src/mesa/drivers/dri/i965/gen6_sol.c

   1 /*
   2  * Copyright © 2011 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
  24 /** \file gen6_sol.c
  25  *
  26  * Code to initialize the binding table entries used by transform feedback.
  27  */
  28
  29 #include "main/bufferobj.h"
  30 #include "main/macros.h"
  31 #include "brw_context.h"
  32 #include "intel_batchbuffer.h"
  33 #include "brw_defines.h"
  34 #include "brw_state.h"
  35 #include "main/transformfeedback.h"
  36
  37 static void
  38 gen6_update_sol_surfaces(struct brw_context *brw)
  39 {
  40    struct gl_context *ctx = &brw->ctx;
  41    /* BRW_NEW_TRANSFORM_FEEDBACK */
  42    struct gl_transform_feedback_object *xfb_obj =
  43       ctx->TransformFeedback.CurrentObject;
  44    const struct gl_shader_program *shaderprog;
  45    const struct gl_transform_feedback_info *linked_xfb_info;
  46    int i;
  47
  48    if (brw->geometry_program) {
  49       /* BRW_NEW_GEOMETRY_PROGRAM */
  50       shaderprog =
  51          ctx->_Shader->CurrentProgram[MESA_SHADER_GEOMETRY];
  52    } else {
  53       /* BRW_NEW_VERTEX_PROGRAM */
  54       shaderprog =
  55          ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX];
  56    }
  57    linked_xfb_info = &shaderprog->LinkedTransformFeedback;
  58
  59    for (i = 0; i < BRW_MAX_SOL_BINDINGS; ++i) {
  60       const int surf_index = SURF_INDEX_GEN6_SOL_BINDING(i);
  61       if (_mesa_is_xfb_active_and_unpaused(ctx) &&
  62           i < linked_xfb_info->NumOutputs) {
  63          unsigned buffer = linked_xfb_info->Outputs[i].OutputBuffer;
  64          unsigned buffer_offset =
  65             xfb_obj->Offset[buffer] / 4 +
  66             linked_xfb_info->Outputs[i].DstOffset;
  67          if (brw->geometry_program) {
  68             brw_update_sol_surface(
  69                brw, xfb_obj->Buffers[buffer],
  70                &brw->gs.base.surf_offset[surf_index],
  71                linked_xfb_info->Outputs[i].NumComponents,
  72                linked_xfb_info->BufferStride[buffer], buffer_offset);
  73          } else {
  74             brw_update_sol_surface(
  75                brw, xfb_obj->Buffers[buffer],
  76                &brw->ff_gs.surf_offset[surf_index],
  77                linked_xfb_info->Outputs[i].NumComponents,
  78                linked_xfb_info->BufferStride[buffer], buffer_offset);
  79          }
  80       } else {
  81          if (!brw->geometry_program)
  82             brw->ff_gs.surf_offset[surf_index] = 0;
  83          else
  84             brw->gs.base.surf_offset[surf_index] = 0;
  85       }
  86    }
  87
  88    brw->state.dirty.brw |= BRW_NEW_SURFACES;
  89 }
  90
  91 const struct brw_tracked_state gen6_sol_surface = {
  92    .dirty = {
  93       .mesa = 0,
  94       .brw = (BRW_NEW_BATCH |
  95               BRW_NEW_VERTEX_PROGRAM |
  96               BRW_NEW_GEOMETRY_PROGRAM |
  97               BRW_NEW_TRANSFORM_FEEDBACK),
  98       .cache = 0
  99    },
 100    .emit = gen6_update_sol_surfaces,
 101 };
 102
 103 /**
 104  * Constructs the binding table for the WM surface state, which maps unit
 105  * numbers to surface state objects.
 106  */
 107 static void
 108 brw_gs_upload_binding_table(struct brw_context *brw)
 109 {
 110    uint32_t *bind;
 111
 112    if (!brw->geometry_program) {
 113       struct gl_context *ctx = &brw->ctx;
 114       /* BRW_NEW_VERTEX_PROGRAM */
 115       const struct gl_shader_program *shaderprog =
 116          ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX];
 117       bool has_surfaces = false;
 118
 119       if (shaderprog) {
 120          const struct gl_transform_feedback_info *linked_xfb_info =
 121             &shaderprog->LinkedTransformFeedback;
 122          /* Currently we only ever upload surfaces for SOL. */
 123          has_surfaces = linked_xfb_info->NumOutputs != 0;
 124
 125          /* Skip making a binding table if we don't have anything to put in it. */
 126          if (!has_surfaces) {
 127             if (brw->ff_gs.bind_bo_offset != 0) {
 128                brw->state.dirty.brw |= BRW_NEW_GS_BINDING_TABLE;
 129                brw->ff_gs.bind_bo_offset = 0;
 130             }
 131             return;
 132          }
 133       }
 134    }
 135
 136    /* Might want to calculate nr_surfaces first, to avoid taking up so much
 137     * space for the binding table.
