edcd89640980237802aabeb8d230eac8f96b66be
[mesa.git] / src / mesa / drivers / dri / i965 / genX_blorp_exec.c
1 /*
2 * Copyright © 2011 Intel Corporation
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 * 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:
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 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
21 * IN THE SOFTWARE.
22 */
23
24 #include <assert.h>
25
26 #include "intel_batchbuffer.h"
27 #include "intel_mipmap_tree.h"
28
29 #include "brw_context.h"
30 #include "brw_state.h"
31
32 #include "blorp/blorp_genX_exec.h"
33
34 #include "brw_blorp.h"
35
36 static void *
37 blorp_emit_dwords(struct blorp_batch *batch, unsigned n)
38 {
39 assert(batch->blorp->driver_ctx == batch->driver_batch);
40 struct brw_context *brw = batch->driver_batch;
41
42 intel_batchbuffer_begin(brw, n, RENDER_RING);
43 uint32_t *map = brw->batch.map_next;
44 brw->batch.map_next += n;
45 intel_batchbuffer_advance(brw);
46 return map;
47 }
48
49 static uint64_t
50 blorp_emit_reloc(struct blorp_batch *batch,
51 void *location, struct blorp_address address, uint32_t delta)
52 {
53 assert(batch->blorp->driver_ctx == batch->driver_batch);
54 struct brw_context *brw = batch->driver_batch;
55
56 uint32_t offset = (char *)location - (char *)brw->batch.map;
57 if (brw->gen >= 8) {
58 return intel_batchbuffer_reloc64(brw, address.buffer, offset,
59 address.read_domains,
60 address.write_domain,
61 address.offset + delta);
62 } else {
63 return intel_batchbuffer_reloc(brw, address.buffer, offset,
64 address.read_domains,
65 address.write_domain,
66 address.offset + delta);
67 }
68 }
69
70 static void
71 blorp_surface_reloc(struct blorp_batch *batch, uint32_t ss_offset,
72 struct blorp_address address, uint32_t delta)
73 {
74 assert(batch->blorp->driver_ctx == batch->driver_batch);
75 struct brw_context *brw = batch->driver_batch;
76 drm_intel_bo *bo = address.buffer;
77
78 drm_intel_bo_emit_reloc(brw->batch.bo, ss_offset,
79 bo, address.offset + delta,
80 address.read_domains, address.write_domain);
81
82 uint64_t reloc_val = bo->offset64 + address.offset + delta;
83 void *reloc_ptr = (void *)brw->batch.map + ss_offset;
84 #if GEN_GEN >= 8
85 *(uint64_t *)reloc_ptr = reloc_val;
86 #else
87 *(uint32_t *)reloc_ptr = reloc_val;
88 #endif
89 }
90
91 static void *
92 blorp_alloc_dynamic_state(struct blorp_batch *batch,
93 enum aub_state_struct_type type,
94 uint32_t size,
95 uint32_t alignment,
96 uint32_t *offset)
97 {
98 assert(batch->blorp->driver_ctx == batch->driver_batch);
99 struct brw_context *brw = batch->driver_batch;
100
101 return brw_state_batch(brw, type, size, alignment, offset);
102 }
103
104 static void
105 blorp_alloc_binding_table(struct blorp_batch *batch, unsigned num_entries,
106 unsigned state_size, unsigned state_alignment,
107 uint32_t *bt_offset, uint32_t *surface_offsets,
108 void **surface_maps)
109 {
110 assert(batch->blorp->driver_ctx == batch->driver_batch);
111 struct brw_context *brw = batch->driver_batch;
112
113 uint32_t *bt_map = brw_state_batch(brw, AUB_TRACE_BINDING_TABLE,
114 num_entries * sizeof(uint32_t), 32,
115 bt_offset);
116
117 for (unsigned i = 0; i < num_entries; i++) {
118 surface_maps[i] = brw_state_batch(brw, AUB_TRACE_SURFACE_STATE,
119 state_size, state_alignment,
120 &(surface_offsets)[i]);
121 bt_map[i] = surface_offsets[i];
122 }
123 }
124
125 static void *
126 blorp_alloc_vertex_buffer(struct blorp_batch *batch, uint32_t size,
127 struct blorp_address *addr)
128 {
129 assert(batch->blorp->driver_ctx == batch->driver_batch);
130 struct brw_context *brw = batch->driver_batch;
131
132 uint32_t offset;
133 void *data = brw_state_batch(brw, AUB_TRACE_VERTEX_BUFFER,
134 size, 32, &offset);
135
136 *addr = (struct blorp_address) {
137 .buffer = brw->batch.bo,
138 .read_domains = I915_GEM_DOMAIN_VERTEX,
139 .write_domain = 0,
140 .offset = offset,
141 };
142
