better MaterializeMergeSchedule.cpp (something still broken downstream)

src/PIM/Conversion/SpatialToPim/BatchCoreLoweringPatterns.cpp

@@ -1,6 +1,8 @@
+#include "mlir/Dialect/Arith/IR/Arith.h"
 #include "mlir/Dialect/Bufferization/IR/Bufferization.h"
 #include "mlir/Dialect/MemRef/IR/MemRef.h"
 #include "mlir/Dialect/Tensor/IR/Tensor.h"
+#include "mlir/Dialect/Func/IR/FuncOps.h"
 #include "mlir/IR/IRMapping.h"
 #include "mlir/IR/Matchers.h"
 
@@ -97,20 +99,73 @@ static LogicalResult lowerChannelReceiveTensorBatch(spatial::SpatChannelReceiveT
   return success();
 }
 
+static FailureOr<unsigned> getDirectReturnOperandIndex(OpResult result) {
+  if (!result.hasOneUse())
+    return failure();
+
+  auto returnOp = dyn_cast<func::ReturnOp>(*result.getUsers().begin());
+  if (!returnOp)
+    return failure();
+  return result.getUses().begin()->getOperandNumber();
+}
+
+static Value createScaledIndexValue(IRRewriter& rewriter, Location loc, Value base, int64_t scale) {
+  if (scale == 1)
+    return base;
+
+  auto scaleValue = arith::ConstantIndexOp::create(rewriter, loc, scale).getResult();
+  return arith::MulIOp::create(rewriter, loc, base, scaleValue).getResult();
+}
+
+static Value createHostTargetOffset(IRRewriter& rewriter,
+                                    tensor::ParallelInsertSliceOp insertSlice,
+                                    ShapedType destinationType,
+                                    IRMapping& mapper) {
+  int64_t elementBytes = destinationType.getElementTypeBitWidth() / 8;
+  SmallVector<int64_t> strides(destinationType.getRank(), 1);
+  ArrayRef<int64_t> shape = destinationType.getShape();
+  for (int64_t dim = destinationType.getRank() - 2; dim >= 0; --dim)
+    strides[dim] = strides[dim + 1] * shape[dim + 1];
+
+  Value totalOffset;
+  Location loc = insertSlice.getLoc();
+  for (auto [dim, offset] : llvm::enumerate(insertSlice.getMixedOffsets())) {
+    int64_t scale = strides[dim] * elementBytes;
+    Value scaledOffset;
+    if (auto attr = dyn_cast<Attribute>(offset)) {
+      auto intAttr = dyn_cast<IntegerAttr>(attr);
+      assert(intAttr && "expected integer offset attribute");
+      scaledOffset = arith::ConstantIndexOp::create(rewriter, loc, intAttr.getInt() * scale).getResult();
+    }
+    else {
+      scaledOffset = createScaledIndexValue(rewriter, loc, mapper.lookupOrDefault(cast<Value>(offset)), scale);
+    }
+
+    totalOffset = totalOffset ? arith::AddIOp::create(rewriter, loc, totalOffset, scaledOffset).getResult()
+                              : scaledOffset;
+  }
+
+  if (!totalOffset)
+    totalOffset = arith::ConstantIndexOp::create(rewriter, loc, 0).getResult();
+  return totalOffset;
+}
+
 } // namespace
 
