip_plane_enable
) {
324 rctx
->clip_misc_state
.pa_cl_clip_cntl
= rs
->pa_cl_clip_cntl
;
325 rctx
->clip_misc_state
.clip_plane_enable
= rs
->clip_plane_enable
;
326 rctx
->clip_misc_state
.atom
.dirty
= true;
329 /* Workaround for a missing scissor enable on r600. */
330 if (rctx
->chip_class
== R600
&&
331 rs
->scissor_enable
!= rctx
->scissor
.enable
) {
332 rctx
->scissor
.enable
= rs
->scissor_enable
;
333 rctx
->scissor
.atom
.dirty
= true;
336 /* Re-emit PA_SC_LINE_STIPPLE. */
337 rctx
->last_primitive_type
= -1;
340 static void r600_delete_rs_state(struct pipe_context
*ctx
, void *state
)
342 struct r600_rasterizer_state
*rs
= (struct r600_rasterizer_state
*)state
;
344 r600_release_command_buffer(&rs
->buffer
);
348 static void r600_sampler_view_destroy(struct pipe_context
*ctx
,
349 struct pipe_sampler_view
*state
)
351 struct r600_pipe_sampler_view
*resource
= (struct r600_pipe_sampler_view
*)state
;
353 pipe_resource_reference(&state
->texture
, NULL
);
357 void r600_sampler_states_dirty(struct r600_context
*rctx
,
358 struct r600_sampler_states
*state
)
360 if (state
->dirty_mask
) {
361 if (state
->dirty_mask
& state
->has_bordercolor_mask
) {
362 rctx
->flags
|= R600_CONTEXT_PS_PARTIAL_FLUSH
;
365 util_bitcount(state
->dirty_mask
& state
->has_bordercolor_mask
) * 11 +
366 util_bitcount(state
->dirty_mask
& ~state
->has_bordercolor_mask
) * 5;
367 state
->atom
.dirty
= true;
371 static void r600_bind_sampler_states(struct pipe_context
*pipe
,
374 unsigned count
, void **states
)
376 struct r600_context
*rctx
= (struct r600_context
*)pipe
;
377 struct r600_textures_info
*dst
= &rctx
->samplers
[shader
];
378 struct r600_pipe_sampler_state
**rstates
= (struct r600_pipe_sampler_state
**)states
;
379 int seamless_cube_map
= -1;
381 /* This sets 1-bit for states with index >= count. */
382 uint32_t disable_mask
= ~((1ull << count
) - 1);
383 /* These are the new states set by this function. */
384 uint32_t new_mask
= 0;
386 assert(start
== 0); /* XXX fix below */
388 for (i
= 0; i
< count
; i
++) {
389 struct r600_pipe_sampler_state
*rstate
= rstates
[i
];
391 if (rstate
== dst
->states
.states
[i
]) {
396 if (rstate
->border_color_use
) {
397 dst
->states
.has_bordercolor_mask
|= 1 << i
;
399 dst
->states
.has_bordercolor_mask
&= ~(1 << i
);
401 seamless_cube_map
= rstate
->seamless_cube_map
;
405 disable_mask
|= 1 << i
;
409 memcpy(dst
->states
.states
, rstates
, sizeof(void*) * count
);
410 memset(dst
->states
.states
+ count
, 0, sizeof(void*) * (NUM_TEX_UNITS
- count
));
412 dst
->states
.enabled_mask
&= ~disable_mask
;
413 dst
->states
.dirty_mask
&= dst
->states
.enabled_mask
;
414 dst
->states
.enabled_mask
|= new_mask
;
415 dst
->states
.dirty_mask
|= new_mask
;
416 dst
->states
.has_bordercolor_mask
&= dst
->states
.enabled_mask
;
418 r600_sampler_states_dirty(rctx
, &dst
->states
);
420 /* Seamless cubemap state. */
421 if (rctx
->chip_class
<= R700
&&
422 seamless_cube_map
!= -1 &&
423 seamless_cube_map
!= rctx
->seamless_cube_map
.enabled
) {
424 /* change in TA_CNTL_AUX need a pipeline flush */
425 rctx
->flags
|= R600_CONTEXT_PS_PARTIAL_FLUSH
;
426 rctx
->seamless_cube_map
.enabled
= seamless_cube_map
;
427 rctx
->seamless_cube_map
.atom
.dirty
= true;
431 static void r600_bind_vs_sampler_states(struct pipe_context
*ctx
, unsigned count
, void **states
)
433 r600_bind_sampler_states(ctx
, PIPE_SHADER_VERTEX
, 0, count
, states
);
436 static void r600_bind_ps_sampler_states(struct pipe_context
*ctx
, unsigned count
, void **states
)
438 r600_bind_sampler_states(ctx
, PIPE_SHADER_FRAGMENT
, 0, count
, states
);
441 static void r600_delete_sampler_state(struct pipe_context
*ctx
, void *state
)
446 static void r600_delete_blend_state(struct pipe_context
*ctx
, void *state
)
448 struct r600_blend_state
*blend
= (struct r600_blend_state
*)state
;
450 r600_release_command_buffer(&blend
->buffer
);
451 r600_release_command_buffer(&blend
->buffer_no_blend
);
455 static void r600_delete_dsa_state(struct pipe_context
*ctx
, void *state
)
457 struct r600_dsa_state
*dsa
= (struct r600_dsa_state
*)state
;
459 r600_release_command_buffer(&dsa
->buffer
);
463 static void r600_bind_vertex_elements(struct pipe_context
*ctx
, void *state
)
465 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
467 r600_set_cso_state(&rctx
->vertex_fetch_shader
, state
);
470 static void r600_delete_vertex_elements(struct pipe_context
*ctx
, void *state
)
472 pipe_resource_reference((struct pipe_resource
**)&state
, NULL
);
475 static void r600_set_index_buffer(struct pipe_context
*ctx
,
476 const struct pipe_index_buffer
*ib
)
478 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
481 pipe_resource_reference(&rctx
->index_buffer
.buffer
, ib
->buffer
);
482 memcpy(&rctx
->index_buffer
, ib
, sizeof(*ib
));
484 pipe_resource_reference(&rctx
->index_buffer
.buffer
, NULL
);
488 void r600_vertex_buffers_dirty(struct r600_context
*rctx
)
490 if (rctx
->vertex_buffer_state
.dirty_mask
) {
491 rctx
->flags
|= rctx
->has_vertex_cache
? R600_CONTEXT_VTX_FLUSH
: R600_CONTEXT_TEX_FLUSH
;
492 rctx
->vertex_buffer_state
.atom
.num_dw
= (rctx
->chip_class
>= EVERGREEN
? 12 : 11) *
493 util_bitcount(rctx
->vertex_buffer_state
.dirty_mask
);
494 rctx
->vertex_buffer_state
.atom
.dirty
= true;
498 static void r600_set_vertex_buffers(struct pipe_context
*ctx
,
499 unsigned start_slot
, unsigned count
,
500 const struct pipe_vertex_buffer
*input
)
502 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
503 struct r600_vertexbuf_state
*state
= &rctx
->vertex_buffer_state
;
504 struct pipe_vertex_buffer
*vb
= state
->vb
+ start_slot
;
506 uint32_t disable_mask
= 0;
507 /* These are the new buffers set by this function. */
508 uint32_t new_buffer_mask
= 0;
510 /* Set vertex buffers. */
512 for (i
= 0; i
< count
; i
++) {
513 if (memcmp(&input
[i
], &vb
[i
], sizeof(struct pipe_vertex_buffer
))) {
514 if (input
[i
].buffer
) {
515 vb
[i
].stride
= input
[i
].stride
;
516 vb
[i
].buffer_offset
= input
[i
].buffer_offset
;
517 pipe_resource_reference(&vb
[i
].buffer
, input
[i
].buffer
);
