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 #include "util/u_prim.h"
  41
  42 #define MAX_PRIM_VERTICES 6
  43 /* fixme: move it from here */
  44 #define MAX_PRIMITIVES 64
  45
  46 boolean
  47 draw_gs_init( struct draw_context *draw )
  48 {
  49    draw->gs.machine = tgsi_exec_machine_create();
  50    if (!draw->gs.machine)
  51       return FALSE;
  52
  53    draw->gs.machine->Primitives = align_malloc(
  54       MAX_PRIMITIVES * sizeof(struct tgsi_exec_vector), 16);
  55    if (!draw->gs.machine->Primitives)
  56       return FALSE;
  57    memset(draw->gs.machine->Primitives, 0,
  58           MAX_PRIMITIVES * sizeof(struct tgsi_exec_vector));
  59
  60    return TRUE;
  61 }
  62
  63 void draw_gs_destroy( struct draw_context *draw )
  64 {
  65    if (!draw->gs.machine)
  66       return;
  67
  68    align_free(draw->gs.machine->Primitives);
  69
  70    tgsi_exec_machine_destroy(draw->gs.machine);
  71 }
  72
  73 void
  74 draw_gs_set_constants(struct draw_context *draw,
  75                       unsigned slot,
  76                       const void *constants,
  77                       unsigned size)
  78 {
  79 }
  80
  81
  82 struct draw_geometry_shader *
  83 draw_create_geometry_shader(struct draw_context *draw,
  84                             const struct pipe_shader_state *state)
  85 {
  86    struct draw_geometry_shader *gs;
  87    int i;
  88
  89    gs = CALLOC_STRUCT(draw_geometry_shader);
  90
  91    if (!gs)
  92       return NULL;
  93
  94    gs->draw = draw;
  95    gs->state = *state;
  96    gs->state.tokens = tgsi_dup_tokens(state->tokens);
  97    if (!gs->state.tokens) {
  98       FREE(gs);
  99       return NULL;
 100    }
 101
 102    tgsi_scan_shader(state->tokens, &gs->info);
 103
 104    /* setup the defaults */
 105    gs->input_primitive = PIPE_PRIM_TRIANGLES;
 106    gs->output_primitive = PIPE_PRIM_TRIANGLE_STRIP;
 107    gs->max_output_vertices = 32;
 108
 109    for (i = 0; i < gs->info.num_properties; ++i) {
 110       if (gs->info.properties[i].name ==
 111           TGSI_PROPERTY_GS_INPUT_PRIM)
 112          gs->input_primitive = gs->info.properties[i].data[0];
 113       else if (gs->info.properties[i].name ==
 114                TGSI_PROPERTY_GS_OUTPUT_PRIM)
 115          gs->output_primitive = gs->info.properties[i].data[0];
 116       else if (gs->info.properties[i].name ==
 117                TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES)
 118          gs->max_output_vertices = gs->info.properties[i].data[0];
 119    }
 120
 121    gs->machine = draw->gs.machine;
 122
 123    if (gs)
 124    {
 125       uint i;
 126       for (i = 0; i < gs->info.num_outputs; i++) {
 127          if (gs->info.output_semantic_name[i] == TGSI_SEMANTIC_POSITION &&
 128              gs->info.output_semantic_index[i] == 0)
 129             gs->position_output = i;
 130       }
 131    }
 132
 133    return gs;
 134 }
 135
 136 void draw_bind_geometry_shader(struct draw_context *draw,
 137                                struct draw_geometry_shader *dgs)
 138 {
 139    draw_do_flush(draw, DRAW_FLUSH_STATE_CHANGE);
 140
 141    if (dgs) {
 142       draw->gs.geometry_shader = dgs;
 143       draw->gs.num_gs_outputs = dgs->info.num_outputs;
 144       draw->gs.position_output = dgs->position_output;
 145       draw_geometry_shader_prepare(dgs, draw);
 146    }
 147    else {
 148       draw->gs.geometry_shader = NULL;
 149       draw->gs.num_gs_outputs = 0;
 150    }
 151 }
 152
 153 void draw_delete_geometry_shader(struct draw_context *draw,
 154                                  struct draw_geometry_shader *dgs)
 155 {
 156    FREE(dgs);
 157 }
 158
