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         /* if no rendering since last flush, ie. app just decided it needed
  53          * a fence, re-use the last one:
  54          */
  55         if (ctx->last_fence) {
  56                 fd_fence_ref(pctx->screen, &fence, ctx->last_fence);
  57                 goto out;
  58         }
  59
  60         if (!batch)
  61                 return;
  62
  63         /* Take a ref to the batch's fence (batch can be unref'd when flushed: */
  64         fd_fence_ref(pctx->screen, &fence, batch->fence);
  65
  66         /* TODO is it worth trying to figure out if app is using fence-fd's, to
  67          * avoid requesting one every batch?
  68          */
  69         batch->needs_out_fence_fd = true;
  70
  71         if (!ctx->screen->reorder) {
  72                 fd_batch_flush(batch, true, false);
  73         } else if (flags & PIPE_FLUSH_DEFERRED) {
  74                 fd_bc_flush_deferred(&ctx->screen->batch_cache, ctx);
  75         } else {
  76                 fd_bc_flush(&ctx->screen->batch_cache, ctx);
  77         }
  78
  79 out:
  80         if (fencep)
  81                 fd_fence_ref(pctx->screen, fencep, fence);
  82
  83         fd_fence_ref(pctx->screen, &ctx->last_fence, fence);
  84
  85         fd_fence_ref(pctx->screen, &fence, NULL);
  86 }
  87
  88 static void
  89 fd_texture_barrier(struct pipe_context *pctx, unsigned flags)
  90 {
  91         /* On devices that could sample from GMEM we could possibly do better.
  92          * Or if we knew that we were doing GMEM bypass we could just emit a
  93          * cache flush, perhaps?  But we don't know if future draws would cause
  94          * us to use GMEM, and a flush in bypass isn't the end of the world.
  95          */
  96         fd_context_flush(pctx, NULL, 0);
  97 }
  98
  99 static void
 100 fd_memory_barrier(struct pipe_context *pctx, unsigned flags)
 101 {
 102         fd_context_flush(pctx, NULL, 0);
 103         /* TODO do we need to check for persistently mapped buffers and fd_bo_cpu_prep()?? */
 104 }
 105
 106 /**
 107  * emit marker string as payload of a no-op packet, which can be
 108  * decoded by cffdump.
 109  */
 110 static void
 111 fd_emit_string_marker(struct pipe_context *pctx, const char *string, int len)
 112 {
 113         struct fd_context *ctx = fd_context(pctx);
 114         struct fd_ringbuffer *ring;
 115         const uint32_t *buf = (const void *)string;
 116
 117         if (!ctx->batch)
 118                 return;
 119
 120         ctx->batch->needs_flush = true;
 121
 122         ring = ctx->batch->draw;
 123
 124         /* max packet size is 0x3fff dwords: */
 125         len = MIN2(len, 0x3fff * 4);
 126
 127         if (ctx->screen->gpu_id >= 500)
 128                 OUT_PKT7(ring, CP_NOP, align(len, 4) / 4);
 129         else
 130                 OUT_PKT3(ring, CP_NOP, align(len, 4) / 4);
 131         while (len >= 4) {
 132                 OUT_RING(ring, *buf);
 133                 buf++;
 134                 len -= 4;
 135         }
 136
 137         /* copy remainder bytes without reading past end of input string: */
 138         if (len > 0) {
 139                 uint32_t w = 0;
 140                 memcpy(&w, buf, len);
 141                 OUT_RING(ring, w);
 142         }
 143 }
 144
 145 void
 146 fd_context_destroy(struct pipe_context *pctx)
 147 {
 148         struct fd_context *ctx = fd_context(pctx);
 149         unsigned i;
 150
 151         DBG("");
 152
 153         fd_fence_ref(pctx->screen, &ctx->last_fence, NULL);
 154
 155         if (ctx->screen->reorder && util_queue_is_initialized(&ctx->flush_queue))
 156                 util_queue_destroy(&ctx->flush_queue);
 157
 158         util_copy_framebuffer_state(&ctx->framebuffer, NULL);
 159         fd_batch_reference(&ctx->batch, NULL);  /* unref current batch */
 160         fd_bc_invalidate_context(ctx);
 161
 162         fd_prog_fini(pctx);
