r300/compiler: Implement KILP opcode.
[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_info.h"
29 #include "tgsi/tgsi_parse.h"
30 #include "tgsi/tgsi_scan.h"
31 #include "tgsi/tgsi_util.h"
32
33 static unsigned translate_opcode(unsigned opcode)
34 {
35 switch(opcode) {
36 case TGSI_OPCODE_ARL: return RC_OPCODE_ARL;
37 case TGSI_OPCODE_MOV: return RC_OPCODE_MOV;
38 case TGSI_OPCODE_LIT: return RC_OPCODE_LIT;
39 case TGSI_OPCODE_RCP: return RC_OPCODE_RCP;
40 case TGSI_OPCODE_RSQ: return RC_OPCODE_RSQ;
41 case TGSI_OPCODE_EXP: return RC_OPCODE_EXP;
42 case TGSI_OPCODE_LOG: return RC_OPCODE_LOG;
43 case TGSI_OPCODE_MUL: return RC_OPCODE_MUL;
44 case TGSI_OPCODE_ADD: return RC_OPCODE_ADD;
45 case TGSI_OPCODE_DP3: return RC_OPCODE_DP3;
46 case TGSI_OPCODE_DP4: return RC_OPCODE_DP4;
47 case TGSI_OPCODE_DST: return RC_OPCODE_DST;
48 case TGSI_OPCODE_MIN: return RC_OPCODE_MIN;
49 case TGSI_OPCODE_MAX: return RC_OPCODE_MAX;
50 case TGSI_OPCODE_SLT: return RC_OPCODE_SLT;
51 case TGSI_OPCODE_SGE: return RC_OPCODE_SGE;
52 case TGSI_OPCODE_MAD: return RC_OPCODE_MAD;
53 case TGSI_OPCODE_SUB: return RC_OPCODE_SUB;
54 case TGSI_OPCODE_LRP: return RC_OPCODE_LRP;
55 /* case TGSI_OPCODE_CND: return RC_OPCODE_CND; */
56 /* case TGSI_OPCODE_CND0: return RC_OPCODE_CND0; */
57 /* case TGSI_OPCODE_DP2A: return RC_OPCODE_DP2A; */
58 /* gap */
59 case TGSI_OPCODE_FRC: return RC_OPCODE_FRC;
60 /* case TGSI_OPCODE_CLAMP: return RC_OPCODE_CLAMP; */
61 case TGSI_OPCODE_FLR: return RC_OPCODE_FLR;
62 /* case TGSI_OPCODE_ROUND: return RC_OPCODE_ROUND; */
63 case TGSI_OPCODE_EX2: return RC_OPCODE_EX2;
64 case TGSI_OPCODE_LG2: return RC_OPCODE_LG2;
65 case TGSI_OPCODE_POW: return RC_OPCODE_POW;
66 case TGSI_OPCODE_XPD: return RC_OPCODE_XPD;
67 /* gap */
68 case TGSI_OPCODE_ABS: return RC_OPCODE_ABS;
69 /* case TGSI_OPCODE_RCC: return RC_OPCODE_RCC; */
70 case TGSI_OPCODE_DPH: return RC_OPCODE_DPH;
71 case TGSI_OPCODE_COS: return RC_OPCODE_COS;
72 case TGSI_OPCODE_DDX: return RC_OPCODE_DDX;
73 case TGSI_OPCODE_DDY: return RC_OPCODE_DDY;
74 case TGSI_OPCODE_KILP: return RC_OPCODE_KILP;
75 /* case TGSI_OPCODE_PK2H: return RC_OPCODE_PK2H; */
76 /* case TGSI_OPCODE_PK2US: return RC_OPCODE_PK2US; */
77 /* case TGSI_OPCODE_PK4B: return RC_OPCODE_PK4B; */
78 /* case TGSI_OPCODE_PK4UB: return RC_OPCODE_PK4UB; */
79 /* case TGSI_OPCODE_RFL: return RC_OPCODE_RFL; */
80 case TGSI_OPCODE_SEQ: return RC_OPCODE_SEQ;
81 case TGSI_OPCODE_SFL: return RC_OPCODE_SFL;
82 case TGSI_OPCODE_SGT: return RC_OPCODE_SGT;
83 case TGSI_OPCODE_SIN: return RC_OPCODE_SIN;
84 case TGSI_OPCODE_SLE: return RC_OPCODE_SLE;
85 case TGSI_OPCODE_SNE: return RC_OPCODE_SNE;
86 /* case TGSI_OPCODE_STR: return RC_OPCODE_STR; */
87 case TGSI_OPCODE_TEX: return RC_OPCODE_TEX;
