r600g/compute: Disable growing the memory pool
[profile/ivi/mesa.git] / src / gallium / drivers / r600 / r600_pipe.h
1 /*
2  * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
3  *
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:
10  *
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
13  * Software.
14  *
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.
22  *
23  * Authors:
24  *      Jerome Glisse
25  */
26 #ifndef R600_PIPE_H
27 #define R600_PIPE_H
28
29 #include "util/u_slab.h"
30 #include "r600.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"
36
37 #define R600_MAX_CONST_BUFFERS 2
38 #define R600_MAX_CONST_BUFFER_SIZE 4096
39
40 #ifdef PIPE_ARCH_BIG_ENDIAN
41 #define R600_BIG_ENDIAN 1
42 #else
43 #define R600_BIG_ENDIAN 0
44 #endif
45
46 enum r600_atom_flags {
47         /* When set, atoms are added at the beginning of the dirty list
48          * instead of the end. */
49         EMIT_EARLY = (1 << 0)
50 };
51
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). */
55 struct r600_atom {
56         void (*emit)(struct r600_context *ctx, struct r600_atom *state);
57
58         unsigned                num_dw;
59         enum r600_atom_flags    flags;
60         bool                    dirty;
61
62         struct list_head        head;
63 };
64
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;
69         uint32_t *buf;
70         unsigned max_num_dw;
71         unsigned pkt_flags;
72 };
73
74 struct r600_surface_sync_cmd {
75         struct r600_atom atom;
76         unsigned flush_flags; /* CP_COHER_CNTL */
77 };
78
79 struct r600_db_misc_state {
80         struct r600_atom atom;
81         bool occlusion_query_enabled;
82         bool flush_depthstencil_enabled;
83 };
84
85 enum r600_pipe_state_id {
86         R600_PIPE_STATE_BLEND = 0,
87         R600_PIPE_STATE_BLEND_COLOR,
88         R600_PIPE_STATE_CONFIG,
89         R600_PIPE_STATE_SEAMLESS_CUBEMAP,
90         R600_PIPE_STATE_CLIP,
91         R600_PIPE_STATE_SCISSOR,
92         R600_PIPE_STATE_VIEWPORT,
93         R600_PIPE_STATE_RASTERIZER,
94         R600_PIPE_STATE_VGT,
95         R600_PIPE_STATE_FRAMEBUFFER,
96         R600_PIPE_STATE_DSA,
97         R600_PIPE_STATE_STENCIL_REF,
98         R600_PIPE_STATE_PS_SHADER,
99         R600_PIPE_STATE_VS_SHADER,
100         R600_PIPE_STATE_CONSTANT,
101         R600_PIPE_STATE_SAMPLER,
102         R600_PIPE_STATE_RESOURCE,
103         R600_PIPE_STATE_POLYGON_OFFSET,
104         R600_PIPE_STATE_FETCH_SHADER,
105         R600_PIPE_STATE_SPI,
106         R600_PIPE_NSTATES
107 };
108
109 struct compute_memory_pool;
110 void compute_memory_pool_delete(struct compute_memory_pool* pool);
111 struct compute_memory_pool* compute_memory_pool_new(
112         struct r600_screen *rscreen);
113
114 struct r600_pipe_fences {
115         struct r600_resource            *bo;
116         unsigned                        *data;
117         unsigned                        next_index;
118         /* linked list of preallocated blocks */
119         struct list_head                blocks;
120         /* linked list of freed fences */
121         struct list_head                pool;
122         pipe_mutex                      mutex;
123 };
124
125 struct r600_screen {
126         struct pipe_screen              screen;
127         struct radeon_winsys            *ws;
128         unsigned                        family;
129         enum chip_class                 chip_class;
130         struct radeon_info              info;
131         bool                            has_streamout;
132         struct r600_tiling_info         tiling_info;
133         struct r600_pipe_fences         fences;
134
135         bool                            use_surface_alloc;
136         int                             glsl_feature_level;
137
138         /*for compute global memory binding, we allocate stuff here, instead of
139          * buffers.
