#include "HECL/HECL.hpp" #include "HECL/Frontend.hpp" /* Combined lexer and semantic analysis system */ namespace HECL { namespace Frontend { static IR::Instruction::ArithmeticOpType ArithType(int aChar) { switch (aChar) { case '+': return IR::Instruction::ArithmeticOpAdd; case '-': return IR::Instruction::ArithmeticOpSubtract; case '*': return IR::Instruction::ArithmeticOpMultiply; case '/': return IR::Instruction::ArithmeticOpDivide; default: return IR::Instruction::ArithmeticOpNone; } } void Lexer::ReconnectArithmetic(OperationNode* sn, OperationNode** lastSub, OperationNode** newSub) const { sn->m_sub = sn->m_prev; sn->m_prev = nullptr; sn->m_sub->m_prev = nullptr; sn->m_sub->m_next = sn->m_next; sn->m_next = sn->m_next->m_next; sn->m_sub->m_next->m_prev = sn->m_sub; sn->m_sub->m_next->m_next = nullptr; if (*lastSub) { (*lastSub)->m_next = sn; sn->m_prev = *lastSub; } *lastSub = sn; if (!*newSub) *newSub = sn; } void Lexer::PrintChain(const Lexer::OperationNode* begin, const Lexer::OperationNode* end) { for (const Lexer::OperationNode* n = begin ; n != end ; n = n->m_next) { printf("%3d %s %s\n", n->m_tok.m_location.col, n->m_tok.typeString(), n->m_tok.m_tokenString.c_str()); } } void Lexer::PrintTree(const Lexer::OperationNode* node, int indent) { for (const Lexer::OperationNode* n = node ; n ; n = n->m_next) { for (int i=0 ; im_tok.m_location.col, n->m_tok.typeString(), n->m_tok.m_tokenString.c_str(), n->m_tok.m_tokenInt, n->m_tok.m_tokenFloat); if (n->m_sub) PrintTree(n->m_sub, indent + 1); } } void Lexer::reset() { m_root = nullptr; m_pool.clear(); } void Lexer::consumeAllTokens(Parser& parser) { reset(); Parser::Token firstTok = parser.consumeToken(); if (firstTok.m_type != Parser::TokenSourceBegin) { m_diag.reportLexerErr(firstTok.m_location, "expected start token"); return; } m_pool.emplace_front(std::move(firstTok)); Lexer::OperationNode* firstNode = &m_pool.front(); Lexer::OperationNode* lastNode = firstNode; /* Build linked-list of nodes parsed in-order */ { std::vector funcStack; std::vector groupStack; while (lastNode->m_tok.m_type != Parser::TokenSourceEnd) { Parser::Token tok = parser.consumeToken(); switch (tok.m_type) { case Parser::TokenEvalGroupStart: groupStack.push_back(tok.m_location); break; case Parser::TokenEvalGroupEnd: if (groupStack.empty()) { m_diag.reportLexerErr(tok.m_location, "unbalanced group detected"); return; } groupStack.pop_back(); break; case Parser::TokenFunctionStart: funcStack.push_back(tok.m_location); break; case Parser::TokenFunctionEnd: if (funcStack.empty()) { m_diag.reportLexerErr(tok.m_location, "unbalanced function detected"); return; } funcStack.pop_back(); break; case Parser::TokenSourceEnd: case Parser::TokenNumLiteral: case Parser::TokenVectorSwizzle: case Parser::TokenFunctionArgDelim: case Parser::TokenArithmeticOp: break; default: m_diag.reportLexerErr(tok.m_location, "invalid token"); return; } m_pool.emplace_front(std::move(tok)); lastNode->m_next = &m_pool.front(); m_pool.front().m_prev = lastNode; lastNode = &m_pool.front(); } /* Ensure functions and groups are balanced */ if (funcStack.size()) { m_diag.reportLexerErr(funcStack.back(), "unclosed function detected"); return; } if (groupStack.size()) { m_diag.reportLexerErr(groupStack.back(), "unclosed group detected"); return; } } /* Ensure first non-start node is a function */ if (firstNode->m_next->m_tok.m_type != Parser::TokenFunctionStart) { m_diag.reportLexerErr(firstNode->m_tok.m_location, "expected root function"); return; } /* Organize marked function args into implicit groups */ for (Lexer::OperationNode* n = firstNode ; n != lastNode ; n = n->m_next) { if (n->m_tok.m_type == Parser::TokenFunctionStart) { if (n->m_next->m_tok.m_type != Parser::TokenFunctionEnd) { if (n->m_next->m_tok.m_type == Parser::TokenFunctionArgDelim) { m_diag.reportLexerErr(n->m_next->m_tok.m_location, "empty function arg"); return; } m_pool.emplace_front(std::move( Parser::Token(Parser::TokenEvalGroupStart, n->m_next->m_tok.m_location))); Lexer::OperationNode* grp = &m_pool.front(); grp->m_next = n->m_next; grp->m_prev = n; n->m_next->m_prev = grp; n->m_next = grp; } } else if (n->m_tok.m_type == Parser::TokenFunctionEnd) { if (n->m_prev->m_tok.m_type != Parser::TokenFunctionStart) { m_pool.emplace_front(std::move( Parser::Token(Parser::TokenEvalGroupEnd, n->m_tok.m_location))); Lexer::OperationNode* grp = &m_pool.front(); grp->m_next = n; grp->m_prev = n->m_prev; n->m_prev->m_next = grp; n->m_prev = grp; } } else if (n->m_tok.m_type == Parser::TokenFunctionArgDelim) { if (n->m_next->m_tok.m_type == Parser::TokenFunctionArgDelim || n->m_next->m_tok.m_type == Parser::TokenFunctionEnd) { m_diag.reportLexerErr(n->m_next->m_tok.m_location, "empty function arg"); return; } m_pool.emplace_front(std::move( Parser::Token(Parser::TokenEvalGroupEnd, n->m_tok.m_location))); Lexer::OperationNode* egrp = &m_pool.front(); m_pool.emplace_front(std::move( Parser::Token(Parser::TokenEvalGroupStart, n->m_next->m_tok.m_location))); Lexer::OperationNode* sgrp = &m_pool.front(); egrp->m_next = sgrp; sgrp->m_prev = egrp; sgrp->m_next = n->m_next; egrp->m_prev = n->m_prev; n->m_next->m_prev = sgrp; n->m_prev->m_next = egrp; } } /* Organize marked groups into tree-hierarchy */ { std::vector groupStack; for (Lexer::OperationNode* n = firstNode ; n != lastNode ; n = n->m_next) { if (n->m_tok.m_type == Parser::TokenEvalGroupStart) groupStack.push_back(n); else if (n->m_tok.m_type == Parser::TokenEvalGroupEnd) { Lexer::OperationNode* start = groupStack.back(); groupStack.pop_back(); if (n->m_prev == start) { m_diag.reportLexerErr(start->m_tok.m_location, "empty group"); return; } start->m_sub = start->m_next; start->m_next = n->m_next; if (n->m_next) n->m_next->m_prev = start; n->m_prev->m_next = nullptr; } } } /* Organize functions into tree-hierarchy */ for (Lexer::OperationNode& n : m_pool) { if (n.m_tok.m_type == Parser::TokenFunctionStart) { for (Lexer::OperationNode* sn = n.m_next ; sn ; sn = sn->m_next) { if (sn->m_tok.m_type == Parser::TokenFunctionEnd) { n.m_sub = n.m_next; if (n.m_next == sn) n.m_sub = nullptr; n.m_next = sn->m_next; if (sn->m_next) sn->m_next->m_prev = &n; if (n.m_sub) n.m_sub->m_prev = nullptr; if (sn->m_prev) sn->m_prev->m_next = nullptr; break; } } } } /* Organize vector swizzles into tree-hierarchy */ for (Lexer::OperationNode& n : m_pool) { if (n.m_tok.m_type == Parser::TokenVectorSwizzle) { if (n.m_prev->m_tok.m_type != Parser::TokenFunctionStart) { m_diag.reportLexerErr(n.m_tok.m_location, "vector swizzles may only follow functions"); return; } Lexer::OperationNode* func = n.m_prev; n.m_sub = func; n.m_prev = func->m_prev; if (func->m_prev) { if (func->m_prev->m_sub == func) func->m_prev->m_sub = &n; else func->m_prev->m_next = &n; } func->m_next = nullptr; func->m_prev = nullptr; } } /* Ensure evaluation groups have proper arithmetic usage */ for (Lexer::OperationNode& n : m_pool) { if (n.m_tok.m_type == Parser::TokenEvalGroupStart) { int idx = 0; for (Lexer::OperationNode* sn = n.m_sub ; sn ; sn = sn->m_next, ++idx) { if ((sn->m_tok.m_type == Parser::TokenArithmeticOp && !(idx & 1)) || (sn->m_tok.m_type != Parser::TokenArithmeticOp && (idx & 1))) { m_diag.reportLexerErr(sn->m_tok.m_location, "improper arithmetic expression"); return; } } } } /* Organize arithmetic usage into tree-hierarchy */ for (Lexer::OperationNode& n : m_pool) { if (n.m_tok.m_type == Parser::TokenEvalGroupStart) { Lexer::OperationNode* newSub = nullptr; Lexer::OperationNode* lastSub = nullptr; for (Lexer::OperationNode* sn = n.m_sub ; sn ; sn = sn->m_next) { if (sn->m_tok.m_type == Parser::TokenArithmeticOp) { IR::Instruction::ArithmeticOpType op = ArithType(sn->m_tok.m_tokenInt); if (op == IR::Instruction::ArithmeticOpMultiply || op == IR::Instruction::ArithmeticOpDivide) ReconnectArithmetic(sn, &lastSub, &newSub); } } for (Lexer::OperationNode* sn = n.m_sub ; sn ; sn = sn->m_next) { if (sn->m_tok.m_type == Parser::TokenArithmeticOp) { IR::Instruction::ArithmeticOpType op = ArithType(sn->m_tok.m_tokenInt); if (op == IR::Instruction::ArithmeticOpAdd || op == IR::Instruction::ArithmeticOpSubtract) ReconnectArithmetic(sn, &lastSub, &newSub); } } if (newSub) n.m_sub = newSub; } } if (HECL::VerbosityLevel > 1) { printf("%s\n", m_diag.getSource().c_str()); PrintTree(firstNode); printf("\n"); } /* Done! */ m_root = firstNode->m_next; } void Lexer::EmitVec3(IR& ir, const Lexer::OperationNode* funcNode, IR::RegID target) const { /* Optimization case: if empty call, emit zero imm load */ const Lexer::OperationNode* gn = funcNode->m_sub; if (!gn) { ir.m_instructions.emplace_back(IR::OpLoadImm, funcNode->m_tok.m_location); ir.m_instructions.back().m_loadImm.m_immVec = {}; return; } /* Optimization case: if all numeric literals, emit vector imm load */ bool opt = true; const Parser::Token* imms[3]; for (int i=0 ; i<3 ; ++i) { if (!gn->m_sub || gn->m_sub->m_tok.m_type != Parser::TokenNumLiteral) { opt = false; break; } imms[i] = &gn->m_sub->m_tok; gn = gn->m_next; } if (opt) { ir.m_instructions.emplace_back(IR::OpLoadImm, funcNode->m_tok.m_location); atVec4f& vec = ir.m_instructions.back().m_loadImm.m_immVec; vec.vec[0] = imms[0]->m_tokenFloat; vec.vec[1] = imms[1]->m_tokenFloat; vec.vec[2] = imms[2]->m_tokenFloat; vec.vec[3] = 1.0; return; } /* Otherwise treat as normal function */ RecursiveFuncCompile(ir, funcNode, target); } void Lexer::EmitVec4(IR& ir, const Lexer::OperationNode* funcNode, IR::RegID target) const { /* Optimization case: if empty call, emit zero imm load */ const Lexer::OperationNode* gn = funcNode->m_sub; if (!gn) { ir.m_instructions.emplace_back(IR::OpLoadImm, funcNode->m_tok.m_location); ir.m_instructions.back().m_loadImm.m_immVec = {}; return; } /* Optimization case: if all numeric literals, emit vector imm load */ bool opt = true; const Parser::Token* imms[4]; for (int i=0 ; i<4 ; ++i) { if (!gn->m_sub || gn->m_sub->m_tok.m_type != Parser::TokenNumLiteral) { opt = false; break; } imms[i] = &gn->m_sub->m_tok; gn = gn->m_next; } if (opt) { ir.m_instructions.emplace_back(IR::OpLoadImm, funcNode->m_tok.m_location); atVec4f& vec = ir.m_instructions.back().m_loadImm.m_immVec; vec.vec[0] = imms[0]->m_tokenFloat; vec.vec[1] = imms[1]->m_tokenFloat; vec.vec[2] = imms[2]->m_tokenFloat; vec.vec[3] = imms[3]->m_tokenFloat; return; } /* Otherwise treat as normal function */ RecursiveFuncCompile(ir, funcNode, target); } void Lexer::EmitArithmetic(IR& ir, const Lexer::OperationNode* arithNode, IR::RegID target) const { /* Evaluate operands */ atVec4f* opt[2] = {nullptr}; size_t instCount = ir.m_instructions.size(); const Lexer::OperationNode* on = arithNode->m_sub; IR::RegID tgt = target; size_t argInsts[2]; for (int i=0 ; i<2 ; ++i, ++tgt) { const Parser::Token& tok = on->m_tok; switch (tok.m_type) { case Parser::TokenFunctionStart: if (!tok.m_tokenString.compare("vec3")) EmitVec3(ir, on, tgt); else if (!tok.m_tokenString.compare("vec4")) EmitVec4(ir, on, tgt); else RecursiveFuncCompile(ir, on, tgt); break; case Parser::TokenEvalGroupStart: RecursiveGroupCompile(ir, on, tgt); break; case Parser::TokenNumLiteral: { ir.m_instructions.emplace_back(IR::OpLoadImm, arithNode->m_tok.m_location); IR::Instruction& inst = ir.m_instructions.back(); inst.m_target = tgt; inst.m_loadImm.m_immVec.vec[0] = tok.m_tokenFloat; inst.m_loadImm.m_immVec.vec[1] = tok.m_tokenFloat; inst.m_loadImm.m_immVec.vec[2] = tok.m_tokenFloat; inst.m_loadImm.m_immVec.vec[3] = tok.m_tokenFloat; break; } case Parser::TokenVectorSwizzle: EmitVectorSwizzle(ir, on, tgt); break; default: m_diag.reportCompileErr(tok.m_location, "invalid lexer node for IR"); break; }; argInsts[i] = ir.m_instructions.size() - 1; if (ir.m_instructions.back().m_op == IR::OpLoadImm) opt[i] = &ir.m_instructions.back().m_loadImm.m_immVec; on = on->m_next; } /* Optimization case: if both operands imm load, pre-evalulate */ if (opt[0] && opt[1] && (ir.m_instructions.size() - instCount == 2)) { atVec4f eval; switch (ArithType(arithNode->m_tok.m_tokenInt)) { case IR::Instruction::ArithmeticOpAdd: eval.vec[0] = opt[0]->vec[0] + opt[1]->vec[0]; eval.vec[1] = opt[0]->vec[1] + opt[1]->vec[1]; eval.vec[2] = opt[0]->vec[2] + opt[1]->vec[2]; eval.vec[3] = opt[0]->vec[3] + opt[1]->vec[3]; break; case IR::Instruction::ArithmeticOpSubtract: eval.vec[0] = opt[0]->vec[0] - opt[1]->vec[0]; eval.vec[1] = opt[0]->vec[1] - opt[1]->vec[1]; eval.vec[2] = opt[0]->vec[2] - opt[1]->vec[2]; eval.vec[3] = opt[0]->vec[3] - opt[1]->vec[3]; break; case