1/**********************************************************
2 * Copyright 2008-2009 VMware, Inc.  All rights reserved.
3 *
4 * Permission is hereby granted, free of charge, to any person
5 * obtaining a copy of this software and associated documentation
6 * files (the "Software"), to deal in the Software without
7 * restriction, including without limitation the rights to use, copy,
8 * modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 *
24 **********************************************************/
25
26
27#include "pipe/p_compiler.h"
28#include "pipe/p_shader_tokens.h"
29#include "pipe/p_defines.h"
30#include "tgsi/tgsi_parse.h"
31#include "tgsi/tgsi_dump.h"
32#include "tgsi/tgsi_scan.h"
33#include "util/u_math.h"
34#include "util/u_memory.h"
35#include "util/u_bitmask.h"
36
37#include "svgadump/svga_shader_dump.h"
38
39#include "svga_context.h"
40#include "svga_tgsi.h"
41#include "svga_tgsi_emit.h"
42#include "svga_debug.h"
43
44#include "svga_hw_reg.h"
45#include "svga3d_shaderdefs.h"
46
47
48/* Sinkhole used only in error conditions.
49 */
50static char err_buf[128];
51
52#if 0
53static void svga_destroy_shader_emitter( struct svga_shader_emitter *emit )
54{
55   if (emit->buf != err_buf)
56      FREE(emit->buf);
57}
58#endif
59
60
61static boolean svga_shader_expand( struct svga_shader_emitter *emit )
62{
63   char *new_buf;
64   unsigned newsize = emit->size * 2;
65
66   if(emit->buf != err_buf)
67      new_buf = REALLOC(emit->buf, emit->size, newsize);
68   else
69      new_buf = NULL;
70
71   if (new_buf == NULL) {
72      emit->ptr = err_buf;
73      emit->buf = err_buf;
74      emit->size = sizeof(err_buf);
75      return FALSE;
76   }
77
78   emit->size = newsize;
79   emit->ptr = new_buf + (emit->ptr - emit->buf);
80   emit->buf = new_buf;
81   return TRUE;
82}
83
84static INLINE boolean reserve(  struct svga_shader_emitter *emit,
85                                unsigned nr_dwords )
86{
87   if (emit->ptr - emit->buf + nr_dwords * sizeof(unsigned) >= emit->size) {
88      if (!svga_shader_expand( emit ))
89         return FALSE;
90   }
91
92   return TRUE;
93}
94
95boolean svga_shader_emit_dword( struct svga_shader_emitter *emit,
96                                unsigned dword )
97{
98   if (!reserve(emit, 1))
99      return FALSE;
100
101   *(unsigned *)emit->ptr = dword;
102   emit->ptr += sizeof dword;
103   return TRUE;
104}
105
106boolean svga_shader_emit_dwords( struct svga_shader_emitter *emit,
107                                 const unsigned *dwords,
108                                 unsigned nr )
109{
110   if (!reserve(emit, nr))
111      return FALSE;
112
113   memcpy( emit->ptr, dwords, nr * sizeof *dwords );
114   emit->ptr += nr * sizeof *dwords;
115   return TRUE;
116}
117
118boolean svga_shader_emit_opcode( struct svga_shader_emitter *emit,
119                                 unsigned opcode )
120{
121   SVGA3dShaderInstToken *here;
122
123   if (!reserve(emit, 1))
124      return FALSE;
125
126   here = (SVGA3dShaderInstToken *)emit->ptr;
127   here->value = opcode;
128
129   if (emit->insn_offset) {
130      SVGA3dShaderInstToken *prev = (SVGA3dShaderInstToken *)(emit->buf +
131                                                              emit->insn_offset);
132      prev->size = (here - prev) - 1;
133   }
134
135   emit->insn_offset = emit->ptr - emit->buf;
136   emit->ptr += sizeof(unsigned);
137   return TRUE;
138}
139
140
141static boolean svga_shader_emit_header( struct svga_shader_emitter *emit )
142{
143   SVGA3dShaderVersion header;
144
145   memset( &header, 0, sizeof header );
146
147   switch (emit->unit) {
148   case PIPE_SHADER_FRAGMENT:
149      header.value = SVGA3D_PS_30;
150      break;
151   case PIPE_SHADER_VERTEX:
152      header.value = SVGA3D_VS_30;
153      break;
154   }
155
156   return svga_shader_emit_dword( emit, header.value );
157}
158
159
160/**
161 * Use the shader info to generate a bitmask indicating which generic
162 * inputs are used by the shader.  A set bit indicates that GENERIC[i]
163 * is used.
