1// Copyright (c) 2015-2016 The Khronos Group Inc.
2//
3// Licensed under the Apache License, Version 2.0 (the "License");
4// you may not use this file except in compliance with the License.
5// You may obtain a copy of the License at
6//
7// http://www.apache.org/licenses/LICENSE-2.0
8//
9// Unless required by applicable law or agreed to in writing, software
10// distributed under the License is distributed on an "AS IS" BASIS,
11// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12// See the License for the specific language governing permissions and
13// limitations under the License.
14
15#include "source/text.h"
16
17#include <algorithm>
18#include <cassert>
19#include <cctype>
20#include <cstdio>
21#include <cstdlib>
22#include <cstring>
23#include <memory>
24#include <set>
25#include <sstream>
26#include <string>
27#include <unordered_map>
28#include <utility>
29#include <vector>
30
31#include "source/assembly_grammar.h"
32#include "source/binary.h"
33#include "source/diagnostic.h"
34#include "source/ext_inst.h"
35#include "source/instruction.h"
36#include "source/opcode.h"
37#include "source/operand.h"
38#include "source/spirv_constant.h"
39#include "source/spirv_target_env.h"
40#include "source/table.h"
41#include "source/text_handler.h"
42#include "source/util/bitutils.h"
43#include "source/util/parse_number.h"
44#include "spirv-tools/libspirv.h"
45
46bool spvIsValidIDCharacter(const char value) {
47 return value == '_' || 0 != ::isalnum(value);
48}
49
50// Returns true if the given string represents a valid ID name.
51bool spvIsValidID(const char* textValue) {
52 const char* c = textValue;
53 for (; *c != '\0'; ++c) {
54 if (!spvIsValidIDCharacter(*c)) {
55 return false;
56 }
57 }
58 // If the string was empty, then the ID also is not valid.
59 return c != textValue;
60}
61
62// Text API
63
64spv_result_t spvTextToLiteral(const char* textValue, spv_literal_t* pLiteral) {
65 bool isSigned = false;
66 int numPeriods = 0;
67 bool isString = false;
68
69 const size_t len = strlen(textValue);
70 if (len == 0) return SPV_FAILED_MATCH;
71
72 for (uint64_t index = 0; index < len; ++index) {
73 switch (textValue[index]) {
74 case '0':
75 case '1':
76 case '2':
77 case '3':
78 case '4':
79 case '5':
80 case '6':
81 case '7':
82 case '8':
83 case '9':
84 break;
85 case '.':
86 numPeriods++;
87 break;
88 case '-':
89 if (index == 0) {
90 isSigned = true;
91 } else {
92 isString = true;
93 }
94 break;
95 default:
96 isString = true;
97 index = len; // break out of the loop too.
98 break;
99 }
100 }
101
102 pLiteral->type = spv_literal_type_t(99);
103
104 if (isString || numPeriods > 1 || (isSigned && len == 1)) {
105 if (len < 2 || textValue[0] != '"' || textValue[len - 1] != '"')
106 return SPV_FAILED_MATCH;
107 bool escaping = false;
108 for (const char* val = textValue + 1; val != textValue + len - 1; ++val) {
109 if ((*val == '\\') && (!escaping)) {
110 escaping = true;
111 } else {
112 // Have to save space for the null-terminator
113 if (pLiteral->str.size() >= SPV_LIMIT_LITERAL_STRING_BYTES_MAX)
114 return SPV_ERROR_OUT_OF_MEMORY;
115 pLiteral->str.push_back(*val);
116 escaping = false;
117 }
118 }
119
120 pLiteral->type = SPV_LITERAL_TYPE_STRING;
121 } else if (numPeriods == 1) {
122 double d = std::strtod(textValue, nullptr);
123 float f = (float)d;
124 if (d == (double)f) {
125 pLiteral->type = SPV_LITERAL_TYPE_FLOAT_32;
126 pLiteral->value.f = f;
127 } else {
128 pLiteral->type = SPV_LITERAL_TYPE_FLOAT_64;
129 pLiteral->value.d = d;
130 }
131 } else if (isSigned) {
132 int64_t i64 = strtoll(textValue, nullptr, 10);
133 int32_t i32 = (int32_t)i64;
134 if (i64 == (int64_t)i32) {
135 pLiteral->type = SPV_LITERAL_TYPE_INT_32;
136 pLiteral->value.i32 = i32;
137 } else {
138 pLiteral->type = SPV_LITERAL_TYPE_INT_64;
139 pLiteral->value.i64 = i64;
140 }
141 } else {
142 uint64_t u64 = strtoull(textValue, nullptr, 10);
143 uint32_t u32 = (uint32_t)u64;
144 if (u64 == (uint64_t)u32) {
145 pLiteral->type = SPV_LITERAL_TYPE_UINT_32;
146 pLiteral->value.u32 = u32;
147 } else {
148 pLiteral->type = SPV_LITERAL_TYPE_UINT_64;
149 pLiteral->value.u64 = u64;
150 }
151 }
152
153 return SPV_SUCCESS;
154}
155
156namespace {
157
158/// Parses an immediate integer from text, guarding against overflow. If
159/// successful, adds the parsed value to pInst, advances the context past it,
160/// and returns SPV_SUCCESS. Otherwise, leaves pInst alone, emits diagnostics,
161/// and returns SPV_ERROR_INVALID_TEXT.
