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 #include "util/compiler.h"
34 
translate_opcode(unsigned opcode)35 static unsigned translate_opcode(unsigned opcode)
36 {
37     switch(opcode) {
38         case TGSI_OPCODE_ARL: return RC_OPCODE_ARL;
39         case TGSI_OPCODE_MOV: return RC_OPCODE_MOV;
40         case TGSI_OPCODE_LIT: return RC_OPCODE_LIT;
41         case TGSI_OPCODE_RCP: return RC_OPCODE_RCP;
42         case TGSI_OPCODE_RSQ: return RC_OPCODE_RSQ;
43         case TGSI_OPCODE_EXP: return RC_OPCODE_EXP;
44         case TGSI_OPCODE_LOG: return RC_OPCODE_LOG;
45         case TGSI_OPCODE_MUL: return RC_OPCODE_MUL;
46         case TGSI_OPCODE_ADD: return RC_OPCODE_ADD;
47         case TGSI_OPCODE_DP3: return RC_OPCODE_DP3;
48         case TGSI_OPCODE_DP4: return RC_OPCODE_DP4;
49         case TGSI_OPCODE_DST: return RC_OPCODE_DST;
50         case TGSI_OPCODE_MIN: return RC_OPCODE_MIN;
51         case TGSI_OPCODE_MAX: return RC_OPCODE_MAX;
52         case TGSI_OPCODE_SLT: return RC_OPCODE_SLT;
53         case TGSI_OPCODE_SGE: return RC_OPCODE_SGE;
54         case TGSI_OPCODE_MAD: return RC_OPCODE_MAD;
55         case TGSI_OPCODE_LRP: return RC_OPCODE_LRP;
56         case TGSI_OPCODE_FRC: return RC_OPCODE_FRC;
57         case TGSI_OPCODE_FLR: return RC_OPCODE_FLR;
58         case TGSI_OPCODE_ROUND: return RC_OPCODE_ROUND;
59         case TGSI_OPCODE_EX2: return RC_OPCODE_EX2;
60         case TGSI_OPCODE_LG2: return RC_OPCODE_LG2;
61         case TGSI_OPCODE_POW: return RC_OPCODE_POW;
62         case TGSI_OPCODE_COS: return RC_OPCODE_COS;
63         case TGSI_OPCODE_DDX: return RC_OPCODE_DDX;
64         case TGSI_OPCODE_DDY: return RC_OPCODE_DDY;
65         case TGSI_OPCODE_KILL: return RC_OPCODE_KILP;
66      /* case TGSI_OPCODE_PK2H: return RC_OPCODE_PK2H; */
67      /* case TGSI_OPCODE_PK2US: return RC_OPCODE_PK2US; */
68      /* case TGSI_OPCODE_PK4B: return RC_OPCODE_PK4B; */
69      /* case TGSI_OPCODE_PK4UB: return RC_OPCODE_PK4UB; */
70         case TGSI_OPCODE_SEQ: return RC_OPCODE_SEQ;
71         case TGSI_OPCODE_SGT: return RC_OPCODE_SGT;
72         case TGSI_OPCODE_SIN: return RC_OPCODE_SIN;
73         case TGSI_OPCODE_SLE: return RC_OPCODE_SLE;
74         case TGSI_OPCODE_SNE: return RC_OPCODE_SNE;
75         case TGSI_OPCODE_TEX: return RC_OPCODE_TEX;
76         case TGSI_OPCODE_TXD: return RC_OPCODE_TXD;
77         case TGSI_OPCODE_TXP: return RC_OPCODE_TXP;
78      /* case TGSI_OPCODE_UP2H: return RC_OPCODE_UP2H; */
79      /* case TGSI_OPCODE_UP2US: return RC_OPCODE_UP2US; */
80      /* case TGSI_OPCODE_UP4B: return RC_OPCODE_UP4B; */
81      /* case TGSI_OPCODE_UP4UB: return RC_OPCODE_UP4UB; */
82         case TGSI_OPCODE_ARR: return RC_OPCODE_ARR;
83      /* case TGSI_OPCODE_CAL: return RC_OPCODE_CAL; */
84      /* case TGSI_OPCODE_RET: return RC_OPCODE_RET; */
85         case TGSI_OPCODE_SSG: return RC_OPCODE_SSG;
86         case TGSI_OPCODE_CMP: return RC_OPCODE_CMP;
87         case TGSI_OPCODE_TXB: return RC_OPCODE_TXB;
88      /* case TGSI_OPCODE_DIV: return RC_OPCODE_DIV; */
89         case TGSI_OPCODE_DP2: return RC_OPCODE_DP2;
90         case TGSI_OPCODE_TXL: return RC_OPCODE_TXL;
91         case TGSI_OPCODE_BRK: return RC_OPCODE_BRK;
92         case TGSI_OPCODE_IF: return RC_OPCODE_IF;
93         case TGSI_OPCODE_BGNLOOP: return RC_OPCODE_BGNLOOP;
94         case TGSI_OPCODE_ELSE: return RC_OPCODE_ELSE;
95         case TGSI_OPCODE_ENDIF: return RC_OPCODE_ENDIF;
96         case TGSI_OPCODE_ENDLOOP: return RC_OPCODE_ENDLOOP;
97      /* case TGSI_OPCODE_PUSHA: return RC_OPCODE_PUSHA; */
