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 #include "util/u_helpers.h"
  35
  36 #include "freedreno_draw.h"
  37 #include "freedreno_context.h"
  38 #include "freedreno_state.h"
  39 #include "freedreno_resource.h"
  40 #include "freedreno_query_hw.h"
  41 #include "freedreno_util.h"
  42
  43 static void
  44 resource_read(struct fd_batch *batch, struct pipe_resource *prsc)
  45 {
  46         if (!prsc)
  47                 return;
  48         fd_batch_resource_used(batch, fd_resource(prsc), false);
  49 }
  50
  51 static void
  52 resource_written(struct fd_batch *batch, struct pipe_resource *prsc)
  53 {
  54         if (!prsc)
  55                 return;
  56         fd_batch_resource_used(batch, fd_resource(prsc), true);
  57 }
  58
  59 static void
  60 fd_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
  61 {
  62         struct fd_context *ctx = fd_context(pctx);
  63         struct fd_batch *batch = ctx->batch;
  64         struct pipe_framebuffer_state *pfb = &batch->framebuffer;
  65         struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
  66         unsigned i, prims, buffers = 0;
  67
  68         /* if we supported transform feedback, we'd have to disable this: */
  69         if (((scissor->maxx - scissor->minx) *
  70                         (scissor->maxy - scissor->miny)) == 0) {
  71                 return;
  72         }
  73
  74         /* TODO: push down the region versions into the tiles */
  75         if (!fd_render_condition_check(pctx))
  76                 return;
  77
  78         /* emulate unsupported primitives: */
  79         if (!fd_supported_prim(ctx, info->mode)) {
  80                 if (ctx->streamout.num_targets > 0)
  81                         debug_error("stream-out with emulated prims");
  82                 util_primconvert_save_index_buffer(ctx->primconvert, &ctx->indexbuf);
  83                 util_primconvert_save_rasterizer_state(ctx->primconvert, ctx->rasterizer);
  84                 util_primconvert_draw_vbo(ctx->primconvert, info);
  85                 return;
  86         }
  87
  88         /* Upload a user index buffer. */
  89         struct pipe_index_buffer ibuffer_saved = {};
  90         if (info->indexed && ctx->indexbuf.user_buffer &&
  91             !util_save_and_upload_index_buffer(pctx, info, &ctx->indexbuf,
  92                                                &ibuffer_saved)) {
  93                 return;
  94         }
  95
  96         if (ctx->in_blit) {
  97                 fd_batch_reset(batch);
  98                 ctx->dirty = ~0;
  99         }
 100
 101         batch->blit = ctx->in_blit;
 102         batch->back_blit = ctx->in_shadow;
 103
 104         /* NOTE: needs to be before resource_written(batch->query_buf), otherwise
 105          * query_buf may not be created yet.
 106          */
 107         fd_hw_query_set_stage(batch, batch->draw, FD_STAGE_DRAW);
 108
 109         /*
 110          * Figure out the buffers/features we need:
 111          */
 112
 113         pipe_mutex_lock(ctx->screen->lock);
 114
 115         if (fd_depth_enabled(ctx)) {
 116                 buffers |= FD_BUFFER_DEPTH;
 117                 resource_written(batch, pfb->zsbuf->texture);
 118                 batch->gmem_reason |= FD_GMEM_DEPTH_ENABLED;
 119         }
 120
 121         if (fd_stencil_enabled(ctx)) {
 122                 buffers |= FD_BUFFER_STENCIL;
 123                 resource_written(batch, pfb->zsbuf->texture);
 124                 batch->gmem_reason |= FD_GMEM_STENCIL_ENABLED;
 125         }
 126
