2 * Copyright (C) 2012 Rob Clark <robclark@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 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * 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 NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
24 * Rob Clark <robclark@freedesktop.org>
27 #ifndef FREEDRENO_CONTEXT_H_
28 #define FREEDRENO_CONTEXT_H_
30 #include "pipe/p_context.h"
31 #include "indices/u_primconvert.h"
32 #include "util/u_blitter.h"
33 #include "util/list.h"
34 #include "util/slab.h"
35 #include "util/u_string.h"
37 #include "freedreno_batch.h"
38 #include "freedreno_screen.h"
39 #include "freedreno_gmem.h"
40 #include "freedreno_util.h"
42 #define BORDER_COLOR_UPLOAD_SIZE (2 * PIPE_MAX_SAMPLERS * BORDERCOLOR_SIZE)
44 struct fd_vertex_stateobj
;
46 struct fd_texture_stateobj
{
47 struct pipe_sampler_view
*textures
[PIPE_MAX_SAMPLERS
];
48 unsigned num_textures
;
49 unsigned valid_textures
;
50 struct pipe_sampler_state
*samplers
[PIPE_MAX_SAMPLERS
];
51 unsigned num_samplers
;
52 unsigned valid_samplers
;
53 /* number of samples per sampler, 2 bits per sampler: */
57 struct fd_program_stateobj
{
58 void *vs
, *hs
, *ds
, *gs
, *fs
;
61 struct fd_constbuf_stateobj
{
62 struct pipe_constant_buffer cb
[PIPE_MAX_CONSTANT_BUFFERS
];
63 uint32_t enabled_mask
;
66 struct fd_shaderbuf_stateobj
{
67 struct pipe_shader_buffer sb
[PIPE_MAX_SHADER_BUFFERS
];
68 uint32_t enabled_mask
;
71 struct fd_shaderimg_stateobj
{
72 struct pipe_image_view si
[PIPE_MAX_SHADER_IMAGES
];
73 uint32_t enabled_mask
;
76 struct fd_vertexbuf_stateobj
{
77 struct pipe_vertex_buffer vb
[PIPE_MAX_ATTRIBS
];
79 uint32_t enabled_mask
;
82 struct fd_vertex_stateobj
{
83 struct pipe_vertex_element pipe
[PIPE_MAX_ATTRIBS
];
84 unsigned num_elements
;
87 struct fd_streamout_stateobj
{
88 struct pipe_stream_output_target
*targets
[PIPE_MAX_SO_BUFFERS
];
89 /* Bitmask of stream that should be reset. */
93 /* Track offset from vtxcnt for streamout data. This counter
94 * is just incremented by # of vertices on each draw until
95 * reset or new streamout buffer bound.
97 * When we eventually have GS, the CPU won't actually know the
98 * number of vertices per draw, so I think we'll have to do
99 * something more clever.
101 unsigned offsets
[PIPE_MAX_SO_BUFFERS
];
104 #define MAX_GLOBAL_BUFFERS 16
105 struct fd_global_bindings_stateobj
{
106 struct pipe_resource
*buf
[MAX_GLOBAL_BUFFERS
];
107 uint32_t enabled_mask
;
110 /* group together the vertex and vertexbuf state.. for ease of passing
111 * around, and because various internal operations (gmem<->mem, etc)
112 * need their own vertex state:
114 struct fd_vertex_state
{
115 struct fd_vertex_stateobj
*vtx
;
116 struct fd_vertexbuf_stateobj vertexbuf
;
119 /* global 3d pipeline dirty state: */
120 enum fd_dirty_3d_state
{
121 FD_DIRTY_BLEND
= BIT(0),
122 FD_DIRTY_RASTERIZER
= BIT(1),
123 FD_DIRTY_ZSA
= BIT(2),
124 FD_DIRTY_BLEND_COLOR
= BIT(3),
125 FD_DIRTY_STENCIL_REF
= BIT(4),
126 FD_DIRTY_SAMPLE_MASK
= BIT(5),
127 FD_DIRTY_FRAMEBUFFER
= BIT(6),
128 FD_DIRTY_STIPPLE
= BIT(7),
129 FD_DIRTY_VIEWPORT
= BIT(8),
130 FD_DIRTY_VTXSTATE
= BIT(9),
131 FD_DIRTY_VTXBUF
= BIT(10),
132 FD_DIRTY_MIN_SAMPLES
= BIT(11),
134 FD_DIRTY_SCISSOR
= BIT(12),
135 FD_DIRTY_STREAMOUT
= BIT(13),
136 FD_DIRTY_UCP
= BIT(14),
137 FD_DIRTY_BLEND_DUAL
= BIT(15),
139 /* These are a bit redundent with fd_dirty_shader_state, and possibly
140 * should be removed. (But OTOH kinda convenient in some places)
