blorp: Pass the VB size to the VF cache workaround
[mesa.git] / src / intel / vulkan / genX_blorp_exec.c
1 /*
2 * Copyright © 2016 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 "anv_private.h"
27
28 /* These are defined in anv_private.h and blorp_genX_exec.h */
29 #undef __gen_address_type
30 #undef __gen_user_data
31 #undef __gen_combine_address
32
33 #include "common/gen_l3_config.h"
34 #include "common/gen_sample_positions.h"
35 #include "blorp/blorp_genX_exec.h"
36
37 static void *
38 blorp_emit_dwords(struct blorp_batch *batch, unsigned n)
39 {
40 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
41 return anv_batch_emit_dwords(&cmd_buffer->batch, n);
42 }
43
44 static uint64_t
45 blorp_emit_reloc(struct blorp_batch *batch,
46 void *location, struct blorp_address address, uint32_t delta)
47 {
48 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
49 assert(cmd_buffer->batch.start <= location &&
50 location < cmd_buffer->batch.end);
51 return anv_batch_emit_reloc(&cmd_buffer->batch, location,
52 address.buffer, address.offset + delta);
53 }
54
55 static void
56 blorp_surface_reloc(struct blorp_batch *batch, uint32_t ss_offset,
57 struct blorp_address address, uint32_t delta)
58 {
59 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
60 uint64_t address_u64 = 0;
61 VkResult result =
62 anv_reloc_list_add(&cmd_buffer->surface_relocs, &cmd_buffer->pool->alloc,
63 ss_offset, address.buffer, address.offset + delta,
64 &address_u64);
65 if (result != VK_SUCCESS)
66 anv_batch_set_error(&cmd_buffer->batch, result);
67
68 void *dest = anv_block_pool_map(
69 &cmd_buffer->device->surface_state_pool.block_pool, ss_offset);
70 write_reloc(cmd_buffer->device, dest, address_u64, false);
71 }
72
73 static uint64_t
74 blorp_get_surface_address(struct blorp_batch *blorp_batch,
75 struct blorp_address address)
76 {
77 /* We'll let blorp_surface_reloc write the address. */
78 return 0ull;
79 }
80
81 #if GEN_GEN >= 7 && GEN_GEN < 10
82 static struct blorp_address
83 blorp_get_surface_base_address(struct blorp_batch *batch)
84 {
85 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
86 return (struct blorp_address) {
87 .buffer = cmd_buffer->device->surface_state_pool.block_pool.bo,
88 .offset = 0,
89 };
90 }
91 #endif
92
93 static void *
94 blorp_alloc_dynamic_state(struct blorp_batch *batch,
95 uint32_t size,
96 uint32_t alignment,
97 uint32_t *offset)
98 {
99 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
100
101 struct anv_state state =
102 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, size, alignment);
103
104 *offset = state.offset;
105 return state.map;
106 }
107
108 static void
109 blorp_alloc_binding_table(struct blorp_batch *batch, unsigned num_entries,
110 unsigned state_size, unsigned state_alignment,
111 uint32_t *bt_offset,
112 uint32_t *surface_offsets, void **surface_maps)
113 {
114 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
115
116 uint32_t state_offset;
117 struct anv_state bt_state;
118
119 VkResult result =
120 anv_cmd_buffer_alloc_blorp_binding_table(cmd_buffer, num_entries,
121 &state_offset, &bt_state);
122 if (result != VK_SUCCESS)
123 return;
124
125 uint32_t *bt_map = bt_state.map;
126 *bt_offset = bt_state.offset;
127
128 for (unsigned i = 0; i < num_entries; i++) {
129 struct anv_state surface_state =
130 anv_cmd_buffer_alloc_surface_state(cmd_buffer);
131 bt_map[i] = surface_state.offset + state_offset;
132 surface_offsets[i] = surface_state.offset;
133 surface_maps[i] = surface_state.map;
134 }
135 }
136
137 static void *
138 blorp_alloc_vertex_buffer(struct blorp_batch *batch, uint32_t size,
139 struct blorp_address *addr)
140 {
141 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
142
143 /* From the Skylake PRM, 3DSTATE_VERTEX_BUFFERS:
144 *
145 * "The VF cache needs to be invalidated before binding and then using
146 * Vertex Buffers that overlap with any previously bound Vertex Buffer
147 * (at a 64B granularity) since the last invalidation. A VF cache
148 * invalidate is performed by setting the "VF Cache Invalidation Enable"
149 * bit in PIPE_CONTROL."
150 *
151 * This restriction first appears in the Skylake PRM but the internal docs
152 * also list it as being an issue on Broadwell. In order to avoid this
153 * problem, we align all vertex buffer allocations to 64 bytes.