 LogicalResult
 lowerComputeBatchOp(spatial::SpatComputeBatch computeBatchOp, CoreLoweringState& state, IRRewriter& rewriter) {
   Location loc = computeBatchOp.getLoc();
   Block& oldBlock = computeBatchOp.getBody().front();
-  if (computeBatchOp.getNumResults() != 0)
-    return computeBatchOp.emitOpError(
-      "batched Spatial-to-PIM lowering currently requires channelized compute_batch with no results; "
-      "materialize explicit communication before lowering to PIM");
-
   auto oldYield = dyn_cast<spatial::SpatYieldOp>(oldBlock.getTerminator());
-  if (!oldYield || oldYield.getNumOperands() != 0)
-    return computeBatchOp.emitOpError("resultless compute_batch lowering requires empty spat.yield");
+  auto inParallelOp = dyn_cast<spatial::SpatInParallelOp>(oldBlock.getTerminator());
+  if (computeBatchOp.getNumResults() == 0) {
+    if (!oldYield || oldYield.getNumOperands() != 0)
+      return computeBatchOp.emitOpError("resultless compute_batch lowering requires empty spat.yield");
+  }
+  else if (!inParallelOp) {
+    return computeBatchOp.emitOpError(
+      "resultful compute_batch lowering currently requires a spat.in_parallel terminator");
+  }
 
   SmallVector<int32_t> coreIds = getPimCoreIdsForBatchOp(computeBatchOp, state.nextCoreId);
   SmallVector<Value> batchWeights(computeBatchOp.getWeights().begin(), computeBatchOp.getWeights().end());
@@ -128,9 +183,24 @@ lowerComputeBatchOp(spatial::SpatComputeBatch computeBatchOp, CoreLoweringState&
     {static_cast<int>(batchWeights.size()), static_cast<int>(batchInputs.size())});
   coreBatchOp->setAttr(onnx_mlir::kCoreIdsAttrName, rewriter.getDenseI32ArrayAttr(coreIds));
 
+  SmallVector<Value> hostOutputTensors;
+  if (computeBatchOp.getNumResults() != 0) {
+    hostOutputTensors.resize(computeBatchOp.getNumResults());
+    for (auto [resultIndex, result] : llvm::enumerate(computeBatchOp.getResults())) {
+      FailureOr<unsigned> returnOperandIndex = getDirectReturnOperandIndex(cast<OpResult>(result));
+      if (failed(returnOperandIndex))
+        return computeBatchOp.emitOpError(
+          "resultful compute_batch lowering currently requires each result to be used directly by func.return");
+
+      hostOutputTensors[resultIndex] = state.outputTensors[*returnOperandIndex](rewriter, loc);
+      result.replaceAllUsesWith(hostOutputTensors[resultIndex]);
+    }
+  }
+
   SmallVector<Type> blockArgTypes;
   SmallVector<Location> blockArgLocs;
-  for (BlockArgument arg : oldBlock.getArguments()) {
+  unsigned inputArgLimit = 1 + computeBatchOp.getWeights().size() + computeBatchOp.getInputs().size();
+  for (BlockArgument arg : oldBlock.getArguments().take_front(inputArgLimit)) {
     blockArgTypes.push_back(arg.getType());
     blockArgLocs.push_back(arg.getLoc());
   }
@@ -183,6 +253,38 @@ lowerComputeBatchOp(spatial::SpatComputeBatch computeBatchOp, CoreLoweringState&
     if (isa<spatial::SpatYieldOp>(op))
       continue;
 
