add shared loop creation helpers

add shared checked arithmetic helpers
refactor pim passes into Pim/Transforms
more robust memory coalescing pass

src/PIM/Conversion/SpatialToPim/SpatialToPimPass.cpp

@@ -25,8 +25,9 @@
 #include <cassert>
 #include <utility>
 
-#include "Common/PimCommon.hpp"
 #include "Common/IR/ConstantUtils.hpp"
+#include "Common/PimCommon.hpp"
+#include "Common/Support/CheckedArithmetic.hpp"
 #include "Conversion/ONNXToSpatial/Common/Common.hpp"
 #include "Conversion/SpatialToPim/Common.hpp"
 #include "Conversion/SpatialToPim/Patterns.hpp"
@@ -75,21 +76,28 @@ static memref::GlobalOp getOrCreateZeroGlobal(IRRewriter& rewriter, Location loc
                                   IntegerAttr {});
 }
 
-static Value createZeroedDeviceHVector(IRRewriter& rewriter,
-                                       Location loc,
-                                       RankedTensorType tensorType,
-                                       OperationFolder& constantFolder) {
+static FailureOr<Value> createZeroedDeviceHVector(IRRewriter& rewriter,
+                                                  Location loc,
+                                                  RankedTensorType tensorType,
+                                                  OperationFolder& constantFolder) {
   auto outputBuffer = createEmptyTensorFromShaped(rewriter, loc, tensorType);
   auto zeroGlobal = getOrCreateZeroGlobal(rewriter, loc, tensorType);
   auto zeroValue = memref::GetGlobalOp::create(rewriter, loc, zeroGlobal.getType(), zeroGlobal.getName());
   auto zeroIndex = getOrCreateIndexConstant(constantFolder, outputBuffer.getOperation(), 0);
-  auto sizeAttr = rewriter.getI32IntegerAttr(static_cast<int32_t>(getShapedTypeSizeInBytes(tensorType)));
+  auto byteSize =
+    pim::getCheckedShapedTypeSizeInBytes(tensorType, outputBuffer.getOperation(), "host-to-device zero copy byte size");
+  if (failed(byteSize))
+    return failure();
+  auto sizeAttr =
+    pim::getCheckedI32Attr(rewriter, outputBuffer.getOperation(), *byteSize, "host-to-device zero copy byte size");
+  if (failed(sizeAttr))
+    return failure();
   return PimMemCopyHostToDevOp::create(
-           rewriter, loc, tensorType, zeroIndex, zeroIndex, outputBuffer, zeroValue, sizeAttr)
+           rewriter, loc, tensorType, zeroIndex, zeroIndex, outputBuffer, zeroValue, *sizeAttr)
     .getOutput();
 }
 
-static Value
+static FailureOr<Value>
 padHVectorInputToCrossbarSize(IRRewriter& rewriter, Location loc, Value vector, OperationFolder& constantFolder) {
   auto vectorType = cast<RankedTensorType>(vector.getType());
   ArrayRef<int64_t> shape = vectorType.getShape();
@@ -101,10 +109,18 @@ padHVectorInputToCrossbarSize(IRRewriter& rewriter, Location loc, Value vector,
 
   auto paddedType = RankedTensorType::get(
     {shape[0], static_cast<int64_t>(crossbarSize)}, vectorType.getElementType(), vectorType.getEncoding());
-  Value zeroed = createZeroedDeviceHVector(rewriter, loc, paddedType, constantFolder);
-  Value zeroIndex = getOrCreateIndexConstant(constantFolder, zeroed.getDefiningOp(), 0);
-  auto sizeAttr = rewriter.getI32IntegerAttr(static_cast<int32_t>(getShapedTypeSizeInBytes(vectorType)));
-  return PimMemCopyOp::create(rewriter, loc, paddedType, zeroIndex, zeroIndex, zeroed, vector, sizeAttr).getOutput();
+  auto zeroed = createZeroedDeviceHVector(rewriter, loc, paddedType, constantFolder);
+  if (failed(zeroed))
+    return failure();
+  Value zeroIndex = getOrCreateIndexConstant(constantFolder, zeroed->getDefiningOp(), 0);
+  auto byteSize =
+    pim::getCheckedShapedTypeSizeInBytes(vectorType, zeroed->getDefiningOp(), "device padding copy byte size");
+  if (failed(byteSize))
+    return failure();
+  auto sizeAttr = pim::getCheckedI32Attr(rewriter, zeroed->getDefiningOp(), *byteSize, "device padding copy byte size");
+  if (failed(sizeAttr))
+    return failure();
+  return PimMemCopyOp::create(rewriter, loc, paddedType, zeroIndex, zeroIndex, *zeroed, vector, *sizeAttr).getOutput();
 }
 
 void onnx_mlir::raptor::SpatialToPimPass::runOnOperation() {
@@ -234,7 +250,11 @@ void onnx_mlir::raptor::SpatialToPimPass::runOnOperation() {
     }
   }
 
-  enlargeVMMOutTensorsToCrossbarSize(funcOp, rewriter);
+  if (failed(enlargeVMMOutTensorsToCrossbarSize(funcOp, rewriter))) {
+    funcOp.emitOpError("failed to enlarge VMM output tensors to crossbar size");
+    signalPassFailure();
+    return;
+  }
   replaceReturnWithOutputBuffers(returnOp, rewriter);
   eraseOpsToRemove();
 
@@ -271,8 +291,9 @@ void onnx_mlir::raptor::SpatialToPimPass::runOnOperation() {
   dumpModule(moduleOp, "pim0");
 }
 
