r600_pipe.h revision d1ca16b2738644d17c10e736ca36981f69f5fa87
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
46enum 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). */
55struct 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. */
67struct r600_command_buffer {
68	struct r600_atom atom;
69	uint32_t *buf;
70	unsigned max_num_dw;
71	unsigned pkt_flags;
72};
73
74struct r600_surface_sync_cmd {
75	struct r600_atom atom;
76	unsigned flush_flags; /* CP_COHER_CNTL */
77};
78
79struct r600_db_misc_state {
80	struct r600_atom atom;
81	bool occlusion_query_enabled;
82	bool flush_depthstencil_through_cb;
83};
84
85struct r600_cb_misc_state {
86	struct r600_atom atom;
87	unsigned cb_color_control; /* this comes from blend state */
88	unsigned blend_colormask; /* 8*4 bits for 8 RGBA colorbuffers */
89	unsigned nr_cbufs;
90	unsigned nr_ps_color_outputs;
91	bool multiwrite;
92	bool dual_src_blend;
93};
94
95enum r600_pipe_state_id {
96	R600_PIPE_STATE_BLEND = 0,
97	R600_PIPE_STATE_BLEND_COLOR,
98	R600_PIPE_STATE_CONFIG,
99	R600_PIPE_STATE_SEAMLESS_CUBEMAP,
100	R600_PIPE_STATE_CLIP,
101	R600_PIPE_STATE_SCISSOR,
102	R600_PIPE_STATE_VIEWPORT,
103	R600_PIPE_STATE_RASTERIZER,
104	R600_PIPE_STATE_VGT,
105	R600_PIPE_STATE_FRAMEBUFFER,
106	R600_PIPE_STATE_DSA,
107	R600_PIPE_STATE_STENCIL_REF,
108	R600_PIPE_STATE_PS_SHADER,
109	R600_PIPE_STATE_VS_SHADER,
110	R600_PIPE_STATE_CONSTANT,
111	R600_PIPE_STATE_SAMPLER,
112	R600_PIPE_STATE_RESOURCE,
113	R600_PIPE_STATE_POLYGON_OFFSET,
114	R600_PIPE_STATE_FETCH_SHADER,
115	R600_PIPE_STATE_SPI,
116	R600_PIPE_NSTATES
117};
118
119struct compute_memory_pool;
120void compute_memory_pool_delete(struct compute_memory_pool* pool);
121struct compute_memory_pool* compute_memory_pool_new(
122	struct r600_screen *rscreen);
123
124struct r600_pipe_fences {
125	struct r600_resource		*bo;
126	unsigned			*data;
127	unsigned			next_index;
128	/* linked list of preallocated blocks */
129	struct list_head		blocks;
130	/* linked list of freed fences */
131	struct list_head		pool;
132	pipe_mutex			mutex;
133};
134
135struct r600_screen {
136	struct pipe_screen		screen;
137	struct radeon_winsys		*ws;
138	unsigned			family;
139	enum chip_class			chip_class;
140	struct radeon_info		info;
141	bool				has_streamout;
142	struct r600_tiling_info		tiling_info;
143	struct r600_pipe_fences		fences;
144
145	bool				use_surface_alloc;
146
147	/*for compute global memory binding, we allocate stuff here, instead of
148	 * buffers.