140          * XXX: Not sure if this is the best place for global_pool.  Also,
141          * it's not thread safe, so it won't work with multiple contexts. */
142         struct compute_memory_pool *global_pool;
143 };
144
145 struct r600_pipe_sampler_view {
146         struct pipe_sampler_view        base;
147         struct r600_pipe_resource_state         state;
148 };
149
150 struct r600_pipe_rasterizer {
151         struct r600_pipe_state          rstate;
152         boolean                         flatshade;
153         boolean                         two_side;
154         unsigned                        sprite_coord_enable;
155         unsigned                        clip_plane_enable;
156         unsigned                        pa_sc_line_stipple;
157         unsigned                        pa_cl_clip_cntl;
158         float                           offset_units;
159         float                           offset_scale;
160         bool                            scissor_enable;
161 };
162
163 struct r600_pipe_blend {
164         struct r600_pipe_state          rstate;
165         unsigned                        cb_target_mask;
166         unsigned                        cb_color_control;
167         bool                            dual_src_blend;
168 };
169
170 struct r600_pipe_dsa {
171         struct r600_pipe_state          rstate;
172         unsigned                        alpha_ref;
173         ubyte                           valuemask[2];
174         ubyte                           writemask[2];
175         bool                            is_flush;
176         unsigned                        sx_alpha_test_control;
177 };
178
179 struct r600_vertex_element
180 {
181         unsigned                        count;
182         struct pipe_vertex_element      elements[PIPE_MAX_ATTRIBS];
183         struct r600_resource            *fetch_shader;
184         unsigned                        fs_size;
185         struct r600_pipe_state          rstate;
186 };
187
188 struct r600_pipe_shader;
189
190 struct r600_pipe_shader_selector {
191         struct r600_pipe_shader *current;
192
193         struct tgsi_token       *tokens;
194         struct pipe_stream_output_info  so;
195
196         unsigned        num_shaders;
197
198         /* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
199         unsigned        type;
200
201         unsigned        nr_ps_max_color_exports;
202 };
203
204 struct r600_pipe_shader {
205         struct r600_pipe_shader_selector *selector;
206         struct r600_pipe_shader *next_variant;
207         struct r600_shader              shader;
208         struct r600_pipe_state          rstate;
209         struct r600_resource            *bo;
210         struct r600_resource            *bo_fetch;
211         struct r600_vertex_element      vertex_elements;
212         unsigned        sprite_coord_enable;
213         unsigned        flatshade;
214         unsigned        pa_cl_vs_out_cntl;
215         unsigned        ps_cb_shader_mask;
216         unsigned        key;
217         unsigned                db_shader_control;
218         unsigned                ps_depth_export;
219 };
220
221 struct r600_pipe_sampler_state {
222         struct r600_pipe_state          rstate;
223         boolean seamless_cube_map;
224 };
225
226 /* needed for blitter save */
227 #define NUM_TEX_UNITS 16
228
229 struct r600_textures_info {
230         struct r600_pipe_sampler_view   *views[NUM_TEX_UNITS];
231         struct r600_pipe_sampler_state  *samplers[NUM_TEX_UNITS];
232         unsigned                        n_views;
233         unsigned                        n_samplers;
234         bool                            samplers_dirty;
235         bool                            is_array_sampler[NUM_TEX_UNITS];
236 };
237
238 struct r600_fence {
239         struct pipe_reference           reference;
240         unsigned                        index; /* in the shared bo */
241         struct r600_resource            *sleep_bo;
242         struct list_head                head;
243 };
244
245 #define FENCE_BLOCK_SIZE 16
246
247 struct r600_fence_block {
248         struct r600_fence               fences[FENCE_BLOCK_SIZE];
249         struct list_head                head;
250 };
251
252 #define R600_CONSTANT_ARRAY_SIZE 256
253 #define R600_RESOURCE_ARRAY_SIZE 160
254
255 struct r600_stencil_ref
256 {
257         ubyte ref_value[2];
258         ubyte valuemask[2];
259         ubyte writemask[2];
260 };
261
262 struct r600_constbuf_state
263 {
264         struct r600_atom                atom;
