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.
30 void si_need_cs_space(struct si_context
*ctx
, unsigned num_dw
,
31 boolean count_draw_in
)
33 struct radeon_winsys_cs
*cs
= ctx
->b
.rings
.gfx
.cs
;
35 /* There are two memory usage counters in the winsys for all buffers
36 * that have been added (cs_add_reloc) and two counters in the pipe
37 * driver for those that haven't been added yet.
39 if (!ctx
->b
.ws
->cs_memory_below_limit(ctx
->b
.rings
.gfx
.cs
, ctx
->b
.vram
, ctx
->b
.gtt
)) {
42 ctx
->b
.rings
.gfx
.flush(ctx
, RADEON_FLUSH_ASYNC
, NULL
);
48 /* If the CS is sufficiently large, don't count the space needed
49 * and just flush if there is less than 8096 dwords left.
51 if (cs
->max_dw
>= 24 * 1024) {
52 if (cs
->cdw
> cs
->max_dw
- 8 * 1024)
53 ctx
->b
.rings
.gfx
.flush(ctx
, RADEON_FLUSH_ASYNC
, NULL
);
57 /* The number of dwords we already used in the CS so far. */
61 unsigned mask
= ctx
->dirty_atoms
;
64 num_dw
+= ctx
->atoms
.array
[u_bit_scan(&mask
)]->num_dw
;
66 /* The number of dwords all the dirty states would take. */
67 num_dw
+= si_pm4_dirty_dw(ctx
);
69 /* The upper-bound of how much a draw command would take. */
70 num_dw
+= SI_MAX_DRAW_CS_DWORDS
;
73 /* Count in queries_suspend. */
74 num_dw
+= ctx
->b
.num_cs_dw_nontimer_queries_suspend
+
75 ctx
->b
.num_cs_dw_timer_queries_suspend
;
77 /* Count in streamout_end at the end of CS. */
78 if (ctx
->b
.streamout
.begin_emitted
) {
79 num_dw
+= ctx
->b
.streamout
.num_dw_for_end
;
82 /* Count in render_condition(NULL) at the end of CS. */
83 if (ctx
->b
.predicate_drawing
) {
87 /* Count in framebuffer cache flushes at the end of CS. */
88 num_dw
+= ctx
->atoms
.s
.cache_flush
->num_dw
;
90 if (ctx
->screen
->b
.trace_bo
)
91 num_dw
+= SI_TRACE_CS_DWORDS
* 2;
93 /* Flush if there's not enough space. */
94 if (num_dw
> cs
->max_dw
) {
95 ctx
->b
.rings
.gfx
.flush(ctx
, RADEON_FLUSH_ASYNC
, NULL
);
99 void si_context_gfx_flush(void *context
, unsigned flags
,
100 struct pipe_fence_handle
**fence
)
102 struct si_context
*ctx
= context
;
103 struct radeon_winsys_cs
*cs
= ctx
->b
.rings
.gfx
.cs
;
104 struct radeon_winsys
*ws
= ctx
->b
.ws
;
106 if (cs
->cdw
== ctx
->b
.initial_gfx_cs_size
&&
107 (!fence
|| ctx
->last_gfx_fence
)) {
109 ws
->fence_reference(fence
, ctx
->last_gfx_fence
);
110 if (!(flags
& RADEON_FLUSH_ASYNC
))
111 ws
->cs_sync_flush(cs
);
115 ctx
->b
.rings
.gfx
.flushing
= true;
117 r600_preflush_suspend_features(&ctx
->b
);
119 ctx
->b
.flags
|= SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER
|
120 SI_CONTEXT_INV_TC_L1
|
121 SI_CONTEXT_INV_TC_L2
|
122 /* this is probably not needed anymore */
123 SI_CONTEXT_PS_PARTIAL_FLUSH
;
124 si_emit_cache_flush(ctx
, NULL
);