162spv_result_t encodeImmediate(spvtools::AssemblyContext* context,
163 const char* text, spv_instruction_t* pInst) {
164 assert(*text == '!');
165 uint32_t parse_result;
166 if (!spvtools::utils::ParseNumber(text + 1, &parse_result)) {
167 return context->diagnostic(SPV_ERROR_INVALID_TEXT)
168 << "Invalid immediate integer: !" << text + 1;
169 }
170 context->binaryEncodeU32(parse_result, pInst);
171 context->seekForward(static_cast<uint32_t>(strlen(text)));
172 return SPV_SUCCESS;
173}
174
175} // anonymous namespace
176
177/// @brief Translate an Opcode operand to binary form
178///
179/// @param[in] grammar the grammar to use for compilation
180/// @param[in, out] context the dynamic compilation info
181/// @param[in] type of the operand
182/// @param[in] textValue word of text to be parsed
183/// @param[out] pInst return binary Opcode
184/// @param[in,out] pExpectedOperands the operand types expected
185///
186/// @return result code
187spv_result_t spvTextEncodeOperand(const spvtools::AssemblyGrammar& grammar,
188 spvtools::AssemblyContext* context,
189 const spv_operand_type_t type,
190 const char* textValue,
191 spv_instruction_t* pInst,
192 spv_operand_pattern_t* pExpectedOperands) {
193 // NOTE: Handle immediate int in the stream
194 if ('!' == textValue[0]) {
195 if (auto error = encodeImmediate(context, textValue, pInst)) {
196 return error;
197 }
198 *pExpectedOperands =
199 spvAlternatePatternFollowingImmediate(*pExpectedOperands);
200 return SPV_SUCCESS;
201 }
202
203 // Optional literal operands can fail to parse. In that case use
204 // SPV_FAILED_MATCH to avoid emitting a diagostic. Use the following
205 // for those situations.
206 spv_result_t error_code_for_literals =
207 spvOperandIsOptional(type) ? SPV_FAILED_MATCH : SPV_ERROR_INVALID_TEXT;
208
209 switch (type) {
210 case SPV_OPERAND_TYPE_ID:
211 case SPV_OPERAND_TYPE_TYPE_ID:
212 case SPV_OPERAND_TYPE_RESULT_ID:
213 case SPV_OPERAND_TYPE_MEMORY_SEMANTICS_ID:
214 case SPV_OPERAND_TYPE_SCOPE_ID:
215 case SPV_OPERAND_TYPE_OPTIONAL_ID: {
216 if ('%' == textValue[0]) {
217 textValue++;
218 } else {
219 return context->diagnostic() << "Expected id to start with %.";
220 }
221 if (!spvIsValidID(textValue)) {
222 return context->diagnostic() << "Invalid ID " << textValue;
223 }
224 const uint32_t id = context->spvNamedIdAssignOrGet(textValue);
225 if (type == SPV_OPERAND_TYPE_TYPE_ID) pInst->resultTypeId = id;
226 spvInstructionAddWord(pInst, id);
227
228 // Set the extended instruction type.
229 // The import set id is the 3rd operand of OpExtInst.