142 FD_DIRTY_PROG
= BIT(16),
143 FD_DIRTY_CONST
= BIT(17),
144 FD_DIRTY_TEX
= BIT(18),
146 /* only used by a2xx.. possibly can be removed.. */
147 FD_DIRTY_TEXSTATE
= BIT(19),
150 /* per shader-stage dirty state: */
151 enum fd_dirty_shader_state
{
152 FD_DIRTY_SHADER_PROG
= BIT(0),
153 FD_DIRTY_SHADER_CONST
= BIT(1),
154 FD_DIRTY_SHADER_TEX
= BIT(2),
155 FD_DIRTY_SHADER_SSBO
= BIT(3),
156 FD_DIRTY_SHADER_IMAGE
= BIT(4),
160 struct pipe_context base
;
162 struct fd_device
*dev
;
163 struct fd_screen
*screen
;
164 struct fd_pipe
*pipe
;
166 struct util_queue flush_queue
;
168 struct blitter_context
*blitter
;
169 void *clear_rs_state
;
170 struct primconvert_context
*primconvert
;
172 /* slab for pipe_transfer allocations: */
173 struct slab_child_pool transfer_pool
;
176 * query related state:
179 /* slabs for fd_hw_sample and fd_hw_sample_period allocations: */
180 struct slab_mempool sample_pool
;
181 struct slab_mempool sample_period_pool
;
183 /* sample-providers for hw queries: */
184 const struct fd_hw_sample_provider
*hw_sample_providers
[MAX_HW_SAMPLE_PROVIDERS
];
186 /* list of active queries: */
187 struct list_head hw_active_queries
;
189 /* sample-providers for accumulating hw queries: */
190 const struct fd_acc_sample_provider
*acc_sample_providers
[MAX_HW_SAMPLE_PROVIDERS
];
192 /* list of active accumulating queries: */
193 struct list_head acc_active_queries
;
196 /* table with PIPE_PRIM_MAX entries mapping PIPE_PRIM_x to
197 * DI_PT_x value to use for draw initiator. There are some
198 * slight differences between generation:
200 const uint8_t *primtypes
;
201 uint32_t primtype_mask
;
203 /* shaders used by clear, and gmem->mem blits: */
204 struct fd_program_stateobj solid_prog
; // TODO move to screen?
206 /* shaders used by mem->gmem blits: */
207 struct fd_program_stateobj blit_prog
[MAX_RENDER_TARGETS
]; // TODO move to screen?
208 struct fd_program_stateobj blit_z
, blit_zs
;
213 uint64_t prims_emitted
;
214 uint64_t prims_generated
;
216 uint64_t batch_total
, batch_sysmem
, batch_gmem
, batch_nondraw
, batch_restore
;
217 uint64_t staging_uploads
, shadow_uploads
;
218 uint64_t vs_regs
, hs_regs
, ds_regs
, gs_regs
, fs_regs
;
221 /* Current batch.. the rule here is that you can deref ctx->batch
222 * in codepaths from pipe_context entrypoints. But not in code-
223 * paths from fd_batch_flush() (basically, the stuff that gets
224 * called from GMEM code), since in those code-paths the batch
225 * you care about is not necessarily the same as ctx->batch.
227 struct fd_batch
*batch
;
229 /* NULL if there has been rendering since last flush. Otherwise
230 * keeps a reference to the last fence so we can re-use it rather
231 * than having to flush no-op batch.
233 struct pipe_fence_handle
*last_fence
;
235 /* track last known reset status globally and per-context to
236 * determine if more resets occurred since then. If global reset
237 * count increases, it means some other context crashed. If
238 * per-context reset count increases, it means we crashed the
241 uint32_t context_reset_count
, global_reset_count
;
243 /* Are we in process of shadowing a resource? Used to detect recursion
244 * in transfer_map, and skip unneeded synchronization.
248 /* Ie. in blit situation where we no longer care about previous framebuffer
249 * contents. Main point is to eliminate blits from fd_try_shadow_resource().
250 * For example, in case of texture upload + gen-mipmaps.