518 new_buffer_mask
|= 1 << i
;
520 pipe_resource_reference(&vb
[i
].buffer
, NULL
);
521 disable_mask
|= 1 << i
;
526 for (i
= 0; i
< count
; i
++) {
527 pipe_resource_reference(&vb
[i
].buffer
, NULL
);
529 disable_mask
= ((1ull << count
) - 1);
532 disable_mask
<<= start_slot
;
533 new_buffer_mask
<<= start_slot
;
535 rctx
->vertex_buffer_state
.enabled_mask
&= ~disable_mask
;
536 rctx
->vertex_buffer_state
.dirty_mask
&= rctx
->vertex_buffer_state
.enabled_mask
;
537 rctx
->vertex_buffer_state
.enabled_mask
|= new_buffer_mask
;
538 rctx
->vertex_buffer_state
.dirty_mask
|= new_buffer_mask
;
540 r600_vertex_buffers_dirty(rctx
);
543 void r600_sampler_views_dirty(struct r600_context
*rctx
,
544 struct r600_samplerview_state
*state
)
546 if (state
->dirty_mask
) {
547 rctx
->flags
|= R600_CONTEXT_TEX_FLUSH
;
548 state
->atom
.num_dw
= (rctx
->chip_class
>= EVERGREEN
? 14 : 13) *
549 util_bitcount(state
->dirty_mask
);
550 state
->atom
.dirty
= true;
554 static void r600_set_sampler_views(struct pipe_context
*pipe
, unsigned shader
,
555 unsigned start
, unsigned count
,
556 struct pipe_sampler_view
**views
)
558 struct r600_context
*rctx
= (struct r600_context
*) pipe
;
559 struct r600_textures_info
*dst
= &rctx
->samplers
[shader
];
560 struct r600_pipe_sampler_view
**rviews
= (struct r600_pipe_sampler_view
**)views
;
561 uint32_t dirty_sampler_states_mask
= 0;
563 /* This sets 1-bit for textures with index >= count. */
564 uint32_t disable_mask
= ~((1ull << count
) - 1);
565 /* These are the new textures set by this function. */
566 uint32_t new_mask
= 0;
568 /* Set textures with index >= count to NULL. */
569 uint32_t remaining_mask
;
571 assert(start
== 0); /* XXX fix below */
573 remaining_mask
= dst
->views
.enabled_mask
& disable_mask
;
575 while (remaining_mask
) {
576 i
= u_bit_scan(&remaining_mask
);
577 assert(dst
->views
.views
[i
]);
579 pipe_sampler_view_reference((struct pipe_sampler_view
**)&dst
->views
.views
[i
], NULL
);
582 for (i
= 0; i
< count
; i
++) {
583 if (rviews
[i
] == dst
->views
.views
[i
]) {
588 struct r600_texture
*rtex
=
589 (struct r600_texture
*)rviews
[i
]->base
.texture
;
591 if (rtex
->is_depth
&& !rtex
->is_flushing_texture
) {
592 dst
->views
.compressed_depthtex_mask
|= 1 << i
;
594 dst
->views
.compressed_depthtex_mask
&= ~(1 << i
);
597 /* Track compressed colorbuffers. */
598 if (rtex
->cmask_size
&& rtex
->fmask_size
) {
599 dst
->views
.compressed_colortex_mask
|= 1 << i
;
601 dst
->views
.compressed_colortex_mask
&= ~(1 << i
);
604 /* Changing from array to non-arrays textures and vice versa requires
605 * updating TEX_ARRAY_OVERRIDE in sampler states on R6xx-R7xx. */
606 if (rctx
->chip_class
<= R700
&&
607 (dst
->states
.enabled_mask
& (1 << i
)) &&
608 (rviews
[i
]->base
.texture
->target
== PIPE_TEXTURE_1D_ARRAY
||
609 rviews
[i
]->base
.texture
->target
== PIPE_TEXTURE_2D_ARRAY
) != dst
->is_array_sampler
[i
]) {
610 dirty_sampler_states_mask
|= 1 << i
;
613 pipe_sampler_view_reference((struct pipe_sampler_view
**)&dst
->views
.views
[i
], views
[i
]);
616 pipe_sampler_view_reference((struct pipe_sampler_view
**)&dst
->views
.views
[i
], NULL
);
617 disable_mask
|= 1 << i
;
621 dst
->views
.enabled_mask
&= ~disable_mask
;
622 dst
->views
.dirty_mask
&= dst
->views
.enabled_mask
;
623 dst
->views
.enabled_mask
|= new_mask
;
624 dst
->views
.dirty_mask
|= new_mask
;
625 dst
->views
.compressed_depthtex_mask
&= dst
->views
.enabled_mask
;
626 dst
->views
.compressed_colortex_mask
&= dst
->views
.enabled_mask
;
627 dst
->views
.dirty_txq_constants
= TRUE
;
628 r600_sampler_views_dirty(rctx
, &dst
->views
);
630 if (dirty_sampler_states_mask
) {
631 dst
->states
.dirty_mask
|= dirty_sampler_states_mask
;
632 r600_sampler_states_dirty(rctx
, &dst
->states
);
636 static void r600_set_vs_sampler_views(struct pipe_context
*ctx
, unsigned count
,
637 struct pipe_sampler_view
**views
)
639 r600_set_sampler_views(ctx
, PIPE_SHADER_VERTEX
, 0, count
, views
);
642 static void r600_set_ps_sampler_views(struct pipe_context
*ctx
, unsigned count
,
643 struct pipe_sampler_view
**views
)
645 r600_set_sampler_views(ctx
, PIPE_SHADER_FRAGMENT
, 0, count
, views
);
648 static void r600_set_viewport_state(struct pipe_context
*ctx
,
649 const struct pipe_viewport_state
*state
)
651 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
653 rctx
->viewport
.state
= *state
;
654 rctx
->viewport
.atom
.dirty
= true;
657 void r600_emit_viewport_state(struct r600_context
*rctx
, struct r600_atom
*atom
)
659 struct radeon_winsys_cs
*cs
= rctx
->cs
;
660 struct pipe_viewport_state
*state
= &rctx
->viewport
.state
;
662 r600_write_context_reg_seq(cs
, R_02843C_PA_CL_VPORT_XSCALE_0
, 6);
663 r600_write_value(cs
, fui(state
->scale
[0])); /* R_02843C_PA_CL_VPORT_XSCALE_0 */
664 r600_write_value(cs
, fui(state
->translate
[0])); /* R_028440_PA_CL_VPORT_XOFFSET_0 */
665 r600_write_value(cs
, fui(state
->scale
[1])); /* R_028444_PA_CL_VPORT_YSCALE_0 */
666 r600_write_value(cs
, fui(state
->translate
[1])); /* R_028448_PA_CL_VPORT_YOFFSET_0 */
667 r600_write_value(cs
, fui(state
->scale
[2])); /* R_02844C_PA_CL_VPORT_ZSCALE_0 */
668 r600_write_value(cs
, fui(state
->translate
[2])); /* R_028450_PA_CL_VPORT_ZOFFSET_0 */
671 /* Compute the key for the hw shader variant */
672 static INLINE
struct r600_shader_key
r600_shader_selector_key(struct pipe_context
* ctx
,
673 struct r600_pipe_shader_selector
* sel
)
675 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
676 struct r600_shader_key key
;
677 memset(&key
, 0, sizeof(key
));
679 if (sel
->type
== PIPE_SHADER_FRAGMENT
) {
680 key
.color_two_side
= rctx
->rasterizer
&& rctx
->rasterizer
->two_side
;
681 key
.alpha_to_one
= rctx
->alpha_to_one
&&
682 rctx
->rasterizer
&& rctx
->rasterizer
->multisample_enable
&&
683 !rctx
->framebuffer
.cb0_is_integer
;
684 key
.nr_cbufs
= rctx
->framebuffer
.state
.nr_cbufs
;
685 /* Dual-source blending only makes sense with nr_cbufs == 1. */
686 if (key
.nr_cbufs
== 1 && rctx
->dual_src_blend
)
692 /* Select the hw shader variant depending on the current state.
