gallium: remove TGSI_OPCODE_ABS
[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 <stdio.h>
24 #include "r300_tgsi_to_rc.h"
25
26 #include "compiler/radeon_compiler.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_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_DPH: return RC_OPCODE_DPH;
65 case TGSI_OPCODE_COS: return RC_OPCODE_COS;
66 case TGSI_OPCODE_DDX: return RC_OPCODE_DDX;
67 case TGSI_OPCODE_DDY: return RC_OPCODE_DDY;
68 case TGSI_OPCODE_KILL: return RC_OPCODE_KILP;
69 /* case TGSI_OPCODE_PK2H: return RC_OPCODE_PK2H; */
70 /* case TGSI_OPCODE_PK2US: return RC_OPCODE_PK2US; */
71 /* case TGSI_OPCODE_PK4B: return RC_OPCODE_PK4B; */
72 /* case TGSI_OPCODE_PK4UB: return RC_OPCODE_PK4UB; */
73 case TGSI_OPCODE_SEQ: return RC_OPCODE_SEQ;
74 case TGSI_OPCODE_SGT: return RC_OPCODE_SGT;
75 case TGSI_OPCODE_SIN: return RC_OPCODE_SIN;
76 case TGSI_OPCODE_SLE: return RC_OPCODE_SLE;
77 case TGSI_OPCODE_SNE: return RC_OPCODE_SNE;
78 case TGSI_OPCODE_TEX: return RC_OPCODE_TEX;
79 case TGSI_OPCODE_TXD: return RC_OPCODE_TXD;
80 case TGSI_OPCODE_TXP: return RC_OPCODE_TXP;
81 /* case TGSI_OPCODE_UP2H: return RC_OPCODE_UP2H; */
82 /* case TGSI_OPCODE_UP2US: return RC_OPCODE_UP2US; */
83 /* case TGSI_OPCODE_UP4B: return RC_OPCODE_UP4B; */
84 /* case TGSI_OPCODE_UP4UB: return RC_OPCODE_UP4UB; */
85 case TGSI_OPCODE_ARR: return RC_OPCODE_ARR;
86 /* case TGSI_OPCODE_CAL: return RC_OPCODE_CAL; */
87 /* case TGSI_OPCODE_RET: return RC_OPCODE_RET; */
88 case TGSI_OPCODE_SSG: return RC_OPCODE_SSG;
89 case TGSI_OPCODE_CMP: return RC_OPCODE_CMP;
90 case TGSI_OPCODE_SCS: return RC_OPCODE_SCS;
91 case TGSI_OPCODE_TXB: return RC_OPCODE_TXB;
92 /* case TGSI_OPCODE_DIV: return RC_OPCODE_DIV; */
93 case TGSI_OPCODE_DP2: return RC_OPCODE_DP2;
94 case TGSI_OPCODE_TXL: return RC_OPCODE_TXL;
95 case TGSI_OPCODE_BRK: return RC_OPCODE_BRK;
96 case TGSI_OPCODE_IF: return RC_OPCODE_IF;
97 case TGSI_OPCODE_BGNLOOP: return RC_OPCODE_BGNLOOP;
98 case TGSI_OPCODE_ELSE: return RC_OPCODE_ELSE;
99 case TGSI_OPCODE_ENDIF: return RC_OPCODE_ENDIF;
100 case TGSI_OPCODE_ENDLOOP: return RC_OPCODE_ENDLOOP;
101 /* case TGSI_OPCODE_PUSHA: return RC_OPCODE_PUSHA; */
102 /* case TGSI_OPCODE_POPA: return RC_OPCODE_POPA; */
103 case TGSI_OPCODE_CEIL: return RC_OPCODE_CEIL;
104 /* case TGSI_OPCODE_I2F: return RC_OPCODE_I2F; */
105 /* case TGSI_OPCODE_NOT: return RC_OPCODE_NOT; */
106 case TGSI_OPCODE_TRUNC: return RC_OPCODE_TRUNC;
107 /* case TGSI_OPCODE_SHL: return RC_OPCODE_SHL; */
108 /* case TGSI_OPCODE_ISHR: return RC_OPCODE_SHR; */
109 /* case TGSI_OPCODE_AND: return RC_OPCODE_AND; */
110 /* case TGSI_OPCODE_OR: return RC_OPCODE_OR; */
111 /* case TGSI_OPCODE_MOD: return RC_OPCODE_MOD; */
112 /* case TGSI_OPCODE_XOR: return RC_OPCODE_XOR; */
113 /* case TGSI_OPCODE_SAD: return RC_OPCODE_SAD; */
114 /* case TGSI_OPCODE_TXF: return RC_OPCODE_TXF; */
115 /* case TGSI_OPCODE_TXQ: return RC_OPCODE_TXQ; */
116 case TGSI_OPCODE_CONT: return RC_OPCODE_CONT;
117 /* case TGSI_OPCODE_EMIT: return RC_OPCODE_EMIT; */
118 /* case TGSI_OPCODE_ENDPRIM: return RC_OPCODE_ENDPRIM; */
119 /* case TGSI_OPCODE_BGNLOOP2: return RC_OPCODE_BGNLOOP2; */
120 /* case TGSI_OPCODE_BGNSUB: return RC_OPCODE_BGNSUB; */
121 /* case TGSI_OPCODE_ENDLOOP2: return RC_OPCODE_ENDLOOP2; */
122 /* case TGSI_OPCODE_ENDSUB: return RC_OPCODE_ENDSUB; */
123 case TGSI_OPCODE_NOP: return RC_OPCODE_NOP;
124 /* case TGSI_OPCODE_CALLNZ: return RC_OPCODE_CALLNZ; */
125 /* case TGSI_OPCODE_BREAKC: return RC_OPCODE_BREAKC; */
126 case TGSI_OPCODE_KILL_IF: return RC_OPCODE_KIL;
127 }
128
129 fprintf(stderr, "r300: Unknown TGSI/RC opcode: %s\n", tgsi_get_opcode_name(opcode));
