SPIRV-Tools/source/text.cpp

842 lines
30 KiB
C++
Raw Normal View History

2016-01-07 18:44:22 +00:00
// Copyright (c) 2015-2016 The Khronos Group Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#include "source/text.h"
2015-09-11 20:34:49 +00:00
#include <algorithm>
#include <cassert>
#include <cctype>
2015-09-11 20:34:49 +00:00
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include <memory>
#include <set>
#include <sstream>
#include <string>
#include <unordered_map>
#include <utility>
#include <vector>
2015-09-11 20:34:49 +00:00
#include "source/assembly_grammar.h"
#include "source/binary.h"
#include "source/diagnostic.h"
#include "source/ext_inst.h"
#include "source/instruction.h"
#include "source/opcode.h"
#include "source/operand.h"
#include "source/spirv_constant.h"
#include "source/spirv_target_env.h"
#include "source/table.h"
#include "source/text_handler.h"
#include "source/util/bitutils.h"
#include "source/util/parse_number.h"
#include "spirv-tools/libspirv.h"
bool spvIsValidIDCharacter(const char value) {
return value == '_' || 0 != ::isalnum(value);
}
// Returns true if the given string represents a valid ID name.
2015-09-29 21:07:21 +00:00
bool spvIsValidID(const char* textValue) {
const char* c = textValue;
for (; *c != '\0'; ++c) {
if (!spvIsValidIDCharacter(*c)) {
return false;
}
}
// If the string was empty, then the ID also is not valid.
return c != textValue;
}
// Text API
spv_result_t spvTextToLiteral(const char* textValue, spv_literal_t* pLiteral) {
bool isSigned = false;
int numPeriods = 0;
bool isString = false;
const size_t len = strlen(textValue);
2015-08-24 20:27:02 +00:00
if (len == 0) return SPV_FAILED_MATCH;
for (uint64_t index = 0; index < len; ++index) {
switch (textValue[index]) {
case '0':
case '1':
case '2':
case '3':
case '4':
case '5':
case '6':
case '7':
case '8':
case '9':
break;
case '.':
numPeriods++;
break;
case '-':
if (index == 0) {
isSigned = true;
} else {
isString = true;
}
break;
default:
isString = true;
2015-09-10 21:07:21 +00:00
index = len; // break out of the loop too.
break;
}
}
pLiteral->type = spv_literal_type_t(99);
2015-09-10 21:07:21 +00:00
if (isString || numPeriods > 1 || (isSigned && len == 1)) {
2015-08-24 20:27:02 +00:00
if (len < 2 || textValue[0] != '"' || textValue[len - 1] != '"')
return SPV_FAILED_MATCH;
bool escaping = false;
for (const char* val = textValue + 1; val != textValue + len - 1; ++val) {
if ((*val == '\\') && (!escaping)) {
escaping = true;
} else {
// Have to save space for the null-terminator
if (pLiteral->str.size() >= SPV_LIMIT_LITERAL_STRING_BYTES_MAX)
return SPV_ERROR_OUT_OF_MEMORY;
pLiteral->str.push_back(*val);
escaping = false;
}
}
pLiteral->type = SPV_LITERAL_TYPE_STRING;
} else if (numPeriods == 1) {
double d = std::strtod(textValue, nullptr);
float f = (float)d;
if (d == (double)f) {
pLiteral->type = SPV_LITERAL_TYPE_FLOAT_32;
pLiteral->value.f = f;
} else {
pLiteral->type = SPV_LITERAL_TYPE_FLOAT_64;
pLiteral->value.d = d;
}
} else if (isSigned) {
int64_t i64 = strtoll(textValue, nullptr, 10);
int32_t i32 = (int32_t)i64;
if (i64 == (int64_t)i32) {
pLiteral->type = SPV_LITERAL_TYPE_INT_32;
pLiteral->value.i32 = i32;
} else {
pLiteral->type = SPV_LITERAL_TYPE_INT_64;
pLiteral->value.i64 = i64;
}
} else {
uint64_t u64 = strtoull(textValue, nullptr, 10);
uint32_t u32 = (uint32_t)u64;
if (u64 == (uint64_t)u32) {
pLiteral->type = SPV_LITERAL_TYPE_UINT_32;
pLiteral->value.u32 = u32;
} else {
pLiteral->type = SPV_LITERAL_TYPE_UINT_64;
pLiteral->value.u64 = u64;
}
}
return SPV_SUCCESS;
}
2015-09-29 21:07:21 +00:00
namespace {
/// Parses an immediate integer from text, guarding against overflow. If
/// successful, adds the parsed value to pInst, advances the context past it,
/// and returns SPV_SUCCESS. Otherwise, leaves pInst alone, emits diagnostics,
/// and returns SPV_ERROR_INVALID_TEXT.