265         struct pipe_constant_buffer     cb[PIPE_MAX_CONSTANT_BUFFERS];
266         uint32_t                        enabled_mask;
267         uint32_t                        dirty_mask;
268 };
269
270 struct r600_context {
271         struct pipe_context             context;
272         struct blitter_context          *blitter;
273         enum radeon_family              family;
274         enum chip_class                 chip_class;
275         boolean                         has_vertex_cache;
276         unsigned                        r6xx_num_clause_temp_gprs;
277         void                            *custom_dsa_flush;
278         struct r600_screen              *screen;
279         struct radeon_winsys            *ws;
280         struct r600_pipe_state          *states[R600_PIPE_NSTATES];
281         struct r600_vertex_element      *vertex_elements;
282         struct pipe_framebuffer_state   framebuffer;
283         unsigned                        cb_target_mask;
284         unsigned                        fb_cb_shader_mask;
285         unsigned                        sx_alpha_test_control;
286         unsigned                        cb_shader_mask;
287         unsigned                        db_shader_control;
288         unsigned                        cb_color_control;
289         unsigned                        pa_sc_line_stipple;
290         unsigned                        pa_cl_clip_cntl;
291         /* for saving when using blitter */
292         struct pipe_stencil_ref         stencil_ref;
293         struct pipe_viewport_state      viewport;
294         struct pipe_clip_state          clip;
295         struct r600_pipe_shader_selector        *ps_shader;
296         struct r600_pipe_shader_selector        *vs_shader;
297         struct r600_pipe_compute        *cs_shader;
298         struct r600_pipe_rasterizer     *rasterizer;
299         struct r600_pipe_state          vgt;
300         struct r600_pipe_state          spi;
301         struct pipe_query               *current_render_cond;
302         unsigned                        current_render_cond_mode;
303         struct pipe_query               *saved_render_cond;
304         unsigned                        saved_render_cond_mode;
305         /* shader information */
306         boolean                         two_side;
307         boolean                         spi_dirty;
308         unsigned                        sprite_coord_enable;
309         boolean                         flatshade;
310         boolean                         export_16bpc;
311         unsigned                        alpha_ref;
312         boolean                         alpha_ref_dirty;
313         unsigned                        nr_cbufs;
314         struct r600_textures_info       vs_samplers;
315         struct r600_textures_info       ps_samplers;
316
317         struct u_upload_mgr             *uploader;
318         struct util_slab_mempool        pool_transfers;
319         boolean                         have_depth_texture, have_depth_fb;
320
321         unsigned default_ps_gprs, default_vs_gprs;
322
323         /* States based on r600_atom. */
324         struct list_head                dirty_states;
325         struct r600_command_buffer      start_cs_cmd; /* invariant state mostly */
326         /** Compute specific registers initializations.  The start_cs_cmd atom
327          *  must be emitted before start_compute_cs_cmd. */
328         struct r600_command_buffer      start_compute_cs_cmd;
329         struct r600_surface_sync_cmd    surface_sync_cmd;
330         struct r600_atom                r6xx_flush_and_inv_cmd;
331         struct r600_db_misc_state       db_misc_state;
332         struct r600_atom                vertex_buffer_state;
333         struct r600_constbuf_state      vs_constbuf_state;
334         struct r600_constbuf_state      ps_constbuf_state;
335
336         struct radeon_winsys_cs *cs;
337
338         struct r600_range       *range;
339         unsigned                nblocks;
340         struct r600_block       **blocks;
341         struct list_head        dirty;
342         struct list_head        resource_dirty;
343         struct list_head        enable_list;
344         unsigned                pm4_dirty_cdwords;
345         unsigned                ctx_pm4_ndwords;
346
347         /* The list of active queries. Only one query of each type can be active. */
348         int                     num_occlusion_queries;
349
350         /* Manage queries in two separate groups:
351          * The timer ones and the others (streamout, occlusion).
