2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
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.
29 #include "util/u_slab.h"
31 #include "r600_llvm.h"
32 #include "r600_public.h"
33 #include "r600_shader.h"
34 #include "r600_resource.h"
35 #include "evergreen_compute.h"
37 #define R600_MAX_CONST_BUFFERS 2
38 #define R600_MAX_CONST_BUFFER_SIZE 4096
40 #ifdef PIPE_ARCH_BIG_ENDIAN
41 #define R600_BIG_ENDIAN 1
43 #define R600_BIG_ENDIAN 0
46 enum r600_atom_flags
{
47 /* When set, atoms are added at the beginning of the dirty list
48 * instead of the end. */
52 /* This encapsulates a state or an operation which can emitted into the GPU
53 * command stream. It's not limited to states only, it can be used for anything
54 * that wants to write commands into the CS (e.g. cache flushes). */
56 void (*emit
)(struct r600_context
*ctx
, struct r600_atom
*state
);
59 enum r600_atom_flags flags
;
62 struct list_head head
;
65 /* This is an atom containing GPU commands that never change.
66 * This is supposed to be copied directly into the CS. */
67 struct r600_command_buffer
{
68 struct r600_atom atom
;
74 struct r600_surface_sync_cmd
{
75 struct r600_atom atom
;
76 unsigned flush_flags
; /* CP_COHER_CNTL */
79 struct r600_db_misc_state
{
80 struct r600_atom atom
;
81 bool occlusion_query_enabled
;
82 bool flush_depthstencil_through_cb
;
83 bool copy_depth
, copy_stencil
;
86 struct r600_cb_misc_state
{
87 struct r600_atom atom
;
88 unsigned cb_color_control
; /* this comes from blend state */
89 unsigned blend_colormask
; /* 8*4 bits for 8 RGBA colorbuffers */
91 unsigned nr_ps_color_outputs
;
96 struct r600_alphatest_state
{
97 struct r600_atom atom
;
98 unsigned sx_alpha_test_control
; /* this comes from dsa state */
99 unsigned sx_alpha_ref
; /* this comes from dsa state */
101 bool cb0_export_16bpc
; /* from set_framebuffer_state */
104 struct r600_cs_shader_state
{
105 struct r600_atom atom
;
106 struct r600_pipe_compute
*shader
;
109 struct r600_sample_mask
{
110 struct r600_atom atom
;
111 uint16_t sample_mask
; /* there are only 8 bits on EG, 16 bits on Cayman */
114 enum r600_pipe_state_id
{
115 R600_PIPE_STATE_BLEND
= 0,
116 R600_PIPE_STATE_BLEND_COLOR
,
117 R600_PIPE_STATE_CONFIG
,
118 R600_PIPE_STATE_SEAMLESS_CUBEMAP
,
119 R600_PIPE_STATE_CLIP
,
120 R600_PIPE_STATE_SCISSOR
,
121 R600_PIPE_STATE_VIEWPORT
,
122 R600_PIPE_STATE_RASTERIZER
,
124 R600_PIPE_STATE_FRAMEBUFFER
,
126 R600_PIPE_STATE_STENCIL_REF
,
127 R600_PIPE_STATE_PS_SHADER
,
128 R600_PIPE_STATE_VS_SHADER
,
129 R600_PIPE_STATE_CONSTANT
,
130 R600_PIPE_STATE_SAMPLER
,
131 R600_PIPE_STATE_RESOURCE
,
132 R600_PIPE_STATE_POLYGON_OFFSET
,
133 R600_PIPE_STATE_FETCH_SHADER
,
138 struct compute_memory_pool
;
139 void compute_memory_pool_delete(struct compute_memory_pool
* pool
);
140 struct compute_memory_pool
* compute_memory_pool_new(
141 struct r600_screen
*rscreen
);
143 struct r600_pipe_fences
{
144 struct r600_resource
*bo
;
147 /* linked list of preallocated blocks */
148 struct list_head blocks
;
149 /* linked list of freed fences */
150 struct list_head pool
;
155 struct pipe_screen screen
;
156 struct radeon_winsys
*ws
;
158 enum chip_class chip_class
;
159 struct radeon_info info
;
161 struct r600_tiling_info tiling_info
;
162 struct r600_pipe_fences fences
;
164 /*for compute global memory binding, we allocate stuff here, instead of
166 * XXX: Not sure if this is the best place for global_pool. Also,
