src/gallium/auxiliary/draw/draw_gs.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2009 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 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 #include "draw_gs.h"
  29
  30 #include "draw_private.h"
  31 #include "draw_context.h"
  32
  33 #include "tgsi/tgsi_parse.h"
  34 #include "tgsi/tgsi_exec.h"
  35
  36 #include "pipe/p_shader_tokens.h"
  37
  38 #include "util/u_math.h"
  39 #include "util/u_memory.h"
  40
  41 #define MAX_PRIM_VERTICES 6
  42 /* fixme: move it from here */
  43 #define MAX_PRIMITIVES 64
  44
  45 boolean
  46 draw_gs_init( struct draw_context *draw )
  47 {
  48    draw->gs.machine = tgsi_exec_machine_create();
  49    if (!draw->gs.machine)
  50       return FALSE;
  51
  52    draw->gs.machine->Primitives = align_malloc(
  53       MAX_PRIMITIVES * sizeof(struct tgsi_exec_vector), 16);
  54    if (!draw->gs.machine->Primitives)
  55       return FALSE;
  56    memset(draw->gs.machine->Primitives, 0,
  57           MAX_PRIMITIVES * sizeof(struct tgsi_exec_vector));
  58
  59    return TRUE;
  60 }
  61
  62 void draw_gs_destroy( struct draw_context *draw )
  63 {
  64    if (!draw->gs.machine)
  65       return;
  66
  67    align_free(draw->gs.machine->Primitives);
  68
  69    tgsi_exec_machine_destroy(draw->gs.machine);
  70 }
  71
  72 void draw_gs_set_constants( struct draw_context *draw,
  73                             const float (*constants)[4],
  74                             unsigned size )
  75 {
  76 }
  77
  78
  79 struct draw_geometry_shader *
  80 draw_create_geometry_shader(struct draw_context *draw,
  81                             const struct pipe_shader_state *state)
  82 {
  83    struct draw_geometry_shader *gs;
  84    int i;
  85
  86    gs = CALLOC_STRUCT(draw_geometry_shader);
  87
  88    if (!gs)
  89       return NULL;
  90
  91    gs->state = *state;
  92    gs->state.tokens = tgsi_dup_tokens(state->tokens);
  93    if (!gs->state.tokens) {
  94       FREE(gs);
  95       return NULL;
  96    }
  97
  98    tgsi_scan_shader(state->tokens, &gs->info);
  99
 100    /* setup the defaults */
 101    gs->input_primitive = PIPE_PRIM_TRIANGLES;
 102    gs->output_primitive = PIPE_PRIM_TRIANGLE_STRIP;
 103    gs->max_output_vertices = 32;
 104
 105    for (i = 0; i < gs->info.num_properties; ++i) {
 106       if (gs->info.properties[i].name ==
 107           TGSI_PROPERTY_GS_INPUT_PRIM)
 108          gs->input_primitive = gs->info.properties[i].data[0];
 109       else if (gs->info.properties[i].name ==
 110                TGSI_PROPERTY_GS_OUTPUT_PRIM)
 111          gs->output_primitive = gs->info.properties[i].data[0];
 112       else if (gs->info.properties[i].name ==
 113                TGSI_PROPERTY_GS_MAX_VERTICES)
 114          gs->max_output_vertices = gs->info.properties[i].data[0];
 115    }
 116
 117    gs->machine = draw->gs.machine;
 118
 119    if (gs)
 120    {
 121       uint i;
 122       for (i = 0; i < gs->info.num_outputs; i++) {
 123          if (gs->info.output_semantic_name[i] == TGSI_SEMANTIC_POSITION &&
 124              gs->info.output_semantic_index[i] == 0)
 125             gs->position_output = i;
 126       }
 127    }
 128
 129    return gs;
 130 }
 131
 132 void draw_bind_geometry_shader(struct draw_context *draw,
 133                                struct draw_geometry_shader *dgs)
 134 {
 135    draw_do_flush(draw, DRAW_FLUSH_STATE_CHANGE);
 136
 137    if (dgs) {
 138       draw->gs.geometry_shader = dgs;
 139       draw->gs.num_gs_outputs = dgs->info.num_outputs;
 140       draw->gs.position_output = dgs->position_output;
 141       draw_geometry_shader_prepare(dgs, draw);
 142    }
 143    else {
 144       draw->gs.geometry_shader = NULL;
 145       draw->gs.num_gs_outputs = 0;
 146    }
 147 }
 148
 149 void draw_delete_geometry_shader(struct draw_context *draw,
 150                                  struct draw_geometry_shader *dgs)
 151 {
 152    FREE(dgs);
 153 }
 154
 155 static INLINE int num_vertices_for_prim(int prim)
 156 {
 157    switch(prim) {
 158    case PIPE_PRIM_POINTS:
 159       return 1;
 160    case PIPE_PRIM_LINES:
 161       return 2;
 162    case PIPE_PRIM_LINE_LOOP:
 163       return 2;