 159 static void draw_fetch_geometry_input(struct draw_geometry_shader *shader,
 160                                       int start_primitive,
 161                                       int num_primitives,
 162                                       const float (*input_ptr)[4],
 163                                       unsigned input_vertex_stride,
 164                                       unsigned inputs_from_vs)
 165 {
 166    struct tgsi_exec_machine *machine = shader->machine;
 167    unsigned slot, vs_slot, k, j;
 168    unsigned num_vertices = u_vertices_per_prim(shader->input_primitive);
 169    int idx = 0;
 170
 171    for (slot = 0, vs_slot = 0; slot < shader->info.num_inputs; slot++) {
 172       /*debug_printf("Slot = %d (semantic = %d)\n", slot,
 173         shader->info.input_semantic_name[slot]);*/
 174       if (shader->info.input_semantic_name[slot] ==
 175           TGSI_SEMANTIC_PRIMID) {
 176          for (j = 0; j < num_primitives; ++j) {
 177             machine->Inputs[idx].xyzw[0].f[j] = (float)start_primitive + j;
 178             machine->Inputs[idx].xyzw[1].f[j] = (float)start_primitive + j;
 179             machine->Inputs[idx].xyzw[2].f[j] = (float)start_primitive + j;
 180             machine->Inputs[idx].xyzw[3].f[j] = (float)start_primitive + j;
 181          }
 182          ++idx;
 183       } else {
 184          for (j = 0; j < num_primitives; ++j) {
 185             int vidx = idx;
 186             const float (*prim_ptr)[4];
 187             /*debug_printf("    %d) Prim (num_verts = %d)\n", start_primitive + j,
 188               num_vertices);*/
 189             prim_ptr = (const float (*)[4])(
 190                (const char *)input_ptr +
 191                (j * num_vertices * input_vertex_stride));
 192
 193             for (k = 0; k < num_vertices; ++k, ++vidx) {
 194                const float (*input)[4];
 195                input = (const float (*)[4])(
 196                   (const char *)prim_ptr + (k * input_vertex_stride));
 197                vidx = k * TGSI_EXEC_MAX_INPUT_ATTRIBS + slot;
 198                /*debug_printf("\t%d)(%d) Input vert:\n", vidx, k);*/
 199 #if 1
 200                assert(!util_is_inf_or_nan(input[vs_slot][0]));
 201                assert(!util_is_inf_or_nan(input[vs_slot][1]));
 202                assert(!util_is_inf_or_nan(input[vs_slot][2]));
 203                assert(!util_is_inf_or_nan(input[vs_slot][3]));
 204 #endif
 205                machine->Inputs[vidx].xyzw[0].f[j] = input[vs_slot][0];
 206                machine->Inputs[vidx].xyzw[1].f[j] = input[vs_slot][1];
 207                machine->Inputs[vidx].xyzw[2].f[j] = input[vs_slot][2];
 208                machine->Inputs[vidx].xyzw[3].f[j] = input[vs_slot][3];
 209 #if 0
 210                debug_printf("\t\t%d %f %f %f %f\n", slot,
 211                             machine->Inputs[vidx].xyzw[0].f[j],
 212                             machine->Inputs[vidx].xyzw[1].f[j],
 213                             machine->Inputs[vidx].xyzw[2].f[j],
 214                             machine->Inputs[vidx].xyzw[3].f[j]);
 215 #endif
 216             }
 217          }
 218          ++vs_slot;
 219          idx += num_vertices;
 220       }
 221    }
 222 }
 223 /*#define DEBUG_OUTPUTS 1*/
 224 static INLINE void
 225 draw_geometry_fetch_outputs(struct draw_geometry_shader *shader,
 226                             int num_primitives,
 227                             float (**p_output)[4],
 228                             unsigned vertex_size)
 229 {
 230    struct tgsi_exec_machine *machine = shader->machine;
 231    unsigned prim_idx, j, slot;
 232    float (*output)[4];
 233
 234    output = *p_output;
 235
 236    /* Unswizzle all output results.
