src/gallium/drivers/freedreno/freedreno_context.c

   1 /*
   2  * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
   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 FROM,
  20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  21  * SOFTWARE.
  22  *
  23  * Authors:
  24  *    Rob Clark <robclark@freedesktop.org>
  25  */
  26
  27 #include "freedreno_context.h"
  28 #include "freedreno_blitter.h"
  29 #include "freedreno_draw.h"
  30 #include "freedreno_fence.h"
  31 #include "freedreno_program.h"
  32 #include "freedreno_resource.h"
  33 #include "freedreno_texture.h"
  34 #include "freedreno_state.h"
  35 #include "freedreno_gmem.h"
  36 #include "freedreno_query.h"
  37 #include "freedreno_query_hw.h"
  38 #include "freedreno_util.h"
  39 #include "util/u_upload_mgr.h"
  40
  41 static void
  42 fd_context_flush(struct pipe_context *pctx, struct pipe_fence_handle **fencep,
  43                 unsigned flags)
  44 {
  45         struct fd_context *ctx = fd_context(pctx);
  46         struct pipe_fence_handle *fence = NULL;
  47         // TODO we want to lookup batch if it exists, but not create one if not.
  48         struct fd_batch *batch = fd_context_batch(ctx);
  49
  50         DBG("%p: flush: flags=%x\n", ctx->batch, flags);
  51
  52         /* In some sequence of events, we can end up with a last_fence that is
  53          * not an "fd" fence, which results in eglDupNativeFenceFDANDROID()
  54          * errors.
  55          *
  56          */
  57         if (flags & PIPE_FLUSH_FENCE_FD)
  58                 fd_fence_ref(&ctx->last_fence, NULL);
  59
  60         /* if no rendering since last flush, ie. app just decided it needed
  61          * a fence, re-use the last one:
  62          */
  63         if (ctx->last_fence) {
  64                 fd_fence_ref(&fence, ctx->last_fence);
  65                 goto out;
  66         }
  67
  68         if (!batch)
  69                 return;
  70
  71         /* Take a ref to the batch's fence (batch can be unref'd when flushed: */
  72         fd_fence_ref(&fence, batch->fence);
  73
  74         if (flags & PIPE_FLUSH_FENCE_FD)
  75                 batch->needs_out_fence_fd = true;
  76
  77         if (!ctx->screen->reorder) {
  78                 fd_batch_flush(batch, true);
  79         } else if (flags & PIPE_FLUSH_DEFERRED) {
  80                 fd_bc_flush_deferred(&ctx->screen->batch_cache, ctx);
  81         } else {
  82                 fd_bc_flush(&ctx->screen->batch_cache, ctx);
  83         }
  84
  85 out:
  86         if (fencep)
  87                 fd_fence_ref(fencep, fence);
  88
  89         fd_fence_ref(&ctx->last_fence, fence);
  90
  91         fd_fence_ref(&fence, NULL);
  92 }
  93
  94 static void
  95 fd_texture_barrier(struct pipe_context *pctx, unsigned flags)
  96 {
  97         if (flags == PIPE_TEXTURE_BARRIER_FRAMEBUFFER) {
  98                 struct fd_context *ctx = fd_context(pctx);
  99
 100                 if (ctx->framebuffer_barrier) {
 101                         ctx->framebuffer_barrier(ctx);
 102                         return;
 103                 }
 104         }
 105
 106         /* On devices that could sample from GMEM we could possibly do better.
 107          * Or if we knew that we were doing GMEM bypass we could just emit a
 108          * cache flush, perhaps?  But we don't know if future draws would cause
 109          * us to use GMEM, and a flush in bypass isn't the end of the world.
 110          */
 111         fd_context_flush(pctx, NULL, 0);
 112 }
 113
 114 static void
 115 fd_memory_barrier(struct pipe_context *pctx, unsigned flags)
 116 {
 117         if (!(flags & ~PIPE_BARRIER_UPDATE))
 118                 return;
 119
 120         fd_context_flush(pctx, NULL, 0);
 121         /* TODO do we need to check for persistently mapped buffers and fd_bo_cpu_prep()?? */
 122 }
 123
 124 /**
 125  * emit marker string as payload of a no-op packet, which can be
 126  * decoded by cffdump.