-void raptor::SpatialToPimPass::enlargeVMMOutTensorsToCrossbarSize(func::FuncOp funcOp, IRRewriter& rewriter) {
+LogicalResult raptor::SpatialToPimPass::enlargeVMMOutTensorsToCrossbarSize(func::FuncOp funcOp, IRRewriter& rewriter) {
   OperationFolder constantFolder(funcOp.getContext());
+  bool hasFailure = false;
   funcOp.walk([&](PimVMMOp vmmOp) {
     auto outputType = cast<RankedTensorType>(vmmOp.getOutput().getType());
     ArrayRef<int64_t> outputShape = outputType.getShape();
@@ -280,19 +301,23 @@ void raptor::SpatialToPimPass::enlargeVMMOutTensorsToCrossbarSize(func::FuncOp f
     assert(outputShape[1] <= static_cast<int64_t>(crossbarSize) && "output width must fit in one crossbar");
 
     rewriter.setInsertionPoint(vmmOp);
-    Value paddedInput = padHVectorInputToCrossbarSize(rewriter, vmmOp.getLoc(), vmmOp.getInput(), constantFolder);
+    auto paddedInput = padHVectorInputToCrossbarSize(rewriter, vmmOp.getLoc(), vmmOp.getInput(), constantFolder);
+    if (failed(paddedInput)) {
+      hasFailure = true;
+      return WalkResult::interrupt();
+    }
     auto paddedOutputType = RankedTensorType::get(
       {outputShape[0], static_cast<int64_t>(crossbarSize)}, outputType.getElementType(), outputType.getEncoding());
     Value paddedOutputBuffer = outputShape[1] == static_cast<int64_t>(crossbarSize)
                                ? vmmOp.getOutputBuffer()
                                : createEmptyTensorFromShaped(rewriter, vmmOp.getLoc(), paddedOutputType).getResult();
-    vmmOp.getInputMutable().assign(paddedInput);
+    vmmOp.getInputMutable().assign(*paddedInput);
     vmmOp.getOutputBufferMutable().assign(paddedOutputBuffer);
 
     vmmOp.getOutput().setType(paddedOutputType);
 
     if (outputShape[1] == static_cast<int64_t>(crossbarSize))
-      return;
+      return WalkResult::advance();
 
     SmallVector<OpFoldResult> offsets = {rewriter.getIndexAttr(0), rewriter.getIndexAttr(0)};
     SmallVector<OpFoldResult> sizes = {rewriter.getIndexAttr(outputShape[0]), rewriter.getIndexAttr(outputShape[1])};
@@ -302,13 +327,16 @@ void raptor::SpatialToPimPass::enlargeVMMOutTensorsToCrossbarSize(func::FuncOp f
       tensor::ExtractSliceOp::create(rewriter, vmmOp.getLoc(), outputType, vmmOp.getOutput(), offsets, sizes, strides);
     SmallPtrSet<Operation*, 2> exceptions = {vmmOp, sliceOp};
     vmmOp.getOutput().replaceAllUsesExcept(sliceOp.getResult(), exceptions);
+    return WalkResult::advance();
   });
+  return success(!hasFailure);
 }
 
 LogicalResult raptor::SpatialToPimPass::allocateAndInitializeCoreLocalVariables(func::FuncOp funcOp,
                                                                                 IRRewriter& rewriter) {
   Location loc = funcOp.getLoc();
   OperationFolder constantFolder(funcOp.getContext());
+  bool hasFailure = false;
 
   auto insertMemCopyHostToDev = [&](Value inputTensor, int64_t elementsOffset) {
     auto tensorType = cast<ShapedType>(inputTensor.getType());
@@ -319,17 +347,28 @@ LogicalResult raptor::SpatialToPimPass::allocateAndInitializeCoreLocalVariables(
     rewriter.setInsertionPointAfter(inputTensor.getDefiningOp());
 
     auto deviceTensor = tensor::EmptyOp::create(rewriter, loc, tensorType.getShape(), elementType);
+    auto offsetBytes = pim::checkedMul(
+      static_cast<size_t>(elementsOffset), elementByteSize, deviceTensor.getOperation(), "host input byte offset");
+    auto byteSize =
+      pim::getCheckedShapedTypeSizeInBytes(tensorType, deviceTensor.getOperation(), "host input copy byte size");
+    auto sizeAttr =
+      succeeded(byteSize)
+        ? pim::getCheckedI32Attr(rewriter, deviceTensor.getOperation(), *byteSize, "host input copy byte size")
+        : FailureOr<IntegerAttr>(failure());
+    if (failed(offsetBytes) || failed(sizeAttr)) {
+      hasFailure = true;
+      return;
+    }
 
     auto memCopyHostToDevOp = PimMemCopyHostToDevOp::create(
       rewriter,
       loc,
       tensorType,
       getOrCreateIndexConstant(constantFolder, deviceTensor.getOperation(), 0),
-      getOrCreateIndexConstant(
-        constantFolder, deviceTensor.getOperation(), static_cast<int64_t>(elementsOffset * elementByteSize)),
+      getOrCreateIndexConstant(constantFolder, deviceTensor.getOperation(), static_cast<int64_t>(*offsetBytes)),
       deviceTensor,
       inputTensor,
-      rewriter.getI32IntegerAttr(static_cast<int32_t>(tensorType.getNumElements() * elementByteSize)));
+      *sizeAttr);
 
     rewriter.replaceAllUsesExcept(inputTensor, memCopyHostToDevOp.getResult(), {memCopyHostToDevOp});
   };
@@ -347,7 +386,7 @@ LogicalResult raptor::SpatialToPimPass::allocateAndInitializeCoreLocalVariables(
       }
     }
 
-  return success();
+  return success(!hasFailure);
 }
 
 void raptor::SpatialToPimPass::markOpToRemove(Operation* op) {