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_gs.h"
  31 #include "draw/draw_context.h"
  32 #include "draw/draw_vbuf.h"
  33 #include "draw/draw_vertex.h"
  34 #include "draw/draw_pt.h"
  35
  36 #include "pipe/p_state.h"
  37
  38 #include "util/u_math.h"
  39 #include "util/u_memory.h"
  40
  41 struct pt_so_emit {
  42    struct draw_context *draw;
  43
  44    unsigned input_vertex_stride;
  45    const float (*inputs)[4];
  46    const float *pre_clip_pos;
  47    boolean has_so;
  48    boolean use_pre_clip_pos;
  49    int pos_idx;
  50    unsigned emitted_primitives;
  51    unsigned emitted_vertices;
  52    unsigned generated_primitives;
  53 };
  54
  55 static const struct pipe_stream_output_info *
  56 draw_so_info(const struct draw_context *draw)
  57 {
  58    const struct pipe_stream_output_info *state = NULL;
  59
  60    if (draw->gs.geometry_shader) {
  61       state = &draw->gs.geometry_shader->state.stream_output;
  62    } else {
  63       state = &draw->vs.vertex_shader->state.stream_output;
  64    }
  65
  66    return state;
  67 }
  68
  69 static INLINE boolean
  70 draw_has_so(const struct draw_context *draw)
  71 {
  72    const struct pipe_stream_output_info *state = draw_so_info(draw);
  73
  74    if (state && state->num_outputs > 0)
  75       return TRUE;
  76
  77    return FALSE;
  78 }
  79
  80 void draw_pt_so_emit_prepare(struct pt_so_emit *emit, boolean use_pre_clip_pos)
  81 {
  82    struct draw_context *draw = emit->draw;
  83
  84    emit->use_pre_clip_pos = use_pre_clip_pos;
  85    emit->has_so = draw_has_so(draw);
  86    if (use_pre_clip_pos)
  87       emit->pos_idx = draw_current_shader_position_output(draw);
  88
  89    /* if we have a state with outputs make sure we have
  90     * buffers to output to */
  91    if (emit->has_so) {
  92       boolean has_valid_buffer = FALSE;
  93       unsigned i;
  94       for (i = 0; i < draw->so.num_targets; ++i) {
  95          if (draw->so.targets[i]) {
  96             has_valid_buffer = TRUE;
  97             break;
  98          }
  99       }
 100       emit->has_so = has_valid_buffer;
 101    }
 102
 103    if (!emit->has_so)
 104       return;
 105
 106    /* XXX: need to flush to get prim_vbuf.c to release its allocation??
 107     */
 108    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 109 }
 110
 111 static void so_emit_prim(struct pt_so_emit *so,
 112                          unsigned *indices,
 113                          unsigned num_vertices)
 114 {
 115    unsigned slot, i;
 116    unsigned input_vertex_stride = so->input_vertex_stride;
 117    struct draw_context *draw = so->draw;
 118    const float (*input_ptr)[4];
 119    const float *pcp_ptr = NULL;
 120    const struct pipe_stream_output_info *state = draw_so_info(draw);
 121    float *buffer;
 122    int buffer_total_bytes[PIPE_MAX_SO_BUFFERS];
 123    boolean buffer_written[PIPE_MAX_SO_BUFFERS] = {0};
 124
 125    input_ptr = so->inputs;
 126    if (so->use_pre_clip_pos)
 127       pcp_ptr = so->pre_clip_pos;
 128
 129    ++so->generated_primitives;
 130
 131    for (i = 0; i < draw->so.num_targets; i++) {
 132       struct draw_so_target *target = draw->so.targets[i];
 133       if (target) {
 134          buffer_total_bytes[i] = target->internal_offset;
 135       } else {
 136          buffer_total_bytes[i] = 0;
 137       }
 138    }
 139
 140    /* check have we space to emit prim first - if not don't do anything */
 141    for (i = 0; i < num_vertices; ++i) {
 142       unsigned ob;
 143       for (slot = 0; slot < state->num_outputs; ++slot) {
 144          unsigned num_comps = state->output[slot].num_components;
 145          int ob = state->output[slot].output_buffer;
 146          unsigned dst_offset = state->output[slot].dst_offset * sizeof(float);
 147          unsigned write_size = num_comps * sizeof(float);
 148          /* If a buffer is missing then that's equivalent to
 149           * an overflow */
 150          if (!draw->so.targets[ob]) {
 151             return;
 152          }
 153          if ((buffer_total_bytes[ob] + write_size + dst_offset) >
 154              draw->so.targets[ob]->target.buffer_size) {
 155             return;
 156          }
 157       }
 158       for (ob = 0; ob < draw->so.num_targets; ++ob) {
 159          buffer_total_bytes[ob] += state->stride[ob] * sizeof(float);
 160       }
 161    }
 162
 163    for (i = 0; i < num_vertices; ++i) {
 164       const float (*input)[4];
 165       const float *pre_clip_pos = NULL;
 166       int ob;
 167
 168       input = (const float (*)[4])(
 169          (const char *)input_ptr + (indices[i] * input_vertex_stride));
 170
 171       if (pcp_ptr)
 172          pre_clip_pos = (const float *)(
 173          (const char *)pcp_ptr + (indices[i] * input_vertex_stride));
 174
 175       for (slot = 0; slot < state->num_outputs; ++slot) {
