2 * Copyright © 2011 Intel Corporation
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
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
26 #include "intel_batchbuffer.h"
27 #include "intel_mipmap_tree.h"
28 #include "intel_fbo.h"
30 #include "brw_context.h"
31 #include "brw_state.h"
33 #include "blorp/blorp_genX_exec.h"
35 #include "brw_blorp.h"
38 blorp_emit_dwords(struct blorp_batch
*batch
, unsigned n
)
40 assert(batch
->blorp
->driver_ctx
== batch
->driver_batch
);
41 struct brw_context
*brw
= batch
->driver_batch
;
43 intel_batchbuffer_begin(brw
, n
, RENDER_RING
);
44 uint32_t *map
= brw
->batch
.map_next
;
45 brw
->batch
.map_next
+= n
;
46 intel_batchbuffer_advance(brw
);
51 blorp_emit_reloc(struct blorp_batch
*batch
,
52 void *location
, struct blorp_address address
, uint32_t delta
)
54 assert(batch
->blorp
->driver_ctx
== batch
->driver_batch
);
55 struct brw_context
*brw
= batch
->driver_batch
;
57 uint32_t offset
= (char *)location
- (char *)brw
->batch
.map
;
59 return intel_batchbuffer_reloc64(&brw
->batch
, address
.buffer
, offset
,
62 address
.offset
+ delta
);
64 return intel_batchbuffer_reloc(&brw
->batch
, address
.buffer
, offset
,
67 address
.offset
+ delta
);
72 blorp_surface_reloc(struct blorp_batch
*batch
, uint32_t ss_offset
,
73 struct blorp_address address
, uint32_t delta
)
75 assert(batch
->blorp
->driver_ctx
== batch
->driver_batch
);
76 struct brw_context
*brw
= batch
->driver_batch
;
77 drm_intel_bo
*bo
= address
.buffer
;
79 drm_intel_bo_emit_reloc(brw
->batch
.bo
, ss_offset
,
80 bo
, address
.offset
+ delta
,
81 address
.read_domains
, address
.write_domain
);
83 uint64_t reloc_val
= bo
->offset64
+ address
.offset
+ delta
;
84 void *reloc_ptr
= (void *)brw
->batch
.map
+ ss_offset
;
86 *(uint64_t *)reloc_ptr
= reloc_val
;
88 *(uint32_t *)reloc_ptr
= reloc_val
;
93 blorp_alloc_dynamic_state(struct blorp_batch
*batch
,
94 enum aub_state_struct_type type
,
99 assert(batch
->blorp
->driver_ctx
== batch
->driver_batch
);
100 struct brw_context
*brw
= batch
->driver_batch
;
102 return brw_state_batch(brw
, type
, size
, alignment
, offset
);
106 blorp_alloc_binding_table(struct blorp_batch
*batch
, unsigned num_entries
,
107 unsigned state_size
, unsigned state_alignment
,
108 uint32_t *bt_offset
, uint32_t *surface_offsets
,
111 assert(batch
->blorp
->driver_ctx
== batch
->driver_batch
);
112 struct brw_context
*brw
= batch
->driver_batch
;
114 uint32_t *bt_map
= brw_state_batch(brw
, AUB_TRACE_BINDING_TABLE
,
115 num_entries
* sizeof(uint32_t), 32,
118 for (unsigned i
= 0; i
< num_entries
; i
++) {
119 surface_maps
[i
] = brw_state_batch(brw
, AUB_TRACE_SURFACE_STATE
,
120 state_size
, state_alignment
,
121 &(surface_offsets
)[i
]);
122 bt_map
[i
] = surface_offsets
[i
];
127 blorp_alloc_vertex_buffer(struct blorp_batch
*batch
, uint32_t size
,
128 struct blorp_address
*addr
)
130 assert(batch
->blorp
->driver_ctx
== batch
->driver_batch
);
131 struct brw_context
*brw
= batch
->driver_batch
;
134 void *data
= brw_state_batch(brw
, AUB_TRACE_VERTEX_BUFFER
,
137 *addr
= (struct blorp_address
) {
138 .buffer
= brw
->batch
.bo
,
139 .read_domains
= I915_GEM_DOMAIN_VERTEX
,
148 blorp_flush_range(struct blorp_batch
*batch
, void *start
, size_t size
)
150 /* All allocated states come from the batch which we will flush before we
151 * submit it. There's nothing for us to do here.
