Merge branch 'master' into r300-compiler
[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
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_LOOP: return RC_OPCODE_LOOP; */
111 /* case TGSI_OPCODE_REP: return RC_OPCODE_REP; */
112 /* case TGSI_OPCODE_ELSE: return RC_OPCODE_ELSE; */
113 /* case TGSI_OPCODE_ENDIF: return RC_OPCODE_ENDIF; */
114 /* case TGSI_OPCODE_ENDLOOP: return RC_OPCODE_ENDLOOP; */
115 /* case TGSI_OPCODE_ENDREP: return RC_OPCODE_ENDREP; */
116 /* case TGSI_OPCODE_PUSHA: return RC_OPCODE_PUSHA; */
117 /* case TGSI_OPCODE_POPA: return RC_OPCODE_POPA; */
118 /* case TGSI_OPCODE_CEIL: return RC_OPCODE_CEIL; */
119 /* case TGSI_OPCODE_I2F: return RC_OPCODE_I2F; */
120 /* case TGSI_OPCODE_NOT: return RC_OPCODE_NOT; */
121 /* case TGSI_OPCODE_TRUNC: return RC_OPCODE_TRUNC; */
122 /* case TGSI_OPCODE_SHL: return RC_OPCODE_SHL; */
123 /* case TGSI_OPCODE_SHR: return RC_OPCODE_SHR; */
124 /* case TGSI_OPCODE_AND: return RC_OPCODE_AND; */
125 /* case TGSI_OPCODE_OR: return RC_OPCODE_OR; */
126 /* case TGSI_OPCODE_MOD: return RC_OPCODE_MOD; */
127 /* case TGSI_OPCODE_XOR: return RC_OPCODE_XOR; */
128 /* case TGSI_OPCODE_SAD: return RC_OPCODE_SAD; */
129 /* case TGSI_OPCODE_TXF: return RC_OPCODE_TXF; */
130 /* case TGSI_OPCODE_TXQ: return RC_OPCODE_TXQ; */
131 /* case TGSI_OPCODE_CONT: return RC_OPCODE_CONT; */
132 /* case TGSI_OPCODE_EMIT: return RC_OPCODE_EMIT; */
133 /* case TGSI_OPCODE_ENDPRIM: return RC_OPCODE_ENDPRIM; */
134 /* case TGSI_OPCODE_BGNLOOP2: return RC_OPCODE_BGNLOOP2; */
135 /* case TGSI_OPCODE_BGNSUB: return RC_OPCODE_BGNSUB; */
136 /* case TGSI_OPCODE_ENDLOOP2: return RC_OPCODE_ENDLOOP2; */
137 /* case TGSI_OPCODE_ENDSUB: return RC_OPCODE_ENDSUB; */
138 /* case TGSI_OPCODE_NOISE1: return RC_OPCODE_NOISE1; */
139 /* case TGSI_OPCODE_NOISE2: return RC_OPCODE_NOISE2; */
140 /* case TGSI_OPCODE_NOISE3: return RC_OPCODE_NOISE3; */
141 /* case TGSI_OPCODE_NOISE4: return RC_OPCODE_NOISE4; */
142 case TGSI_OPCODE_NOP: return RC_OPCODE_NOP;
143 /* gap */
144 /* case TGSI_OPCODE_NRM4: return RC_OPCODE_NRM4; */
145 /* case TGSI_OPCODE_CALLNZ: return RC_OPCODE_CALLNZ; */
146 /* case TGSI_OPCODE_IFC: return RC_OPCODE_IFC; */
147 /* case TGSI_OPCODE_BREAKC: return RC_OPCODE_BREAKC; */
148 case TGSI_OPCODE_KIL: return RC_OPCODE_KIL;
149 case TGSI_OPCODE_SWZ: return RC_OPCODE_SWZ;
150 }
151
152 fprintf(stderr, "Unknown opcode: %i\n", opcode);
153 return RC_OPCODE_ILLEGAL_OPCODE;
154 }
155
156 static unsigned translate_saturate(unsigned saturate)
157 {
158 switch(saturate) {
159 default:
160 fprintf(stderr, "Unknown saturate mode: %i\n", saturate);
161 /* fall-through */
162 case TGSI_SAT_NONE: return RC_SATURATE_NONE;
163 case TGSI_SAT_ZERO_ONE: return RC_SATURATE_ZERO_ONE;