 164    case PIPE_PRIM_LINE_STRIP:
 165       return 2;
 166    case PIPE_PRIM_TRIANGLES:
 167       return 3;
 168    case PIPE_PRIM_TRIANGLE_STRIP:
 169       return 3;
 170    case PIPE_PRIM_TRIANGLE_FAN:
 171       return 3;
 172    case PIPE_PRIM_LINES_ADJACENCY:
 173    case PIPE_PRIM_LINE_STRIP_ADJACENCY:
 174       return 4;
 175    case PIPE_PRIM_TRIANGLES_ADJACENCY:
 176    case PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY:
 177       return 6;
 178    default:
 179       assert(!"Bad geometry shader input");
 180       return 0;
 181    }
 182 }
 183
 184 static void draw_fetch_geometry_input(struct draw_geometry_shader *shader,
 185                                       int start_primitive,
 186                                       int num_primitives,
 187                                       const float (*input_ptr)[4],
 188                                       unsigned input_vertex_stride,
 189                                       unsigned inputs_from_vs)
 190 {
 191    struct tgsi_exec_machine *machine = shader->machine;
 192    unsigned slot, vs_slot, k, j;
 193    unsigned num_vertices = num_vertices_for_prim(shader->input_primitive);
 194    int idx = 0;
 195
 196    for (slot = 0, vs_slot = 0; slot < shader->info.num_inputs; slot++) {
 197       /*debug_printf("Slot = %d (semantic = %d)\n", slot,
 198         shader->info.input_semantic_name[slot]);*/
 199       if (shader->info.input_semantic_name[slot] ==
 200           TGSI_SEMANTIC_PRIMID) {
 201          for (j = 0; j < num_primitives; ++j) {
 202             machine->Inputs[idx].xyzw[0].f[j] = (float)start_primitive + j;
 203             machine->Inputs[idx].xyzw[1].f[j] = (float)start_primitive + j;
 204             machine->Inputs[idx].xyzw[2].f[j] = (float)start_primitive + j;
 205             machine->Inputs[idx].xyzw[3].f[j] = (float)start_primitive + j;
 206          }
 207          ++idx;
 208       } else {
 209          for (j = 0; j < num_primitives; ++j) {
 210             int vidx = idx;
 211             const float (*prim_ptr)[4];
 212             /*debug_printf("    %d) Prim (num_verts = %d)\n", start_primitive + j,
 213               num_vertices);*/
 214             prim_ptr = (const float (*)[4])(
 215                (const char *)input_ptr +
 216                (j * num_vertices * input_vertex_stride));
 217
 218             for (k = 0; k < num_vertices; ++k, ++vidx) {
 219                const float (*input)[4];
 220                input = (const float (*)[4])(
 221                   (const char *)prim_ptr + (k * input_vertex_stride));
 222                vidx = k * TGSI_EXEC_MAX_INPUT_ATTRIBS + slot;
 223                /*debug_printf("\t%d)(%d) Input vert:\n", vidx, k);*/
 224 #if 1
 225                assert(!util_is_inf_or_nan(input[vs_slot][0]));
 226                assert(!util_is_inf_or_nan(input[vs_slot][1]));
 227                assert(!util_is_inf_or_nan(input[vs_slot][2]));
 228                assert(!util_is_inf_or_nan(input[vs_slot][3]));
 229 #endif
 230                machine->Inputs[vidx].xyzw[0].f[j] = input[vs_slot][0];
 231                machine->Inputs[vidx].xyzw[1].f[j] = input[vs_slot][1];
 232                machine->Inputs[vidx].xyzw[2].f[j] = input[vs_slot][2];
 233                machine->Inputs[vidx].xyzw[3].f[j] = input[vs_slot][3];
 234 #if 0
 235                debug_printf("\t\t%d %f %f %f %f\n", slot,
 236                             machine->Inputs[vidx].xyzw[0].f[j],
 237                             machine->Inputs[vidx].xyzw[1].f[j],
 238                             machine->Inputs[vidx].xyzw[2].f[j],
 239                             machine->Inputs[vidx].xyzw[3].f[j]);
 240 #endif
 241             }
 242          }
 243          ++vs_slot;
 244          idx += num_vertices;
 245       }
 246    }
 247 }
 248
 249 static INLINE void
 250 draw_geometry_fetch_outputs(struct draw_geometry_shader *shader,
 251                             int num_primitives,
 252                             float (*output)[4],
 253                             unsigned vertex_size)
 254 {
 255    struct tgsi_exec_machine *machine = shader->machine;
 256    unsigned prim_idx, j, slot;
 257
 258    /* Unswizzle all output results.
