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         if (!(flags & ~PIPE_BARRIER_UPDATE))
 103                 return;
 104
 105         fd_context_flush(pctx, NULL, 0);
 106         /* TODO do we need to check for persistently mapped buffers and fd_bo_cpu_prep()?? */
 107 }
 108
 109 /**
 110  * emit marker string as payload of a no-op packet, which can be
 111  * decoded by cffdump.
 112  */
 113 static void
 114 fd_emit_string_marker(struct pipe_context *pctx, const char *string, int len)
 115 {
 116         struct fd_context *ctx = fd_context(pctx);
 117         struct fd_ringbuffer *ring;
 118         const uint32_t *buf = (const void *)string;
 119
 120         if (!ctx->batch)
 121                 return;
 122
 123         ctx->batch->needs_flush = true;
 124
 125         ring = ctx->batch->draw;
 126
 127         /* max packet size is 0x3fff dwords: */
 128         len = MIN2(len, 0x3fff * 4);
 129
 130         if (ctx->screen->gpu_id >= 500)
 131                 OUT_PKT7(ring, CP_NOP, align(len, 4) / 4);
 132         else
 133                 OUT_PKT3(ring, CP_NOP, align(len, 4) / 4);
 134         while (len >= 4) {
 135                 OUT_RING(ring, *buf);
 136                 buf++;
 137                 len -= 4;
 138         }
 139
 140         /* copy remainder bytes without reading past end of input string: */
 141         if (len > 0) {
 142                 uint32_t w = 0;
 143                 memcpy(&w, buf, len);
 144                 OUT_RING(ring, w);
 145         }
 146 }
 147
 148 void
 149 fd_context_destroy(struct pipe_context *pctx)
 150 {
 151         struct fd_context *ctx = fd_context(pctx);
 152         unsigned i;
 153
 154         DBG("");
 155
 156         fd_fence_ref(pctx->screen, &ctx->last_fence, NULL);
 157
 158         if (ctx->screen->reorder && util_queue_is_initialized(&ctx->flush_queue))
 159                 util_queue_destroy(&ctx->flush_queue);
 160
 161         util_copy_framebuffer_state(&ctx->framebuffer, NULL);
 162         fd_batch_reference(&ctx->batch, NULL);  /* unref current batch */
 163         fd_bc_invalidate_context(ctx);
 164
 165         fd_prog_fini(pctx);
 166
 167         if (ctx->blitter)
 168                 util_blitter_destroy(ctx->blitter);
 169
 170         if (pctx->stream_uploader)
 171                 u_upload_destroy(pctx->stream_uploader);
 172
 173         if (ctx->clear_rs_state)
 174                 pctx->delete_rasterizer_state(pctx, ctx->clear_rs_state);
 175
 176         if (ctx->primconvert)
 177                 util_primconvert_destroy(ctx->primconvert);
 178
 179         slab_destroy_child(&ctx->transfer_pool);
 180
 181         for (i = 0; i < ARRAY_SIZE(ctx->vsc_pipe); i++) {
 182                 struct fd_vsc_pipe *pipe = &ctx->vsc_pipe[i];
 183                 if (!pipe->bo)
 184                         break;
 185                 fd_bo_del(pipe->bo);
 186         }
 187
 188         fd_device_del(ctx->dev);
 189         fd_pipe_del(ctx->pipe);
 190
 191         if (fd_mesa_debug & (FD_DBG_BSTAT | FD_DBG_MSGS)) {
 192                 printf("batch_total=%u, batch_sysmem=%u, batch_gmem=%u, batch_nondraw=%u, batch_restore=%u\n",
 193                         (uint32_t)ctx->stats.batch_total, (uint32_t)ctx->stats.batch_sysmem,
 194                         (uint32_t)ctx->stats.batch_gmem, (uint32_t)ctx->stats.batch_nondraw,
 195                         (uint32_t)ctx->stats.batch_restore);
 196         }
 197 }
 198
 199 static void
 200 fd_set_debug_callback(struct pipe_context *pctx,
 201                 const struct pipe_debug_callback *cb)
 202 {
 203         struct fd_context *ctx = fd_context(pctx);
 204
 205         if (cb)
 206                 ctx->debug = *cb;
 207         else
 208                 memset(&ctx->debug, 0, sizeof(ctx->debug));
 209 }
 210
 211 /* TODO we could combine a few of these small buffers (solid_vbuf,
 212  * blit_texcoord_vbuf, and vsc_size_mem, into a single buffer and
 213  * save a tiny bit of memory
 214  */
 215
 216 static struct pipe_resource *
 217 create_solid_vertexbuf(struct pipe_context *pctx)
 218 {
 219         static const float init_shader_const[] = {
 220                         -1.000000, +1.000000, +1.000000,
 221                         +1.000000, -1.000000, +1.000000,
 222         };
 223         struct pipe_resource *prsc = pipe_buffer_create(pctx->screen,
 224                         PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE, sizeof(init_shader_const));
 225         pipe_buffer_write(pctx, prsc, 0,
 226                         sizeof(init_shader_const), init_shader_const);
 227         return prsc;
 228 }
 229
 230 static struct pipe_resource *
 231 create_blit_texcoord_vertexbuf(struct pipe_context *pctx)
 232 {
 233         struct pipe_resource *prsc = pipe_buffer_create(pctx->screen,
 234                         PIPE_BIND_CUSTOM, PIPE_USAGE_DYNAMIC, 16);
 235         return prsc;
 236 }
 237
 238 void
 239 fd_context_setup_common_vbos(struct fd_context *ctx)
 240 {
 241         struct pipe_context *pctx = &ctx->base;