352          *
353          * We do this because we should only suspend non-timer queries for u_blitter,
354          * and later if the non-timer queries are suspended, the context flush should
355          * only suspend and resume the timer queries. */
356         struct list_head        active_timer_queries;
357         unsigned                num_cs_dw_timer_queries_suspend;
358         struct list_head        active_nontimer_queries;
359         unsigned                num_cs_dw_nontimer_queries_suspend;
360
361         unsigned                num_cs_dw_streamout_end;
362
363         unsigned                backend_mask;
364         unsigned                max_db; /* for OQ */
365         unsigned                flags;
366         boolean                 predicate_drawing;
367         struct r600_range       ps_resources;
368         struct r600_range       vs_resources;
369         int                     num_ps_resources, num_vs_resources;
370
371         unsigned                num_so_targets;
372         struct r600_so_target   *so_targets[PIPE_MAX_SO_BUFFERS];
373         boolean                 streamout_start;
374         unsigned                streamout_append_bitmask;
375
376         /* There is no scissor enable bit on r6xx, so we must use a workaround.
377          * These track the current scissor state. */
378         bool                    scissor_enable;
379         struct pipe_scissor_state scissor_state;
380
381         /* With rasterizer discard, there doesn't have to be a pixel shader.
382          * In that case, we bind this one: */
383         void                    *dummy_pixel_shader;
384
385         boolean                 dual_src_blend;
386
387         /* Vertex and index buffers. */
388         bool                    vertex_buffers_dirty;
389         struct pipe_index_buffer index_buffer;
390         struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
391         unsigned                nr_vertex_buffers;
392 };
393
394 static INLINE void r600_emit_atom(struct r600_context *rctx, struct r600_atom *atom)
395 {
396         atom->emit(rctx, atom);
397         atom->dirty = false;
398         if (atom->head.next && atom->head.prev)
399                 LIST_DELINIT(&atom->head);
400 }
401
402 static INLINE void r600_atom_dirty(struct r600_context *rctx, struct r600_atom *state)
403 {
404         if (!state->dirty) {
405                 if (state->flags & EMIT_EARLY) {
406                         LIST_ADD(&state->head, &rctx->dirty_states);
407                 } else {
408                         LIST_ADDTAIL(&state->head, &rctx->dirty_states);
409                 }
410                 state->dirty = true;
411         }
412 }
413
414 /* evergreen_state.c */
415 void evergreen_init_state_functions(struct r600_context *rctx);
416 void evergreen_init_atom_start_cs(struct r600_context *rctx);
417 void evergreen_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader *shader);
418 void evergreen_pipe_shader_vs(struct pipe_context *ctx, struct r600_pipe_shader *shader);
419 void evergreen_fetch_shader(struct pipe_context *ctx, struct r600_vertex_element *ve);
420 void *evergreen_create_db_flush_dsa(struct r600_context *rctx);
421 void evergreen_polygon_offset_update(struct r600_context *rctx);
422 boolean evergreen_is_format_supported(struct pipe_screen *screen,
423                                       enum pipe_format format,
424                                       enum pipe_texture_target target,
425                                       unsigned sample_count,
426                                       unsigned usage);
427 void evergreen_cb(struct r600_context *rctx, struct r600_pipe_state *rstate,
428                          const struct pipe_framebuffer_state *state, int cb);
429
430
431 void evergreen_update_dual_export_state(struct r600_context * rctx);
432
433 /* r600_blit.c */
434 void r600_init_blit_functions(struct r600_context *rctx);
435 void r600_blit_uncompress_depth(struct pipe_context *ctx,
436                 struct r600_resource_texture *texture,
437                 struct r600_resource_texture *staging);
438 void r600_flush_depth_textures(struct r600_context *rctx);
439
440 /* r600_buffer.c */