IR::Instruction::ArithmeticOpMultiply: eval.vec[0] = opt[0]->vec[0] * opt[1]->vec[0]; eval.vec[1] = opt[0]->vec[1] * opt[1]->vec[1]; eval.vec[2] = opt[0]->vec[2] * opt[1]->vec[2]; eval.vec[3] = opt[0]->vec[3] * opt[1]->vec[3]; break; case IR::Instruction::ArithmeticOpDivide: eval.vec[0] = opt[0]->vec[0] / opt[1]->vec[0]; eval.vec[1] = opt[0]->vec[1] / opt[1]->vec[1]; eval.vec[2] = opt[0]->vec[2] / opt[1]->vec[2]; eval.vec[3] = opt[0]->vec[3] / opt[1]->vec[3]; break; default: m_diag.reportCompileErr(arithNode->m_tok.m_location, "invalid arithmetic type"); break; } ir.m_instructions.pop_back(); ir.m_instructions.pop_back(); ir.m_instructions.emplace_back(IR::OpLoadImm, arithNode->m_tok.m_location); IR::Instruction& inst = ir.m_instructions.back(); inst.m_target = target; inst.m_loadImm.m_immVec = eval; } else { ir.m_instructions.emplace_back(IR::OpArithmetic, arithNode->m_tok.m_location); IR::Instruction& inst = ir.m_instructions.back(); inst.m_target = target; inst.m_arithmetic.m_instIdxs[0] = argInsts[0]; inst.m_arithmetic.m_instIdxs[1] = argInsts[1]; inst.m_arithmetic.m_op = ArithType(arithNode->m_tok.m_tokenInt); if (tgt > ir.m_regCount) ir.m_regCount = tgt; } } static int SwizzleCompIdx(char aChar, Diagnostics& diag, const SourceLocation& loc) { switch (aChar) { case 'x': case 'r': return 0; case 'y': case 'g': return 1; case 'z': case 'b': return 2; case 'w': case 'a': return 3; default: diag.reportCompileErr(loc, "invalid swizzle char %c", aChar); } return -1; } void Lexer::EmitVectorSwizzle(IR& ir, const Lexer::OperationNode* swizNode, IR::RegID target) const { const std::string& str = swizNode->m_tok.m_tokenString; if (str.size() != 1 && str.size() != 3 && str.size() != 4) m_diag.reportCompileErr(swizNode->m_tok.m_location, "%d component swizzles not supported", int(str.size())); size_t instCount = ir.m_instructions.size(); const Lexer::OperationNode* on = swizNode->m_sub; const Parser::Token& tok = on->m_tok; switch (tok.m_type) { case Parser::TokenFunctionStart: if (!tok.m_tokenString.compare("vec3")) EmitVec3(ir, on, target); else if (!tok.m_tokenString.compare("vec4")) EmitVec4(ir, on, target); else RecursiveFuncCompile(ir, on, target); break; case Parser::TokenEvalGroupStart: RecursiveGroupCompile(ir, on, target); break; case Parser::TokenNumLiteral: { ir.m_instructions.emplace_back(IR::OpLoadImm, swizNode->m_tok.m_location); IR::Instruction& inst = ir.m_instructions.back(); inst.m_target = target; inst.m_loadImm.m_immVec.vec[0] = tok.m_tokenFloat; inst.m_loadImm.m_immVec.vec[1] = tok.m_tokenFloat; inst.m_loadImm.m_immVec.vec[2] = tok.m_tokenFloat; inst.m_loadImm.m_immVec.vec[3] = tok.m_tokenFloat; break; } case Parser::TokenVectorSwizzle: EmitVectorSwizzle(ir, on, target); break; default: m_diag.reportCompileErr(tok.m_location, "invalid lexer node for IR"); break; }; /* Optimization case: if operand imm load, pre-evalulate */ if (ir.m_instructions.back().m_op == IR::OpLoadImm && (ir.m_instructions.size() - instCount == 1)) { atVec4f* opt = &ir.m_instructions.back().m_loadImm.m_immVec; const SourceLocation& loc = ir.m_instructions.back().m_loc; atVec4f