230 if (pInst->opcode == SpvOpExtInst && pInst->words.size() == 4) {
231 auto ext_inst_type = context->getExtInstTypeForId(pInst->words[3]);
232 if (ext_inst_type == SPV_EXT_INST_TYPE_NONE) {
233 return context->diagnostic()
234 << "Invalid extended instruction import Id "
235 << pInst->words[2];
236 }
237 pInst->extInstType = ext_inst_type;
238 }
239 } break;
240
241 case SPV_OPERAND_TYPE_EXTENSION_INSTRUCTION_NUMBER: {
242 // The assembler accepts the symbolic name for an extended instruction,
243 // and emits its corresponding number.
244 spv_ext_inst_desc extInst;
245 if (grammar.lookupExtInst(pInst->extInstType, textValue, &extInst) ==
246 SPV_SUCCESS) {
247 // if we know about this extended instruction, push the numeric value
248 spvInstructionAddWord(pInst, extInst->ext_inst);
249
250 // Prepare to parse the operands for the extended instructions.
251 spvPushOperandTypes(extInst->operandTypes, pExpectedOperands);
252 } else {
253 // if we don't know this extended instruction and the set isn't
254 // non-semantic, we cannot process further
255 if (!spvExtInstIsNonSemantic(pInst->extInstType)) {
256 return context->diagnostic()
257 << "Invalid extended instruction name '" << textValue << "'.";
258 } else {
259 // for non-semantic instruction sets, as long as the text name is an
260 // integer value we can encode it since we know the form of all such
261 // extended instructions
262 spv_literal_t extInstValue;
263 if (spvTextToLiteral(textValue, &extInstValue) ||
264 extInstValue.type != SPV_LITERAL_TYPE_UINT_32) {
265 return context->diagnostic()
266 << "Couldn't translate unknown extended instruction name '"
267 << textValue << "' to unsigned integer.";
268 }
269
270 spvInstructionAddWord(pInst, extInstValue.value.u32);
271
272 // opcode contains an unknown number of IDs.
273 pExpectedOperands->push_back(SPV_OPERAND_TYPE_VARIABLE_ID);
274 }
275 }
276 } break;
277
278 case SPV_OPERAND_TYPE_SPEC_CONSTANT_OP_NUMBER: {
279 // The assembler accepts the symbolic name for the opcode, but without
280 // the "Op" prefix. For example, "IAdd" is accepted. The number
281 // of the opcode is emitted.
282 SpvOp opcode;
283 if (grammar.lookupSpecConstantOpcode(textValue, &opcode)) {
284 return context->diagnostic() << "Invalid " << spvOperandTypeStr(type)
285 << " '" << textValue << "'.";
286 }
287 spv_opcode_desc opcodeEntry = nullptr;
288 if (grammar.lookupOpcode(opcode, &opcodeEntry)) {
289 return context->diagnostic(SPV_ERROR_INTERNAL)
290 << "OpSpecConstant opcode table out of sync";
291 }
292 spvInstructionAddWord(pInst, uint32_t(opcodeEntry->opcode));
293
294 // Prepare to parse the operands for the opcode. Except skip the
295 // type Id and result Id, since they've already been processed.
296 assert(opcodeEntry->hasType);
297 assert(opcodeEntry->hasResult);
298 assert(opcodeEntry->numTypes >= 2);
299 spvPushOperandTypes(opcodeEntry->operandTypes + 2, pExpectedOperands);
300 } break;
301
302 case SPV_OPERAND_TYPE_LITERAL_INTEGER:
303 case SPV_OPERAND_TYPE_OPTIONAL_LITERAL_INTEGER: {
304 // The current operand is an *unsigned* 32-bit integer.
305 // That's just how the grammar works.
306 spvtools::IdType expected_type = {
307 32, false, spvtools::IdTypeClass::kScalarIntegerType};
308 if (auto error = context->binaryEncodeNumericLiteral(
309 textValue, error_code_for_literals, expected_type, pInst)) {
310 return error;
311 }
312 } break;
313
314 case SPV_OPERAND_TYPE_OPTIONAL_LITERAL_NUMBER:
315 // This is a context-independent literal number which can be a 32-bit
316 // number of floating point value.
