src/gallium/auxiliary/draw/draw_pt_so_emit.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2010 VMware, Inc.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * 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, sub license, 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 portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 #include "draw/draw_private.h"
  29 #include "draw/draw_vs.h"
  30 #include "draw/draw_context.h"
  31 #include "draw/draw_vbuf.h"
  32 #include "draw/draw_vertex.h"
  33 #include "draw/draw_pt.h"
  34
  35 #include "pipe/p_state.h"
  36
  37 #include "util/u_math.h"
  38 #include "util/u_memory.h"
  39
  40 struct pt_so_emit {
  41    struct draw_context *draw;
  42
  43    unsigned input_vertex_stride;
  44    const float (*inputs)[4];
  45
  46    boolean has_so;
  47
  48    unsigned emitted_primitives;
  49    unsigned emitted_vertices;
  50    unsigned generated_primitives;
  51 };
  52
  53
  54 void draw_pt_so_emit_prepare(struct pt_so_emit *emit)
  55 {
  56    struct draw_context *draw = emit->draw;
  57
  58    emit->has_so = (draw->vs.vertex_shader->state.stream_output.num_outputs > 0);
  59
  60    /* if we have a state with outputs make sure we have
  61     * buffers to output to */
  62    if (emit->has_so) {
  63       boolean has_valid_buffer = FALSE;
  64       unsigned i;
  65       for (i = 0; i < draw->so.num_targets; ++i) {
  66          if (draw->so.targets[i]) {
  67             has_valid_buffer = TRUE;
  68             break;
  69          }
  70       }
  71       emit->has_so = has_valid_buffer;
  72    }
  73
  74    if (!emit->has_so)
  75       return;
  76
  77    /* XXX: need to flush to get prim_vbuf.c to release its allocation??
  78     */
  79    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
  80 }
  81
  82 static boolean
  83 is_component_writable(unsigned mask,
  84                       unsigned compo)
  85 {
  86    switch (mask) {
  87    case TGSI_WRITEMASK_NONE:
  88       return FALSE;
  89    case TGSI_WRITEMASK_X:
  90       return compo == 0;
  91    case TGSI_WRITEMASK_Y:
  92       return compo == 1;
  93    case TGSI_WRITEMASK_XY:
  94       return compo == 0 || compo == 1;
  95    case TGSI_WRITEMASK_Z:
  96       return compo == 2;
  97    case TGSI_WRITEMASK_XZ:
  98       return compo == 0 || compo == 2;
  99    case TGSI_WRITEMASK_YZ:
 100       return compo == 1 || compo == 2;
 101    case TGSI_WRITEMASK_XYZ:
 102       return compo == 0 || compo == 1 || compo == 2;
 103    case TGSI_WRITEMASK_W:
 104       return compo == 3;
 105    case TGSI_WRITEMASK_XW:
 106       return compo == 0 || compo == 3;
 107    case TGSI_WRITEMASK_YW:
 108       return compo == 1 || compo == 3;
 109    case TGSI_WRITEMASK_XYW:
 110       return compo == 0 || compo == 1 || compo == 3;
 111    case TGSI_WRITEMASK_ZW:
 112       return compo == 2 || compo == 3;
 113    case TGSI_WRITEMASK_XZW:
 114       return compo == 0 || compo == 1 || compo == 3;
 115    case TGSI_WRITEMASK_YZW:
 116       return compo == 1 || compo == 2 || compo == 4;
 117    case TGSI_WRITEMASK_XYZW:
 118       return compo < 4;
 119    default:
 120       debug_assert(!"Unknown writemask in stream out");
 121       return compo < 4;
 122    }
 123 }
 124
 125 static INLINE int mask_num_comps(int register_mask)
 126 {
 127    int comps = 0;
 128    switch (register_mask) {
 129    case TGSI_WRITEMASK_XYZW:
 130       comps = 4;
 131       break;
 132    case TGSI_WRITEMASK_XYZ:
 133       comps = 3;
 134       break;
 135    case TGSI_WRITEMASK_XY:
 136       comps = 2;
 137       break;
 138    case TGSI_WRITEMASK_X:
 139       comps = 1;
 140       break;
 141    default:
 142       assert(0);
 143       break;
 144    }
 145    return comps;
 146 }
 147
 148 static void so_emit_prim(struct pt_so_emit *so,
 149                          unsigned *indices,
 150                          unsigned num_vertices)
 151 {
 152    unsigned slot, i;
 153    unsigned input_vertex_stride = so->input_vertex_stride;
 154    struct draw_context *draw = so->draw;
 155    const float (*input_ptr)[4];
 156    const struct pipe_stream_output_info *state =
 157       &draw->vs.vertex_shader->state.stream_output;
 158    float *buffer;
 159    int buffer_total_bytes[PIPE_MAX_SO_BUFFERS];
 160
 161    input_ptr = so->inputs;
 162
 163    ++so->generated_primitives;
 164
 165    for (i = 0; i < draw->so.num_targets; i++) {
 166       struct draw_so_target *target = draw->so.targets[i];
 167       buffer_total_bytes[i] = target->internal_offset;
 168    }
 169
 170    /* check have we space to emit prim first - if not don't do anything */
 171    for (i = 0; i < num_vertices; ++i) {
