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_batch *batch, struct pipe_resource *prsc)
  44 {
  45         if (!prsc)
  46                 return;
  47         fd_batch_resource_used(batch, fd_resource(prsc), false);
  48 }
  49
  50 static void
  51 resource_written(struct fd_batch *batch, struct pipe_resource *prsc)
  52 {
  53         if (!prsc)
  54                 return;
  55         fd_batch_resource_used(batch, fd_resource(prsc), true);
  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 fd_batch *batch = ctx->batch;
  63         struct pipe_framebuffer_state *pfb = &batch->framebuffer;
  64         struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
  65         unsigned i, prims, buffers = 0;
  66
  67         /* if we supported transform feedback, we'd have to disable this: */
  68         if (((scissor->maxx - scissor->minx) *
  69                         (scissor->maxy - scissor->miny)) == 0) {
  70                 return;
  71         }
  72
  73         /* TODO: push down the region versions into the tiles */
  74         if (!fd_render_condition_check(pctx))
  75                 return;
  76
  77         /* emulate unsupported primitives: */
  78         if (!fd_supported_prim(ctx, info->mode)) {
  79                 if (ctx->streamout.num_targets > 0)
  80                         debug_error("stream-out with emulated prims");
  81                 util_primconvert_save_index_buffer(ctx->primconvert, &ctx->indexbuf);
  82                 util_primconvert_save_rasterizer_state(ctx->primconvert, ctx->rasterizer);
  83                 util_primconvert_draw_vbo(ctx->primconvert, info);
  84                 return;
  85         }
  86
  87         if (ctx->discard) {
  88                 fd_batch_reset(ctx->batch);
  89                 ctx->discard = false;
  90         }
  91
  92         /* NOTE: needs to be before resource_written(batch->query_buf), otherwise
  93          * query_buf may not be created yet.
  94          */
  95         fd_hw_query_set_stage(batch, batch->draw, FD_STAGE_DRAW);
  96         /*
  97          * Figure out the buffers/features we need:
  98          */
  99
 100         if (fd_depth_enabled(ctx)) {
 101                 buffers |= FD_BUFFER_DEPTH;
 102                 resource_written(batch, pfb->zsbuf->texture);
 103                 batch->gmem_reason |= FD_GMEM_DEPTH_ENABLED;
 104         }
 105
 106         if (fd_stencil_enabled(ctx)) {
 107                 buffers |= FD_BUFFER_STENCIL;
 108                 resource_written(batch, pfb->zsbuf->texture);
 109                 batch->gmem_reason |= FD_GMEM_STENCIL_ENABLED;
 110         }
 111
 112         if (fd_logicop_enabled(ctx))
 113                 batch->gmem_reason |= FD_GMEM_LOGICOP_ENABLED;
 114
 115         for (i = 0; i < pfb->nr_cbufs; i++) {
 116                 struct pipe_resource *surf;
 117
 118                 if (!pfb->cbufs[i])
 119                         continue;
 120
 121                 surf = pfb->cbufs[i]->texture;
 122
 123                 resource_written(batch, surf);
 124                 buffers |= PIPE_CLEAR_COLOR0 << i;
 125
 126                 if (surf->nr_samples > 1)
 127                         batch->gmem_reason |= FD_GMEM_MSAA_ENABLED;
 128
 129                 if (fd_blend_enabled(ctx, i))
 130                         batch->gmem_reason |= FD_GMEM_BLEND_ENABLED;
 131         }
 132
