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
  41 static void
  42 fd_context_flush(struct pipe_context *pctx, struct pipe_fence_handle **fence,
  43                 unsigned flags)
  44 {
  45         struct fd_context *ctx = fd_context(pctx);
  46
  47         if (flags & PIPE_FLUSH_FENCE_FD)
  48                 ctx->batch->needs_out_fence_fd = true;
  49
  50         if (!ctx->screen->reorder) {
  51                 fd_batch_flush(ctx->batch, true);
  52         } else {
  53                 fd_bc_flush(&ctx->screen->batch_cache, ctx);
  54         }
  55
  56         if (fence) {
  57                 /* if there hasn't been any rendering submitted yet, we might not
  58                  * have actually created a fence
  59                  */
  60                 if (!ctx->last_fence || ctx->batch->needs_out_fence_fd) {
  61                         ctx->batch->needs_flush = true;
  62                         fd_gmem_render_noop(ctx->batch);
  63                         fd_batch_reset(ctx->batch);
  64                 }
  65                 fd_fence_ref(pctx->screen, fence, ctx->last_fence);
  66         }
  67 }
  68
  69 /**
  70  * emit marker string as payload of a no-op packet, which can be
  71  * decoded by cffdump.
  72  */
  73 static void
  74 fd_emit_string_marker(struct pipe_context *pctx, const char *string, int len)
  75 {
  76         struct fd_context *ctx = fd_context(pctx);
  77         struct fd_ringbuffer *ring;
  78         const uint32_t *buf = (const void *)string;
  79
  80         if (!ctx->batch)
  81                 return;
  82
  83         ring = ctx->batch->draw;
  84
  85         /* max packet size is 0x3fff dwords: */
  86         len = MIN2(len, 0x3fff * 4);
  87
  88         if (ctx->screen->gpu_id >= 500)
  89                 OUT_PKT7(ring, CP_NOP, align(len, 4) / 4);
  90         else
  91                 OUT_PKT3(ring, CP_NOP, align(len, 4) / 4);
  92         while (len >= 4) {
  93                 OUT_RING(ring, *buf);
  94                 buf++;
  95                 len -= 4;
  96         }
  97
  98         /* copy remainder bytes without reading past end of input string: */
  99         if (len > 0) {
 100                 uint32_t w = 0;
 101                 memcpy(&w, buf, len);
 102                 OUT_RING(ring, w);
 103         }
 104 }
 105
 106 void
 107 fd_context_destroy(struct pipe_context *pctx)
 108 {
 109         struct fd_context *ctx = fd_context(pctx);
 110         unsigned i;
 111
 112         DBG("");
 113
 114         if (ctx->screen->reorder)
 115                 util_queue_destroy(&ctx->flush_queue);
 116
 117         fd_batch_reference(&ctx->batch, NULL);  /* unref current batch */
 118         fd_bc_invalidate_context(ctx);
 119
 120         fd_fence_ref(pctx->screen, &ctx->last_fence, NULL);
 121
 122         fd_prog_fini(pctx);
 123         fd_hw_query_fini(pctx);
 124
 125         if (ctx->blitter)
 126                 util_blitter_destroy(ctx->blitter);
 127
 128         if (ctx->clear_rs_state)
 129                 pctx->delete_rasterizer_state(pctx, ctx->clear_rs_state);
 130
 131         if (ctx->primconvert)
 132                 util_primconvert_destroy(ctx->primconvert);
 133
 134         slab_destroy_child(&ctx->transfer_pool);
 135
 136         for (i = 0; i < ARRAY_SIZE(ctx->pipe); i++) {
 137                 struct fd_vsc_pipe *pipe = &ctx->pipe[i];
 138                 if (!pipe->bo)
 139                         break;
 140                 fd_bo_del(pipe->bo);
 141         }
 142
 143         fd_device_del(ctx->dev);
 144
 145         if (fd_mesa_debug & (FD_DBG_BSTAT | FD_DBG_MSGS)) {
 146                 printf("batch_total=%u, batch_sysmem=%u, batch_gmem=%u, batch_restore=%u\n",
 147                         (uint32_t)ctx->stats.batch_total, (uint32_t)ctx->stats.batch_sysmem,
 148                         (uint32_t)ctx->stats.batch_gmem, (uint32_t)ctx->stats.batch_restore);
 149         }
 150
 151         FREE(ctx);
 152 }
 153
 154 static void
 155 fd_set_debug_callback(struct pipe_context *pctx,
 156                 const struct pipe_debug_callback *cb)
 157 {
 158         struct fd_context *ctx = fd_context(pctx);
 159
 160         if (cb)
 161                 ctx->debug = *cb;
 162         else
 163                 memset(&ctx->debug, 0, sizeof(ctx->debug));
 164 }
 165
 166 /* TODO we could combine a few of these small buffers (solid_vbuf,
 167  * blit_texcoord_vbuf, and vsc_size_mem, into a single buffer and
 168  * save a tiny bit of memory
 169  */
 170
 171 static struct pipe_resource *
 172 create_solid_vertexbuf(struct pipe_context *pctx)
 173 {
 174         static const float init_shader_const[] = {
 175                         -1.000000, +1.000000, +1.000000,
 176                         +1.000000, -1.000000, +1.000000,
 177         };
 178         struct pipe_resource *prsc = pipe_buffer_create(pctx->screen,
 179                         PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE, sizeof(init_shader_const));
