src/mesa/drivers/dri/i965/gen6_vs_state.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 "brw_context.h"
  29 #include "brw_state.h"
  30 #include "brw_defines.h"
  31 #include "brw_util.h"
  32 #include "program/prog_parameter.h"
  33 #include "program/prog_statevars.h"
  34 #include "intel_batchbuffer.h"
  35
  36 static void
  37 gen6_prepare_vs_push_constants(struct brw_context *brw)
  38 {
  39    struct intel_context *intel = &brw->intel;
  40    struct gl_context *ctx = &intel->ctx;
  41    /* _BRW_NEW_VERTEX_PROGRAM */
  42    const struct brw_vertex_program *vp =
  43       brw_vertex_program_const(brw->vertex_program);
  44    unsigned int nr_params = brw->vs.prog_data->nr_params / 4;
  45    bool uses_clip_distance = vp->program.UsesClipDistance;
  46
  47    if (brw->vertex_program->IsNVProgram)
  48       _mesa_load_tracked_matrices(ctx);
  49
  50    /* Updates the ParamaterValues[i] pointers for all parameters of the
  51     * basic type of PROGRAM_STATE_VAR.
  52     */
  53    /* XXX: Should this happen somewhere before to get our state flag set? */
  54    _mesa_load_state_parameters(ctx, vp->program.Base.Parameters);
  55
  56    /* CACHE_NEW_VS_PROG | _NEW_TRANSFORM */
  57    if (brw->vs.prog_data->nr_params == 0 && !ctx->Transform.ClipPlanesEnabled) {
  58       brw->vs.push_const_size = 0;
  59    } else {
  60       int params_uploaded = 0;
  61       float *param;
  62       int i;
  63
  64       param = brw_state_batch(brw, AUB_TRACE_VS_CONSTANTS,
  65                               (MAX_CLIP_PLANES + nr_params) *
  66                               4 * sizeof(float),
  67                               32, &brw->vs.push_const_offset);
  68
  69       /* This should be loaded like any other param, but it's ad-hoc
  70        * until we redo the VS backend.
  71        */
  72       if (!uses_clip_distance) {
  73          for (i = 0; i < MAX_CLIP_PLANES; i++) {
  74             if (ctx->Transform.ClipPlanesEnabled & (1 << i)) {
  75                memcpy(param, ctx->Transform._ClipUserPlane[i], 4 * sizeof(float));
  76                param += 4;
  77                params_uploaded++;
  78             }
  79          }
  80       }
  81       /* Align to a reg for convenience for brw_vs_emit.c */
  82       if (params_uploaded & 1) {
  83          param += 4;
  84          params_uploaded++;
  85       }
  86
  87       if (brw->vs.prog_data->uses_new_param_layout) {
  88          for (i = 0; i < brw->vs.prog_data->nr_params; i++) {
  89             *param = *brw->vs.prog_data->param[i];
  90             param++;
  91          }
  92          params_uploaded += brw->vs.prog_data->nr_params / 4;
  93       } else {
  94          for (i = 0; i < vp->program.Base.Parameters->NumParameters; i++) {
  95             if (brw->vs.constant_map[i] != -1) {
  96                memcpy(param + brw->vs.constant_map[i] * 4,
  97                       vp->program.Base.Parameters->ParameterValues[i],
  98                       4 * sizeof(float));
  99                params_uploaded++;
 100             }
 101          }
 102       }
 103
 104       if (0) {
 105          printf("VS constant buffer:\n");
 106          for (i = 0; i < params_uploaded; i++) {
 107             float *buf = param + i * 4;
 108             printf("%d: %f %f %f %f\n",
 109                    i, buf[0], buf[1], buf[2], buf[3]);
 110          }
 111       }
 112
 113       brw->vs.push_const_size = (params_uploaded + 1) / 2;
 114       /* We can only push 32 registers of constants at a time. */
 115       assert(brw->vs.push_const_size <= 32);
 116    }
 117 }
 118
 119 const struct brw_tracked_state gen6_vs_constants = {
 120    .dirty = {
 121       .mesa  = _NEW_TRANSFORM | _NEW_PROGRAM_CONSTANTS,
 122       .brw   = (BRW_NEW_BATCH |
 123                 BRW_NEW_VERTEX_PROGRAM),
 124       .cache = CACHE_NEW_VS_PROG,
 125    },
