src/gallium/drivers/freedreno/freedreno_draw.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 "pipe/p_state.h"
  30 #include "util/u_string.h"
  31 #include "util/u_memory.h"
  32 #include "util/u_prim.h"
  33 #include "util/u_format.h"
  34
  35 #include "freedreno_draw.h"
  36 #include "freedreno_context.h"
  37 #include "freedreno_state.h"
  38 #include "freedreno_resource.h"
  39 #include "freedreno_query_hw.h"
  40 #include "freedreno_util.h"
  41
  42 static void
  43 resource_read(struct fd_context *ctx, struct pipe_resource *prsc)
  44 {
  45         if (!prsc)
  46                 return;
  47         fd_batch_resource_used(ctx->batch, fd_resource(prsc), FD_PENDING_READ);
  48 }
  49
  50 static void
  51 resource_written(struct fd_context *ctx, struct pipe_resource *prsc)
  52 {
  53         if (!prsc)
  54                 return;
  55         fd_batch_resource_used(ctx->batch, fd_resource(prsc), FD_PENDING_WRITE);
  56 }
  57
  58 static void
  59 fd_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
  60 {
  61         struct fd_context *ctx = fd_context(pctx);
  62         struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
  63         struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
  64         unsigned i, prims, buffers = 0;
  65
  66         /* if we supported transform feedback, we'd have to disable this: */
  67         if (((scissor->maxx - scissor->minx) *
  68                         (scissor->maxy - scissor->miny)) == 0) {
  69                 return;
  70         }
  71
  72         /* TODO: push down the region versions into the tiles */
  73         if (!fd_render_condition_check(pctx))
  74                 return;
  75
  76         /* emulate unsupported primitives: */
  77         if (!fd_supported_prim(ctx, info->mode)) {
  78                 if (ctx->streamout.num_targets > 0)
  79                         debug_error("stream-out with emulated prims");
  80                 util_primconvert_save_index_buffer(ctx->primconvert, &ctx->indexbuf);
  81                 util_primconvert_save_rasterizer_state(ctx->primconvert, ctx->rasterizer);
  82                 util_primconvert_draw_vbo(ctx->primconvert, info);
  83                 return;
  84         }
  85
  86         /*
  87          * Figure out the buffers/features we need:
  88          */
  89
  90         if (fd_depth_enabled(ctx)) {
  91                 buffers |= FD_BUFFER_DEPTH;
  92                 resource_written(ctx, pfb->zsbuf->texture);
  93                 ctx->gmem_reason |= FD_GMEM_DEPTH_ENABLED;
  94         }
  95
  96         if (fd_stencil_enabled(ctx)) {
  97                 buffers |= FD_BUFFER_STENCIL;
  98                 resource_written(ctx, pfb->zsbuf->texture);
  99                 ctx->gmem_reason |= FD_GMEM_STENCIL_ENABLED;
 100         }
 101
 102         if (fd_logicop_enabled(ctx))
 103                 ctx->gmem_reason |= FD_GMEM_LOGICOP_ENABLED;
 104
 105         for (i = 0; i < pfb->nr_cbufs; i++) {
 106                 struct pipe_resource *surf;
 107
 108                 if (!pfb->cbufs[i])
 109                         continue;
 110
 111                 surf = pfb->cbufs[i]->texture;
 112
 113                 resource_written(ctx, surf);
 114                 buffers |= PIPE_CLEAR_COLOR0 << i;
 115
 116                 if (surf->nr_samples > 1)
 117                         ctx->gmem_reason |= FD_GMEM_MSAA_ENABLED;
 118
 119                 if (fd_blend_enabled(ctx, i))
