diff --git a/include/aidge/quantization/QAT/QAT_LSQ.hpp b/include/aidge/quantization/QAT/QAT_LSQ.hpp
index b1e7b6fcf99a50e707da2fdc7f7c35cdb2d778f7..7919b1af10647379f11d8819d1c3583a6c1fe9cb 100644
--- a/include/aidge/quantization/QAT/QAT_LSQ.hpp
+++ b/include/aidge/quantization/QAT/QAT_LSQ.hpp
@@ -9,30 +9,30 @@
*
********************************************************************************/
- #ifndef AIDGE_QUANTIZATION_QUANTIZATION_QAT_LSQ_H_
- #define AIDGE_QUANTIZATION_QUANTIZATION_QAT_LSQ_H_
-
- #include <cstddef> // std::size_t
- #include <memory>
-
- #include "aidge/data/Tensor.hpp"
- #include "aidge/graph/GraphView.hpp"
-
- namespace Aidge {
- namespace QuantLSQ {
-
- /**
- * @brief Given a GraphView with parameters properly initialized, insert
- * the LSQ quantizer nodes, and setup the adjustment their step-sizes.
- * @param graphView The GraphView containing the network to quantize.
- * @param nbBits Number of quantization bits.
- */
-
- void setupQuantizers(std::shared_ptr<GraphView> graphView, size_t nbBits);
-
- } // namespace QuantLSQ
- } // namespace Aidge
-
- #endif /* AIDGE_QUANTIZATION_QUANTIZATION_QAT_LSQ_H_ */
+#ifndef AIDGE_QUANTIZATION_QUANTIZATION_QAT_LSQ_H_
+#define AIDGE_QUANTIZATION_QUANTIZATION_QAT_LSQ_H_
+
+#include <cstddef> // std::size_t
+#include <memory>
+
+#include "aidge/data/Tensor.hpp"
+#include "aidge/graph/GraphView.hpp"
+
+namespace Aidge {
+namespace QuantLSQ {
+
+/**
+ * @brief Given a GraphView with parameters properly initialized, insert
+ * the LSQ quantizer nodes, and setup the adjustment their step-sizes.
+ * @param graphView The GraphView containing the network to quantize.
+ * @param nbBits Number of quantization bits.
+ */
+
+void setupQuantizers(std::shared_ptr<GraphView> graphView, size_t nbBits);
+
+} // namespace QuantLSQ
+} // namespace Aidge
+
+#endif /* AIDGE_QUANTIZATION_QUANTIZATION_QAT_LSQ_H_ */
\ No newline at end of file
diff --git a/include/aidge/recipes/QuantRecipes.hpp b/include/aidge/recipes/QuantRecipes.hpp
index 39349f962d61970020741ba533403ba03559a53f..1e78699c579d53549ada884247ff545ac451f737 100644
--- a/include/aidge/recipes/QuantRecipes.hpp
+++ b/include/aidge/recipes/QuantRecipes.hpp
@@ -40,6 +40,14 @@ namespace Aidge
* @param graphView The GraphView to process.
*/
void sanitizeNodeNames(std::shared_ptr<GraphView> graphView);
+
+ /**
+ * @brief Given a GraphView, set all it's MatMul weights to index 1 (required for the PTQ)
+ * This operation involve the insertion of Transpose nodes as well as the transposition of
+ * the MatMul weight tensors.
+ * @param graphView The GraphView to process.
