draw_gs.c revision b85a361ccbac956d2842251395c048a4b3f4c440 
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
46boolean
47draw_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
63void 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
73void
74draw_gs_set_constants(struct draw_context *draw,
75                      unsigned slot,
76                      const void *constants,
77                      unsigned size)
78{
79}
80
81
82struct draw_geometry_shader *
83draw_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
136void 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
153void draw_delete_geometry_shader(struct draw_context *draw,
154                                 struct draw_geometry_shader *dgs)
155{
156   FREE(dgs);
157}
158
159/*#define DEBUG_OUTPUTS 1*/
160static INLINE void
161draw_geometry_fetch_outputs(struct draw_geometry_shader *shader,
162                            int num_primitives,
163                            float (**p_output)[4])
164{
165   struct tgsi_exec_machine *machine = shader->machine;
166   unsigned prim_idx, j, slot;
167   float (*output)[4];
168
169   output = *p_output;
170
171   /* Unswizzle all output results.
172    */
173
174   shader->emitted_primitives += num_primitives;
175   for (prim_idx = 0; prim_idx < num_primitives; ++prim_idx) {
176      unsigned num_verts_per_prim = machine->Primitives[prim_idx];
177      shader->emitted_vertices += num_verts_per_prim;
178      for (j = 0; j < num_verts_per_prim; j++) {
179         int idx = (prim_idx * num_verts_per_prim + j) *
180                   shader->info.num_outputs;
181#ifdef DEBUG_OUTPUTS
182         debug_printf("%d) Output vert:\n", idx / shader->info.num_outputs);
183#endif
184         for (slot = 0; slot < shader->info.num_outputs; slot++) {
185            output[slot][0] = machine->Outputs[idx + slot].xyzw[0].f[0];
186            output[slot][1] = machine->Outputs[idx + slot].xyzw[1].f[0];
187            output[slot][2] = machine->Outputs[idx + slot].xyzw[2].f[0];
188            output[slot][3] = machine->Outputs[idx + slot].xyzw[3].f[0];
189#ifdef DEBUG_OUTPUTS
190            debug_printf("\t%d: %f %f %f %f\n", slot,
191                         output[slot][0],
192                         output[slot][1],
193                         output[slot][2],
194                         output[slot][3]);
195#endif
196            debug_assert(!util_is_inf_or_nan(output[slot][0]));
197         }
198         output = (float (*)[4])((char *)output + shader->vertex_size);
199      }
200   }
201   *p_output = output;
202}
203
204
205static void draw_fetch_gs_input(struct draw_geometry_shader *shader,
206                                unsigned *indices,
207                                unsigned num_vertices,
208                                unsigned prim_idx)
209{
210   struct tgsi_exec_machine *machine = shader->machine;
211   unsigned slot, vs_slot, i;
212   unsigned input_vertex_stride = shader->input_vertex_stride;
213   const float (*input_ptr)[4];
214
215   input_ptr = shader->input;
216
217   for (i = 0; i < num_vertices; ++i) {
218      const float (*input)[4];
219      /*debug_printf("%d) vertex index = %d (prim idx = %d)\n", i, indices[i], prim_idx);*/
220      input = (const float (*)[4])(
221         (const char *)input_ptr + (indices[i] * input_vertex_stride));
222      for (slot = 0, vs_slot = 0; slot < shader->info.num_inputs; ++slot) {
223         unsigned idx = i * TGSI_EXEC_MAX_INPUT_ATTRIBS + slot;
224         if (shader->info.input_semantic_name[slot] == TGSI_SEMANTIC_PRIMID) {
225            machine->Inputs[idx].xyzw[0].f[prim_idx] = (float)shader->in_prim_idx;
226            machine->Inputs[idx].xyzw[1].f[prim_idx] = (float)shader->in_prim_idx;
227            machine->Inputs[idx].xyzw[2].f[prim_idx] = (float)shader->in_prim_idx;
228            machine->Inputs[idx].xyzw[3].f[prim_idx] = (float)shader->in_prim_idx;
229         } else {
230            /*debug_printf("\tSlot = %d, vs_slot = %d, idx = %d:\n",
231              slot, vs_slot, idx);*/
232#if 1
233            assert(!util_is_inf_or_nan(input[vs_slot][0]));
234            assert(!util_is_inf_or_nan(input[vs_slot][1]));
235            assert(!util_is_inf_or_nan(input[vs_slot][2]));
236            assert(!util_is_inf_or_nan(input[vs_slot][3]));
237#endif
238            machine->Inputs[idx].xyzw[0].f[prim_idx] = input[vs_slot][0];
239            machine->Inputs[idx].xyzw[1].f[prim_idx] = input[vs_slot][1];
240            machine->Inputs[idx].xyzw[2].f[prim_idx] = input[vs_slot][2];
241            machine->Inputs[idx].xyzw[3].f[prim_idx] = input[vs_slot][3];
242#if 0
243            debug_printf("\t\t%f %f %f %f\n",
244                         machine->Inputs[idx].xyzw[0].f[prim_idx],
245                         machine->Inputs[idx].xyzw[1].f[prim_idx],
246                         machine->Inputs[idx].xyzw[2].f[prim_idx],
247                         machine->Inputs[idx].xyzw[3].f[prim_idx]);
248#endif
249            ++vs_slot;
250         }
251      }
252   }
253}
254
255
256static void gs_flush(struct draw_geometry_shader *shader,
257                     unsigned input_primitives)
258{
259   unsigned out_prim_count;
260   struct tgsi_exec_machine *machine = shader->machine;
261
262   debug_assert(input_primitives > 0 &&
263                input_primitives < 4);
264
265   tgsi_set_exec_mask(machine,
266                      1,
267                      input_primitives > 1,
