r300g: One less assert.
[mesa.git] / src / gallium / drivers / r300 / r300_tgsi_to_rc.c
1 /*
2 * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com>
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE COPYRIGHT HOLDER(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
22
23 #include "r300_tgsi_to_rc.h"
24
25 #include "radeon_compiler.h"
26 #include "radeon_program.h"
27
28 #include "tgsi/tgsi_parse.h"
29 #include "tgsi/tgsi_scan.h"
30 #include "tgsi/tgsi_util.h"
31
32 #include "util/u_debug.h"
33
34 static unsigned translate_opcode(unsigned opcode)
35 {
36 switch(opcode) {
37 case TGSI_OPCODE_ARL: return RC_OPCODE_ARL;
38 case TGSI_OPCODE_MOV: return RC_OPCODE_MOV;
39 case TGSI_OPCODE_LIT: return RC_OPCODE_LIT;
40 case TGSI_OPCODE_RCP: return RC_OPCODE_RCP;
41 case TGSI_OPCODE_RSQ: return RC_OPCODE_RSQ;
42 case TGSI_OPCODE_EXP: return RC_OPCODE_EXP;
43 case TGSI_OPCODE_LOG: return RC_OPCODE_LOG;
44 case TGSI_OPCODE_MUL: return RC_OPCODE_MUL;
45 case TGSI_OPCODE_ADD: return RC_OPCODE_ADD;
46 case TGSI_OPCODE_DP3: return RC_OPCODE_DP3;
47 case TGSI_OPCODE_DP4: return RC_OPCODE_DP4;
48 case TGSI_OPCODE_DST: return RC_OPCODE_DST;
49 case TGSI_OPCODE_MIN: return RC_OPCODE_MIN;
50 case TGSI_OPCODE_MAX: return RC_OPCODE_MAX;
51 case TGSI_OPCODE_SLT: return RC_OPCODE_SLT;
52 case TGSI_OPCODE_SGE: return RC_OPCODE_SGE;
53 case TGSI_OPCODE_MAD: return RC_OPCODE_MAD;
54 case TGSI_OPCODE_SUB: return RC_OPCODE_SUB;
55 case TGSI_OPCODE_LRP: return RC_OPCODE_LRP;
56 /* case TGSI_OPCODE_CND: return RC_OPCODE_CND; */
57 /* case TGSI_OPCODE_CND0: return RC_OPCODE_CND0; */
58 /* case TGSI_OPCODE_DP2A: return RC_OPCODE_DP2A; */
59 /* gap */
60 case TGSI_OPCODE_FRC: return RC_OPCODE_FRC;
61 /* case TGSI_OPCODE_CLAMP: return RC_OPCODE_CLAMP; */
62 case TGSI_OPCODE_FLR: return RC_OPCODE_FLR;
63 /* case TGSI_OPCODE_ROUND: return RC_OPCODE_ROUND; */
64 case TGSI_OPCODE_EX2: return RC_OPCODE_EX2;
65 case TGSI_OPCODE_LG2: return RC_OPCODE_LG2;
66 case TGSI_OPCODE_POW: return RC_OPCODE_POW;
67 case TGSI_OPCODE_XPD: return RC_OPCODE_XPD;
68 /* gap */
69 case TGSI_OPCODE_ABS: return RC_OPCODE_ABS;
70 /* case TGSI_OPCODE_RCC: return RC_OPCODE_RCC; */
71 case TGSI_OPCODE_DPH: return RC_OPCODE_DPH;
72 case TGSI_OPCODE_COS: return RC_OPCODE_COS;
73 case TGSI_OPCODE_DDX: return RC_OPCODE_DDX;
74 case TGSI_OPCODE_DDY: return RC_OPCODE_DDY;
75 /* case TGSI_OPCODE_KILP: return RC_OPCODE_KILP; */
76 /* case TGSI_OPCODE_PK2H: return RC_OPCODE_PK2H; */
77 /* case TGSI_OPCODE_PK2US: return RC_OPCODE_PK2US; */
78 /* case TGSI_OPCODE_PK4B: return RC_OPCODE_PK4B; */
79 /* case TGSI_OPCODE_PK4UB: return RC_OPCODE_PK4UB; */
80 /* case TGSI_OPCODE_RFL: return RC_OPCODE_RFL; */
81 case TGSI_OPCODE_SEQ: return RC_OPCODE_SEQ;
82 case TGSI_OPCODE_SFL: return RC_OPCODE_SFL;
83 case TGSI_OPCODE_SGT: return RC_OPCODE_SGT;
84 case TGSI_OPCODE_SIN: return RC_OPCODE_SIN;
85 case TGSI_OPCODE_SLE: return RC_OPCODE_SLE;
86 case TGSI_OPCODE_SNE: return RC_OPCODE_SNE;
87 /* case TGSI_OPCODE_STR: return RC_OPCODE_STR; */
88 case TGSI_OPCODE_TEX: return RC_OPCODE_TEX;
89 case TGSI_OPCODE_TXD: return RC_OPCODE_TXD;
90 case TGSI_OPCODE_TXP: return RC_OPCODE_TXP;
91 /* case TGSI_OPCODE_UP2H: return RC_OPCODE_UP2H; */
92 /* case TGSI_OPCODE_UP2US: return RC_OPCODE_UP2US; */
93 /* case TGSI_OPCODE_UP4B: return RC_OPCODE_UP4B; */