441 bool r600_init_resource(struct r600_screen *rscreen,
442                         struct r600_resource *res,
443                         unsigned size, unsigned alignment,
444                         unsigned bind, unsigned usage);
445 struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
446                                          const struct pipe_resource *templ);
447
448 /* r600_pipe.c */
449 void r600_flush(struct pipe_context *ctx, struct pipe_fence_handle **fence,
450                 unsigned flags);
451
452 /* r600_query.c */
453 void r600_init_query_functions(struct r600_context *rctx);
454 void r600_suspend_nontimer_queries(struct r600_context *ctx);
455 void r600_resume_nontimer_queries(struct r600_context *ctx);
456 void r600_suspend_timer_queries(struct r600_context *ctx);
457 void r600_resume_timer_queries(struct r600_context *ctx);
458
459 /* r600_resource.c */
460 void r600_init_context_resource_functions(struct r600_context *r600);
461
462 /* r600_shader.c */
463 int r600_pipe_shader_create(struct pipe_context *ctx, struct r600_pipe_shader *shader);
464 #ifdef HAVE_OPENCL
465 int r600_compute_shader_create(struct pipe_context * ctx,
466         LLVMModuleRef mod,  struct r600_bytecode * bytecode);
467 #endif
468 void r600_pipe_shader_destroy(struct pipe_context *ctx, struct r600_pipe_shader *shader);
469
470 /* r600_state.c */
471 void r600_set_scissor_state(struct r600_context *rctx,
472                             const struct pipe_scissor_state *state);
473 void r600_update_sampler_states(struct r600_context *rctx);
474 void r600_init_state_functions(struct r600_context *rctx);
475 void r600_init_atom_start_cs(struct r600_context *rctx);
476 void r600_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader *shader);
477 void r600_pipe_shader_vs(struct pipe_context *ctx, struct r600_pipe_shader *shader);
478 void r600_fetch_shader(struct pipe_context *ctx, struct r600_vertex_element *ve);
479 void *r600_create_db_flush_dsa(struct r600_context *rctx);
480 void r600_polygon_offset_update(struct r600_context *rctx);
481 void r600_adjust_gprs(struct r600_context *rctx);
482 boolean r600_is_format_supported(struct pipe_screen *screen,
483                                  enum pipe_format format,
484                                  enum pipe_texture_target target,
485                                  unsigned sample_count,
486                                  unsigned usage);
487 void r600_update_dual_export_state(struct r600_context * rctx);
488
489 /* r600_texture.c */
490 void r600_init_screen_texture_functions(struct pipe_screen *screen);
491 void r600_init_surface_functions(struct r600_context *r600);
492 uint32_t r600_translate_texformat(struct pipe_screen *screen, enum pipe_format format,
493                                   const unsigned char *swizzle_view,
494                                   uint32_t *word4_p, uint32_t *yuv_format_p);
495 unsigned r600_texture_get_offset(struct r600_resource_texture *rtex,
496                                         unsigned level, unsigned layer);
497
498 /* r600_translate.c */
499 void r600_translate_index_buffer(struct r600_context *r600,
500                                  struct pipe_index_buffer *ib,
501                                  unsigned count);
502
503 /* r600_state_common.c */
504 void r600_init_atom(struct r600_atom *atom,
505                     void (*emit)(struct r600_context *ctx, struct r600_atom *state),
506                     unsigned num_dw, enum r600_atom_flags flags);
507 void r600_init_common_atoms(struct r600_context *rctx);
508 unsigned r600_get_cb_flush_flags(struct r600_context *rctx);
509 void r600_texture_barrier(struct pipe_context *ctx);
510 void r600_set_index_buffer(struct pipe_context *ctx,
511                            const struct pipe_index_buffer *ib);
512 void r600_set_vertex_buffers(struct pipe_context *ctx, unsigned count,
513                              const struct pipe_vertex_buffer *buffers);
514 void *r600_create_vertex_elements(struct pipe_context *ctx,
515                                   unsigned count,
516                                   const struct pipe_vertex_element *elements);