 242
 243         ctx->solid_vbuf = create_solid_vertexbuf(pctx);
 244         ctx->blit_texcoord_vbuf = create_blit_texcoord_vertexbuf(pctx);
 245
 246         /* setup solid_vbuf_state: */
 247         ctx->solid_vbuf_state.vtx = pctx->create_vertex_elements_state(
 248                         pctx, 1, (struct pipe_vertex_element[]){{
 249                                 .vertex_buffer_index = 0,
 250                                 .src_offset = 0,
 251                                 .src_format = PIPE_FORMAT_R32G32B32_FLOAT,
 252                         }});
 253         ctx->solid_vbuf_state.vertexbuf.count = 1;
 254         ctx->solid_vbuf_state.vertexbuf.vb[0].stride = 12;
 255         ctx->solid_vbuf_state.vertexbuf.vb[0].buffer.resource = ctx->solid_vbuf;
 256
 257         /* setup blit_vbuf_state: */
 258         ctx->blit_vbuf_state.vtx = pctx->create_vertex_elements_state(
 259                         pctx, 2, (struct pipe_vertex_element[]){{
 260                                 .vertex_buffer_index = 0,
 261                                 .src_offset = 0,
 262                                 .src_format = PIPE_FORMAT_R32G32_FLOAT,
 263                         }, {
 264                                 .vertex_buffer_index = 1,
 265                                 .src_offset = 0,
 266                                 .src_format = PIPE_FORMAT_R32G32B32_FLOAT,
 267                         }});
 268         ctx->blit_vbuf_state.vertexbuf.count = 2;
 269         ctx->blit_vbuf_state.vertexbuf.vb[0].stride = 8;
 270         ctx->blit_vbuf_state.vertexbuf.vb[0].buffer.resource = ctx->blit_texcoord_vbuf;
 271         ctx->blit_vbuf_state.vertexbuf.vb[1].stride = 12;
 272         ctx->blit_vbuf_state.vertexbuf.vb[1].buffer.resource = ctx->solid_vbuf;
 273 }
 274
 275 void
 276 fd_context_cleanup_common_vbos(struct fd_context *ctx)
 277 {
 278         struct pipe_context *pctx = &ctx->base;
 279
 280         pctx->delete_vertex_elements_state(pctx, ctx->solid_vbuf_state.vtx);
 281         pctx->delete_vertex_elements_state(pctx, ctx->blit_vbuf_state.vtx);
 282
 283         pipe_resource_reference(&ctx->solid_vbuf, NULL);
 284         pipe_resource_reference(&ctx->blit_texcoord_vbuf, NULL);
 285 }
 286
 287 struct pipe_context *
 288 fd_context_init(struct fd_context *ctx, struct pipe_screen *pscreen,
 289                 const uint8_t *primtypes, void *priv, unsigned flags)
 290 {
 291         struct fd_screen *screen = fd_screen(pscreen);
 292         struct pipe_context *pctx;
 293         unsigned prio = 1;
 294         int i;
 295
 296         /* lower numerical value == higher priority: */
 297         if (fd_mesa_debug & FD_DBG_HIPRIO)
 298                 prio = 0;
 299         else if (flags & PIPE_CONTEXT_HIGH_PRIORITY)
 300                 prio = 0;
 301         else if (flags & PIPE_CONTEXT_LOW_PRIORITY)
 302                 prio = 2;
 303
 304         ctx->screen = screen;
 305         ctx->pipe = fd_pipe_new2(screen->dev, FD_PIPE_3D, prio);
 306
 307         ctx->primtypes = primtypes;
 308         ctx->primtype_mask = 0;
 309         for (i = 0; i < PIPE_PRIM_MAX; i++)
 310                 if (primtypes[i])
 311                         ctx->primtype_mask |= (1 << i);
 312
 313         /* need some sane default in case state tracker doesn't
 314          * set some state:
 315          */
 316         ctx->sample_mask = 0xffff;
 317
 318         pctx = &ctx->base;
 319         pctx->screen = pscreen;
 320         pctx->priv = priv;
 321         pctx->flush = fd_context_flush;
 322         pctx->emit_string_marker = fd_emit_string_marker;
 323         pctx->set_debug_callback = fd_set_debug_callback;
 324         pctx->create_fence_fd = fd_create_fence_fd;
 325         pctx->fence_server_sync = fd_fence_server_sync;
 326         pctx->texture_barrier = fd_texture_barrier;
 327         pctx->memory_barrier = fd_memory_barrier;
 328
 329         pctx->stream_uploader = u_upload_create_default(pctx);
 330         if (!pctx->stream_uploader)
 331                 goto fail;
 332         pctx->const_uploader = pctx->stream_uploader;
 333
 334         if (!ctx->screen->reorder)
 335                 ctx->batch = fd_bc_alloc_batch(&screen->batch_cache, ctx, false);
 336
 337         slab_create_child(&ctx->transfer_pool, &screen->transfer_pool);
 338
 339         fd_draw_init(pctx);
 340         fd_resource_context_init(pctx);
 341         fd_query_context_init(pctx);
 342         fd_texture_init(pctx);
 343         fd_state_init(pctx);
 344
 345         ctx->blitter = util_blitter_create(pctx);
 346         if (!ctx->blitter)
 347                 goto fail;
 348
 349         ctx->primconvert = util_primconvert_create(pctx, ctx->primtype_mask);
 350         if (!ctx->primconvert)
 351                 goto fail;
 352
 353         list_inithead(&ctx->hw_active_queries);
 354         list_inithead(&ctx->acc_active_queries);
 355
 356         return pctx;
 357
 358 fail:
 359         pctx->destroy(pctx);
 360         return NULL;
 361 }