r300: Drop the "/* gap */" notes.
[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 "compiler/radeon_compiler.h"
26
27 #include "tgsi/tgsi_info.h"
28 #include "tgsi/tgsi_parse.h"
29 #include "tgsi/tgsi_scan.h"
30 #include "tgsi/tgsi_util.h"
31
32 static unsigned translate_opcode(unsigned opcode)
33 {
34 switch(opcode) {
35 case TGSI_OPCODE_ARL: return RC_OPCODE_ARL;
36 case TGSI_OPCODE_MOV: return RC_OPCODE_MOV;
37 case TGSI_OPCODE_LIT: return RC_OPCODE_LIT;
38 case TGSI_OPCODE_RCP: return RC_OPCODE_RCP;
39 case TGSI_OPCODE_RSQ: return RC_OPCODE_RSQ;
40 case TGSI_OPCODE_EXP: return RC_OPCODE_EXP;
41 case TGSI_OPCODE_LOG: return RC_OPCODE_LOG;
42 case TGSI_OPCODE_MUL: return RC_OPCODE_MUL;
43 case TGSI_OPCODE_ADD: return RC_OPCODE_ADD;
44 case TGSI_OPCODE_DP3: return RC_OPCODE_DP3;
45 case TGSI_OPCODE_DP4: return RC_OPCODE_DP4;
46 case TGSI_OPCODE_DST: return RC_OPCODE_DST;
47 case TGSI_OPCODE_MIN: return RC_OPCODE_MIN;
48 case TGSI_OPCODE_MAX: return RC_OPCODE_MAX;
49 case TGSI_OPCODE_SLT: return RC_OPCODE_SLT;
50 case TGSI_OPCODE_SGE: return RC_OPCODE_SGE;
51 case TGSI_OPCODE_MAD: return RC_OPCODE_MAD;
52 case TGSI_OPCODE_SUB: return RC_OPCODE_SUB;
53 case TGSI_OPCODE_LRP: return RC_OPCODE_LRP;
54 case TGSI_OPCODE_CND: return RC_OPCODE_CND;
55 /* case TGSI_OPCODE_DP2A: return RC_OPCODE_DP2A; */
56 case TGSI_OPCODE_FRC: return RC_OPCODE_FRC;
57 case TGSI_OPCODE_CLAMP: return RC_OPCODE_CLAMP;
58 case TGSI_OPCODE_FLR: return RC_OPCODE_FLR;
59 case TGSI_OPCODE_ROUND: return RC_OPCODE_ROUND;
60 case TGSI_OPCODE_EX2: return RC_OPCODE_EX2;
61 case TGSI_OPCODE_LG2: return RC_OPCODE_LG2;
62 case TGSI_OPCODE_POW: return RC_OPCODE_POW;
63 case TGSI_OPCODE_XPD: return RC_OPCODE_XPD;
64 case TGSI_OPCODE_ABS: return RC_OPCODE_ABS;
65 /* case TGSI_OPCODE_RCC: return RC_OPCODE_RCC; */
66 case TGSI_OPCODE_DPH: return RC_OPCODE_DPH;
67 case TGSI_OPCODE_COS: return RC_OPCODE_COS;
68 case TGSI_OPCODE_DDX: return RC_OPCODE_DDX;
69 case TGSI_OPCODE_DDY: return RC_OPCODE_DDY;
70 case TGSI_OPCODE_KILL: return RC_OPCODE_KILP;
71 /* case TGSI_OPCODE_PK2H: return RC_OPCODE_PK2H; */
72 /* case TGSI_OPCODE_PK2US: return RC_OPCODE_PK2US; */
73 /* case TGSI_OPCODE_PK4B: return RC_OPCODE_PK4B; */
74 /* case TGSI_OPCODE_PK4UB: return RC_OPCODE_PK4UB; */
75 /* case TGSI_OPCODE_RFL: return RC_OPCODE_RFL; */
76 case TGSI_OPCODE_SEQ: return RC_OPCODE_SEQ;
77 case TGSI_OPCODE_SFL: return RC_OPCODE_SFL;
78 case TGSI_OPCODE_SGT: return RC_OPCODE_SGT;
79 case TGSI_OPCODE_SIN: return RC_OPCODE_SIN;
80 case TGSI_OPCODE_SLE: return RC_OPCODE_SLE;
81 case TGSI_OPCODE_SNE: return RC_OPCODE_SNE;
82 /* case TGSI_OPCODE_STR: return RC_OPCODE_STR; */
83 case TGSI_OPCODE_TEX: return RC_OPCODE_TEX;
84 case TGSI_OPCODE_TXD: return RC_OPCODE_TXD;
85 case TGSI_OPCODE_TXP: return RC_OPCODE_TXP;
86 /* case TGSI_OPCODE_UP2H: return RC_OPCODE_UP2H; */
87 /* case TGSI_OPCODE_UP2US: return RC_OPCODE_UP2US; */
88 /* case TGSI_OPCODE_UP4B: return RC_OPCODE_UP4B; */
89 /* case TGSI_OPCODE_UP4UB: return RC_OPCODE_UP4UB; */
90 /* case TGSI_OPCODE_X2D: return RC_OPCODE_X2D; */
91 /* case TGSI_OPCODE_ARA: return RC_OPCODE_ARA; */
92 /* case TGSI_OPCODE_ARR: return RC_OPCODE_ARR; */
93 /* case TGSI_OPCODE_BRA: return RC_OPCODE_BRA; */
94 /* case TGSI_OPCODE_CAL: return RC_OPCODE_CAL; */