 133         /* Skip over buffer 0, that is sent along with the command stream */
 134         for (i = 1; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
 135                 resource_read(batch, ctx->constbuf[PIPE_SHADER_VERTEX].cb[i].buffer);
 136                 resource_read(batch, ctx->constbuf[PIPE_SHADER_FRAGMENT].cb[i].buffer);
 137         }
 138
 139         /* Mark VBOs as being read */
 140         for (i = 0; i < ctx->vtx.vertexbuf.count; i++) {
 141                 assert(!ctx->vtx.vertexbuf.vb[i].user_buffer);
 142                 resource_read(batch, ctx->vtx.vertexbuf.vb[i].buffer);
 143         }
 144
 145         /* Mark index buffer as being read */
 146         resource_read(batch, ctx->indexbuf.buffer);
 147
 148         /* Mark textures as being read */
 149         for (i = 0; i < ctx->verttex.num_textures; i++)
 150                 if (ctx->verttex.textures[i])
 151                         resource_read(batch, ctx->verttex.textures[i]->texture);
 152         for (i = 0; i < ctx->fragtex.num_textures; i++)
 153                 if (ctx->fragtex.textures[i])
 154                         resource_read(batch, ctx->fragtex.textures[i]->texture);
 155
 156         /* Mark streamout buffers as being written.. */
 157         for (i = 0; i < ctx->streamout.num_targets; i++)
 158                 if (ctx->streamout.targets[i])
 159                         resource_written(batch, ctx->streamout.targets[i]->buffer);
 160
 161         resource_written(batch, batch->query_buf);
 162
 163         batch->num_draws++;
 164
 165         prims = u_reduced_prims_for_vertices(info->mode, info->count);
 166
 167         ctx->stats.draw_calls++;
 168
 169         /* TODO prims_emitted should be clipped when the stream-out buffer is
 170          * not large enough.  See max_tf_vtx().. probably need to move that
 171          * into common code.  Although a bit more annoying since a2xx doesn't
 172          * use ir3 so no common way to get at the pipe_stream_output_info
 173          * which is needed for this calculation.
 174          */
 175         if (ctx->streamout.num_targets > 0)
 176                 ctx->stats.prims_emitted += prims;
 177         ctx->stats.prims_generated += prims;
 178
 179         /* any buffers that haven't been cleared yet, we need to restore: */
 180         batch->restore |= buffers & (FD_BUFFER_ALL & ~batch->cleared);
 181         /* and any buffers used, need to be resolved: */
 182         batch->resolve |= buffers;
 183
 184         DBG("%p: %x %ux%u num_draws=%u (%s/%s)", batch, buffers,
 185                 pfb->width, pfb->height, batch->num_draws,
 186                 util_format_short_name(pipe_surface_format(pfb->cbufs[0])),
 187                 util_format_short_name(pipe_surface_format(pfb->zsbuf)));
 188
 189         if (ctx->draw_vbo(ctx, info))
 190                 batch->needs_flush = true;
 191
 192         for (i = 0; i < ctx->streamout.num_targets; i++)
 193                 ctx->streamout.offsets[i] += info->count;
 194
 195         if (fd_mesa_debug & FD_DBG_DDRAW)
 196                 ctx->dirty = 0xffffffff;
 197
 198         fd_batch_check_size(batch);
 199 }
 200
 201 /* TODO figure out how to make better use of existing state mechanism
 202  * for clear (and possibly gmem->mem / mem->gmem) so we can (a) keep
 203  * track of what state really actually changes, and (b) reduce the code
 204  * in the a2xx/a3xx parts.