167 * it's not thread safe, so it won't work with multiple contexts. */
168 struct compute_memory_pool
*global_pool
;
171 struct r600_pipe_sampler_view
{
172 struct pipe_sampler_view base
;
173 struct r600_resource
*tex_resource
;
174 uint32_t tex_resource_words
[8];
177 struct r600_pipe_rasterizer
{
178 struct r600_pipe_state rstate
;
181 unsigned sprite_coord_enable
;
182 unsigned clip_plane_enable
;
183 unsigned pa_sc_line_stipple
;
184 unsigned pa_cl_clip_cntl
;
188 bool multisample_enable
;
191 struct r600_pipe_blend
{
192 struct r600_pipe_state rstate
;
193 unsigned cb_target_mask
;
194 unsigned cb_color_control
;
199 struct r600_pipe_dsa
{
200 struct r600_pipe_state rstate
;
204 unsigned sx_alpha_test_control
;
207 struct r600_vertex_element
210 struct pipe_vertex_element elements
[PIPE_MAX_ATTRIBS
];
211 struct r600_resource
*fetch_shader
;
213 struct r600_pipe_state rstate
;
216 struct r600_pipe_shader
;
218 struct r600_pipe_shader_selector
{
219 struct r600_pipe_shader
*current
;
221 struct tgsi_token
*tokens
;
222 struct pipe_stream_output_info so
;
224 unsigned num_shaders
;
226 /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
229 unsigned nr_ps_max_color_exports
;
232 struct r600_pipe_shader
{
233 struct r600_pipe_shader_selector
*selector
;
234 struct r600_pipe_shader
*next_variant
;
235 struct r600_shader shader
;
236 struct r600_pipe_state rstate
;
237 struct r600_resource
*bo
;
238 struct r600_resource
*bo_fetch
;
239 struct r600_vertex_element vertex_elements
;
240 unsigned sprite_coord_enable
;
242 unsigned pa_cl_vs_out_cntl
;
243 unsigned nr_ps_color_outputs
;
245 unsigned db_shader_control
;
246 unsigned ps_depth_export
;
249 struct r600_pipe_sampler_state
{
250 uint32_t tex_sampler_words
[3];
251 uint32_t border_color
[4];
252 bool border_color_use
;
253 bool seamless_cube_map
;
256 /* needed for blitter save */
257 #define NUM_TEX_UNITS 16
259 struct r600_seamless_cube_map
{
260 struct r600_atom atom
;
264 struct r600_samplerview_state
{
265 struct r600_atom atom
;
266 struct r600_pipe_sampler_view
*views
[NUM_TEX_UNITS
];
267 uint32_t enabled_mask
;
269 uint32_t depth_texture_mask
; /* which textures are depth */
272 struct r600_textures_info
{
273 struct r600_samplerview_state views
;
274 struct r600_atom atom_sampler
;
275 struct r600_pipe_sampler_state
*samplers
[NUM_TEX_UNITS
];
277 bool is_array_sampler
[NUM_TEX_UNITS
];
281 struct pipe_reference reference
;
282 unsigned index
; /* in the shared bo */
283 struct r600_resource
*sleep_bo
;
284 struct list_head head
;
287 #define FENCE_BLOCK_SIZE 16
289 struct r600_fence_block
{
290 struct r600_fence fences
[FENCE_BLOCK_SIZE
];
291 struct list_head head
;
294 #define R600_CONSTANT_ARRAY_SIZE 256
295 #define R600_RESOURCE_ARRAY_SIZE 160
297 struct r600_stencil_ref
304 struct r600_constbuf_state
306 struct r600_atom atom
;
307 struct pipe_constant_buffer cb
[PIPE_MAX_CONSTANT_BUFFERS
];
308 uint32_t enabled_mask
;
312 struct r600_vertexbuf_state
314 struct r600_atom atom
;
315 struct pipe_vertex_buffer vb
[PIPE_MAX_ATTRIBS
];
316 uint32_t enabled_mask
; /* non-NULL buffers */
320 struct r600_context
{
321 struct pipe_context context
;
322 struct blitter_context
*blitter
;
323 enum radeon_family family
;
324 enum chip_class chip_class
;
325 boolean has_vertex_cache
;
326 unsigned r6xx_num_clause_temp_gprs
;
327 void *custom_dsa_flush
;
328 struct r600_screen
*screen
;