 127  */
 128 static void
 129 fd_emit_string_marker(struct pipe_context *pctx, const char *string, int len)
 130 {
 131         struct fd_context *ctx = fd_context(pctx);
 132         struct fd_ringbuffer *ring;
 133         const uint32_t *buf = (const void *)string;
 134
 135         if (!ctx->batch)
 136                 return;
 137
 138         ctx->batch->needs_flush = true;
 139
 140         ring = ctx->batch->draw;
 141
 142         /* max packet size is 0x3fff dwords: */
 143         len = MIN2(len, 0x3fff * 4);
 144
 145         if (ctx->screen->gpu_id >= 500)
 146                 OUT_PKT7(ring, CP_NOP, align(len, 4) / 4);
 147         else
 148                 OUT_PKT3(ring, CP_NOP, align(len, 4) / 4);
 149         while (len >= 4) {
 150                 OUT_RING(ring, *buf);
 151                 buf++;
 152                 len -= 4;
 153         }
 154
 155         /* copy remainder bytes without reading past end of input string: */
 156         if (len > 0) {
 157                 uint32_t w = 0;
 158                 memcpy(&w, buf, len);
 159                 OUT_RING(ring, w);
 160         }
 161 }
 162
 163 void
 164 fd_context_destroy(struct pipe_context *pctx)
 165 {
 166         struct fd_context *ctx = fd_context(pctx);
 167         unsigned i;
 168
 169         DBG("");
 170
 171         fd_fence_ref(&ctx->last_fence, NULL);
 172
 173         if (ctx->screen->reorder && util_queue_is_initialized(&ctx->flush_queue))
 174                 util_queue_destroy(&ctx->flush_queue);
 175
 176         util_copy_framebuffer_state(&ctx->framebuffer, NULL);
 177         fd_batch_reference(&ctx->batch, NULL);  /* unref current batch */
 178         fd_bc_invalidate_context(ctx);
 179
 180         fd_prog_fini(pctx);
 181
 182         if (ctx->blitter)
 183                 util_blitter_destroy(ctx->blitter);
 184
 185         if (pctx->stream_uploader)
 186                 u_upload_destroy(pctx->stream_uploader);
 187
 188         if (ctx->clear_rs_state)
 189                 pctx->delete_rasterizer_state(pctx, ctx->clear_rs_state);
 190
 191         if (ctx->primconvert)
 192                 util_primconvert_destroy(ctx->primconvert);
 193
 194         slab_destroy_child(&ctx->transfer_pool);
 195
 196         for (i = 0; i < ARRAY_SIZE(ctx->vsc_pipe); i++) {
 197                 struct fd_vsc_pipe *pipe = &ctx->vsc_pipe[i];
 198                 if (!pipe->bo)
 199                         break;
 200                 fd_bo_del(pipe->bo);
 201         }
 202
 203         fd_device_del(ctx->dev);
 204         fd_pipe_del(ctx->pipe);
 205
 206         if (fd_mesa_debug & (FD_DBG_BSTAT | FD_DBG_MSGS)) {
 207                 printf("batch_total=%u, batch_sysmem=%u, batch_gmem=%u, batch_nondraw=%u, batch_restore=%u\n",
 208                         (uint32_t)ctx->stats.batch_total, (uint32_t)ctx->stats.batch_sysmem,
 209                         (uint32_t)ctx->stats.batch_gmem, (uint32_t)ctx->stats.batch_nondraw,
 210                         (uint32_t)ctx->stats.batch_restore);
 211         }
 212 }
 213
 214 static void
 215 fd_set_debug_callback(struct pipe_context *pctx,
 216                 const struct pipe_debug_callback *cb)
 217 {
 218         struct fd_context *ctx = fd_context(pctx);
 219
 220         if (cb)
 221                 ctx->debug = *cb;
 222         else
 223                 memset(&ctx->debug, 0, sizeof(ctx->debug));
 224 }
 225
 226 static uint32_t
 227 fd_get_reset_count(struct fd_context *ctx, bool per_context)
 228 {
 229         uint64_t val;
 230         enum fd_param_id param =
 231                 per_context ? FD_CTX_FAULTS : FD_GLOBAL_FAULTS;
 232         int ret = fd_pipe_get_param(ctx->pipe, param, &val);
 233         debug_assert(!ret);