149	 * XXX: Not sure if this is the best place for global_pool.  Also,
150	 * it's not thread safe, so it won't work with multiple contexts. */
151	struct compute_memory_pool *global_pool;
152};
153
154struct r600_pipe_sampler_view {
155	struct pipe_sampler_view	base;
156	struct r600_pipe_resource_state		state;
157};
158
159struct r600_pipe_rasterizer {
160	struct r600_pipe_state		rstate;
161	boolean				flatshade;
162	boolean				two_side;
163	unsigned			sprite_coord_enable;
164	unsigned                        clip_plane_enable;
165	unsigned			pa_sc_line_stipple;
166	unsigned			pa_cl_clip_cntl;
167	float				offset_units;
168	float				offset_scale;
169	bool				scissor_enable;
170};
171
172struct r600_pipe_blend {
173	struct r600_pipe_state		rstate;
174	unsigned			cb_target_mask;
175	unsigned			cb_color_control;
176	bool				dual_src_blend;
177};
178
179struct r600_pipe_dsa {
180	struct r600_pipe_state		rstate;
181	unsigned			alpha_ref;
182	ubyte				valuemask[2];
183	ubyte				writemask[2];
184	unsigned                        sx_alpha_test_control;
185};
186
187struct r600_vertex_element
188{
189	unsigned			count;
190	struct pipe_vertex_element	elements[PIPE_MAX_ATTRIBS];
191	struct r600_resource		*fetch_shader;
192	unsigned			fs_size;
193	struct r600_pipe_state		rstate;
194};
195
196struct r600_pipe_shader;
197
198struct r600_pipe_shader_selector {
199	struct r600_pipe_shader *current;
200
201	struct tgsi_token       *tokens;
202	struct pipe_stream_output_info  so;
203
204	unsigned	num_shaders;
205
206	/* PIPE_SHADER_[VERTEX|FRAGMENT|...] */
207	unsigned	type;
208
209	unsigned	nr_ps_max_color_exports;
210};
211
212struct r600_pipe_shader {
213	struct r600_pipe_shader_selector *selector;
214	struct r600_pipe_shader	*next_variant;
215	struct r600_shader		shader;
216	struct r600_pipe_state		rstate;
217	struct r600_resource		*bo;
218	struct r600_resource		*bo_fetch;
219	struct r600_vertex_element	vertex_elements;
220	unsigned	sprite_coord_enable;
221	unsigned	flatshade;
222	unsigned	pa_cl_vs_out_cntl;
223	unsigned	nr_ps_color_outputs;
224	unsigned	key;
225	unsigned		db_shader_control;
226	unsigned		ps_depth_export;
227};
228
229struct r600_pipe_sampler_state {
230	struct r600_pipe_state		rstate;
231	boolean seamless_cube_map;
232};
233
234/* needed for blitter save */
235#define NUM_TEX_UNITS 16
236
237struct r600_textures_info {
238	struct r600_pipe_sampler_view	*views[NUM_TEX_UNITS];
239	struct r600_pipe_sampler_state	*samplers[NUM_TEX_UNITS];
240	unsigned			n_views;
241	unsigned			n_samplers;
242	bool				samplers_dirty;
243	bool				is_array_sampler[NUM_TEX_UNITS];
244};
245
246struct r600_fence {
247	struct pipe_reference		reference;
248	unsigned			index; /* in the shared bo */
249	struct r600_resource            *sleep_bo;
250	struct list_head		head;
251};
252
253#define FENCE_BLOCK_SIZE 16
254
255struct r600_fence_block {
256	struct r600_fence		fences[FENCE_BLOCK_SIZE];
257	struct list_head		head;
258};
259
260#define R600_CONSTANT_ARRAY_SIZE 256
261#define R600_RESOURCE_ARRAY_SIZE 160
262
263struct r600_stencil_ref
264{
265	ubyte ref_value[2];
266	ubyte valuemask[2];
267	ubyte writemask[2];
268};
269
270struct r600_constbuf_state
271{
272	struct r600_atom		atom;
273	struct pipe_constant_buffer	cb[PIPE_MAX_CONSTANT_BUFFERS];
274	uint32_t			enabled_mask;
275	uint32_t			dirty_mask;