164 }
165 }
166
167 static unsigned translate_register_file(unsigned file)
168 {
169 switch(file) {
170 case TGSI_FILE_CONSTANT: return RC_FILE_CONSTANT;
171 case TGSI_FILE_IMMEDIATE: return RC_FILE_CONSTANT;
172 case TGSI_FILE_INPUT: return RC_FILE_INPUT;
173 case TGSI_FILE_OUTPUT: return RC_FILE_OUTPUT;
174 default:
175 fprintf(stderr, "Unhandled register file: %i\n", file);
176 /* fall-through */
177 case TGSI_FILE_TEMPORARY: return RC_FILE_TEMPORARY;
178 case TGSI_FILE_ADDRESS: return RC_FILE_ADDRESS;
179 }
180 }
181
182 static int translate_register_index(
183 struct tgsi_to_rc * ttr,
184 unsigned file,
185 int index)
186 {
187 if (file == TGSI_FILE_IMMEDIATE)
188 return ttr->immediate_offset + index;
189
190 return index;
191 }
192
193 static void transform_dstreg(
194 struct tgsi_to_rc * ttr,
195 struct rc_dst_register * dst,
196 struct tgsi_full_dst_register * src)
197 {
198 dst->File = translate_register_file(src->DstRegister.File);
199 dst->Index = translate_register_index(ttr, src->DstRegister.File, src->DstRegister.Index);
200 dst->WriteMask = src->DstRegister.WriteMask;
201 dst->RelAddr = src->DstRegister.Indirect;
202 }
203
204 static void transform_srcreg(
205 struct tgsi_to_rc * ttr,
206 struct rc_src_register * dst,
207 struct tgsi_full_src_register * src)
208 {
209 dst->File = translate_register_file(src->SrcRegister.File);
210 dst->Index = translate_register_index(ttr, src->SrcRegister.File, src->SrcRegister.Index);
211 dst->RelAddr = src->SrcRegister.Indirect;
212 dst->Swizzle = tgsi_util_get_full_src_register_extswizzle(src, 0);
213 dst->Swizzle |= tgsi_util_get_full_src_register_extswizzle(src, 1) << 3;
214 dst->Swizzle |= tgsi_util_get_full_src_register_extswizzle(src, 2) << 6;
215 dst->Swizzle |= tgsi_util_get_full_src_register_extswizzle(src, 3) << 9;
216 dst->Abs = src->SrcRegisterExtMod.Absolute;
217 dst->Negate =
218 src->SrcRegisterExtSwz.NegateX |
219 (src->SrcRegisterExtSwz.NegateY << 1) |
220 (src->SrcRegisterExtSwz.NegateZ << 2) |
221 (src->SrcRegisterExtSwz.NegateW << 3);
222 dst->Negate ^= src->SrcRegister.Negate ? RC_MASK_XYZW : 0;
223 }
224
225 static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_ext_texture src)
226 {
227 switch(src.Texture) {
228 case TGSI_TEXTURE_1D:
229 dst->I.TexSrcTarget = RC_TEXTURE_1D;
230 break;
231 case TGSI_TEXTURE_2D:
232 dst->I.TexSrcTarget = RC_TEXTURE_2D;
233 break;
234 case TGSI_TEXTURE_3D:
235 dst->I.TexSrcTarget = RC_TEXTURE_3D;
236 break;
237 case TGSI_TEXTURE_CUBE:
238 dst->I.TexSrcTarget = RC_TEXTURE_CUBE;
239 break;
240 case TGSI_TEXTURE_RECT:
241 dst->I.TexSrcTarget = RC_TEXTURE_RECT;
242 break;
243 case TGSI_TEXTURE_SHADOW1D:
244 dst->I.TexSrcTarget = RC_TEXTURE_1D;
245 dst->I.TexShadow = 1;