 176          unsigned idx = state->output[slot].register_index;
 177          unsigned start_comp = state->output[slot].start_component;
 178          unsigned num_comps = state->output[slot].num_components;
 179
 180          ob = state->output[slot].output_buffer;
 181          buffer_written[ob] = TRUE;
 182
 183          buffer = (float *)((char *)draw->so.targets[ob]->mapping +
 184                             draw->so.targets[ob]->target.buffer_offset +
 185                             draw->so.targets[ob]->internal_offset) +
 186             state->output[slot].dst_offset;
 187
 188          if (idx == so->pos_idx && pcp_ptr)
 189             memcpy(buffer, &pre_clip_pos[start_comp],
 190                    num_comps * sizeof(float));
 191          else
 192             memcpy(buffer, &input[idx][start_comp],
 193                    num_comps * sizeof(float));
 194 #if 0
 195          {
 196             int j;
 197             debug_printf("VERT[%d], offset = %d, slot[%d] sc = %d, num_c = %d, idx = %d = [",
 198                          i + draw->so.targets[ob]->emitted_vertices,
 199                          draw->so.targets[ob]->internal_offset,
 200                          slot, start_comp, num_comps, idx);
 201             for (j = 0; j < num_comps; ++j) {
 202                unsigned *ubuffer = (unsigned*)buffer;
 203                debug_printf("%d (0x%x), ", ubuffer[j], ubuffer[j]);
 204             }
 205             debug_printf("]\n");
 206          }
 207 #endif
 208       }
 209       for (ob = 0; ob < draw->so.num_targets; ++ob) {
 210          struct draw_so_target *target = draw->so.targets[ob];
 211          if (target && buffer_written[ob]) {
 212             target->internal_offset += state->stride[ob] * sizeof(float);
 213             target->emitted_vertices += 1;
 214          }
 215       }
 216    }
 217    so->emitted_vertices += num_vertices;
 218    ++so->emitted_primitives;
 219 }
 220
 221 static void so_point(struct pt_so_emit *so, int idx)
 222 {
 223    unsigned indices[1];
 224
 225    indices[0] = idx;
 226
 227    so_emit_prim(so, indices, 1);
 228 }
 229
 230 static void so_line(struct pt_so_emit *so, int i0, int i1)
 231 {
 232    unsigned indices[2];
 233
 234    indices[0] = i0;
 235    indices[1] = i1;
 236
 237    so_emit_prim(so, indices, 2);
 238 }
 239
 240 static void so_tri(struct pt_so_emit *so, int i0, int i1, int i2)
 241 {
 242    unsigned indices[3];
 243
 244    indices[0] = i0;
 245    indices[1] = i1;
 246    indices[2] = i2;
 247
 248    so_emit_prim(so, indices, 3);
 249 }
 250
 251
 252 #define FUNC         so_run_linear
 253 #define GET_ELT(idx) (start + (idx))
 254 #include "draw_so_emit_tmp.h"
 255
 256
 257 #define FUNC         so_run_elts
 258 #define LOCAL_VARS   const ushort *elts = input_prims->elts;
 259 #define GET_ELT(idx) (elts[start + (idx)])
 260 #include "draw_so_emit_tmp.h"
 261
 262
 263 void draw_pt_so_emit( struct pt_so_emit *emit,
 264                       const struct draw_vertex_info *input_verts,
 265                       const struct draw_prim_info *input_prims )
 266 {
 267    struct draw_context *draw = emit->draw;
 268    struct vbuf_render *render = draw->render;
 269    unsigned start, i;
 270
 271    if (!emit->has_so)
 272       return;
 273
 274    if (!draw->so.num_targets)
 275       return;
 276
 277    emit->emitted_vertices = 0;
 278    emit->emitted_primitives = 0;
 279    emit->generated_primitives = 0;
 280    emit->input_vertex_stride = input_verts->stride;
 281    if (emit->use_pre_clip_pos)
 282       emit->pre_clip_pos = input_verts->verts->pre_clip_pos;
 283
 284    emit->inputs = (const float (*)[4])input_verts->verts->data;
 285
 286    /* XXX: need to flush to get prim_vbuf.c to release its allocation??*/
 287    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 288
 289    for (start = i = 0; i < input_prims->primitive_count;
 290         start += input_prims->primitive_lengths[i], i++)
 291    {
 292       unsigned count = input_prims->primitive_lengths[i];
 293
 294       if (input_prims->linear) {
 295          so_run_linear(emit, input_prims, input_verts,
 296                        start, count);
 297       } else {
 298          so_run_elts(emit, input_prims, input_verts,
 299                      start, count);
 300       }
 301    }
 302
 303    render->set_stream_output_info(render,
 304                                   emit->emitted_primitives,
 305                                   emit->emitted_vertices,
 306                                   emit->generated_primitives);
 307 }
 308
 309
 310 struct pt_so_emit *draw_pt_so_emit_create( struct draw_context *draw )
 311 {
 312    struct pt_so_emit *emit = CALLOC_STRUCT(pt_so_emit);
 313    if (!emit)
 314       return NULL;
 315
 316    emit->draw = draw;
 317
 318    return emit;
 319 }
 320
 321 void draw_pt_so_emit_destroy( struct pt_so_emit *emit )
 322 {
 323    FREE(emit);
 324 }