94 /* case TGSI_OPCODE_UP4UB: return RC_OPCODE_UP4UB; */
95 /* case TGSI_OPCODE_X2D: return RC_OPCODE_X2D; */
96 /* case TGSI_OPCODE_ARA: return RC_OPCODE_ARA; */
97 /* case TGSI_OPCODE_ARR: return RC_OPCODE_ARR; */
98 /* case TGSI_OPCODE_BRA: return RC_OPCODE_BRA; */
99 /* case TGSI_OPCODE_CAL: return RC_OPCODE_CAL; */
100 /* case TGSI_OPCODE_RET: return RC_OPCODE_RET; */
101 /* case TGSI_OPCODE_SSG: return RC_OPCODE_SSG; */
102 case TGSI_OPCODE_CMP: return RC_OPCODE_CMP;
103 case TGSI_OPCODE_SCS: return RC_OPCODE_SCS;
104 case TGSI_OPCODE_TXB: return RC_OPCODE_TXB;
105 /* case TGSI_OPCODE_NRM: return RC_OPCODE_NRM; */
106 /* case TGSI_OPCODE_DIV: return RC_OPCODE_DIV; */
107 /* case TGSI_OPCODE_DP2: return RC_OPCODE_DP2; */
108 case TGSI_OPCODE_TXL: return RC_OPCODE_TXL;
109 /* case TGSI_OPCODE_BRK: return RC_OPCODE_BRK; */
110 case TGSI_OPCODE_IF: return RC_OPCODE_IF;
111 /* case TGSI_OPCODE_LOOP: return RC_OPCODE_LOOP; */
112 /* case TGSI_OPCODE_REP: return RC_OPCODE_REP; */
113 case TGSI_OPCODE_ELSE: return RC_OPCODE_ELSE;
114 case TGSI_OPCODE_ENDIF: return RC_OPCODE_ENDIF;
115 /* case TGSI_OPCODE_ENDLOOP: return RC_OPCODE_ENDLOOP; */
116 /* case TGSI_OPCODE_ENDREP: return RC_OPCODE_ENDREP; */
117 /* case TGSI_OPCODE_PUSHA: return RC_OPCODE_PUSHA; */
118 /* case TGSI_OPCODE_POPA: return RC_OPCODE_POPA; */
119 /* case TGSI_OPCODE_CEIL: return RC_OPCODE_CEIL; */
120 /* case TGSI_OPCODE_I2F: return RC_OPCODE_I2F; */
121 /* case TGSI_OPCODE_NOT: return RC_OPCODE_NOT; */
122 /* case TGSI_OPCODE_TRUNC: return RC_OPCODE_TRUNC; */
123 /* case TGSI_OPCODE_SHL: return RC_OPCODE_SHL; */
124 /* case TGSI_OPCODE_ISHR: return RC_OPCODE_SHR; */
125 /* case TGSI_OPCODE_AND: return RC_OPCODE_AND; */
126 /* case TGSI_OPCODE_OR: return RC_OPCODE_OR; */
127 /* case TGSI_OPCODE_MOD: return RC_OPCODE_MOD; */
128 /* case TGSI_OPCODE_XOR: return RC_OPCODE_XOR; */
129 /* case TGSI_OPCODE_SAD: return RC_OPCODE_SAD; */
130 /* case TGSI_OPCODE_TXF: return RC_OPCODE_TXF; */
131 /* case TGSI_OPCODE_TXQ: return RC_OPCODE_TXQ; */
132 /* case TGSI_OPCODE_CONT: return RC_OPCODE_CONT; */
133 /* case TGSI_OPCODE_EMIT: return RC_OPCODE_EMIT; */
134 /* case TGSI_OPCODE_ENDPRIM: return RC_OPCODE_ENDPRIM; */
135 /* case TGSI_OPCODE_BGNLOOP2: return RC_OPCODE_BGNLOOP2; */
136 /* case TGSI_OPCODE_BGNSUB: return RC_OPCODE_BGNSUB; */
137 /* case TGSI_OPCODE_ENDLOOP2: return RC_OPCODE_ENDLOOP2; */
138 /* case TGSI_OPCODE_ENDSUB: return RC_OPCODE_ENDSUB; */
139 case TGSI_OPCODE_NOP: return RC_OPCODE_NOP;
140 /* gap */
141 /* case TGSI_OPCODE_NRM4: return RC_OPCODE_NRM4; */
142 /* case TGSI_OPCODE_CALLNZ: return RC_OPCODE_CALLNZ; */
143 /* case TGSI_OPCODE_IFC: return RC_OPCODE_IFC; */
144 /* case TGSI_OPCODE_BREAKC: return RC_OPCODE_BREAKC; */
145 case TGSI_OPCODE_KIL: return RC_OPCODE_KIL;
146 }
147
148 debug_printf("r300: Unknown TGSI/RC opcode: %i\n", opcode);
149 return RC_OPCODE_ILLEGAL_OPCODE;
150 }
151
152 static unsigned translate_saturate(unsigned saturate)
153 {
154 switch(saturate) {
155 default:
156 fprintf(stderr, "Unknown saturate mode: %i\n", saturate);
157 /* fall-through */
158 case TGSI_SAT_NONE: return RC_SATURATE_NONE;
159 case TGSI_SAT_ZERO_ONE: return RC_SATURATE_ZERO_ONE;
160 }
161 }
162
163 static unsigned translate_register_file(unsigned file)
164 {
165 switch(file) {
166 case TGSI_FILE_CONSTANT: return RC_FILE_CONSTANT;
