diff options
Diffstat (limited to 'source/slang')
| -rw-r--r-- | source/slang/emit.cpp | 114 | ||||
| -rw-r--r-- | source/slang/ir-inst-defs.h | 2 | ||||
| -rw-r--r-- | source/slang/ir-insts.h | 16 | ||||
| -rw-r--r-- | source/slang/ir-ssa.cpp | 848 | ||||
| -rw-r--r-- | source/slang/ir-ssa.h | 9 | ||||
| -rw-r--r-- | source/slang/ir.cpp | 261 | ||||
| -rw-r--r-- | source/slang/ir.h | 62 | ||||
| -rw-r--r-- | source/slang/lower-to-ir.cpp | 37 | ||||
| -rw-r--r-- | source/slang/slang.vcxproj | 2 | ||||
| -rw-r--r-- | source/slang/slang.vcxproj.filters | 2 | ||||
| -rw-r--r-- | source/slang/vm.cpp | 40 |
11 files changed, 1381 insertions, 12 deletions
diff --git a/source/slang/emit.cpp b/source/slang/emit.cpp index f9f24fcf6..eaabd520c 100644 --- a/source/slang/emit.cpp +++ b/source/slang/emit.cpp @@ -6074,6 +6074,14 @@ emitDeclImpl(decl, nullptr); emit(";\n"); break; + case kIROp_undefined: + { + auto type = inst->getType(); + emitIRType(ctx, type, getIRName(inst)); + emit(";\n"); + } + break; + case kIROp_Var: { auto ptrType = inst->getType(); @@ -6202,6 +6210,37 @@ emitDeclImpl(decl, nullptr); } } + // When we are about to traverse an edge from one block to another, + // we need to emit the assignments that conceptually occur "along" + // the edge. In traditional SSA these are the phi nodes in the + // target block, while in our representation these use the arguments + // to the branch instruction to fill in the parameters of the target. + void emitPhiVarAssignments( + EmitContext* ctx, + UInt argCount, + IRUse* args, + IRBlock* targetBlock) + { + UInt argCounter = 0; + for (auto pp = targetBlock->getFirstParam(); pp; pp = pp->getNextParam()) + { + UInt argIndex = argCounter++; + + if (argIndex >= argCount) + { + assert(!"not enough arguments for branch"); + break; + } + + IRValue* arg = args[argIndex].usedValue; + + emitIROperand(ctx, pp); + emit(" = "); + emitIROperand(ctx, arg); + emit(";\n"); + } + } + // We want to emit a range of code in the IR, represented // by the blocks that are logically in the interval [begin, end) // which we consider as a single-entry multiple-exit region. @@ -6305,6 +6344,14 @@ emitDeclImpl(decl, nullptr); auto breakBlock = t->getBreakBlock(); auto continueBlock = t->getContinueBlock(); + UInt argCount = t->getArgCount(); + static const UInt kFixedArgCount = 3; + emitPhiVarAssignments( + ctx, + argCount - kFixedArgCount, + t->getArgs() + kFixedArgCount, + targetBlock); + if (auto loopControlDecoration = t->findDecoration<IRLoopControlDecoration>()) { switch (loopControlDecoration->mode) @@ -6388,7 +6435,20 @@ emitDeclImpl(decl, nullptr); break; case kIROp_break: - emit("break;\n"); + { + auto t = (IRBreak*)terminator; + auto targetBlock = t->getTargetBlock(); + + UInt argCount = t->getArgCount(); + static const UInt kFixedArgCount = 1; + emitPhiVarAssignments( + ctx, + argCount - kFixedArgCount, + t->getArgs() + kFixedArgCount, + targetBlock); + + emit("break;\n"); + } return; case kIROp_continue: @@ -6405,6 +6465,15 @@ emitDeclImpl(decl, nullptr); { auto continueInst = (IRContinue*) terminator; auto targetBlock = continueInst->getTargetBlock(); + + UInt argCount = continueInst->getArgCount(); + static const UInt kFixedArgCount = 1; + emitPhiVarAssignments( + ctx, + argCount - kFixedArgCount, + continueInst->getArgs() + kFixedArgCount, + targetBlock); + IRBlock* loopHead = nullptr; ctx->shared->irMapContinueTargetToLoopHead.TryGetValue(targetBlock, loopHead); SLANG_ASSERT(loopHead); @@ -6422,10 +6491,16 @@ emitDeclImpl(decl, nullptr); auto t = (IRLoopTest*)terminator; auto afterBlock = t->getTrueBlock(); + auto exitBlock = t->getFalseBlock(); emit("if("); emitIROperand(ctx, t->getCondition()); - emit(")\n{} else break;\n"); + emit(")\n{} else {\n"); + emitIRStmtsForBlocks( + ctx, + exitBlock, + nullptr); + emit("}\n"); // Continue with the block after the test block = afterBlock; @@ -6440,6 +6515,16 @@ emitDeclImpl(decl, nullptr); // block, or a backward edge to the top // of a loop. auto t = (IRUnconditionalBranch*)terminator; + auto targetBlock = t->getTargetBlock(); + + UInt argCount = t->getArgCount(); + static const UInt kFixedArgCount = 1; + emitPhiVarAssignments( + ctx, + argCount - kFixedArgCount, + t->getArgs() + kFixedArgCount, + targetBlock); + block = t->getTargetBlock(); } break; @@ -6758,6 +6843,27 @@ emitDeclImpl(decl, nullptr); } } + void emitPhiVarDecls( + EmitContext* ctx, + IRFunc* func) + { + // We will skip the first block, since its parameters are + // the parameters of the whole function. + auto bb = func->getFirstBlock(); + if (!bb) + return; + bb = bb->getNextBlock(); + + for (; bb; bb = bb->getNextBlock()) + { + for (auto pp = bb->getFirstParam(); pp; pp = pp->getNextParam()) + { + emitIRType(ctx, pp->getType(), getIRName(pp)); + emit(";\n"); + } + } + } + void emitIRSimpleFunc( EmitContext* ctx, IRFunc* func) @@ -6816,6 +6922,10 @@ emitDeclImpl(decl, nullptr); { emit("\n{\n"); + // HACK: forward-declare all the local variables needed for the + // prameters of non-entry blocks. + emitPhiVarDecls(ctx, func); + // Need to emit the operations in the blocks of the function emitIRStmtsForBlocks(ctx, func->getFirstBlock(), nullptr); diff --git a/source/slang/ir-inst-defs.h b/source/slang/ir-inst-defs.h index fa02da683..38515e6f0 100644 --- a/source/slang/ir-inst-defs.h +++ b/source/slang/ir-inst-defs.h @@ -96,6 +96,8 @@ INST(IntLit, integer_constant, 0, 0) INST(FloatLit, float_constant, 