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->in_blit) {
  88                 fd_batch_reset(batch);
  89                 ctx->dirty = ~0;
  90         }
  91
  92         batch->blit = ctx->in_blit;
  93         batch->back_blit = ctx->in_shadow;
  94
  95         /* NOTE: needs to be before resource_written(batch->query_buf), otherwise
  96          * query_buf may not be created yet.
  97          */
  98         fd_hw_query_set_stage(batch, batch->draw, FD_STAGE_DRAW);
  99
 100         /*
 101          * Figure out the buffers/features we need:
 102          */
 103
 104         pipe_mutex_lock(ctx->screen->lock);
 105
 106         if (fd_depth_enabled(ctx)) {
 107                 buffers |= FD_BUFFER_DEPTH;
 108                 resource_written(batch, pfb->zsbuf->texture);
 109                 batch->gmem_reason |= FD_GMEM_DEPTH_ENABLED;
 110         }
 111
 112         if (fd_stencil_enabled(ctx)) {
 113                 buffers |= FD_BUFFER_STENCIL;
 114                 resource_written(batch, pfb->zsbuf->texture);
 115                 batch->gmem_reason |= FD_GMEM_STENCIL_ENABLED;
 116         }
 117
 118         if (fd_logicop_enabled(ctx))
 119                 batch->gmem_reason |= FD_GMEM_LOGICOP_ENABLED;
 120
 121         for (i = 0; i < pfb->nr_cbufs; i++) {
 122                 struct pipe_resource *surf;
 123
 124                 if (!pfb->cbufs[i])
 125                         continue;
 126
 127                 surf = pfb->cbufs[i]->texture;
 128
 129                 resource_written(batch, surf);
 130                 buffers |= PIPE_CLEAR_COLOR0 << i;
 131
 132                 if (surf->nr_samples > 1)
 133                         batch->gmem_reason |= FD_GMEM_MSAA_ENABLED;
 134
 135                 if (fd_blend_enabled(ctx, i))
 136                         batch->gmem_reason |= FD_GMEM_BLEND_ENABLED;
 137         }
 138
 139         foreach_bit(i, ctx->constbuf[PIPE_SHADER_VERTEX].enabled_mask)
 140                 resource_read(batch, ctx->constbuf[PIPE_SHADER_VERTEX].cb[i].buffer);
 141         foreach_bit(i, ctx->constbuf[PIPE_SHADER_FRAGMENT].enabled_mask)
 142                 resource_read(batch, ctx->constbuf[PIPE_SHADER_FRAGMENT].cb[i].buffer);
 143
 144         /* Mark VBOs as being read */
 145         foreach_bit(i, ctx->vtx.vertexbuf.enabled_mask) {
 146                 assert(!ctx->vtx.vertexbuf.vb[i].user_buffer);
 147                 resource_read(batch, ctx->vtx.vertexbuf.vb[i].buffer);
 148         }
 149
 150         /* Mark index buffer as being read */
 151         resource_read(batch, ctx->indexbuf.buffer);
 152
 153         /* Mark textures as being read */
 154         foreach_bit(i, ctx->verttex.valid_textures)
 155                 resource_read(batch, ctx->verttex.textures[i]->texture);
 156         foreach_bit(i, ctx->fragtex.valid_textures)
 157                 resource_read(batch, ctx->fragtex.textures[i]->texture);
 158
 159         /* Mark streamout buffers as being written.. */
 160         for (i = 0; i < ctx->streamout.num_targets; i++)
 161                 if (ctx->streamout.targets[i])
 162                         resource_written(batch, ctx->streamout.targets[i]->buffer);
 163
 164         resource_written(batch, batch->query_buf);
 165
 166         pipe_mutex_unlock(ctx->screen->lock);
 167
 168         batch->num_draws++;
 169
 170         prims = u_reduced_prims_for_vertices(info->mode, info->count);
 171
 172         ctx->stats.draw_calls++;
 173
 174         /* TODO prims_emitted should be clipped when the stream-out buffer is
 175          * not large enough.  See max_tf_vtx().. probably need to move that
 176          * into common code.  Although a bit more annoying since a2xx doesn't
 177          * use ir3 so no common way to get at the pipe_stream_output_info
 178          * which is needed for this calculation.
 179          */
 180         if (ctx->streamout.num_targets > 0)
 181                 ctx->stats.prims_emitted += prims;
 182         ctx->stats.prims_generated += prims;
 183
 184         /* any buffers that haven't been cleared yet, we need to restore: */
 185         batch->restore |= buffers & (FD_BUFFER_ALL & ~batch->cleared);
 186         /* and any buffers used, need to be resolved: */
 187         batch->resolve |= buffers;
 188
 189         DBG("%p: %x %ux%u num_draws=%u (%s/%s)", batch, buffers,
 190                 pfb->width, pfb->height, batch->num_draws,
 191                 util_format_short_name(pipe_surface_format(pfb->cbufs[0])),
 192                 util_format_short_name(pipe_surface_format(pfb->zsbuf)));
 193
 194         if (ctx->draw_vbo(ctx, info))
 195                 batch->needs_flush = true;
 196
 197         for (i = 0; i < ctx->streamout.num_targets; i++)
 198                 ctx->streamout.offsets[i] += info->count;
 199
 200         if (fd_mesa_debug & FD_DBG_DDRAW)
 201                 ctx->dirty = 0xffffffff;
 202
 203         fd_batch_check_size(batch);
 204 }
 205
 206 /* TODO figure out how to make better use of existing state mechanism
 207  * for clear (and possibly gmem->mem / mem->gmem) so we can (a) keep
 208  * track of what state really actually changes, and (b) reduce the code
 209  * in the a2xx/a3xx parts.