 120                         ctx->gmem_reason |= FD_GMEM_BLEND_ENABLED;
 121         }
 122
 123         /* Skip over buffer 0, that is sent along with the command stream */
 124         for (i = 1; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
 125                 resource_read(ctx, ctx->constbuf[PIPE_SHADER_VERTEX].cb[i].buffer);
 126                 resource_read(ctx, ctx->constbuf[PIPE_SHADER_FRAGMENT].cb[i].buffer);
 127         }
 128
 129         /* Mark VBOs as being read */
 130         for (i = 0; i < ctx->vtx.vertexbuf.count; i++) {
 131                 assert(!ctx->vtx.vertexbuf.vb[i].user_buffer);
 132                 resource_read(ctx, ctx->vtx.vertexbuf.vb[i].buffer);
 133         }
 134
 135         /* Mark index buffer as being read */
 136         resource_read(ctx, ctx->indexbuf.buffer);
 137
 138         /* Mark textures as being read */
 139         for (i = 0; i < ctx->verttex.num_textures; i++)
 140                 if (ctx->verttex.textures[i])
 141                         resource_read(ctx, ctx->verttex.textures[i]->texture);
 142         for (i = 0; i < ctx->fragtex.num_textures; i++)
 143                 if (ctx->fragtex.textures[i])
 144                         resource_read(ctx, ctx->fragtex.textures[i]->texture);
 145
 146         /* Mark streamout buffers as being written.. */
 147         for (i = 0; i < ctx->streamout.num_targets; i++)
 148                 if (ctx->streamout.targets[i])
 149                         resource_written(ctx, ctx->streamout.targets[i]->buffer);
 150
 151         ctx->num_draws++;
 152
 153         prims = u_reduced_prims_for_vertices(info->mode, info->count);
 154
 155         ctx->stats.draw_calls++;
 156
 157         /* TODO prims_emitted should be clipped when the stream-out buffer is
 158          * not large enough.  See max_tf_vtx().. probably need to move that
 159          * into common code.  Although a bit more annoying since a2xx doesn't
 160          * use ir3 so no common way to get at the pipe_stream_output_info
 161          * which is needed for this calculation.
 162          */
 163         if (ctx->streamout.num_targets > 0)
 164                 ctx->stats.prims_emitted += prims;
 165         ctx->stats.prims_generated += prims;
 166
 167         /* any buffers that haven't been cleared yet, we need to restore: */
 168         ctx->restore |= buffers & (FD_BUFFER_ALL & ~ctx->cleared);
 169         /* and any buffers used, need to be resolved: */
 170         ctx->resolve |= buffers;
 171
 172         DBG("%x num_draws=%u (%s/%s)", buffers, ctx->num_draws,
 173                 util_format_short_name(pipe_surface_format(pfb->cbufs[0])),
 174                 util_format_short_name(pipe_surface_format(pfb->zsbuf)));
 175
 176         fd_hw_query_set_stage(ctx, ctx->batch->draw, FD_STAGE_DRAW);
 177         if (ctx->draw_vbo(ctx, info))
 178                 ctx->needs_flush = true;
 179
 180         for (i = 0; i < ctx->streamout.num_targets; i++)
 181                 ctx->streamout.offsets[i] += info->count;
 182
 183         if (fd_mesa_debug & FD_DBG_DDRAW)
 184                 ctx->dirty = 0xffffffff;
 185
 186         fd_batch_check_size(ctx->batch);
 187 }
 188
 189 /* TODO figure out how to make better use of existing state mechanism
 190  * for clear (and possibly gmem->mem / mem->gmem) so we can (a) keep
 191  * track of what state really actually changes, and (b) reduce the code
 192  * in the a2xx/a3xx parts.