0, 0) INST(decl_ref, decl_ref, 0, 0) +INST(undefined, undefined, 0, 0) + INST(specialize, specialize, 2, 0) INST(lookup_interface_method, lookup_interface_method, 2, 0) INST(lookup_witness_table, lookup_witness_table, 2, 0) diff --git a/source/slang/ir-insts.h b/source/slang/ir-insts.h index 46a804a1c..266374545 100644 --- a/source/slang/ir-insts.h +++ b/source/slang/ir-insts.h @@ -326,6 +326,15 @@ struct IRWitnessTable : IRGlobalValue IRValueList<IRWitnessTableEntry> entries; }; +// An instruction that yields an undefined value. +// +// Note that we make this an instruction rather than a value, +// so that we will be able to identify a variable that is +// used when undefined. +struct IRUndefined : IRInst +{ +}; + // Description of an instruction to be used for global value numbering struct IRInstKey { @@ -394,6 +403,7 @@ struct IRBuilder IRValue* getBoolValue(bool value); IRValue* getIntValue(IRType* type, IRIntegerValue value); IRValue* getFloatValue(IRType* type, IRFloatingPointValue value); + IRValue* getDeclRefVal( DeclRefBase const& declRef); IRValue* getTypeVal(IRType* type); // create an IR value that represents a type @@ -443,6 +453,10 @@ struct IRBuilder UInt argCount, IRValue* const* args); + IRUndefined* emitUndefined(IRType* type); + + + IRModule* createModule(); IRFunc* createFunc(); @@ -458,6 +472,8 @@ struct IRBuilder IRBlock* createBlock(); IRBlock* emitBlock(); + IRParam* createParam( + IRType* type); IRParam* emitParam( IRType* type); diff --git a/source/slang/ir-ssa.cpp b/source/slang/ir-ssa.cpp new file mode 100644 index 000000000..366ecc279 --- /dev/null +++ b/source/slang/ir-ssa.cpp @@ -0,0 +1,848 @@ +// ir-ssa.cpp +#include "ir-ssa.h" + +#include "ir.h" +#include "ir-insts.h" + +namespace Slang { + +// Track information on a phi node we are in +// the process of constructing. +struct PhiInfo : RefObject +{ + // The phi node will be represented as a parameter + // to a (non-entry) basic block. + IRParam* phi; + + // The original variable that this phi will be replacing. + IRVar* var; + + // The operands to the phi will be stored as uses here, + // because our IR parameters don't have operands. + // + // Once we've collected all the values we plan to use, + // we will turn this into argument in predecessor blocks + // that branch to this one. + // + // The order of elements in this list must match the + // order in which the predecessor blocks get enumerated. + List<IRUse> operands; + + // Did we end up removing this phi because it was determined + // to be trivial? + bool wasRemoved = false; +}; + +// Information about a basic block that we generate/use +// during SSA construction. +struct SSABlockInfo : RefObject +{ + // Map a promotable variable to the value to + // use for that variable + Dictionary<IRVar*, IRValue*> valueForVar; + + // The underlying basic block. + IRBlock* block; + + // Have we processed all the instructions in the + // body of this block (so that we would have + // found any stores to SSA variables)? + bool isFilled = false; + + // Have we filled all the predecessors of + // this block, so that we can actually perform + // look up in them? + bool isSealed = false; + + // An IR builder to use when we want to construct + // stuff in the context of this block + IRBuilder builder; + + // Phi nodes we are creating for this block. + List<PhiInfo*> phis; + + // Arguments that this block needs to pass along + // to the phi nodes defined by is sucessor + List<IRValue*> successorArgs; +}; + +// State for constructing SSA form for a global value +// with code (usually a function). +struct ConstructSSAContext +{ + // The value that we want to rewrite into SSA form + // (usually an IR function) + IRGlobalValueWithCode* globalVal; + + // Variables that we've identified for promotion + // to SSA values. + List<IRVar*> promotableVars; + + // Information about each basic block + Dictionary<IRBlock*, RefPtr<SSABlockInfo>> blockInfos; + + // IR building state to use during the operation + SharedIRBuilder sharedBuilder; + + + Dictionary<IRParam*, RefPtr<PhiInfo>> phiInfos; + + PhiInfo* getPhiInfo(IRParam* phi) + { + return *phiInfos.TryGetValue(phi); + } +}; + +// Is the given variable one that we can promote to SSA form? +bool isPromotableVar( + ConstructSSAContext* /*context*/, + IRVar* var) +{ + // If the variable is only used directly as the pointer + // operand of load and store instructions, then it should + // be promote-able. + // + // For now, we won't deal with cases where the variable + // is an aggregate an we sometimes pull out individual + // fields or elements. This is an important extension, + // but we probably also need to think about scalarizing + // any aggregates when we promote them to registers. + + for (auto u = var->firstUse; u; u = u->nextUse) + { + auto user = u->user; + switch (user->op) + { + default: + // If the variable gets used by any operation + // we can't account for directly, then it isn't + // promotable. + return false; + + case kIROp_Load: + { + // A load has only a single argument, so + // it had better be our pointer. + assert(u == &((IRLoad*) user)->ptr); + } + break; + + case kIROp_Store: + { + auto storeInst = (IRStore*)user; + + // We don't want to promote a variable if + // its address gets stored into another + // variable, so check for that case. + if (u == &storeInst->val) + return false; + + // Otherwise our variable is being used + // as the destination for