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