167 case TGSI_FILE_IMMEDIATE: return RC_FILE_CONSTANT;
168 case TGSI_FILE_INPUT: return RC_FILE_INPUT;
169 case TGSI_FILE_OUTPUT: return RC_FILE_OUTPUT;
170 default:
171 fprintf(stderr, "Unhandled register file: %i\n", file);
172 /* fall-through */
173 case TGSI_FILE_TEMPORARY: return RC_FILE_TEMPORARY;
174 case TGSI_FILE_ADDRESS: return RC_FILE_ADDRESS;
175 }
176 }
177
178 static int translate_register_index(
179 struct tgsi_to_rc * ttr,
180 unsigned file,
181 int index)
182 {
183 if (file == TGSI_FILE_IMMEDIATE)
184 return ttr->immediate_offset + index;
185
186 return index;
187 }
188
189 static void transform_dstreg(
190 struct tgsi_to_rc * ttr,
191 struct rc_dst_register * dst,
192 struct tgsi_full_dst_register * src)
193 {
194 dst->File = translate_register_file(src->Register.File);
195 dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
196 dst->WriteMask = src->Register.WriteMask;
197 dst->RelAddr = src->Register.Indirect;
198 }
199
200 static void transform_srcreg(
201 struct tgsi_to_rc * ttr,
202 struct rc_src_register * dst,
203 struct tgsi_full_src_register * src)
204 {
205 unsigned i, j;
206
207 dst->File = translate_register_file(src->Register.File);
208 dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
209 dst->RelAddr = src->Register.Indirect;
210 dst->Swizzle = tgsi_util_get_full_src_register_swizzle(src, 0);
211 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 1) << 3;
212 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 2) << 6;
213 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 3) << 9;
214 dst->Abs = src->Register.Absolute;
215 dst->Negate = src->Register.Negate ? RC_MASK_XYZW : 0;
216
217 if (src->Register.File == TGSI_FILE_IMMEDIATE) {
218 for (i = 0; i < ttr->imms_to_swizzle_count; i++) {
219 if (ttr->imms_to_swizzle[i].index == src->Register.Index) {
220 dst->File = RC_FILE_TEMPORARY;
221 dst->Index = 0;
222 dst->Swizzle = 0;
223 for (j = 0; j < 4; j++) {
224 dst->Swizzle |= GET_SWZ(ttr->imms_to_swizzle[i].swizzle,
225 tgsi_util_get_full_src_register_swizzle(src, j)) << (j * 3);
226 }
227 break;
228 }
229 }
230 }
231 }
232
233 static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_texture src,
234 uint32_t *shadowSamplers)
235 {
236 switch(src.Texture) {
237 case TGSI_TEXTURE_1D:
238 dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
239 break;
240 case TGSI_TEXTURE_2D:
241 dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
242 break;
243 case TGSI_TEXTURE_3D:
244 dst->U.I.TexSrcTarget = RC_TEXTURE_3D;
245 break;
246 case TGSI_TEXTURE_CUBE:
247 dst->U.I.TexSrcTarget = RC_TEXTURE_CUBE;
248 break;
249 case TGSI_TEXTURE_RECT:
250 dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
251 break;
252 case TGSI_TEXTURE_SHADOW1D:
253 dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
254 dst->U.I.TexShadow = 1;
255 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
256 break;
257 case TGSI_TEXTURE_SHADOW2D:
258 dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
259 dst->U.I.TexShadow = 1;
260 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
261 break;
262 case TGSI_TEXTURE_SHADOWRECT:
263 dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
264 dst->U.I.TexShadow = 1;
265 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
266 break;
267 }
268 }
269