317 if (auto error = context->binaryEncodeNumericLiteral(
318 textValue, error_code_for_literals, spvtools::kUnknownType,
319 pInst)) {
320 return error;
321 }
322 break;
323
324 case SPV_OPERAND_TYPE_OPTIONAL_TYPED_LITERAL_INTEGER:
325 case SPV_OPERAND_TYPE_TYPED_LITERAL_NUMBER: {
326 spvtools::IdType expected_type = spvtools::kUnknownType;
327 // The encoding for OpConstant, OpSpecConstant and OpSwitch all
328 // depend on either their own result-id or the result-id of
329 // one of their parameters.
330 if (SpvOpConstant == pInst->opcode ||
331 SpvOpSpecConstant == pInst->opcode) {
332 // The type of the literal is determined by the type Id of the
333 // instruction.
334 expected_type =
335 context->getTypeOfTypeGeneratingValue(pInst->resultTypeId);
336 if (!spvtools::isScalarFloating(expected_type) &&
337 !spvtools::isScalarIntegral(expected_type)) {
338 spv_opcode_desc d;
339 const char* opcode_name = "opcode";
340 if (SPV_SUCCESS == grammar.lookupOpcode(pInst->opcode, &d)) {
341 opcode_name = d->name;
342 }
343 return context->diagnostic()
344 << "Type for " << opcode_name
345 << " must be a scalar floating point or integer type";
346 }
347 } else if (pInst->opcode == SpvOpSwitch) {
348 // The type of the literal is the same as the type of the selector.
349 expected_type = context->getTypeOfValueInstruction(pInst->words[1]);
350 if (!spvtools::isScalarIntegral(expected_type)) {
351 return context->diagnostic()
352 << "The selector operand for OpSwitch must be the result"
353 " of an instruction that generates an integer scalar";
354 }
355 }
356 if (auto error = context->binaryEncodeNumericLiteral(
357 textValue, error_code_for_literals, expected_type, pInst)) {
358 return error;
359 }
360 } break;
361
362 case SPV_OPERAND_TYPE_LITERAL_STRING:
363 case SPV_OPERAND_TYPE_OPTIONAL_LITERAL_STRING: {
364 spv_literal_t literal = {};
365 spv_result_t error = spvTextToLiteral(textValue, &literal);
366 if (error != SPV_SUCCESS) {
367 if (error == SPV_ERROR_OUT_OF_MEMORY) return error;
368 return context->diagnostic(error_code_for_literals)
369 << "Invalid literal string '" << textValue << "'.";
370 }
371 if (literal.type != SPV_LITERAL_TYPE_STRING) {
372 return context->diagnostic()
373 << "Expected literal string, found literal number '" << textValue
374 << "'.";
375 }
376
377 // NOTE: Special case for extended instruction library import
378 if (SpvOpExtInstImport == pInst->opcode) {
379 const spv_ext_inst_type_t ext_inst_type =
380 spvExtInstImportTypeGet(literal.str.c_str());
381 if (SPV_EXT_INST_TYPE_NONE == ext_inst_type) {
382 return context->diagnostic()
383 << "Invalid extended instruction import '" << literal.str
384 << "'";
385 }
386 if ((error = context->recordIdAsExtInstImport(pInst->words[1],
387 ext_inst_type)))
388 return error;
389 }
390
391 if (context->binaryEncodeString(literal.str.c_str(), pInst))
392 return SPV_ERROR_INVALID_TEXT;
393 } break;
394
395 // Masks.
396 case SPV_OPERAND_TYPE_FP_FAST_MATH_MODE:
397 case SPV_OPERAND_TYPE_FUNCTION_CONTROL:
398 case SPV_OPERAND_TYPE_LOOP_CONTROL:
399 case SPV_OPERAND_TYPE_IMAGE:
400 case SPV_OPERAND_TYPE_OPTIONAL_IMAGE:
401 case SPV_OPERAND_TYPE_OPTIONAL_MEMORY_ACCESS:
402 case SPV_OPERAND_TYPE_SELECTION_CONTROL:
403 case SPV_OPERAND_TYPE_DEBUG_INFO_FLAGS:
404 case SPV_OPERAND_TYPE_CLDEBUG100_DEBUG_INFO_FLAGS: {
405 uint32_t value;
406 if (grammar.parseMaskOperand(type, textValue, &value)) {
407 return context->diagnostic() << "Invalid " << spvOperandTypeStr(type)
408 << " operand '" << textValue << "'.";
409 }
410 if (auto error = context->binaryEncodeU32(value, pInst)) return error;
411 // Prepare to parse the operands for this logical operand.