 127         if (fd_logicop_enabled(ctx))
 128                 batch->gmem_reason |= FD_GMEM_LOGICOP_ENABLED;
 129
 130         for (i = 0; i < pfb->nr_cbufs; i++) {
 131                 struct pipe_resource *surf;
 132
 133                 if (!pfb->cbufs[i])
 134                         continue;
 135
 136                 surf = pfb->cbufs[i]->texture;
 137
 138                 resource_written(batch, surf);
 139                 buffers |= PIPE_CLEAR_COLOR0 << i;
 140
 141                 if (surf->nr_samples > 1)
 142                         batch->gmem_reason |= FD_GMEM_MSAA_ENABLED;
 143
 144                 if (fd_blend_enabled(ctx, i))
 145                         batch->gmem_reason |= FD_GMEM_BLEND_ENABLED;
 146         }
 147
 148         foreach_bit(i, ctx->constbuf[PIPE_SHADER_VERTEX].enabled_mask)
 149                 resource_read(batch, ctx->constbuf[PIPE_SHADER_VERTEX].cb[i].buffer);
 150         foreach_bit(i, ctx->constbuf[PIPE_SHADER_FRAGMENT].enabled_mask)
 151                 resource_read(batch, ctx->constbuf[PIPE_SHADER_FRAGMENT].cb[i].buffer);
 152
 153         /* Mark VBOs as being read */
 154         foreach_bit(i, ctx->vtx.vertexbuf.enabled_mask) {
 155                 assert(!ctx->vtx.vertexbuf.vb[i].user_buffer);
 156                 resource_read(batch, ctx->vtx.vertexbuf.vb[i].buffer);
 157         }
 158
 159         /* Mark index buffer as being read */
 160         resource_read(batch, ctx->indexbuf.buffer);
 161
 162         /* Mark textures as being read */
 163         foreach_bit(i, ctx->verttex.valid_textures)
 164                 resource_read(batch, ctx->verttex.textures[i]->texture);
 165         foreach_bit(i, ctx->fragtex.valid_textures)
 166                 resource_read(batch, ctx->fragtex.textures[i]->texture);
 167
 168         /* Mark streamout buffers as being written.. */
 169         for (i = 0; i < ctx->streamout.num_targets; i++)
 170                 if (ctx->streamout.targets[i])
 171                         resource_written(batch, ctx->streamout.targets[i]->buffer);
 172
 173         resource_written(batch, batch->query_buf);
 174
 175         pipe_mutex_unlock(ctx->screen->lock);
 176
 177         batch->num_draws++;
 178
 179         prims = u_reduced_prims_for_vertices(info->mode, info->count);
 180
 181         ctx->stats.draw_calls++;
 182
 183         /* TODO prims_emitted should be clipped when the stream-out buffer is
 184          * not large enough.  See max_tf_vtx().. probably need to move that
 185          * into common code.  Although a bit more annoying since a2xx doesn't
 186          * use ir3 so no common way to get at the pipe_stream_output_info
 187          * which is needed for this calculation.
 188          */
 189         if (ctx->streamout.num_targets > 0)
 190                 ctx->stats.prims_emitted += prims;
 191         ctx->stats.prims_generated += prims;
 192
 193         /* any buffers that haven't been cleared yet, we need to restore: */
 194         batch->restore |= buffers & (FD_BUFFER_ALL & ~batch->cleared);
 195         /* and any buffers used, need to be resolved: */
 196         batch->resolve |= buffers;
 197
 198         DBG("%p: %x %ux%u num_draws=%u (%s/%s)", batch, buffers,
 199                 pfb->width, pfb->height, batch->num_draws,
 200                 util_format_short_name(pipe_surface_format(pfb->cbufs[0])),
 201                 util_format_short_name(pipe_surface_format(pfb->zsbuf)));
 202
 203         if (ctx->draw_vbo(ctx, info))
 204                 batch->needs_flush = true;
 205
 206         for (i = 0; i < ctx->streamout.num_targets; i++)