130 return RC_OPCODE_ILLEGAL_OPCODE;
131 }
132
133 static unsigned translate_saturate(unsigned saturate)
134 {
135 return saturate ? RC_SATURATE_ZERO_ONE : RC_SATURATE_NONE;
136 }
137
138 static unsigned translate_register_file(unsigned file)
139 {
140 switch(file) {
141 case TGSI_FILE_CONSTANT: return RC_FILE_CONSTANT;
142 case TGSI_FILE_IMMEDIATE: return RC_FILE_CONSTANT;
143 case TGSI_FILE_INPUT: return RC_FILE_INPUT;
144 case TGSI_FILE_OUTPUT: return RC_FILE_OUTPUT;
145 default:
146 fprintf(stderr, "Unhandled register file: %i\n", file);
147 /* fall-through */
148 case TGSI_FILE_TEMPORARY: return RC_FILE_TEMPORARY;
149 case TGSI_FILE_ADDRESS: return RC_FILE_ADDRESS;
150 }
151 }
152
153 static int translate_register_index(
154 struct tgsi_to_rc * ttr,
155 unsigned file,
156 int index)
157 {
158 if (file == TGSI_FILE_IMMEDIATE)
159 return ttr->immediate_offset + index;
160
161 return index;
162 }
163
164 static void transform_dstreg(
165 struct tgsi_to_rc * ttr,
166 struct rc_dst_register * dst,
167 struct tgsi_full_dst_register * src)
168 {
169 dst->File = translate_register_file(src->Register.File);
170 dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
171 dst->WriteMask = src->Register.WriteMask;
172
173 if (src->Register.Indirect) {
174 ttr->error = TRUE;
175 fprintf(stderr, "r300: Relative addressing of destination operands "
176 "is unsupported.\n");
177 }
178 }
179
180 static void transform_srcreg(
181 struct tgsi_to_rc * ttr,
182 struct rc_src_register * dst,
183 struct tgsi_full_src_register * src)
184 {
185 unsigned i, j;
186
187 dst->File = translate_register_file(src->Register.File);
188 dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
189 dst->RelAddr = src->Register.Indirect;
190 dst->Swizzle = tgsi_util_get_full_src_register_swizzle(src, 0);
191 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 1) << 3;
192 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 2) << 6;
193 dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 3) << 9;
194 dst->Abs = src->Register.Absolute;
195 dst->Negate = src->Register.Negate ? RC_MASK_XYZW : 0;
196
197 if (src->Register.File == TGSI_FILE_IMMEDIATE) {
198 for (i = 0; i < ttr->imms_to_swizzle_count; i++) {
199 if (ttr->imms_to_swizzle[i].index == src->Register.Index) {
200 dst->File = RC_FILE_TEMPORARY;
201 dst->Index = 0;
202 dst->Swizzle = 0;
203 for (j = 0; j < 4; j++) {
204 dst->Swizzle |= GET_SWZ(ttr->imms_to_swizzle[i].swizzle,
205 tgsi_util_get_full_src_register_swizzle(src, j)) << (j * 3);
206 }
207 break;
208 }
209 }
210 }
211 }
212
213 static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_texture src,
214 uint32_t *shadowSamplers)
215 {
216 switch(src.Texture) {
217 case TGSI_TEXTURE_1D:
218 dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
219 break;
220 case TGSI_TEXTURE_2D:
221 dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
222 break;
223 case TGSI_TEXTURE_3D:
224 dst->U.I.TexSrcTarget = RC_TEXTURE_3D;
225 break;
226 case TGSI_TEXTURE_CUBE:
227 dst->U.I.TexSrcTarget = RC_TEXTURE_CUBE;
228 break;
229 case TGSI_TEXTURE_RECT:
230 dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
231 break;
232 case TGSI_TEXTURE_SHADOW1D:
233 dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
234 dst->U.I.TexShadow = 1;
235 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
236 break;
237 case TGSI_TEXTURE_SHADOW2D:
238 dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
239 dst->U.I.TexShadow = 1;
240 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
241 break;
242 case TGSI_TEXTURE_SHADOWRECT:
243 dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
244 dst->U.I.TexShadow = 1;