 163
 164         if (ctx->blitter)
 165                 util_blitter_destroy(ctx->blitter);
 166
 167         if (pctx->stream_uploader)
 168                 u_upload_destroy(pctx->stream_uploader);
 169
 170         if (ctx->clear_rs_state)
 171                 pctx->delete_rasterizer_state(pctx, ctx->clear_rs_state);
 172
 173         if (ctx->primconvert)
 174                 util_primconvert_destroy(ctx->primconvert);
 175
 176         slab_destroy_child(&ctx->transfer_pool);
 177
 178         for (i = 0; i < ARRAY_SIZE(ctx->vsc_pipe); i++) {
 179                 struct fd_vsc_pipe *pipe = &ctx->vsc_pipe[i];
 180                 if (!pipe->bo)
 181                         break;
 182                 fd_bo_del(pipe->bo);
 183         }
 184
 185         fd_device_del(ctx->dev);
 186         fd_pipe_del(ctx->pipe);
 187
 188         if (fd_mesa_debug & (FD_DBG_BSTAT | FD_DBG_MSGS)) {
 189                 printf("batch_total=%u, batch_sysmem=%u, batch_gmem=%u, batch_nondraw=%u, batch_restore=%u\n",
 190                         (uint32_t)ctx->stats.batch_total, (uint32_t)ctx->stats.batch_sysmem,
 191                         (uint32_t)ctx->stats.batch_gmem, (uint32_t)ctx->stats.batch_nondraw,
 192                         (uint32_t)ctx->stats.batch_restore);
 193         }
 194 }
 195
 196 static void
 197 fd_set_debug_callback(struct pipe_context *pctx,
 198                 const struct pipe_debug_callback *cb)
 199 {
 200         struct fd_context *ctx = fd_context(pctx);
 201
 202         if (cb)
 203                 ctx->debug = *cb;
 204         else
 205                 memset(&ctx->debug, 0, sizeof(ctx->debug));
 206 }
 207
 208 /* TODO we could combine a few of these small buffers (solid_vbuf,
 209  * blit_texcoord_vbuf, and vsc_size_mem, into a single buffer and
 210  * save a tiny bit of memory
 211  */
 212
 213 static struct pipe_resource *
 214 create_solid_vertexbuf(struct pipe_context *pctx)
 215 {
 216         static const float init_shader_const[] = {
 217                         -1.000000, +1.000000, +1.000000,
 218                         +1.000000, -1.000000, +1.000000,
 219         };
 220         struct pipe_resource *prsc = pipe_buffer_create(pctx->screen,
 221                         PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE, sizeof(init_shader_const));
 222         pipe_buffer_write(pctx, prsc, 0,
 223                         sizeof(init_shader_const), init_shader_const);
 224         return prsc;
 225 }
 226
 227 static struct pipe_resource *
 228 create_blit_texcoord_vertexbuf(struct pipe_context *pctx)
 229 {
 230         struct pipe_resource *prsc = pipe_buffer_create(pctx->screen,
 231                         PIPE_BIND_CUSTOM, PIPE_USAGE_DYNAMIC, 16);
 232         return prsc;
 233 }
 234
 235 void
 236 fd_context_setup_common_vbos(struct fd_context *ctx)
 237 {
 238         struct pipe_context *pctx = &ctx->base;
 239
 240         ctx->solid_vbuf = create_solid_vertexbuf(pctx);
 241         ctx->blit_texcoord_vbuf = create_blit_texcoord_vertexbuf(pctx);
 242
 243         /* setup solid_vbuf_state: */
 244         ctx->solid_vbuf_state.vtx = pctx->create_vertex_elements_state(
 245                         pctx, 1, (struct pipe_vertex_element[]){{
 246                                 .vertex_buffer_index = 0,
 247                                 .src_offset = 0,
 248                                 .src_format = PIPE_FORMAT_R32G32B32_FLOAT,
 249                         }});
 250         ctx->solid_vbuf_state.vertexbuf.count = 1;
 251         ctx->solid_vbuf_state.vertexbuf.vb[0].stride = 12;
 252         ctx->solid_vbuf_state.vertexbuf.vb[0].buffer.resource = ctx->solid_vbuf;
 253
 254         /* setup blit_vbuf_state: */
 255         ctx->blit_vbuf_state.vtx = pctx->create_vertex_elements_state(
 256                         pctx, 2, (struct pipe_vertex_element[]){{
 257                                 .vertex_buffer_index = 0,
 258                                 .src_offset = 0,