the store, and + // that is okay by us. + assert(u == &storeInst->ptr); + } + break; + } + } + + // If all of the uses passed our checking, then + // we are good to go. + return true; +} + +// Identify local variables that can be promoted to SSA form +void identifyPromotableVars( + ConstructSSAContext* context) +{ + for (auto bb = context->globalVal->getFirstBlock(); bb; bb = bb->getNextBlock()) + { + for (auto ii = bb->getFirstInst(); ii; ii = ii->getNextInst()) + { + if (ii->op != kIROp_Var) + continue; + + IRVar* var = (IRVar*)ii; + + if (isPromotableVar(context, var)) + { + context->promotableVars.Add(var); + } + } + } +} + +IRVar* asPromotableVar( + ConstructSSAContext* context, + IRValue* value) +{ + if (value->op != kIROp_Var) + return nullptr; + + IRVar* var = (IRVar*)value; + if (!context->promotableVars.Contains(var)) + return nullptr; + + return var; +} + +// Try to read the value of an SSA variable +// in the context of the given block. If +// the variable is defined in the block, then +// that value will be used. If not, this all +// may recursively work its way up through +// the predecessors of the block. +IRValue* readVar( + ConstructSSAContext* context, + SSABlockInfo* blockInfo, + IRVar* var); + + +// Add a phi node to represent the given variable +PhiInfo* addPhi( + ConstructSSAContext* context, + SSABlockInfo* blockInfo, + IRVar* var) +{ + auto builder = &blockInfo->builder; + IRParam* phi = builder->createParam(var->getType()->getValueType()); + + RefPtr<PhiInfo> phiInfo = new PhiInfo(); + context->phiInfos.Add(phi, phiInfo); + + phiInfo->phi = phi; + phiInfo->var = var; + + blockInfo->phis.Add(phiInfo); + + return phiInfo; +} + +IRValue* tryRemoveTrivialPhi( + ConstructSSAContext* /*context*/, + PhiInfo* phiInfo) +{ + auto phi = phiInfo->phi; + + // We are going to check if all of the operands + // to the phi are either the same, or are equal + // to the phi itself. + + IRValue* same = nullptr; + for (auto u : phiInfo->operands) + { + auto usedVal = u.usedValue; + assert(usedVal); + + if (usedVal == same || usedVal == phi) + { + // Either this is a self-reference, or it refers + // to the same value we've seen already. + continue; + } + if (same != nullptr) + { + // We've found at least two distinct values + // other than the phi itself, so this phi + // indeed appears to be non-trivial. + // + // We will keep the phi around. + return phi; + } + else + { + // This value is distinct from the phi itself, + // so we need to track its value. + same = usedVal; + } + } + + if (!same) + { + // There were no operands other than the phi itself. + // This implies that the value at the use sites should + // actually be undefined. + + assert(!"unimplemented"); + } + + // replace uses of the phi (including its possible uses + // of itself) with the unique non-phi value. + phi->replaceUsesWith(same); + + // We will mark the phi as removed in the `PhiInfo` structure, + // so that we can avoid adding it to the block later. + phiInfo->wasRemoved = true; + + // TODO: we need a way to record that this parameter + // is no longer to be used... + + // TODO: need to recursively consider chances to simplify + // other phi nodes now. + + return same; +} + +IRValue* addPhiOperands( + ConstructSSAContext* context, + SSABlockInfo* blockInfo, + PhiInfo* phiInfo) +{ + auto var = phiInfo->var; + + auto block = blockInfo->block; + + List<IRValue*> operandValues; + for (auto predBlock : block->getPredecessors()) + { + // Precondition: if we have multiple predecessors, then + // each must have only one successor (no critical edges). + // + assert(predBlock->getSuccessors().getCount() == 1); + + auto predInfo = *context->blockInfos.TryGetValue(predBlock); + + auto phiOperand = readVar(context, predInfo, var); + + operandValues.Add(phiOperand); + } + + // The `IRUse` type needs to stay at a stable location + // since they get threaded into lists. We allocate the + // list with its final size so that we can preserve the + // required invariant. + + UInt operandCount = operandValues.Count(); + phiInfo->operands.SetSize(operandCount); + for(UInt ii = 0; ii < operandCount; ++ii) + { + phiInfo->operands[ii].init(nullptr, operandValues[ii]); + } + + return tryRemoveTrivialPhi(context, phiInfo); +} + +void writeVar( + ConstructSSAContext* /*context*/, + SSABlockInfo* blockInfo, + IRVar* var, + IRValue* val) +{ + blockInfo->valueForVar[var] = val; +} + +void maybeSealBlock( + ConstructSSAContext* context, + SSABlockInfo* blockInfo) +{ + // We can't seal a block that has already been sealed. + if (blockInfo->isSealed) + return; + + // We can't seal a block until all of its predecessors + // have been filled. + for (auto pp : blockInfo->block->getPredecessors()) + { + auto predInfo = *context->blockInfos.TryGetValue(pp); + if (!predInfo->isFilled) + return; + } + + // All the checks passed, so it seems like we can be sealed. + + // We will loop over any incomplete phis that have been recoreded + // for this block, and complete them here. + // + // Note that we are doing the "inefficient" loop where we compute + // the count on each iteration to account for the possibility that + // new incomplete phis will get added while we are working. + for (UInt ii = 0; ii < blockInfo->phis.Count(); ++ii) + { + auto incompletePhi = blockInfo->phis[ii]; + addPhiOperands(context, blockInfo, incompletePhi); + } + + // After we've completed all our incomplete phis, we can mark