276};
277
278struct r600_vertexbuf_state
279{
280	struct r600_atom		atom;
281	struct pipe_vertex_buffer	vb[PIPE_MAX_ATTRIBS];
282	uint32_t			enabled_mask; /* non-NULL buffers */
283	uint32_t			dirty_mask;
284};
285
286struct r600_context {
287	struct pipe_context		context;
288	struct blitter_context		*blitter;
289	enum radeon_family		family;
290	enum chip_class			chip_class;
291	boolean				has_vertex_cache;
292	unsigned			r6xx_num_clause_temp_gprs;
293	void				*custom_dsa_flush;
294	struct r600_screen		*screen;
295	struct radeon_winsys		*ws;
296	struct r600_pipe_state		*states[R600_PIPE_NSTATES];
297	struct r600_vertex_element	*vertex_elements;
298	struct pipe_framebuffer_state	framebuffer;
299	unsigned			compute_cb_target_mask;
300	unsigned			sx_alpha_test_control;
301	unsigned			db_shader_control;
302	unsigned			pa_sc_line_stipple;
303	unsigned			pa_cl_clip_cntl;
304	/* for saving when using blitter */
305	struct pipe_stencil_ref		stencil_ref;
306	struct pipe_viewport_state	viewport;
307	struct pipe_clip_state		clip;
308	struct r600_pipe_shader_selector 	*ps_shader;
309	struct r600_pipe_shader_selector 	*vs_shader;
310	struct r600_pipe_compute	*cs_shader;
311	struct r600_pipe_rasterizer	*rasterizer;
312	struct r600_pipe_state          vgt;
313	struct r600_pipe_state          spi;
314	struct pipe_query		*current_render_cond;
315	unsigned			current_render_cond_mode;
316	struct pipe_query		*saved_render_cond;
317	unsigned			saved_render_cond_mode;
318	/* shader information */
319	boolean				two_side;
320	boolean				spi_dirty;
321	unsigned			sprite_coord_enable;
322	boolean				flatshade;
323	boolean				export_16bpc;
324	unsigned			alpha_ref;
325	boolean				alpha_ref_dirty;
326	unsigned			nr_cbufs;
327	struct r600_textures_info	vs_samplers;
328	struct r600_textures_info	ps_samplers;
329
330	struct u_upload_mgr	        *uploader;
331	struct util_slab_mempool	pool_transfers;
332	boolean				have_depth_texture;
333
334	unsigned default_ps_gprs, default_vs_gprs;
335
336	/* States based on r600_atom. */
337	struct list_head		dirty_states;
338	struct r600_command_buffer	start_cs_cmd; /* invariant state mostly */
339	/** Compute specific registers initializations.  The start_cs_cmd atom
340	 *  must be emitted before start_compute_cs_cmd. */
341        struct r600_command_buffer      start_compute_cs_cmd;
342	struct r600_surface_sync_cmd	surface_sync_cmd;
343	struct r600_atom		r6xx_flush_and_inv_cmd;
344	struct r600_cb_misc_state	cb_misc_state;
345	struct r600_db_misc_state	db_misc_state;
346	/** Vertex buffers for fetch shaders */
347	struct r600_vertexbuf_state	vertex_buffer_state;
348	/** Vertex buffers for compute shaders */
349	struct r600_vertexbuf_state	cs_vertex_buffer_state;
350	struct r600_constbuf_state	vs_constbuf_state;
351	struct r600_constbuf_state	ps_constbuf_state;
352
353	struct radeon_winsys_cs	*cs;
354
355	struct r600_range	*range;
356	unsigned		nblocks;
357	struct r600_block	**blocks;
358	struct list_head	dirty;
359	struct list_head	resource_dirty;
360	struct list_head	enable_list;
361	unsigned		pm4_dirty_cdwords;
362	unsigned		ctx_pm4_ndwords;
363
364	/* The list of active queries. Only one query of each type can be active. */
365	int			num_occlusion_queries;
366
367	/* Manage queries in two separate groups:
368	 * The timer ones and the others (streamout, occlusion).