164 */
165unsigned
166svga_get_generic_inputs_mask(const struct tgsi_shader_info *info)
167{
168   unsigned i, mask = 0x0;
169
170   for (i = 0; i < info->num_inputs; i++) {
171      if (info->input_semantic_name[i] == TGSI_SEMANTIC_GENERIC) {
172         unsigned j = info->input_semantic_index[i];
173         assert(j < sizeof(mask) * 8);
174         mask |= 1 << j;
175      }
176   }
177
178   return mask;
179}
180
181
182/**
183 * Given a mask of used generic variables (as returned by the above functions)
184 * fill in a table which maps those indexes to small integers.
185 * This table is used by the remap_generic_index() function in
186 * svga_tgsi_decl_sm30.c
187 * Example: if generics_mask = binary(1010) it means that GENERIC[1] and
188 * GENERIC[3] are used.  The remap_table will contain:
189 *   table[1] = 0;
190 *   table[3] = 1;
191 * The remaining table entries will be filled in with the next unused
192 * generic index (in this example, 2).
193 */
194void
195svga_remap_generics(unsigned generics_mask,
196                    int8_t remap_table[MAX_GENERIC_VARYING])
197{
198   /* Note texcoord[0] is reserved so start at 1 */
199   unsigned count = 1, i;
200
201   for (i = 0; i < MAX_GENERIC_VARYING; i++) {
202      remap_table[i] = -1;
203   }
204
205   /* for each bit set in generic_mask */
206   while (generics_mask) {
207      unsigned index = ffs(generics_mask) - 1;
208      remap_table[index] = count++;
209      generics_mask &= ~(1 << index);
210   }
211}
212
213
214/**
215 * Use the generic remap table to map a TGSI generic varying variable
216 * index to a small integer.  If the remapping table doesn't have a
217 * valid value for the given index (the table entry is -1) it means
218 * the fragment shader doesn't use that VS output.  Just allocate
219 * the next free value in that case.  Alternately, we could cull
220 * VS instructions that write to register, or replace the register
221 * with a dummy temp register.
222 * XXX TODO: we should do one of the later as it would save precious
223 * texcoord registers.
224 */
225int
226svga_remap_generic_index(int8_t remap_table[MAX_GENERIC_VARYING],
227                         int generic_index)
228{
229   assert(generic_index < MAX_GENERIC_VARYING);
230
231   if (generic_index >= MAX_GENERIC_VARYING) {
232      /* just don't return a random/garbage value */
233      generic_index = MAX_GENERIC_VARYING - 1;
234   }
235
236   if (remap_table[generic_index] == -1) {
237      /* This is a VS output that has no matching PS input.  Find a
238       * free index.
239       */
240      int i, max = 0;
241      for (i = 0; i < MAX_GENERIC_VARYING; i++) {
242         max = MAX2(max, remap_table[i]);
243      }
244      remap_table[generic_index] = max + 1;
245   }
246
247   return remap_table[generic_index];
248}
249
250
251/* Parse TGSI shader and translate to SVGA/DX9 serialized
252 * representation.
253 *
254 * In this function SVGA shader is emitted to an in-memory buffer that
255 * can be dynamically grown.  Once we've finished and know how large
256 * it is, it will be copied to a hardware buffer for upload.