 193  */
 194
 195 static void
 196 fd_clear(struct pipe_context *pctx, unsigned buffers,
 197                 const union pipe_color_union *color, double depth, unsigned stencil)
 198 {
 199         struct fd_context *ctx = fd_context(pctx);
 200         struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
 201         struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
 202         unsigned cleared_buffers;
 203         int i;
 204
 205         /* TODO: push down the region versions into the tiles */
 206         if (!fd_render_condition_check(pctx))
 207                 return;
 208
 209         /* for bookkeeping about which buffers have been cleared (and thus
 210          * can fully or partially skip mem2gmem) we need to ignore buffers
 211          * that have already had a draw, in case apps do silly things like
 212          * clear after draw (ie. if you only clear the color buffer, but
 213          * something like alpha-test causes side effects from the draw in
 214          * the depth buffer, etc)
 215          */
 216         cleared_buffers = buffers & (FD_BUFFER_ALL & ~ctx->restore);
 217
 218         /* do we have full-screen scissor? */
 219         if (!memcmp(scissor, &ctx->disabled_scissor, sizeof(*scissor))) {
 220                 ctx->cleared |= cleared_buffers;
 221         } else {
 222                 ctx->partial_cleared |= cleared_buffers;
 223                 if (cleared_buffers & PIPE_CLEAR_COLOR)
 224                         ctx->cleared_scissor.color = *scissor;
 225                 if (cleared_buffers & PIPE_CLEAR_DEPTH)
 226                         ctx->cleared_scissor.depth = *scissor;
 227                 if (cleared_buffers & PIPE_CLEAR_STENCIL)
 228                         ctx->cleared_scissor.stencil = *scissor;
 229         }
 230         ctx->resolve |= buffers;
 231         ctx->needs_flush = true;
 232
 233         if (buffers & PIPE_CLEAR_COLOR)
 234                 for (i = 0; i < pfb->nr_cbufs; i++)
 235                         if (buffers & (PIPE_CLEAR_COLOR0 << i))
 236                                 resource_written(ctx, pfb->cbufs[i]->texture);
 237
 238         if (buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) {
 239                 resource_written(ctx, pfb->zsbuf->texture);
 240                 ctx->gmem_reason |= FD_GMEM_CLEARS_DEPTH_STENCIL;
 241         }
 242
 243         DBG("%x depth=%f, stencil=%u (%s/%s)", buffers, depth, stencil,
 244                 util_format_short_name(pipe_surface_format(pfb->cbufs[0])),
 245                 util_format_short_name(pipe_surface_format(pfb->zsbuf)));
 246
 247         fd_hw_query_set_stage(ctx, ctx->batch->draw, FD_STAGE_CLEAR);
 248
 249         ctx->clear(ctx, buffers, color, depth, stencil);
 250
 251         ctx->dirty |= FD_DIRTY_ZSA |
 252                         FD_DIRTY_VIEWPORT |
 253                         FD_DIRTY_RASTERIZER |
 254                         FD_DIRTY_SAMPLE_MASK |
 255                         FD_DIRTY_PROG |
 256                         FD_DIRTY_CONSTBUF |
 257                         FD_DIRTY_BLEND |
 258                         FD_DIRTY_FRAMEBUFFER;
 259
 260         if (fd_mesa_debug & FD_DBG_DCLEAR)
 261                 ctx->dirty = 0xffffffff;
 262 }
 263
 264 static void
 265 fd_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
 266                 const union pipe_color_union *color,
 267                 unsigned x, unsigned y, unsigned w, unsigned h)
 268 {
 269         DBG("TODO: x=%u, y=%u, w=%u, h=%u", x, y, w, h);
 270 }
 271
 272 static void
 273 fd_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
 274                 unsigned buffers, double depth, unsigned stencil,
 275                 unsigned x, unsigned y, unsigned w, unsigned h)
 276 {
 277         DBG("TODO: buffers=%u, depth=%f, stencil=%u, x=%u, y=%u, w=%u, h=%u",
 278                         buffers, depth, stencil, x, y, w, h);
 279 }
 280
 281 void
 282 fd_draw_init(struct pipe_context *pctx)
 283 {
 284         pctx->draw_vbo = fd_draw_vbo;
 285         pctx->clear = fd_clear;
 286         pctx->clear_render_target = fd_clear_render_target;
 287         pctx->clear_depth_stencil = fd_clear_depth_stencil;
 288 }