 259                                 .src_format = PIPE_FORMAT_R32G32_FLOAT,
 260                         }, {
 261                                 .vertex_buffer_index = 1,
 262                                 .src_offset = 0,
 263                                 .src_format = PIPE_FORMAT_R32G32B32_FLOAT,
 264                         }});
 265         ctx->blit_vbuf_state.vertexbuf.count = 2;
 266         ctx->blit_vbuf_state.vertexbuf.vb[0].stride = 8;
 267         ctx->blit_vbuf_state.vertexbuf.vb[0].buffer.resource = ctx->blit_texcoord_vbuf;
 268         ctx->blit_vbuf_state.vertexbuf.vb[1].stride = 12;
 269         ctx->blit_vbuf_state.vertexbuf.vb[1].buffer.resource = ctx->solid_vbuf;
 270 }
 271
 272 void
 273 fd_context_cleanup_common_vbos(struct fd_context *ctx)
 274 {
 275         struct pipe_context *pctx = &ctx->base;
 276
 277         pctx->delete_vertex_elements_state(pctx, ctx->solid_vbuf_state.vtx);
 278         pctx->delete_vertex_elements_state(pctx, ctx->blit_vbuf_state.vtx);
 279
 280         pipe_resource_reference(&ctx->solid_vbuf, NULL);
 281         pipe_resource_reference(&ctx->blit_texcoord_vbuf, NULL);
 282 }
 283
 284 struct pipe_context *
 285 fd_context_init(struct fd_context *ctx, struct pipe_screen *pscreen,
 286                 const uint8_t *primtypes, void *priv, unsigned flags)
 287 {
 288         struct fd_screen *screen = fd_screen(pscreen);
 289         struct pipe_context *pctx;
 290         unsigned prio = 1;
 291         int i;
 292
 293         /* lower numerical value == higher priority: */
 294         if (fd_mesa_debug & FD_DBG_HIPRIO)
 295                 prio = 0;
 296         else if (flags & PIPE_CONTEXT_HIGH_PRIORITY)
 297                 prio = 0;
 298         else if (flags & PIPE_CONTEXT_LOW_PRIORITY)
 299                 prio = 2;
 300
 301         ctx->screen = screen;
 302         ctx->pipe = fd_pipe_new2(screen->dev, FD_PIPE_3D, prio);
 303
 304         ctx->primtypes = primtypes;
 305         ctx->primtype_mask = 0;
 306         for (i = 0; i < PIPE_PRIM_MAX; i++)
 307                 if (primtypes[i])
 308                         ctx->primtype_mask |= (1 << i);
 309
 310         /* need some sane default in case state tracker doesn't
 311          * set some state:
 312          */
 313         ctx->sample_mask = 0xffff;
 314
 315         pctx = &ctx->base;
 316         pctx->screen = pscreen;
 317         pctx->priv = priv;
 318         pctx->flush = fd_context_flush;
 319         pctx->emit_string_marker = fd_emit_string_marker;
 320         pctx->set_debug_callback = fd_set_debug_callback;
 321         pctx->create_fence_fd = fd_create_fence_fd;
 322         pctx->fence_server_sync = fd_fence_server_sync;
 323         pctx->texture_barrier = fd_texture_barrier;
 324         pctx->memory_barrier = fd_memory_barrier;
 325
 326         pctx->stream_uploader = u_upload_create_default(pctx);
 327         if (!pctx->stream_uploader)
 328                 goto fail;
 329         pctx->const_uploader = pctx->stream_uploader;
 330
 331         if (!ctx->screen->reorder)
 332                 ctx->batch = fd_bc_alloc_batch(&screen->batch_cache, ctx, false);
 333
 334         slab_create_child(&ctx->transfer_pool, &screen->transfer_pool);
 335
 336         fd_draw_init(pctx);
 337         fd_resource_context_init(pctx);
 338         fd_query_context_init(pctx);
 339         fd_texture_init(pctx);
 340         fd_state_init(pctx);
 341
 342         ctx->blitter = util_blitter_create(pctx);
 343         if (!ctx->blitter)
 344                 goto fail;
 345
 346         ctx->primconvert = util_primconvert_create(pctx, ctx->primtype_mask);
 347         if (!ctx->primconvert)
 348                 goto fail;
 349
 350         list_inithead(&ctx->hw_active_queries);
 351         list_inithead(&ctx->acc_active_queries);
 352
 353         return pctx;
 354
 355 fail:
 356         pctx->destroy(pctx);
 357         return NULL;
 358 }