329 struct radeon_winsys
*ws
;
330 struct r600_pipe_state
*states
[R600_PIPE_NSTATES
];
331 struct r600_vertex_element
*vertex_elements
;
332 struct pipe_framebuffer_state framebuffer
;
333 unsigned compute_cb_target_mask
;
334 unsigned db_shader_control
;
335 unsigned pa_sc_line_stipple
;
336 unsigned pa_cl_clip_cntl
;
337 /* for saving when using blitter */
338 struct pipe_stencil_ref stencil_ref
;
339 struct pipe_viewport_state viewport
;
340 struct pipe_clip_state clip
;
341 struct r600_pipe_shader_selector
*ps_shader
;
342 struct r600_pipe_shader_selector
*vs_shader
;
343 struct r600_pipe_rasterizer
*rasterizer
;
344 struct r600_pipe_state vgt
;
345 struct r600_pipe_state spi
;
346 struct pipe_query
*current_render_cond
;
347 unsigned current_render_cond_mode
;
348 struct pipe_query
*saved_render_cond
;
349 unsigned saved_render_cond_mode
;
350 /* shader information */
353 unsigned sprite_coord_enable
;
355 boolean export_16bpc
;
358 bool multisample_enable
;
361 struct u_upload_mgr
*uploader
;
362 struct util_slab_mempool pool_transfers
;
364 unsigned default_ps_gprs
, default_vs_gprs
;
366 /* States based on r600_atom. */
367 struct list_head dirty_states
;
368 struct r600_command_buffer start_cs_cmd
; /* invariant state mostly */
369 /** Compute specific registers initializations. The start_cs_cmd atom
370 * must be emitted before start_compute_cs_cmd. */
371 struct r600_command_buffer start_compute_cs_cmd
;
372 struct r600_surface_sync_cmd surface_sync_cmd
;
373 struct r600_atom r6xx_flush_and_inv_cmd
;
374 struct r600_alphatest_state alphatest_state
;
375 struct r600_cb_misc_state cb_misc_state
;
376 struct r600_db_misc_state db_misc_state
;
377 /** Vertex buffers for fetch shaders */
378 struct r600_vertexbuf_state vertex_buffer_state
;
379 /** Vertex buffers for compute shaders */
380 struct r600_vertexbuf_state cs_vertex_buffer_state
;
381 struct r600_constbuf_state vs_constbuf_state
;
382 struct r600_constbuf_state ps_constbuf_state
;
383 struct r600_textures_info vs_samplers
;
384 struct r600_textures_info ps_samplers
;
385 struct r600_seamless_cube_map seamless_cube_map
;
386 struct r600_cs_shader_state cs_shader_state
;
387 struct r600_sample_mask sample_mask
;
389 struct radeon_winsys_cs
*cs
;
391 struct r600_range
*range
;
393 struct r600_block
**blocks
;
394 struct list_head dirty
;
395 struct list_head enable_list
;
396 unsigned pm4_dirty_cdwords
;
397 unsigned ctx_pm4_ndwords
;
399 /* The list of active queries. Only one query of each type can be active. */
400 int num_occlusion_queries
;
402 /* Manage queries in two separate groups:
403 * The timer ones and the others (streamout, occlusion).
405 * We do this because we should only suspend non-timer queries for u_blitter,
406 * and later if the non-timer queries are suspended, the context flush should
407 * only suspend and resume the timer queries. */
408 struct list_head active_timer_queries
;
409 unsigned num_cs_dw_timer_queries_suspend
;
410 struct list_head active_nontimer_queries
;
411 unsigned num_cs_dw_nontimer_queries_suspend
;
413 unsigned num_cs_dw_streamout_end
;
415 unsigned backend_mask
;
416 unsigned max_db
; /* for OQ */
418 boolean predicate_drawing
;
420 unsigned num_so_targets
;
421 struct r600_so_target
*so_targets
[PIPE_MAX_SO_BUFFERS
];
422 boolean streamout_start
;
423 unsigned streamout_append_bitmask
;
425 /* There is no scissor enable bit on r6xx, so we must use a workaround.