95 /* case TGSI_OPCODE_RET: return RC_OPCODE_RET; */
96 case TGSI_OPCODE_SSG: return RC_OPCODE_SSG;
97 case TGSI_OPCODE_CMP: return RC_OPCODE_CMP;
98 case TGSI_OPCODE_SCS: return RC_OPCODE_SCS;
99 case TGSI_OPCODE_TXB: return RC_OPCODE_TXB;
100 /* case TGSI_OPCODE_NRM: return RC_OPCODE_NRM; */
101 /* case TGSI_OPCODE_DIV: return RC_OPCODE_DIV; */
102 case TGSI_OPCODE_DP2: return RC_OPCODE_DP2;
103 case TGSI_OPCODE_TXL: return RC_OPCODE_TXL;
104 case TGSI_OPCODE_BRK: return RC_OPCODE_BRK;
105 case TGSI_OPCODE_IF: return RC_OPCODE_IF;
106 case TGSI_OPCODE_BGNLOOP: return RC_OPCODE_BGNLOOP;
107 case TGSI_OPCODE_ELSE: return RC_OPCODE_ELSE;
108 case TGSI_OPCODE_ENDIF: return RC_OPCODE_ENDIF;
109 case TGSI_OPCODE_ENDLOOP: return RC_OPCODE_ENDLOOP;
110 /* case TGSI_OPCODE_PUSHA: return RC_OPCODE_PUSHA; */
111 /* case TGSI_OPCODE_POPA: return RC_OPCODE_POPA; */
112 case TGSI_OPCODE_CEIL: return RC_OPCODE_CEIL;
113 /* case TGSI_OPCODE_I2F: return RC_OPCODE_I2F; */
114 /* case TGSI_OPCODE_NOT: return RC_OPCODE_NOT; */
115 case TGSI_OPCODE_TRUNC: return RC_OPCODE_TRUNC;
116 /* case TGSI_OPCODE_SHL: return RC_OPCODE_SHL; */
117 /* case TGSI_OPCODE_ISHR: return RC_OPCODE_SHR; */
118 /* case TGSI_OPCODE_AND: return RC_OPCODE_AND; */
119 /* case TGSI_OPCODE_OR: return RC_OPCODE_OR; */
120 /* case TGSI_OPCODE_MOD: return RC_OPCODE_MOD; */
121 /* case TGSI_OPCODE_XOR: return RC_OPCODE_XOR; */
122 /* case TGSI_OPCODE_SAD: return RC_OPCODE_SAD; */
123 /* case TGSI_OPCODE_TXF: return RC_OPCODE_TXF; */
124 /* case TGSI_OPCODE_TXQ: return RC_OPCODE_TXQ; */
125 case TGSI_OPCODE_CONT: return RC_OPCODE_CONT;
126 /* case TGSI_OPCODE_EMIT: return RC_OPCODE_EMIT; */
127 /* case TGSI_OPCODE_ENDPRIM: return RC_OPCODE_ENDPRIM; */
128 /* case TGSI_OPCODE_BGNLOOP2: return RC_OPCODE_BGNLOOP2; */
129 /* case TGSI_OPCODE_BGNSUB: return RC_OPCODE_BGNSUB; */
130 /* case TGSI_OPCODE_ENDLOOP2: return RC_OPCODE_ENDLOOP2; */
131 /* case TGSI_OPCODE_ENDSUB: return RC_OPCODE_ENDSUB; */
132 case TGSI_OPCODE_NOP: return RC_OPCODE_NOP;
133 /* case TGSI_OPCODE_NRM4: return RC_OPCODE_NRM4; */
134 /* case TGSI_OPCODE_CALLNZ: return RC_OPCODE_CALLNZ; */
135 /* case TGSI_OPCODE_BREAKC: return RC_OPCODE_BREAKC; */
136 case TGSI_OPCODE_KILL_IF: return RC_OPCODE_KIL;
137 }
138
139 fprintf(stderr, "r300: Unknown TGSI/RC opcode: %s\n", tgsi_get_opcode_name(opcode));
140 return RC_OPCODE_ILLEGAL_OPCODE;
141 }
142
143 static unsigned translate_saturate(unsigned saturate)
144 {
145 switch(saturate) {
146 default:
147 fprintf(stderr, "Unknown saturate mode: %i\n", saturate);
148 /* fall-through */
149 case TGSI_SAT_NONE: return RC_SATURATE_NONE;
150 case TGSI_SAT_ZERO_ONE: return RC_SATURATE_ZERO_ONE;
151 }
152 }
153
154 static unsigned translate_register_file(unsigned file)
155 {
156 switch(file) {
157 case TGSI_FILE_CONSTANT: return RC_FILE_CONSTANT;
158 case TGSI_FILE_IMMEDIATE: return RC_FILE_CONSTANT;
159 case TGSI_FILE_INPUT: return RC_FILE_INPUT;
160 case TGSI_FILE_OUTPUT: return RC_FILE_OUTPUT;
161 default:
162 fprintf(stderr, "Unhandled register file: %i\n", file);
163 /* fall-through */
164 case TGSI_FILE_TEMPORARY: return RC_FILE_TEMPORARY;
165 case TGSI_FILE_ADDRESS: return RC_FILE_ADDRESS;
166 }
167 }
168
169 static int translate_register_index(