eval; switch (str.size()) { case 1: eval = {opt->vec[SwizzleCompIdx(str[0], m_diag, loc)]}; break; case 3: eval.vec[0] = opt->vec[SwizzleCompIdx(str[0], m_diag, loc)]; eval.vec[1] = opt->vec[SwizzleCompIdx(str[1], m_diag, loc)]; eval.vec[2] = opt->vec[SwizzleCompIdx(str[2], m_diag, loc)]; eval.vec[3] = 1.0; break; case 4: eval.vec[0] = opt->vec[SwizzleCompIdx(str[0], m_diag, loc)]; eval.vec[1] = opt->vec[SwizzleCompIdx(str[1], m_diag, loc)]; eval.vec[2] = opt->vec[SwizzleCompIdx(str[2], m_diag, loc)]; eval.vec[3] = opt->vec[SwizzleCompIdx(str[3], m_diag, loc)]; break; default: break; } ir.m_instructions.pop_back(); ir.m_instructions.emplace_back(IR::OpLoadImm, swizNode->m_tok.m_location); IR::Instruction& inst = ir.m_instructions.back(); inst.m_target = target; inst.m_loadImm.m_immVec = eval; } else { ir.m_instructions.emplace_back(IR::OpSwizzle, swizNode->m_tok.m_location); IR::Instruction& inst = ir.m_instructions.back(); inst.m_swizzle.m_instIdx = ir.m_instructions.size() - 2; inst.m_target = target; for (int i=0 ; im_tok.m_location); } } void Lexer::RecursiveGroupCompile(IR& ir, const Lexer::OperationNode* groupNode, IR::RegID target) const { IR::RegID tgt = target; for (const Lexer::OperationNode* sn = groupNode->m_sub ; sn ; sn = sn->m_next, ++tgt) { const Parser::Token& tok = sn->m_tok; switch (tok.m_type) { case Parser::TokenFunctionStart: if (!tok.m_tokenString.compare("vec3")) EmitVec3(ir, sn, tgt); else if (!tok.m_tokenString.compare("vec4")) EmitVec4(ir, sn, tgt); else RecursiveFuncCompile(ir, sn, tgt); break; case Parser::TokenEvalGroupStart: RecursiveGroupCompile(ir, sn, tgt); break; case Parser::TokenNumLiteral: { ir.m_instructions.emplace_back(IR::OpLoadImm, tok.m_location); IR::Instruction& inst = ir.m_instructions.back(); inst.m_target = tgt; inst.m_loadImm.m_immVec.vec[0] = tok.m_tokenFloat; inst.m_loadImm.m_immVec.vec[1] = tok.m_tokenFloat; inst.m_loadImm.m_immVec.vec[2] = tok.m_tokenFloat; inst.m_loadImm.m_immVec.vec[3] = tok.m_tokenFloat; break; } case Parser::TokenArithmeticOp: EmitArithmetic(ir, sn, tgt); break; case Parser::TokenVectorSwizzle: EmitVectorSwizzle(ir, sn, tgt); break; default: m_diag.reportCompileErr(tok.m_location, "invalid lexer node for IR"); break; }; } if (tgt > ir.m_regCount) ir.m_regCount = tgt; } void Lexer::RecursiveFuncCompile(IR& ir, const Lexer::OperationNode* funcNode, IR::RegID target) const { IR::RegID tgt = target; std::vector instIdxs; for (const Lexer::OperationNode* gn = funcNode->m_sub ; gn ; gn = gn->m_next, ++tgt) { RecursiveGroupCompile(ir, gn, tgt); instIdxs.push_back(ir.m_instructions.size() - 1); } ir.m_instructions.emplace_back(IR::OpCall, funcNode->m_tok.m_location); IR::Instruction& inst = ir.m_instructions.back(); inst.m_call.m_name = funcNode->m_tok.m_tokenString; inst.m_call.m_argInstIdxs = std::move(instIdxs); inst.m_target = target; if (tgt > ir.m_regCount) ir.m_regCount = tgt; } IR Lexer::compileIR() const { if (!m_root) m_diag.reportCompileErr(SourceLocation(), "unable to compile HECL-IR for invalid source"); IR ir; RecursiveFuncCompile(ir, m_root, 0); return ir; } } }