src/gallium/auxiliary/draw/draw_vs_exec.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
   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 TUNGSTEN GRAPHICS 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  /*
  29   * Authors:
  30   *   Keith Whitwell <keith@tungstengraphics.com>
  31   *   Brian Paul
  32   */
  33
  34 #include "pipe/p_util.h"
  35 #include "pipe/p_shader_tokens.h"
  36
  37 #include "draw_private.h"
  38 #include "draw_context.h"
  39 #include "draw_vs.h"
  40
  41 #include "tgsi/util/tgsi_parse.h"
  42
  43
  44 struct exec_vertex_shader {
  45    struct draw_vertex_shader base;
  46    struct tgsi_exec_machine *machine;
  47 };
  48
  49 static struct exec_vertex_shader *exec_vertex_shader( struct draw_vertex_shader *vs )
  50 {
  51    return (struct exec_vertex_shader *)vs;
  52 }
  53
  54
  55 /* Not required for run_linear.
  56  */
  57 static void
  58 vs_exec_prepare( struct draw_vertex_shader *shader,
  59                  struct draw_context *draw )
  60 {
  61    struct exec_vertex_shader *evs = exec_vertex_shader(shader);
  62
  63    /* specify the vertex program to interpret/execute */
  64    tgsi_exec_machine_bind_shader(evs->machine,
  65                                  shader->state.tokens,
  66                                  PIPE_MAX_SAMPLERS,
  67                                  NULL /*samplers*/ );
  68
  69    draw_update_vertex_fetch( draw );
  70 }
  71
  72
  73 /**
  74  * Transform vertices with the current vertex program/shader
  75  * Up to four vertices can be shaded at a time.
  76  * \param vbuffer  the input vertex data
  77  * \param elts  indexes of four input vertices
  78  * \param count  number of vertices to shade [1..4]
  79  * \param vOut  array of pointers to four output vertices
  80  */
  81 static boolean
  82 vs_exec_run( struct draw_vertex_shader *shader,
  83              struct draw_context *draw,
  84              const unsigned *elts,
  85              unsigned count,
  86              void *vOut,
  87              unsigned vertex_size)
  88 {
  89    struct exec_vertex_shader *evs = exec_vertex_shader(shader);
  90    struct tgsi_exec_machine *machine = evs->machine;
  91    unsigned int i, j;
  92    unsigned int clipped = 0;
  93    struct tgsi_exec_vector *outputs = 0;
  94    const float *scale = draw->viewport.scale;
  95    const float *trans = draw->viewport.translate;
  96
  97    assert(shader->info.output_semantic_name[0] == TGSI_SEMANTIC_POSITION);
  98
  99    machine->Consts = (const float (*)[4]) draw->user.constants;
 100
 101    if (draw->rasterizer->bypass_vs) {
 102       /* outputs are just the inputs */
 103       outputs = machine->Inputs;
 104    }
 105    else {
 106       outputs = machine->Outputs;
 107    }
 108
 109    for (i = 0; i < count; i += MAX_TGSI_VERTICES) {
 110       unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i);
 111       draw->vertex_fetch.fetch_func( draw, machine, &elts[i], max_vertices );
 112
 113 #if 0
 114       for (j = 0; j < max_vertices; j++) {
 115          unsigned slot;
 116          debug_printf("%d) Input vert:\n", i + j);
 117          for (slot = 0; slot < shader->info.num_inputs; slot++) {
 118             debug_printf("\t%d: %f %f %f %f\n", slot,
 119                          machine->Inputs[slot].xyzw[0].f[j],
 120                          machine->Inputs[slot].xyzw[1].f[j],
 121                          machine->Inputs[slot].xyzw[2].f[j],
 122                          machine->Inputs[slot].xyzw[3].f[j]);
 123          }
 124       }
 125 #endif
 126
 127
 128       if (!draw->rasterizer->bypass_vs) {
 129          /* run interpreter */
 130          tgsi_exec_machine_run( machine );
 131       }
 132
 133       /* store machine results */
 134       for (j = 0; j < max_vertices; j++) {
 135          unsigned slot;
 136          float x, y, z, w;
 137          struct vertex_header *out =
 138             draw_header_from_block(vOut, vertex_size, i + j);
 139
 140          /* Handle attr[0] (position) specially:
 141           *
 142           * XXX: Computing the clipmask should be done in the vertex
 143           * program as a set of DP4 instructions appended to the
 144           * user-provided code.
 145           */
 146          x = out->clip[0] = outputs[0].xyzw[0].f[j];
 147          y = out->clip[1] = outputs[0].xyzw[1].f[j];
 148          z = out->clip[2] = outputs[0].xyzw[2].f[j];
 149          w = out->clip[3] = outputs[0].xyzw[3].f[j];
 150
 151          if (!draw->rasterizer->bypass_clipping) {
 152             out->clipmask = compute_clipmask(out->clip, draw->plane,
 153                                              draw->nr_planes);
 154             clipped += out->clipmask;
 155
 156             /* divide by w */
 157             w = 1.0f / w;
 158             x *= w;
 159             y *= w;
 160             z *= w;
 161          }
 162          else {
 163             out->clipmask = 0;
 164          }
 165          out->edgeflag = 1;
 166          out->vertex_id = UNDEFINED_VERTEX_ID;
 167
 168          if (!draw->identity_viewport) {
 169             /* Viewport mapping */
 170             out->data[0][0] = x * scale[0] + trans[0];
 171             out->data[0][1] = y * scale[1] + trans[1];
 172             out->data[0][2] = z * scale[2] + trans[2];
 173             out->data[0][3] = w;
 174          }
 175          else
 176          {
 177             out->data[0][0] = x;
 178             out->data[0][1] = y;
 179             out->data[0][2] = z;
 180             out->data[0][3] = w;
 181          }
 182
 183          /* Remaining attributes are packed into sequential post-transform
 184           * vertex attrib slots.