170 struct tgsi_to_rc * ttr,
171 unsigned file,
172 int index)
173 {
174 if (file == TGSI_FILE_IMMEDIATE)
175 return ttr->immediate_offset + index;
176
177 return index;
178 }
179
180 static void transform_dstreg(
181 struct tgsi_to_rc * ttr,
182 struct rc_dst_register * dst,
183 struct tgsi_full_dst_register * src)
184 {
185 dst->File = translate_register_file(src->Register.File);
186 dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
187 dst->WriteMask = src->Register.WriteMask;
188
189 if (src->Register.Indirect) {
190 ttr->error = TRUE;
191 fprintf(stderr, "r300: Relative addressing of destination operands "
192 "is unsupported.\n");
193 }
194 }
195
196 static void transform_srcreg(
197 struct tgsi_to_rc * ttr,
198 struct rc_src_register * dst,
199 struct tgsi_full_src_register * src)
200 {
201 unsigned i, j;
202
203 dst->File = translate_register_file(src->Register.File);
204 dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
205 dst->RelAddr = src->Register.Indirect;
206 dst->Swizzle = tgsi_util_get_full_src_register_swizzle(src, 0);
207 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 1) << 3;
208 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 2) << 6;
209 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 3) << 9;
210 dst->Abs = src->Register.Absolute;
211 dst->Negate = src->Register.Negate ? RC_MASK_XYZW : 0;
212
213 if (src->Register.File == TGSI_FILE_IMMEDIATE) {
214 for (i = 0; i < ttr->imms_to_swizzle_count; i++) {
215 if (ttr->imms_to_swizzle[i].index == src->Register.Index) {
216 dst->File = RC_FILE_TEMPORARY;
217 dst->Index = 0;
218 dst->Swizzle = 0;
219 for (j = 0; j < 4; j++) {
220 dst->Swizzle |= GET_SWZ(ttr->imms_to_swizzle[i].swizzle,
221 tgsi_util_get_full_src_register_swizzle(src, j)) << (j * 3);
222 }
223 break;
224 }
225 }
226 }
227 }
228
229 static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_texture src,
230 uint32_t *shadowSamplers)
231 {
232 switch(src.Texture) {
233 case TGSI_TEXTURE_1D:
234 dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
235 break;
236 case TGSI_TEXTURE_2D:
237 dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
238 break;
239 case TGSI_TEXTURE_3D:
240 dst->U.I.TexSrcTarget = RC_TEXTURE_3D;
241 break;
242 case TGSI_TEXTURE_CUBE:
243 dst->U.I.TexSrcTarget = RC_TEXTURE_CUBE;
244 break;
245 case TGSI_TEXTURE_RECT:
246 dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
247 break;
248 case TGSI_TEXTURE_SHADOW1D:
249 dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
250 dst->U.I.TexShadow = 1;
251 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
252 break;
253 case TGSI_TEXTURE_SHADOW2D:
254 dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
255 dst->U.I.TexShadow = 1;
256 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
257 break;
258 case TGSI_TEXTURE_SHADOWRECT:
259 dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
260 dst->U.I.TexShadow = 1;
261 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
262 break;
263 }
264 dst->U.I.TexSwizzle = RC_SWIZZLE_XYZW;
265 }
266
267 static void transform_instruction(struct tgsi_to_rc * ttr, struct tgsi_full_instruction * src)
268 {
269 struct rc_instruction * dst;
270 int i;
271
272 dst = rc_insert_new_instruction(ttr->compiler, ttr->compiler->Program.Instructions.Prev);
