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_shader.h"
41#include "svga_tgsi.h"
42#include "svga_tgsi_emit.h"
43#include "svga_debug.h"
44
45#include "svga_hw_reg.h"
46#include "svga3d_shaderdefs.h"
47
48
49/* Sinkhole used only in error conditions.
50 */
51static char err_buf[128];
52
53
54static boolean
55svga_shader_expand(struct svga_shader_emitter *emit)
56{
57   char *new_buf;
58   unsigned newsize = emit->size * 2;
59
60   if (emit->buf != err_buf)
61      new_buf = REALLOC(emit->buf, emit->size, newsize);
62   else
63      new_buf = NULL;
64
65   if (!new_buf) {
66      emit->ptr = err_buf;
67      emit->buf = err_buf;
68      emit->size = sizeof(err_buf);
69      return FALSE;
70   }
71
72   emit->size = newsize;
73   emit->ptr = new_buf + (emit->ptr - emit->buf);
74   emit->buf = new_buf;
75   return TRUE;
76}
77
78
79static inline boolean
80reserve(struct svga_shader_emitter *emit, unsigned nr_dwords)
81{
82   if (emit->ptr - emit->buf + nr_dwords * sizeof(unsigned) >= emit->size) {
83      if (!svga_shader_expand(emit)) {
84         return FALSE;
85      }
86   }
87
88   return TRUE;
89}
90
91
92boolean
93svga_shader_emit_dword(struct svga_shader_emitter * emit, unsigned dword)
94{
95   if (!reserve(emit, 1))
96      return FALSE;
97
98   *(unsigned *) emit->ptr = dword;
99   emit->ptr += sizeof dword;
100   return TRUE;
101}
102
103
104boolean
105svga_shader_emit_dwords(struct svga_shader_emitter * emit,
106                        const unsigned *dwords, unsigned nr)
107{
108   if (!reserve(emit, nr))
109      return FALSE;
110
111   memcpy(emit->ptr, dwords, nr * sizeof *dwords);
112   emit->ptr += nr * sizeof *dwords;
113   return TRUE;
114}
115
116
117boolean
118svga_shader_emit_opcode(struct svga_shader_emitter * emit, unsigned opcode)
119{
120   SVGA3dShaderInstToken *here;
121
122   if (!reserve(emit, 1))
123      return FALSE;
124
125   here = (SVGA3dShaderInstToken *) emit->ptr;
126   here->value = opcode;
127
128   if (emit->insn_offset) {
129      SVGA3dShaderInstToken *prev =
130         (SVGA3dShaderInstToken *) (emit->buf + emit->insn_offset);
131      prev->size = (here - prev) - 1;
132   }
133
134   emit->insn_offset = emit->ptr - emit->buf;
135   emit->ptr += sizeof(unsigned);
136   return TRUE;
137}
138
139
140static boolean
141svga_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 * Parse TGSI shader and translate to SVGA/DX9 serialized
162 * representation.
163 *
164 * In this function SVGA shader is emitted to an in-memory buffer that
165 * can be dynamically grown.  Once we've finished and know how large
166 * it is, it will be copied to a hardware buffer for upload.
167 */
168struct svga_shader_variant *
169svga_tgsi_vgpu9_translate(struct svga_context *svga,
170                          const struct svga_shader *shader,
171                          const struct svga_compile_key *key,
172                          enum pipe_shader_type unit)
173{
174   struct svga_shader_variant *variant = NULL;
175   struct svga_shader_emitter emit;
176
177   SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_TGSIVGPU9TRANSLATE);
178
179   memset(&emit, 0, sizeof(emit));
180
181   emit.size = 1024;
182   emit.buf = MALLOC(emit.size);
183   if (emit.buf == NULL) {
184      goto fail;
185   }
186
187   emit.ptr = emit.buf;
188   emit.unit = unit;
189   emit.key = *key;
190
191   tgsi_scan_shader(shader->tokens, &emit.info);
192
193   emit.imm_start = emit.info.file_max[TGSI_FILE_CONSTANT] + 1;
194
195   if (unit == PIPE_SHADER_FRAGMENT)
196      emit.imm_start += key->num_unnormalized_coords;
197
198   if (unit == PIPE_SHADER_VERTEX) {
199      emit.imm_start += key->vs.need_prescale ? 2 : 0;
200   }
201
202   emit.nr_hw_float_const =
203      (emit.imm_start + emit.info.file_max[TGSI_FILE_IMMEDIATE] + 1);
204
205   emit.nr_hw_temp = emit.info.file_max[TGSI_FILE_TEMPORARY] + 1;
206
207   if (emit.nr_hw_temp >= SVGA3D_TEMPREG_MAX) {
208      debug_printf("svga: too many temporary registers (%u)\n",
209                   emit.nr_hw_temp);
210      goto fail;
211   }
212
213   if (emit.info.indirect_files & (1 << TGSI_FILE_TEMPORARY)) {
214      debug_printf(
215         "svga: indirect indexing of temporary registers is not supported.\n");
216      goto fail;
217   }
218
219   emit.in_main_func = TRUE;
220
221   if (!svga_shader_emit_header(&emit)) {
222      debug_printf("svga: emit header failed\n");
223      goto fail;
224   }
225
226   if (!svga_shader_emit_instructions(&emit, shader->tokens)) {
227      debug_printf("svga: emit instructions failed\n");
228      goto fail;
229   }
230
231   variant = svga_new_shader_variant(svga, unit);
232   if (!variant)
233      goto fail;
234
235   variant->shader = shader;
236   variant->tokens = (const unsigned *) emit.buf;
237   variant->nr_tokens = (emit.ptr - emit.buf) / sizeof(unsigned);
238   memcpy(&variant->key, key, sizeof(*key));
239   variant->id = UTIL_BITMASK_INVALID_INDEX;
240
241   if (unit == PIPE_SHADER_FRAGMENT) {
242      struct svga_fs_variant *fs_variant = svga_fs_variant(variant);
243
244      fs_variant->pstipple_sampler_unit = emit.pstipple_sampler_unit;
245
246      /* If there was exactly one write to a fragment shader output register
247       * and it came from a constant buffer, we know all fragments will have
248       * the same color (except for blending).
249       */
250      fs_variant->constant_color_output =
251         emit.constant_color_output && emit.num_output_writes == 1;
252   }
253
254#if 0
255   if (!svga_shader_verify(variant->tokens, variant->nr_tokens) ||
256       SVGA_DEBUG & DEBUG_TGSI) {
257      debug_printf("#####################################\n");
258      debug_printf("Shader %u below\n", shader->id);
259      tgsi_dump(shader->tokens, 0);
260      if (SVGA_DEBUG & DEBUG_TGSI) {
261         debug_printf("Shader %u compiled below\n", shader->id);
262         svga_shader_dump(variant->tokens, variant->nr_tokens, FALSE);
263      }
264      debug_printf("#####################################\n");
265   }
266#endif
267
268   goto done;
269
270fail:
271   FREE(variant);
272   if (emit.buf != err_buf)
273      FREE(emit.buf);
274   variant = NULL;
275
276done:
277   SVGA_STATS_TIME_POP(svga_sws(svga));
278   return variant;
279}
280