412 grammar.pushOperandTypesForMask(type, value, pExpectedOperands);
413 } break;
414 case SPV_OPERAND_TYPE_OPTIONAL_CIV: {
415 auto error = spvTextEncodeOperand(
416 grammar, context, SPV_OPERAND_TYPE_OPTIONAL_LITERAL_NUMBER, textValue,
417 pInst, pExpectedOperands);
418 if (error == SPV_FAILED_MATCH) {
419 // It's not a literal number -- is it a literal string?
420 error = spvTextEncodeOperand(grammar, context,
421 SPV_OPERAND_TYPE_OPTIONAL_LITERAL_STRING,
422 textValue, pInst, pExpectedOperands);
423 }
424 if (error == SPV_FAILED_MATCH) {
425 // It's not a literal -- is it an ID?
426 error =
427 spvTextEncodeOperand(grammar, context, SPV_OPERAND_TYPE_OPTIONAL_ID,
428 textValue, pInst, pExpectedOperands);
429 }
430 if (error) {
431 return context->diagnostic(error)
432 << "Invalid word following !<integer>: " << textValue;
433 }
434 if (pExpectedOperands->empty()) {
435 pExpectedOperands->push_back(SPV_OPERAND_TYPE_OPTIONAL_CIV);
436 }
437 } break;
438 default: {
439 // NOTE: All non literal operands are handled here using the operand
440 // table.
441 spv_operand_desc entry;
442 if (grammar.lookupOperand(type, textValue, strlen(textValue), &entry)) {
443 return context->diagnostic() << "Invalid " << spvOperandTypeStr(type)
444 << " '" << textValue << "'.";
445 }
446 if (context->binaryEncodeU32(entry->value, pInst)) {
447 return context->diagnostic() << "Invalid " << spvOperandTypeStr(type)
448 << " '" << textValue << "'.";
449 }
450
451 // Prepare to parse the operands for this logical operand.
452 spvPushOperandTypes(entry->operandTypes, pExpectedOperands);
453 } break;
454 }
455 return SPV_SUCCESS;
456}
457
458namespace {
459
460/// Encodes an instruction started by !<integer> at the given position in text.
461///
462/// Puts the encoded words into *pInst. If successful, moves position past the
463/// instruction and returns SPV_SUCCESS. Otherwise, returns an error code and
464/// leaves position pointing to the error in text.
465spv_result_t encodeInstructionStartingWithImmediate(
466 const spvtools::AssemblyGrammar& grammar,
467 spvtools::AssemblyContext* context, spv_instruction_t* pInst) {
468 std::string firstWord;
469 spv_position_t nextPosition = {};
470 auto error = context->getWord(&firstWord, &nextPosition);
471 if (error) return context->diagnostic(error) << "Internal Error";
472
473 if ((error = encodeImmediate(context, firstWord.c_str(), pInst))) {
474 return error;
475 }
476 while (context->advance() != SPV_END_OF_STREAM) {
477 // A beginning of a new instruction means we're done.
478 if (context->isStartOfNewInst()) return SPV_SUCCESS;
479
480 // Otherwise, there must be an operand that's either a literal, an ID, or
481 // an immediate.
482 std::string operandValue;
483 if ((error = context->getWord(&operandValue, &nextPosition)))
484 return context->diagnostic(error) << "Internal Error";
485
486 if (operandValue == "=")
487 return context->diagnostic() << firstWord << " not allowed before =.";
488
489 // Needed to pass to spvTextEncodeOpcode(), but it shouldn't ever be
490 // expanded.
491 spv_operand_pattern_t dummyExpectedOperands;
492 error = spvTextEncodeOperand(
493 grammar, context, SPV_OPERAND_TYPE_OPTIONAL_CIV, operandValue.c_str(),
494 pInst, &dummyExpectedOperands);
495 if (error) return error;
496 context->setPosition(nextPosition);
497 }
498 return SPV_SUCCESS;
499}
500
501/// @brief Translate single Opcode and operands to binary form
502///
503/// @param[in] grammar the grammar to use for compilation
504/// @param[in, out] context the dynamic compilation info
505/// @param[in] text stream to translate
506/// @param[out] pInst returned binary Opcode
507/// @param[in,out] pPosition in the text stream
508///
509/// @return result code
510spv_result_t spvTextEncodeOpcode(const spvtools::AssemblyGrammar& grammar,
511 spvtools::AssemblyContext* context,
512 spv_instruction_t* pInst) {
513 // Check for !<integer> first.