 237     */
 238
 239    shader->emitted_primitives += num_primitives;
 240    for (prim_idx = 0; prim_idx < num_primitives; ++prim_idx) {
 241       unsigned num_verts_per_prim = machine->Primitives[prim_idx];
 242       shader->emitted_vertices += num_verts_per_prim;
 243       for (j = 0; j < num_verts_per_prim; j++) {
 244          int idx = (prim_idx * num_verts_per_prim + j) *
 245                    shader->info.num_outputs;
 246 #ifdef DEBUG_OUTPUTS
 247          debug_printf("%d) Output vert:\n", idx / shader->info.num_outputs);
 248 #endif
 249          for (slot = 0; slot < shader->info.num_outputs; slot++) {
 250             output[slot][0] = machine->Outputs[idx + slot].xyzw[0].f[0];
 251             output[slot][1] = machine->Outputs[idx + slot].xyzw[1].f[0];
 252             output[slot][2] = machine->Outputs[idx + slot].xyzw[2].f[0];
 253             output[slot][3] = machine->Outputs[idx + slot].xyzw[3].f[0];
 254 #ifdef DEBUG_OUTPUTS
 255             debug_printf("\t%d: %f %f %f %f\n", slot,
 256                          output[slot][0],
 257                          output[slot][1],
 258                          output[slot][2],
 259                          output[slot][3]);
 260 #endif
 261             debug_assert(!util_is_inf_or_nan(output[slot][0]));
 262          }
 263          output = (float (*)[4])((char *)output + vertex_size);
 264       }
 265    }
 266    *p_output = output;
 267 }
 268
 269 int draw_geometry_shader_run(struct draw_geometry_shader *shader,
 270                              unsigned pipe_prim,
 271                              const float (*input)[4],
 272                              float (*output)[4],
 273                              const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
 274                              unsigned count,
 275                              unsigned input_stride,
 276                              unsigned vertex_size)
 277 {
 278    struct tgsi_exec_machine *machine = shader->machine;
 279    unsigned int i;
 280    unsigned num_in_primitives =
 281       u_gs_prims_for_vertices(pipe_prim, count);
 282    unsigned inputs_from_vs = 0;
 283    float (*tmp_output)[4];
 284    unsigned alloc_count = draw_max_output_vertices(shader->draw,
 285                                                    pipe_prim,
 286                                                    count);
 287    /* this is bad, but we can't be overwriting the output array
 288     * because it's the same as input array here */
 289    struct vertex_header *pipeline_verts =
 290       (struct vertex_header *)MALLOC(vertex_size * alloc_count);
 291
 292    if (0) debug_printf("%s count = %d (prims = %d)\n", __FUNCTION__,
 293                        count, num_in_primitives);
 294
 295    shader->emitted_vertices = 0;
 296    shader->emitted_primitives = 0;
 297
 298    for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
 299       machine->Consts[i] = constants[i];
 300    }
 301
 302    for (i = 0; i < shader->info.num_inputs; ++i) {
 303       if (shader->info.input_semantic_name[i] != TGSI_SEMANTIC_PRIMID)
 304          ++inputs_from_vs;
 305    }
 306
 307    tmp_output = (      float (*)[4])pipeline_verts->data;
 308
 309    for (i = 0; i < num_in_primitives; ++i) {
 310       unsigned int max_input_primitives = 1;
 311       /* FIXME: handle all the primitives produced by the gs, not just
 312        * the first one */
 313       unsigned out_prim_count;
 314
 315       draw_fetch_geometry_input(shader, i, max_input_primitives, input,
 316                                 input_stride, inputs_from_vs);
 317
 318       tgsi_set_exec_mask(machine,
 319                          1,
 320                          max_input_primitives > 1,
 321                          max_input_primitives > 2,
 322                          max_input_primitives > 3);
 323
 324       /* run interpreter */
 325       tgsi_exec_machine_run(machine);
 326
 327       out_prim_count =
 328          machine->Temps[TGSI_EXEC_TEMP_PRIMITIVE_I].xyzw[TGSI_EXEC_TEMP_PRIMITIVE_C].u[0];
 329
 330       draw_geometry_fetch_outputs(shader, out_prim_count,
 331                                   &tmp_output, vertex_size);
 332    }
 333
 334    memcpy(output, pipeline_verts->data,
 335           shader->info.num_outputs * 4 * sizeof(float) +
 336           vertex_size * (shader->emitted_vertices -1));
 337
 338    FREE(pipeline_verts);
 339    return shader->emitted_vertices;
 340 }
 341
 342 void draw_geometry_shader_delete(struct draw_geometry_shader *shader)
 343 {
 344    FREE((void*) shader->state.tokens);
 345    FREE(shader);
 346 }
 347
 348 void draw_geometry_shader_prepare(struct draw_geometry_shader *shader,
 349                                   struct draw_context *draw)
 350 {
 351    if (shader && shader->machine->Tokens != shader->state.tokens) {
 352       tgsi_exec_machine_bind_shader(shader->machine,
 353                                     shader->state.tokens,
 354                                     draw->gs.num_samplers,
 355                                     draw->gs.samplers);
 356    }
 357 }
 358
 359 int draw_max_output_vertices(struct draw_context *draw,
 360                              unsigned pipe_prim,
 361                              unsigned count)
 362 {
 363    unsigned alloc_count = align( count, 4 );
 364
 365    if (draw->gs.geometry_shader) {
 366       unsigned input_primitives = u_gs_prims_for_vertices(pipe_prim,
 367                                                           count);
 368       /* max GS output is number of input primitives * max output
 369        * vertices per each invocation */
 370       unsigned gs_max_verts = input_primitives *
 371                               draw->gs.geometry_shader->max_output_vertices;
 372       if (gs_max_verts > count)
 373          alloc_count = align(gs_max_verts, 4);
 374    }
 375    /*debug_printf("------- alloc count = %d (input = %d)\n",
 376                   alloc_count, count);*/
 377    return alloc_count;
 378 }