98      /* case TGSI_OPCODE_POPA: return RC_OPCODE_POPA; */
99         case TGSI_OPCODE_CEIL: return RC_OPCODE_CEIL;
100      /* case TGSI_OPCODE_I2F: return RC_OPCODE_I2F; */
101      /* case TGSI_OPCODE_NOT: return RC_OPCODE_NOT; */
102         case TGSI_OPCODE_TRUNC: return RC_OPCODE_TRUNC;
103      /* case TGSI_OPCODE_SHL: return RC_OPCODE_SHL; */
104      /* case TGSI_OPCODE_ISHR: return RC_OPCODE_SHR; */
105      /* case TGSI_OPCODE_AND: return RC_OPCODE_AND; */
106      /* case TGSI_OPCODE_OR: return RC_OPCODE_OR; */
107      /* case TGSI_OPCODE_MOD: return RC_OPCODE_MOD; */
108      /* case TGSI_OPCODE_XOR: return RC_OPCODE_XOR; */
109      /* case TGSI_OPCODE_TXF: return RC_OPCODE_TXF; */
110      /* case TGSI_OPCODE_TXQ: return RC_OPCODE_TXQ; */
111         case TGSI_OPCODE_CONT: return RC_OPCODE_CONT;
112      /* case TGSI_OPCODE_EMIT: return RC_OPCODE_EMIT; */
113      /* case TGSI_OPCODE_ENDPRIM: return RC_OPCODE_ENDPRIM; */
114      /* case TGSI_OPCODE_BGNLOOP2: return RC_OPCODE_BGNLOOP2; */
115      /* case TGSI_OPCODE_BGNSUB: return RC_OPCODE_BGNSUB; */
116      /* case TGSI_OPCODE_ENDLOOP2: return RC_OPCODE_ENDLOOP2; */
117      /* case TGSI_OPCODE_ENDSUB: return RC_OPCODE_ENDSUB; */
118         case TGSI_OPCODE_NOP: return RC_OPCODE_NOP;
119         case TGSI_OPCODE_KILL_IF: return RC_OPCODE_KIL;
120     }
121 
122     fprintf(stderr, "r300: Unknown TGSI/RC opcode: %s\n", tgsi_get_opcode_name(opcode));
123     return RC_OPCODE_ILLEGAL_OPCODE;
124 }
125 
translate_saturate(unsigned saturate)126 static unsigned translate_saturate(unsigned saturate)
127 {
128     return saturate ? RC_SATURATE_ZERO_ONE : RC_SATURATE_NONE;
129 }
130 
translate_register_file(unsigned file)131 static unsigned translate_register_file(unsigned file)
132 {
133     switch(file) {
134         case TGSI_FILE_CONSTANT: return RC_FILE_CONSTANT;
135         case TGSI_FILE_IMMEDIATE: return RC_FILE_CONSTANT;
136         case TGSI_FILE_INPUT: return RC_FILE_INPUT;
137         case TGSI_FILE_OUTPUT: return RC_FILE_OUTPUT;
138         default:
139             fprintf(stderr, "Unhandled register file: %i\n", file);
140             FALLTHROUGH;
141         case TGSI_FILE_TEMPORARY: return RC_FILE_TEMPORARY;
142         case TGSI_FILE_ADDRESS: return RC_FILE_ADDRESS;
143     }
144 }
145 
translate_register_index(struct tgsi_to_rc * ttr,unsigned file,int index)146 static int translate_register_index(
147     struct tgsi_to_rc * ttr,
148     unsigned file,
149     int index)
150 {
151     if (file == TGSI_FILE_IMMEDIATE)
152         return ttr->immediate_offset + index;
153 
154     return index;
155 }
156 
transform_dstreg(struct tgsi_to_rc * ttr,struct rc_dst_register * dst,struct tgsi_full_dst_register * src)157 static void transform_dstreg(
158     struct tgsi_to_rc * ttr,
159     struct rc_dst_register * dst,
160     struct tgsi_full_dst_register * src)
161 {
162     dst->File = translate_register_file(src->Register.File);
163     dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
164     dst->WriteMask = src->Register.WriteMask;
165 
166     if (src->Register.Indirect) {
167         ttr->error = TRUE;
168         fprintf(stderr, "r300: Relative addressing of destination operands "
169                 "is unsupported.\n");
170     }
171 }
172 
transform_srcreg(struct tgsi_to_rc * ttr,struct rc_src_register * dst,struct tgsi_full_src_register * src)173 static void transform_srcreg(
174     struct tgsi_to_rc * ttr,
175     struct rc_src_register * dst,
176     struct tgsi_full_src_register * src)
177 {