246 break;
247 case TGSI_TEXTURE_SHADOW2D:
248 dst->I.TexSrcTarget = RC_TEXTURE_2D;
249 dst->I.TexShadow = 1;
250 break;
251 case TGSI_TEXTURE_SHADOWRECT:
252 dst->I.TexSrcTarget = RC_TEXTURE_RECT;
253 dst->I.TexShadow = 1;
254 break;
255 }
256 }
257
258 static void transform_instruction(struct tgsi_to_rc * ttr, struct tgsi_full_instruction * src)
259 {
260 if (src->Instruction.Opcode == TGSI_OPCODE_END)
261 return;
262
263 struct rc_instruction * dst = rc_insert_new_instruction(ttr->compiler, ttr->compiler->Program.Instructions.Prev);
264 int i;
265
266 dst->I.Opcode = translate_opcode(src->Instruction.Opcode);
267 dst->I.SaturateMode = translate_saturate(src->Instruction.Saturate);
268
269 if (src->Instruction.NumDstRegs)
270 transform_dstreg(ttr, &dst->I.DstReg, &src->FullDstRegisters[0]);
271
272 for(i = 0; i < src->Instruction.NumSrcRegs; ++i) {
273 if (src->FullSrcRegisters[i].SrcRegister.File == TGSI_FILE_SAMPLER)
274 dst->I.TexSrcUnit = src->FullSrcRegisters[i].SrcRegister.Index;
275 else
276 transform_srcreg(ttr, &dst->I.SrcReg[i], &src->FullSrcRegisters[i]);
277 }
278
279 /* Texturing. */
280 transform_texture(dst, src->InstructionExtTexture);
281 }
282
283 static void handle_immediate(struct tgsi_to_rc * ttr, struct tgsi_full_immediate * imm)
284 {
285 struct rc_constant constant;
286 int i;
287
288 constant.Type = RC_CONSTANT_IMMEDIATE;
289 constant.Size = 4;
290 for(i = 0; i < 4; ++i)
291 constant.u.Immediate[i] = imm->u[i].Float;
292 rc_constants_add(&ttr->compiler->Program.Constants, &constant);
293 }
294
295 void r300_tgsi_to_rc(struct tgsi_to_rc * ttr, const struct tgsi_token * tokens)
296 {
297 struct tgsi_parse_context parser;
298 int i;
299
300 /* Allocate constants placeholders.
301 *
302 * Note: What if declared constants are not contiguous? */
303 for(i = 0; i <= ttr->info->file_max[TGSI_FILE_CONSTANT]; ++i) {
304 struct rc_constant constant;
305 memset(&constant, 0, sizeof(constant));
306 constant.Type = RC_CONSTANT_EXTERNAL;
307 constant.Size = 4;
308 constant.u.External = i;
309 rc_constants_add(&ttr->compiler->Program.Constants, &constant);
310 }
311
312 ttr->immediate_offset = ttr->compiler->Program.Constants.Count;
313
314 tgsi_parse_init(&parser, tokens);
315
316 while (!tgsi_parse_end_of_tokens(&parser)) {
317 tgsi_parse_token(&parser);
318
319 switch (parser.FullToken.Token.Type) {
320 case TGSI_TOKEN_TYPE_DECLARATION:
321 break;
322 case TGSI_TOKEN_TYPE_IMMEDIATE:
323 handle_immediate(ttr, &parser.FullToken.FullImmediate);
324 break;
325 case TGSI_TOKEN_TYPE_INSTRUCTION:
326 transform_instruction(ttr, &parser.FullToken.FullInstruction);
327 break;
328 }
329 }
330
331 tgsi_parse_free(&parser);
332
333 rc_calculate_inputs_outputs(ttr->compiler);
334 }
335