+    if (auto parallelOp = dyn_cast<spatial::SpatInParallelOp>(op)) {
+      unsigned firstOutputArg = computeBatchOp.getOutputArgument(0).getArgNumber();
+      for (Operation& nestedOp : parallelOp.getRegion().front()) {
+        auto insertSlice = dyn_cast<tensor::ParallelInsertSliceOp>(&nestedOp);
+        if (!insertSlice)
+          return parallelOp.emitOpError("expected only tensor.parallel_insert_slice in spat.in_parallel");
+
+        auto outputArg = dyn_cast<BlockArgument>(insertSlice.getDest());
+        if (!outputArg || outputArg.getOwner() != &oldBlock)
+          return insertSlice.emitOpError("expected compute_batch output block argument destination");
+
+        unsigned resultIndex = outputArg.getArgNumber() - firstOutputArg;
+        if (resultIndex >= hostOutputTensors.size())
+          return insertSlice.emitOpError("result index out of range while lowering host batch output");
+
+        Value mappedSource = mapper.lookup(insertSlice.getSource());
+        auto hostTarget = hostOutputTensors[resultIndex];
+        auto hostTargetType = cast<ShapedType>(hostTarget.getType());
+        Value hostTargetOffset = createHostTargetOffset(rewriter, insertSlice, hostTargetType, mapper);
+        Value zeroOffset = arith::ConstantIndexOp::create(rewriter, insertSlice.getLoc(), 0).getResult();
+        pim::PimMemCopyDevToHostOp::create(rewriter,
+                                           insertSlice.getLoc(),
+                                           hostTarget.getType(),
+                                           hostTargetOffset,
+                                           zeroOffset,
+                                           hostTarget,
+                                           mappedSource,
+                                           getTensorSizeInBytesAttr(rewriter, mappedSource));
+      }
+      continue;
+    }
+
     if (auto sendBatchOp = dyn_cast<spatial::SpatChannelSendBatchOp>(op)) {
       FailureOr<SmallVector<int32_t>> targetCoreIds = getConstantI32Values(sendBatchOp.getTargetCoreIds());
       if (failed(targetCoreIds))

src/PIM/Conversion/SpatialToPim/CMakeLists.txt

@@ -6,7 +6,6 @@ add_pim_library(OMSpatialToPim
   SpatialToPimPass.cpp
   BatchCoreLoweringPatterns.cpp
   ChannelLoweringPatterns.cpp
-  Cleanup.cpp
   Common.cpp
   ComputeLikeRegionUtils.cpp
   CoreLoweringPatterns.cpp

src/PIM/Conversion/SpatialToPim/Cleanup.cpp

@@ -1,42 +0,0 @@
-#include "llvm/ADT/STLExtras.h"
-
-#include "src/Accelerators/PIM/Conversion/SpatialToPim/Cleanup.hpp"
-
-using namespace mlir;
-
-namespace onnx_mlir {
-
-LogicalResult erasePendingOps(SmallVectorImpl<Operation*>& pendingOps, IRRewriter& rewriter) {
-  while (!pendingOps.empty()) {
-    bool erasedAnyOp = false;
-    for (auto it = pendingOps.begin(); it != pendingOps.end();) {
-      Operation* opToRemove = *it;
-      if (!opToRemove->use_empty()) {
-        ++it;
-        continue;
-      }
-
-      rewriter.eraseOp(opToRemove);
-      it = pendingOps.erase(it);
-      erasedAnyOp = true;
-    }
-
-    if (erasedAnyOp)
-      continue;
-
-    for (Operation* opToRemove : pendingOps) {
-      InFlightDiagnostic diag = opToRemove->emitError("pending Spatial-to-PIM cleanup could not erase operation");
-      diag << "; op has " << llvm::range_size(opToRemove->getUsers()) << " remaining user(s)";
-      for (Operation* user : opToRemove->getUsers()) {
-        bool userPendingRemoval = llvm::is_contained(pendingOps, user);
-        opToRemove->emitRemark() << "remaining user `" << user->getName() << "`"
-                                 << (userPendingRemoval ? " is also pending removal" : " is not pending removal");
-      }
-    }
-    return failure();
-  }
-
-  return success();
-}
-
-} // namespace onnx_mlir

src/PIM/Conversion/SpatialToPim/Cleanup.hpp

@@ -1,11 +0,0 @@
-#pragma once
-
-#include "mlir/IR/Operation.h"
-#include "mlir/IR/PatternMatch.h"
-#include "mlir/Support/LLVM.h"
-
-namespace onnx_mlir {
-
-mlir::LogicalResult erasePendingOps(llvm::SmallVectorImpl<mlir::Operation*>& pendingOps, mlir::IRRewriter& rewriter);
-
-} // namespace onnx_mlir

src/PIM/Conversion/SpatialToPim/GlobalTensorMaterialization.cpp

@@ -141,152 +141,6 @@ struct MoveExtractSliceIntoCompute final : OpRewritePattern<mlir::tensor::Extrac
   }
 };
 