257 */
258static struct svga_shader_result *
259svga_tgsi_translate( const struct svga_shader *shader,
260                     struct svga_compile_key key,
261                     unsigned unit )
262{
263   struct svga_shader_result *result = NULL;
264   struct svga_shader_emitter emit;
265
266   memset(&emit, 0, sizeof(emit));
267
268   emit.size = 1024;
269   emit.buf = MALLOC(emit.size);
270   if (emit.buf == NULL) {
271      goto fail;
272   }
273
274   emit.ptr = emit.buf;
275   emit.unit = unit;
276   emit.key = key;
277
278   tgsi_scan_shader( shader->tokens, &emit.info);
279
280   emit.imm_start = emit.info.file_max[TGSI_FILE_CONSTANT] + 1;
281
282   if (unit == PIPE_SHADER_FRAGMENT)
283      emit.imm_start += key.fkey.num_unnormalized_coords;
284
285   if (unit == PIPE_SHADER_VERTEX) {
286      emit.imm_start += key.vkey.need_prescale ? 2 : 0;
287   }
288
289   emit.nr_hw_float_const = (emit.imm_start + emit.info.file_max[TGSI_FILE_IMMEDIATE] + 1);
290
291   emit.nr_hw_temp = emit.info.file_max[TGSI_FILE_TEMPORARY] + 1;
292
293   if (emit.nr_hw_temp >= SVGA3D_TEMPREG_MAX) {
294      debug_printf("svga: too many temporary registers (%u)\n", emit.nr_hw_temp);
295      goto fail;
296   }
297
298   emit.in_main_func = TRUE;
299
300   if (!svga_shader_emit_header( &emit )) {
301      debug_printf("svga: emit header failed\n");
302      goto fail;
303   }
304
305   if (!svga_shader_emit_instructions( &emit, shader->tokens )) {
306      debug_printf("svga: emit instructions failed\n");
307      goto fail;
308   }
309
310   result = CALLOC_STRUCT(svga_shader_result);
311   if (result == NULL)
312      goto fail;
313
314   result->shader = shader;
315   result->tokens = (const unsigned *)emit.buf;
316   result->nr_tokens = (emit.ptr - emit.buf) / sizeof(unsigned);
317   memcpy(&result->key, &key, sizeof key);
318   result->id = UTIL_BITMASK_INVALID_INDEX;
319
320   if (SVGA_DEBUG & DEBUG_TGSI)
321   {
322      debug_printf( "#####################################\n" );
323      debug_printf( "Shader %u below\n", shader->id );
324      tgsi_dump( shader->tokens, 0 );
325      if (SVGA_DEBUG & DEBUG_TGSI) {
326         debug_printf( "Shader %u compiled below\n", shader->id );
327         svga_shader_dump( result->tokens,
328                           result->nr_tokens ,
329                           FALSE );
330      }
331      debug_printf( "#####################################\n" );
332   }
333
334   return result;
335
336fail:
337   FREE(result);
338   FREE(emit.buf);
339   return NULL;
340}
341
342
343
344
345struct svga_shader_result *
346svga_translate_fragment_program( const struct svga_fragment_shader *fs,
347                                 const struct svga_fs_compile_key *fkey )
348{
349   struct svga_compile_key key;
350
351   memset(&key, 0, sizeof(key));
352
353   memcpy(&key.fkey, fkey, sizeof *fkey);
354
355   memcpy(key.generic_remap_table, fs->generic_remap_table,
356          sizeof(fs->generic_remap_table));
357
358   return svga_tgsi_translate( &fs->base,
359                               key,
360                               PIPE_SHADER_FRAGMENT );
361}
362
363struct svga_shader_result *
364svga_translate_vertex_program( const struct svga_vertex_shader *vs,
365                               const struct svga_vs_compile_key *vkey )
366{
367   struct svga_compile_key key;
368
369   memset(&key, 0, sizeof(key));
370
371   memcpy(&key.vkey, vkey, sizeof *vkey);
372
373   /* Note: we could alternately store the remap table in the vkey but
374    * that would make it larger.  We just regenerate it here instead.
375    */
376   svga_remap_generics(vkey->fs_generic_inputs, key.generic_remap_table);
377
378   return svga_tgsi_translate( &vs->base,
379                               key,
380                               PIPE_SHADER_VERTEX );
381}
382
383
384void svga_destroy_shader_result( struct svga_shader_result *result )
385{
386   FREE((unsigned *)result->tokens);
387   FREE(result);
388}
389
390