426 * These track the current scissor state. */
428 struct pipe_scissor_state scissor_state
;
430 /* With rasterizer discard, there doesn't have to be a pixel shader.
431 * In that case, we bind this one: */
432 void *dummy_pixel_shader
;
434 boolean dual_src_blend
;
437 struct pipe_index_buffer index_buffer
;
440 static INLINE
void r600_emit_atom(struct r600_context
*rctx
, struct r600_atom
*atom
)
442 atom
->emit(rctx
, atom
);
444 if (atom
->head
.next
&& atom
->head
.prev
)
445 LIST_DELINIT(&atom
->head
);
448 static INLINE
void r600_atom_dirty(struct r600_context
*rctx
, struct r600_atom
*state
)
451 if (state
->flags
& EMIT_EARLY
) {
452 LIST_ADD(&state
->head
, &rctx
->dirty_states
);
454 LIST_ADDTAIL(&state
->head
, &rctx
->dirty_states
);
460 /* evergreen_state.c */
461 void evergreen_init_state_functions(struct r600_context
*rctx
);
462 void evergreen_init_atom_start_cs(struct r600_context
*rctx
);
463 void evergreen_pipe_shader_ps(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
464 void evergreen_pipe_shader_vs(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
465 void evergreen_fetch_shader(struct pipe_context
*ctx
, struct r600_vertex_element
*ve
);
466 void *evergreen_create_db_flush_dsa(struct r600_context
*rctx
);
467 void evergreen_polygon_offset_update(struct r600_context
*rctx
);
468 boolean
evergreen_is_format_supported(struct pipe_screen
*screen
,
469 enum pipe_format format
,
470 enum pipe_texture_target target
,
471 unsigned sample_count
,
473 void evergreen_init_color_surface(struct r600_context
*rctx
,
474 struct r600_surface
*surf
);
475 void evergreen_update_dual_export_state(struct r600_context
* rctx
);
478 void r600_copy_buffer(struct pipe_context
*ctx
, struct
479 pipe_resource
*dst
, unsigned dstx
,
480 struct pipe_resource
*src
, const struct pipe_box
*src_box
);
481 void r600_init_blit_functions(struct r600_context
*rctx
);
482 void r600_blit_uncompress_depth(struct pipe_context
*ctx
,
483 struct r600_resource_texture
*texture
,
484 struct r600_resource_texture
*staging
,
485 unsigned first_level
, unsigned last_level
,
486 unsigned first_layer
, unsigned last_layer
);
487 void r600_flush_depth_textures(struct r600_context
*rctx
,
488 struct r600_samplerview_state
*textures
);
490 bool r600_init_resource(struct r600_screen
*rscreen
,
491 struct r600_resource
*res
,
492 unsigned size
, unsigned alignment
,
493 unsigned bind
, unsigned usage
);
494 struct pipe_resource
*r600_buffer_create(struct pipe_screen
*screen
,
495 const struct pipe_resource
*templ
);
498 void r600_flush(struct pipe_context
*ctx
, struct pipe_fence_handle
**fence
,
502 void r600_init_query_functions(struct r600_context
*rctx
);
503 void r600_suspend_nontimer_queries(struct r600_context
*ctx
);
504 void r600_resume_nontimer_queries(struct r600_context
*ctx
);
505 void r600_suspend_timer_queries(struct r600_context
*ctx
);
506 void r600_resume_timer_queries(struct r600_context
*ctx
);
508 /* r600_resource.c */
509 void r600_init_context_resource_functions(struct r600_context
*r600
);
512 int r600_pipe_shader_create(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
514 int r600_compute_shader_create(struct pipe_context
* ctx
,
515 LLVMModuleRef mod
, struct r600_bytecode
* bytecode
);