-// Turns runtime constants consumed by compute regions into private globals and local loads.
-struct ArithConstToGlobalMemoryPattern final : OpRewritePattern<mlir::arith::ConstantOp> {
-  using OpRewritePattern::OpRewritePattern;
-
-  LogicalResult matchAndRewrite(mlir::arith::ConstantOp constantOp, PatternRewriter& rewriter) const override {
-    Location loc = constantOp.getLoc();
-
-    if (hasWeightAlways(constantOp))
-      return failure();
-
-    if (!isa<func::FuncOp>(constantOp->getParentOp()))
-      return failure();
-
-    if (llvm::all_of(constantOp->getUsers(), [](Operation* op) {
-          if (isa<spatial::SpatCompute>(op))
-            return false;
-          if (isa<func::FuncOp>(op->getParentOp()))
-            return true;
-          return false;
-        }))
-      return failure();
-
-    rewriter.setInsertionPoint(constantOp->getParentOfType<func::FuncOp>());
-
-    auto constRankedTensorType = llvm::dyn_cast<mlir::RankedTensorType>(constantOp.getType());
-
-    if (constRankedTensorType) {
-      mlir::MemRefType memRefType =
-        mlir::MemRefType::get(constRankedTensorType.getShape(), constRankedTensorType.getElementType());
-      auto globalOp = createPrivateMemrefGlobalWithUniqueName(rewriter,
-                                                              loc,
-                                                              constantOp->getParentOfType<ModuleOp>(),
-                                                              "const",
-                                                              memRefType,
-                                                              constantOp.getValueAttr(),
-                                                              rewriter.getUnitAttr());
-      std::string argName = globalOp.getSymName().str();
-
-      llvm::DenseMap<Operation*, Value> mapSpatComputeToConst;
-
-      for (auto& constUses : llvm::make_early_inc_range(constantOp->getUses())) {
-        auto constUsers = constUses.getOwner();
-
-        if (auto spatCompute = llvm::dyn_cast<spatial::SpatCompute>(constUsers)) {
-          auto inputIndex = getDirectComputeLikeInputIndex(spatCompute, constUses.getOperandNumber());
-          if (!inputIndex)
-            return failure();
-          auto BBArgIndex = *inputIndex;
-          rewriter.setInsertionPoint(&spatCompute.getBody().front().front());
-          if (!mapSpatComputeToConst.contains(spatCompute.getOperation())) {
-            auto getGlobalOp = memref::GetGlobalOp::create(rewriter, loc, memRefType, argName);
-            auto toTensor = bufferization::ToTensorOp::create(
-              rewriter, loc, constRankedTensorType, getGlobalOp, rewriter.getUnitAttr(), rewriter.getUnitAttr());
-            mapSpatComputeToConst.insert({spatCompute.getOperation(), toTensor.getResult()});
-          }
-
-          replaceAndEraseDirectComputeLikeInput(
-            rewriter, spatCompute.getOperation(), BBArgIndex, mapSpatComputeToConst[spatCompute.getOperation()]);
-        }
-        else if (auto spatComputeBatch = llvm::dyn_cast<spatial::SpatComputeBatch>(constUsers)) {
-          auto inputIndex = getDirectComputeLikeInputIndex(spatComputeBatch, constUses.getOperandNumber());
-          if (!inputIndex)
-            return failure();
-          auto BBArgIndex = *inputIndex;
-          rewriter.setInsertionPoint(&spatComputeBatch.getBody().front().front());
-          if (!mapSpatComputeToConst.contains(spatComputeBatch.getOperation())) {
-            auto getGlobalOp = memref::GetGlobalOp::create(rewriter, loc, memRefType, argName);
-            auto toTensor = bufferization::ToTensorOp::create(
-              rewriter, loc, constRankedTensorType, getGlobalOp, rewriter.getUnitAttr(), rewriter.getUnitAttr());
-            mapSpatComputeToConst.insert({spatComputeBatch.getOperation(), toTensor.getResult()});
-          }
-
-          replaceAndEraseDirectComputeLikeInput(rewriter,
-                                                spatComputeBatch.getOperation(),