517 void r600_delete_vertex_element(struct pipe_context *ctx, void *state);
518 void r600_bind_blend_state(struct pipe_context *ctx, void *state);
519 void r600_set_blend_color(struct pipe_context *ctx,
520                           const struct pipe_blend_color *state);
521 void r600_bind_dsa_state(struct pipe_context *ctx, void *state);
522 void r600_set_max_scissor(struct r600_context *rctx);
523 void r600_bind_rs_state(struct pipe_context *ctx, void *state);
524 void r600_delete_rs_state(struct pipe_context *ctx, void *state);
525 void r600_sampler_view_destroy(struct pipe_context *ctx,
526                                struct pipe_sampler_view *state);
527 void r600_delete_state(struct pipe_context *ctx, void *state);
528 void r600_bind_vertex_elements(struct pipe_context *ctx, void *state);
529 void *r600_create_shader_state_ps(struct pipe_context *ctx,
530                    const struct pipe_shader_state *state);
531 void *r600_create_shader_state_vs(struct pipe_context *ctx,
532                    const struct pipe_shader_state *state);
533 void r600_bind_ps_shader(struct pipe_context *ctx, void *state);
534 void r600_bind_vs_shader(struct pipe_context *ctx, void *state);
535 void r600_delete_ps_shader(struct pipe_context *ctx, void *state);
536 void r600_delete_vs_shader(struct pipe_context *ctx, void *state);
537 void r600_constant_buffers_dirty(struct r600_context *rctx, struct r600_constbuf_state *state);
538 void r600_set_constant_buffer(struct pipe_context *ctx, uint shader, uint index,
539                               struct pipe_constant_buffer *cb);
540 struct pipe_stream_output_target *
541 r600_create_so_target(struct pipe_context *ctx,
542                       struct pipe_resource *buffer,
543                       unsigned buffer_offset,
544                       unsigned buffer_size);
545 void r600_so_target_destroy(struct pipe_context *ctx,
546                             struct pipe_stream_output_target *target);
547 void r600_set_so_targets(struct pipe_context *ctx,
548                          unsigned num_targets,
549                          struct pipe_stream_output_target **targets,
550                          unsigned append_bitmask);
551 void r600_set_pipe_stencil_ref(struct pipe_context *ctx,
552                                const struct pipe_stencil_ref *state);
553 void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info);
554 uint32_t r600_translate_stencil_op(int s_op);
555 uint32_t r600_translate_fill(uint32_t func);
556 unsigned r600_tex_wrap(unsigned wrap);
557 unsigned r600_tex_filter(unsigned filter);
558 unsigned r600_tex_mipfilter(unsigned filter);
559 unsigned r600_tex_compare(unsigned compare);
560
561 /*
562  * Helpers for building command buffers
563  */
564
565 #define PKT3_SET_CONFIG_REG     0x68
566 #define PKT3_SET_CONTEXT_REG    0x69
567 #define PKT3_SET_CTL_CONST      0x6F
568 #define PKT3_SET_LOOP_CONST                    0x6C
569
570 #define R600_CONFIG_REG_OFFSET  0x08000
571 #define R600_CONTEXT_REG_OFFSET 0x28000
572 #define R600_CTL_CONST_OFFSET   0x3CFF0
573 #define R600_LOOP_CONST_OFFSET                 0X0003E200
574 #define EG_LOOP_CONST_OFFSET               0x0003A200
575
576 #define PKT_TYPE_S(x)                   (((x) & 0x3) << 30)
577 #define PKT_COUNT_S(x)                  (((x) & 0x3FFF) << 16)
578 #define PKT3_IT_OPCODE_S(x)             (((x) & 0xFF) << 8)
579 #define PKT3_PREDICATE(x)               (((x) >> 0) & 0x1)
580 #define PKT3(op, count, predicate) (PKT_TYPE_S(3) | PKT_COUNT_S(count) | PKT3_IT_OPCODE_S(op) | PKT3_PREDICATE(predicate))
581
582 static INLINE void r600_store_value(struct r600_command_buffer *cb, unsigned value)
583 {
584         cb->buf[cb->atom.num_dw++] = value;
585 }
586
587 static INLINE void r600_store_config_reg_seq(struct r600_command_buffer *cb, unsigned reg, unsigned num)
588 {
589         assert(reg < R600_CONTEXT_REG_OFFSET);
590         assert(cb->atom.num_dw+2+num <= cb->max_num_dw);
591         cb->buf[cb->atom.num_dw++] = PKT3(PKT3_SET_CONFIG_REG, num, 0);
592         cb->buf[cb->atom.num_dw++] = (reg - R600_CONFIG_REG_OFFSET) >> 2;
593 }
594
595 /**
596  * Needs cb->pkt_flags set to  RADEON_CP_PACKET3_COMPUTE_MODE for compute
597  * shaders.