 172       for (slot = 0; slot < state->num_outputs; ++slot) {
 173          unsigned writemask = state->output[slot].register_mask;
 174          int ob = state->output[slot].output_buffer;
 175
 176          if ((buffer_total_bytes[ob] + mask_num_comps(writemask) * sizeof(float)) >
 177              draw->so.targets[ob]->target.buffer_size) {
 178             return;
 179          }
 180          buffer_total_bytes[ob] += mask_num_comps(writemask) * sizeof(float);
 181       }
 182    }
 183
 184    for (i = 0; i < num_vertices; ++i) {
 185       const float (*input)[4];
 186       unsigned total_written_compos = 0;
 187       /*debug_printf("%d) vertex index = %d (prim idx = %d)\n", i, indices[i], prim_idx);*/
 188       input = (const float (*)[4])(
 189          (const char *)input_ptr + (indices[i] * input_vertex_stride));
 190
 191       for (slot = 0; slot < state->num_outputs; ++slot) {
 192          unsigned idx = state->output[slot].register_index;
 193          unsigned writemask = state->output[slot].register_mask;
 194          unsigned written_compos = 0;
 195          unsigned compo;
 196          int ob = state->output[slot].output_buffer;
 197
 198          buffer = (float *)((char *)draw->so.targets[ob]->mapping +
 199                             draw->so.targets[ob]->target.buffer_offset +
 200                             draw->so.targets[ob]->internal_offset);
 201          for (compo = 0; compo < 4; ++compo) {
 202             if (is_component_writable(writemask, compo)) {
 203                buffer[written_compos++] = input[idx][compo];
 204             }
 205          }
 206          draw->so.targets[ob]->internal_offset += written_compos * sizeof(float);
 207          total_written_compos += written_compos;
 208       }
 209    }
 210    so->emitted_vertices += num_vertices;
 211    ++so->emitted_primitives;
 212 }
 213
 214 static void so_point(struct pt_so_emit *so, int idx)
 215 {
 216    unsigned indices[1];
 217
 218    indices[0] = idx;
 219
 220    so_emit_prim(so, indices, 1);
 221 }
 222
 223 static void so_line(struct pt_so_emit *so, int i0, int i1)
 224 {
 225    unsigned indices[2];
 226
 227    indices[0] = i0;
 228    indices[1] = i1;
 229
 230    so_emit_prim(so, indices, 2);
 231 }
 232
 233 static void so_tri(struct pt_so_emit *so, int i0, int i1, int i2)
 234 {
 235    unsigned indices[3];
 236
 237    indices[0] = i0;
 238    indices[1] = i1;
 239    indices[2] = i2;
 240
 241    so_emit_prim(so, indices, 3);
 242 }
 243
 244
 245 #define FUNC         so_run_linear
 246 #define GET_ELT(idx) (start + (idx))
 247 #include "draw_so_emit_tmp.h"
 248
 249
 250 #define FUNC         so_run_elts
 251 #define LOCAL_VARS   const ushort *elts = input_prims->elts;
 252 #define GET_ELT(idx) (elts[start + (idx)])
 253 #include "draw_so_emit_tmp.h"
 254
 255
 256 void draw_pt_so_emit( struct pt_so_emit *emit,
 257                       const struct draw_vertex_info *input_verts,
 258                       const struct draw_prim_info *input_prims )
 259 {
 260    struct draw_context *draw = emit->draw;
 261    struct vbuf_render *render = draw->render;
 262    unsigned start, i;
 263
 264    if (!emit->has_so)
 265       return;
 266
 267    emit->emitted_vertices = 0;
 268    emit->emitted_primitives = 0;
 269    emit->generated_primitives = 0;
 270    emit->input_vertex_stride = input_verts->stride;
 271    emit->inputs = (const float (*)[4])input_verts->verts->data;
 272
 273    /* XXX: need to flush to get prim_vbuf.c to release its allocation??*/
 274    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 275
 276    for (start = i = 0; i < input_prims->primitive_count;
 277         start += input_prims->primitive_lengths[i], i++)
 278    {
 279       unsigned count = input_prims->primitive_lengths[i];
 280
 281       if (input_prims->linear) {
 282          so_run_linear(emit, input_prims, input_verts,
 283                        start, count);
 284       } else {
 285          so_run_elts(emit, input_prims, input_verts,
 286                      start, count);
 287       }
 288    }
 289
 290    render->set_stream_output_info(render,
 291                                   emit->emitted_primitives,
 292                                   emit->emitted_vertices,
 293                                   emit->generated_primitives);
 294 }
 295
 296
 297 struct pt_so_emit *draw_pt_so_emit_create( struct draw_context *draw )
 298 {
 299    struct pt_so_emit *emit = CALLOC_STRUCT(pt_so_emit);
 300    if (!emit)
 301       return NULL;
 302
 303    emit->draw = draw;
 304
 305    return emit;
 306 }
 307
 308 void draw_pt_so_emit_destroy( struct pt_so_emit *emit )
 309 {
 310    FREE(emit);
 311 }