369	 *
370	 * We do this because we should only suspend non-timer queries for u_blitter,
371	 * and later if the non-timer queries are suspended, the context flush should
372	 * only suspend and resume the timer queries. */
373	struct list_head	active_timer_queries;
374	unsigned		num_cs_dw_timer_queries_suspend;
375	struct list_head	active_nontimer_queries;
376	unsigned		num_cs_dw_nontimer_queries_suspend;
377
378	unsigned		num_cs_dw_streamout_end;
379
380	unsigned		backend_mask;
381	unsigned                max_db; /* for OQ */
382	unsigned		flags;
383	boolean                 predicate_drawing;
384	struct r600_range	ps_resources;
385	struct r600_range	vs_resources;
386	int			num_ps_resources, num_vs_resources;
387
388	unsigned		num_so_targets;
389	struct r600_so_target	*so_targets[PIPE_MAX_SO_BUFFERS];
390	boolean			streamout_start;
391	unsigned		streamout_append_bitmask;
392
393	/* There is no scissor enable bit on r6xx, so we must use a workaround.
394	 * These track the current scissor state. */
395	bool			scissor_enable;
396	struct pipe_scissor_state scissor_state;
397
398	/* With rasterizer discard, there doesn't have to be a pixel shader.
399	 * In that case, we bind this one: */
400	void			*dummy_pixel_shader;
401
402	boolean			dual_src_blend;
403
404	/* Index buffer. */
405	struct pipe_index_buffer index_buffer;
406};
407
408static INLINE void r600_emit_atom(struct r600_context *rctx, struct r600_atom *atom)
409{
410	atom->emit(rctx, atom);
411	atom->dirty = false;
412	if (atom->head.next && atom->head.prev)
413		LIST_DELINIT(&atom->head);
414}
415
416static INLINE void r600_atom_dirty(struct r600_context *rctx, struct r600_atom *state)
417{
418	if (!state->dirty) {
419		if (state->flags & EMIT_EARLY) {
420			LIST_ADD(&state->head, &rctx->dirty_states);
421		} else {
422			LIST_ADDTAIL(&state->head, &rctx->dirty_states);
423		}
424		state->dirty = true;
425	}
426}
427
428/* evergreen_state.c */
429void evergreen_init_state_functions(struct r600_context *rctx);
430void evergreen_init_atom_start_cs(struct r600_context *rctx);
431void evergreen_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader *shader);
432void evergreen_pipe_shader_vs(struct pipe_context *ctx, struct r600_pipe_shader *shader);
433void evergreen_fetch_shader(struct pipe_context *ctx, struct r600_vertex_element *ve);
434void *evergreen_create_db_flush_dsa(struct r600_context *rctx);
435void evergreen_polygon_offset_update(struct r600_context *rctx);
436boolean evergreen_is_format_supported(struct pipe_screen *screen,
437				      enum pipe_format format,
438				      enum pipe_texture_target target,
439				      unsigned sample_count,
440				      unsigned usage);
441void evergreen_cb(struct r600_context *rctx, struct r600_pipe_state *rstate,
442                         const struct pipe_framebuffer_state *state, int cb);
443
444
445void evergreen_update_dual_export_state(struct r600_context * rctx);
446
447/* r600_blit.c */
448void r600_init_blit_functions(struct r600_context *rctx);
449void r600_blit_uncompress_depth(struct pipe_context *ctx,
450		struct r600_resource_texture *texture,
451		struct r600_resource_texture *staging,
452		unsigned first_level, unsigned last_level,
453		unsigned first_layer, unsigned last_layer);
454void r600_flush_all_depth_textures(struct r600_context *rctx);
455
456/* r600_buffer.c */
457bool r600_init_resource(struct r600_screen *rscreen,
458			struct r600_resource *res,
459			unsigned size, unsigned alignment,
460			unsigned bind, unsigned usage);
461struct pipe_resource *r600_buffer_create(struct pipe_screen *screen,
462					 const struct pipe_resource *templ);
463
