src/gallium/drivers/freedreno/freedreno_context.c

   1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
   2
   3 /*
   4  * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the next
  14  * paragraph) shall be included in all copies or substantial portions of the
  15  * Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  23  * SOFTWARE.
  24  *
  25  * Authors:
  26  *    Rob Clark <robclark@freedesktop.org>
  27  */
  28
  29 #include "freedreno_context.h"
  30 #include "freedreno_draw.h"
  31 #include "freedreno_fence.h"
  32 #include "freedreno_program.h"
  33 #include "freedreno_resource.h"
  34 #include "freedreno_texture.h"
  35 #include "freedreno_state.h"
  36 #include "freedreno_gmem.h"
  37 #include "freedreno_query.h"
  38 #include "freedreno_query_hw.h"
  39 #include "freedreno_util.h"
  40 #include "util/u_upload_mgr.h"
  41
  42 static void
  43 fd_context_flush(struct pipe_context *pctx, struct pipe_fence_handle **fence,
  44                 unsigned flags)
  45 {
  46         struct fd_context *ctx = fd_context(pctx);
  47
  48         if (flags & PIPE_FLUSH_FENCE_FD)
  49                 ctx->batch->needs_out_fence_fd = true;
  50
  51         if (!ctx->screen->reorder) {
  52                 fd_batch_flush(ctx->batch, true);
  53         } else {
  54                 fd_bc_flush(&ctx->screen->batch_cache, ctx);
  55         }
  56
  57         if (fence) {
  58                 /* if there hasn't been any rendering submitted yet, we might not
  59                  * have actually created a fence
  60                  */
  61                 if (!ctx->last_fence || ctx->batch->needs_out_fence_fd) {
  62                         ctx->batch->needs_flush = true;
  63                         fd_gmem_render_noop(ctx->batch);
  64                         fd_batch_reset(ctx->batch);
  65                 }
  66                 fd_fence_ref(pctx->screen, fence, ctx->last_fence);
  67         }
  68 }
  69
  70 static void
  71 fd_texture_barrier(struct pipe_context *pctx, unsigned flags)
  72 {
  73         /* On devices that could sample from GMEM we could possibly do better.
  74          * Or if we knew that we were doing GMEM bypass we could just emit a
  75          * cache flush, perhaps?  But we don't know if future draws would cause
  76          * us to use GMEM, and a flush in bypass isn't the end of the world.
  77          */
  78         fd_context_flush(pctx, NULL, 0);
  79 }
  80
  81 /**
  82  * emit marker string as payload of a no-op packet, which can be
  83  * decoded by cffdump.
  84  */
  85 static void
  86 fd_emit_string_marker(struct pipe_context *pctx, const char *string, int len)
  87 {
  88         struct fd_context *ctx = fd_context(pctx);
  89         struct fd_ringbuffer *ring;
  90         const uint32_t *buf = (const void *)string;
  91
  92         if (!ctx->batch)
  93                 return;
  94
  95         ring = ctx->batch->draw;
  96
  97         /* max packet size is 0x3fff dwords: */
  98         len = MIN2(len, 0x3fff * 4);
  99
 100         if (ctx->screen->gpu_id >= 500)
 101                 OUT_PKT7(ring, CP_NOP, align(len, 4) / 4);
 102         else
 103                 OUT_PKT3(ring, CP_NOP, align(len, 4) / 4);
 104         while (len >= 4) {
 105                 OUT_RING(ring, *buf);
 106                 buf++;
 107                 len -= 4;
 108         }
 109
 110         /* copy remainder bytes without reading past end of input string: */
 111         if (len > 0) {
 112                 uint32_t w = 0;
 113                 memcpy(&w, buf, len);
 114                 OUT_RING(ring, w);
 115         }
 116 }
 117
 118 void
 119 fd_context_destroy(struct pipe_context *pctx)
 120 {
 121         struct fd_context *ctx = fd_context(pctx);
 122         unsigned i;
 123
 124         DBG("");
 125
 126         if (ctx->screen->reorder && util_queue_is_initialized(&ctx->flush_queue))