598  */
599 static INLINE void r600_store_context_reg_seq(struct r600_command_buffer *cb, unsigned reg, unsigned num)
600 {
601         assert(reg >= R600_CONTEXT_REG_OFFSET && reg < R600_CTL_CONST_OFFSET);
602         assert(cb->atom.num_dw+2+num <= cb->max_num_dw);
603         cb->buf[cb->atom.num_dw++] = PKT3(PKT3_SET_CONTEXT_REG, num, 0) | cb->pkt_flags;
604         cb->buf[cb->atom.num_dw++] = (reg - R600_CONTEXT_REG_OFFSET) >> 2;
605 }
606
607 /**
608  * Needs cb->pkt_flags set to  RADEON_CP_PACKET3_COMPUTE_MODE for compute
609  * shaders.
610  */
611 static INLINE void r600_store_ctl_const_seq(struct r600_command_buffer *cb, unsigned reg, unsigned num)
612 {
613         assert(reg >= R600_CTL_CONST_OFFSET);
614         assert(cb->atom.num_dw+2+num <= cb->max_num_dw);
615         cb->buf[cb->atom.num_dw++] = PKT3(PKT3_SET_CTL_CONST, num, 0) | cb->pkt_flags;
616         cb->buf[cb->atom.num_dw++] = (reg - R600_CTL_CONST_OFFSET) >> 2;
617 }
618
619 static INLINE void r600_store_loop_const_seq(struct r600_command_buffer *cb, unsigned reg, unsigned num)
620 {
621         assert(reg >= R600_LOOP_CONST_OFFSET);
622         assert(cb->atom.num_dw+2+num <= cb->max_num_dw);
623         cb->buf[cb->atom.num_dw++] = PKT3(PKT3_SET_LOOP_CONST, num, 0);
624         cb->buf[cb->atom.num_dw++] = (reg - R600_LOOP_CONST_OFFSET) >> 2;
625 }
626
627 /**
628  * Needs cb->pkt_flags set to  RADEON_CP_PACKET3_COMPUTE_MODE for compute
629  * shaders.
630  */
631 static INLINE void eg_store_loop_const_seq(struct r600_command_buffer *cb, unsigned reg, unsigned num)
632 {
633         assert(reg >= EG_LOOP_CONST_OFFSET);
634         assert(cb->atom.num_dw+2+num <= cb->max_num_dw);
635         cb->buf[cb->atom.num_dw++] = PKT3(PKT3_SET_LOOP_CONST, num, 0) | cb->pkt_flags;
636         cb->buf[cb->atom.num_dw++] = (reg - EG_LOOP_CONST_OFFSET) >> 2;
637 }
638
639 static INLINE void r600_store_config_reg(struct r600_command_buffer *cb, unsigned reg, unsigned value)
640 {
641         r600_store_config_reg_seq(cb, reg, 1);
642         r600_store_value(cb, value);
643 }
644
645 static INLINE void r600_store_context_reg(struct r600_command_buffer *cb, unsigned reg, unsigned value)
646 {
647         r600_store_context_reg_seq(cb, reg, 1);
648         r600_store_value(cb, value);
649 }
650
651 static INLINE void r600_store_ctl_const(struct r600_command_buffer *cb, unsigned reg, unsigned value)
652 {
653         r600_store_ctl_const_seq(cb, reg, 1);
654         r600_store_value(cb, value);
655 }
656
657 static INLINE void r600_store_loop_const(struct r600_command_buffer *cb, unsigned reg, unsigned value)
658 {
659         r600_store_loop_const_seq(cb, reg, 1);
660         r600_store_value(cb, value);
661 }
662
663 static INLINE void eg_store_loop_const(struct r600_command_buffer *cb, unsigned reg, unsigned value)
664 {
665         eg_store_loop_const_seq(cb, reg, 1);
666         r600_store_value(cb, value);
667 }
668
669 void r600_init_command_buffer(struct r600_command_buffer *cb, unsigned num_dw, enum r600_atom_flags flags);
670 void r600_release_command_buffer(struct r600_command_buffer *cb);
671
672 /*
673  * Helpers for emitting state into a command stream directly.