254 struct pipe_scissor_state scissor
;
256 /* we don't have a disable/enable bit for scissor, so instead we keep
257 * a disabled-scissor state which matches the entire bound framebuffer
258 * and use that when scissor is not enabled.
260 struct pipe_scissor_state disabled_scissor
;
262 /* Current gmem/tiling configuration.. gets updated on render_tiles()
263 * if out of date with current maximal-scissor/cpp:
265 * (NOTE: this is kind of related to the batch, but moving it there
266 * means we'd always have to recalc tiles ever batch)
268 struct fd_gmem_stateobj gmem
;
270 /* Per vsc pipe bo's (a2xx-a5xx): */
271 struct fd_bo
*vsc_pipe_bo
[32];
273 /* which state objects need to be re-emit'd: */
274 enum fd_dirty_3d_state dirty
;
276 /* per shader-stage dirty status: */
277 enum fd_dirty_shader_state dirty_shader
[PIPE_SHADER_TYPES
];
280 struct pipe_blend_state
*blend
;
281 struct pipe_rasterizer_state
*rasterizer
;
282 struct pipe_depth_stencil_alpha_state
*zsa
;
284 struct fd_texture_stateobj tex
[PIPE_SHADER_TYPES
];
286 struct fd_program_stateobj prog
;
288 struct fd_vertex_state vtx
;
290 struct pipe_blend_color blend_color
;
291 struct pipe_stencil_ref stencil_ref
;
292 unsigned sample_mask
;
293 unsigned min_samples
;
294 /* local context fb state, for when ctx->batch is null: */
295 struct pipe_framebuffer_state framebuffer
;
296 struct pipe_poly_stipple stipple
;
297 struct pipe_viewport_state viewport
;
298 struct pipe_scissor_state viewport_scissor
;
299 struct fd_constbuf_stateobj constbuf
[PIPE_SHADER_TYPES
];
300 struct fd_shaderbuf_stateobj shaderbuf
[PIPE_SHADER_TYPES
];
301 struct fd_shaderimg_stateobj shaderimg
[PIPE_SHADER_TYPES
];
302 struct fd_streamout_stateobj streamout
;
303 struct fd_global_bindings_stateobj global_bindings
;
304 struct pipe_clip_state ucp
;
306 struct pipe_query
*cond_query
;
307 bool cond_cond
; /* inverted rendering condition */
310 struct pipe_debug_callback debug
;
312 /* GMEM/tile handling fxns: */
313 void (*emit_tile_init
)(struct fd_batch
*batch
);
314 void (*emit_tile_prep
)(struct fd_batch
*batch
, const struct fd_tile
*tile
);
315 void (*emit_tile_mem2gmem
)(struct fd_batch
*batch
, const struct fd_tile
*tile
);
316 void (*emit_tile_renderprep
)(struct fd_batch
*batch
, const struct fd_tile
*tile
);
317 void (*emit_tile
)(struct fd_batch
*batch
, const struct fd_tile
*tile
);
318 void (*emit_tile_gmem2mem
)(struct fd_batch
*batch
, const struct fd_tile
*tile
);
319 void (*emit_tile_fini
)(struct fd_batch
*batch
); /* optional */
321 /* optional, for GMEM bypass: */
322 void (*emit_sysmem_prep
)(struct fd_batch
*batch
);
323 void (*emit_sysmem_fini
)(struct fd_batch
*batch
);
326 bool (*draw_vbo
)(struct fd_context
*ctx
, const struct pipe_draw_info
*info
,
327 unsigned index_offset
);
328 bool (*clear
)(struct fd_context
*ctx
, unsigned buffers
,
329 const union pipe_color_union
*color
, double depth
, unsigned stencil
);
332 void (*launch_grid
)(struct fd_context
*ctx
, const struct pipe_grid_info
*info
);
335 struct fd_query
* (*create_query
)(struct fd_context
*ctx
, unsigned query_type
, unsigned index
);
336 void (*query_prepare
)(struct fd_batch
*batch
, uint32_t num_tiles
);
337 void (*query_prepare_tile
)(struct fd_batch
*batch
, uint32_t n
,
338 struct fd_ringbuffer
*ring
);
339 void (*query_set_stage
)(struct fd_batch
*batch
, enum fd_render_stage stage
);
342 bool (*blit
)(struct fd_context
*ctx
, const struct pipe_blit_info
*info
);
344 /* handling for barriers: */
345 void (*framebuffer_barrier
