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