+ */
+ void reorderMatMulInputs(std::shared_ptr<GraphView> graphView);
}
#endif /* AIDGE_QUANTIZATION_QUANTRECIPES_H_ */
diff --git a/python_binding/recipes/pybind_QuantRecipes.cpp b/python_binding/recipes/pybind_QuantRecipes.cpp
index 0b96aef775a32cd362013998dd786a9985cc3fc1..15257b0a6b292d3205b6256fecb221ea0a7c7297 100644
--- a/python_binding/recipes/pybind_QuantRecipes.cpp
+++ b/python_binding/recipes/pybind_QuantRecipes.cpp
@@ -18,13 +18,15 @@
namespace py = pybind11;
-namespace Aidge {
-
-void init_QuantRecipes(py::module &m) {
+namespace Aidge
+{
+void init_QuantRecipes(py::module &m)
+{
m.def("pop_softmax", &popSoftMax, py::arg("network"));
m.def("insert_batchnorm_nodes", &insertBatchNormNodes, py::arg("network"));
m.def("sanitize_node_names", &sanitizeNodeNames, py::arg("network"));
+ m.def("reorder_matmul_inputs", &reorderMatMulInputs, py::arg("network"));
}
} // namespace Aidge
diff --git a/src/PTQ/CLE.cpp b/src/PTQ/CLE.cpp
index 7115a2f17726c21666306aad8f75bd51eed3eb29..5ffc8eb29a04313d91d94069eedc29697a47e4b0 100644
--- a/src/PTQ/CLE.cpp
+++ b/src/PTQ/CLE.cpp
@@ -109,11 +109,15 @@ static double getTensorAbsoluteMax(std::shared_ptr<Tensor> tensor)
return localFlatTensor.get<double>(maxIndex);
}
-//Function used to extraxt the local tensor (from a ProducerScalingNode)
-std::shared_ptr<Aidge::Tensor> getLocalTensor(std::shared_ptr<Node> node) {
- if (node->getParent(1)->attributes()->hasAttr("quantization.ptq.isProducerScaling")) {
+
+// What is this thing ???
+// Function used to extract the local tensor (from a ProducerScalingNode)
+std::shared_ptr<Aidge::Tensor> getLocalTensor(std::shared_ptr<Node> node)
+{
+ if (node->getParent(1)->attributes()->hasAttr("quantization.ptq.isProducerScaling"))
+ {
std::shared_ptr<Aidge::OperatorTensor> operatorTensor = std::static_pointer_cast<OperatorTensor>(node->getParent(1)->getOperator());
- operatorTensor->forward();// We need the forward pass to compute the scaled value of the Tensor
+ operatorTensor->forward(); // We need the forward pass to compute the scaled value of the Tensor
return operatorTensor->getOutput(0);
} else {
return getWeightTensor(node);
@@ -129,16 +133,16 @@ void crossLayerEqualization(std::shared_ptr<GraphView> graphView, double targetD
for (std::shared_ptr<Node> node : nodeVector)
{
if (node->getChildren().size() > 1) {
- Log::notice(" Network have multiple branches, skipping the CLE ... ");
+ Log::warn(" Network have multiple branches, skipping the CLE ... ");
return;
}
if (isNotQuantized(node)) {
- Log::notice(" Network contains non linear nodes, skipping the CLE ... ");
+ Log::warn(" Network contains non linear nodes, skipping the CLE ... ");
return;
}
}
- Log::info(" Applying the Cross-Layer Equalization ... ");
+ Log::notice(" Applying the Cross-Layer Equalization ... ");
// Get the vector of affine nodes
@@ -148,13 +152,14 @@ void crossLayerEqualization(std::shared_ptr<GraphView> graphView, double targetD
affineNodeVector.push_back(node);
double maxRangeDelta;
-
do
{
maxRangeDelta = 0.0;
for (size_t i = 0; i < (affineNodeVector.size() - 1); i++)
{
+ // Log::notice(" node index : {} ", i);
+
std::shared_ptr<Node> n1 = affineNodeVector[i];
std::shared_ptr<Node> n2 = affineNodeVector[i+1];
@@ -168,8 +173,12 @@ void crossLayerEqualization(std::shared_ptr<GraphView> graphView, double targetD
double s2 = std::sqrt(r1 * r2) / r2;
insertScalingBelowProducer(n1->getParent(1), s1, graphView);
+
+ if (n1->type() != "MatMul") // TODO : enhance this !