517 void r600_pipe_shader_destroy(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
520 void r600_set_scissor_state(struct r600_context
*rctx
,
521 const struct pipe_scissor_state
*state
);
522 void r600_init_state_functions(struct r600_context
*rctx
);
523 void r600_init_atom_start_cs(struct r600_context
*rctx
);
524 void r600_pipe_shader_ps(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
525 void r600_pipe_shader_vs(struct pipe_context
*ctx
, struct r600_pipe_shader
*shader
);
526 void r600_fetch_shader(struct pipe_context
*ctx
, struct r600_vertex_element
*ve
);
527 void *r600_create_db_flush_dsa(struct r600_context
*rctx
);
528 void r600_polygon_offset_update(struct r600_context
*rctx
);
529 void r600_adjust_gprs(struct r600_context
*rctx
);
530 boolean
r600_is_format_supported(struct pipe_screen
*screen
,
531 enum pipe_format format
,
532 enum pipe_texture_target target
,
533 unsigned sample_count
,
535 void r600_update_dual_export_state(struct r600_context
* rctx
);
538 void r600_init_screen_texture_functions(struct pipe_screen
*screen
);
539 void r600_init_surface_functions(struct r600_context
*r600
);
540 uint32_t r600_translate_texformat(struct pipe_screen
*screen
, enum pipe_format format
,
541 const unsigned char *swizzle_view
,
542 uint32_t *word4_p
, uint32_t *yuv_format_p
);
543 unsigned r600_texture_get_offset(struct r600_resource_texture
*rtex
,
544 unsigned level
, unsigned layer
);
546 /* r600_translate.c */
547 void r600_translate_index_buffer(struct r600_context
*r600
,
548 struct pipe_index_buffer
*ib
,
551 /* r600_state_common.c */
552 void r600_init_atom(struct r600_atom
*atom
,
553 void (*emit
)(struct r600_context
*ctx
, struct r600_atom
*state
),
554 unsigned num_dw
, enum r600_atom_flags flags
);
555 void r600_init_common_atoms(struct r600_context
*rctx
);
556 unsigned r600_get_cb_flush_flags(struct r600_context
*rctx
);
557 void r600_texture_barrier(struct pipe_context
*ctx
);
558 void r600_set_index_buffer(struct pipe_context
*ctx
,
559 const struct pipe_index_buffer
*ib
);
560 void r600_vertex_buffers_dirty(struct r600_context
*rctx
);
561 void r600_set_vertex_buffers(struct pipe_context
*ctx
, unsigned count
,
562 const struct pipe_vertex_buffer
*input
);
563 void r600_sampler_views_dirty(struct r600_context
*rctx
,
564 struct r600_samplerview_state
*state
);
565 void r600_set_sampler_views(struct r600_context
*rctx
,
566 struct r600_textures_info
*dst
,
568 struct pipe_sampler_view
**views
);
569 void r600_bind_vs_samplers(struct pipe_context
*ctx
, unsigned count
, void **states
);
570 void r600_bind_ps_samplers(struct pipe_context
*ctx
, unsigned count
, void **states
);
571 void *r600_create_vertex_elements(struct pipe_context
*ctx
,
573 const struct pipe_vertex_element
*elements
);
574 void r600_delete_vertex_element(struct pipe_context
*ctx
, void *state
);
575 void r600_bind_blend_state(struct pipe_context
*ctx
, void *state
);
576 void r600_set_blend_color(struct pipe_context
*ctx
,
577 const struct pipe_blend_color
*state
);
578 void r600_bind_dsa_state(struct pipe_context
*ctx
, void *state
);
579 void r600_set_max_scissor(struct r600_context
*rctx
);
580 void r600_bind_rs_state(struct pipe_context
*ctx
, void *state
);
581 void r600_delete_rs_state(struct pipe_context
*ctx