spv_result_t encodeImmediate(spvtools::AssemblyContext* context,
2015-09-29 21:07:21 +00:00
const char* text, spv_instruction_t* pInst) {
assert(*text == '!');
uint32_t parse_result;
if (!spvtools::utils::ParseNumber(text + 1, &parse_result)) {
Pull out the number parsing logic Pull out the number parsing logic from AssemblyContext::binaryEncodeNumericLiteral() to utilities. The new utility function: `ParseAndEncodeNumber()` now accepts: * number text to parse * number type * a emit function, which is a function which will be called with each parsed uint32 word. * a pointer to std::string to be overwritten with error messages. (pass nullptr if expect no error message) and returns: * an enum result type to indicate the status Type/Structs moved to utility: * template<typename T> class ClampToZeroIfUnsignedType New type: * enum EncodeNumberStatus: success or error code * NumberType: hold the number type information for the number to be parsed. * several helper functions are also added for NumberType. Functions moved to utility: * Helpers: * template<typename T> checkRangeAndIfHexThenSignExtend() -> CheckRangeAndIfHex....() * Interfaces: * template<typename T> parseNumber() -> ParseNumber() * binaryEncodeIntegerLiteral() -> ParseAndEncodeIntegerNumber() * binaryEncodeFloatingPointLiteral() -> ParseAndEncodeFloatingPointNumber() * binaryEncodeNumericLiteral() -> ParseAndEncodeNumber() Tests added/moved to test/ParseNumber.cpp, including tests for: * ParseNumber(): This is moved from TextToBinary.cpp to ParseNumber.cpp * ParseAndEncodeIntegerNumber(): New added * ParseAndEncodeFloatingPointNumber(): New added * ParseAndEncodeNumber(): New added Note that the error messages are kept almost the same as before, but they may be inappropriate for an utility function. Those will be fixed in another CL.
2016-09-01 18:27:04 +00:00
return context->diagnostic(SPV_ERROR_INVALID_TEXT)
<< "Invalid immediate integer: !" << text + 1;
}
context->binaryEncodeU32(parse_result, pInst);
context->seekForward(static_cast<uint32_t>(strlen(text)));
2015-09-29 21:07:21 +00:00
return SPV_SUCCESS;
}
} // anonymous namespace
/// @brief Translate an Opcode operand to binary form
///
/// @param[in] grammar the grammar to use for compilation
/// @param[in, out] context the dynamic compilation info
/// @param[in] type of the operand
/// @param[in] textValue word of text to be parsed
/// @param[out] pInst return binary Opcode
/// @param[in,out] pExpectedOperands the operand types expected
///
/// @return result code
spv_result_t spvTextEncodeOperand(const spvtools::AssemblyGrammar& grammar,
spvtools::AssemblyContext* context,
const spv_operand_type_t type,
const char* textValue,
spv_instruction_t* pInst,
spv_operand_pattern_t* pExpectedOperands) {
// NOTE: Handle immediate int in the stream
if ('!' == textValue[0]) {
2015-09-29 21:07:21 +00:00
if (auto error = encodeImmediate(context, textValue, pInst)) {
return error;
}
*pExpectedOperands =
spvAlternatePatternFollowingImmediate(*pExpectedOperands);
return SPV_SUCCESS;
}
// Optional literal operands can fail to parse. In that case use
// SPV_FAILED_MATCH to avoid emitting a diagostic. Use the following
// for those situations.
spv_result_t error_code_for_literals =
spvOperandIsOptional(type) ? SPV_FAILED_MATCH : SPV_ERROR_INVALID_TEXT;
switch (type) {
case SPV_OPERAND_TYPE_ID:
case SPV_OPERAND_TYPE_TYPE_ID:
case SPV_OPERAND_TYPE_RESULT_ID:
case SPV_OPERAND_TYPE_MEMORY_SEMANTICS_ID:
case SPV_OPERAND_TYPE_SCOPE_ID:
case SPV_OPERAND_TYPE_OPTIONAL_ID: {
if ('%' == textValue[0]) {
textValue++;
} else {
return context->diagnostic() << "Expected id to start with %.";
}
if (!spvIsValidID(textValue)) {
return context->diagnostic() << "Invalid ID " << textValue;
}
const uint32_t id = context->spvNamedIdAssignOrGet(textValue);
if (type == SPV_OPERAND_TYPE_TYPE_ID) pInst->resultTypeId = id;
spvInstructionAddWord(pInst, id);
// Set the extended instruction type.