 127                 util_queue_destroy(&ctx->flush_queue);
 128
 129         fd_batch_reference(&ctx->batch, NULL);  /* unref current batch */
 130         fd_bc_invalidate_context(ctx);
 131
 132         fd_fence_ref(pctx->screen, &ctx->last_fence, NULL);
 133
 134         fd_prog_fini(pctx);
 135
 136         if (ctx->blitter)
 137                 util_blitter_destroy(ctx->blitter);
 138
 139         if (pctx->stream_uploader)
 140                 u_upload_destroy(pctx->stream_uploader);
 141
 142         if (ctx->clear_rs_state)
 143                 pctx->delete_rasterizer_state(pctx, ctx->clear_rs_state);
 144
 145         if (ctx->primconvert)
 146                 util_primconvert_destroy(ctx->primconvert);
 147
 148         slab_destroy_child(&ctx->transfer_pool);
 149
 150         for (i = 0; i < ARRAY_SIZE(ctx->vsc_pipe); i++) {
 151                 struct fd_vsc_pipe *pipe = &ctx->vsc_pipe[i];
 152                 if (!pipe->bo)
 153                         break;
 154                 fd_bo_del(pipe->bo);
 155         }
 156
 157         fd_device_del(ctx->dev);
 158         fd_pipe_del(ctx->pipe);
 159
 160         if (fd_mesa_debug & (FD_DBG_BSTAT | FD_DBG_MSGS)) {
 161                 printf("batch_total=%u, batch_sysmem=%u, batch_gmem=%u, batch_restore=%u\n",
 162                         (uint32_t)ctx->stats.batch_total, (uint32_t)ctx->stats.batch_sysmem,
 163                         (uint32_t)ctx->stats.batch_gmem, (uint32_t)ctx->stats.batch_restore);
 164         }
 165
 166         FREE(ctx);
 167 }
 168
 169 static void
 170 fd_set_debug_callback(struct pipe_context *pctx,
 171                 const struct pipe_debug_callback *cb)
 172 {
 173         struct fd_context *ctx = fd_context(pctx);
 174
 175         if (cb)
 176                 ctx->debug = *cb;
 177         else
 178                 memset(&ctx->debug, 0, sizeof(ctx->debug));
 179 }
 180
 181 /* TODO we could combine a few of these small buffers (solid_vbuf,
 182  * blit_texcoord_vbuf, and vsc_size_mem, into a single buffer and
 183  * save a tiny bit of memory
 184  */
 185
 186 static struct pipe_resource *
 187 create_solid_vertexbuf(struct pipe_context *pctx)
 188 {
 189         static const float init_shader_const[] = {
 190                         -1.000000, +1.000000, +1.000000,
 191                         +1.000000, -1.000000, +1.000000,
 192         };
 193         struct pipe_resource *prsc = pipe_buffer_create(pctx->screen,
 194                         PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE, sizeof(init_shader_const));
 195         pipe_buffer_write(pctx, prsc, 0,
 196                         sizeof(init_shader_const), init_shader_const);
 197         return prsc;
 198 }
 199
 200 static struct pipe_resource *
 201 create_blit_texcoord_vertexbuf(struct pipe_context *pctx)
 202 {
 203         struct pipe_resource *prsc = pipe_buffer_create(pctx->screen,
 204                         PIPE_BIND_CUSTOM, PIPE_USAGE_DYNAMIC, 16);
 205         return prsc;
 206 }
 207
 208 void
 209 fd_context_setup_common_vbos(struct fd_context *ctx)
 210 {
 211         struct pipe_context *pctx = &ctx->base;
 212
 213         ctx->solid_vbuf = create_solid_vertexbuf(pctx);
 214         ctx->blit_texcoord_vbuf = create_blit_texcoord_vertexbuf(pctx);
 215
 216         /* setup solid_vbuf_state: */
 217         ctx->solid_vbuf_state.vtx = pctx->create_vertex_elements_state(
 218                         pctx, 1, (struct pipe_vertex_element[]){{
 219                                 .vertex_buffer_index = 0,
 220                                 .src_offset = 0,
 221                                 .src_format = PIPE_FORMAT_R32G32B32_FLOAT,
 222                         }});
 223         ctx->solid_vbuf_state.vertexbuf.count = 1;
 224         ctx->solid_vbuf_state.vertexbuf.vb[0].stride = 12;