270 static void transform_instruction(struct tgsi_to_rc * ttr, struct tgsi_full_instruction * src)
271 {
272 struct rc_instruction * dst;
273 int i;
274
275 if (src->Instruction.Opcode == TGSI_OPCODE_END)
276 return;
277
278 dst = rc_insert_new_instruction(ttr->compiler, ttr->compiler->Program.Instructions.Prev);
279 dst->U.I.Opcode = translate_opcode(src->Instruction.Opcode);
280 dst->U.I.SaturateMode = translate_saturate(src->Instruction.Saturate);
281
282 if (src->Instruction.NumDstRegs)
283 transform_dstreg(ttr, &dst->U.I.DstReg, &src->Dst[0]);
284
285 for(i = 0; i < src->Instruction.NumSrcRegs; ++i) {
286 if (src->Src[i].Register.File == TGSI_FILE_SAMPLER)
287 dst->U.I.TexSrcUnit = src->Src[i].Register.Index;
288 else
289 transform_srcreg(ttr, &dst->U.I.SrcReg[i], &src->Src[i]);
290 }
291
292 /* Texturing. */
293 if (src->Instruction.Texture)
294 transform_texture(dst, src->Texture,
295 &ttr->compiler->Program.ShadowSamplers);
296 }
297
298 static void handle_immediate(struct tgsi_to_rc * ttr,
299 struct tgsi_full_immediate * imm,
300 unsigned index)
301 {
302 struct rc_constant constant;
303 unsigned swizzle = 0;
304 boolean can_swizzle = TRUE;
305 unsigned i;
306
307 for (i = 0; i < 4; i++) {
308 if (imm->u[i].Float == 0.0f) {
309 swizzle |= RC_SWIZZLE_ZERO << (i * 3);
310 } else if (imm->u[i].Float == 0.5f) {
311 swizzle |= RC_SWIZZLE_HALF << (i * 3);
312 } else if (imm->u[i].Float == 1.0f) {
313 swizzle |= RC_SWIZZLE_ONE << (i * 3);
314 } else {
315 can_swizzle = FALSE;
316 break;
317 }
318 }
319
320 if (can_swizzle) {
321 ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].index = index;
322 ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].swizzle = swizzle;
323 ttr->imms_to_swizzle_count++;
324 } else {
325 constant.Type = RC_CONSTANT_IMMEDIATE;
326 constant.Size = 4;
327 for(i = 0; i < 4; ++i)
328 constant.u.Immediate[i] = imm->u[i].Float;
329 rc_constants_add(&ttr->compiler->Program.Constants, &constant);
330 }
331 }
332
333 void r300_tgsi_to_rc(struct tgsi_to_rc * ttr, const struct tgsi_token * tokens)
334 {
335 struct tgsi_parse_context parser;
336 unsigned imm_index = 0;
337 int i;
338
339 /* Allocate constants placeholders.
340 *
341 * Note: What if declared constants are not contiguous? */
342 for(i = 0; i <= ttr->info->file_max[TGSI_FILE_CONSTANT]; ++i) {
343 struct rc_constant constant;
344 memset(&constant, 0, sizeof(constant));
345 constant.Type = RC_CONSTANT_EXTERNAL;
346 constant.Size = 4;
347 constant.u.External = i;
348 rc_constants_add(&ttr->compiler->Program.Constants, &constant);
349 }
350
351 ttr->immediate_offset = ttr->compiler->Program.Constants.Count;
352
353 ttr->imms_to_swizzle = malloc(ttr->info->immediate_count * sizeof(struct swizzled_imms));
354 ttr->imms_to_swizzle_count = 0;
355
356 tgsi_parse_init(&parser, tokens);
357
358 while (!tgsi_parse_end_of_tokens(&parser)) {
359 tgsi_parse_token(&parser);
360
361 switch (parser.FullToken.Token.Type) {
362 case TGSI_TOKEN_TYPE_DECLARATION:
363 break;
364 case TGSI_TOKEN_TYPE_IMMEDIATE:
365 handle_immediate(ttr, &parser.FullToken.FullImmediate, imm_index);
366 imm_index++;
367 break;
368 case TGSI_TOKEN_TYPE_INSTRUCTION:
369 transform_instruction(ttr, &parser.FullToken.FullInstruction);
370 break;
371 }
372 }
373
374 tgsi_parse_free(&parser);
375
376 free(ttr->imms_to_swizzle);
377
378 rc_calculate_inputs_outputs(ttr->compiler);
379 }
380