88 case TGSI_OPCODE_TXD: return RC_OPCODE_TXD;
89 case TGSI_OPCODE_TXP: return RC_OPCODE_TXP;
90 /* case TGSI_OPCODE_UP2H: return RC_OPCODE_UP2H; */
91 /* case TGSI_OPCODE_UP2US: return RC_OPCODE_UP2US; */
92 /* case TGSI_OPCODE_UP4B: return RC_OPCODE_UP4B; */
93 /* case TGSI_OPCODE_UP4UB: return RC_OPCODE_UP4UB; */
94 /* case TGSI_OPCODE_X2D: return RC_OPCODE_X2D; */
95 /* case TGSI_OPCODE_ARA: return RC_OPCODE_ARA; */
96 /* case TGSI_OPCODE_ARR: return RC_OPCODE_ARR; */
97 /* case TGSI_OPCODE_BRA: return RC_OPCODE_BRA; */
98 /* case TGSI_OPCODE_CAL: return RC_OPCODE_CAL; */
99 /* case TGSI_OPCODE_RET: return RC_OPCODE_RET; */
100 /* case TGSI_OPCODE_SSG: return RC_OPCODE_SSG; */
101 case TGSI_OPCODE_CMP: return RC_OPCODE_CMP;
102 case TGSI_OPCODE_SCS: return RC_OPCODE_SCS;
103 case TGSI_OPCODE_TXB: return RC_OPCODE_TXB;
104 /* case TGSI_OPCODE_NRM: return RC_OPCODE_NRM; */
105 /* case TGSI_OPCODE_DIV: return RC_OPCODE_DIV; */
106 /* case TGSI_OPCODE_DP2: return RC_OPCODE_DP2; */
107 case TGSI_OPCODE_TXL: return RC_OPCODE_TXL;
108 case TGSI_OPCODE_BRK: return RC_OPCODE_BRK;
109 case TGSI_OPCODE_IF: return RC_OPCODE_IF;
110 case TGSI_OPCODE_BGNLOOP: return RC_OPCODE_BGNLOOP;
111 case TGSI_OPCODE_ELSE: return RC_OPCODE_ELSE;
112 case TGSI_OPCODE_ENDIF: return RC_OPCODE_ENDIF;
113 case TGSI_OPCODE_ENDLOOP: return RC_OPCODE_ENDLOOP;
114 /* case TGSI_OPCODE_PUSHA: return RC_OPCODE_PUSHA; */
115 /* case TGSI_OPCODE_POPA: return RC_OPCODE_POPA; */
116 case TGSI_OPCODE_CEIL: return RC_OPCODE_CEIL;
117 /* case TGSI_OPCODE_I2F: return RC_OPCODE_I2F; */
118 /* case TGSI_OPCODE_NOT: return RC_OPCODE_NOT; */
119 case TGSI_OPCODE_TRUNC: return RC_OPCODE_FLR;
120 /* case TGSI_OPCODE_SHL: return RC_OPCODE_SHL; */
121 /* case TGSI_OPCODE_ISHR: return RC_OPCODE_SHR; */
122 /* case TGSI_OPCODE_AND: return RC_OPCODE_AND; */
123 /* case TGSI_OPCODE_OR: return RC_OPCODE_OR; */
124 /* case TGSI_OPCODE_MOD: return RC_OPCODE_MOD; */
125 /* case TGSI_OPCODE_XOR: return RC_OPCODE_XOR; */
126 /* case TGSI_OPCODE_SAD: return RC_OPCODE_SAD; */
127 /* case TGSI_OPCODE_TXF: return RC_OPCODE_TXF; */
128 /* case TGSI_OPCODE_TXQ: return RC_OPCODE_TXQ; */
129 /* case TGSI_OPCODE_CONT: return RC_OPCODE_CONT; */
130 /* case TGSI_OPCODE_EMIT: return RC_OPCODE_EMIT; */
131 /* case TGSI_OPCODE_ENDPRIM: return RC_OPCODE_ENDPRIM; */
132 /* case TGSI_OPCODE_BGNLOOP2: return RC_OPCODE_BGNLOOP2; */
133 /* case TGSI_OPCODE_BGNSUB: return RC_OPCODE_BGNSUB; */
134 /* case TGSI_OPCODE_ENDLOOP2: return RC_OPCODE_ENDLOOP2; */
135 /* case TGSI_OPCODE_ENDSUB: return RC_OPCODE_ENDSUB; */
136 case TGSI_OPCODE_NOP: return RC_OPCODE_NOP;
137 /* gap */
138 /* case TGSI_OPCODE_NRM4: return RC_OPCODE_NRM4; */
139 /* case TGSI_OPCODE_CALLNZ: return RC_OPCODE_CALLNZ; */