693 * (*dirty) is set to 1 if current variant was changed */
694 static int r600_shader_select(struct pipe_context
*ctx
,
695 struct r600_pipe_shader_selector
* sel
,
698 struct r600_shader_key key
;
699 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
700 struct r600_pipe_shader
* shader
= NULL
;
703 key
= r600_shader_selector_key(ctx
, sel
);
705 /* Check if we don't need to change anything.
706 * This path is also used for most shaders that don't need multiple
707 * variants, it will cost just a computation of the key and this
709 if (likely(sel
->current
&& memcmp(&sel
->current
->key
, &key
, sizeof(key
)) == 0)) {
713 /* lookup if we have other variants in the list */
714 if (sel
->num_shaders
> 1) {
715 struct r600_pipe_shader
*p
= sel
->current
, *c
= p
->next_variant
;
717 while (c
&& memcmp(&c
->key
, &key
, sizeof(key
)) != 0) {
723 p
->next_variant
= c
->next_variant
;
728 if (unlikely(!shader
)) {
729 shader
= CALLOC(1, sizeof(struct r600_pipe_shader
));
730 shader
->selector
= sel
;
732 r
= r600_pipe_shader_create(ctx
, shader
, key
);
734 R600_ERR("Failed to build shader variant (type=%u) %d\n",
740 /* We don't know the value of nr_ps_max_color_exports until we built
741 * at least one variant, so we may need to recompute the key after
742 * building first variant. */
743 if (sel
->type
== PIPE_SHADER_FRAGMENT
&&
744 sel
->num_shaders
== 0) {
745 sel
->nr_ps_max_color_exports
= shader
->shader
.nr_ps_max_color_exports
;
746 key
= r600_shader_selector_key(ctx
, sel
);
756 shader
->next_variant
= sel
->current
;
757 sel
->current
= shader
;
759 if (rctx
->ps_shader
&&
760 rctx
->cb_misc_state
.nr_ps_color_outputs
!= rctx
->ps_shader
->current
->nr_ps_color_outputs
) {
761 rctx
->cb_misc_state
.nr_ps_color_outputs
= rctx
->ps_shader
->current
->nr_ps_color_outputs
;
762 rctx
->cb_misc_state
.atom
.dirty
= true;
767 static void *r600_create_shader_state(struct pipe_context
*ctx
,
768 const struct pipe_shader_state
*state
,
769 unsigned pipe_shader_type
)
771 struct r600_pipe_shader_selector
*sel
= CALLOC_STRUCT(r600_pipe_shader_selector
);
774 sel
->type
= pipe_shader_type
;
775 sel
->tokens
= tgsi_dup_tokens(state
->tokens
);
776 sel
->so
= state
->stream_output
;
778 r
= r600_shader_select(ctx
, sel
, NULL
);
785 static void *r600_create_ps_state(struct pipe_context
*ctx
,
786 const struct pipe_shader_state
*state
)
788 return r600_create_shader_state(ctx
, state
, PIPE_SHADER_FRAGMENT
);
791 static void *r600_create_vs_state(struct pipe_context
*ctx
,
792 const struct pipe_shader_state
*state
)
794 return r600_create_shader_state(ctx
, state
, PIPE_SHADER_VERTEX
);
797 static void r600_bind_ps_state(struct pipe_context
*ctx
, void *state
)
799 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
802 state
= rctx
->dummy_pixel_shader
;
804 rctx
->ps_shader
= (struct r600_pipe_shader_selector
*)state
;
805 r600_context_pipe_state_set(rctx
, &rctx
->ps_shader
->current
->rstate
);
807 if (rctx
->chip_class
<= R700
) {
808 bool multiwrite
= rctx
->ps_shader
->current
->shader
.fs_write_all
;
810 if (rctx
->cb_misc_state
.multiwrite
!= multiwrite
) {
811 rctx
->cb_misc_state
.multiwrite
= multiwrite
;
812 rctx
->cb_misc_state
.atom
.dirty
= true;
816 if (rctx
->cb_misc_state
.nr_ps_color_outputs
!= rctx
->ps_shader
->current
->nr_ps_color_outputs
) {
817 rctx
->cb_misc_state
.nr_ps_color_outputs
= rctx
->ps_shader
->current
->nr_ps_color_outputs
;
818 rctx
->cb_misc_state
.atom
.dirty
= true;
821 if (rctx
->chip_class
>= EVERGREEN
) {
822 evergreen_update_db_shader_control(rctx
);
824 r600_update_db_shader_control(rctx
);
828 static void r600_bind_vs_state(struct pipe_context
*ctx
, void *state
)
830 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
832 rctx
->vs_shader
= (struct r600_pipe_shader_selector
*)state
;
834 r600_context_pipe_state_set(rctx
, &rctx
->vs_shader
->current
->rstate
);
836 /* Update clip misc state. */
837 if (rctx
->vs_shader
->current
->pa_cl_vs_out_cntl
!= rctx
->clip_misc_state
.pa_cl_vs_out_cntl
||
838 rctx
->vs_shader
->current
->shader
.clip_dist_write
!= rctx
->clip_misc_state
.clip_dist_write
) {
839 rctx
->clip_misc_state
.pa_cl_vs_out_cntl
= rctx
->vs_shader
->current
->pa_cl_vs_out_cntl
;
840 rctx
->clip_misc_state
.clip_dist_write
= rctx
->vs_shader
->current
->shader
.clip_dist_write
;
841 rctx
->clip_misc_state
.atom
.dirty
= true;
846 static void r600_delete_shader_selector(struct pipe_context
*ctx
,
847 struct r600_pipe_shader_selector
*sel
)
849 struct r600_pipe_shader
*p
= sel
->current
, *c
;
852 r600_pipe_shader_destroy(ctx
, p
);
862 static void r600_delete_ps_state(struct pipe_context
*ctx
, void *state
)
864 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
865 struct r600_pipe_shader_selector
*sel
= (struct r600_pipe_shader_selector
*)state
;
867 if (rctx
->ps_shader
== sel
) {
868 rctx
->ps_shader
= NULL
;
871 r600_delete_shader_selector(ctx
, sel
);
874 static void r600_delete_vs_state(struct pipe_context
*ctx
, void *state
)
876 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
877 struct r600_pipe_shader_selector
*sel