156 blorp_emit_urb_config(struct blorp_batch
*batch
, unsigned vs_entry_size
)
158 assert(batch
->blorp
->driver_ctx
== batch
->driver_batch
);
159 struct brw_context
*brw
= batch
->driver_batch
;
162 if (!(brw
->ctx
.NewDriverState
& (BRW_NEW_CONTEXT
| BRW_NEW_URB_SIZE
)) &&
163 brw
->urb
.vsize
>= vs_entry_size
)
166 brw
->ctx
.NewDriverState
|= BRW_NEW_URB_SIZE
;
168 gen7_upload_urb(brw
, vs_entry_size
, false, false);
170 gen6_upload_urb(brw
, vs_entry_size
, false, 0);
175 genX(blorp_exec
)(struct blorp_batch
*batch
,
176 const struct blorp_params
*params
)
178 assert(batch
->blorp
->driver_ctx
== batch
->driver_batch
);
179 struct brw_context
*brw
= batch
->driver_batch
;
180 struct gl_context
*ctx
= &brw
->ctx
;
181 const uint32_t estimated_max_batch_usage
= GEN_GEN
>= 8 ? 1800 : 1500;
182 bool check_aperture_failed_once
= false;
184 /* Flush the sampler and render caches. We definitely need to flush the
185 * sampler cache so that we get updated contents from the render cache for
186 * the glBlitFramebuffer() source. Also, we are sometimes warned in the
187 * docs to flush the cache between reinterpretations of the same surface
188 * data with different formats, which blorp does for stencil and depth
191 if (params
->src
.enabled
)
192 brw_render_cache_set_check_flush(brw
, params
->src
.addr
.buffer
);
193 brw_render_cache_set_check_flush(brw
, params
->dst
.addr
.buffer
);
195 brw_select_pipeline(brw
, BRW_RENDER_PIPELINE
);
198 intel_batchbuffer_require_space(brw
, estimated_max_batch_usage
, RENDER_RING
);
199 intel_batchbuffer_save_state(brw
);
200 drm_intel_bo
*saved_bo
= brw
->batch
.bo
;
201 uint32_t saved_used
= USED_BATCH(brw
->batch
);
202 uint32_t saved_state_batch_offset
= brw
->batch
.state_batch_offset
;
205 /* Emit workaround flushes when we switch from drawing to blorping. */
206 brw_emit_post_sync_nonzero_flush(brw
);
209 brw_upload_state_base_address(brw
);
212 gen7_l3_state
.emit(brw
);
215 brw_emit_depth_stall_flushes(brw
);
218 gen8_write_pma_stall_bits(brw
, 0);
221 blorp_emit(batch
, GENX(3DSTATE_DRAWING_RECTANGLE
), rect
) {
222 rect
.ClippedDrawingRectangleXMax
= MAX2(params
->x1
, params
->x0
) - 1;
223 rect
.ClippedDrawingRectangleYMax
= MAX2(params
->y1
, params
->y0
) - 1;
226 blorp_exec(batch
, params
);
228 /* Make sure we didn't wrap the batch unintentionally, and make sure we
229 * reserved enough space that a wrap will never happen.
231 assert(brw
->batch
.bo
== saved_bo
);
232 assert((USED_BATCH(brw
->batch
) - saved_used
) * 4 +
233 (saved_state_batch_offset
- brw
->batch
.state_batch_offset
) <
234 estimated_max_batch_usage
);
235 /* Shut up compiler warnings on release build */
238 (void)saved_state_batch_offset
;
240 /* Check if the blorp op we just did would make our batch likely to fail to
241 * map all the BOs into the GPU at batch exec time later. If so, flush the
242 * batch and try again with nothing else in the batch.
244 if (dri_bufmgr_check_aperture_space(&brw
->batch
.bo
, 1)) {
245 if (!check_aperture_failed_once
) {
246 check_aperture_failed_once
= true;
247 intel_batchbuffer_reset_to_saved(brw
);
248 intel_batchbuffer_flush(brw
);
251 int ret
= intel_batchbuffer_flush(brw
);
252 WARN_ONCE(ret
== -ENOSPC
,
253 "i965: blorp emit exceeded available aperture space\n");
257 if (unlikely(brw
->always_flush_batch
))
258 intel_batchbuffer_flush(brw
);
260 /* We've smashed all state compared to what the normal 3D pipeline
261 * rendering tracks for GL.
263 brw
->ctx
.NewDriverState
|= BRW_NEW_BLORP
;
264 brw
->no_depth_or_stencil
= false;
267 if (params
->dst
.enabled
)
268 brw_render_cache_set_add_bo(brw
, params
->dst
.addr
.buffer
);
269 if (params
->depth
.enabled
)
270 brw_render_cache_set_add_bo(brw
, params
->depth
.addr
.buffer
);
271 if (params
->stencil
.enabled
)
272 brw_render_cache_set_add_bo(brw
, params
->stencil
.addr
.buffer
);