, void *state
);
582 void r600_sampler_view_destroy(struct pipe_context
*ctx
,
583 struct pipe_sampler_view
*state
);
584 void r600_delete_sampler(struct pipe_context
*ctx
, void *state
);
585 void r600_delete_state(struct pipe_context
*ctx
, void *state
);
586 void r600_bind_vertex_elements(struct pipe_context
*ctx
, void *state
);
587 void *r600_create_shader_state_ps(struct pipe_context
*ctx
,
588 const struct pipe_shader_state
*state
);
589 void *r600_create_shader_state_vs(struct pipe_context
*ctx
,
590 const struct pipe_shader_state
*state
);
591 void r600_bind_ps_shader(struct pipe_context
*ctx
, void *state
);
592 void r600_bind_vs_shader(struct pipe_context
*ctx
, void *state
);
593 void r600_delete_ps_shader(struct pipe_context
*ctx
, void *state
);
594 void r600_delete_vs_shader(struct pipe_context
*ctx
, void *state
);
595 void r600_constant_buffers_dirty(struct r600_context
*rctx
, struct r600_constbuf_state
*state
);
596 void r600_set_constant_buffer(struct pipe_context
*ctx
, uint shader
, uint index
,
597 struct pipe_constant_buffer
*cb
);
598 struct pipe_stream_output_target
*
599 r600_create_so_target(struct pipe_context
*ctx
,
600 struct pipe_resource
*buffer
,
601 unsigned buffer_offset
,
602 unsigned buffer_size
);
603 void r600_so_target_destroy(struct pipe_context
*ctx
,
604 struct pipe_stream_output_target
*target
);
605 void r600_set_so_targets(struct pipe_context
*ctx
,
606 unsigned num_targets
,
607 struct pipe_stream_output_target
**targets
,
608 unsigned append_bitmask
);
609 void r600_set_sample_mask(struct pipe_context
*pipe
, unsigned sample_mask
);
610 void r600_set_pipe_stencil_ref(struct pipe_context
*ctx
,
611 const struct pipe_stencil_ref
*state
);
612 void r600_draw_vbo(struct pipe_context
*ctx
, const struct pipe_draw_info
*info
);
613 uint32_t r600_translate_stencil_op(int s_op
);
614 uint32_t r600_translate_fill(uint32_t func
);
615 unsigned r600_tex_wrap(unsigned wrap
);
616 unsigned r600_tex_filter(unsigned filter
);
617 unsigned r600_tex_mipfilter(unsigned filter
);
618 unsigned r600_tex_compare(unsigned compare
);
621 * Helpers for building command buffers
624 #define PKT3_SET_CONFIG_REG 0x68
625 #define PKT3_SET_CONTEXT_REG 0x69
626 #define PKT3_SET_CTL_CONST 0x6F
627 #define PKT3_SET_LOOP_CONST 0x6C
629 #define R600_CONFIG_REG_OFFSET 0x08000
630 #define R600_CONTEXT_REG_OFFSET 0x28000
631 #define R600_CTL_CONST_OFFSET 0x3CFF0
632 #define R600_LOOP_CONST_OFFSET 0X0003E200
633 #define EG_LOOP_CONST_OFFSET 0x0003A200
635 #define PKT_TYPE_S(x) (((x) & 0x3) << 30)
636 #define PKT_COUNT_S(x) (((x) & 0x3FFF) << 16)
637 #define PKT3_IT_OPCODE_S(x) (((x) & 0xFF) << 8)
638 #define PKT3_PREDICATE(x) (((x) >> 0) & 0x1)
639 #define PKT3(op, count, predicate) (PKT_TYPE_S(3) | PKT_COUNT_S(count) | PKT3_IT_OPCODE_S(op) | PKT3_PREDICATE(predicate))
641 #define RADEON_CP_PACKET3_COMPUTE_MODE 0x00000002
643 /*Evergreen Compute packet3*/
644 #define PKT3C(op, count, predicate) (PKT_TYPE_S(3) | PKT3_IT_OPCODE_S(op) | PKT_COUNT_S(count) | PKT3_PREDICATE(predicate) | RADEON_CP_PACKET3_COMPUTE_MODE)
646 static INLINE
void r600_store_value(struct r600_command_buffer
*cb
, unsigned value
)
648 cb
->buf
[cb
->atom
.num_dw
++] = value
;
651 static INLINE
void r600_store_config_reg_seq(struct r600_command_buffer