 234         return val;
 235 }
 236
 237 static enum pipe_reset_status
 238 fd_get_device_reset_status(struct pipe_context *pctx)
 239 {
 240         struct fd_context *ctx = fd_context(pctx);
 241         int context_faults = fd_get_reset_count(ctx, true);
 242         int global_faults  = fd_get_reset_count(ctx, false);
 243         enum pipe_reset_status status;
 244
 245         if (context_faults != ctx->context_reset_count) {
 246                 status = PIPE_GUILTY_CONTEXT_RESET;
 247         } else if (global_faults != ctx->global_reset_count) {
 248                 status = PIPE_INNOCENT_CONTEXT_RESET;
 249         } else {
 250                 status = PIPE_NO_RESET;
 251         }
 252
 253         ctx->context_reset_count = context_faults;
 254         ctx->global_reset_count = global_faults;
 255
 256         return status;
 257 }
 258
 259 /* TODO we could combine a few of these small buffers (solid_vbuf,
 260  * blit_texcoord_vbuf, and vsc_size_mem, into a single buffer and
 261  * save a tiny bit of memory
 262  */
 263
 264 static struct pipe_resource *
 265 create_solid_vertexbuf(struct pipe_context *pctx)
 266 {
 267         static const float init_shader_const[] = {
 268                         -1.000000, +1.000000, +1.000000,
 269                         +1.000000, -1.000000, +1.000000,
 270         };
 271         struct pipe_resource *prsc = pipe_buffer_create(pctx->screen,
 272                         PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE, sizeof(init_shader_const));
 273         pipe_buffer_write(pctx, prsc, 0,
 274                         sizeof(init_shader_const), init_shader_const);
 275         return prsc;
 276 }
 277
 278 static struct pipe_resource *
 279 create_blit_texcoord_vertexbuf(struct pipe_context *pctx)
 280 {
 281         struct pipe_resource *prsc = pipe_buffer_create(pctx->screen,
 282                         PIPE_BIND_CUSTOM, PIPE_USAGE_DYNAMIC, 16);
 283         return prsc;
 284 }
 285
 286 void
 287 fd_context_setup_common_vbos(struct fd_context *ctx)
 288 {
 289         struct pipe_context *pctx = &ctx->base;
 290
 291         ctx->solid_vbuf = create_solid_vertexbuf(pctx);
 292         ctx->blit_texcoord_vbuf = create_blit_texcoord_vertexbuf(pctx);
 293
 294         /* setup solid_vbuf_state: */
 295         ctx->solid_vbuf_state.vtx = pctx->create_vertex_elements_state(
 296                         pctx, 1, (struct pipe_vertex_element[]){{
 297                                 .vertex_buffer_index = 0,
 298                                 .src_offset = 0,
 299                                 .src_format = PIPE_FORMAT_R32G32B32_FLOAT,
 300                         }});
 301         ctx->solid_vbuf_state.vertexbuf.count = 1;
 302         ctx->solid_vbuf_state.vertexbuf.vb[0].stride = 12;
 303         ctx->solid_vbuf_state.vertexbuf.vb[0].buffer.resource = ctx->solid_vbuf;
 304
 305         /* setup blit_vbuf_state: */
 306         ctx->blit_vbuf_state.vtx = pctx->create_vertex_elements_state(
 307                         pctx, 2, (struct pipe_vertex_element[]){{
 308                                 .vertex_buffer_index = 0,
 309                                 .src_offset = 0,
 310                                 .src_format = PIPE_FORMAT_R32G32_FLOAT,
 311                         }, {
 312                                 .vertex_buffer_index = 1,
 313                                 .src_offset = 0,
 314                                 .src_format = PIPE_FORMAT_R32G32B32_FLOAT,
 315                         }});
 316         ctx->blit_vbuf_state.vertexbuf.count = 2;
 317         ctx->blit_vbuf_state.vertexbuf.vb[0].stride = 8;
 318         ctx->blit_vbuf_state.vertexbuf.vb[0].buffer.resource = ctx->blit_texcoord_vbuf;