140 /* case TGSI_OPCODE_IFC: return RC_OPCODE_IFC; */
141 /* case TGSI_OPCODE_BREAKC: return RC_OPCODE_BREAKC; */
142 case TGSI_OPCODE_KIL: return RC_OPCODE_KIL;
143 }
144
145 fprintf(stderr, "r300: Unknown TGSI/RC opcode: %s\n", tgsi_get_opcode_name(opcode));
146 return RC_OPCODE_ILLEGAL_OPCODE;
147 }
148
149 static unsigned translate_saturate(unsigned saturate)
150 {
151 switch(saturate) {
152 default:
153 fprintf(stderr, "Unknown saturate mode: %i\n", saturate);
154 /* fall-through */
155 case TGSI_SAT_NONE: return RC_SATURATE_NONE;
156 case TGSI_SAT_ZERO_ONE: return RC_SATURATE_ZERO_ONE;
157 }
158 }
159
160 static unsigned translate_register_file(unsigned file)
161 {
162 switch(file) {
163 case TGSI_FILE_CONSTANT: return RC_FILE_CONSTANT;
164 case TGSI_FILE_IMMEDIATE: return RC_FILE_CONSTANT;
165 case TGSI_FILE_INPUT: return RC_FILE_INPUT;
166 case TGSI_FILE_OUTPUT: return RC_FILE_OUTPUT;
167 default:
168 fprintf(stderr, "Unhandled register file: %i\n", file);
169 /* fall-through */
170 case TGSI_FILE_TEMPORARY: return RC_FILE_TEMPORARY;
171 case TGSI_FILE_ADDRESS: return RC_FILE_ADDRESS;
172 }
173 }
174
175 static int translate_register_index(
176 struct tgsi_to_rc * ttr,
177 unsigned file,
178 int index)
179 {
180 if (file == TGSI_FILE_IMMEDIATE)
181 return ttr->immediate_offset + index;
182
183 return index;
184 }
185
186 static void transform_dstreg(
187 struct tgsi_to_rc * ttr,
188 struct rc_dst_register * dst,
189 struct tgsi_full_dst_register * src)
190 {
191 dst->File = translate_register_file(src->Register.File);
192 dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
193 dst->WriteMask = src->Register.WriteMask;
194 dst->RelAddr = src->Register.Indirect;
195 }
196
197 static void transform_srcreg(
198 struct tgsi_to_rc * ttr,
199 struct rc_src_register * dst,
200 struct tgsi_full_src_register * src)
201 {
202 unsigned i, j;
203
204 dst->File = translate_register_file(src->Register.File);
205 dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
206 dst->RelAddr = src->Register.Indirect;
207 dst->Swizzle = tgsi_util_get_full_src_register_swizzle(src, 0);
208 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 1) << 3;
209 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 2) << 6;
210 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 3) << 9;
211 dst->Abs = src->Register.Absolute;
212 dst->Negate = src->Register.Negate ? RC_MASK_XYZW : 0;
213
214 if (src->Register.File == TGSI_FILE_IMMEDIATE) {
215 for (i = 0; i < ttr->imms_to_swizzle_count; i++) {
216 if (ttr->imms_to_swizzle[i].index == src->Register.Index) {
217 dst->File = RC_FILE_TEMPORARY;
218 dst->Index = 0;
219 dst->Swizzle = 0;
220 for (j = 0; j < 4; j++) {
221 dst->Swizzle |= GET_SWZ(ttr->imms_to_swizzle[i].swizzle,