464/* r600_pipe.c */
465void r600_flush(struct pipe_context *ctx, struct pipe_fence_handle **fence,
466		unsigned flags);
467
468/* r600_query.c */
469void r600_init_query_functions(struct r600_context *rctx);
470void r600_suspend_nontimer_queries(struct r600_context *ctx);
471void r600_resume_nontimer_queries(struct r600_context *ctx);
472void r600_suspend_timer_queries(struct r600_context *ctx);
473void r600_resume_timer_queries(struct r600_context *ctx);
474
475/* r600_resource.c */
476void r600_init_context_resource_functions(struct r600_context *r600);
477
478/* r600_shader.c */
479int r600_pipe_shader_create(struct pipe_context *ctx, struct r600_pipe_shader *shader);
480#ifdef HAVE_OPENCL
481int r600_compute_shader_create(struct pipe_context * ctx,
482	LLVMModuleRef mod,  struct r600_bytecode * bytecode);
483#endif
484void r600_pipe_shader_destroy(struct pipe_context *ctx, struct r600_pipe_shader *shader);
485
486/* r600_state.c */
487void r600_set_scissor_state(struct r600_context *rctx,
488			    const struct pipe_scissor_state *state);
489void r600_update_sampler_states(struct r600_context *rctx);
490void r600_init_state_functions(struct r600_context *rctx);
491void r600_init_atom_start_cs(struct r600_context *rctx);
492void r600_pipe_shader_ps(struct pipe_context *ctx, struct r600_pipe_shader *shader);
493void r600_pipe_shader_vs(struct pipe_context *ctx, struct r600_pipe_shader *shader);
494void r600_fetch_shader(struct pipe_context *ctx, struct r600_vertex_element *ve);
495void *r600_create_db_flush_dsa(struct r600_context *rctx);
496void r600_polygon_offset_update(struct r600_context *rctx);
497void r600_adjust_gprs(struct r600_context *rctx);
498boolean r600_is_format_supported(struct pipe_screen *screen,
499				 enum pipe_format format,
500				 enum pipe_texture_target target,
501				 unsigned sample_count,
502				 unsigned usage);
503void r600_update_dual_export_state(struct r600_context * rctx);
504
505/* r600_texture.c */
506void r600_init_screen_texture_functions(struct pipe_screen *screen);
507void r600_init_surface_functions(struct r600_context *r600);
508uint32_t r600_translate_texformat(struct pipe_screen *screen, enum pipe_format format,
509				  const unsigned char *swizzle_view,
510				  uint32_t *word4_p, uint32_t *yuv_format_p);
511unsigned r600_texture_get_offset(struct r600_resource_texture *rtex,
512					unsigned level, unsigned layer);
513
514/* r600_translate.c */
515void r600_translate_index_buffer(struct r600_context *r600,
516				 struct pipe_index_buffer *ib,
517				 unsigned count);
518
519/* r600_state_common.c */
520void r600_init_atom(struct r600_atom *atom,
521		    void (*emit)(struct r600_context *ctx, struct r600_atom *state),
522		    unsigned num_dw, enum r600_atom_flags flags);
523void r600_init_common_atoms(struct r600_context *rctx);
524unsigned r600_get_cb_flush_flags(struct r600_context *rctx);
525void r600_texture_barrier(struct pipe_context *ctx);
526void r600_set_index_buffer(struct pipe_context *ctx,
527			   const struct pipe_index_buffer *ib);
528void r600_vertex_buffers_dirty(struct r600_context *rctx);
529void r600_set_vertex_buffers(struct pipe_context *ctx, unsigned count,
530			     const struct pipe_vertex_buffer *input);
531void *r600_create_vertex_elements(struct pipe_context *ctx,
532				  unsigned count,
533				  const struct pipe_vertex_element *elements);
534void r600_delete_vertex_element(struct pipe_context *ctx, void *state);
535void r600_bind_blend_state(struct pipe_context *ctx, void *state);
536void r600_set_blend_color(struct pipe_context *ctx,
537			  const struct pipe_blend_color *state);