+ if (n1->getParent(2))
+ insertScalingBelowProducer(n1->getParent(2), s1, graphView);
+
insertScalingBelowProducer(n2->getParent(1), s2, graphView);
- insertScalingBelowProducer(n1->getParent(2), s1, graphView);
double rangeDelta = std::abs(r1 - r2);
if (rangeDelta > maxRangeDelta)
diff --git a/src/PTQ/PTQ.cpp b/src/PTQ/PTQ.cpp
index 05ab48594709f79099ad643b2960b03c96ef7e75..8a36e4f44fb3cdfe41981004907cc76996326ed0 100644
--- a/src/PTQ/PTQ.cpp
+++ b/src/PTQ/PTQ.cpp
@@ -29,6 +29,7 @@
#include "aidge/operator/Conv.hpp"
#include "aidge/operator/ArgMax.hpp"
#include "aidge/operator/Reshape.hpp"
+#include "aidge/operator/MatMul.hpp"
#include "aidge/recipes/Recipes.hpp"
#include "aidge/recipes/QuantRecipes.hpp"
@@ -36,9 +37,25 @@
namespace Aidge
{
-bool isAffine(std::shared_ptr<Node> node)
+static bool hasAttr(std::shared_ptr<Aidge::Node> node, std::string attr)
{
- return (affineNodeTypes.find(node->type()) != affineNodeTypes.end());
+ return node->attributes()->hasAttr("quantization.ptq." + attr);
+}
+
+static void addAttr(std::shared_ptr<Aidge::Node> node, std::string attr, double value = 0.0)
+{
+ node->attributes()->addAttr("quantization.ptq." + attr, value);
+}
+
+bool isAffine(std::shared_ptr<Node> node)
+{
+ if (affineNodeTypes.find(node->type()) != affineNodeTypes.end())
+ return true;
+
+ if ((node->type() == "MatMul") && hasAttr(node, "isWeighted"))
+ return true;
+
+ return false;
}
bool isSeamless(std::shared_ptr<Node> node)
@@ -48,7 +65,13 @@ bool isSeamless(std::shared_ptr<Node> node)
bool isMerging(std::shared_ptr<Node> node)
{
- return (mergingNodeTypes.find(node->type()) != mergingNodeTypes.end());
+ if (mergingNodeTypes.find(node->type()) != mergingNodeTypes.end())
+ return true;
+
+ if ((node->type() == "MatMul") && !hasAttr(node, "isWeighted"))
+ return true;
+
+ return false;
}
bool isNotQuantized(std::shared_ptr<Node> node)
@@ -58,14 +81,17 @@ bool isNotQuantized(std::shared_ptr<Node> node)
bool checkArchitecture(std::shared_ptr<GraphView> graphView)
{
- std::set<std::string> otherNodeTypes({"Flatten", "Softmax", "BatchNorm2D", "ReLU", "Producer"});
+ std::set<std::string> removedNodeTypes({"Flatten", "Softmax", "BatchNorm2D"});
+
+ std::set<std::string> specialNodeTypes({"MatMul", "ReLU", "Producer"});
std::set<std::string> notQuantizedNodesTypes;
for (std::shared_ptr<Node> node : graphView->getNodes())
{
- bool isOther = otherNodeTypes.find(node->type()) != otherNodeTypes.end();
- if (!isOther && !isAffine(node) && !isSeamless(node) && !isMerging(node) && !isNotQuantized(node)) {
+ bool isRemoved = removedNodeTypes.find(node->type()) != removedNodeTypes.end();
+ bool isSpecial = specialNodeTypes.find(node->type()) != specialNodeTypes.end();
+ if (!isRemoved && !isSpecial && !isAffine(node) && !isSeamless(node) && !isMerging(node) && !isNotQuantized(node)) {
Log::warn(" GraphView can't be quantized : node type {} is not supported !", node->type());
return false;
}
@@ -86,25 +112,52 @@ bool checkArchitecture(std::shared_ptr<GraphView> graphView)
void prepareNetwork(std::shared_ptr<GraphView> graphView)
{
- removeFlatten(graphView);
+ // XXX remove this !