// The import set id is the 3rd operand of OpExtInst.
if (pInst->opcode == SpvOpExtInst && pInst->words.size() == 4) {
auto ext_inst_type = context->getExtInstTypeForId(pInst->words[3]);
if (ext_inst_type == SPV_EXT_INST_TYPE_NONE) {
return context->diagnostic()
<< "Invalid extended instruction import Id "
<< pInst->words[2];
}
pInst->extInstType = ext_inst_type;
}
} break;
case SPV_OPERAND_TYPE_EXTENSION_INSTRUCTION_NUMBER: {
// The assembler accepts the symbolic name for an extended instruction,
// and emits its corresponding number.
spv_ext_inst_desc extInst;
2019-12-18 23:10:29 +00:00
if (grammar.lookupExtInst(pInst->extInstType, textValue, &extInst) ==
SPV_SUCCESS) {
// if we know about this extended instruction, push the numeric value
spvInstructionAddWord(pInst, extInst->ext_inst);
// Prepare to parse the operands for the extended instructions.
spvPushOperandTypes(extInst->operandTypes, pExpectedOperands);
} else {
// if we don't know this extended instruction and the set isn't
// non-semantic, we cannot process further
if (!spvExtInstIsNonSemantic(pInst->extInstType)) {
return context->diagnostic()
<< "Invalid extended instruction name '" << textValue << "'.";
} else {
// for non-semantic instruction sets, as long as the text name is an
// integer value we can encode it since we know the form of all such
// extended instructions
spv_literal_t extInstValue;
if (spvTextToLiteral(textValue, &extInstValue) ||
extInstValue.type != SPV_LITERAL_TYPE_UINT_32) {
return context->diagnostic()
<< "Couldn't translate unknown extended instruction name '"
<< textValue << "' to unsigned integer.";
}
spvInstructionAddWord(pInst, extInstValue.value.u32);
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
2019-12-18 23:10:29 +00:00
// opcode contains an unknown number of IDs.
pExpectedOperands->push_back(SPV_OPERAND_TYPE_VARIABLE_ID);
}
}
} break;
case SPV_OPERAND_TYPE_SPEC_CONSTANT_OP_NUMBER: {
// The assembler accepts the symbolic name for the opcode, but without
// the "Op" prefix. For example, "IAdd" is accepted. The number
// of the opcode is emitted.
SpvOp opcode;
if (grammar.lookupSpecConstantOpcode(textValue, &opcode)) {
return context->diagnostic() << "Invalid " << spvOperandTypeStr(type)
<< " '" << textValue << "'.";
}
spv_opcode_desc opcodeEntry = nullptr;
if (grammar.lookupOpcode(opcode, &opcodeEntry)) {
return context->diagnostic(SPV_ERROR_INTERNAL)
<< "OpSpecConstant opcode table out of sync";
}
spvInstructionAddWord(pInst, uint32_t(opcodeEntry->opcode));
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
// Prepare to parse the operands for the opcode. Except skip the
// type Id and result Id, since they've already been processed.
assert(opcodeEntry->hasType);
assert(opcodeEntry->hasResult);
assert(opcodeEntry->numTypes >= 2);
spvPushOperandTypes(opcodeEntry->operandTypes + 2, pExpectedOperands);
} break;
case SPV_OPERAND_TYPE_LITERAL_INTEGER:
case SPV_OPERAND_TYPE_OPTIONAL_LITERAL_INTEGER: {
// The current operand is an *unsigned* 32-bit integer.