538void r600_bind_dsa_state(struct pipe_context *ctx, void *state);
539void r600_set_max_scissor(struct r600_context *rctx);
540void r600_bind_rs_state(struct pipe_context *ctx, void *state);
541void r600_delete_rs_state(struct pipe_context *ctx, void *state);
542void r600_sampler_view_destroy(struct pipe_context *ctx,
543			       struct pipe_sampler_view *state);
544void r600_delete_state(struct pipe_context *ctx, void *state);
545void r600_bind_vertex_elements(struct pipe_context *ctx, void *state);
546void *r600_create_shader_state_ps(struct pipe_context *ctx,
547                   const struct pipe_shader_state *state);
548void *r600_create_shader_state_vs(struct pipe_context *ctx,
549                   const struct pipe_shader_state *state);
550void r600_bind_ps_shader(struct pipe_context *ctx, void *state);
551void r600_bind_vs_shader(struct pipe_context *ctx, void *state);
552void r600_delete_ps_shader(struct pipe_context *ctx, void *state);
553void r600_delete_vs_shader(struct pipe_context *ctx, void *state);
554void r600_constant_buffers_dirty(struct r600_context *rctx, struct r600_constbuf_state *state);
555void r600_set_constant_buffer(struct pipe_context *ctx, uint shader, uint index,
556			      struct pipe_constant_buffer *cb);
557struct pipe_stream_output_target *
558r600_create_so_target(struct pipe_context *ctx,
559		      struct pipe_resource *buffer,
560		      unsigned buffer_offset,
561		      unsigned buffer_size);
562void r600_so_target_destroy(struct pipe_context *ctx,
563			    struct pipe_stream_output_target *target);
564void r600_set_so_targets(struct pipe_context *ctx,
565			 unsigned num_targets,
566			 struct pipe_stream_output_target **targets,
567			 unsigned append_bitmask);
568void r600_set_pipe_stencil_ref(struct pipe_context *ctx,
569			       const struct pipe_stencil_ref *state);
570void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info);
571uint32_t r600_translate_stencil_op(int s_op);
572uint32_t r600_translate_fill(uint32_t func);
573unsigned r600_tex_wrap(unsigned wrap);
574unsigned r600_tex_filter(unsigned filter);
575unsigned r600_tex_mipfilter(unsigned filter);
576unsigned r600_tex_compare(unsigned compare);
577
578/*
579 * Helpers for building command buffers
580 */
581
582#define PKT3_SET_CONFIG_REG	0x68
583#define PKT3_SET_CONTEXT_REG	0x69
584#define PKT3_SET_CTL_CONST      0x6F
585#define PKT3_SET_LOOP_CONST                    0x6C
586
587#define R600_CONFIG_REG_OFFSET	0x08000
588#define R600_CONTEXT_REG_OFFSET 0x28000
589#define R600_CTL_CONST_OFFSET   0x3CFF0
590#define R600_LOOP_CONST_OFFSET                 0X0003E200
591#define EG_LOOP_CONST_OFFSET               0x0003A200
592
593#define PKT_TYPE_S(x)                   (((x) & 0x3) << 30)
594#define PKT_COUNT_S(x)                  (((x) & 0x3FFF) << 16)
595#define PKT3_IT_OPCODE_S(x)             (((x) & 0xFF) << 8)
596#define PKT3_PREDICATE(x)               (((x) >> 0) & 0x1)
597#define PKT3(op, count, predicate) (PKT_TYPE_S(3) | PKT_COUNT_S(count) | PKT3_IT_OPCODE_S(op) | PKT3_PREDICATE(predicate))
598
599static INLINE void r600_store_value(struct r600_command_buffer *cb, unsigned value)
600{
601	cb->buf[cb->atom.num_dw++] = value;
602}
603
604static INLINE void r600_store_config_reg_seq(struct r600_command_buffer *cb, unsigned reg, unsigned num)
605{
606	assert(reg < R600_CONTEXT_REG_OFFSET);
607	assert(cb->atom.num_dw+2+num <= cb->max_num_dw);
608	cb->buf[cb->atom.num_dw++] = PKT3(PKT3_SET_CONFIG_REG, num, 0);
609	cb->buf[cb->atom.num_dw++] = (reg - R600_CONFIG_REG_OFFSET) >> 2;
610}
611
612/**
613 * Needs cb->pkt_flags set to  RADEON_CP_PACKET3_COMPUTE_MODE for compute
614 * shaders.