+
sanitizeNodeNames(graphView);
- bool containsBatchNorm = false;
+ // remove the flatten nodes
+
+ removeFlatten(graphView);
+
std::vector<std::shared_ptr<Node>> nodeVector = retrieveNodeVector(graphView);
+ // tag the weighted nodes
+
for (std::shared_ptr<Node> node : nodeVector)
+ {
+ bool isWeighted = isAffine(node);
+ if (node->type() == "MatMul")
+ {
+ std::shared_ptr<Node> parent = node->getParent(1);
+ if (parent)
+ if (parent->type() == "Producer")
+ isWeighted = true;
+ }
+
+ if (isWeighted)
+ addAttr(node, "isWeighted");
+ }
+
+ // fuse the batchnorms
+
+ bool containsBatchNorm = false;
+ for (std::shared_ptr<Node> node : nodeVector) {
if (node->type() == "BatchNorm") {
containsBatchNorm = true;
break;
}
+ }
if (containsBatchNorm)
fuseBatchNorm(graphView);
+ // pop the softmax
+
popSoftMax(graphView);
}
-static std::shared_ptr<Aidge::Node> getUniqueChildren(std::shared_ptr<Aidge::Node> node)
+static std::shared_ptr<Aidge::Node> getUniqueChild(std::shared_ptr<Aidge::Node> node)
{
std::set<std::shared_ptr<Aidge::Node>> childrenSet = node->getChildren();
AIDGE_ASSERT(childrenSet.size() == 1, " Attempted to access to a unique child while the parent have multiple ones ! ");
@@ -121,8 +174,8 @@ static int getInputIndex(std::shared_ptr<Node> node, std::shared_ptr<Node> paren
void multiplyScalingFactor(std::shared_ptr<Aidge::Node> node, double coeff)
{
- AIDGE_ASSERT(node->type() == "Mul" && (node->attributes()->hasAttr("quantization.ptq.isProducerScaling") || node->attributes()->hasAttr("quantization.ptq.isScaling")),
- "Cannot update the scaling factor on Node of type {} with no scaling tag", node->type());
+ AIDGE_ASSERT(node->type() == "Mul" && hasAttr(node, "isProducerScaling") || hasAttr(node, "isScaling"),
+ "Cannot update the scaling factor on Node of type {} with no scaling tag", node->type());
auto scalingFactorTensor = std::static_pointer_cast<OperatorTensor>(node->getOperator())->getInput(1);
@@ -167,14 +220,15 @@ static void insertChildren(std::shared_ptr<Node> parent, std::shared_ptr<Node> n
bool insertRoundBelowProducer(std::shared_ptr<Node> node, std::shared_ptr<GraphView> graphView)
{
- if (node->attributes()->hasAttr("quantization.ptq.isProducerScaling") && node->type() != "Round")
+ if (hasAttr(node, "isProducerScaling") && node->type() != "Round")
{
std::shared_ptr<Aidge::Node> roundNode = Round(node->name() + "_Round");
roundNode->getOperator()->setDataType(DataType::Float64); // getDataType(parentNode)
roundNode->getOperator()->setBackend(node->getOperator()->backend());
insertChildren(node, roundNode, graphView);
- roundNode->attributes()->addAttr("quantization.ptq.isProducerRounding", 0.0);
+ addAttr(roundNode, "isProducerRounding");
+
return true;
}
return false;
@@ -233,7 +287,7 @@ static std::vector<std::shared_ptr<Node>> removeProdScalingNodes(std::vector<std
{
std::vector<std::shared_ptr<Node>> remainingNodes;
for (std::shared_ptr<Node> node : nodeVector)
- if (!node->attributes()->hasAttr("quantization.ptq.isProducerScaling"))
+ if (!hasAttr(node, "isProducerScaling"))
remainingNodes.push_back(node);
return remainingNodes;
@@ -268,25 +322,16 @@ static std::shared_ptr<Tensor> getBiasTensor(std::shared_ptr<Node> node)
std::vector<std::shared_ptr<Node>> retrieveNodeVector(std::shared_ptr<GraphView> graphView, bool newSchedule, bool verbose)
{
- std::vector<std::shared_ptr<Node>> nodeVector;
-
- SequentialScheduler scheduler(graphView);
-
- if (newSchedule)
- {
- scheduler.resetScheduling();
- scheduler.generateScheduling(); // old way : scheduler.forward();
- }
-
- nodeVector = scheduler.getSequentialStaticScheduling();
+ std::vector<std::shared_ptr<Node>> nodeVector = graphView->getOrderedNodes();
+
+ fixScheduling(nodeVector);
- fixScheduling(nodeVector);
nodeVector = removeMatchingNodes(nodeVector, "Producer");
nodeVector = removeProdScalingNodes(nodeVector);
if (verbose)
{
- Log::info("NB OF NODES = {}", nodeVector.size());
+ Log::info(" NB OF NODES = {}", nodeVector.size());
for (std::shared_ptr<Node> node : nodeVector)
Log::info("{} {}", node->type(), node->name());