143 return data;
144 }
145
146 static void
147 blorp_emit_urb_config(struct blorp_batch *batch, unsigned vs_entry_size)
148 {
149 assert(batch->blorp->driver_ctx == batch->driver_batch);
150 struct brw_context *brw = batch->driver_batch;
151
152 #if GEN_GEN >= 7
153 if (!(brw->ctx.NewDriverState & (BRW_NEW_CONTEXT | BRW_NEW_URB_SIZE)) &&
154 brw->urb.vsize >= vs_entry_size)
155 return;
156
157 brw->ctx.NewDriverState |= BRW_NEW_URB_SIZE;
158
159 gen7_upload_urb(brw, vs_entry_size, false, false);
160 #else
161 gen6_upload_urb(brw, vs_entry_size, false, 0);
162 #endif
163 }
164
165 void
166 genX(blorp_exec)(struct blorp_batch *batch,
167 const struct blorp_params *params)
168 {
169 assert(batch->blorp->driver_ctx == batch->driver_batch);
170 struct brw_context *brw = batch->driver_batch;
171 struct gl_context *ctx = &brw->ctx;
172 const uint32_t estimated_max_batch_usage = GEN_GEN >= 8 ? 1800 : 1500;
173 bool check_aperture_failed_once = false;
174
175 /* Flush the sampler and render caches. We definitely need to flush the
176 * sampler cache so that we get updated contents from the render cache for
177 * the glBlitFramebuffer() source. Also, we are sometimes warned in the
178 * docs to flush the cache between reinterpretations of the same surface
179 * data with different formats, which blorp does for stencil and depth
180 * data.
181 */
182 brw_emit_mi_flush(brw);
183
184 brw_select_pipeline(brw, BRW_RENDER_PIPELINE);
185
186 retry:
187 intel_batchbuffer_require_space(brw, estimated_max_batch_usage, RENDER_RING);
188 intel_batchbuffer_save_state(brw);
189 drm_intel_bo *saved_bo = brw->batch.bo;
190 uint32_t saved_used = USED_BATCH(brw->batch);
191 uint32_t saved_state_batch_offset = brw->batch.state_batch_offset;
192
193 #if GEN_GEN == 6
194 /* Emit workaround flushes when we switch from drawing to blorping. */
195 brw_emit_post_sync_nonzero_flush(brw);
196 #endif
197
198 brw_upload_state_base_address(brw);
199
200 #if GEN_GEN >= 8
201 gen7_l3_state.emit(brw);
202 #endif
203
204 if (brw->use_resource_streamer)
205 gen7_disable_hw_binding_tables(brw);
206
207 brw_emit_depth_stall_flushes(brw);
208
209 blorp_emit(batch, GENX(3DSTATE_DRAWING_RECTANGLE), rect) {
210 rect.ClippedDrawingRectangleXMax = MAX2(params->x1, params->x0) - 1;
211 rect.ClippedDrawingRectangleYMax = MAX2(params->y1, params->y0) - 1;
212 }
213
214 blorp_exec(batch, params);
215
216 /* Make sure we didn't wrap the batch unintentionally, and make sure we
217 * reserved enough space that a wrap will never happen.
218 */
219 assert(brw->batch.bo == saved_bo);
220 assert((USED_BATCH(brw->batch) - saved_used) * 4 +
221 (saved_state_batch_offset - brw->batch.state_batch_offset) <
222 estimated_max_batch_usage);
223 /* Shut up compiler warnings on release build */
224 (void)saved_bo;
225 (void)saved_used;
226 (void)saved_state_batch_offset;
227
228 /* Check if the blorp op we just did would make our batch likely to fail to
229 * map all the BOs into the GPU at batch exec time later. If so, flush the
230 * batch and try again with nothing else in the batch.
231 */
232 if (dri_bufmgr_check_aperture_space(&brw->batch.bo, 1)) {
233 if (!check_aperture_failed_once) {
234 check_aperture_failed_once = true;
235 intel_batchbuffer_reset_to_saved(brw);
236 intel_batchbuffer_flush(brw);
237 goto retry;
238 } else {
239 int ret = intel_batchbuffer_flush(brw);
240 WARN_ONCE(ret == -ENOSPC,
241 "i965: blorp emit exceeded available aperture space\n");
242 }
243 }
244
245 if (unlikely(brw->always_flush_batch))
246 intel_batchbuffer_flush(brw);
247
248 /* We've smashed all state compared to what the normal 3D pipeline
249 * rendering tracks for GL.
250 */
251 brw->ctx.NewDriverState |= BRW_NEW_BLORP;
252 brw->no_depth_or_stencil = false;
253 brw->ib.type = -1;
254
255 /* Flush the sampler cache so any texturing from the destination is
256 * coherent.
257 */
258 brw_emit_mi_flush(brw);
259 }