= (struct r600_pipe_shader_selector
*)state
;
879 if (rctx
->vs_shader
== sel
) {
880 rctx
->vs_shader
= NULL
;
883 r600_delete_shader_selector(ctx
, sel
);
886 void r600_constant_buffers_dirty(struct r600_context
*rctx
, struct r600_constbuf_state
*state
)
888 if (state
->dirty_mask
) {
889 rctx
->flags
|= R600_CONTEXT_SHADERCONST_FLUSH
;
890 state
->atom
.num_dw
= rctx
->chip_class
>= EVERGREEN
? util_bitcount(state
->dirty_mask
)*20
891 : util_bitcount(state
->dirty_mask
)*19;
892 state
->atom
.dirty
= true;
896 static void r600_set_constant_buffer(struct pipe_context
*ctx
, uint shader
, uint index
,
897 struct pipe_constant_buffer
*input
)
899 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
900 struct r600_constbuf_state
*state
= &rctx
->constbuf_state
[shader
];
901 struct pipe_constant_buffer
*cb
;
904 /* Note that the state tracker can unbind constant buffers by
907 if (unlikely(!input
)) {
908 state
->enabled_mask
&= ~(1 << index
);
909 state
->dirty_mask
&= ~(1 << index
);
910 pipe_resource_reference(&state
->cb
[index
].buffer
, NULL
);
914 cb
= &state
->cb
[index
];
915 cb
->buffer_size
= input
->buffer_size
;
917 ptr
= input
->user_buffer
;
920 /* Upload the user buffer. */
921 if (R600_BIG_ENDIAN
) {
923 unsigned i
, size
= input
->buffer_size
;
925 if (!(tmpPtr
= malloc(size
))) {
926 R600_ERR("Failed to allocate BE swap buffer.\n");
930 for (i
= 0; i
< size
/ 4; ++i
) {
931 tmpPtr
[i
] = bswap_32(((uint32_t *)ptr
)[i
]);
934 u_upload_data(rctx
->uploader
, 0, size
, tmpPtr
, &cb
->buffer_offset
, &cb
->buffer
);
937 u_upload_data(rctx
->uploader
, 0, input
->buffer_size
, ptr
, &cb
->buffer_offset
, &cb
->buffer
);
940 /* Setup the hw buffer. */
941 cb
->buffer_offset
= input
->buffer_offset
;
942 pipe_resource_reference(&cb
->buffer
, input
->buffer
);
945 state
->enabled_mask
|= 1 << index
;
946 state
->dirty_mask
|= 1 << index
;
947 r600_constant_buffers_dirty(rctx
, state
);
950 static struct pipe_stream_output_target
*
951 r600_create_so_target(struct pipe_context
*ctx
,
952 struct pipe_resource
*buffer
,
953 unsigned buffer_offset
,
954 unsigned buffer_size
)
956 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
957 struct r600_so_target
*t
;
960 t
= CALLOC_STRUCT(r600_so_target
);
965 t
->b
.reference
.count
= 1;
967 pipe_resource_reference(&t
->b
.buffer
, buffer
);
968 t
->b
.buffer_offset
= buffer_offset
;
969 t
->b
.buffer_size
= buffer_size
;
971 t
->filled_size
= (struct r600_resource
*)
972 pipe_buffer_create(ctx
->screen
, PIPE_BIND_CUSTOM
, PIPE_USAGE_STATIC
, 4);
973 ptr
= rctx
->ws
->buffer_map(t
->filled_size
->cs_buf
, rctx
->cs
, PIPE_TRANSFER_WRITE
);
974 memset(ptr
, 0, t
->filled_size
->buf
->size
);
975 rctx
->ws
->buffer_unmap(t
->filled_size
->cs_buf
);
980 static void r600_so_target_destroy(struct pipe_context
*ctx
,
981 struct pipe_stream_output_target
*target
)
983 struct r600_so_target
*t
= (struct r600_so_target
*)target
;
984 pipe_resource_reference(&t
->b
.buffer
, NULL
);
985 pipe_resource_reference((struct pipe_resource
**)&t
->filled_size
, NULL
);
989 static void r600_set_so_targets(struct pipe_context
*ctx
,
990 unsigned num_targets
,
991 struct pipe_stream_output_target
**targets
,
992 unsigned append_bitmask
)
994 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
997 /* Stop streamout. */
998 if (rctx
->num_so_targets
&& !rctx
->streamout_start
) {
999 r600_context_streamout_end(rctx
);
1002 /* Set the new targets. */
1003 for (i
= 0; i
< num_targets
; i
++) {
1004 pipe_so_target_reference((struct pipe_stream_output_target
**)&rctx
->so_targets
[i
], targets
[i
]);
1006 for (; i
< rctx
->num_so_targets
; i
++) {
1007 pipe_so_target_reference((struct pipe_stream_output_target
**)&rctx
->so_targets
[i
], NULL
);
1010 rctx
->num_so_targets
= num_targets
;
1011 rctx
->streamout_start
= num_targets
!= 0;
1012 rctx
->streamout_append_bitmask
= append_bitmask
;
1015 static void r600_set_sample_mask(struct pipe_context
*pipe
, unsigned sample_mask
)
1017 struct r600_context
*rctx
= (struct r600_context
*)pipe
;
1019 if (rctx
->sample_mask
.sample_mask
== (uint16_t)sample_mask
)
1022 rctx
->sample_mask
.sample_mask
= sample_mask
;
1023 rctx
->sample_mask
.atom
.dirty
= true;
1026 static void r600_setup_txq_cube_array_constants(struct r600_context
*rctx
, int shader_type
)
1028 struct r600_textures_info
*samplers
= &rctx
->samplers
[shader_type
];
1030 uint32_t array_size
;
1031 struct pipe_constant_buffer cb
;
1034 if (!samplers
->views
.dirty_txq_constants
)
1037 samplers
->views
.dirty_txq_constants
= FALSE
;
1039 bits
= util_last_bit(samplers
->views
.enabled_mask
);
1040 array_size
= bits
* sizeof(uint32_t) * 4;
1041 samplers
->txq_constants
= realloc(samplers
->txq_constants
, array_size
);
1042 memset(samplers
->txq_constants
, 0, array_size
);
1043 for (i
= 0; i
< bits
; i
++)
1044 if (samplers
->views
.enabled_mask
& (1 << i
))
1045 samplers
->txq_constants
[i
] = samplers
->views
.views
[i
]->base
.texture
->array_size
/ 6;
1048 cb
.user_buffer
= samplers
->txq_constants
;
1049 cb
.buffer_offset
= 0;
1050 cb
.buffer_size
= array_size
;