 210  */
 211
 212 static void
 213 fd_clear(struct pipe_context *pctx, unsigned buffers,
 214                 const union pipe_color_union *color, double depth, unsigned stencil)
 215 {
 216         struct fd_context *ctx = fd_context(pctx);
 217         struct fd_batch *batch = ctx->batch;
 218         struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 219         struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
 220         unsigned cleared_buffers;
 221         int i;
 222
 223         /* TODO: push down the region versions into the tiles */
 224         if (!fd_render_condition_check(pctx))
 225                 return;
 226
 227         if (ctx->in_blit) {
 228                 fd_batch_reset(batch);
 229                 ctx->dirty = ~0;
 230         }
 231
 232         /* for bookkeeping about which buffers have been cleared (and thus
 233          * can fully or partially skip mem2gmem) we need to ignore buffers
 234          * that have already had a draw, in case apps do silly things like
 235          * clear after draw (ie. if you only clear the color buffer, but
 236          * something like alpha-test causes side effects from the draw in
 237          * the depth buffer, etc)
 238          */
 239         cleared_buffers = buffers & (FD_BUFFER_ALL & ~batch->restore);
 240
 241         /* do we have full-screen scissor? */
 242         if (!memcmp(scissor, &ctx->disabled_scissor, sizeof(*scissor))) {
 243                 batch->cleared |= cleared_buffers;
 244         } else {
 245                 batch->partial_cleared |= cleared_buffers;
 246                 if (cleared_buffers & PIPE_CLEAR_COLOR)
 247                         batch->cleared_scissor.color = *scissor;
 248                 if (cleared_buffers & PIPE_CLEAR_DEPTH)
 249                         batch->cleared_scissor.depth = *scissor;
 250                 if (cleared_buffers & PIPE_CLEAR_STENCIL)
 251                         batch->cleared_scissor.stencil = *scissor;
 252         }
 253         batch->resolve |= buffers;
 254         batch->needs_flush = true;
 255
 256         pipe_mutex_lock(ctx->screen->lock);
 257
 258         if (buffers & PIPE_CLEAR_COLOR)
 259                 for (i = 0; i < pfb->nr_cbufs; i++)
 260                         if (buffers & (PIPE_CLEAR_COLOR0 << i))
 261                                 resource_written(batch, pfb->cbufs[i]->texture);
 262
 263         if (buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) {
 264                 resource_written(batch, pfb->zsbuf->texture);
 265                 batch->gmem_reason |= FD_GMEM_CLEARS_DEPTH_STENCIL;
 266         }
 267
 268         resource_written(batch, batch->query_buf);
 269
 270         pipe_mutex_unlock(ctx->screen->lock);
 271
 272         DBG("%p: %x %ux%u depth=%f, stencil=%u (%s/%s)", batch, buffers,
 273                 pfb->width, pfb->height, depth, stencil,
 274                 util_format_short_name(pipe_surface_format(pfb->cbufs[0])),
 275                 util_format_short_name(pipe_surface_format(pfb->zsbuf)));
 276
 277         fd_hw_query_set_stage(batch, batch->draw, FD_STAGE_CLEAR);
 278
 279         ctx->clear(ctx, buffers, color, depth, stencil);
 280
 281         ctx->dirty |= FD_DIRTY_ZSA |
 282                         FD_DIRTY_VIEWPORT |
 283                         FD_DIRTY_RASTERIZER |
 284                         FD_DIRTY_SAMPLE_MASK |
 285                         FD_DIRTY_PROG |
 286                         FD_DIRTY_CONSTBUF |
 287                         FD_DIRTY_BLEND |
 288                         FD_DIRTY_FRAMEBUFFER;
 289
 290         if (fd_mesa_debug & FD_DBG_DCLEAR)
 291                 ctx->dirty = 0xffffffff;
 292 }
 293
 294 static void
 295 fd_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
 296                 const union pipe_color_union *color,
 297                 unsigned x, unsigned y, unsigned w, unsigned h,
 298                 bool render_condition_enabled)
 299 {
 300         DBG("TODO: x=%u, y=%u, w=%u, h=%u", x, y, w, h);
 301 }
 302
 303 static void
 304 fd_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
 305                 unsigned buffers, double depth, unsigned stencil,
 306                 unsigned x, unsigned y, unsigned w, unsigned h,
 307                 bool render_condition_enabled)
 308 {
 309         DBG("TODO: buffers=%u, depth=%f, stencil=%u, x=%u, y=%u, w=%u, h=%u",
 310                         buffers, depth, stencil, x, y, w, h);
 311 }
 312
 313 void
 314 fd_draw_init(struct pipe_context *pctx)
 315 {
 316         pctx->draw_vbo = fd_draw_vbo;
 317         pctx->clear = fd_clear;
 318         pctx->clear_render_target = fd_clear_render_target;
 319         pctx->clear_depth_stencil = fd_clear_depth_stencil;
 320 }