615 */
616static INLINE void r600_store_context_reg_seq(struct r600_command_buffer *cb, unsigned reg, unsigned num)
617{
618	assert(reg >= R600_CONTEXT_REG_OFFSET && reg < R600_CTL_CONST_OFFSET);
619	assert(cb->atom.num_dw+2+num <= cb->max_num_dw);
620	cb->buf[cb->atom.num_dw++] = PKT3(PKT3_SET_CONTEXT_REG, num, 0) | cb->pkt_flags;
621	cb->buf[cb->atom.num_dw++] = (reg - R600_CONTEXT_REG_OFFSET) >> 2;
622}
623
624/**
625 * Needs cb->pkt_flags set to  RADEON_CP_PACKET3_COMPUTE_MODE for compute
626 * shaders.
627 */
628static INLINE void r600_store_ctl_const_seq(struct r600_command_buffer *cb, unsigned reg, unsigned num)
629{
630	assert(reg >= R600_CTL_CONST_OFFSET);
631	assert(cb->atom.num_dw+2+num <= cb->max_num_dw);
632	cb->buf[cb->atom.num_dw++] = PKT3(PKT3_SET_CTL_CONST, num, 0) | cb->pkt_flags;
633	cb->buf[cb->atom.num_dw++] = (reg - R600_CTL_CONST_OFFSET) >> 2;
634}
635
636static INLINE void r600_store_loop_const_seq(struct r600_command_buffer *cb, unsigned reg, unsigned num)
637{
638	assert(reg >= R600_LOOP_CONST_OFFSET);
639	assert(cb->atom.num_dw+2+num <= cb->max_num_dw);
640	cb->buf[cb->atom.num_dw++] = PKT3(PKT3_SET_LOOP_CONST, num, 0);
641	cb->buf[cb->atom.num_dw++] = (reg - R600_LOOP_CONST_OFFSET) >> 2;
642}
643
644/**
645 * Needs cb->pkt_flags set to  RADEON_CP_PACKET3_COMPUTE_MODE for compute
646 * shaders.
647 */
648static INLINE void eg_store_loop_const_seq(struct r600_command_buffer *cb, unsigned reg, unsigned num)
649{
650	assert(reg >= EG_LOOP_CONST_OFFSET);
651	assert(cb->atom.num_dw+2+num <= cb->max_num_dw);
652	cb->buf[cb->atom.num_dw++] = PKT3(PKT3_SET_LOOP_CONST, num, 0) | cb->pkt_flags;
653	cb->buf[cb->atom.num_dw++] = (reg - EG_LOOP_CONST_OFFSET) >> 2;
654}
655
656static INLINE void r600_store_config_reg(struct r600_command_buffer *cb, unsigned reg, unsigned value)
657{
658	r600_store_config_reg_seq(cb, reg, 1);
659	r600_store_value(cb, value);
660}
661
662static INLINE void r600_store_context_reg(struct r600_command_buffer *cb, unsigned reg, unsigned value)
663{
664	r600_store_context_reg_seq(cb, reg, 1);
665	r600_store_value(cb, value);
666}
667
668static INLINE void r600_store_ctl_const(struct r600_command_buffer *cb, unsigned reg, unsigned value)
669{
670	r600_store_ctl_const_seq(cb, reg, 1);
671	r600_store_value(cb, value);
672}
673
674static INLINE void r600_store_loop_const(struct r600_command_buffer *cb, unsigned reg, unsigned value)
675{
676	r600_store_loop_const_seq(cb, reg, 1);
677	r600_store_value(cb, value);
678}
679
680static INLINE void eg_store_loop_const(struct r600_command_buffer *cb, unsigned reg, unsigned value)
681{
682	eg_store_loop_const_seq(cb, reg, 1);
683	r600_store_value(cb, value);
684}
685
686void r600_init_command_buffer(struct r600_command_buffer *cb, unsigned num_dw, enum r600_atom_flags flags);
687void r600_release_command_buffer(struct r600_command_buffer *cb);
688
689/*
690 * Helpers for emitting state into a command stream directly.