*cb
, unsigned reg
, unsigned num
)
653 assert(reg
< R600_CONTEXT_REG_OFFSET
);
654 assert(cb
->atom
.num_dw
+2+num
<= cb
->max_num_dw
);
655 cb
->buf
[cb
->atom
.num_dw
++] = PKT3(PKT3_SET_CONFIG_REG
, num
, 0);
656 cb
->buf
[cb
->atom
.num_dw
++] = (reg
- R600_CONFIG_REG_OFFSET
) >> 2;
660 * Needs cb->pkt_flags set to RADEON_CP_PACKET3_COMPUTE_MODE for compute
663 static INLINE
void r600_store_context_reg_seq(struct r600_command_buffer
*cb
, unsigned reg
, unsigned num
)
665 assert(reg
>= R600_CONTEXT_REG_OFFSET
&& reg
< R600_CTL_CONST_OFFSET
);
666 assert(cb
->atom
.num_dw
+2+num
<= cb
->max_num_dw
);
667 cb
->buf
[cb
->atom
.num_dw
++] = PKT3(PKT3_SET_CONTEXT_REG
, num
, 0) | cb
->pkt_flags
;
668 cb
->buf
[cb
->atom
.num_dw
++] = (reg
- R600_CONTEXT_REG_OFFSET
) >> 2;
672 * Needs cb->pkt_flags set to RADEON_CP_PACKET3_COMPUTE_MODE for compute
675 static INLINE
void r600_store_ctl_const_seq(struct r600_command_buffer
*cb
, unsigned reg
, unsigned num
)
677 assert(reg
>= R600_CTL_CONST_OFFSET
);
678 assert(cb
->atom
.num_dw
+2+num
<= cb
->max_num_dw
);
679 cb
->buf
[cb
->atom
.num_dw
++] = PKT3(PKT3_SET_CTL_CONST
, num
, 0) | cb
->pkt_flags
;
680 cb
->buf
[cb
->atom
.num_dw
++] = (reg
- R600_CTL_CONST_OFFSET
) >> 2;
683 static INLINE
void r600_store_loop_const_seq(struct r600_command_buffer
*cb
, unsigned reg
, unsigned num
)
685 assert(reg
>= R600_LOOP_CONST_OFFSET
);
686 assert(cb
->atom
.num_dw
+2+num
<= cb
->max_num_dw
);
687 cb
->buf
[cb
->atom
.num_dw
++] = PKT3(PKT3_SET_LOOP_CONST
, num
, 0);
688 cb
->buf
[cb
->atom
.num_dw
++] = (reg
- R600_LOOP_CONST_OFFSET
) >> 2;
692 * Needs cb->pkt_flags set to RADEON_CP_PACKET3_COMPUTE_MODE for compute
695 static INLINE
void eg_store_loop_const_seq(struct r600_command_buffer
*cb
, unsigned reg
, unsigned num
)
697 assert(reg
>= EG_LOOP_CONST_OFFSET
);
698 assert(cb
->atom
.num_dw
+2+num
<= cb
->max_num_dw
);
699 cb
->buf
[cb
->atom
.num_dw
++] = PKT3(PKT3_SET_LOOP_CONST
, num
, 0) | cb
->pkt_flags
;
700 cb
->buf
[cb
->atom
.num_dw
++] = (reg
- EG_LOOP_CONST_OFFSET
) >> 2;
703 static INLINE
void r600_store_config_reg(struct r600_command_buffer
*cb
, unsigned reg
, unsigned value
)
705 r600_store_config_reg_seq(cb
, reg
, 1);
706 r600_store_value(cb
, value
);
709 static INLINE
void r600_store_context_reg(struct r600_command_buffer
*cb
, unsigned reg
, unsigned value
)
711 r600_store_context_reg_seq(cb
, reg
, 1);
712 r600_store_value(cb
, value
);
715 static INLINE
void r600_store_ctl_const(struct r600_command_buffer
*cb
, unsigned reg
, unsigned value
)
717 r600_store_ctl_const_seq(cb
, reg
, 1);
718 r600_store_value(cb
, value
);
721 static INLINE
void r600_store_loop_const(struct r600_command_buffer
*cb
, unsigned reg
, unsigned value
)
723 r600_store_loop_const_seq(cb
, reg
, 1);
724 r600_store_value(cb
, value
);
727 static INLINE
void eg_store_loop_const(struct r600_command_buffer
*cb
, unsigned reg
, unsigned value
)
729 eg_store_loop_const_seq(cb
, reg
, 1);
730 r600_store_value(cb
, value
);
733 void r600_init_command_buffer(struct r600_command_buffer
*cb
, unsigned num_dw
, enum r600_atom_flags flags
);
734 void r600_release_command_buffer(struct r600_command_buffer
*cb
);
737 * Helpers for emitting state into a command stream directly.
