1 /* 2 * Copyright 2017 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 shall be included 12 * in all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 15 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 18 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 19 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER 20 * DEALINGS IN THE SOFTWARE. 21 */ 22 23 #include <stdio.h> 24 #include <time.h> 25 #include "pipe/p_defines.h" 26 #include "pipe/p_state.h" 27 #include "util/debug.h" 28 #include "util/ralloc.h" 29 #include "util/u_inlines.h" 30 #include "util/format/u_format.h" 31 #include "util/u_upload_mgr.h" 32 #include "drm-uapi/i915_drm.h" 33 #include "iris_context.h" 34 #include "iris_resource.h" 35 #include "iris_screen.h" 36 #include "common/intel_defines.h" 37 #include "common/intel_sample_positions.h" 38 39 /** 40 * The pipe->set_debug_callback() driver hook. 41 */ 42 static void 43 iris_set_debug_callback(struct pipe_context *ctx, 44 const struct pipe_debug_callback *cb) 45 { 46 struct iris_context *ice = (struct iris_context *)ctx; 47 struct iris_screen *screen = (struct iris_screen *)ctx->screen; 48 49 util_queue_finish(&screen->shader_compiler_queue); 50 51 if (cb) 52 ice->dbg = *cb; 53 else 54 memset(&ice->dbg, 0, sizeof(ice->dbg)); 55 } 56 57 /** 58 * Called from the batch module when it detects a GPU hang. 59 * 60 * In this case, we've lost our GEM context, and can't rely on any existing 61 * state on the GPU. We must mark everything dirty and wipe away any saved 62 * assumptions about the last known state of the GPU. 63 */ 64 void 65 iris_lost_context_state(struct iris_batch *batch) 66 { 67 struct iris_context *ice = batch->ice; 68 69 if (batch->name == IRIS_BATCH_RENDER) { 70 batch->screen->vtbl.init_render_context(batch); 71 } else if (batch->name == IRIS_BATCH_COMPUTE) { 72 batch->screen->vtbl.init_compute_context(batch); 73 } else { 74 unreachable("unhandled batch reset"); 75 } 76 77 ice->state.dirty = ~0ull; 78 ice->state.stage_dirty = ~0ull; 79 ice->state.current_hash_scale = 0; 80 memset(&ice->shaders.urb, 0, sizeof(ice->shaders.urb)); 81 memset(ice->state.last_block, 0, sizeof(ice->state.last_block)); 82 memset(ice->state.last_grid, 0, sizeof(ice->state.last_grid)); 83 batch->last_surface_base_address = ~0ull; 84 batch->last_aux_map_state = 0; 85 batch->screen->vtbl.lost_genx_state(ice, batch); 86 } 87 88 static enum pipe_reset_status 89 iris_get_device_reset_status(struct pipe_context *ctx) 90 { 91 struct iris_context *ice = (struct iris_context *)ctx; 92 93 enum pipe_reset_status worst_reset = PIPE_NO_RESET; 94 95 /* Check the reset status of each batch's hardware context, and take the 96 * worst status (if one was guilty, proclaim guilt). 97 */ 98 for (int i = 0; i < IRIS_BATCH_COUNT; i++) { 99 /* This will also recreate the hardware contexts as necessary, so any 100 * future queries will show no resets. We only want to report once. 101 */ 102 enum pipe_reset_status batch_reset = 103 iris_batch_check_for_reset(&ice->batches[i]); 104 105 if (batch_reset == PIPE_NO_RESET) 106 continue; 107 108 if (worst_reset == PIPE_NO_RESET) { 109 worst_reset = batch_reset; 110 } else { 111 /* GUILTY < INNOCENT < UNKNOWN */ 112 worst_reset = MIN2(worst_reset, batch_reset); 113 } 114 } 115 116 if (worst_reset != PIPE_NO_RESET && ice->reset.reset) 117 ice->reset.reset(ice->reset.data, worst_reset); 118 119 return worst_reset; 120 } 121 122 static void 123 iris_set_device_reset_callback(struct pipe_context *ctx, 124 const struct pipe_device_reset_callback *cb) 125 { 126 struct iris_context *ice = (struct iris_context *)ctx; 127 128 if (cb) 129 ice->reset = *cb; 130 else 131 memset(&ice->reset, 0, sizeof(ice->reset)); 132 } 133 134 static void 135 iris_get_sample_position(struct