// That's just how the grammar works.
spvtools::IdType expected_type = {
32, false, spvtools::IdTypeClass::kScalarIntegerType};
if (auto error = context->binaryEncodeNumericLiteral(
textValue, error_code_for_literals, expected_type, pInst)) {
return error;
}
} break;
case SPV_OPERAND_TYPE_OPTIONAL_LITERAL_NUMBER:
// This is a context-independent literal number which can be a 32-bit
// number of floating point value.
if (auto error = context->binaryEncodeNumericLiteral(
textValue, error_code_for_literals, spvtools::kUnknownType,
pInst)) {
return error;
}
break;
case SPV_OPERAND_TYPE_OPTIONAL_TYPED_LITERAL_INTEGER:
case SPV_OPERAND_TYPE_TYPED_LITERAL_NUMBER: {
spvtools::IdType expected_type = spvtools::kUnknownType;
// The encoding for OpConstant, OpSpecConstant and OpSwitch all
// depend on either their own result-id or the result-id of
// one of their parameters.
if (SpvOpConstant == pInst->opcode ||
SpvOpSpecConstant == pInst->opcode) {
// The type of the literal is determined by the type Id of the
// instruction.
expected_type =
context->getTypeOfTypeGeneratingValue(pInst->resultTypeId);
if (!spvtools::isScalarFloating(expected_type) &&
!spvtools::isScalarIntegral(expected_type)) {
spv_opcode_desc d;
const char* opcode_name = "opcode";
if (SPV_SUCCESS == grammar.lookupOpcode(pInst->opcode, &d)) {
opcode_name = d->name;
}
return context->diagnostic()
<< "Type for " << opcode_name
<< " must be a scalar floating point or integer type";
}
} else if (pInst->opcode == SpvOpSwitch) {
// The type of the literal is the same as the type of the selector.
expected_type = context->getTypeOfValueInstruction(pInst->words[1]);
if (!spvtools::isScalarIntegral(expected_type)) {
return context->diagnostic()
<< "The selector operand for OpSwitch must be the result"
" of an instruction that generates an integer scalar";
}
}
if (auto error = context->binaryEncodeNumericLiteral(
textValue, error_code_for_literals, expected_type, pInst)) {
return error;
}
} break;
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
case SPV_OPERAND_TYPE_LITERAL_STRING:
case SPV_OPERAND_TYPE_OPTIONAL_LITERAL_STRING: {
spv_literal_t literal = {};
spv_result_t error = spvTextToLiteral(textValue, &literal);
if (error != SPV_SUCCESS) {
if (error == SPV_ERROR_OUT_OF_MEMORY) return error;
return context->diagnostic(error_code_for_literals)
<< "Invalid literal string '" << textValue << "'.";
}
if (literal.type != SPV_LITERAL_TYPE_STRING) {
return context->diagnostic()
<< "Expected literal string, found literal number '" << textValue
<< "'.";
}
// NOTE: Special case for extended instruction library import
if (SpvOpExtInstImport == pInst->opcode) {
const spv_ext_inst_type_t ext_inst_type =
spvExtInstImportTypeGet(literal.str.c_str());
if (SPV_EXT_INST_TYPE_NONE == ext_inst_type) {
return context->diagnostic()
<< "Invalid extended instruction import '" << literal.str
<< "'";
}
2016-01-05 19:56:02 +00:00
if ((error = context->recordIdAsExtInstImport(pInst->words[1],
ext_inst_type)))
return error;
}
if (context->binaryEncodeString(literal.str.c_str(), pInst))
return SPV_ERROR_INVALID_TEXT;
} break;
// Masks.
case SPV_OPERAND_TYPE_FP_FAST_MATH_MODE:
case SPV_OPERAND_TYPE_FUNCTION_CONTROL:
case SPV_OPERAND_TYPE_LOOP_CONTROL:
case SPV_OPERAND_TYPE_IMAGE:
case SPV_OPERAND_TYPE_OPTIONAL_IMAGE:
case SPV_OPERAND_TYPE_OPTIONAL_MEMORY_ACCESS:
case SPV_OPERAND_TYPE_SELECTION_CONTROL:
case SPV_OPERAND_TYPE_DEBUG_INFO_FLAGS:
case SPV_OPERAND_TYPE_CLDEBUG100_DEBUG_INFO_FLAGS: {
uint32_t value;
if (grammar.parseMaskOperand(type, textValue, &value)) {
return context->diagnostic() << "Invalid " << spvOperandTypeStr(type)
<< " operand '" << textValue << "'.";
}
if (auto error = context->binaryEncodeU32(value, pInst)) return error;
// Prepare to parse the operands for this logical operand.