514 if ('!' == context->peek()) {
515 return encodeInstructionStartingWithImmediate(grammar, context, pInst);
516 }
517
518 std::string firstWord;
519 spv_position_t nextPosition = {};
520 spv_result_t error = context->getWord(&firstWord, &nextPosition);
521 if (error) return context->diagnostic() << "Internal Error";
522
523 std::string opcodeName;
524 std::string result_id;
525 spv_position_t result_id_position = {};
526 if (context->startsWithOp()) {
527 opcodeName = firstWord;
528 } else {
529 result_id = firstWord;
530 if ('%' != result_id.front()) {
531 return context->diagnostic()
532 << "Expected <opcode> or <result-id> at the beginning "
533 "of an instruction, found '"
534 << result_id << "'.";
535 }
536 result_id_position = context->position();
537
538 // The '=' sign.
539 context->setPosition(nextPosition);
540 if (context->advance())
541 return context->diagnostic() << "Expected '=', found end of stream.";
542 std::string equal_sign;
543 error = context->getWord(&equal_sign, &nextPosition);
544 if ("=" != equal_sign)
545 return context->diagnostic() << "'=' expected after result id.";
546
547 // The <opcode> after the '=' sign.
548 context->setPosition(nextPosition);
549 if (context->advance())
550 return context->diagnostic() << "Expected opcode, found end of stream.";
551 error = context->getWord(&opcodeName, &nextPosition);
552 if (error) return context->diagnostic(error) << "Internal Error";
553 if (!context->startsWithOp()) {
554 return context->diagnostic()
555 << "Invalid Opcode prefix '" << opcodeName << "'.";
556 }
557 }
558
559 // NOTE: The table contains Opcode names without the "Op" prefix.
560 const char* pInstName = opcodeName.data() + 2;
561
562 spv_opcode_desc opcodeEntry;
563 error = grammar.lookupOpcode(pInstName, &opcodeEntry);
564 if (error) {
565 return context->diagnostic(error)
566 << "Invalid Opcode name '" << opcodeName << "'";
567 }
568 if (opcodeEntry->hasResult && result_id.empty()) {
569 return context->diagnostic()
570 << "Expected <result-id> at the beginning of an instruction, found '"
571 << firstWord << "'.";
572 }
573 if (!opcodeEntry->hasResult && !result_id.empty()) {
574 return context->diagnostic()
575 << "Cannot set ID " << result_id << " because " << opcodeName
576 << " does not produce a result ID.";
577 }
578 pInst->opcode = opcodeEntry->opcode;
579 context->setPosition(nextPosition);
580 // Reserve the first word for the instruction.
581 spvInstructionAddWord(pInst, 0);
582
583 // Maintains the ordered list of expected operand types.
584 // For many instructions we only need the {numTypes, operandTypes}
585 // entries in opcodeEntry. However, sometimes we need to modify
586 // the list as we parse the operands. This occurs when an operand
587 // has its own logical operands (such as the LocalSize operand for
588 // ExecutionMode), or for extended instructions that may have their
589 // own operands depending on the selected extended instruction.
590 spv_operand_pattern_t expectedOperands;
591 expectedOperands.reserve(opcodeEntry->numTypes);
592 for (auto i = 0; i < opcodeEntry->numTypes; i++)
593 expectedOperands.push_back(
594 opcodeEntry->operandTypes[opcodeEntry->numTypes - i - 1]);
595
596 while (!expectedOperands.empty()) {
597 const spv_operand_type_t type = expectedOperands.back();
598 expectedOperands.pop_back();
599
600 // Expand optional tuples lazily.
601 if (spvExpandOperandSequenceOnce(type, &expectedOperands)) continue;
602
603 if (type == SPV_OPERAND_TYPE_RESULT_ID && !result_id.empty()) {
604 // Handle the <result-id> for value generating instructions.
605 // We've already consumed it from the text stream. Here
606 // we inject its words into the instruction.
607 spv_position_t temp_pos = context->position();
608 error = spvTextEncodeOperand(grammar, context, SPV_OPERAND_TYPE_RESULT_ID,
609 result_id.c_str(), pInst, nullptr);
610 result_id_position = context->position();
611 // Because we are injecting we have to reset the position afterwards.