pipe_context *ctx, 136 unsigned sample_count, 137 unsigned sample_index, 138 float *out_value) 139 { 140 union { 141 struct { 142 float x[16]; 143 float y[16]; 144 } a; 145 struct { 146 float _0XOffset, _1XOffset, _2XOffset, _3XOffset, 147 _4XOffset, _5XOffset, _6XOffset, _7XOffset, 148 _8XOffset, _9XOffset, _10XOffset, _11XOffset, 149 _12XOffset, _13XOffset, _14XOffset, _15XOffset; 150 float _0YOffset, _1YOffset, _2YOffset, _3YOffset, 151 _4YOffset, _5YOffset, _6YOffset, _7YOffset, 152 _8YOffset, _9YOffset, _10YOffset, _11YOffset, 153 _12YOffset, _13YOffset, _14YOffset, _15YOffset; 154 } v; 155 } u; 156 switch (sample_count) { 157 case 1: INTEL_SAMPLE_POS_1X(u.v._); break; 158 case 2: INTEL_SAMPLE_POS_2X(u.v._); break; 159 case 4: INTEL_SAMPLE_POS_4X(u.v._); break; 160 case 8: INTEL_SAMPLE_POS_8X(u.v._); break; 161 case 16: INTEL_SAMPLE_POS_16X(u.v._); break; 162 default: unreachable("invalid sample count"); 163 } 164 165 out_value[0] = u.a.x[sample_index]; 166 out_value[1] = u.a.y[sample_index]; 167 } 168 169 static bool 170 create_dirty_dmabuf_set(struct iris_context *ice) 171 { 172 assert(ice->dirty_dmabufs == NULL); 173 174 ice->dirty_dmabufs = _mesa_pointer_set_create(ice); 175 return ice->dirty_dmabufs != NULL; 176 } 177 178 void 179 iris_mark_dirty_dmabuf(struct iris_context *ice, 180 struct pipe_resource *res) 181 { 182 if (!_mesa_set_search(ice->dirty_dmabufs, res)) { 183 _mesa_set_add(ice->dirty_dmabufs, res); 184 pipe_reference(NULL, &res->reference); 185 } 186 } 187 188 static void 189 clear_dirty_dmabuf_set(struct iris_context *ice) 190 { 191 set_foreach(ice->dirty_dmabufs, entry) { 192 struct pipe_resource *res = (struct pipe_resource *)entry->key; 193 if (pipe_reference(&res->reference, NULL)) 194 res->screen->resource_destroy(res->screen, res); 195 } 196 197 _mesa_set_clear(ice->dirty_dmabufs, NULL); 198 } 199 200 void 201 iris_flush_dirty_dmabufs(struct iris_context *ice) 202 { 203 set_foreach(ice->dirty_dmabufs, entry) { 204 struct pipe_resource *res = (struct pipe_resource *)entry->key; 205 ice->ctx.flush_resource(&ice->ctx, res); 206 } 207 208 clear_dirty_dmabuf_set(ice); 209 } 210 211 212 /** 213 * Destroy a context, freeing any associated memory. 214 */ 215 void 216 iris_destroy_context(struct pipe_context *ctx) 217 { 218 struct iris_context *ice = (struct iris_context *)ctx; 219 struct iris_screen *screen = (struct iris_screen *)ctx->screen; 220 221 if (ctx->stream_uploader) 222 u_upload_destroy(ctx->stream_uploader); 223 if (ctx->const_uploader) 224 u_upload_destroy(ctx->const_uploader); 225 226 clear_dirty_dmabuf_set(ice); 227 228 screen->vtbl.destroy_state(ice); 229 230 for (unsigned i = 0; i < ARRAY_SIZE(ice->shaders.scratch_surfs); i++) 231 pipe_resource_reference(&ice->shaders.scratch_surfs[i].res, NULL); 232 233 iris_destroy_program_cache(ice); 234 iris_destroy_border_color_pool(ice); 235 if (screen->measure.config) 236 iris_destroy_ctx_measure(ice); 237 238 u_upload_destroy(ice->state.surface_uploader); 239 u_upload_destroy(ice->state.bindless_uploader); 240 u_upload_destroy(ice->state.dynamic_uploader); 241 u_upload_destroy(ice->query_buffer_uploader); 242 243 iris_batch_free(&ice->batches[IRIS_BATCH_RENDER]); 244 iris_batch_free(&ice->batches[IRIS_BATCH_COMPUTE]); 245 iris_destroy_binder(&ice->state.binder); 246 247 slab_destroy_child(&ice->transfer_pool); 248 slab_destroy_child(&ice->transfer_pool_unsync); 249 250 ralloc_free(ice); 251 } 252 253 #define genX_call(devinfo, func, ...) \ 254 switch ((devinfo)->verx10) { \ 255 case 125: \ 256 gfx125_##func(__VA_ARGS__); \ 257 break; \ 258 case 120: \ 259 gfx12_##func(__VA_ARGS__); \ 260 break; \ 261 case 110: \ 262 gfx11_##func(__VA_ARGS__); \ 263 break; \ 264 case 90: \ 265 gfx9_##func(__VA_ARGS__); \ 266 break; \ 267 case 80: \ 268 gfx8_##func(__VA_ARGS__); \ 269 break; \ 270 default: \ 271 unreachable("Unknown hardware generation"); \ 272 } 273 274 /** 275 * Create a context. 