 207                 ctx->streamout.offsets[i] += info->count;
 208
 209         if (fd_mesa_debug & FD_DBG_DDRAW)
 210                 ctx->dirty = 0xffffffff;
 211
 212         fd_batch_check_size(batch);
 213
 214         if (info->indexed && ibuffer_saved.user_buffer)
 215            pctx->set_index_buffer(pctx, &ibuffer_saved);
 216 }
 217
 218 /* Generic clear implementation (partially) using u_blitter: */
 219 static void
 220 fd_blitter_clear(struct pipe_context *pctx, unsigned buffers,
 221                 const union pipe_color_union *color, double depth, unsigned stencil)
 222 {
 223         struct fd_context *ctx = fd_context(pctx);
 224         struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
 225         struct blitter_context *blitter = ctx->blitter;
 226
 227         fd_blitter_pipe_begin(ctx, false, true, FD_STAGE_CLEAR);
 228
 229         util_blitter_common_clear_setup(blitter, pfb->width, pfb->height,
 230                         buffers, NULL, NULL);
 231
 232         struct pipe_stencil_ref sr = {
 233                 .ref_value = { stencil & 0xff }
 234         };
 235         pctx->set_stencil_ref(pctx, &sr);
 236
 237         struct pipe_constant_buffer cb = {
 238                 .buffer_size = 16,
 239                 .user_buffer = &color->ui,
 240         };
 241         pctx->set_constant_buffer(pctx, PIPE_SHADER_FRAGMENT, 0, &cb);
 242
 243         if (!ctx->clear_rs_state) {
 244                 const struct pipe_rasterizer_state tmpl = {
 245                         .cull_face = PIPE_FACE_NONE,
 246                         .half_pixel_center = 1,
 247                         .bottom_edge_rule = 1,
 248                         .flatshade = 1,
 249                         .depth_clip = 1,
 250                 };
 251                 ctx->clear_rs_state = pctx->create_rasterizer_state(pctx, &tmpl);
 252         }
 253         pctx->bind_rasterizer_state(pctx, ctx->clear_rs_state);
 254
 255         struct pipe_viewport_state vp = {
 256                 .scale     = { 0.5f * pfb->width, -0.5f * pfb->height, depth },
 257                 .translate = { 0.5f * pfb->width,  0.5f * pfb->height, 0.0f },
 258         };
 259         pctx->set_viewport_states(pctx, 0, 1, &vp);
 260
 261         pctx->bind_vertex_elements_state(pctx, ctx->solid_vbuf_state.vtx);
 262         pctx->set_vertex_buffers(pctx, blitter->vb_slot, 1,
 263                         &ctx->solid_vbuf_state.vertexbuf.vb[0]);
 264         pctx->set_stream_output_targets(pctx, 0, NULL, NULL);
 265         pctx->bind_vs_state(pctx, ctx->solid_prog.vp);
 266         pctx->bind_fs_state(pctx, ctx->solid_prog.fp);
 267
 268         struct pipe_draw_info info = {
 269                 .mode = PIPE_PRIM_MAX,    /* maps to DI_PT_RECTLIST */
 270                 .count = 2,
 271                 .max_index = 1,
 272                 .instance_count = 1,
 273         };
 274         ctx->draw_vbo(ctx, &info);
 275
 276         util_blitter_restore_constant_buffer_state(blitter);
 277         util_blitter_restore_vertex_states(blitter);
 278         util_blitter_restore_fragment_states(blitter);
 279         util_blitter_restore_render_cond(blitter);
 280         util_blitter_unset_running_flag(blitter);
 281
 282         fd_blitter_pipe_end(ctx);
 283 }
 284
 285 /* TODO figure out how to make better use of existing state mechanism
 286  * for clear (and possibly gmem->mem / mem->gmem) so we can (a) keep
 287  * track of what state really actually changes, and (b) reduce the code
 288  * in the a2xx/a3xx parts.