674  */
675
676 static INLINE unsigned r600_context_bo_reloc(struct r600_context *ctx, struct r600_resource *rbo,
677                                              enum radeon_bo_usage usage)
678 {
679         assert(usage);
680         return ctx->ws->cs_add_reloc(ctx->cs, rbo->cs_buf, usage, rbo->domains) * 4;
681 }
682
683 static INLINE void r600_write_value(struct radeon_winsys_cs *cs, unsigned value)
684 {
685         cs->buf[cs->cdw++] = value;
686 }
687
688 static INLINE void r600_write_config_reg_seq(struct radeon_winsys_cs *cs, unsigned reg, unsigned num)
689 {
690         assert(reg < R600_CONTEXT_REG_OFFSET);
691         assert(cs->cdw+2+num <= RADEON_MAX_CMDBUF_DWORDS);
692         cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONFIG_REG, num, 0);
693         cs->buf[cs->cdw++] = (reg - R600_CONFIG_REG_OFFSET) >> 2;
694 }
695
696 static INLINE void r600_write_context_reg_seq(struct radeon_winsys_cs *cs, unsigned reg, unsigned num)
697 {
698         assert(reg >= R600_CONTEXT_REG_OFFSET && reg < R600_CTL_CONST_OFFSET);
699         assert(cs->cdw+2+num <= RADEON_MAX_CMDBUF_DWORDS);
700         cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, num, 0);
701         cs->buf[cs->cdw++] = (reg - R600_CONTEXT_REG_OFFSET) >> 2;
702 }
703
704 static INLINE void r600_write_ctl_const_seq(struct radeon_winsys_cs *cs, unsigned reg, unsigned num)
705 {
706         assert(reg >= R600_CTL_CONST_OFFSET);
707         assert(cs->cdw+2+num <= RADEON_MAX_CMDBUF_DWORDS);
708         cs->buf[cs->cdw++] = PKT3(PKT3_SET_CTL_CONST, num, 0);
709         cs->buf[cs->cdw++] = (reg - R600_CTL_CONST_OFFSET) >> 2;
710 }
711
712 static INLINE void r600_write_config_reg(struct radeon_winsys_cs *cs, unsigned reg, unsigned value)
713 {
714         r600_write_config_reg_seq(cs, reg, 1);
715         r600_write_value(cs, value);
716 }
717
718 static INLINE void r600_write_context_reg(struct radeon_winsys_cs *cs, unsigned reg, unsigned value)
719 {
720         r600_write_context_reg_seq(cs, reg, 1);
721         r600_write_value(cs, value);
722 }
723
724 static INLINE void r600_write_ctl_const(struct radeon_winsys_cs *cs, unsigned reg, unsigned value)
725 {
726         r600_write_ctl_const_seq(cs, reg, 1);
727         r600_write_value(cs, value);
728 }
729
730 /*
731  * common helpers
732  */
733 static INLINE uint32_t S_FIXED(float value, uint32_t frac_bits)
734 {
735         return value * (1 << frac_bits);
736 }
737 #define ALIGN_DIVUP(x, y) (((x) + (y) - 1) / (y))
738
739 static inline unsigned r600_tex_aniso_filter(unsigned filter)
740 {
741         if (filter <= 1)   return 0;
742         if (filter <= 2)   return 1;
743         if (filter <= 4)   return 2;
744         if (filter <= 8)   return 3;
745          /* else */        return 4;
746 }
747
748 /* 12.4 fixed-point */
749 static INLINE unsigned r600_pack_float_12p4(float x)
750 {
751         return x <= 0    ? 0 :
752                x >= 4096 ? 0xffff : x * 16;
753 }
754
755 static INLINE uint64_t r600_resource_va(struct pipe_screen *screen, struct pipe_resource *resource)
756 {
757         struct r600_screen *rscreen = (struct r600_screen*)screen;
758         struct r600_resource *rresource = (struct r600_resource*)resource;
759
760         return rscreen->ws->buffer_get_virtual_address(rresource->cs_buf);
761 }
762
763 #endif