126 /* force to keep tiling flags */
127 flags
|= RADEON_FLUSH_KEEP_TILING_FLAGS
;
132 /* Save the IB for debug contexts. */
135 ctx
->last_ib_dw_size
= cs
->cdw
;
136 ctx
->last_ib
= malloc(cs
->cdw
* 4);
137 memcpy(ctx
->last_ib
, cs
->buf
, cs
->cdw
* 4);
138 r600_resource_reference(&ctx
->last_trace_buf
, ctx
->trace_buf
);
139 r600_resource_reference(&ctx
->trace_buf
, NULL
);
143 ws
->cs_flush(cs
, flags
, &ctx
->last_gfx_fence
,
144 ctx
->screen
->b
.cs_count
++);
145 ctx
->b
.rings
.gfx
.flushing
= false;
148 ws
->fence_reference(fence
, ctx
->last_gfx_fence
);
150 si_begin_new_cs(ctx
);
153 void si_begin_new_cs(struct si_context
*ctx
)
158 /* Create a buffer used for writing trace IDs and initialize it to 0. */
159 assert(!ctx
->trace_buf
);
160 ctx
->trace_buf
= (struct r600_resource
*)
161 pipe_buffer_create(ctx
->b
.b
.screen
, PIPE_BIND_CUSTOM
,
162 PIPE_USAGE_STAGING
, 4);
164 pipe_buffer_write_nooverlap(&ctx
->b
.b
, &ctx
->trace_buf
->b
.b
,
165 0, sizeof(zero
), &zero
);
172 /* Flush read caches at the beginning of CS. */
173 ctx
->b
.flags
|= SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER
|
174 SI_CONTEXT_INV_TC_L1
|
175 SI_CONTEXT_INV_TC_L2
|
176 SI_CONTEXT_INV_KCACHE
|
177 SI_CONTEXT_INV_ICACHE
;
179 /* set all valid group as dirty so they get reemited on
182 si_pm4_reset_emitted(ctx
);
184 /* The CS initialization should be emitted before everything else. */
185 si_pm4_emit(ctx
, ctx
->init_config
);
187 si_mark_atom_dirty(ctx
, &ctx
->clip_regs
);
188 si_mark_atom_dirty(ctx
, &ctx
->framebuffer
.atom
);
189 si_mark_atom_dirty(ctx
, &ctx
->msaa_sample_locs
);
190 si_mark_atom_dirty(ctx
, &ctx
->msaa_config
);
191 si_mark_atom_dirty(ctx
, &ctx
->db_render_state
);
192 si_mark_atom_dirty(ctx
, &ctx
->b
.streamout
.enable_atom
);
193 si_all_descriptors_begin_new_cs(ctx
);
195 ctx
->scissors
.dirty_mask
= (1 << SI_MAX_VIEWPORTS
) - 1;
196 ctx
->viewports
.dirty_mask
= (1 << SI_MAX_VIEWPORTS
) - 1;
197 si_mark_atom_dirty(ctx
, &ctx
->scissors
.atom
);
198 si_mark_atom_dirty(ctx
, &ctx
->viewports
.atom
);
200 r600_postflush_resume_features(&ctx
->b
);
202 ctx
->b
.initial_gfx_cs_size
= ctx
->b
.rings
.gfx
.cs
->cdw
;
204 /* Invalidate various draw states so that they are emitted before
205 * the first draw call. */
206 si_invalidate_draw_sh_constants(ctx
);
207 ctx
->last_primitive_restart_en
= -1;
208 ctx
->last_restart_index
= SI_RESTART_INDEX_UNKNOWN
;
209 ctx
->last_gs_out_prim
= -1;
211 ctx
->last_multi_vgt_param
= -1;
212 ctx
->last_ls_hs_config
= -1;
213 ctx
->last_rast_prim
= -1;
214 ctx
->last_sc_line_stipple
= ~0;
215 ctx
->emit_scratch_reloc
= true;
217 ctx
->last_tcs
= NULL
;
218 ctx
->last_tes_sh_base
= -1;
219 ctx
->last_num_tcs_input_cp
= -1;