this + // block as sealed and move along. + blockInfo->isSealed = true; +} + +IRValue* readVarRec( + ConstructSSAContext* context, + SSABlockInfo* blockInfo, + IRVar* var) +{ + IRValue* val = nullptr; + if (!blockInfo->isSealed) + { + // If block isn't sealed, we need to + // speculatively add a phi to it. + // This phi may get removed later, once + // we are able to seal this block. + + PhiInfo* phiInfo = addPhi(context, blockInfo, var); + val = phiInfo->phi; + } + else + { + // If the block is sealed, then we are free to look at + // it predecessor list, and use that to decide what to do. + auto predecessors = blockInfo->block->getPredecessors(); + + // + IRBlock* firstPred = nullptr; + bool multiplePreds = false; + for (auto pp : predecessors) + { + if (!firstPred) + { + // A candidate for the sole predecessor + firstPred = pp; + } + else if (pp == firstPred) + { + // Same as existing predecessor + } + else + { + // Multiple unique predecessors + multiplePreds = true; + } + } + + if (!firstPred) + { + // The block had *no* predecssors. This will commonly + // happen for the entry block, but could also conceivably + // happen for a block that is somehow disconnected + // from the CFG and thus unreachable. + + // We would only reach this function (`readVarRec`) if + // a local lookup in the block had already failed, so + // at this point we are dealing with an undefined value. + + auto type = var->getType()->getValueType(); + val = blockInfo->builder.emitUndefined(type); + } + else if (!multiplePreds) + { + // There is only a single predecessor for this block, + // so there is no need to insert a phi. Instead, we + // just perform the lookup step recursively in + // the predecessor. + auto predInfo = *context->blockInfos.TryGetValue(firstPred); + val = readVar(context, predInfo, var); + } + else + { + // The default/fallback case requires us to create + // a phi node in the current block, and then look + // up the appropriate operands in the predecessor + // blocks, which will eventually become the operands + // that drive the phi. + + // Create the phi node for the given variable + PhiInfo* phiInfo = addPhi(context, blockInfo, var); + + // Mark the phi as the value for the variable inside + // this block + writeVar(context, blockInfo, var, phiInfo->phi); + + // Now add operands to the phi and maybe simplify + // it, based on what gets found. + + val = addPhiOperands(context, blockInfo, phiInfo); + } + } + + // Whatever value we find, we need to mark it as the + // value for the given variable in this block + writeVar(context, blockInfo, var, val); + + return val; +} + + +IRValue* readVar( + ConstructSSAContext* context, + SSABlockInfo* blockInfo, + IRVar* var) +{ + // In the easy case, there will be a preceeding + // store in the same block, so we can use + // that local value. + IRValue* val = nullptr; + if (blockInfo->valueForVar.TryGetValue(var, val)) + { + // Hooray, we found a value to use, and we + // can proceed without too many complications. + return val; + } + + // Otherwise we need to try to non-trivial/recursive + // case of lookup. + return readVarRec(context, blockInfo, var); +} + +void processBlock( + ConstructSSAContext* context, + IRBlock* block, + SSABlockInfo* blockInfo) +{ + // Before starting, check if this block can be sealed + maybeSealBlock(context, blockInfo); + + // Walk the instructions in the block, and either + // leave them as-is, or replace them with a value + // that we look up with local/global value numbering + + IRInst* next = nullptr; + for (auto ii = block->getFirstInst(); ii; ii = next) + { + next = ii->getNextInst(); + + // Any new instructions we create to represent + // the new value will get inserted before whatever + // instruction we are working with. + blockInfo->builder.insertBeforeInst = ii; + + switch (ii->op) + { + default: + // Ordinary instruction -> leave as-is + break; + + case kIROp_Store: + { + auto storeInst = (IRStore*)ii; + auto ptrArg = storeInst->ptr.usedValue; + auto valArg = storeInst->val.usedValue; + + if (auto var = asPromotableVar(context, ptrArg)) + { + // We are storing to a promotable variable, + // so we want to register the value being + // stored as the value for the given SSA + // variable. + writeVar(context, blockInfo, var, valArg); + + // Also eliminate the store instruction, + // since it is no longer needed. + storeInst->removeAndDeallocate(); + } + } + break; + + case kIROp_Load: + { + IRLoad* loadInst = (IRLoad*)ii; + auto ptrArg = loadInst->ptr.usedValue; + + if (auto var = asPromotableVar(context, ptrArg)) + { + // We are loading from a promotable variable. + // Look up the value in the context of this + // block. + auto val = readVar(context, blockInfo, var); + + // We can just replace all uses of this + // load instruction with the given value. + loadInst->replaceUsesWith(val); + + // Also eliminate the load instruction, + // since it is no longer needed. + loadInst->removeAndDeallocate(); + } + } + break; + } + } + + blockInfo->builder.insertBeforeInst = block->lastInst; + + // Once we are done with all of the instructions + // in a block, we can mark it as "filled," which + // means we can actually consider lookups into + // it. + blockInfo->isFilled = true; + + // Having filled this block might allow us to seal some + // of its successor(s) + for (auto ss : block->getSuccessors()) + { + auto successorInfo = *context->blockInfos.TryGetValue(ss); + maybeSealBlock(context, successorInfo); + } +} + +static void breakCriticalEdges( + ConstructSSAContext* context) +{ + // A critical edge is an edge P -> S where + // P has multiple sucessors, and S has multiple + // predecessors. + // + // In the context of our CFG representation, such an edge + // will be an `IRUse` in the terminator instruction of block P, + // which refers to block S. + // + // We will make a pass over the CFG to collect all the critical + // edges, and then we will break them in a follow-up pass. + + List<IRUse*> criticalEdges; + + auto globalVal = context->globalVal; + for (auto pred = globalVal->getFirstBlock(); pred; pred = pred->getNextBlock()) + { + auto successors = pred->getSuccessors(); + if (successors.getCount() <= 1) + continue; + + auto succIter = successors.begin(); + auto succEnd = successors.end(); + + for (; succIter != succEnd; ++succIter) + { + auto succ = *succIter; + + // For the edge to be critical, the successor must have + // more than one predecessor. + // More than that, we require that it has more than one + // *unique* predecessor, to handle the case where multiple + // cases of a `switch` might lead to the same block. + // + // To implement this, we test if it has any predecessor + // other than `pred` which we already know about. + + bool multiplePreds = false; + for (auto pp : succ->getPredecessors()) + { + if (pp != pred) + { + multiplePreds = true; + break; + } + } + if (!multiplePreds) + continue; + + // We have found a critical edge from `pred` to `succ`. + // + // Furthermore, the `IRUse` embedded in `succIter` represents + // that edge directly. + auto edgeUse = succIter.use; + criticalEdges.Add(edgeUse); + } + } + + // Now we will iterate over the critical edges and break each + // one by inserting a new block. Note that we do not try + // to break the edges while doing the initial walk, because + // that would change the CFG while we are walking it. + + for (auto edgeUse : criticalEdges) + { + auto pred = (IRBlock*) edgeUse->user->parent; + assert(pred->op == kIROp_Block); + + auto succ = (IRBlock*)edgeUse->usedValue; + assert(succ->op == kIROp_Block); + + IRBuilder builder; + builder.sharedBuilder = &context->sharedBuilder; + builder.curFunc = globalVal; + builder.curBlock = pred; + + // Create a new block that will sit "along" the edge + IRBlock* edgeBlock = builder.createBlock(); + + // The predecessor block should now branch to + // the edge block. + edgeUse->set(edgeBlock); + + // The edge block should branch (unconditionally) + // to the successor block. + builder.curBlock = edgeBlock; + builder.emitBranch(succ); + + // Insert the new block into the block list + // for the function. + // + // In principle, the order of this list shouldn't + // affect the semantics of a program, but we + // might want to be careful about ordering anyway. + edgeBlock->insertAfter(pred); + } +} + +// Construct SSA form for a global value with code +void constructSSA(ConstructSSAContext* context) +{ + // First, detect and and break any critical edges in the CFG, + // because our representation of SSA form doesn't allow for them. + breakCriticalEdges(context); + + + // Figure out what variables we can promote to + // SSA temporaries. + identifyPromotableVars(context); + + // If none of the variables are promote-able, + // then we can exit without making any changes + if (context->promotableVars.Count() == 0) + return; + + // We are going to walk the blocks in order, + // and try to process each, by replacing loads + // and stores of promotable variables with simple values. + + auto globalVal = context->globalVal; + for (auto bb = globalVal->firstBlock; bb; bb = bb->nextBlock) + { + auto blockInfo = new SSABlockInfo(); + blockInfo->block = bb; + + blockInfo->builder.sharedBuilder = &context->sharedBuilder; + blockInfo->builder.curBlock = bb; + blockInfo->builder.curFunc = globalVal; + blockInfo->builder.insertBeforeInst = bb->lastInst; + + context->blockInfos.Add(bb, blockInfo); + } + for (auto bb = globalVal->firstBlock; bb; bb = bb->nextBlock) + { + auto blockInfo = * context->blockInfos.TryGetValue(bb); + processBlock(context, bb, blockInfo); + } + + // We need to transfer the logical arguments to our phi nodes + // from the phi nodes back to the predecessor blocks that will + // pass them in. + for (auto bb = globalVal->firstBlock; bb; bb = bb->nextBlock) + { + auto blockInfo = *context->blockInfos.TryGetValue(bb); + + for (auto phiInfo : blockInfo->phis) + { + // If we eliminated this phi, then we had better not + // include it in the result. + if (phiInfo->wasRemoved) + continue; + + // We should add the phi as an explicit parameter of + // the given block. + bb->addParam(phiInfo->phi); + + UInt predCounter = 0; + for (auto pp : bb->getPredecessors()) + { + UInt predIndex = predCounter++; + auto predInfo = *context->blockInfos.TryGetValue(pp); + + IRValue* operandVal = phiInfo->operands[predIndex].usedValue; + + phiInfo->operands[predIndex].clear(); + + predInfo->successorArgs.Add(operandVal); + } + } + } + + // Some blocks may now need to pass along arguments to their sucessor, + // which have been stored into the `SSABlockInfo::successorArgs` field. + for (auto bb = globalVal->firstBlock; bb; bb = bb->nextBlock) + { + auto blockInfo = * context->blockInfos.TryGetValue(bb); + + // Sanity check: all blocks should be filled and sealed. + assert(blockInfo->isSealed); + assert(blockInfo->isFilled); + + // Don't