 126    .prepare = gen6_prepare_vs_push_constants,
 127 };
 128
 129 static void
 130 upload_vs_state(struct brw_context *brw)
 131 {
 132    struct intel_context *intel = &brw->intel;
 133
 134    if (brw->vs.push_const_size == 0) {
 135       /* Disable the push constant buffers. */
 136       BEGIN_BATCH(5);
 137       OUT_BATCH(_3DSTATE_CONSTANT_VS << 16 | (5 - 2));
 138       OUT_BATCH(0);
 139       OUT_BATCH(0);
 140       OUT_BATCH(0);
 141       OUT_BATCH(0);
 142       ADVANCE_BATCH();
 143    } else {
 144       BEGIN_BATCH(5);
 145       OUT_BATCH(_3DSTATE_CONSTANT_VS << 16 |
 146                 GEN6_CONSTANT_BUFFER_0_ENABLE |
 147                 (5 - 2));
 148       /* Pointer to the VS constant buffer.  Covered by the set of
 149        * state flags from gen6_prepare_wm_constants
 150        */
 151       OUT_BATCH(brw->vs.push_const_offset +
 152                 brw->vs.push_const_size - 1);
 153       OUT_BATCH(0);
 154       OUT_BATCH(0);
 155       OUT_BATCH(0);
 156       ADVANCE_BATCH();
 157    }
 158
 159    BEGIN_BATCH(6);
 160    OUT_BATCH(_3DSTATE_VS << 16 | (6 - 2));
 161    OUT_BATCH(brw->vs.prog_offset);
 162    OUT_BATCH((0 << GEN6_VS_SAMPLER_COUNT_SHIFT) |
 163              GEN6_VS_FLOATING_POINT_MODE_ALT |
 164              (brw->vs.nr_surfaces << GEN6_VS_BINDING_TABLE_ENTRY_COUNT_SHIFT));
 165
 166    if (brw->vs.prog_data->total_scratch) {
 167       OUT_RELOC(brw->vs.scratch_bo,
 168                 I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
 169                 ffs(brw->vs.prog_data->total_scratch) - 11);
 170    } else {
 171       OUT_BATCH(0);
 172    }
 173
 174    OUT_BATCH((1 << GEN6_VS_DISPATCH_START_GRF_SHIFT) |
 175              (brw->vs.prog_data->urb_read_length << GEN6_VS_URB_READ_LENGTH_SHIFT) |
 176              (0 << GEN6_VS_URB_ENTRY_READ_OFFSET_SHIFT));
 177
 178    OUT_BATCH(((brw->vs_max_threads - 1) << GEN6_VS_MAX_THREADS_SHIFT) |
 179              GEN6_VS_STATISTICS_ENABLE |
 180              GEN6_VS_ENABLE);
 181    ADVANCE_BATCH();
 182
 183    /* Based on my reading of the simulator, the VS constants don't get
 184     * pulled into the VS FF unit until an appropriate pipeline flush
 185     * happens, and instead the 3DSTATE_CONSTANT_VS packet just adds
 186     * references to them into a little FIFO.  The flushes are common,
 187     * but don't reliably happen between this and a 3DPRIMITIVE, causing
 188     * the primitive to use the wrong constants.  Then the FIFO
 189     * containing the constant setup gets added to again on the next
 190     * constants change, and eventually when a flush does happen the
 191     * unit is overwhelmed by constant changes and dies.
 192     *
 193     * To avoid this, send a PIPE_CONTROL down the line that will
 194     * update the unit immediately loading the constants.  The flush
 195     * type bits here were those set by the STATE_BASE_ADDRESS whose
 196     * move in a82a43e8d99e1715dd11c9c091b5ab734079b6a6 triggered the
 197     * bug reports that led to this workaround, and may be more than
 198     * what is strictly required to avoid the issue.
 199     */
 200    BEGIN_BATCH(4);
 201    OUT_BATCH(_3DSTATE_PIPE_CONTROL);
 202    OUT_BATCH(PIPE_CONTROL_DEPTH_STALL |
 203              PIPE_CONTROL_INSTRUCTION_FLUSH |
 204              PIPE_CONTROL_STATE_CACHE_INVALIDATE);
 205    OUT_BATCH(0); /* address */
 206    OUT_BATCH(0); /* write data */
 207    ADVANCE_BATCH();
 208 }
 209
 210 const struct brw_tracked_state gen6_vs_state = {
 211    .dirty = {
 212       .mesa  = _NEW_TRANSFORM | _NEW_PROGRAM_CONSTANTS,
 213       .brw   = (BRW_NEW_NR_VS_SURFACES |
 214                 BRW_NEW_URB_FENCE |
 215                 BRW_NEW_CONTEXT |
 216                 BRW_NEW_VERTEX_PROGRAM |
 217                 BRW_NEW_BATCH),
 218       .cache = CACHE_NEW_VS_PROG
 219    },
 220    .emit = upload_vs_state,
 221 };