691 */
692
693static INLINE unsigned r600_context_bo_reloc(struct r600_context *ctx, struct r600_resource *rbo,
694					     enum radeon_bo_usage usage)
695{
696	assert(usage);
697	return ctx->ws->cs_add_reloc(ctx->cs, rbo->cs_buf, usage, rbo->domains) * 4;
698}
699
700static INLINE void r600_write_value(struct radeon_winsys_cs *cs, unsigned value)
701{
702	cs->buf[cs->cdw++] = value;
703}
704
705static INLINE void r600_write_config_reg_seq(struct radeon_winsys_cs *cs, unsigned reg, unsigned num)
706{
707	assert(reg < R600_CONTEXT_REG_OFFSET);
708	assert(cs->cdw+2+num <= RADEON_MAX_CMDBUF_DWORDS);
709	cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONFIG_REG, num, 0);
710	cs->buf[cs->cdw++] = (reg - R600_CONFIG_REG_OFFSET) >> 2;
711}
712
713static INLINE void r600_write_context_reg_seq(struct radeon_winsys_cs *cs, unsigned reg, unsigned num)
714{
715	assert(reg >= R600_CONTEXT_REG_OFFSET && reg < R600_CTL_CONST_OFFSET);
716	assert(cs->cdw+2+num <= RADEON_MAX_CMDBUF_DWORDS);
717	cs->buf[cs->cdw++] = PKT3(PKT3_SET_CONTEXT_REG, num, 0);
718	cs->buf[cs->cdw++] = (reg - R600_CONTEXT_REG_OFFSET) >> 2;
719}
720
721static INLINE void r600_write_ctl_const_seq(struct radeon_winsys_cs *cs, unsigned reg, unsigned num)
722{
723	assert(reg >= R600_CTL_CONST_OFFSET);
724	assert(cs->cdw+2+num <= RADEON_MAX_CMDBUF_DWORDS);
725	cs->buf[cs->cdw++] = PKT3(PKT3_SET_CTL_CONST, num, 0);
726	cs->buf[cs->cdw++] = (reg - R600_CTL_CONST_OFFSET) >> 2;
727}
728
729static INLINE void r600_write_config_reg(struct radeon_winsys_cs *cs, unsigned reg, unsigned value)
730{
731	r600_write_config_reg_seq(cs, reg, 1);
732	r600_write_value(cs, value);
733}
734
735static INLINE void r600_write_context_reg(struct radeon_winsys_cs *cs, unsigned reg, unsigned value)
736{
737	r600_write_context_reg_seq(cs, reg, 1);
738	r600_write_value(cs, value);
739}
740
741static INLINE void r600_write_ctl_const(struct radeon_winsys_cs *cs, unsigned reg, unsigned value)
742{
743	r600_write_ctl_const_seq(cs, reg, 1);
744	r600_write_value(cs, value);
745}
746
747/*
748 * common helpers
749 */
750static INLINE uint32_t S_FIXED(float value, uint32_t frac_bits)
751{
752	return value * (1 << frac_bits);
753}
754#define ALIGN_DIVUP(x, y) (((x) + (y) - 1) / (y))
755
756static inline unsigned r600_tex_aniso_filter(unsigned filter)
757{
758	if (filter <= 1)   return 0;
759	if (filter <= 2)   return 1;
760	if (filter <= 4)   return 2;
761	if (filter <= 8)   return 3;
762	 /* else */        return 4;
763}
764
765/* 12.4 fixed-point */
766static INLINE unsigned r600_pack_float_12p4(float x)
767{
768	return x <= 0    ? 0 :
769	       x >= 4096 ? 0xffff : x * 16;
770}
771
772static INLINE uint64_t r600_resource_va(struct pipe_screen *screen, struct pipe_resource *resource)
773{
774	struct r600_screen *rscreen = (struct r600_screen*)screen;
775	struct r600_resource *rresource = (struct r600_resource*)resource;
776
777	return rscreen->ws->buffer_get_virtual_address(rresource->cs_buf);
778}
779
780#endif
781