1051 rctx
->context
.set_constant_buffer(&rctx
->context
, shader_type
, R600_TXQ_CONST_BUFFER
, &cb
);
1052 pipe_resource_reference(&cb
.buffer
, NULL
);
1055 static bool r600_update_derived_state(struct r600_context
*rctx
)
1057 struct pipe_context
* ctx
= (struct pipe_context
*)rctx
;
1058 unsigned ps_dirty
= 0;
1061 if (!rctx
->blitter
->running
) {
1064 /* Decompress textures if needed. */
1065 for (i
= 0; i
< PIPE_SHADER_TYPES
; i
++) {
1066 struct r600_samplerview_state
*views
= &rctx
->samplers
[i
].views
;
1067 if (views
->compressed_depthtex_mask
) {
1068 r600_decompress_depth_textures(rctx
, views
);
1070 if (views
->compressed_colortex_mask
) {
1071 r600_decompress_color_textures(rctx
, views
);
1076 r600_shader_select(ctx
, rctx
->ps_shader
, &ps_dirty
);
1078 if (rctx
->ps_shader
&& rctx
->rasterizer
&&
1079 ((rctx
->rasterizer
->sprite_coord_enable
!= rctx
->ps_shader
->current
->sprite_coord_enable
) ||
1080 (rctx
->rasterizer
->flatshade
!= rctx
->ps_shader
->current
->flatshade
))) {
1082 if (rctx
->chip_class
>= EVERGREEN
)
1083 evergreen_pipe_shader_ps(ctx
, rctx
->ps_shader
->current
);
1085 r600_pipe_shader_ps(ctx
, rctx
->ps_shader
->current
);
1091 r600_context_pipe_state_set(rctx
, &rctx
->ps_shader
->current
->rstate
);
1093 if (rctx
->ps_shader
&& rctx
->ps_shader
->current
->shader
.has_txq_cube_array_z_comp
)
1094 r600_setup_txq_cube_array_constants(rctx
, PIPE_SHADER_FRAGMENT
);
1095 if (rctx
->vs_shader
&& rctx
->vs_shader
->current
->shader
.has_txq_cube_array_z_comp
)
1096 r600_setup_txq_cube_array_constants(rctx
, PIPE_SHADER_VERTEX
);
1098 if (rctx
->chip_class
< EVERGREEN
&& rctx
->ps_shader
&& rctx
->vs_shader
) {
1099 if (!r600_adjust_gprs(rctx
)) {
1100 /* discard rendering */
1105 blend_disable
= (rctx
->dual_src_blend
&&
1106 rctx
->ps_shader
->current
->nr_ps_color_outputs
< 2);
1108 if (blend_disable
!= rctx
->force_blend_disable
) {
1109 rctx
->force_blend_disable
= blend_disable
;
1110 r600_bind_blend_state_internal(rctx
,
1111 rctx
->blend_state
.cso
,
1117 static unsigned r600_conv_prim_to_gs_out(unsigned mode
)
1119 static const int prim_conv
[] = {
1120 V_028A6C_OUTPRIM_TYPE_POINTLIST
,
1121 V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
1122 V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
1123 V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
1124 V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
1125 V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
1126 V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
1127 V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
1128 V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
1129 V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
1130 V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
1131 V_028A6C_OUTPRIM_TYPE_LINESTRIP
,
1132 V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
1133 V_028A6C_OUTPRIM_TYPE_TRISTRIP
,
1134 V_028A6C_OUTPRIM_TYPE_TRISTRIP
1136 assert(mode
< Elements(prim_conv
));
1138 return prim_conv
[mode
];
1141 void r600_emit_clip_misc_state(struct r600_context
*rctx
, struct r600_atom
*atom
)
1143 struct radeon_winsys_cs
*cs
= rctx
->cs
;
1144 struct r600_clip_misc_state
*state
= &rctx
->clip_misc_state
;
1146 r600_write_context_reg(cs
, R_028810_PA_CL_CLIP_CNTL
,
1147 state
->pa_cl_clip_cntl
|
1148 (state
->clip_dist_write
? 0 : state
->clip_plane_enable
& 0x3F));
1149 r600_write_context_reg(cs
, R_02881C_PA_CL_VS_OUT_CNTL
,
1150 state
->pa_cl_vs_out_cntl
|
1151 (state
->clip_plane_enable
& state
->clip_dist_write
));
1154 static void r600_draw_vbo(struct pipe_context
*ctx
, const struct pipe_draw_info
*dinfo
)
1156 struct r600_context
*rctx
= (struct r600_context
*)ctx
;
1157 struct pipe_draw_info info
= *dinfo
;
1158 struct pipe_index_buffer ib
= {};
1160 struct r600_block
*dirty_block
= NULL
, *next_block
= NULL
;
1161 struct radeon_winsys_cs
*cs
= rctx
->cs
;
1163 if (!info
.count
&& (info
.indexed
|| !info
.count_from_stream_output
)) {
1168 if (!rctx
->vs_shader
) {
1173 if (!r600_update_derived_state(rctx
)) {
1174 /* useless to render because current rendering command
1181 /* Initialize the index buffer struct. */
1182 pipe_resource_reference(&ib
.buffer
, rctx
->index_buffer
.buffer
);
1183 ib
.user_buffer
= rctx
->index_buffer
.user_buffer
;
1184 ib
.index_size
= rctx
->index_buffer
.index_size
;
1185 ib
.offset
= rctx
->index_buffer
.offset
+ info
.start
* ib
.index_size
;
1187 /* Translate 8-bit indices to 16-bit. */
1188 if (ib
.index_size
== 1) {
1189 struct pipe_resource
*out_buffer
= NULL
;
1190 unsigned out_offset
;
1193 u_upload_alloc(rctx
->uploader
, 0, info
.count
* 2,
1194 &out_offset
, &out_buffer
, &ptr
);
1196 util_shorten_ubyte_elts_to_userptr(
1197 &rctx
->context
, &ib
, 0, ib
.offset
, info
.count
, ptr
);
1199 pipe_resource_reference(&ib
.buffer
, NULL
);
1200 ib
.user_buffer
= NULL
;
1201 ib
.buffer
= out_buffer
;
1202 ib
.offset
= out_offset
;
1206 /* Upload the index buffer.
1207 * The upload is skipped for small index counts on little-endian machines
1208 * and the indices are emitted via PKT3_DRAW_INDEX_IMMD.