740 static INLINE
unsigned r600_context_bo_reloc(struct r600_context
*ctx
, struct r600_resource
*rbo
,
741 enum radeon_bo_usage usage
)
744 return ctx
->ws
->cs_add_reloc(ctx
->cs
, rbo
->cs_buf
, usage
, rbo
->domains
) * 4;
747 static INLINE
void r600_write_value(struct radeon_winsys_cs
*cs
, unsigned value
)
749 cs
->buf
[cs
->cdw
++] = value
;
752 static INLINE
void r600_write_array(struct radeon_winsys_cs
*cs
, unsigned num
, unsigned *ptr
)
754 assert(cs
->cdw
+num
<= RADEON_MAX_CMDBUF_DWORDS
);
755 memcpy(&cs
->buf
[cs
->cdw
], ptr
, num
* sizeof(ptr
[0]));
759 static INLINE
void r600_write_config_reg_seq(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned num
)
761 assert(reg
< R600_CONTEXT_REG_OFFSET
);
762 assert(cs
->cdw
+2+num
<= RADEON_MAX_CMDBUF_DWORDS
);
763 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_SET_CONFIG_REG
, num
, 0);
764 cs
->buf
[cs
->cdw
++] = (reg
- R600_CONFIG_REG_OFFSET
) >> 2;
767 static INLINE
void r600_write_context_reg_seq(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned num
)
769 assert(reg
>= R600_CONTEXT_REG_OFFSET
&& reg
< R600_CTL_CONST_OFFSET
);
770 assert(cs
->cdw
+2+num
<= RADEON_MAX_CMDBUF_DWORDS
);
771 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_SET_CONTEXT_REG
, num
, 0);
772 cs
->buf
[cs
->cdw
++] = (reg
- R600_CONTEXT_REG_OFFSET
) >> 2;
775 static INLINE
void r600_write_compute_context_reg_seq(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned num
)
777 r600_write_context_reg_seq(cs
, reg
, num
);
778 /* Set the compute bit on the packet header */
779 cs
->buf
[cs
->cdw
- 2] |= RADEON_CP_PACKET3_COMPUTE_MODE
;
782 static INLINE
void r600_write_ctl_const_seq(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned num
)
784 assert(reg
>= R600_CTL_CONST_OFFSET
);
785 assert(cs
->cdw
+2+num
<= RADEON_MAX_CMDBUF_DWORDS
);
786 cs
->buf
[cs
->cdw
++] = PKT3(PKT3_SET_CTL_CONST
, num
, 0);
787 cs
->buf
[cs
->cdw
++] = (reg
- R600_CTL_CONST_OFFSET
) >> 2;
790 static INLINE
void r600_write_config_reg(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned value
)
792 r600_write_config_reg_seq(cs
, reg
, 1);
793 r600_write_value(cs
, value
);
796 static INLINE
void r600_write_context_reg(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned value
)
798 r600_write_context_reg_seq(cs
, reg
, 1);
799 r600_write_value(cs
, value
);
802 static INLINE
void r600_write_compute_context_reg(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned value
)
804 r600_write_compute_context_reg_seq(cs
, reg
, 1);
805 r600_write_value(cs
, value
);
808 static INLINE
void r600_write_ctl_const(struct radeon_winsys_cs
*cs
, unsigned reg
, unsigned value
)
810 r600_write_ctl_const_seq(cs
, reg
, 1);
811 r600_write_value(cs
, value
);
817 static INLINE
uint32_t S_FIXED(float value
, uint32_t frac_bits
)
819 return value
* (1 << frac_bits
);
821 #define ALIGN_DIVUP(x, y) (((x) + (y) - 1) / (y))
823 static inline unsigned r600_tex_aniso_filter(unsigned filter
)
825 if (filter
<= 1) return 0;
826 if (filter
<= 2) return 1;
827 if (filter
<= 4) return 2;
828 if (filter
<= 8) return 3;
832 /* 12.4 fixed-point */
833 static INLINE
unsigned r600_pack_float_12p4(float x
)
836 x
>= 4096 ? 0xffff : x
* 16;
839 static INLINE
uint64_t r600_resource_va(struct pipe_screen
*screen
, struct pipe_resource
*resource
)
841 struct r600_screen
*rscreen
= (struct r600_screen
*)screen
;
842 struct r600_resource
*rresource
= (struct r600_resource
*)resource
;
844 return rscreen
->ws
->buffer_get_virtual_address(rresource
->cs_buf
);