grammar.pushOperandTypesForMask(type, value, pExpectedOperands);
} break;
case SPV_OPERAND_TYPE_OPTIONAL_CIV: {
auto error = spvTextEncodeOperand(
grammar, context, SPV_OPERAND_TYPE_OPTIONAL_LITERAL_NUMBER, textValue,
pInst, pExpectedOperands);
if (error == SPV_FAILED_MATCH) {
// It's not a literal number -- is it a literal string?
error = spvTextEncodeOperand(grammar, context,
SPV_OPERAND_TYPE_OPTIONAL_LITERAL_STRING,
textValue, pInst, pExpectedOperands);
}
if (error == SPV_FAILED_MATCH) {
// It's not a literal -- is it an ID?
error =
spvTextEncodeOperand(grammar, context, SPV_OPERAND_TYPE_OPTIONAL_ID,
textValue, pInst, pExpectedOperands);
}
if (error) {
return context->diagnostic(error)
<< "Invalid word following !<integer>: " << textValue;
}
if (pExpectedOperands->empty()) {
pExpectedOperands->push_back(SPV_OPERAND_TYPE_OPTIONAL_CIV);
}
} break;
default: {
// NOTE: All non literal operands are handled here using the operand
// table.
spv_operand_desc entry;
2015-09-29 21:07:21 +00:00
if (grammar.lookupOperand(type, textValue, strlen(textValue), &entry)) {
return context->diagnostic() << "Invalid " << spvOperandTypeStr(type)
<< " '" << textValue << "'.";
}
if (context->binaryEncodeU32(entry->value, pInst)) {
return context->diagnostic() << "Invalid " << spvOperandTypeStr(type)
<< " '" << textValue << "'.";
}
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
// Prepare to parse the operands for this logical operand.
spvPushOperandTypes(entry->operandTypes, pExpectedOperands);
} break;
}
return SPV_SUCCESS;
}
namespace {
/// Encodes an instruction started by !<integer> at the given position in text.
///
/// Puts the encoded words into *pInst. If successful, moves position past the
/// instruction and returns SPV_SUCCESS. Otherwise, returns an error code and
/// leaves position pointing to the error in text.
spv_result_t encodeInstructionStartingWithImmediate(
const spvtools::AssemblyGrammar& grammar,
spvtools::AssemblyContext* context, spv_instruction_t* pInst) {
std::string firstWord;
spv_position_t nextPosition = {};
auto error = context->getWord(&firstWord, &nextPosition);
if (error) return context->diagnostic(error) << "Internal Error";
2015-09-29 21:07:21 +00:00
if ((error = encodeImmediate(context, firstWord.c_str(), pInst))) {
return error;
}
while (context->advance() != SPV_END_OF_STREAM) {
// A beginning of a new instruction means we're done.
if (context->isStartOfNewInst()) return SPV_SUCCESS;
// Otherwise, there must be an operand that's either a literal, an ID, or
// an immediate.
std::string operandValue;
if ((error = context->getWord(&operandValue, &nextPosition)))
return context->diagnostic(error) << "Internal Error";
if (operandValue == "=")
return context->diagnostic() << firstWord << " not allowed before =.";
// Needed to pass to spvTextEncodeOpcode(), but it shouldn't ever be
// expanded.
spv_operand_pattern_t dummyExpectedOperands;
error = spvTextEncodeOperand(
grammar, context, SPV_OPERAND_TYPE_OPTIONAL_CIV, operandValue.c_str(),
pInst, &dummyExpectedOperands);
if (error) return error;
context->setPosition(nextPosition);
}
return SPV_SUCCESS;
}
/// @brief Translate single Opcode and operands to binary form
///
/// @param[in] grammar the grammar to use for compilation
/// @param[in, out] context the dynamic compilation info
/// @param[in] text stream to translate
/// @param[out] pInst returned binary Opcode
/// @param[in,out] pPosition in the text stream
///
/// @return result code
spv_result_t spvTextEncodeOpcode(const spvtools::AssemblyGrammar& grammar,
spvtools::AssemblyContext* context,
spv_instruction_t* pInst) {
// Check for !<integer> first.