612 context->setPosition(temp_pos);
613 if (error) return error;
614 } else {
615 // Find the next word.
616 error = context->advance();
617 if (error == SPV_END_OF_STREAM) {
618 if (spvOperandIsOptional(type)) {
619 // This would have been the last potential operand for the
620 // instruction,
621 // and we didn't find one. We're finished parsing this instruction.
622 break;
623 } else {
624 return context->diagnostic()
625 << "Expected operand, found end of stream.";
626 }
627 }
628 assert(error == SPV_SUCCESS && "Somebody added another way to fail");
629
630 if (context->isStartOfNewInst()) {
631 if (spvOperandIsOptional(type)) {
632 break;
633 } else {
634 return context->diagnostic()
635 << "Expected operand, found next instruction instead.";
636 }
637 }
638
639 std::string operandValue;
640 error = context->getWord(&operandValue, &nextPosition);
641 if (error) return context->diagnostic(error) << "Internal Error";
642
643 error = spvTextEncodeOperand(grammar, context, type, operandValue.c_str(),
644 pInst, &expectedOperands);
645
646 if (error == SPV_FAILED_MATCH && spvOperandIsOptional(type))
647 return SPV_SUCCESS;
648
649 if (error) return error;
650
651 context->setPosition(nextPosition);
652 }
653 }
654
655 if (spvOpcodeGeneratesType(pInst->opcode)) {
656 if (context->recordTypeDefinition(pInst) != SPV_SUCCESS) {
657 return SPV_ERROR_INVALID_TEXT;
658 }
659 } else if (opcodeEntry->hasType) {
660 // SPIR-V dictates that if an instruction has both a return value and a
661 // type ID then the type id is first, and the return value is second.
662 assert(opcodeEntry->hasResult &&
663 "Unknown opcode: has a type but no result.");
664 context->recordTypeIdForValue(pInst->words[2], pInst->words[1]);
665 }
666
667 if (pInst->words.size() > SPV_LIMIT_INSTRUCTION_WORD_COUNT_MAX) {
668 return context->diagnostic()
669 << "Instruction too long: " << pInst->words.size()
670 << " words, but the limit is "
671 << SPV_LIMIT_INSTRUCTION_WORD_COUNT_MAX;
672 }
673
674 pInst->words[0] =
675 spvOpcodeMake(uint16_t(pInst->words.size()), opcodeEntry->opcode);
676
677 return SPV_SUCCESS;
678}
679
680enum { kAssemblerVersion = 0 };
681
682// Populates a binary stream's |header|. The target environment is specified via
683// |env| and Id bound is via |bound|.
684spv_result_t SetHeader(spv_target_env env, const uint32_t bound,
685 uint32_t* header) {
686 if (!header) return SPV_ERROR_INVALID_BINARY;
687
688 header[SPV_INDEX_MAGIC_NUMBER] = SpvMagicNumber;
689 header[SPV_INDEX_VERSION_NUMBER] = spvVersionForTargetEnv(env);
690 header[SPV_INDEX_GENERATOR_NUMBER] =
691 SPV_GENERATOR_WORD(SPV_GENERATOR_KHRONOS_ASSEMBLER, kAssemblerVersion);
692 header[SPV_INDEX_BOUND] = bound;
693 header[SPV_INDEX_SCHEMA] = 0; // NOTE: Reserved
694
695 return SPV_SUCCESS;
696}
697
698// Collects all numeric ids in the module source into |numeric_ids|.
699// This function is essentially a dry-run of spvTextToBinary.
700spv_result_t GetNumericIds(const spvtools::AssemblyGrammar& grammar,
701 const spvtools::MessageConsumer& consumer,
702 const spv_text text,
703 std::set<uint32_t>* numeric_ids) {
704 spvtools::AssemblyContext context(text, consumer);
705
706 if (!text->str) return context.diagnostic() << "Missing assembly text.";
707
708 if (!grammar.isValid()) {
709 return SPV_ERROR_INVALID_TABLE;
710 }
711
712 // Skip past whitespace and comments.