276 * 277 * This is where each context begins. 278 */ 279 struct pipe_context * 280 iris_create_context(struct pipe_screen *pscreen, void *priv, unsigned flags) 281 { 282 struct iris_screen *screen = (struct iris_screen*)pscreen; 283 const struct intel_device_info *devinfo = &screen->devinfo; 284 struct iris_context *ice = rzalloc(NULL, struct iris_context); 285 286 if (!ice) 287 return NULL; 288 289 struct pipe_context *ctx = &ice->ctx; 290 291 ctx->screen = pscreen; 292 ctx->priv = priv; 293 294 ctx->stream_uploader = u_upload_create_default(ctx); 295 if (!ctx->stream_uploader) { 296 free(ctx); 297 return NULL; 298 } 299 ctx->const_uploader = u_upload_create(ctx, 1024 * 1024, 300 PIPE_BIND_CONSTANT_BUFFER, 301 PIPE_USAGE_IMMUTABLE, 302 IRIS_RESOURCE_FLAG_DEVICE_MEM); 303 if (!ctx->const_uploader) { 304 u_upload_destroy(ctx->stream_uploader); 305 free(ctx); 306 return NULL; 307 } 308 309 if (!create_dirty_dmabuf_set(ice)) { 310 ralloc_free(ice); 311 return NULL; 312 } 313 314 ctx->destroy = iris_destroy_context; 315 ctx->set_debug_callback = iris_set_debug_callback; 316 ctx->set_device_reset_callback = iris_set_device_reset_callback; 317 ctx->get_device_reset_status = iris_get_device_reset_status; 318 ctx->get_sample_position = iris_get_sample_position; 319 320 iris_init_context_fence_functions(ctx); 321 iris_init_blit_functions(ctx); 322 iris_init_clear_functions(ctx); 323 iris_init_program_functions(ctx); 324 iris_init_resource_functions(ctx); 325 iris_init_flush_functions(ctx); 326 iris_init_perfquery_functions(ctx); 327 328 iris_init_program_cache(ice); 329 iris_init_border_color_pool(ice); 330 iris_init_binder(ice); 331 332 slab_create_child(&ice->transfer_pool, &screen->transfer_pool); 333 slab_create_child(&ice->transfer_pool_unsync, &screen->transfer_pool); 334 335 ice->state.surface_uploader = 336 u_upload_create(ctx, 64 * 1024, PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE, 337 IRIS_RESOURCE_FLAG_SURFACE_MEMZONE | 338 IRIS_RESOURCE_FLAG_DEVICE_MEM); 339 ice->state.bindless_uploader = 340 u_upload_create(ctx, 64 * 1024, PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE, 341 IRIS_RESOURCE_FLAG_BINDLESS_MEMZONE | 342 IRIS_RESOURCE_FLAG_DEVICE_MEM); 343 ice->state.dynamic_uploader = 344 u_upload_create(ctx, 64 * 1024, PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE, 345 IRIS_RESOURCE_FLAG_DYNAMIC_MEMZONE | 346 IRIS_RESOURCE_FLAG_DEVICE_MEM); 347 348 ice->query_buffer_uploader = 349 u_upload_create(ctx, 16 * 1024, PIPE_BIND_CUSTOM, PIPE_USAGE_STAGING, 350 0); 351 352 genX_call(devinfo, init_state, ice); 353 genX_call(devinfo, init_blorp, ice); 354 genX_call(devinfo, init_query, ice); 355 356 int priority = 0; 357 if (flags & PIPE_CONTEXT_HIGH_PRIORITY) 358 priority = INTEL_CONTEXT_HIGH_PRIORITY; 359 if (flags & PIPE_CONTEXT_LOW_PRIORITY) 360 priority = INTEL_CONTEXT_LOW_PRIORITY; 361 362 if (INTEL_DEBUG(DEBUG_BATCH)) 363 ice->state.sizes = _mesa_hash_table_u64_create(ice); 364 365 for (int i = 0; i < IRIS_BATCH_COUNT; i++) { 366 iris_init_batch(ice, (enum iris_batch_name) i, priority); 367 } 368 369 screen->vtbl.init_render_context(&ice->batches[IRIS_BATCH_RENDER]); 370 screen->vtbl.init_compute_context(&ice->batches[IRIS_BATCH_COMPUTE]); 371 372 if (!(flags & PIPE_CONTEXT_PREFER_THREADED)) 373 return ctx; 374 375 /* Clover doesn't support u_threaded_context */ 376 if (flags & PIPE_CONTEXT_COMPUTE_ONLY) 377 return ctx; 378 379 return threaded_context_create(ctx, &screen->transfer_pool, 380 iris_replace_buffer_storage, 381 NULL, /* TODO: asynchronous flushes? */ 382 &ice->thrctx); 383 } 384