273 dst->U.I.Opcode = translate_opcode(src->Instruction.Opcode);
274 dst->U.I.SaturateMode = translate_saturate(src->Instruction.Saturate);
275
276 if (src->Instruction.NumDstRegs)
277 transform_dstreg(ttr, &dst->U.I.DstReg, &src->Dst[0]);
278
279 for(i = 0; i < src->Instruction.NumSrcRegs; ++i) {
280 if (src->Src[i].Register.File == TGSI_FILE_SAMPLER)
281 dst->U.I.TexSrcUnit = src->Src[i].Register.Index;
282 else
283 transform_srcreg(ttr, &dst->U.I.SrcReg[i], &src->Src[i]);
284 }
285
286 /* Texturing. */
287 if (src->Instruction.Texture)
288 transform_texture(dst, src->Texture,
289 &ttr->compiler->Program.ShadowSamplers);
290 }
291
292 static void handle_immediate(struct tgsi_to_rc * ttr,
293 struct tgsi_full_immediate * imm,
294 unsigned index)
295 {
296 struct rc_constant constant;
297 unsigned swizzle = 0;
298 boolean can_swizzle = TRUE;
299 unsigned i;
300
301 for (i = 0; i < 4; i++) {
302 if (imm->u[i].Float == 0.0f) {
303 swizzle |= RC_SWIZZLE_ZERO << (i * 3);
304 } else if (imm->u[i].Float == 0.5f && ttr->use_half_swizzles) {
305 swizzle |= RC_SWIZZLE_HALF << (i * 3);
306 } else if (imm->u[i].Float == 1.0f) {
307 swizzle |= RC_SWIZZLE_ONE << (i * 3);
308 } else {
309 can_swizzle = FALSE;
310 break;
311 }
312 }
313
314 if (can_swizzle) {
315 ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].index = index;
316 ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].swizzle = swizzle;
317 ttr->imms_to_swizzle_count++;
318 } else {
319 constant.Type = RC_CONSTANT_IMMEDIATE;
320 constant.Size = 4;
321 for(i = 0; i < 4; ++i)
322 constant.u.Immediate[i] = imm->u[i].Float;
323 rc_constants_add(&ttr->compiler->Program.Constants, &constant);
324 }
325 }
326
327 void r300_tgsi_to_rc(struct tgsi_to_rc * ttr,
328 const struct tgsi_token * tokens)
329 {
330 struct tgsi_full_instruction *inst;
331 struct tgsi_parse_context parser;
332 unsigned imm_index = 0;
333 int i;
334
335 ttr->error = FALSE;
336
337 /* Allocate constants placeholders.
338 *
339 * Note: What if declared constants are not contiguous? */
340 for(i = 0; i <= ttr->info->file_max[TGSI_FILE_CONSTANT]; ++i) {
341 struct rc_constant constant;
342 memset(&constant, 0, sizeof(constant));
343 constant.Type = RC_CONSTANT_EXTERNAL;
344 constant.Size = 4;
345 constant.u.External = i;
346 rc_constants_add(&ttr->compiler->Program.Constants, &constant);
347 }
348
349 ttr->immediate_offset = ttr->compiler->Program.Constants.Count;
350
351 ttr->imms_to_swizzle = malloc(ttr->info->immediate_count * sizeof(struct swizzled_imms));
352 ttr->imms_to_swizzle_count = 0;
353
354 tgsi_parse_init(&parser, tokens);
355
356 while (!tgsi_parse_end_of_tokens(&parser)) {
357 tgsi_parse_token(&parser);
358
359 switch (parser.FullToken.Token.Type) {
360 case TGSI_TOKEN_TYPE_DECLARATION:
361 break;
362 case TGSI_TOKEN_TYPE_IMMEDIATE:
363 handle_immediate(ttr, &parser.FullToken.FullImmediate, imm_index);
364 imm_index++;
365 break;
366 case TGSI_TOKEN_TYPE_INSTRUCTION:
367 inst = &parser.FullToken.FullInstruction;
368 if (inst->Instruction.Opcode == TGSI_OPCODE_END) {
369 break;
370 }
371
372 transform_instruction(ttr, inst);
373 break;
374 }
375 }
376
377 tgsi_parse_free(&parser);
378
379 free(ttr->imms_to_swizzle);
380
381 rc_calculate_inputs_outputs(ttr->compiler);
382 }