-                                                BBArgIndex,
-                                                mapSpatComputeToConst[spatComputeBatch.getOperation()]);
-        }
-        else {
-          {
-
-            if (auto spatCompute = constUses.getOwner()->getParentOfType<spatial::SpatCompute>()) {
-              rewriter.setInsertionPoint(&spatCompute.getBody().front().front());
-              if (!mapSpatComputeToConst.contains(spatCompute.getOperation())) {
-                auto getGlobalOp = memref::GetGlobalOp::create(rewriter, loc, memRefType, argName);
-                auto toTensor = bufferization::ToTensorOp::create(
-                  rewriter, loc, constRankedTensorType, getGlobalOp, rewriter.getUnitAttr(), rewriter.getUnitAttr());
-                mapSpatComputeToConst.insert({spatCompute.getOperation(), toTensor.getResult()});
-              }
-
-              rewriter.startOpModification(spatCompute.getOperation());
-              constUses.set(mapSpatComputeToConst[spatCompute.getOperation()]);
-              rewriter.finalizeOpModification(spatCompute.getOperation());
-            }
-            else if (auto spatComputeBatch = constUses.getOwner()->getParentOfType<spatial::SpatComputeBatch>()) {
-              rewriter.setInsertionPoint(&spatComputeBatch.getBody().front().front());
-              if (!mapSpatComputeToConst.contains(spatComputeBatch.getOperation())) {
-                auto getGlobalOp = memref::GetGlobalOp::create(rewriter, loc, memRefType, argName);
-                auto toTensor = bufferization::ToTensorOp::create(
-                  rewriter, loc, constRankedTensorType, getGlobalOp, rewriter.getUnitAttr(), rewriter.getUnitAttr());
-                mapSpatComputeToConst.insert({spatComputeBatch.getOperation(), toTensor.getResult()});
-              }
-
-              rewriter.startOpModification(spatComputeBatch.getOperation());
-              constUses.set(mapSpatComputeToConst[spatComputeBatch.getOperation()]);
-              rewriter.finalizeOpModification(spatComputeBatch.getOperation());
-            }
-          }
-        }
-      }
-    }
-    else if (constantOp.getType().isIntOrIndexOrFloat()) {
-      Value hostConstant = constantOp.getResult();
-
-      for (auto& constUses : llvm::make_early_inc_range(constantOp->getUses())) {
-        auto constUsers = constUses.getOwner();
-
-        if (auto spatCompute = llvm::dyn_cast<spatial::SpatCompute>(constUsers)) {
-          auto inputIndex = getDirectComputeLikeInputIndex(spatCompute, constUses.getOperandNumber());
-          if (!inputIndex)
-            return failure();
-          auto BBArgIndex = *inputIndex;
-          replaceAndEraseDirectComputeLikeInput(rewriter, spatCompute.getOperation(), BBArgIndex, hostConstant);
-        }
-        else if (auto spatComputeBatch = llvm::dyn_cast<spatial::SpatComputeBatch>(constUsers)) {
-          auto inputIndex = getDirectComputeLikeInputIndex(spatComputeBatch, constUses.getOperandNumber());
-          if (!inputIndex)
-            return failure();
-          auto BBArgIndex = *inputIndex;
-          replaceAndEraseDirectComputeLikeInput(rewriter, spatComputeBatch.getOperation(), BBArgIndex, hostConstant);
-        }
-        else if (constUsers->getParentOfType<spatial::SpatCompute>()) {
-          constUses.set(hostConstant);
-        }
-        else {
-          auto batchParent = constUsers->getParentOfType<spatial::SpatComputeBatch>();
-          assert(batchParent && "Global Constant used direcly not within a compute");
-          constUses.set(hostConstant);
-        }
-      }
-    }
-    if (constantOp->use_empty())
-      rewriter.eraseOp(constantOp);
-    return success();
-  }
-};
-
 // Materializes public function tensor inputs as globals so compute bodies can load them uniformly.
 struct FuncOpArgToGlobalMemoryPattern final : OpRewritePattern<mlir::func::FuncOp> {
   using OpRewritePattern::OpRewritePattern;
@@ -363,7 +217,7 @@ struct FuncOpArgToGlobalMemoryPattern final : OpRewritePattern<mlir::func::FuncO
 