if ('!' == context->peek()) {
return encodeInstructionStartingWithImmediate(grammar, context, pInst);
}
std::string firstWord;
spv_position_t nextPosition = {};
spv_result_t error = context->getWord(&firstWord, &nextPosition);
if (error) return context->diagnostic() << "Internal Error";
std::string opcodeName;
std::string result_id;
spv_position_t result_id_position = {};
if (context->startsWithOp()) {
opcodeName = firstWord;
} else {
result_id = firstWord;
if ('%' != result_id.front()) {
return context->diagnostic()
<< "Expected <opcode> or <result-id> at the beginning "
"of an instruction, found '"
<< result_id << "'.";
}
result_id_position = context->position();
// The '=' sign.
context->setPosition(nextPosition);
if (context->advance())
return context->diagnostic() << "Expected '=', found end of stream.";
std::string equal_sign;
error = context->getWord(&equal_sign, &nextPosition);
if ("=" != equal_sign)
return context->diagnostic() << "'=' expected after result id.";
// The <opcode> after the '=' sign.
context->setPosition(nextPosition);
if (context->advance())
return context->diagnostic() << "Expected opcode, found end of stream.";
error = context->getWord(&opcodeName, &nextPosition);
if (error) return context->diagnostic(error) << "Internal Error";
if (!context->startsWithOp()) {
return context->diagnostic()
<< "Invalid Opcode prefix '" << opcodeName << "'.";
}
}
// NOTE: The table contains Opcode names without the "Op" prefix.
const char* pInstName = opcodeName.data() + 2;
spv_opcode_desc opcodeEntry;
error = grammar.lookupOpcode(pInstName, &opcodeEntry);
if (error) {
return context->diagnostic(error)
<< "Invalid Opcode name '" << opcodeName << "'";
}
if (opcodeEntry->hasResult && result_id.empty()) {
return context->diagnostic()
<< "Expected <result-id> at the beginning of an instruction, found '"
<< firstWord << "'.";
}
if (!opcodeEntry->hasResult && !result_id.empty()) {
return context->diagnostic()
<< "Cannot set ID " << result_id << " because " << opcodeName
<< " does not produce a result ID.";
}
pInst->opcode = opcodeEntry->opcode;
context->setPosition(nextPosition);
// Reserve the first word for the instruction.
spvInstructionAddWord(pInst, 0);
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
// Maintains the ordered list of expected operand types.
// For many instructions we only need the {numTypes, operandTypes}
// entries in opcodeEntry. However, sometimes we need to modify
// the list as we parse the operands. This occurs when an operand
// has its own logical operands (such as the LocalSize operand for
// ExecutionMode), or for extended instructions that may have their
// own operands depending on the selected extended instruction.
spv_operand_pattern_t expectedOperands;
expectedOperands.reserve(opcodeEntry->numTypes);
for (auto i = 0; i < opcodeEntry->numTypes; i++)
expectedOperands.push_back(
opcodeEntry->operandTypes[opcodeEntry->numTypes - i - 1]);
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
while (!expectedOperands.empty()) {
const spv_operand_type_t type = expectedOperands.back();
expectedOperands.pop_back();
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
// Expand optional tuples lazily.
2015-09-10 21:07:21 +00:00
if (spvExpandOperandSequenceOnce(type, &expectedOperands)) continue;
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
if (type == SPV_OPERAND_TYPE_RESULT_ID && !result_id.empty()) {
// Handle the <result-id> for value generating instructions.
// We've already consumed it from the text stream. Here
// we inject its words into the instruction.
spv_position_t temp_pos = context->position();
error = spvTextEncodeOperand(grammar, context, SPV_OPERAND_TYPE_RESULT_ID,
result_id.c_str(), pInst, nullptr);
result_id_position = context->position();
// Because we are injecting we have to reset the position afterwards.
context->setPosition(temp_pos);
if (error) return error;
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
} else {
// Find the next word.
error = context->advance();
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
if (error == SPV_END_OF_STREAM) {
if (spvOperandIsOptional(type)) {
2015-09-10 21:07:21 +00:00
// This would have been the last potential operand for the
// instruction,
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
// and we didn't find one. We're finished parsing this instruction.