154 */
155 struct anv_state vb_state =
156 anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, size, 64);
157
158 *addr = (struct blorp_address) {
159 .buffer = cmd_buffer->device->dynamic_state_pool.block_pool.bo,
160 .offset = vb_state.offset,
161 .mocs = cmd_buffer->device->isl_dev.mocs.internal,
162 };
163
164 return vb_state.map;
165 }
166
167 static void
168 blorp_vf_invalidate_for_vb_48b_transitions(struct blorp_batch *batch,
169 const struct blorp_address *addrs,
170 uint32_t *sizes,
171 unsigned num_vbs)
172 {
173 /* anv forces all vertex buffers into the low 4GB so there are never any
174 * transitions that require a VF invalidation.
175 */
176 }
177
178 #if GEN_GEN >= 8
179 static struct blorp_address
180 blorp_get_workaround_page(struct blorp_batch *batch)
181 {
182 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
183
184 return (struct blorp_address) {
185 .buffer = cmd_buffer->device->workaround_bo,
186 };
187 }
188 #endif
189
190 static void
191 blorp_flush_range(struct blorp_batch *batch, void *start, size_t size)
192 {
193 /* We don't need to flush states anymore, since everything will be snooped.
194 */
195 }
196
197 static void
198 blorp_emit_urb_config(struct blorp_batch *batch,
199 unsigned vs_entry_size, unsigned sf_entry_size)
200 {
201 struct anv_device *device = batch->blorp->driver_ctx;
202 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
203
204 assert(sf_entry_size == 0);
205
206 const unsigned entry_size[4] = { vs_entry_size, 1, 1, 1 };
207
208 genX(emit_urb_setup)(device, &cmd_buffer->batch,
209 cmd_buffer->state.current_l3_config,
210 VK_SHADER_STAGE_VERTEX_BIT |
211 VK_SHADER_STAGE_FRAGMENT_BIT,
212 entry_size);
213 }
214
215 void
216 genX(blorp_exec)(struct blorp_batch *batch,
217 const struct blorp_params *params)
218 {
219 struct anv_cmd_buffer *cmd_buffer = batch->driver_batch;
220
221 if (!cmd_buffer->state.current_l3_config) {
222 const struct gen_l3_config *cfg =
223 gen_get_default_l3_config(&cmd_buffer->device->info);
224 genX(cmd_buffer_config_l3)(cmd_buffer, cfg);
225 }
226
227 const unsigned scale = params->fast_clear_op ? UINT_MAX : 1;
228 genX(cmd_buffer_emit_hashing_mode)(cmd_buffer, params->x1 - params->x0,
229 params->y1 - params->y0, scale);
230
231 #if GEN_GEN >= 11
232 /* The PIPE_CONTROL command description says:
233 *
234 * "Whenever a Binding Table Index (BTI) used by a Render Taget Message
235 * points to a different RENDER_SURFACE_STATE, SW must issue a Render
236 * Target Cache Flush by enabling this bit. When render target flush
237 * is set due to new association of BTI, PS Scoreboard Stall bit must
238 * be set in this packet."
239 */
240 cmd_buffer->state.pending_pipe_bits |=
241 ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT |
242 ANV_PIPE_STALL_AT_SCOREBOARD_BIT;
243 #endif
244
245 #if GEN_GEN == 7
246 /* The MI_LOAD/STORE_REGISTER_MEM commands which BLORP uses to implement
247 * indirect fast-clear colors can cause GPU hangs if we don't stall first.
248 * See genX(cmd_buffer_mi_memcpy) for more details.
249 */
250 if (params->src.clear_color_addr.buffer ||
251 params->dst.clear_color_addr.buffer)
252 cmd_buffer->state.pending_pipe_bits |= ANV_PIPE_CS_STALL_BIT;
253 #endif
254
255 genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer);
256
257 genX(flush_pipeline_select_3d)(cmd_buffer);
258
259 #if GEN_GEN >= 12
260 genX(cmd_buffer_aux_map_state)(cmd_buffer);
261 #endif
262
263 genX(cmd_buffer_emit_gen7_depth_flush)(cmd_buffer);
264
265 /* BLORP doesn't do anything fancy with depth such as discards, so we want
266 * the PMA fix off. Also, off is always the safe option.
267 */
268 genX(cmd_buffer_enable_pma_fix)(cmd_buffer, false);
269
270 blorp_exec(batch, params);
271
272 cmd_buffer->state.gfx.vb_dirty = ~0;
273 cmd_buffer->state.gfx.dirty = ~0;
274 cmd_buffer->state.push_constants_dirty = ~0;
275 }