 } // namespace
 void populateGlobalTensorMaterializationPatterns(RewritePatternSet& patterns) {
-  patterns.add<MoveExtractSliceIntoCompute, FuncOpArgToGlobalMemoryPattern, ArithConstToGlobalMemoryPattern>(
+  patterns.add<MoveExtractSliceIntoCompute, FuncOpArgToGlobalMemoryPattern>(
     patterns.getContext());
 }
 

src/PIM/Conversion/SpatialToPim/SpatialToPimPass.cpp

@@ -14,7 +14,6 @@
 #include "mlir/IR/Value.h"
 #include "mlir/Pass/Pass.h"
 #include "mlir/Transforms/FoldUtils.h"
-#include "mlir/Transforms/GreedyPatternRewriteDriver.h"
 #include "mlir/Transforms/WalkPatternRewriteDriver.h"
 
 #include "llvm/ADT/StringRef.h"
@@ -28,7 +27,6 @@
 #include "src/Accelerators/PIM/Common/PimCommon.hpp"
 #include "src/Accelerators/PIM/Conversion/SpatialToPim/BatchCoreLoweringPatterns.hpp"
 #include "src/Accelerators/PIM/Conversion/SpatialToPim/ChannelLoweringPatterns.hpp"
-#include "src/Accelerators/PIM/Conversion/SpatialToPim/Cleanup.hpp"
 #include "src/Accelerators/PIM/Conversion/SpatialToPim/Common.hpp"
 #include "src/Accelerators/PIM/Conversion/SpatialToPim/CoreLoweringPatterns.hpp"
 #include "src/Accelerators/PIM/Conversion/SpatialToPim/GlobalTensorMaterialization.hpp"
@@ -67,6 +65,7 @@ private:
   LogicalResult allocateAndInitializeCoreLocalVariables(func::FuncOp funcOp, IRRewriter& rewriter);
 
   void markOpToRemove(Operation* op);
+  void eraseOpsToRemove();
 
   void enlargeVMMOutTensorsToCrossbarSize(func::FuncOp funcOp, IRRewriter& rewriter);
 };
@@ -268,13 +267,7 @@ void SpatialToPimPass::runOnOperation() {
 
   enlargeVMMOutTensorsToCrossbarSize(funcOp, rewriter);
   replaceReturnWithOutputBuffers(returnOp, rewriter, returnPathState);
-
-  SmallVector<Operation*> pendingRemovals(operationsToRemove.begin(), operationsToRemove.end());
-  if (failed(erasePendingOps(pendingRemovals, rewriter))) {
-    funcOp.emitOpError("failed to erase obsolete Spatial ops after lowering to PIM");
-    signalPassFailure();
-    return;
-  }
+  eraseOpsToRemove();
 
   RewritePatternSet finalTensorPackingPatterns(ctx);
   populateTensorPackingPatterns(finalTensorPackingPatterns);
@@ -399,6 +392,13 @@ void SpatialToPimPass::markOpToRemove(Operation* op) {
     operationsToRemove.push_back(op);
 }
 
+void SpatialToPimPass::eraseOpsToRemove() {
+  for (Operation* op : operationsToRemove) {
+    op->dropAllUses();
+    op->erase();
+  }
+}
+
 std::unique_ptr<Pass> createSpatialToPimPass() { return std::make_unique<SpatialToPimPass>(); }
 
 } // namespace onnx_mlir