)(struct fd_context
*ctx
);
348 * Common pre-cooked VBO state (used for a3xx and later):
351 /* for clear/gmem->mem vertices, and mem->gmem */
352 struct pipe_resource
*solid_vbuf
;
354 /* for mem->gmem tex coords: */
355 struct pipe_resource
*blit_texcoord_vbuf
;
357 /* vertex state for solid_vbuf:
358 * - solid_vbuf / 12 / R32G32B32_FLOAT
360 struct fd_vertex_state solid_vbuf_state
;
362 /* vertex state for blit_prog:
363 * - blit_texcoord_vbuf / 8 / R32G32_FLOAT
364 * - solid_vbuf / 12 / R32G32B32_FLOAT
366 struct fd_vertex_state blit_vbuf_state
;
369 * Info about state of previous draw, for state that comes from
370 * pipe_draw_info (ie. not part of a CSO). This allows us to
371 * skip some register emit when the state doesn't change from
375 bool dirty
; /* last draw state unknown */
376 bool primitive_restart
;
377 uint32_t index_start
;
378 uint32_t instance_start
;
379 uint32_t restart_index
;
383 static inline struct fd_context
*
384 fd_context(struct pipe_context
*pctx
)
386 return (struct fd_context
*)pctx
;
390 fd_context_assert_locked(struct fd_context
*ctx
)
392 pipe_mutex_assert_locked(ctx
->screen
->lock
);
396 fd_context_lock(struct fd_context
*ctx
)
398 mtx_lock(&ctx
->screen
->lock
);
402 fd_context_unlock(struct fd_context
*ctx
)
404 mtx_unlock(&ctx
->screen
->lock
);
407 /* mark all state dirty: */
409 fd_context_all_dirty(struct fd_context
*ctx
)
411 ctx
->last
.dirty
= true;
413 for (unsigned i
= 0; i
< PIPE_SHADER_TYPES
; i
++)
414 ctx
->dirty_shader
[i
] = ~0;
418 fd_context_all_clean(struct fd_context
*ctx
)
421 for (unsigned i
= 0; i
< PIPE_SHADER_TYPES
; i
++) {
422 /* don't mark compute state as clean, since it is not emitted
423 * during normal draw call. The places that call _all_dirty(),
424 * it is safe to mark compute state dirty as well, but the
425 * inverse is not true.
427 if (i
== PIPE_SHADER_COMPUTE
)
429 ctx
->dirty_shader
[i
] = 0;
433 static inline struct pipe_scissor_state
*
434 fd_context_get_scissor(struct fd_context
*ctx
)
436 if (ctx
->rasterizer
&& ctx
->rasterizer
->scissor
)
437 return &ctx
->scissor
;
438 return &ctx
->disabled_scissor
;
442 fd_supported_prim(struct fd_context
*ctx
, unsigned prim
)
444 return (1 << prim
) & ctx
->primtype_mask
;
447 static inline struct fd_batch
*
448 fd_context_batch(struct fd_context
*ctx
)
450 if (unlikely(!ctx
->batch
)) {
451 struct fd_batch
*batch
=
452 fd_batch_from_fb(&ctx
->screen
->batch_cache
, ctx
, &ctx
->framebuffer
);
453 util_copy_framebuffer_state(&batch
->framebuffer
, &ctx
->framebuffer
);
455 fd_context_all_dirty(ctx
);
461 fd_batch_set_stage(struct fd_batch
*batch
, enum fd_render_stage stage
)
463 struct fd_context
*ctx
= batch
->ctx
;
465 /* special case: internal blits (like mipmap level generation)
466 * go through normal draw path (via util_blitter_blit()).. but
467 * we need to ignore the FD_STAGE_DRAW which will be set, so we
468 * don't enable queries which should be paused during internal
471 if ((batch
->stage
== FD_STAGE_BLIT
) &&
472 (stage
!= FD_STAGE_NULL
))
475 if (ctx
->query_set_stage
)
476 ctx
->query_set_stage(batch
, stage
);
478 batch
->stage
= stage
;
481 void fd_context_setup_common_vbos(struct fd_context
*ctx
);
482 void fd_context_cleanup_common_vbos(struct fd_context
*ctx
);
484 struct pipe_context
* fd_context_init(struct fd_context
*ctx
,
485 struct pipe_screen
*pscreen
, const uint8_t *primtypes
,
486 void *priv
, unsigned flags
);
488 void fd_context_destroy(struct pipe_context
*pctx
);
490 #endif /* FREEDRENO_CONTEXT_H_ */