 319         ctx->blit_vbuf_state.vertexbuf.vb[1].stride = 12;
 320         ctx->blit_vbuf_state.vertexbuf.vb[1].buffer.resource = ctx->solid_vbuf;
 321 }
 322
 323 void
 324 fd_context_cleanup_common_vbos(struct fd_context *ctx)
 325 {
 326         struct pipe_context *pctx = &ctx->base;
 327
 328         pctx->delete_vertex_elements_state(pctx, ctx->solid_vbuf_state.vtx);
 329         pctx->delete_vertex_elements_state(pctx, ctx->blit_vbuf_state.vtx);
 330
 331         pipe_resource_reference(&ctx->solid_vbuf, NULL);
 332         pipe_resource_reference(&ctx->blit_texcoord_vbuf, NULL);
 333 }
 334
 335 struct pipe_context *
 336 fd_context_init(struct fd_context *ctx, struct pipe_screen *pscreen,
 337                 const uint8_t *primtypes, void *priv, unsigned flags)
 338 {
 339         struct fd_screen *screen = fd_screen(pscreen);
 340         struct pipe_context *pctx;
 341         unsigned prio = 1;
 342         int i;
 343
 344         /* lower numerical value == higher priority: */
 345         if (fd_mesa_debug & FD_DBG_HIPRIO)
 346                 prio = 0;
 347         else if (flags & PIPE_CONTEXT_HIGH_PRIORITY)
 348                 prio = 0;
 349         else if (flags & PIPE_CONTEXT_LOW_PRIORITY)
 350                 prio = 2;
 351
 352         ctx->screen = screen;
 353         ctx->pipe = fd_pipe_new2(screen->dev, FD_PIPE_3D, prio);
 354
 355         if (fd_device_version(screen->dev) >= FD_VERSION_ROBUSTNESS) {
 356                 ctx->context_reset_count = fd_get_reset_count(ctx, true);
 357                 ctx->global_reset_count = fd_get_reset_count(ctx, false);
 358         }
 359
 360         ctx->primtypes = primtypes;
 361         ctx->primtype_mask = 0;
 362         for (i = 0; i < PIPE_PRIM_MAX; i++)
 363                 if (primtypes[i])
 364                         ctx->primtype_mask |= (1 << i);
 365
 366         /* need some sane default in case state tracker doesn't
 367          * set some state:
 368          */
 369         ctx->sample_mask = 0xffff;
 370
 371         pctx = &ctx->base;
 372         pctx->screen = pscreen;
 373         pctx->priv = priv;
 374         pctx->flush = fd_context_flush;
 375         pctx->emit_string_marker = fd_emit_string_marker;
 376         pctx->set_debug_callback = fd_set_debug_callback;
 377         pctx->get_device_reset_status = fd_get_device_reset_status;
 378         pctx->create_fence_fd = fd_create_fence_fd;
 379         pctx->fence_server_sync = fd_fence_server_sync;
 380         pctx->texture_barrier = fd_texture_barrier;
 381         pctx->memory_barrier = fd_memory_barrier;
 382
 383         pctx->stream_uploader = u_upload_create_default(pctx);
 384         if (!pctx->stream_uploader)
 385                 goto fail;
 386         pctx->const_uploader = pctx->stream_uploader;
 387
 388         if (!ctx->screen->reorder)
 389                 ctx->batch = fd_bc_alloc_batch(&screen->batch_cache, ctx, false);
 390
 391         slab_create_child(&ctx->transfer_pool, &screen->transfer_pool);
 392
 393         fd_draw_init(pctx);
 394         fd_resource_context_init(pctx);
 395         fd_query_context_init(pctx);
 396         fd_texture_init(pctx);
 397         fd_state_init(pctx);
 398
 399         ctx->blitter = util_blitter_create(pctx);
 400         if (!ctx->blitter)
 401                 goto fail;
 402
 403         ctx->primconvert = util_primconvert_create(pctx, ctx->primtype_mask);
 404         if (!ctx->primconvert)
 405                 goto fail;
 406
 407         list_inithead(&ctx->hw_active_queries);
 408         list_inithead(&ctx->acc_active_queries);
 409
 410         return pctx;
 411
 412 fail:
 413         pctx->destroy(pctx);
 414         return NULL;
 415 }