178     unsigned i, j;
179 
180     dst->File = translate_register_file(src->Register.File);
181     dst->Index = translate_register_index(ttr, src->Register.File, src->Register.Index);
182     dst->RelAddr = src->Register.Indirect;
183     dst->Swizzle = tgsi_util_get_full_src_register_swizzle(src, 0);
184     dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 1) << 3;
185     dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 2) << 6;
186     dst->Swizzle |= tgsi_util_get_full_src_register_swizzle(src, 3) << 9;
187     dst->Abs = src->Register.Absolute;
188     dst->Negate = src->Register.Negate ? RC_MASK_XYZW : 0;
189 
190     if (src->Register.File == TGSI_FILE_IMMEDIATE) {
191         for (i = 0; i < ttr->imms_to_swizzle_count; i++) {
192             if (ttr->imms_to_swizzle[i].index == src->Register.Index) {
193                 dst->File = RC_FILE_TEMPORARY;
194                 dst->Index = 0;
195                 dst->Swizzle = 0;
196                 for (j = 0; j < 4; j++) {
197                     dst->Swizzle |= GET_SWZ(ttr->imms_to_swizzle[i].swizzle,
198                         tgsi_util_get_full_src_register_swizzle(src, j)) << (j * 3);
199                 }
200                 break;
201             }
202         }
203     }
204 }
205 
transform_texture(struct rc_instruction * dst,struct tgsi_instruction_texture src,uint32_t * shadowSamplers)206 static void transform_texture(struct rc_instruction * dst, struct tgsi_instruction_texture src,
207                               uint32_t *shadowSamplers)
208 {
209     switch(src.Texture) {
210         case TGSI_TEXTURE_1D:
211             dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
212             break;
213         case TGSI_TEXTURE_2D:
214             dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
215             break;
216         case TGSI_TEXTURE_3D:
217             dst->U.I.TexSrcTarget = RC_TEXTURE_3D;
218             break;
219         case TGSI_TEXTURE_CUBE:
220             dst->U.I.TexSrcTarget = RC_TEXTURE_CUBE;
221             break;
222         case TGSI_TEXTURE_RECT:
223             dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
224             break;
225         case TGSI_TEXTURE_SHADOW1D:
226             dst->U.I.TexSrcTarget = RC_TEXTURE_1D;
227             dst->U.I.TexShadow = 1;
228             *shadowSamplers |= 1U << dst->U.I.TexSrcUnit;
229             break;
230         case TGSI_TEXTURE_SHADOW2D:
231             dst->U.I.TexSrcTarget = RC_TEXTURE_2D;
232             dst->U.I.TexShadow = 1;
233             *shadowSamplers |= 1U << dst->U.I.TexSrcUnit;
234             break;
235         case TGSI_TEXTURE_SHADOWRECT:
236             dst->U.I.TexSrcTarget = RC_TEXTURE_RECT;
237             dst->U.I.TexShadow = 1;
238             *shadowSamplers |= 1U << dst->U.I.TexSrcUnit;
239             break;
240         case TGSI_TEXTURE_1D_ARRAY:
241             dst->U.I.TexSrcTarget = RC_TEXTURE_1D_ARRAY;
242             break;
243         case TGSI_TEXTURE_2D_ARRAY:
244             dst->U.I.TexSrcTarget = RC_TEXTURE_2D_ARRAY;
245             break;
246         case TGSI_TEXTURE_SHADOW1D_ARRAY:
247             dst->U.I.TexSrcTarget = RC_TEXTURE_1D_ARRAY;
248             dst->U.I.TexShadow = 1;
249             *shadowSamplers |= 1U << dst->U.I.TexSrcUnit;
250             break;
251         case TGSI_TEXTURE_SHADOW2D_ARRAY:
252             dst->U.I.TexSrcTarget = RC_TEXTURE_2D_ARRAY;
253             dst->U.I.TexShadow = 1;
254             *shadowSamplers |= 1U << dst->U.I.TexSrcUnit;
255             break;