 180         pipe_buffer_write(pctx, prsc, 0,
 181                         sizeof(init_shader_const), init_shader_const);
 182         return prsc;
 183 }
 184
 185 static struct pipe_resource *
 186 create_blit_texcoord_vertexbuf(struct pipe_context *pctx)
 187 {
 188         struct pipe_resource *prsc = pipe_buffer_create(pctx->screen,
 189                         PIPE_BIND_CUSTOM, PIPE_USAGE_DYNAMIC, 16);
 190         return prsc;
 191 }
 192
 193 void
 194 fd_context_setup_common_vbos(struct fd_context *ctx)
 195 {
 196         struct pipe_context *pctx = &ctx->base;
 197
 198         ctx->solid_vbuf = create_solid_vertexbuf(pctx);
 199         ctx->blit_texcoord_vbuf = create_blit_texcoord_vertexbuf(pctx);
 200
 201         /* setup solid_vbuf_state: */
 202         ctx->solid_vbuf_state.vtx = pctx->create_vertex_elements_state(
 203                         pctx, 1, (struct pipe_vertex_element[]){{
 204                                 .vertex_buffer_index = 0,
 205                                 .src_offset = 0,
 206                                 .src_format = PIPE_FORMAT_R32G32B32_FLOAT,
 207                         }});
 208         ctx->solid_vbuf_state.vertexbuf.count = 1;
 209         ctx->solid_vbuf_state.vertexbuf.vb[0].stride = 12;
 210         ctx->solid_vbuf_state.vertexbuf.vb[0].buffer = ctx->solid_vbuf;
 211
 212         /* setup blit_vbuf_state: */
 213         ctx->blit_vbuf_state.vtx = pctx->create_vertex_elements_state(
 214                         pctx, 2, (struct pipe_vertex_element[]){{
 215                                 .vertex_buffer_index = 0,
 216                                 .src_offset = 0,
 217                                 .src_format = PIPE_FORMAT_R32G32_FLOAT,
 218                         }, {
 219                                 .vertex_buffer_index = 1,
 220                                 .src_offset = 0,
 221                                 .src_format = PIPE_FORMAT_R32G32B32_FLOAT,
 222                         }});
 223         ctx->blit_vbuf_state.vertexbuf.count = 2;
 224         ctx->blit_vbuf_state.vertexbuf.vb[0].stride = 8;
 225         ctx->blit_vbuf_state.vertexbuf.vb[0].buffer = ctx->blit_texcoord_vbuf;
 226         ctx->blit_vbuf_state.vertexbuf.vb[1].stride = 12;
 227         ctx->blit_vbuf_state.vertexbuf.vb[1].buffer = ctx->solid_vbuf;
 228 }
 229
 230 void
 231 fd_context_cleanup_common_vbos(struct fd_context *ctx)
 232 {
 233         struct pipe_context *pctx = &ctx->base;
 234
 235         pctx->delete_vertex_elements_state(pctx, ctx->solid_vbuf_state.vtx);
 236         pctx->delete_vertex_elements_state(pctx, ctx->blit_vbuf_state.vtx);
 237
 238         pipe_resource_reference(&ctx->solid_vbuf, NULL);
 239         pipe_resource_reference(&ctx->blit_texcoord_vbuf, NULL);
 240 }
 241
 242 struct pipe_context *
 243 fd_context_init(struct fd_context *ctx, struct pipe_screen *pscreen,
 244                 const uint8_t *primtypes, void *priv)
 245 {
 246         struct fd_screen *screen = fd_screen(pscreen);
 247         struct pipe_context *pctx;
 248         int i;
 249
 250         ctx->screen = screen;
 251
 252         ctx->primtypes = primtypes;
 253         ctx->primtype_mask = 0;
 254         for (i = 0; i < PIPE_PRIM_MAX; i++)
 255                 if (primtypes[i])
 256                         ctx->primtype_mask |= (1 << i);
 257
 258         /* need some sane default in case state tracker doesn't
 259          * set some state:
 260          */
 261         ctx->sample_mask = 0xffff;
 262
 263         pctx = &ctx->base;
 264         pctx->screen = pscreen;
 265         pctx->priv = priv;
 266         pctx->flush = fd_context_flush;
 267         pctx->emit_string_marker = fd_emit_string_marker;
 268         pctx->set_debug_callback = fd_set_debug_callback;
 269         pctx->create_fence_fd = fd_create_fence_fd;
 270         pctx->fence_server_sync = fd_fence_server_sync;
 271
 272         /* TODO what about compute?  Ideally it creates it's own independent
 273          * batches per compute job (since it isn't using tiling, so no point
 274          * in getting involved with the re-ordering madness)..
 275          */
 276         if (!screen->reorder) {
 277                 ctx->batch = fd_bc_alloc_batch(&screen->batch_cache, ctx);
 278         }
 279
 280         slab_create_child(&ctx->transfer_pool, &screen->transfer_pool);
 281
 282         fd_draw_init(pctx);
 283         fd_resource_context_init(pctx);
 284         fd_query_context_init(pctx);
 285         fd_texture_init(pctx);
 286         fd_state_init(pctx);
 287         fd_hw_query_init(pctx);
 288
 289         ctx->blitter = util_blitter_create(pctx);
 290         if (!ctx->blitter)
 291                 goto fail;
 292
 293         ctx->primconvert = util_primconvert_create(pctx, ctx->primtype_mask);
 294         if (!ctx->primconvert)
 295                 goto fail;
 296
 297         return pctx;
 298
 299 fail:
 300         pctx->destroy(pctx);
 301         return NULL;
 302 }