 185           */
 186          for (slot = 1; slot < draw->num_vs_outputs; slot++) {
 187             out->data[slot][0] = outputs[slot].xyzw[0].f[j];
 188             out->data[slot][1] = outputs[slot].xyzw[1].f[j];
 189             out->data[slot][2] = outputs[slot].xyzw[2].f[j];
 190             out->data[slot][3] = outputs[slot].xyzw[3].f[j];
 191          }
 192
 193 #if 0 /*DEBUG*/
 194          printf("%d) Post xform vert:\n", i + j);
 195          for (slot = 0; slot < draw->num_vs_outputs; slot++) {
 196             printf("\t%d: %f %f %f %f\n", slot,
 197                    out->data[slot][0],
 198                    out->data[slot][1],
 199                    out->data[slot][2],
 200                    out->data[slot][3]);
 201          }
 202 #endif
 203       } /* loop over vertices */
 204    }
 205    return clipped != 0;
 206 }
 207
 208
 209
 210 /* Simplified vertex shader interface for the pt paths.  Given the
 211  * complexity of code-generating all the above operations together,
 212  * it's time to try doing all the other stuff separately.
 213  */
 214 static void
 215 vs_exec_run_linear( struct draw_vertex_shader *shader,
 216                     const float (*input)[4],
 217                     float (*output)[4],
 218                     const float (*constants)[4],
 219                     unsigned count,
 220                     unsigned input_stride,
 221                     unsigned output_stride )
 222 {
 223    struct exec_vertex_shader *evs = exec_vertex_shader(shader);
 224    struct tgsi_exec_machine *machine = evs->machine;
 225    unsigned int i, j;
 226    unsigned slot;
 227
 228    machine->Consts = constants;
 229
 230    for (i = 0; i < count; i += MAX_TGSI_VERTICES) {
 231       unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i);
 232
 233       /* Swizzle inputs.
 234        */
 235       for (j = 0; j < max_vertices; j++) {
 236 #if 0
 237          debug_printf("%d) Input vert:\n", i + j);
 238          for (slot = 0; slot < shader->info.num_inputs; slot++) {
 239             debug_printf("\t%d: %f %f %f %f\n", slot,
 240                          input[slot][0],
 241                          input[slot][1],
 242                          input[slot][2],
 243                          input[slot][3]);
 244          }
 245 #endif
 246
 247          for (slot = 0; slot < shader->info.num_inputs; slot++) {
 248             machine->Inputs[slot].xyzw[0].f[j] = input[slot][0];
 249             machine->Inputs[slot].xyzw[1].f[j] = input[slot][1];
 250             machine->Inputs[slot].xyzw[2].f[j] = input[slot][2];
 251             machine->Inputs[slot].xyzw[3].f[j] = input[slot][3];
 252          }
 253
 254          input = (const float (*)[4])((const char *)input + input_stride);
 255       }
 256
 257       /* run interpreter */
 258       tgsi_exec_machine_run( machine );
 259
 260       /* Unswizzle all output results.
 261        */
 262       for (j = 0; j < max_vertices; j++) {
 263          for (slot = 0; slot < shader->info.num_outputs; slot++) {
 264             output[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
 265             output[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
 266             output[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
 267             output[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
 268
 269          }
 270
 271 #if 0
 272          debug_printf("%d) Post xform vert:\n", i + j);
 273          for (slot = 0; slot < shader->info.num_outputs; slot++) {
 274             debug_printf("\t%d: %f %f %f %f\n", slot,
 275                          output[slot][0],
 276                          output[slot][1],
 277                          output[slot][2],
 278                          output[slot][3]);
 279          }
 280 #endif
 281
 282          output = (float (*)[4])((char *)output + output_stride);
 283       }
 284
 285    }
 286 }
 287
 288
 289
 290
 291 static void
 292 vs_exec_delete( struct draw_vertex_shader *dvs )
 293 {
 294    FREE((void*) dvs->state.tokens);
 295    FREE( dvs );
 296 }
 297
 298
 299 struct draw_vertex_shader *
 300 draw_create_vs_exec(struct draw_context *draw,
 301                     const struct pipe_shader_state *state)
 302 {
 303    struct exec_vertex_shader *vs = CALLOC_STRUCT( exec_vertex_shader );
 304    uint nt = tgsi_num_tokens(state->tokens);
 305
 306    if (vs == NULL)
 307       return NULL;
 308
 309    /* we make a private copy of the tokens */
 310    vs->base.state.tokens = mem_dup(state->tokens, nt * sizeof(state->tokens[0]));
 311    tgsi_scan_shader(state->tokens, &vs->base.info);
 312
 313
 314    vs->base.prepare = vs_exec_prepare;
 315    vs->base.run = vs_exec_run;
 316    vs->base.run_linear = vs_exec_run_linear;
 317    vs->base.delete = vs_exec_delete;
 318    vs->machine = &draw->machine;
 319
 320
 321    return &vs->base;
 322 }