break;
} else {
return context->diagnostic()
<< "Expected operand, found end of stream.";
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
}
}
assert(error == SPV_SUCCESS && "Somebody added another way to fail");
if (context->isStartOfNewInst()) {
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
if (spvOperandIsOptional(type)) {
break;
} else {
return context->diagnostic()
<< "Expected operand, found next instruction instead.";
}
}
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
std::string operandValue;
error = context->getWord(&operandValue, &nextPosition);
if (error) return context->diagnostic(error) << "Internal Error";
error = spvTextEncodeOperand(grammar, context, type, operandValue.c_str(),
pInst, &expectedOperands);
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
if (error == SPV_FAILED_MATCH && spvOperandIsOptional(type))
return SPV_SUCCESS;
if (error) return error;
Use opcode operand definitions from SPIR-V specification generator. The assembler and disassembler now use a dynamically adjusted sequence of expected operand types. (Internally, it is a deque, for readability.) Both parsers repeatedly pull an expected operand type from the left of this pattern list, and try to match the next input token against it. The expected pattern is adjusted during the parse to accommodate: - an extended instruction's expected operands, depending on the extended instruction's index. - when an operand itself has operands - to handle sequences of zero or more operands, or pairs of operands. These are expanded lazily during the parse. Adds spv::OperandClass from the SPIR-V specification generator. Modifies spv_operand_desc_t: - adds hasResult, hasType, and operandClass array to the opcode description type. - "wordCount" is replaced with "numTypes", which counts the number of entries in operandTypes. And each of those describes a *logical* operand, including the type id for the instruction, and the result id for the instruction. A logical operand could be variable-width, such as a literal string. Adds opcode.inc, an automatically-generated table of operation descriptions, with one line to describe each core instruction. Externally, we have modified the SPIR-V spec doc generator to emit this file. (We have hacked this copy to use the old semantics for OpLine.) Inside the assembler, parsing an operand may fail with new error code SPV_FAIL_MATCH. For an optional operand, this is not fatal, but should trigger backtracking at a higher level. The spvTextIsStartOfNewInst checks the case of the third letter of what might be an opcode. So now, "OpenCL" does not look like an opcode name. In assembly, the EntryPoint name field is mandatory, but can be an empty string. Adjust tests for changes to: - OpSampedImage - OpTypeSampler
2015-08-27 17:03:52 +00:00
context->setPosition(nextPosition);
}
}
if (spvOpcodeGeneratesType(pInst->opcode)) {
if (context->recordTypeDefinition(pInst) != SPV_SUCCESS) {
return SPV_ERROR_INVALID_TEXT;
}
} else if (opcodeEntry->hasType) {
// SPIR-V dictates that if an instruction has both a return value and a
// type ID then the type id is first, and the return value is second.
assert(opcodeEntry->hasResult &&
"Unknown opcode: has a type but no result.");
context->recordTypeIdForValue(pInst->words[2], pInst->words[1]);
}
if (pInst->words.size() > SPV_LIMIT_INSTRUCTION_WORD_COUNT_MAX) {
return context->diagnostic()
<< "Instruction too long: " << pInst->words.size()
<< " words, but the limit is "
<< SPV_LIMIT_INSTRUCTION_WORD_COUNT_MAX;
}
pInst->words[0] =
spvOpcodeMake(uint16_t(pInst->words.size()), opcodeEntry->opcode);
return SPV_SUCCESS;
}
enum { kAssemblerVersion = 0 };
// Populates a binary stream's |header|. The target environment is specified via
// |env| and Id bound is via |bound|.
spv_result_t SetHeader(spv_target_env env, const uint32_t bound,
uint32_t* header) {
if (!header) return SPV_ERROR_INVALID_BINARY;
header[SPV_INDEX_MAGIC_NUMBER] = SpvMagicNumber;
header[SPV_INDEX_VERSION_NUMBER] = spvVersionForTargetEnv(env);
header[SPV_INDEX_GENERATOR_NUMBER] =
SPV_GENERATOR_WORD(SPV_GENERATOR_KHRONOS_ASSEMBLER, kAssemblerVersion);
header[SPV_INDEX_BOUND] = bound;
header[SPV_INDEX_SCHEMA] = 0; // NOTE: Reserved
return SPV_SUCCESS;
}
// Collects all numeric ids in the module source into |numeric_ids|.