1209 * Note: Instanced rendering in combination with immediate indices hangs. */
1210 if (ib
.user_buffer
&& (R600_BIG_ENDIAN
|| info
.instance_count
> 1 ||
1211 info
.count
*ib
.index_size
> 20)) {
1212 u_upload_data(rctx
->uploader
, 0, info
.count
* ib
.index_size
,
1213 ib
.user_buffer
, &ib
.offset
, &ib
.buffer
);
1214 ib
.user_buffer
= NULL
;
1217 info
.index_bias
= info
.start
;
1220 /* Enable stream out if needed. */
1221 if (rctx
->streamout_start
) {
1222 r600_context_streamout_begin(rctx
);
1223 rctx
->streamout_start
= FALSE
;
1226 /* Set the index offset and multi primitive */
1227 if (rctx
->vgt2_state
.vgt_indx_offset
!= info
.index_bias
) {
1228 rctx
->vgt2_state
.vgt_indx_offset
= info
.index_bias
;
1229 rctx
->vgt2_state
.atom
.dirty
= true;
1231 if (rctx
->vgt_state
.vgt_multi_prim_ib_reset_en
!= info
.primitive_restart
||
1232 rctx
->vgt_state
.vgt_multi_prim_ib_reset_indx
!= info
.restart_index
) {
1233 rctx
->vgt_state
.vgt_multi_prim_ib_reset_en
= info
.primitive_restart
;
1234 rctx
->vgt_state
.vgt_multi_prim_ib_reset_indx
= info
.restart_index
;
1235 rctx
->vgt_state
.atom
.dirty
= true;
1239 r600_need_cs_space(rctx
, ib
.user_buffer
? 5 : 0, TRUE
);
1240 r600_flush_emit(rctx
);
1242 for (i
= 0; i
< R600_NUM_ATOMS
; i
++) {
1243 if (rctx
->atoms
[i
] == NULL
|| !rctx
->atoms
[i
]->dirty
) {
1246 r600_emit_atom(rctx
, rctx
->atoms
[i
]);
1248 LIST_FOR_EACH_ENTRY_SAFE(dirty_block
, next_block
, &rctx
->dirty
,list
) {
1249 r600_context_block_emit_dirty(rctx
, dirty_block
, 0 /* pkt_flags */);
1251 rctx
->pm4_dirty_cdwords
= 0;
1253 /* Update start instance. */
1254 if (rctx
->last_start_instance
!= info
.start_instance
) {
1255 r600_write_ctl_const(cs
, R_03CFF4_SQ_VTX_START_INST_LOC
, info
.start_instance
);
1256 rctx
->last_start_instance
= info
.start_instance
;
1259 /* Update the primitive type. */
1260 if (rctx
->last_primitive_type
!= info
.mode
) {
1261 unsigned ls_mask
= 0;
1263 if (info
.mode
== PIPE_PRIM_LINES
)
1265 else if (info
.mode
== PIPE_PRIM_LINE_STRIP
||
1266 info
.mode
== PIPE_PRIM_LINE_LOOP
)
1269 r600_write_context_reg(cs
, R_028A0C_PA_SC_LINE_STIPPLE
,
1270 S_028A0C_AUTO_RESET_CNTL(ls_mask
) |
1271 (rctx
->rasterizer
? rctx
->rasterizer
->pa_sc_line_stipple
: 0));
1272 r600_write_context_reg(cs
, R_028A6C_VGT_GS_OUT_PRIM_TYPE
,
1273 r600_conv_prim_to_gs_out(info
.mode
));
1274 r600_write_config_reg(cs
, R_008958_VGT_PRIMITIVE_TYPE
,
1275 r600_conv_pipe_prim(info
.mode
));
1277 rctx
->last_primitive_type
= info
.mode
;
1281 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_NUM_INSTANCES
, 0, rctx
->predicate_drawing
);
1282 cs
->buf
[cs
->cdw
++] = info
.instance_count
;
1284 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_INDEX_TYPE
, 0, rctx
->predicate_drawing
);
1285 cs
->buf
[cs
->cdw
++] = ib
.index_size
== 4 ?
1286 (VGT_INDEX_32
| (R600_BIG_ENDIAN
? VGT_DMA_SWAP_32_BIT
: 0)) :
1287 (VGT_INDEX_16
| (R600_BIG_ENDIAN
? VGT_DMA_SWAP_16_BIT
: 0));
1289 if (ib
.user_buffer
) {
1290 unsigned size_bytes
= info
.count
*ib
.index_size
;
1291 unsigned size_dw
= align(size_bytes
, 4) / 4;
1292 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_DRAW_INDEX_IMMD
, 1 + size_dw
, rctx
->predicate_drawing
);
1293 cs
->buf
[cs
->cdw
++] = info
.count
;
1294 cs
->buf
[cs
->cdw
++] = V_0287F0_DI_SRC_SEL_IMMEDIATE
;
1295 memcpy(cs
->buf
+cs
->cdw
, ib
.user_buffer
, size_bytes
);
1298 uint64_t va
= r600_resource_va(ctx
->screen
, ib
.buffer
) + ib
.offset
;
1299 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_DRAW_INDEX
, 3, rctx
->predicate_drawing
);
1300 cs
->buf
[cs
->cdw
++] = va
;
1301 cs
->buf
[cs
->cdw
++] = (va
>> 32UL) & 0xFF;
1302 cs
->buf
[cs
->cdw
++] = info
.count
;
1303 cs
->buf
[cs
->cdw
++] = V_0287F0_DI_SRC_SEL_DMA
;
1304 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_NOP
, 0, rctx
->predicate_drawing
);
1305 cs
->buf
[cs
->cdw
++] = r600_context_bo_reloc(rctx
, (struct r600_resource
*)ib
.buffer
, RADEON_USAGE_READ
);
1308 if (info
.count_from_stream_output
) {
1309 struct r600_so_target
*t
= (struct r600_so_target
*)info
.count_from_stream_output
;
1310 uint64_t va
= r600_resource_va(&rctx
->screen
->screen
, (void*)t
->filled_size
);
1312 r600_write_context_reg(cs
, R_028B30_VGT_STRMOUT_DRAW_OPAQUE_VERTEX_STRIDE
, t
->stride_in_dw
);
1314 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_COPY_DW
, 4, 0);
1315 cs
->buf
[cs
->cdw
++] = COPY_DW_SRC_IS_MEM
| COPY_DW_DST_IS_REG
;
1316 cs
->buf
[cs
->cdw
++] = va
& 0xFFFFFFFFUL
; /* src address lo */
1317 cs
->buf
[cs
->cdw
++] = (va
>> 32UL) & 0xFFUL
; /* src address hi */
1318 cs
->buf
[cs
->cdw
++] = R_028B2C_VGT_STRMOUT_DRAW_OPAQUE_BUFFER_FILLED_SIZE
>> 2; /* dst register */
1319 cs
->buf
[cs
->cdw
++] = 0; /* unused */
1321 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_NOP
, 0, 0);
1322 cs
->buf
[cs
->cdw
++] = r600_context_bo_reloc(rctx
, t
->filled_size
, RADEON_USAGE_READ
);
