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