 138     */
 139    if (brw->geometry_program) {
 140       bind = brw_state_batch(brw, AUB_TRACE_BINDING_TABLE,
 141                              sizeof(uint32_t) * BRW_MAX_GEN6_GS_SURFACES,
 142                              32, &brw->gs.base.bind_bo_offset);
 143
 144       /* BRW_NEW_SURFACES */
 145       memcpy(bind, brw->gs.base.surf_offset, BRW_MAX_GEN6_GS_SURFACES * sizeof(uint32_t));
 146    } else {
 147       bind = brw_state_batch(brw, AUB_TRACE_BINDING_TABLE,
 148                              sizeof(uint32_t) * BRW_MAX_GEN6_GS_SURFACES,
 149                              32, &brw->ff_gs.bind_bo_offset);
 150
 151       /* BRW_NEW_SURFACES */
 152       memcpy(bind, brw->ff_gs.surf_offset, BRW_MAX_GEN6_GS_SURFACES * sizeof(uint32_t));
 153    }
 154
 155    brw->state.dirty.brw |= BRW_NEW_GS_BINDING_TABLE;
 156 }
 157
 158 const struct brw_tracked_state gen6_gs_binding_table = {
 159    .dirty = {
 160       .mesa = 0,
 161       .brw = (BRW_NEW_BATCH |
 162               BRW_NEW_VERTEX_PROGRAM |
 163               BRW_NEW_GEOMETRY_PROGRAM |
 164               BRW_NEW_SURFACES),
 165       .cache = 0
 166    },
 167    .emit = brw_gs_upload_binding_table,
 168 };
 169
 170 struct gl_transform_feedback_object *
 171 brw_new_transform_feedback(struct gl_context *ctx, GLuint name)
 172 {
 173    struct brw_context *brw = brw_context(ctx);
 174    struct brw_transform_feedback_object *brw_obj =
 175       CALLOC_STRUCT(brw_transform_feedback_object);
 176    if (!brw_obj)
 177       return NULL;
 178
 179    _mesa_init_transform_feedback_object(&brw_obj->base, name);
 180
 181    brw_obj->offset_bo =
 182       drm_intel_bo_alloc(brw->bufmgr, "transform feedback offsets", 16, 64);
 183    brw_obj->prim_count_bo =
 184       drm_intel_bo_alloc(brw->bufmgr, "xfb primitive counts", 4096, 64);
 185
 186    return &brw_obj->base;
 187 }
 188
 189 void
 190 brw_delete_transform_feedback(struct gl_context *ctx,
 191                               struct gl_transform_feedback_object *obj)
 192 {
 193    struct brw_transform_feedback_object *brw_obj =
 194       (struct brw_transform_feedback_object *) obj;
 195
 196    for (unsigned i = 0; i < Elements(obj->Buffers); i++) {
 197       _mesa_reference_buffer_object(ctx, &obj->Buffers[i], NULL);
 198    }
 199
 200    drm_intel_bo_unreference(brw_obj->offset_bo);
 201    drm_intel_bo_unreference(brw_obj->prim_count_bo);
 202
 203    free(brw_obj);
 204 }
 205
 206 void
 207 brw_begin_transform_feedback(struct gl_context *ctx, GLenum mode,
 208                              struct gl_transform_feedback_object *obj)
 209 {
 210    struct brw_context *brw = brw_context(ctx);
 211    const struct gl_shader_program *shaderprog;
 212    const struct gl_transform_feedback_info *linked_xfb_info;
 213    struct gl_transform_feedback_object *xfb_obj =
 214       ctx->TransformFeedback.CurrentObject;
 215
 216    assert(brw->gen == 6);
 217
 218    if (brw->geometry_program) {
 219       /* BRW_NEW_GEOMETRY_PROGRAM */
 220       shaderprog =
 221          ctx->_Shader->CurrentProgram[MESA_SHADER_GEOMETRY];
 222    } else {
 223       /* BRW_NEW_VERTEX_PROGRAM */
 224       shaderprog =
 225          ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX];
 226    }
 227    linked_xfb_info = &shaderprog->LinkedTransformFeedback;
 228
 229    /* Compute the maximum number of vertices that we can write without
 230     * overflowing any of the buffers currently being used for feedback.
 231     */
 232    unsigned max_index
 233       = _mesa_compute_max_transform_feedback_vertices(xfb_obj,
 234                                                       linked_xfb_info);
 235
 236    /* 3DSTATE_GS_SVB_INDEX is non-pipelined. */
 237    intel_emit_post_sync_nonzero_flush(brw);
 238
 239    /* Initialize the SVBI 0 register to zero and set the maximum index. */
 240    BEGIN_BATCH(4);
 241    OUT_BATCH(_3DSTATE_GS_SVB_INDEX << 16 | (4 - 2));
 242    OUT_BATCH(0); /* SVBI 0 */
 243    OUT_BATCH(0); /* starting index */
 244    OUT_BATCH(max_index);
 245    ADVANCE_BATCH();
 246
 247    /* Initialize the rest of the unused streams to sane values.  Otherwise,
 248     * they may indicate that there is no room to write data and prevent
 249     * anything from happening at all.
 250     */
 251    for (int i = 1; i < 4; i++) {
 252       BEGIN_BATCH(4);
 253       OUT_BATCH(_3DSTATE_GS_SVB_INDEX << 16 | (4 - 2));
 254       OUT_BATCH(i << SVB_INDEX_SHIFT);
 255       OUT_BATCH(0); /* starting index */
 256       OUT_BATCH(0xffffffff);
 257       ADVANCE_BATCH();
 258    }
 259 }
 260
 261 void
 262 brw_end_transform_feedback(struct gl_context *ctx,
 263                            struct gl_transform_feedback_object *obj)
 264 {
 265    /* After EndTransformFeedback, it's likely that the client program will try
 266     * to draw using the contents of the transform feedback buffer as vertex
 267     * input.  In order for this to work, we need to flush the data through at
 268     * least the GS stage of the pipeline, and flush out the render cache.  For
 269     * simplicity, just do a full flush.
 270     */
 271    struct brw_context *brw = brw_context(ctx);
 272    intel_batchbuffer_emit_mi_flush(brw);
 273 }