do any work for blocks that don't need to pass along + // values to the sucessor block. + auto addedArgCount = blockInfo->successorArgs.Count(); + if (addedArgCount == 0) + continue; + + // We need to replace the terminator instruction with one that + // has additional arguments. + + IRTerminatorInst* oldTerminator = (IRTerminatorInst*) bb->getLastInst(); + assert(isTerminatorInst(oldTerminator)); + + blockInfo->builder.insertBeforeInst = nullptr; + + auto oldArgCount = oldTerminator->argCount; + auto newArgCount = oldArgCount + addedArgCount; + + List<IRValue*> newArgs; + for (UInt aa = 0; aa < oldArgCount; ++aa) + { + newArgs.Add(oldTerminator->getArg(aa)); + } + for (UInt aa = 0; aa < addedArgCount; ++aa) + { + newArgs.Add(blockInfo->successorArgs[aa]); + } + + IRTerminatorInst* newTerminator = (IRTerminatorInst*)blockInfo->builder.emitIntrinsicInst( + oldTerminator->type, + oldTerminator->op, + newArgCount, + newArgs.Buffer()); + + // Swap decorations over to the new instruction + newTerminator->firstDecoration = oldTerminator->firstDecoration; + oldTerminator->firstDecoration = nullptr; + + // A terminator better not have uses, so we shouldn't have + // to replace them. + assert(!oldTerminator->firstUse); + + + // Okay, we should be clear to remove the old terminator + oldTerminator->removeAndDeallocate(); + } + + // Now we should be able to go through and remove + // of of the variables + for (auto var : context->promotableVars) + { + var->removeAndDeallocate(); + } +} + +// Construct SSA form for a global value with code +void constructSSA(IRModule* module, IRGlobalValueWithCode* globalVal) +{ + ConstructSSAContext context; + context.globalVal = globalVal; + + context.sharedBuilder.module = module; + context.sharedBuilder.session = module->session; + + constructSSA(&context); +} + +void constructSSA(IRModule* module, IRGlobalValue* globalVal) +{ + switch (globalVal->op) + { + case kIROp_Func: + case kIROp_global_var: + constructSSA(module, (IRGlobalValueWithCode*)globalVal); + + default: + break; + } +} + +void constructSSA(IRModule* module) +{ + for (auto gv = module->getFirstGlobalValue(); gv; gv = gv->getNextValue()) + { + constructSSA(module, gv); + } +} + +} diff --git a/source/slang/ir-ssa.h b/source/slang/ir-ssa.h new file mode 100644 index 000000000..b9a05c2ff --- /dev/null +++ b/source/slang/ir-ssa.h @@ -0,0 +1,9 @@ +// ir-ssa.h +#pragma once + +struct IRModule; + +namespace Slang +{ + void constructSSA(IRModule* module); +} diff --git a/source/slang/ir.cpp b/source/slang/ir.cpp index 77aac3f5a..634632f0d 100644 --- a/source/slang/ir.cpp +++ b/source/slang/ir.cpp @@ -28,16 +28,27 @@ namespace Slang { // TODO: need to make this faster by using a dictionary... - if (0) {} + static const struct { + char const* mnemonic; + IROp op; + } kOps[] = { #define INST(ID, MNEMONIC, ARG_COUNT, FLAGS) \ - else if(strcmp(name, #MNEMONIC) == 0) return kIROp_##ID; + { #MNEMONIC, kIROp_##ID }, #define PSEUDO_INST(ID) \ - else if(strcmp(name, #ID) == 0) return kIRPseudoOp_##ID; + { #ID, kIRPseudoOp_##ID }, #include "ir-inst-defs.h" + }; + + for (auto ee : kOps) + { + if (strcmp(name, ee.mnemonic) == 0) + return ee.op; + } + return IROp(kIROp_Invalid); } @@ -125,6 +136,228 @@ namespace Slang lastParam = param; } + // The predecessors of a block should all show up as users + // of its value, so rather than explicitly store the CFG, + // we will recover it on demand from the use-def information. + // + // Note: we are really iterating over incoming/outgoing *edges* + // for a block, because there might be multiple uses of a block, + // if more than one way of an N-way branch targets the same block. + + // Get the list of successor blocks for an instruction, + // which we expect to be the last instruction in a block. + static IRBlock::SuccessorList getSuccessors(IRInst* terminator) + { + // If the block somehow isn't terminated, then + // there is no way to read its successors, so + // we return an empty list. + if (!terminator || !isTerminatorInst(terminator)) + return IRBlock::SuccessorList(nullptr, nullptr); + + // Otherwise, based on the opcode of the terminator + // instruction, we will build up our list of uses. + IRUse* begin = nullptr; + IRUse* end = nullptr; + UInt stride = 1; + + auto args = terminator->getArgs(); + switch (terminator->op) + { + case kIROp_ReturnVal: + case kIROp_ReturnVoid: + case kIROp_unreachable: + case kIROp_discard: + break; + + case kIROp_unconditionalBranch: + case kIROp_break: + case kIROp_continue: + case kIROp_loop: + // unconditonalBranch <block> + begin = args + 0; + end = begin + 1; + break; + + case kIROp_conditionalBranch: + case kIROp_if: + case kIROp_ifElse: + case kIROp_loopTest: + // conditionalBranch <condition> <trueBlock> <falseBlock> + begin = args + 1; + end = begin + 2; + break; + + case kIROp_switch: + // switch <val> <break> <default> <caseVal1> <caseBlock1> ... + begin = args + 4; + end = args + terminator->getArgCount() + 1; + stride = 2; + break; + + default: + assert(!"unepxected"); + return IRBlock::SuccessorList(nullptr, nullptr); + } + + return IRBlock::SuccessorList(begin, end, stride); + } + + void IRBlock::insertAfter(IRBlock* other) + { + assert(other); + insertAfter(other, other->parentFunc); + } + + void IRBlock::insertAfter(IRBlock* other, IRGlobalValueWithCode* func) + { + assert(other || func); + + if (!other) other = func->lastBlock; + if (!func) func = other->parentFunc; + + assert(func); + + auto pp = other; + auto nn = other ? other->nextBlock : nullptr; + + if (pp) + { + pp->nextBlock = this; + } + else + { + func->firstBlock = this; + } + + if (nn) + { + nn->prevBlock = this; + } + else + { + func->lastBlock = this; + } + + this->prevBlock = pp; + this->nextBlock = nn; + this->parentFunc = func; + } + + static IRUse* adjustPredecessorUse(IRUse* use) + { + // We will search until we either find a + // suitable use, or run out of uses. + for (;use; use = use->nextUse) + { + // We only want to deal with uses that represent + // a "sucessor" operand to some terminator instruction. + // We will re-use the logic for getting the successor + // list from such an instruction. + + auto successorList = getSuccessors((IRInst*) use->user); + + if(use >= successorList.begin_ + && use < successorList.end_) + { + UInt index = (use - successorList.begin_); + if ((index % successorList.stride) == 0) + { + // This use is in the range of the sucessor list, + // and so it represents a real edge between + // blocks. + return use; + } + } + } + + // If we ran out of uses, then we are at the end + // of the list of incoming edges. + return nullptr; + } + + IRBlock::PredecessorList IRBlock::getPredecessors() + { + // We want to iterate over the predecessors of this block. + // First, we resign ourselves to iterating over the + // incoming edges, rather than the blocks themselves. + // This might sound like a trival distinction, but it is + // possible for there to be multiple edges between two + // blocks (as for a `switch` with multiple cases that + // map to the same code). Any client that wants just + // the unique predecessor blocks needs to deal with + // the deduplication themselves. + // + // Next, we note that for any predecessor edge, there will + // be a use of this block in the terminator instruction of + // the predecessor. We basically just want to iterate over + // the users of this block, then, but we need to be careful + // to rule out anything that doesn't actually represent + // an edge. The `adjustPredecessorUse` function will be + // used to search for a use that actually represents an edge. + + return PredecessorList( + adjustPredecessorUse(firstUse)); + } + + UInt IRBlock::PredecessorList::getCount() + { + UInt count = 0; + for (auto ii : *this) + { + (void)ii; + count++; + } + return count; + } + + void IRBlock::PredecessorList::Iterator::operator++() + { + if (!use) return; + use = adjustPredecessorUse(use->nextUse); + } + + IRBlock* IRBlock::PredecessorList::Iterator::operator*() + { + if (!use) return nullptr; + return (IRBlock*)use->user->parent; + } + + IRBlock::SuccessorList IRBlock::getSuccessors() + { + // The successors of a block will all be listed + // as operands of its terminator instruction. + // Depending on the terminator, we might have + // different numbers of operands to deal with. + // + // (We might also have to deal with a "stride" + // in the case where the basic-block operands + // are mixed up with non-block operands) + + auto terminator = getLastInst(); + return Slang::getSuccessors(terminator); + } + + UInt IRBlock::SuccessorList::getCount() + { + UInt count = 0; + for (auto ii : *this) + { + (void)ii; + count++; + } + return count; + } + + void IRBlock::SuccessorList::Iterator::operator++() + { + use += stride; + } + + IRBlock* IRBlock::SuccessorList::Iterator::operator*() + { + return (IRBlock*)use->usedValue; + } + // IRFunc IRType* IRFunc::getResultType() { return getType()->getResultType(); } @@ -609,6 +842,18 @@ namespace Slang &value); } + IRUndefined* IRBuilder::emitUndefined(IRType* type) + { + auto inst = createInst<IRUndefined>( + this, + kIROp_undefined, + type); + + addInst(inst); + + return inst; + } + IRValue* IRBuilder::getDeclRefVal( DeclRefBase const& declRef) { @@ -1004,13 +1249,20 @@ namespace Slang return bb; } - IRParam* IRBuilder::emitParam( + IRParam* IRBuilder::createParam( IRType* type) { auto param = createValue<IRParam>( this, kIROp_Param, type); + return param; + } + + IRParam* IRBuilder::emitParam( + IRType* type) + { + auto param = createParam(type); if (auto bb = curBlock) { @@ -3283,6 +3535,7 @@ namespace Slang case kIROp_Func: case kIROp_witness_table: return cloneGlobalValue(this, (IRGlobalValue*) originalValue); + case kIROp_boolConst: { IRConstant* c = (IRConstant*)originalValue; diff --git a/source/slang/ir.h b/source/slang/ir.h index 1b4529a3c..028468308 100644 --- a/source/slang/ir.h +++ b/source/slang/ir.h @@ -375,6 +375,68 @@ struct IRBlock : IRValue IRGlobalValueWithCode* getParent() { return parentFunc; } + void insertAfter(IRBlock* other); + void insertAfter(IRBlock* other, IRGlobalValueWithCode* func); + + struct PredecessorList + { + PredecessorList(IRUse* begin) : b(begin) {} + IRUse* b; + + UInt getCount(); + + struct Iterator + { + Iterator(IRUse* use) : use(use) {} + + IRBlock* operator*(); + + void operator++(); + + bool operator!=(Iterator const& that) + { + return use != that.use; + } + + IRUse* use; + }; + + Iterator begin() { return Iterator(b); } + Iterator end() { return Iterator(nullptr); } + }; + + struct SuccessorList + { + SuccessorList(IRUse* begin, IRUse* end, UInt stride = 1) : begin_(begin), end_(end), stride(stride) {} + IRUse* begin_; + IRUse* end_; + UInt stride; + + UInt getCount(); + + struct Iterator + { + Iterator(IRUse* use, UInt stride) : use(use), stride(stride) {} + + IRBlock* operator*(); + + void operator++(); + + bool operator!