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

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
   4  develop this 3D driver.
   5
   6  Permission is hereby granted, free of charge, to any person obtaining
   7  a copy of this software and associated documentation files (the
   8  "Software"), to deal in the Software without restriction, including
   9  without limitation the rights to use, copy, modify, merge, publish,
  10  distribute, sublicense, and/or sell copies of the Software, and to
  11  permit persons to whom the Software is furnished to do so, subject to
  12  the following conditions:
  13
  14  The above copyright notice and this permission notice (including the
  15  next paragraph) shall be included in all copies or substantial
  16  portions of the Software.
  17
  18  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  19  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  21  IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  22  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  23  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  24  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25
  26  **********************************************************************/
  27  /*
  28   * Authors:
  29   *   Keith Whitwell <keith@tungstengraphics.com>
  30   */
  31
  32
  33
  34 #include "brw_context.h"
  35 #include "brw_state.h"
  36 #include "brw_defines.h"
  37 #include "main/macros.h"
  38
  39 static void
  40 brw_upload_vs_unit(struct brw_context *brw)
  41 {
  42    struct intel_context *intel = &brw->intel;
  43    struct gl_context *ctx = &intel->ctx;
  44    struct brw_vs_unit_state *vs;
  45
  46    vs = brw_state_batch(brw, AUB_TRACE_VS_STATE,
  47                         sizeof(*vs), 32, &brw->vs.state_offset);
  48    memset(vs, 0, sizeof(*vs));
  49
  50    /* BRW_NEW_PROGRAM_CACHE | CACHE_NEW_VS_PROG */
  51    vs->thread0.grf_reg_count = ALIGN(brw->vs.prog_data->total_grf, 16) / 16 - 1;
  52    vs->thread0.kernel_start_pointer =
  53       brw_program_reloc(brw,
  54                         brw->vs.state_offset +
  55                         offsetof(struct brw_vs_unit_state, thread0),
  56                         brw->vs.prog_offset +
  57                         (vs->thread0.grf_reg_count << 1)) >> 6;
  58
  59    vs->thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
  60    /* Choosing multiple program flow means that we may get 2-vertex threads,
  61     * which will have the channel mask for dwords 4-7 enabled in the thread,
  62     * and those dwords will be written to the second URB handle when we
  63     * brw_urb_WRITE() results.
  64     */
  65    /* Disable single program flow on Ironlake.  We cannot reliably get
  66     * all applications working without it.  See:
  67     * https://bugs.freedesktop.org/show_bug.cgi?id=29172
  68     *
  69     * The most notable and reliably failing application is the Humus
  70     * demo "CelShading"
  71    */
  72    vs->thread1.single_program_flow = (intel->gen == 5);
  73
  74    vs->thread1.binding_table_entry_count = 0;
  75
  76    if (brw->vs.prog_data->total_scratch != 0) {
  77       vs->thread2.scratch_space_base_pointer =
  78          brw->vs.scratch_bo->offset >> 10; /* reloc */
  79       vs->thread2.per_thread_scratch_space =
  80          ffs(brw->vs.prog_data->total_scratch) - 11;
  81    } else {
  82       vs->thread2.scratch_space_base_pointer = 0;
  83       vs->thread2.per_thread_scratch_space = 0;
  84    }
  85
  86    vs->thread3.urb_entry_read_length = brw->vs.prog_data->urb_read_length;
  87    vs->thread3.const_urb_entry_read_length = brw->vs.prog_data->curb_read_length;
  88    vs->thread3.dispatch_grf_start_reg = 1;
  89    vs->thread3.urb_entry_read_offset = 0;
  90
  91    /* BRW_NEW_CURBE_OFFSETS, _NEW_TRANSFORM, BRW_NEW_VERTEX_PROGRAM */
  92    if (ctx->Transform.ClipPlanesEnabled && !brw->vs.prog_data->uses_new_param_layout) {
  93       /* Note that we read in the userclip planes as well, hence
  94        * clip_start:
  95        */
  96       vs->thread3.const_urb_entry_read_offset = brw->curbe.clip_start * 2;
  97    }
  98    else {
  99       vs->thread3.const_urb_entry_read_offset = brw->curbe.vs_start * 2;
 100    }
 101
 102
 103    /* BRW_NEW_URB_FENCE */
 104    if (intel->gen == 5) {
 105       switch (brw->urb.nr_vs_entries) {
 106       case 8:
 107       case 12:
 108       case 16:
 109       case 32:
 110       case 64:
 111       case 96:
 112       case 128:
 113       case 168:
 114       case 192:
 115       case 224:
 116       case 256:
 117          vs->thread4.nr_urb_entries = brw->urb.nr_vs_entries >> 2;
 118          break;
 119       default:
 120          assert(0);
 121       }
 122    } else {
 123       switch (brw->urb.nr_vs_entries) {
 124       case 8:
 125       case 12:
 126       case 16:
 127       case 32:
 128          break;
 129       case 64:
 130          assert(intel->is_g4x);
 131          break;
 132       default:
 133          assert(0);
 134       }
 135       vs->thread4.nr_urb_entries = brw->urb.nr_vs_entries;
 136    }
 137
 138    vs->thread4.urb_entry_allocation_size = brw->urb.vsize - 1;
 139
 140    vs->thread4.max_threads = CLAMP(brw->urb.nr_vs_entries / 2,
 141                                    1, brw->max_vs_threads) - 1;
 142
 143    /* No samplers for ARB_vp programs:
 144     */
 145    /* It has to be set to 0 for Ironlake
 146     */
 147    vs->vs5.sampler_count = 0;
 148
 149    if (unlikely(INTEL_DEBUG & DEBUG_STATS))
 150       vs->thread4.stats_enable = 1;
 151
 152    /* Vertex program always enabled:
 153     */
 154    vs->vs6.vs_enable = 1;
 155
 156    /* Emit scratch space relocation */
 157    if (brw->vs.prog_data->total_scratch != 0) {
 158       drm_intel_bo_emit_reloc(intel->batch.bo,
 159                               brw->vs.state_offset +
 160                               offsetof(struct brw_vs_unit_state, thread2),
 161                               brw->vs.scratch_bo,
 162                               vs->thread2.per_thread_scratch_space,
 163                               I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER);
 164    }
 165
 166    brw->state.dirty.cache |= CACHE_NEW_VS_UNIT;
 167 }
 168
 169 const struct brw_tracked_state brw_vs_unit = {
 170    .dirty = {
 171       .mesa  = _NEW_TRANSFORM,
 172       .brw   = (BRW_NEW_BATCH |
 173                 BRW_NEW_PROGRAM_CACHE |
 174                 BRW_NEW_CURBE_OFFSETS |
 175                 BRW_NEW_URB_FENCE |
 176                 BRW_NEW_VERTEX_PROGRAM),
 177       .cache = CACHE_NEW_VS_PROG
 178    },
 179    .emit = brw_upload_vs_unit,
 180 };