268                      input_primitives > 2,
269                      input_primitives > 3);
270
271   /* run interpreter */
272   tgsi_exec_machine_run(machine);
273
274   out_prim_count =
275      machine->Temps[TGSI_EXEC_TEMP_PRIMITIVE_I].xyzw[TGSI_EXEC_TEMP_PRIMITIVE_C].u[0];
276
277   draw_geometry_fetch_outputs(shader, out_prim_count,
278                               &shader->tmp_output);
279}
280
281static void gs_point(struct draw_geometry_shader *shader,
282                     int idx)
283{
284   unsigned indices[1];
285
286   indices[0] = idx;
287
288   draw_fetch_gs_input(shader, indices, 1, 0);
289   ++shader->in_prim_idx;
290
291   gs_flush(shader, 1);
292}
293
294static void gs_line(struct draw_geometry_shader *shader,
295                    int i0, int i1)
296{
297   unsigned indices[2];
298
299   indices[0] = i0;
300   indices[1] = i1;
301
302   draw_fetch_gs_input(shader, indices, 2, 0);
303   ++shader->in_prim_idx;
304
305   gs_flush(shader, 1);
306}
307
308static void gs_tri(struct draw_geometry_shader *shader,
309                   int i0, int i1, int i2)
310{
311   unsigned indices[3];
312
313   indices[0] = i0;
314   indices[1] = i1;
315   indices[2] = i2;
316
317   draw_fetch_gs_input(shader, indices, 3, 0);
318   ++shader->in_prim_idx;
319
320   gs_flush(shader, 1);
321}
322
323#define TRIANGLE(gs,i0,i1,i2) gs_tri(gs,i0,i1,i2)
324#define LINE(gs,i0,i1)  gs_line(gs,i0,i1)
325#define POINT(gs,i0)          gs_point(gs,i0)
326#define FUNC gs_run
327#include "draw_gs_tmp.h"
328
329int draw_geometry_shader_run(struct draw_geometry_shader *shader,
330                             const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
331                             const struct draw_vertex_info *input_verts,
332                             const struct draw_prim_info *input_prim,
333                             struct draw_vertex_info *output_verts,
334                             struct draw_prim_info *output_prims )
335{
336   const float (*input)[4] = input_verts->verts;
337   unsigned input_stride = input_verts->vertex_size;
338   unsigned vertex_size = input_verts->vertex_size;
339   struct tgsi_exec_machine *machine = shader->machine;
340   unsigned int i;
341   unsigned num_in_primitives = u_gs_prims_for_vertices(input_prim->prim,
342                                                        input_verts->count);
343
344   output_verts->vertex_size = input_verts->vertex_size;
345   output_verts->stride = input_verts->vertex_size;
346
347   output_verts->count = draw_max_output_vertices(draw,
348                                                  input_prim->prim,
349                                                  input_verts->count);
350
351   output_verts->verts =
352         (struct vertex_header *)MALLOC(vert_info.vertex_size *
353                                        vert_info.count);
354
355
356
357   if (0) debug_printf("%s count = %d (prims = %d)\n", __FUNCTION__,
358                       count, num_in_primitives);
359
360   shader->emitted_vertices = 0;
361   shader->emitted_primitives = 0;
362   shader->vertex_size = vertex_size;
363   shader->tmp_output = (float (*)[4])input_verts->verts->data;
364   shader->in_prim_idx = 0;
365   shader->input_vertex_stride = input_stride;
366   shader->input = input;
367
368   for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
369      machine->Consts[i] = constants[i];
370   }
371
372   gs_run(shader, pipe_prim, count);
373
374   if (shader->emitted_vertices > 0) {
375      memcpy(output, pipeline_verts->data,
376             shader->info.num_outputs * 4 * sizeof(float) +
377             vertex_size * (shader->emitted_vertices -1));
378   }
379
380
381   /* Update prim_info:
382    */
383   output_prims->linear = TRUE;
384   output_prims->elts = NULL;
385   output_prims->primitive_lengths = machine->foo;
386   output_prims->primitive_count = machine->bar;
387
388   return shader->emitted_vertices;
389}
390
391void draw_geometry_shader_delete(struct draw_geometry_shader *shader)
392{
393   FREE((void*) shader->state.tokens);
394   FREE(shader);
395}
396
397void draw_geometry_shader_prepare(struct draw_geometry_shader *shader,
398                                  struct draw_context *draw)
399{
400   if (shader && shader->machine->Tokens != shader->state.tokens) {
401      tgsi_exec_machine_bind_shader(shader->machine,
402                                    shader->state.tokens,
403                                    draw->gs.num_samplers,
404                                    draw->gs.samplers);
405   }
406}
407
408int draw_max_output_vertices(struct draw_context *draw,
409                             unsigned pipe_prim,
410                             unsigned count)
411{
412   unsigned alloc_count = align( count, 4 );
413
414   if (draw->gs.geometry_shader) {
415      unsigned input_primitives = u_gs_prims_for_vertices(pipe_prim,
416                                                          count);
417      /* max GS output is number of input primitives * max output
418       * vertices per each invocation */
419      unsigned gs_max_verts = input_primitives *
420                              draw->gs.geometry_shader->max_output_vertices;
421      if (gs_max_verts > count)
422         alloc_count = align(gs_max_verts, 4);
423   }
424   /*debug_printf("------- alloc count = %d (input = %d)\n",
425                  alloc_count, count);*/
426   return alloc_count;
427}
428