1325 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_DRAW_INDEX_AUTO
, 1, rctx
->predicate_drawing
);
1326 cs
->buf
[cs
->cdw
++] = info
.count
;
1327 cs
->buf
[cs
->cdw
++] = V_0287F0_DI_SRC_SEL_AUTO_INDEX
|
1328 (info
.count_from_stream_output
? S_0287F0_USE_OPAQUE(1) : 0);
1331 /* Set the depth buffer as dirty. */
1332 if (rctx
->framebuffer
.state
.zsbuf
) {
1333 struct pipe_surface
*surf
= rctx
->framebuffer
.state
.zsbuf
;
1334 struct r600_texture
*rtex
= (struct r600_texture
*)surf
->texture
;
1336 rtex
->dirty_level_mask
|= 1 << surf
->u
.tex
.level
;
1338 if (rctx
->framebuffer
.compressed_cb_mask
) {
1339 struct pipe_surface
*surf
;
1340 struct r600_texture
*rtex
;
1341 unsigned mask
= rctx
->framebuffer
.compressed_cb_mask
;
1344 unsigned i
= u_bit_scan(&mask
);
1345 surf
= rctx
->framebuffer
.state
.cbufs
[i
];
1346 rtex
= (struct r600_texture
*)surf
->texture
;
1348 rtex
->dirty_level_mask
|= 1 << surf
->u
.tex
.level
;
1353 pipe_resource_reference(&ib
.buffer
, NULL
);
1356 void r600_draw_rectangle(struct blitter_context
*blitter
,
1357 int x1
, int y1
, int x2
, int y2
, float depth
,
1358 enum blitter_attrib_type type
, const union pipe_color_union
*attrib
)
1360 struct r600_context
*rctx
= (struct r600_context
*)util_blitter_get_pipe(blitter
);
1361 struct pipe_viewport_state viewport
;
1362 struct pipe_resource
*buf
= NULL
;
1363 unsigned offset
= 0;
1366 if (type
== UTIL_BLITTER_ATTRIB_TEXCOORD
) {
1367 util_blitter_draw_rectangle(blitter
, x1
, y1
, x2
, y2
, depth
, type
, attrib
);
1371 /* Some operations (like color resolve on r6xx) don't work
1372 * with the conventional primitive types.
1373 * One that works is PT_RECTLIST, which we use here. */
1375 /* setup viewport */
1376 viewport
.scale
[0] = 1.0f
;
1377 viewport
.scale
[1] = 1.0f
;
1378 viewport
.scale
[2] = 1.0f
;
1379 viewport
.scale
[3] = 1.0f
;
1380 viewport
.translate
[0] = 0.0f
;
1381 viewport
.translate
[1] = 0.0f
;
1382 viewport
.translate
[2] = 0.0f
;
1383 viewport
.translate
[3] = 0.0f
;
1384 rctx
->context
.set_viewport_state(&rctx
->context
, &viewport
);
1386 /* Upload vertices. The hw rectangle has only 3 vertices,
1387 * I guess the 4th one is derived from the first 3.
1388 * The vertex specification should match u_blitter's vertex element state. */
1389 u_upload_alloc(rctx
->uploader
, 0, sizeof(float) * 24, &offset
, &buf
, (void**)&vb
);
1406 memcpy(vb
+4, attrib
->f
, sizeof(float)*4);
1407 memcpy(vb
+12, attrib
->f
, sizeof(float)*4);
1408 memcpy(vb
+20, attrib
->f
, sizeof(float)*4);
1412 util_draw_vertex_buffer(&rctx
->context
, NULL
, buf
, rctx
->blitter
->vb_slot
, offset
,
1413 R600_PRIM_RECTANGLE_LIST
, 3, 2);
1414 pipe_resource_reference(&buf
, NULL
);
1417 void _r600_pipe_state_add_reg_bo(struct r600_context
*ctx
,
1418 struct r600_pipe_state
*state
,
1419 uint32_t offset
, uint32_t value
,
1420 uint32_t range_id
, uint32_t block_id
,
1421 struct r600_resource
*bo
,
1422 enum radeon_bo_usage usage
)
1425 struct r600_range
*range
;
1426 struct r600_block
*block
;
1428 if (bo
) assert(usage
);
1430 range
= &ctx
->range
[range_id
];
1431 block
= range
->blocks
[block_id
];
1432 state
->regs
[state
->nregs
].block
= block
;
1433 state
->regs
[state
->nregs
].id
= (offset
- block
->start_offset
) >> 2;
1435 state
->regs
[state
->nregs
].value
= value
;
1436 state
->regs
[state
->nregs
].bo
= bo
;
1437 state
->regs
[state
->nregs
].bo_usage
= usage
;
1440 assert(state
->nregs
< R600_BLOCK_MAX_REG
);
1443 void _r600_pipe_state_add_reg(struct r600_context
*ctx
,
1444 struct r600_pipe_state
*state
,
1445 uint32_t offset
, uint32_t value
,
1446 uint32_t range_id
, uint32_t block_id
)
1448 _r600_pipe_state_add_reg_bo(ctx
, state
, offset
, value
,
1449 range_id
, block_id
, NULL
, 0);
1452 uint32_t r600_translate_stencil_op(int s_op
)
1455 case PIPE_STENCIL_OP_KEEP
:
1456 return V_028800_STENCIL_KEEP
;
1457 case PIPE_STENCIL_OP_ZERO
:
1458 return V_028800_STENCIL_ZERO
;
1459 case PIPE_STENCIL_OP_REPLACE
:
1460 return V_028800_STENCIL_REPLACE
;
1461 case PIPE_STENCIL_OP_INCR
:
1462 return V_028800_STENCIL_INCR
;
1463 case PIPE_STENCIL_OP_DECR
:
1464 return V_028800_STENCIL_DECR
;
1465 case PIPE_STENCIL_OP_INCR_WRAP
:
1466 return V_028800_STENCIL_INCR_WRAP
;
1467 case PIPE_STENCIL_OP_DECR_WRAP
:
1468 return V_028800_STENCIL_DECR_WRAP
;
1469 case PIPE_STENCIL_OP_INVERT
:
1470 return V_028800_STENCIL_INVERT
;
1472 R600_ERR("Unknown stencil op %d", s_op
);
1479 uint32_t r600_translate_fill(uint32_t func
)
1482 case PIPE_POLYGON_MODE_FILL
:
1484 case PIPE_POLYGON_MODE_LINE
:
1486 case PIPE_POLYGON_MODE_POINT
:
1494 unsigned r600_tex_wrap(unsigned wrap
)
1498 case PIPE_TEX_WRAP_REPEAT
:
1499 return V_03C000_SQ_TEX_WRAP
;
1500 case PIPE_TEX_WRAP_CLAMP
:
1501 return V_03C000_SQ_TEX_CLAMP_HALF_BORDER
;
1502 case PIPE_TEX_WRAP_CLAMP_TO_EDGE
:
1503 return V_03C000_SQ_TEX_CLAMP_LAST_TEXEL
;
1504 case PIPE_TEX_WRAP_CLAMP_TO_BORDER
:
1505 return V_03C000_SQ_TEX_CLAMP_BORDER
;