 225         ctx->solid_vbuf_state.vertexbuf.vb[0].buffer.resource = ctx->solid_vbuf;
 226
 227         /* setup blit_vbuf_state: */
 228         ctx->blit_vbuf_state.vtx = pctx->create_vertex_elements_state(
 229                         pctx, 2, (struct pipe_vertex_element[]){{
 230                                 .vertex_buffer_index = 0,
 231                                 .src_offset = 0,
 232                                 .src_format = PIPE_FORMAT_R32G32_FLOAT,
 233                         }, {
 234                                 .vertex_buffer_index = 1,
 235                                 .src_offset = 0,
 236                                 .src_format = PIPE_FORMAT_R32G32B32_FLOAT,
 237                         }});
 238         ctx->blit_vbuf_state.vertexbuf.count = 2;
 239         ctx->blit_vbuf_state.vertexbuf.vb[0].stride = 8;
 240         ctx->blit_vbuf_state.vertexbuf.vb[0].buffer.resource = ctx->blit_texcoord_vbuf;
 241         ctx->blit_vbuf_state.vertexbuf.vb[1].stride = 12;
 242         ctx->blit_vbuf_state.vertexbuf.vb[1].buffer.resource = ctx->solid_vbuf;
 243 }
 244
 245 void
 246 fd_context_cleanup_common_vbos(struct fd_context *ctx)
 247 {
 248         struct pipe_context *pctx = &ctx->base;
 249
 250         pctx->delete_vertex_elements_state(pctx, ctx->solid_vbuf_state.vtx);
 251         pctx->delete_vertex_elements_state(pctx, ctx->blit_vbuf_state.vtx);
 252
 253         pipe_resource_reference(&ctx->solid_vbuf, NULL);
 254         pipe_resource_reference(&ctx->blit_texcoord_vbuf, NULL);
 255 }
 256
 257 struct pipe_context *
 258 fd_context_init(struct fd_context *ctx, struct pipe_screen *pscreen,
 259                 const uint8_t *primtypes, void *priv, unsigned flags)
 260 {
 261         struct fd_screen *screen = fd_screen(pscreen);
 262         struct pipe_context *pctx;
 263         int i;
 264
 265         ctx->screen = screen;
 266         ctx->pipe = fd_pipe_new(screen->dev, FD_PIPE_3D);
 267
 268         ctx->primtypes = primtypes;
 269         ctx->primtype_mask = 0;
 270         for (i = 0; i < PIPE_PRIM_MAX; i++)
 271                 if (primtypes[i])
 272                         ctx->primtype_mask |= (1 << i);
 273
 274         /* need some sane default in case state tracker doesn't
 275          * set some state:
 276          */
 277         ctx->sample_mask = 0xffff;
 278
 279         pctx = &ctx->base;
 280         pctx->screen = pscreen;
 281         pctx->priv = priv;
 282         pctx->flush = fd_context_flush;
 283         pctx->emit_string_marker = fd_emit_string_marker;
 284         pctx->set_debug_callback = fd_set_debug_callback;
 285         pctx->create_fence_fd = fd_create_fence_fd;
 286         pctx->fence_server_sync = fd_fence_server_sync;
 287         pctx->texture_barrier = fd_texture_barrier;
 288
 289         pctx->stream_uploader = u_upload_create_default(pctx);
 290         if (!pctx->stream_uploader)
 291                 goto fail;
 292         pctx->const_uploader = pctx->stream_uploader;
 293
 294         ctx->batch = fd_bc_alloc_batch(&screen->batch_cache, ctx);
 295
 296         slab_create_child(&ctx->transfer_pool, &screen->transfer_pool);
 297
 298         fd_draw_init(pctx);
 299         fd_resource_context_init(pctx);
 300         fd_query_context_init(pctx);
 301         fd_texture_init(pctx);
 302         fd_state_init(pctx);
 303
 304         ctx->blitter = util_blitter_create(pctx);
 305         if (!ctx->blitter)
 306                 goto fail;
 307
 308         ctx->primconvert = util_primconvert_create(pctx, ctx->primtype_mask);
 309         if (!ctx->primconvert)
 310                 goto fail;
 311
 312         list_inithead(&ctx->hw_active_queries);
 313         list_inithead(&ctx->acc_active_queries);
 314
 315         return pctx;
 316
 317 fail:
 318         pctx->destroy(pctx);
 319         return NULL;
 320 }