245 *shadowSamplers |= 1 << dst->U.I.TexSrcUnit;
246 break;
247 }
248 dst->U.I.TexSwizzle = RC_SWIZZLE_XYZW;
249 }
250
251 static void transform_instruction(struct tgsi_to_rc * ttr, struct tgsi_full_instruction * src)
252 {
253 struct rc_instruction * dst;
254 int i;
255
256 dst = rc_insert_new_instruction(ttr->compiler, ttr->compiler->Program.Instructions.Prev);
257 dst->U.I.Opcode = translate_opcode(src->Instruction.Opcode);
258 dst->U.I.SaturateMode = translate_saturate(src->Instruction.Saturate);
259
260 if (src->Instruction.NumDstRegs)
261 transform_dstreg(ttr, &dst->U.I.DstReg, &src->Dst[0]);
262
263 for(i = 0; i < src->Instruction.NumSrcRegs; ++i) {
264 if (src->Src[i].Register.File == TGSI_FILE_SAMPLER)
265 dst->U.I.TexSrcUnit = src->Src[i].Register.Index;
266 else
267 transform_srcreg(ttr, &dst->U.I.SrcReg[i], &src->Src[i]);
268 }
269
270 /* Texturing. */
271 if (src->Instruction.Texture)
272 transform_texture(dst, src->Texture,
273 &ttr->compiler->Program.ShadowSamplers);
274 }
275
276 static void handle_immediate(struct tgsi_to_rc * ttr,
277 struct tgsi_full_immediate * imm,
278 unsigned index)
279 {
280 struct rc_constant constant;
281 unsigned swizzle = 0;
282 boolean can_swizzle = TRUE;
283 unsigned i;
284
285 for (i = 0; i < 4; i++) {
286 if (imm->u[i].Float == 0.0f) {
287 swizzle |= RC_SWIZZLE_ZERO << (i * 3);
288 } else if (imm->u[i].Float == 0.5f && ttr->use_half_swizzles) {
289 swizzle |= RC_SWIZZLE_HALF << (i * 3);
290 } else if (imm->u[i].Float == 1.0f) {
291 swizzle |= RC_SWIZZLE_ONE << (i * 3);
292 } else {
293 can_swizzle = FALSE;
294 break;
295 }
296 }
297
298 if (can_swizzle) {
299 ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].index = index;
300 ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].swizzle = swizzle;
301 ttr->imms_to_swizzle_count++;
302 } else {
303 constant.Type = RC_CONSTANT_IMMEDIATE;
304 constant.Size = 4;
305 for(i = 0; i < 4; ++i)
306 constant.u.Immediate[i] = imm->u[i].Float;
307 rc_constants_add(&ttr->compiler->Program.Constants, &constant);
308 }
309 }
310
311 void r300_tgsi_to_rc(struct tgsi_to_rc * ttr,
312 const struct tgsi_token * tokens)
313 {
314 struct tgsi_full_instruction *inst;
315 struct tgsi_parse_context parser;
316 unsigned imm_index = 0;
317 int i;
318
319 ttr->error = FALSE;
320
321 /* Allocate constants placeholders.
322 *
323 * Note: What if declared constants are not contiguous? */
324 for(i = 0; i <= ttr->info->file_max[TGSI_FILE_CONSTANT]; ++i) {
325 struct rc_constant constant;
326 memset(&constant, 0, sizeof(constant));
327 constant.Type = RC_CONSTANT_EXTERNAL;
328 constant.Size = 4;
329 constant.u.External = i;
330 rc_constants_add(&ttr->compiler->Program.Constants, &constant);
331 }
332
333 ttr->immediate_offset = ttr->compiler->Program.Constants.Count;
334
335 ttr->imms_to_swizzle = malloc(ttr->info->immediate_count * sizeof(struct swizzled_imms));
336 ttr->imms_to_swizzle_count = 0;
337
338 tgsi_parse_init(&parser, tokens);
339
340 while (!tgsi_parse_end_of_tokens(&parser)) {
341 tgsi_parse_token(&parser);
342
343 switch (parser.FullToken.Token.Type) {
344 case TGSI_TOKEN_TYPE_DECLARATION:
345 break;
346 case TGSI_TOKEN_TYPE_IMMEDIATE:
347 handle_immediate(ttr, &parser.FullToken.FullImmediate, imm_index);
348 imm_index++;
349 break;
350 case TGSI_TOKEN_TYPE_INSTRUCTION:
351 inst = &parser.FullToken.FullInstruction;
352 if (inst->Instruction.Opcode == TGSI_OPCODE_END) {
353 break;
354 }
355
356 transform_instruction(ttr, inst);
357 break;
358 }
359 }
360
361 tgsi_parse_free(&parser);
362
363 free(ttr->imms_to_swizzle);
364
365 rc_calculate_inputs_outputs(ttr->compiler);
366 }