1506 case PIPE_TEX_WRAP_MIRROR_REPEAT
:
1507 return V_03C000_SQ_TEX_MIRROR
;
1508 case PIPE_TEX_WRAP_MIRROR_CLAMP
:
1509 return V_03C000_SQ_TEX_MIRROR_ONCE_HALF_BORDER
;
1510 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
:
1511 return V_03C000_SQ_TEX_MIRROR_ONCE_LAST_TEXEL
;
1512 case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
:
1513 return V_03C000_SQ_TEX_MIRROR_ONCE_BORDER
;
1517 unsigned r600_tex_filter(unsigned filter
)
1521 case PIPE_TEX_FILTER_NEAREST
:
1522 return V_03C000_SQ_TEX_XY_FILTER_POINT
;
1523 case PIPE_TEX_FILTER_LINEAR
:
1524 return V_03C000_SQ_TEX_XY_FILTER_BILINEAR
;
1528 unsigned r600_tex_mipfilter(unsigned filter
)
1531 case PIPE_TEX_MIPFILTER_NEAREST
:
1532 return V_03C000_SQ_TEX_Z_FILTER_POINT
;
1533 case PIPE_TEX_MIPFILTER_LINEAR
:
1534 return V_03C000_SQ_TEX_Z_FILTER_LINEAR
;
1536 case PIPE_TEX_MIPFILTER_NONE
:
1537 return V_03C000_SQ_TEX_Z_FILTER_NONE
;
1541 unsigned r600_tex_compare(unsigned compare
)
1545 case PIPE_FUNC_NEVER
:
1546 return V_03C000_SQ_TEX_DEPTH_COMPARE_NEVER
;
1547 case PIPE_FUNC_LESS
:
1548 return V_03C000_SQ_TEX_DEPTH_COMPARE_LESS
;
1549 case PIPE_FUNC_EQUAL
:
1550 return V_03C000_SQ_TEX_DEPTH_COMPARE_EQUAL
;
1551 case PIPE_FUNC_LEQUAL
:
1552 return V_03C000_SQ_TEX_DEPTH_COMPARE_LESSEQUAL
;
1553 case PIPE_FUNC_GREATER
:
1554 return V_03C000_SQ_TEX_DEPTH_COMPARE_GREATER
;
1555 case PIPE_FUNC_NOTEQUAL
:
1556 return V_03C000_SQ_TEX_DEPTH_COMPARE_NOTEQUAL
;
1557 case PIPE_FUNC_GEQUAL
:
1558 return V_03C000_SQ_TEX_DEPTH_COMPARE_GREATEREQUAL
;
1559 case PIPE_FUNC_ALWAYS
:
1560 return V_03C000_SQ_TEX_DEPTH_COMPARE_ALWAYS
;
1564 static bool wrap_mode_uses_border_color(unsigned wrap
, bool linear_filter
)
1566 return wrap
== PIPE_TEX_WRAP_CLAMP_TO_BORDER
||
1567 wrap
== PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER
||
1569 (wrap
== PIPE_TEX_WRAP_CLAMP
||
1570 wrap
== PIPE_TEX_WRAP_MIRROR_CLAMP
));
1573 bool sampler_state_needs_border_color(const struct pipe_sampler_state
*state
)
1575 bool linear_filter
= state
->min_img_filter
!= PIPE_TEX_FILTER_NEAREST
||
1576 state
->mag_img_filter
!= PIPE_TEX_FILTER_NEAREST
;
1578 return (state
->border_color
.ui
[0] || state
->border_color
.ui
[1] ||
1579 state
->border_color
.ui
[2] || state
->border_color
.ui
[3]) &&
1580 (wrap_mode_uses_border_color(state
->wrap_s
, linear_filter
) ||
1581 wrap_mode_uses_border_color(state
->wrap_t
, linear_filter
) ||
1582 wrap_mode_uses_border_color(state
->wrap_r
, linear_filter
));
1585 /* keep this at the end of this file, please */
1586 void r600_init_common_state_functions(struct r600_context
*rctx
)
1588 rctx
->context
.create_fs_state
= r600_create_ps_state
;
1589 rctx
->context
.create_vs_state
= r600_create_vs_state
;
1590 rctx
->context
.create_vertex_elements_state
= r600_create_vertex_fetch_shader
;
1591 rctx
->context
.bind_blend_state
= r600_bind_blend_state
;
1592 rctx
->context
.bind_depth_stencil_alpha_state
= r600_bind_dsa_state
;
1593 rctx
->context
.bind_fragment_sampler_states
= r600_bind_ps_sampler_states
;
1594 rctx
->context
.bind_fs_state
= r600_bind_ps_state
;
1595 rctx
->context
.bind_rasterizer_state
= r600_bind_rs_state
;
1596 rctx
->context
.bind_vertex_elements_state
= r600_bind_vertex_elements
;
1597 rctx
->context
.bind_vertex_sampler_states
= r600_bind_vs_sampler_states
;
1598 rctx
->context
.bind_vs_state
= r600_bind_vs_state
;
1599 rctx
->context
.delete_blend_state
= r600_delete_blend_state
;
1600 rctx
->context
.delete_depth_stencil_alpha_state
= r600_delete_dsa_state
;
1601 rctx
->context
.delete_fs_state
= r600_delete_ps_state
;
1602 rctx
->context
.delete_rasterizer_state
= r600_delete_rs_state
;
1603 rctx
->context
.delete_sampler_state
= r600_delete_sampler_state
;
1604 rctx
->context
.delete_vertex_elements_state
= r600_delete_vertex_elements
;
1605 rctx
->context
.delete_vs_state
= r600_delete_vs_state
;
1606 rctx
->context
.set_blend_color
= r600_set_blend_color
;
1607 rctx
->context
.set_clip_state
= r600_set_clip_state
;
1608 rctx
->context
.set_constant_buffer
= r600_set_constant_buffer
;
1609 rctx
->context
.set_sample_mask
= r600_set_sample_mask
;
1610 rctx
->context
.set_stencil_ref
= r600_set_pipe_stencil_ref
;
1611 rctx
->context
.set_viewport_state
= r600_set_viewport_state
;
1612 rctx
->context
.set_vertex_buffers
= r600_set_vertex_buffers
;
1613 rctx
->context
.set_index_buffer
= r600_set_index_buffer
;
1614 rctx
->context
.set_fragment_sampler_views
= r600_set_ps_sampler_views
;
1615 rctx
->context
.set_vertex_sampler_views
= r600_set_vs_sampler_views
;
1616 rctx
->context
.sampler_view_destroy
= r600_sampler_view_destroy
;
1617 rctx
->context
.texture_barrier
= r600_texture_barrier
;
1618 rctx
->context
.create_stream_output_target
= r600_create_so_target
;
1619 rctx
->context
.stream_output_target_destroy
= r600_so_target_destroy
;
1620 rctx
->context
.set_stream_output_targets
= r600_set_so_targets
;
1621 rctx
->context
.draw_vbo
= r600_draw_vbo
;