222 tgsi_util_get_full_src_register_swizzle(src, j)) << (j * 3);
223 }
224 break;
225 }
226 }
227 }
228 }
229
230 static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_texture src,
231 uint32_t *shadowSamplers)
232 {
233 switch(src.Texture) {
234 case TGSI_TEXTURE_1D:
235 dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
236 break;
237 case TGSI_TEXTURE_2D:
238 dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
239 break;
240 case TGSI_TEXTURE_3D:
241 dst->U.I.TexSrcTarget = RC_TEXTURE_3D;
242 break;
243 case TGSI_TEXTURE_CUBE:
244 dst->U.I.TexSrcTarget = RC_TEXTURE_CUBE;
245 break;
246 case TGSI_TEXTURE_RECT:
247 dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
248 break;
249 case TGSI_TEXTURE_SHADOW1D:
250 dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
251 dst->U.I.TexShadow = 1;
252 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
253 break;
254 case TGSI_TEXTURE_SHADOW2D:
255 dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
256 dst->U.I.TexShadow = 1;
257 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
258 break;
259 case TGSI_TEXTURE_SHADOWRECT:
260 dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
261 dst->U.I.TexShadow = 1;
262 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
263 break;
264 }
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 /* Allocate constants placeholders.
336 *
337 * Note: What if declared constants are not contiguous? */
338 for(i = 0; i <= ttr->info->file_max[TGSI_FILE_CONSTANT]; ++i) {
339 struct rc_constant constant;
340 memset(&constant, 0, sizeof(constant));
341 constant.Type = RC_CONSTANT_EXTERNAL;
342 constant.Size = 4;
343 constant.u.External = i;
344 rc_constants_add(&ttr->compiler->Program.Constants, &constant);
345 }
346
347 ttr->immediate_offset = ttr->compiler->Program.Constants.Count;
348
349 ttr->imms_to_swizzle = malloc(ttr->info->immediate_count * sizeof(struct swizzled_imms));
350 ttr->imms_to_swizzle_count = 0;
351
352 tgsi_parse_init(&parser, tokens);
353
354 while (!tgsi_parse_end_of_tokens(&parser)) {
355 tgsi_parse_token(&parser);
356
357 switch (parser.FullToken.Token.Type) {
358 case TGSI_TOKEN_TYPE_DECLARATION:
359 break;
360 case TGSI_TOKEN_TYPE_IMMEDIATE:
361 handle_immediate(ttr, &parser.FullToken.FullImmediate, imm_index);
362 imm_index++;
363 break;
364 case TGSI_TOKEN_TYPE_INSTRUCTION:
365 inst = &parser.FullToken.FullInstruction;
366 /* This hack with the RET opcode woudn't work with
367 * conditionals. */
368 if (inst->Instruction.Opcode == TGSI_OPCODE_END ||
369 inst->Instruction.Opcode == TGSI_OPCODE_RET) {
370 break;
371 }
372
373 transform_instruction(ttr, inst);
374 break;
375 }
376 }
377
378 tgsi_parse_free(&parser);
379
380 free(ttr->imms_to_swizzle);
381
382 rc_calculate_inputs_outputs(ttr->compiler);
383 }