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 generated_primitives;
  52    unsigned stream;
  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
 149          if (state->output[slot].stream != so->stream)
 150             continue;
 151          /* If a buffer is missing then that's equivalent to
 152           * an overflow */
 153          if (!draw->so.targets[ob]) {
 154             return;
 155          }
 156          if ((buffer_total_bytes[ob] + write_size + dst_offset) >
 157              draw->so.targets[ob]->target.buffer_size) {
 158             return;
 159          }
 160       }
 161       for (ob = 0; ob < draw->so.num_targets; ++ob) {
 162          buffer_total_bytes[ob] += state->stride[ob] * sizeof(float);
 163       }
 164    }
 165
 166    for (i = 0; i < num_vertices; ++i) {
 167       const float (*input)[4];
 168       const float *pre_clip_pos = NULL;
 169       unsigned  ob;
 170
 171       input = (const float (*)[4])(
 172          (const char *)input_ptr + (indices[i] * input_vertex_stride));
 173
 174       if (pcp_ptr)
 175          pre_clip_pos = (const float *)(
 176          (const char *)pcp_ptr + (indices[i] * input_vertex_stride));
 177
 178       for (slot = 0; slot < state->num_outputs; ++slot) {
 179          unsigned idx = state->output[slot].register_index;
 180          unsigned start_comp = state->output[slot].start_component;
 181          unsigned num_comps = state->output[slot].num_components;
 182          unsigned stream = state->output[slot].stream;
 183
 184          if (stream != so->stream)
 185             continue;
 186          ob = state->output[slot].output_buffer;
 187          buffer_written[ob] = TRUE;
 188
 189          buffer = (float *)((char *)draw->so.targets[ob]->mapping +
 190                             draw->so.targets[ob]->target.buffer_offset +
 191                             draw->so.targets[ob]->internal_offset) +
 192             state->output[slot].dst_offset;
 193
 194          if (idx == so->pos_idx && pcp_ptr && so->stream == 0)
 195             memcpy(buffer, &pre_clip_pos[start_comp],
 196                    num_comps * sizeof(float));
 197          else
 198             memcpy(buffer, &input[idx][start_comp],
 199                    num_comps * sizeof(float));
 200 #if 0
 201          {
 202             int j;
 203             debug_printf("VERT[%d], stream = %d, offset = %d, slot[%d] sc = %d, num_c = %d, idx = %d = [",
 204                          i, stream,
 205                          draw->so.targets[ob]->internal_offset,
 206                          slot, start_comp, num_comps, idx);
 207             for (j = 0; j < num_comps; ++j) {
 208                unsigned *ubuffer = (unsigned*)buffer;
 209                debug_printf("%d (0x%x), ", ubuffer[j], ubuffer[j]);
 210             }
 211             debug_printf("]\n");
 212          }
 213 #endif
 214       }
 215       for (ob = 0; ob < draw->so.num_targets; ++ob) {
 216          struct draw_so_target *target = draw->so.targets[ob];
 217          if (target && buffer_written[ob]) {
 218             target->internal_offset += state->stride[ob] * sizeof(float);
 219          }
 220       }
 221    }
 222    ++so->emitted_primitives;
 223 }
 224
 225 static void so_point(struct pt_so_emit *so, int idx)
 226 {
 227    unsigned indices[1];
 228
 229    indices[0] = idx;
 230
 231    so_emit_prim(so, indices, 1);
 232 }
 233
 234 static void so_line(struct pt_so_emit *so, int i0, int i1)
 235 {
 236    unsigned indices[2];
 237
 238    indices[0] = i0;
 239    indices[1] = i1;
 240
 241    so_emit_prim(so, indices, 2);
 242 }
 243
 244 static void so_tri(struct pt_so_emit *so, int i0, int i1, int i2)
 245 {
 246    unsigned indices[3];
 247
 248    indices[0] = i0;
 249    indices[1] = i1;
 250    indices[2] = i2;
 251
 252    so_emit_prim(so, indices, 3);
 253 }
 254
 255
 256 #define FUNC         so_run_linear
 257 #define GET_ELT(idx) (start + (idx))
 258 #include "draw_so_emit_tmp.h"
 259
 260
 261 #define FUNC         so_run_elts
 262 #define LOCAL_VARS   const ushort *elts = input_prims->elts;
 263 #define GET_ELT(idx) (elts[start + (idx)])
 264 #include "draw_so_emit_tmp.h"
 265
 266
 267 void draw_pt_so_emit( struct pt_so_emit *emit,
 268                       int num_vertex_streams,
 269                       const struct draw_vertex_info *input_verts,
 270                       const struct draw_prim_info *input_prims )
 271 {
 272    struct draw_context *draw = emit->draw;
 273    struct vbuf_render *render = draw->render;
 274    unsigned start, i, stream;
 275
 276    if (!emit->has_so)
 277       return;
 278
 279    if (!draw->so.num_targets)
 280       return;
 281
 282    /* XXX: need to flush to get prim_vbuf.c to release its allocation??*/
 283    draw_do_flush( draw, DRAW_FLUSH_BACKEND );
 284
 285    for (stream = 0; stream < num_vertex_streams; stream++) {
 286       emit->emitted_primitives = 0;
 287       emit->generated_primitives = 0;
 288       if (emit->use_pre_clip_pos)
 289          emit->pre_clip_pos = input_verts[stream].verts->clip_pos;
 290
 291       emit->input_vertex_stride = input_verts[stream].stride;
 292       emit->inputs = (const float (*)[4])input_verts[stream].verts->data;
 293       emit->stream = stream;
 294       for (start = i = 0; i < input_prims[stream].primitive_count;
 295            start += input_prims[stream].primitive_lengths[i], i++)
 296       {
 297          unsigned count = input_prims[stream].primitive_lengths[i];
 298
 299          if (input_prims->linear) {
 300             so_run_linear(emit, &input_prims[stream], &input_verts[stream],
 301                           start, count);
 302          } else {
 303             so_run_elts(emit, &input_prims[stream], &input_verts[stream],
 304                         start, count);
 305          }
 306       }
 307       render->set_stream_output_info(render,
 308                                      stream,
 309                                      emit->emitted_primitives,
 310                                      emit->generated_primitives);
 311    }
 312 }
 313
 314
 315 struct pt_so_emit *draw_pt_so_emit_create( struct draw_context *draw )
 316 {
 317    struct pt_so_emit *emit = CALLOC_STRUCT(pt_so_emit);
 318    if (!emit)
 319       return NULL;
 320
 321    emit->draw = draw;
 322
 323    return emit;
 324 }
 325
 326 void draw_pt_so_emit_destroy( struct pt_so_emit *emit )
 327 {
 328    FREE(emit);
 329 }