// This function is essentially a dry-run of spvTextToBinary.
spv_result_t GetNumericIds(const spvtools::AssemblyGrammar& grammar,
const spvtools::MessageConsumer& consumer,
const spv_text text,
std::set<uint32_t>* numeric_ids) {
spvtools::AssemblyContext context(text, consumer);
if (!text->str) return context.diagnostic() << "Missing assembly text.";
if (!grammar.isValid()) {
return SPV_ERROR_INVALID_TABLE;
}
// Skip past whitespace and comments.
context.advance();
while (context.hasText()) {
spv_instruction_t inst;
if (spvTextEncodeOpcode(grammar, &context, &inst)) {
return SPV_ERROR_INVALID_TEXT;
}
if (context.advance()) break;
}
*numeric_ids = context.GetNumericIds();
return SPV_SUCCESS;
}
// Translates a given assembly language module into binary form.
// If a diagnostic is generated, it is not yet marked as being
// for a text-based input.
spv_result_t spvTextToBinaryInternal(const spvtools::AssemblyGrammar& grammar,
const spvtools::MessageConsumer& consumer,
const spv_text text,
const uint32_t options,
spv_binary* pBinary) {
// The ids in this set will have the same values both in source and binary.
// All other ids will be generated by filling in the gaps.
std::set<uint32_t> ids_to_preserve;
if (options & SPV_TEXT_TO_BINARY_OPTION_PRESERVE_NUMERIC_IDS) {
// Collect all numeric ids from the source into ids_to_preserve.
const spv_result_t result =
GetNumericIds(grammar, consumer, text, &ids_to_preserve);
if (result != SPV_SUCCESS) return result;
}
spvtools::AssemblyContext context(text, consumer, std::move(ids_to_preserve));
if (!text->str) return context.diagnostic() << "Missing assembly text.";
if (!grammar.isValid()) {
return SPV_ERROR_INVALID_TABLE;
}
if (!pBinary) return SPV_ERROR_INVALID_POINTER;
std::vector<spv_instruction_t> instructions;
// Skip past whitespace and comments.
context.advance();
while (context.hasText()) {
instructions.push_back({});
spv_instruction_t& inst = instructions.back();
if (spvTextEncodeOpcode(grammar, &context, &inst)) {
return SPV_ERROR_INVALID_TEXT;
}
if (context.advance()) break;
}
size_t totalSize = SPV_INDEX_INSTRUCTION;
for (auto& inst : instructions) {
totalSize += inst.words.size();
}
uint32_t* data = new uint32_t[totalSize];
if (!data) return SPV_ERROR_OUT_OF_MEMORY;
uint64_t currentIndex = SPV_INDEX_INSTRUCTION;
for (auto& inst : instructions) {
memcpy(data + currentIndex, inst.words.data(),
sizeof(uint32_t) * inst.words.size());
currentIndex += inst.words.size();
}
if (auto error = SetHeader(grammar.target_env(), context.getBound(), data))
return error;
spv_binary binary = new spv_binary_t();
if (!binary) {
delete[] data;
return SPV_ERROR_OUT_OF_MEMORY;
}
binary->code = data;
binary->wordCount = totalSize;
*pBinary = binary;
return SPV_SUCCESS;
}
2015-09-10 21:07:21 +00:00
} // anonymous namespace
spv_result_t spvTextToBinary(const spv_const_context context,
const char* input_text,
const size_t input_text_size, spv_binary* pBinary,
spv_diagnostic* pDiagnostic) {
return spvTextToBinaryWithOptions(context, input_text, input_text_size,
SPV_TEXT_TO_BINARY_OPTION_NONE, pBinary,
pDiagnostic);
}
spv_result_t spvTextToBinaryWithOptions(const spv_const_context context,
const char* input_text,
const size_t input_text_size,
const uint32_t options,
spv_binary* pBinary,
spv_diagnostic* pDiagnostic) {
spv_context_t hijack_context = *context;
if (pDiagnostic) {
*pDiagnostic = nullptr;
spvtools::UseDiagnosticAsMessageConsumer(&hijack_context, pDiagnostic);
}
spv_text_t text = {input_text, input_text_size};
spvtools::AssemblyGrammar grammar(&hijack_context);
spv_result_t result = spvTextToBinaryInternal(
grammar, hijack_context.consumer, &text, options, pBinary);
if (pDiagnostic && *pDiagnostic) (*pDiagnostic)->isTextSource = true;
return result;
}
void spvTextDestroy(spv_text text) {
if (text) {
if (text->str) delete[] text->str;
delete text;
}
}