 259     */
 260    /* FIXME: handle all the primitives produced by the gs, not just
 261     * the first one
 262     unsigned prim_count =
 263     mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0];*/
 264    for (prim_idx = 0; prim_idx < num_primitives; ++prim_idx) {
 265       unsigned num_verts_per_prim = machine->Primitives[0];
 266       for (j = 0; j < num_verts_per_prim; j++) {
 267          int idx = (prim_idx * num_verts_per_prim + j) *
 268                    shader->info.num_outputs;
 269 #ifdef DEBUG_OUTPUTS
 270          debug_printf("%d) Output vert:\n", idx);
 271 #endif
 272          for (slot = 0; slot < shader->info.num_outputs; slot++) {
 273             output[slot][0] = machine->Outputs[idx + slot].xyzw[0].f[prim_idx];
 274             output[slot][1] = machine->Outputs[idx + slot].xyzw[1].f[prim_idx];
 275             output[slot][2] = machine->Outputs[idx + slot].xyzw[2].f[prim_idx];
 276             output[slot][3] = machine->Outputs[idx + slot].xyzw[3].f[prim_idx];
 277 #ifdef DEBUG_OUTPUTS
 278             debug_printf("\t%d: %f %f %f %f\n", slot,
 279                          output[slot][0],
 280                          output[slot][1],
 281                          output[slot][2],
 282                          output[slot][3]);
 283 #endif
 284             debug_assert(!util_is_inf_or_nan(output[slot][0]));
 285          }
 286          output = (float (*)[4])((char *)output + vertex_size);
 287       }
 288    }
 289 }
 290
 291 void draw_geometry_shader_run(struct draw_geometry_shader *shader,
 292                               const float (*input)[4],
 293                               float (*output)[4],
 294                               const float (*constants)[4],
 295                               unsigned count,
 296                               unsigned input_stride,
 297                               unsigned vertex_size)
 298 {
 299    struct tgsi_exec_machine *machine = shader->machine;
 300    unsigned int i;
 301    unsigned num_vertices = num_vertices_for_prim(shader->input_primitive);
 302    unsigned num_primitives = count/num_vertices;
 303    unsigned inputs_from_vs = 0;
 304
 305    machine->Consts = constants;
 306
 307    for (i = 0; i < shader->info.num_inputs; ++i) {
 308       if (shader->info.input_semantic_name[i] != TGSI_SEMANTIC_PRIMID)
 309          ++inputs_from_vs;
 310    }
 311
 312    for (i = 0; i < num_primitives; ++i) {
 313       unsigned int max_primitives = 1;
 314
 315       draw_fetch_geometry_input(shader, i, max_primitives, input,
 316                                 input_stride, inputs_from_vs);
 317
 318       tgsi_set_exec_mask(machine,
 319                          1,
 320                          max_primitives > 1,
 321                          max_primitives > 2,
 322                          max_primitives > 3);
 323
 324       /* run interpreter */
 325       tgsi_exec_machine_run(machine);
 326
 327       draw_geometry_fetch_outputs(shader, max_primitives,
 328                                   output, vertex_size);
 329    }
 330 }
 331
 332 void draw_geometry_shader_delete(struct draw_geometry_shader *shader)
 333 {
 334    FREE((void*) shader->state.tokens);
 335    FREE(shader);
 336 }
 337
 338 void draw_geometry_shader_prepare(struct draw_geometry_shader *shader,
 339                                   struct draw_context *draw)
 340 {
 341     if (shader->machine->Tokens != shader->state.tokens) {
 342        tgsi_exec_machine_bind_shader(shader->machine,
 343                                      shader->state.tokens,
 344                                      draw->gs.num_samplers,
 345                                      draw->gs.samplers);
 346     }
 347 }