=(Iterator const& that) + { + return use != that.use; + } + + IRUse* use; + UInt stride; + }; + + Iterator begin() { return Iterator(begin_, stride); } + Iterator end() { return Iterator(end_, stride); } + }; + + PredecessorList getPredecessors(); + SuccessorList getSuccessors(); }; // For right now, we will represent the type of diff --git a/source/slang/lower-to-ir.cpp b/source/slang/lower-to-ir.cpp index bde50aa89..5c6f62c38 100644 --- a/source/slang/lower-to-ir.cpp +++ b/source/slang/lower-to-ir.cpp @@ -5,6 +5,7 @@ #include "ir.h" #include "ir-insts.h" +#include "ir-ssa.h" #include "mangle.h" #include "type-layout.h" #include "visitor.h" @@ -4201,6 +4202,42 @@ IRModule* generateIRForTranslationUnit( ensureDecl(context, decl); } + // We will perform certain "mandatory" optimization passes now. + // These passes serve two purposes: + // + // 1. To simplify the code that we use in backend compilation, + // or when serializing/deserializing modules, so that we can + // amortize this effort when we compile multiple entry points + // that use the same module(s). + // + // 2. To ensure certain semantic properties that can't be + // validated without dataflow information. For example, we want + // to detect when a variable might be used before it is initialized. + + // Note: if you need to debug the IR that is created before + // any mandatory optimizations have been applied, then + // uncomment this line while debugging. + + // dumpIR(module); + + // First, attempt to promote local variables to SSA + // temporaries whenever possible. + constructSSA(module); + + // TODO: Do basic constant folding and DCE + + // TODO: give error messages if any `undefined` or + // `unreachable` instructions remain. + + // TODO: consider doing some more aggressive optimizations + // (in particular specialization of generics) here, so + // that we can avoid doing them downstream. + // + // Note: doing specialization or inlining involving code + // from other modules potentially makes the IR we generate + // "fragile" in that we'd now need to recompile when + // a module we depend on changes. + // If we are being sked to dump IR during compilation, // then we can dump the initial IR for the module here. if(compileRequest->shouldDumpIR) diff --git a/source/slang/slang.vcxproj b/source/slang/slang.vcxproj index 23400bb23..1ff88020f 100644 --- a/source/slang/slang.vcxproj +++ b/source/slang/slang.vcxproj @@ -179,6 +179,7 @@ <ClInclude Include="expr-defs.h" /> <ClInclude Include="ir-inst-defs.h" /> <ClInclude Include="ir-insts.h" /> + <ClInclude Include="ir-ssa.h" /> <ClInclude Include="ir.h" /> <ClInclude Include="legalize-types.h" /> <ClInclude Include="lexer.h" /> @@ -217,6 +218,7 @@ <ClCompile Include="dxc-support.cpp" /> <ClCompile Include="emit.cpp" /> <ClCompile Include="ir-legalize-types.cpp" /> + <ClCompile Include="ir-ssa.cpp" /> <ClCompile Include="ir.cpp" /> <ClCompile Include="legalize-types.cpp" /> <ClCompile Include="lexer.cpp" /> diff --git a/source/slang/slang.vcxproj.filters b/source/slang/slang.vcxproj.filters index 4c463a62a..e0bf59025 100644 --- a/source/slang/slang.vcxproj.filters +++ b/source/slang/slang.vcxproj.filters @@ -44,6 +44,7 @@ <ClInclude Include="vm.h" /> <ClInclude Include="mangle.h" /> <ClInclude Include="legalize-types.h" /> + <ClInclude Include="ir-ssa.h" /> </ItemGroup> <ItemGroup> <ClCompile Include="check.cpp" /> @@ -73,6 +74,7 @@ <ClCompile Include="dxc-support.cpp" /> <ClCompile Include="ir-legalize-types.cpp" /> <ClCompile Include="legalize-types.cpp" /> + <ClCompile Include="ir-ssa.cpp" /> </ItemGroup> <ItemGroup> <CustomBuild Include="core.meta.slang" /> diff --git a/source/slang/vm.cpp b/source/slang/vm.cpp index ffc455232..c486615c9 100644 --- a/source/slang/vm.cpp +++ b/source/slang/vm.cpp @@ -1003,17 +1003,45 @@ void resumeThread( VMType type = decodeType(frame, &ip); UInt argCount = decodeUInt(&ip); - Int destinationBlock = decodeSInt(&ip); - for( UInt aa = 2; aa < argCount; ++aa ) + Int destinationBlockIndex = decodeSInt(&ip); + + auto func = frame->func; + auto& destinationBlock = func->bcFunc->blocks[destinationBlockIndex]; + + // We may be passing arguments through to the destination + // block, so any remaining operands of the branch instruction + // need to be used to fill in the parameter registers of + // the destination block. + + UInt paramCount = destinationBlock.paramCount; + + // There might be additional operands, because some branches + // also include information on merge points and break/continue labels. + UInt remainingArgCount = argCount - 1; + assert(remainingArgCount >= paramCount); + + UInt extraArgCount = remainingArgCount - paramCount; + + for (UInt ee = 0; ee < extraArgCount; ++ee) { decodeOperandPtr<void>(frame, &ip); } - // TODO: we need to deal with the case of - // passing arguments to the block, which - // means copying between the registers... + auto baseRegIndex = destinationBlock.params - func->bcFunc->regs; + for( UInt pp = 0; pp < paramCount; ++pp ) + { + auto regIndex = baseRegIndex + pp; + + void* argPtr = decodeOperandPtr<void>(frame, &ip); + void* regPtr = getRegPtrImpl(frame, regIndex); + + VMType regType = func->regs[regIndex].type; + memcpy(regPtr, argPtr, regType.getSize()); + } + + // Now simply jump to the destination block. // - ip = frame->func->bcFunc->blocks[destinationBlock].code; + ip = destinationBlock.code; } break; |