 289  */
 290
 291 static void
 292 fd_clear(struct pipe_context *pctx, unsigned buffers,
 293                 const union pipe_color_union *color, double depth, unsigned stencil)
 294 {
 295         struct fd_context *ctx = fd_context(pctx);
 296         struct fd_batch *batch = ctx->batch;
 297         struct pipe_framebuffer_state *pfb = &batch->framebuffer;
 298         struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
 299         unsigned cleared_buffers;
 300         int i;
 301
 302         /* TODO: push down the region versions into the tiles */
 303         if (!fd_render_condition_check(pctx))
 304                 return;
 305
 306         if (ctx->in_blit) {
 307                 fd_batch_reset(batch);
 308                 ctx->dirty = ~0;
 309         }
 310
 311         /* for bookkeeping about which buffers have been cleared (and thus
 312          * can fully or partially skip mem2gmem) we need to ignore buffers
 313          * that have already had a draw, in case apps do silly things like
 314          * clear after draw (ie. if you only clear the color buffer, but
 315          * something like alpha-test causes side effects from the draw in
 316          * the depth buffer, etc)
 317          */
 318         cleared_buffers = buffers & (FD_BUFFER_ALL & ~batch->restore);
 319
 320         /* do we have full-screen scissor? */
 321         if (!memcmp(scissor, &ctx->disabled_scissor, sizeof(*scissor))) {
 322                 batch->cleared |= cleared_buffers;
 323         } else {
 324                 batch->partial_cleared |= cleared_buffers;
 325                 if (cleared_buffers & PIPE_CLEAR_COLOR)
 326                         batch->cleared_scissor.color = *scissor;
 327                 if (cleared_buffers & PIPE_CLEAR_DEPTH)
 328                         batch->cleared_scissor.depth = *scissor;
 329                 if (cleared_buffers & PIPE_CLEAR_STENCIL)
 330                         batch->cleared_scissor.stencil = *scissor;
 331         }
 332         batch->resolve |= buffers;
 333         batch->needs_flush = true;
 334
 335         pipe_mutex_lock(ctx->screen->lock);
 336
 337         if (buffers & PIPE_CLEAR_COLOR)
 338                 for (i = 0; i < pfb->nr_cbufs; i++)
 339                         if (buffers & (PIPE_CLEAR_COLOR0 << i))
 340                                 resource_written(batch, pfb->cbufs[i]->texture);
 341
 342         if (buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) {
 343                 resource_written(batch, pfb->zsbuf->texture);
 344                 batch->gmem_reason |= FD_GMEM_CLEARS_DEPTH_STENCIL;
 345         }
 346
 347         resource_written(batch, batch->query_buf);
 348
 349         pipe_mutex_unlock(ctx->screen->lock);
 350
 351         DBG("%p: %x %ux%u depth=%f, stencil=%u (%s/%s)", batch, buffers,
 352                 pfb->width, pfb->height, depth, stencil,
 353                 util_format_short_name(pipe_surface_format(pfb->cbufs[0])),
 354                 util_format_short_name(pipe_surface_format(pfb->zsbuf)));
 355
 356         /* if per-gen backend doesn't implement ctx->clear() generic
 357          * blitter clear:
 358          */
 359         if (!ctx->clear) {
 360                 fd_blitter_clear(pctx, buffers, color, depth, stencil);
 361                 return;
 362         }
 363
 364         fd_hw_query_set_stage(batch, batch->draw, FD_STAGE_CLEAR);
 365
 366         ctx->clear(ctx, buffers, color, depth, stencil);
 367
 368         ctx->dirty |= FD_DIRTY_ZSA |
 369                         FD_DIRTY_VIEWPORT |
 370                         FD_DIRTY_RASTERIZER |
 371                         FD_DIRTY_SAMPLE_MASK |
 372                         FD_DIRTY_PROG |
 373                         FD_DIRTY_CONSTBUF |
 374                         FD_DIRTY_BLEND |
 375                         FD_DIRTY_FRAMEBUFFER;
 376
 377         if (fd_mesa_debug & FD_DBG_DCLEAR)
 378                 ctx->dirty = 0xffffffff;
 379 }
 380
 381 static void
 382 fd_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
 383                 const union pipe_color_union *color,
 384                 unsigned x, unsigned y, unsigned w, unsigned h,
 385                 bool render_condition_enabled)
 386 {
 387         DBG("TODO: x=%u, y=%u, w=%u, h=%u", x, y, w, h);
 388 }
 389
 390 static void
 391 fd_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
 392                 unsigned buffers, double depth, unsigned stencil,
 393                 unsigned x, unsigned y, unsigned w, unsigned h,
 394                 bool render_condition_enabled)
 395 {
 396         DBG("TODO: buffers=%u, depth=%f, stencil=%u, x=%u, y=%u, w=%u, h=%u",
 397                         buffers, depth, stencil, x, y, w, h);
 398 }
 399
 400 void
 401 fd_draw_init(struct pipe_context *pctx)
 402 {
 403         pctx->draw_vbo = fd_draw_vbo;
 404         pctx->clear = fd_clear;
 405         pctx->clear_render_target = fd_clear_render_target;
 406         pctx->clear_depth_stencil = fd_clear_depth_stencil;
 407 }