f7638bcb3d823d75f2b8270a7a11306ecb50f84f
1 #ifndef INTEL_BATCHBUFFER_H
2 #define INTEL_BATCHBUFFER_H
4 #include "main/mtypes.h"
6 #include "brw_context.h"
7 #include "intel_bufmgr.h"
15 * Number of bytes to reserve for commands necessary to complete a batch.
18 * - MI_BATCHBUFFER_END (4 bytes)
19 * - Optional MI_NOOP for ensuring the batch length is qword aligned (4 bytes)
20 * - Any state emitted by vtbl->finish_batch():
21 * - Gen4-5 record ending occlusion query values (4 * 4 = 16 bytes)
23 #define BATCH_RESERVED 24
25 struct intel_batchbuffer
;
27 void intel_batchbuffer_init(struct brw_context
*brw
);
28 void intel_batchbuffer_free(struct brw_context
*brw
);
29 void intel_batchbuffer_save_state(struct brw_context
*brw
);
30 void intel_batchbuffer_reset_to_saved(struct brw_context
*brw
);
31 void intel_batchbuffer_clear_cache(struct brw_context
*brw
);
33 int _intel_batchbuffer_flush(struct brw_context
*brw
,
34 const char *file
, int line
);
36 #define intel_batchbuffer_flush(intel) \
37 _intel_batchbuffer_flush(intel, __FILE__, __LINE__)
41 /* Unlike bmBufferData, this currently requires the buffer be mapped.
42 * Consider it a convenience function wrapping multple
43 * intel_buffer_dword() calls.
45 void intel_batchbuffer_data(struct brw_context
*brw
,
46 const void *data
, GLuint bytes
,
47 enum brw_gpu_ring ring
);
49 bool intel_batchbuffer_emit_reloc(struct brw_context
*brw
,
51 uint32_t read_domains
,
52 uint32_t write_domain
,
54 bool intel_batchbuffer_emit_reloc_fenced(struct brw_context
*brw
,
56 uint32_t read_domains
,
57 uint32_t write_domain
,
59 void intel_batchbuffer_emit_mi_flush(struct brw_context
*brw
);
60 void intel_emit_post_sync_nonzero_flush(struct brw_context
*brw
);
61 void intel_emit_depth_stall_flushes(struct brw_context
*brw
);
62 void gen7_emit_vs_workaround_flush(struct brw_context
*brw
);
63 void gen7_emit_cs_stall_flush(struct brw_context
*brw
);
65 static INLINE
uint32_t float_as_int(float f
)
76 /* Inline functions - might actually be better off with these
77 * non-inlined. Certainly better off switching all command packets to
78 * be passed as structs rather than dwords, but that's a little bit of
81 static INLINE
unsigned
82 intel_batchbuffer_space(struct brw_context
*brw
)
84 return (brw
->batch
.state_batch_offset
- brw
->batch
.reserved_space
)
90 intel_batchbuffer_emit_dword(struct brw_context
*brw
, GLuint dword
)
93 assert(intel_batchbuffer_space(brw
) >= 4);
95 brw
->batch
.map
[brw
->batch
.used
++] = dword
;
96 assert(brw
->batch
.ring
!= UNKNOWN_RING
);
100 intel_batchbuffer_emit_float(struct brw_context
*brw
, float f
)
102 intel_batchbuffer_emit_dword(brw
, float_as_int(f
));
106 intel_batchbuffer_require_space(struct brw_context
*brw
, GLuint sz
,
107 enum brw_gpu_ring ring
)
109 /* If we're switching rings, implicitly flush the batch. */
110 if (unlikely(ring
!= brw
->batch
.ring
) && brw
->batch
.ring
!= UNKNOWN_RING
&&
112 intel_batchbuffer_flush(brw
);
116 assert(sz
< BATCH_SZ
- BATCH_RESERVED
);
118 if (intel_batchbuffer_space(brw
) < sz
)
119 intel_batchbuffer_flush(brw
);
121 /* The intel_batchbuffer_flush() calls above might have changed
122 * brw->batch.ring to UNKNOWN_RING, so we need to set it here at the end.
124 brw
->batch
.ring
= ring
;
128 intel_batchbuffer_begin(struct brw_context
*brw
, int n
, enum brw_gpu_ring ring
)
130 intel_batchbuffer_require_space(brw
, n
* 4, ring
);
132 brw
->batch
.emit
= brw
->batch
.used
;
134 brw
->batch
.total
= n
;
139 intel_batchbuffer_advance(struct brw_context
*brw
)
142 struct intel_batchbuffer
*batch
= &brw
->batch
;
143 unsigned int _n
= batch
->used
- batch
->emit
;
144 assert(batch
->total
!= 0);
145 if (_n
!= batch
->total
) {
146 fprintf(stderr
, "ADVANCE_BATCH: %d of %d dwords emitted\n",
154 void intel_batchbuffer_cached_advance(struct brw_context
*brw
);
156 #define BEGIN_BATCH(n) intel_batchbuffer_begin(brw, n, RENDER_RING)
157 #define BEGIN_BATCH_BLT(n) intel_batchbuffer_begin(brw, n, BLT_RING)
158 #define OUT_BATCH(d) intel_batchbuffer_emit_dword(brw, d)
159 #define OUT_BATCH_F(f) intel_batchbuffer_emit_float(brw, f)
160 #define OUT_RELOC(buf, read_domains, write_domain, delta) do { \
161 intel_batchbuffer_emit_reloc(brw, buf, \
162 read_domains, write_domain, delta); \
164 #define OUT_RELOC_FENCED(buf, read_domains, write_domain, delta) do { \
165 intel_batchbuffer_emit_reloc_fenced(brw, buf, \
166 read_domains, write_domain, delta); \
169 #define ADVANCE_BATCH() intel_batchbuffer_advance(brw);
170 #define CACHED_BATCH() intel_batchbuffer_cached_advance(brw);