713 context.advance();
714
715 while (context.hasText()) {
716 spv_instruction_t inst;
717
718 if (spvTextEncodeOpcode(grammar, &context, &inst)) {
719 return SPV_ERROR_INVALID_TEXT;
720 }
721
722 if (context.advance()) break;
723 }
724
725 *numeric_ids = context.GetNumericIds();
726 return SPV_SUCCESS;
727}
728
729// Translates a given assembly language module into binary form.
730// If a diagnostic is generated, it is not yet marked as being
731// for a text-based input.
732spv_result_t spvTextToBinaryInternal(const spvtools::AssemblyGrammar& grammar,
733 const spvtools::MessageConsumer& consumer,
734 const spv_text text,
735 const uint32_t options,
736 spv_binary* pBinary) {
737 // The ids in this set will have the same values both in source and binary.
738 // All other ids will be generated by filling in the gaps.
739 std::set<uint32_t> ids_to_preserve;
740
741 if (options & SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS) {
742 // Collect all numeric ids from the source into ids_to_preserve.
743 const spv_result_t result =
744 GetNumericIds(grammar, consumer, text, &ids_to_preserve);
745 if (result != SPV_SUCCESS) return result;
746 }
747
748 spvtools::AssemblyContext context(text, consumer, std::move(ids_to_preserve));
749
750 if (!text->str) return context.diagnostic() << "Missing assembly text.";
751
752 if (!grammar.isValid()) {
753 return SPV_ERROR_INVALID_TABLE;
754 }
755 if (!pBinary) return SPV_ERROR_INVALID_POINTER;
756
757 std::vector<spv_instruction_t> instructions;
758
759 // Skip past whitespace and comments.
760 context.advance();
761
762 while (context.hasText()) {
763 instructions.push_back({});
764 spv_instruction_t& inst = instructions.back();
765
766 if (spvTextEncodeOpcode(grammar, &context, &inst)) {
767 return SPV_ERROR_INVALID_TEXT;
768 }
769
770 if (context.advance()) break;
771 }
772
773 size_t totalSize = SPV_INDEX_INSTRUCTION;
774 for (auto& inst : instructions) {
775 totalSize += inst.words.size();
776 }
777
778 uint32_t* data = new uint32_t[totalSize];
779 if (!data) return SPV_ERROR_OUT_OF_MEMORY;
780 uint64_t currentIndex = SPV_INDEX_INSTRUCTION;
781 for (auto& inst : instructions) {
782 memcpy(data + currentIndex, inst.words.data(),
783 sizeof(uint32_t) * inst.words.size());
784 currentIndex += inst.words.size();
785 }
786
787 if (auto error = SetHeader(grammar.target_env(), context.getBound(), data))
788 return error;
789
790 spv_binary binary = new spv_binary_t();
791 if (!binary) {
792 delete[] data;
793 return SPV_ERROR_OUT_OF_MEMORY;
794 }
795 binary->code = data;
796 binary->wordCount = totalSize;
797
798 *pBinary = binary;
799
800 return SPV_SUCCESS;
801}
802
803} // anonymous namespace
804
805spv_result_t spvTextToBinary(const spv_const_context context,
806 const char* input_text,
807 const size_t input_text_size, spv_binary* pBinary,
808 spv_diagnostic* pDiagnostic) {
809 return spvTextToBinaryWithOptions(context, input_text, input_text_size,
810 SPV_TEXT_TO_BINARY_OPTION_NONE, pBinary,
811 pDiagnostic);
812}
813
814spv_result_t spvTextToBinaryWithOptions(const spv_const_context context,
815 const char* input_text,
816 const size_t input_text_size,
817 const uint32_t options,
818 spv_binary* pBinary,
819 spv_diagnostic* pDiagnostic) {
820 spv_context_t hijack_context = *context;
821 if (pDiagnostic) {
822 *pDiagnostic = nullptr;
823 spvtools::UseDiagnosticAsMessageConsumer(&hijack_context, pDiagnostic);
824 }
825
826 spv_text_t text = {input_text, input_text_size};
827 spvtools::AssemblyGrammar grammar(&hijack_context);
828
829 spv_result_t result = spvTextToBinaryInternal(
830 grammar, hijack_context.consumer, &text, options, pBinary);
831 if (pDiagnostic && *pDiagnostic) (*pDiagnostic)->isTextSource = true;
832
833 return result;
834}
835
836void spvTextDestroy(spv_text text) {
837 if (text) {
838 if (text->str) delete[] text->str;
839 delete text;
840 }
841}
842