256         case TGSI_TEXTURE_SHADOWCUBE:
257             dst->U.I.TexSrcTarget = RC_TEXTURE_CUBE;
258             dst->U.I.TexShadow = 1;
259             *shadowSamplers |= 1U << dst->U.I.TexSrcUnit;
260             break;
261     }
262     dst->U.I.TexSwizzle = RC_SWIZZLE_XYZW;
263 }
264 
transform_instruction(struct tgsi_to_rc * ttr,struct tgsi_full_instruction * src)265 static void transform_instruction(struct tgsi_to_rc * ttr, struct tgsi_full_instruction * src)
266 {
267     struct rc_instruction * dst;
268     int i;
269 
270     dst = rc_insert_new_instruction(ttr->compiler, ttr->compiler->Program.Instructions.Prev);
271     dst->U.I.Opcode = translate_opcode(src->Instruction.Opcode);
272     dst->U.I.SaturateMode = translate_saturate(src->Instruction.Saturate);
273 
274     if (src->Instruction.NumDstRegs)
275         transform_dstreg(ttr, &dst->U.I.DstReg, &src->Dst[0]);
276 
277     for(i = 0; i < src->Instruction.NumSrcRegs; ++i) {
278         if (src->Src[i].Register.File == TGSI_FILE_SAMPLER)
279             dst->U.I.TexSrcUnit = src->Src[i].Register.Index;
280         else
281             transform_srcreg(ttr, &dst->U.I.SrcReg[i], &src->Src[i]);
282     }
283 
284     /* Texturing. */
285     if (src->Instruction.Texture)
286         transform_texture(dst, src->Texture,
287                           &ttr->compiler->Program.ShadowSamplers);
288 }
289 
handle_immediate(struct tgsi_to_rc * ttr,struct tgsi_full_immediate * imm,unsigned index)290 static void handle_immediate(struct tgsi_to_rc * ttr,
291                              struct tgsi_full_immediate * imm,
292                              unsigned index)
293 {
294     struct rc_constant constant;
295     unsigned swizzle = 0;
296     boolean can_swizzle = TRUE;
297     unsigned i;
298 
299     for (i = 0; i < 4; i++) {
300         if (imm->u[i].Float == 0.0f) {
301             swizzle |= RC_SWIZZLE_ZERO << (i * 3);
302         } else if (imm->u[i].Float == 0.5f && ttr->use_half_swizzles) {
303             swizzle |= RC_SWIZZLE_HALF << (i * 3);
304         } else if (imm->u[i].Float == 1.0f) {
305             swizzle |= RC_SWIZZLE_ONE << (i * 3);
306         } else {
307             can_swizzle = FALSE;
308             break;
309         }
310     }
311 
312     if (can_swizzle) {
313         ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].index = index;
314         ttr->imms_to_swizzle[ttr->imms_to_swizzle_count].swizzle = swizzle;
315         ttr->imms_to_swizzle_count++;
316     } else {
317         constant.Type = RC_CONSTANT_IMMEDIATE;
318         constant.Size = 4;
319         for(i = 0; i < 4; ++i)
320             constant.u.Immediate[i] = imm->u[i].Float;
321         rc_constants_add(&ttr->compiler->Program.Constants, &constant);
322     }
323 }
324 
r300_tgsi_to_rc(struct tgsi_to_rc * ttr,const struct tgsi_token * tokens)325 void r300_tgsi_to_rc(struct tgsi_to_rc * ttr,
326                      const struct tgsi_token * tokens)
327 {
328     struct tgsi_full_instruction *inst;
329     struct tgsi_parse_context parser;
330     unsigned imm_index = 0;
331     int i;
332 
333     ttr->error = FALSE;
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                 if (inst->Instruction.Opcode == TGSI_OPCODE_END) {
367                     break;
368                 }
369 
370                 transform_instruction(ttr, inst);
371                 break;
372         }
373     }
374 
375     tgsi_parse_free(&parser);
376 
377     free(ttr->imms_to_swizzle);
378 
379     rc_calculate_inputs_outputs(ttr->compiler);
380 }
381