src/mesa/drivers/dri/i965/gen6_urb.c

   1 /*
   2  * Copyright © 2009 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  *
  23  * Authors:
  24  *    Eric Anholt <eric@anholt.net>
  25  *
  26  */
  27
  28 #include "main/macros.h"
  29 #include "intel_batchbuffer.h"
  30 #include "brw_context.h"
  31 #include "brw_state.h"
  32 #include "brw_defines.h"
  33
  34 /**
  35  * When the GS is not in use, we assign the entire URB space to the VS.  When
  36  * the GS is in use, we split the URB space evenly between the VS and the GS.
  37  * This is not ideal, but it's simple.
  38  *
  39  *           URB size / 2                   URB size / 2
  40  *   _____________-______________   _____________-______________
  41  *  /                            \ /                            \
  42  * +-------------------------------------------------------------+
  43  * | Vertex Shader Entries        | Geometry Shader Entries      |
  44  * +-------------------------------------------------------------+
  45  *
  46  * Sandybridge GT1 has 32kB of URB space, while GT2 has 64kB.
  47  * (See the Sandybridge PRM, Volume 2, Part 1, Section 1.4.7: 3DSTATE_URB.)
  48  */
  49 static void
  50 gen6_upload_urb( struct brw_context *brw )
  51 {
  52    int nr_vs_entries, nr_gs_entries;
  53    int total_urb_size = brw->urb.size * 1024; /* in bytes */
  54
  55    bool gs_present = brw->ff_gs.prog_active || brw->geometry_program;
  56
  57    /* BRW_NEW_VS_PROG_DATA */
  58    unsigned vs_size = MAX2(brw->vs.prog_data->base.urb_entry_size, 1);
  59
  60    /* Whe using GS to do transform feedback only we use the same VUE layout for
  61     * VS outputs and GS outputs (as it's what the SF and Clipper expect), so we
  62     * can simply make the GS URB entry size the same as for the VS.  This may
  63     * technically be too large in cases where we have few vertex attributes and
  64     * a lot of varyings, since the VS size is determined by the larger of the
  65     * two. For now, it's safe.
  66     *
  67     * For user-provided GS the assumption above does not hold since the GS
  68     * outputs can be different from the VS outputs.
  69     */
  70    unsigned gs_size = vs_size;
  71    if (brw->geometry_program) {
  72       gs_size = brw->gs.prog_data->base.urb_entry_size;
  73       assert(gs_size >= 1);
  74    }
  75
  76    /* Calculate how many entries fit in each stage's section of the URB */
  77    if (gs_present) {
  78       nr_vs_entries = (total_urb_size/2) / (vs_size * 128);
  79       nr_gs_entries = (total_urb_size/2) / (gs_size * 128);
  80    } else {
  81       nr_vs_entries = total_urb_size / (vs_size * 128);
  82       nr_gs_entries = 0;
  83    }
  84
  85    /* Then clamp to the maximum allowed by the hardware */
  86    if (nr_vs_entries > brw->urb.max_vs_entries)
  87       nr_vs_entries = brw->urb.max_vs_entries;
  88
  89    if (nr_gs_entries > brw->urb.max_gs_entries)
  90       nr_gs_entries = brw->urb.max_gs_entries;
  91
  92    /* Finally, both must be a multiple of 4 (see 3DSTATE_URB in the PRM). */
  93    brw->urb.nr_vs_entries = ROUND_DOWN_TO(nr_vs_entries, 4);
  94    brw->urb.nr_gs_entries = ROUND_DOWN_TO(nr_gs_entries, 4);
  95
  96    assert(brw->urb.nr_vs_entries >= brw->urb.min_vs_entries);
  97    assert(brw->urb.nr_vs_entries % 4 == 0);
  98    assert(brw->urb.nr_gs_entries % 4 == 0);
  99    assert(vs_size <= 5);
 100    assert(gs_size <= 5);
 101
 102    BEGIN_BATCH(3);
 103    OUT_BATCH(_3DSTATE_URB << 16 | (3 - 2));
 104    OUT_BATCH(((vs_size - 1) << GEN6_URB_VS_SIZE_SHIFT) |
 105              ((brw->urb.nr_vs_entries) << GEN6_URB_VS_ENTRIES_SHIFT));
 106    OUT_BATCH(((gs_size - 1) << GEN6_URB_GS_SIZE_SHIFT) |
 107              ((brw->urb.nr_gs_entries) << GEN6_URB_GS_ENTRIES_SHIFT));
 108    ADVANCE_BATCH();
 109
 110    /* From the PRM Volume 2 part 1, section 1.4.7:
 111     *
 112     *   Because of a urb corruption caused by allocating a previous gsunit’s
 113     *   urb entry to vsunit software is required to send a "GS NULL
 114     *   Fence"(Send URB fence with VS URB size == 1 and GS URB size == 0) plus
 115     *   a dummy DRAW call before any case where VS will be taking over GS URB
 116     *   space.
 117     *
 118     * It is not clear exactly what this means ("URB fence" is a command that
 119     * doesn't exist on Gen6).  So for now we just do a full pipeline flush as
 120     * a workaround.
 121     */
 122    if (brw->urb.gs_present && !gs_present)
 123       intel_batchbuffer_emit_mi_flush(brw);
 124    brw->urb.gs_present = gs_present;
 125 }
 126
 127 const struct brw_tracked_state gen6_urb = {
 128    .dirty = {
 129       .mesa = 0,
 130       .brw = BRW_NEW_CONTEXT |
 131              BRW_NEW_GEOMETRY_PROGRAM,
 132       .cache = BRW_NEW_GS_PROG_DATA |
 133                BRW_NEW_VS_PROG_DATA |
 134                BRW_NEW_FF_GS_PROG_DATA,
 135    },
 136    .emit = gen6_upload_urb,
 137 };