 205  */
 206
 207 static void
 208 fd_clear(struct pipe_context *pctx, unsigned buffers,
 209                 const union pipe_color_union *color, double depth, unsigned stencil)
 210 {
 211         struct fd_context *ctx = fd_context(pctx);
 212         struct fd_batch *batch = ctx->batch;
 213         struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 214         struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
 215         unsigned cleared_buffers;
 216         int i;
 217
 218         /* TODO: push down the region versions into the tiles */
 219         if (!fd_render_condition_check(pctx))
 220                 return;
 221
 222         if (ctx->discard) {
 223                 fd_batch_reset(ctx->batch);
 224                 ctx->discard = false;
 225         }
 226
 227         /* for bookkeeping about which buffers have been cleared (and thus
 228          * can fully or partially skip mem2gmem) we need to ignore buffers
 229          * that have already had a draw, in case apps do silly things like
 230          * clear after draw (ie. if you only clear the color buffer, but
 231          * something like alpha-test causes side effects from the draw in
 232          * the depth buffer, etc)
 233          */
 234         cleared_buffers = buffers & (FD_BUFFER_ALL & ~batch->restore);
 235
 236         /* do we have full-screen scissor? */
 237         if (!memcmp(scissor, &ctx->disabled_scissor, sizeof(*scissor))) {
 238                 batch->cleared |= cleared_buffers;
 239         } else {
 240                 batch->partial_cleared |= cleared_buffers;
 241                 if (cleared_buffers & PIPE_CLEAR_COLOR)
 242                         batch->cleared_scissor.color = *scissor;
 243                 if (cleared_buffers & PIPE_CLEAR_DEPTH)
 244                         batch->cleared_scissor.depth = *scissor;
 245                 if (cleared_buffers & PIPE_CLEAR_STENCIL)
 246                         batch->cleared_scissor.stencil = *scissor;
 247         }
 248         batch->resolve |= buffers;
 249         batch->needs_flush = true;
 250
 251         if (buffers & PIPE_CLEAR_COLOR)
 252                 for (i = 0; i < pfb->nr_cbufs; i++)
 253                         if (buffers & (PIPE_CLEAR_COLOR0 << i))
 254                                 resource_written(batch, pfb->cbufs[i]->texture);
 255
 256         if (buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) {
 257                 resource_written(batch, pfb->zsbuf->texture);
 258                 batch->gmem_reason |= FD_GMEM_CLEARS_DEPTH_STENCIL;
 259         }
 260
 261         resource_written(batch, batch->query_buf);
 262
 263         DBG("%p: %x %ux%u depth=%f, stencil=%u (%s/%s)", batch, buffers,
 264                 pfb->width, pfb->height, depth, stencil,
 265                 util_format_short_name(pipe_surface_format(pfb->cbufs[0])),
 266                 util_format_short_name(pipe_surface_format(pfb->zsbuf)));
 267
 268         fd_hw_query_set_stage(batch, batch->draw, FD_STAGE_CLEAR);
 269
 270         ctx->clear(ctx, buffers, color, depth, stencil);
 271
 272         ctx->dirty |= FD_DIRTY_ZSA |
 273                         FD_DIRTY_VIEWPORT |
 274                         FD_DIRTY_RASTERIZER |
 275                         FD_DIRTY_SAMPLE_MASK |
 276                         FD_DIRTY_PROG |
 277                         FD_DIRTY_CONSTBUF |
 278                         FD_DIRTY_BLEND |
 279                         FD_DIRTY_FRAMEBUFFER;
 280
 281         if (fd_mesa_debug & FD_DBG_DCLEAR)
 282                 ctx->dirty = 0xffffffff;
 283 }
 284
 285 static void
 286 fd_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
 287                 const union pipe_color_union *color,
 288                 unsigned x, unsigned y, unsigned w, unsigned h)
 289 {
 290         DBG("TODO: x=%u, y=%u, w=%u, h=%u", x, y, w, h);
 291 }
 292
 293 static void
 294 fd_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
 295                 unsigned buffers, double depth, unsigned stencil,
 296                 unsigned x, unsigned y, unsigned w, unsigned h)
 297 {
 298         DBG("TODO: buffers=%u, depth=%f, stencil=%u, x=%u, y=%u, w=%u, h=%u",
 299                         buffers, depth, stencil, x, y, w, h);
 300 }
 301
 302 void
 303 fd_draw_init(struct pipe_context *pctx)
 304 {
 305         pctx->draw_vbo = fd_draw_vbo;
 306         pctx->clear = fd_clear;
 307         pctx->clear_render_target = fd_clear_render_target;
 308         pctx->clear_depth_stencil = fd_clear_depth_stencil;
 309 }