}
@@ -310,8 +355,8 @@ static std::shared_ptr<Aidge::Node> createScalingNode(std::string name, std::vec
{
std::shared_ptr<Node> scalingNode = Mul(name);
- for (std::string attr : attributes)
- scalingNode->attributes()->addAttr("quantization.ptq." + attr, 0.0);
+ for (std::string a : attributes)
+ addAttr(scalingNode, a);
// Add the scaling factor as a producer of the node
@@ -327,14 +372,14 @@ static std::shared_ptr<Aidge::Node> createScalingNode(std::string name, std::vec
bool insertScalingBelowProducer(std::shared_ptr<Node> producerNode, double scalingFactor, std::shared_ptr<GraphView> graphView)
{
- if (producerNode->attributes()->hasAttr("quantization.ptq.isProducerRounding"))
+ if (hasAttr(producerNode, "isProducerRounding"))
{
// In this case we 'bump' the node to the one above him (an actual ProducerScaling)
// because the round node is not usable (only used when SSA is enabled)
producerNode = producerNode->getParent(0);
}
- if (producerNode->attributes()->hasAttr("quantization.ptq.isProducerScaling"))
+ if (hasAttr(producerNode, "isProducerScaling"))
{
// We accumulate the previous scaling factors by multiplying the SF of the ProducerScaling node
// (adding new nodes each time would make the graph unusable)
@@ -399,7 +444,7 @@ void insertResidualScalingNodes(std::shared_ptr<GraphView> graphView)
static std::shared_ptr<Node> getPreviousScalingNode(std::shared_ptr<Node> node)
{
std::shared_ptr<Node> currNode = node;
- while(!currNode->attributes()->hasAttr("quantization.ptq.isScaling"))
+ while(!hasAttr(currNode, "isScaling"))
{
if (currNode->getParents().size() == 0)
{
@@ -451,7 +496,6 @@ void insertScalingNodes(std::shared_ptr<GraphView> graphView)
graphView->insertParent(parentNode, prevScalingNode, 0, 0, 0);
graphView->add(prevScalingNode->getParent(1)); // add the scaling factor producer
}
-
}
}
}
@@ -484,7 +528,7 @@ void normalizeParameters(std::shared_ptr<GraphView> graphView)
for (std::shared_ptr<Node> node : nodeVector)
{
// Scaling nodes still have a ratio of 1, so they are seamless ...
- if (node->type() == "ReLU" || node->attributes()->hasAttr("quantization.ptq.isScaling") || isSeamless(node))
+ if (node->type() == "ReLU" || hasAttr(node, "isScaling") || isSeamless(node))
{
if (node != firstNode)
{
@@ -496,6 +540,8 @@ void normalizeParameters(std::shared_ptr<GraphView> graphView)
// Residual nodes should enter in this category but their ratio is 1 ...
if (isAffine(node))
{
+ Log::warn(" affine : {} ", node->name());
+
// Rescale the weight tensor
std::shared_ptr<Tensor> weightTensor = getWeightTensor(node);
@@ -538,37 +584,55 @@ void normalizeParameters(std::shared_ptr<GraphView> graphView)
// Revert the canceling by using the next scaling node
accumulatedRatios[node] = prevRatio;
- std::shared_ptr<Node> nextScalingNode = getUniqueChildren(node);
+ std::shared_ptr<Node> nextScalingNode = getUniqueChild(node);
multiplyScalingFactor(nextScalingNode, prevRatio);
}
if (isMerging(node))
{
- std::vector<std::shared_ptr<Node>> mergingNodes = node->getParents();
+ if (node->type() == "MatMul")
+ {
+ Log::warn(" matmul : {} ", node->name());
- // Compute the max ratio ...
+ // Multiply the input scaling factors !
- double maxRatio = 0;
- for (std::shared_ptr<Node> mergingNode : mergingNodes)
- {
- double merginNodeRatio = accumulatedRatios[mergingNode];
- if (merginNodeRatio > maxRatio)
- maxRatio = merginNodeRatio;
+ double leftRatio = accumulatedRatios[node->getParent(0)];
+ double rightRatio = accumulatedRatios[node->getParent(1)];
+
+ accumulatedRatios[node] = leftRatio * rightRatio;
}
+ else
+ {
+ // Use a maximum arbitration !
- accumulatedRatios[node] = maxRatio;
+ Log::warn(" merging : {} ", node->name());
- // Rescale the previous scaling Nodes
- for (std::shared_ptr<Node> mergingNode : mergingNodes)
- {
- double mergingNodeRatio = accumulatedRatios[mergingNode];
- double rescaling = mergingNodeRatio / maxRatio;
+ std::vector<std::shared_ptr<Node>> mergingNodes = node->getParents();
+
+ // Compute the max ratio ...
+
+ double maxRatio = 0;
+ for (std::shared_ptr<Node> mergingNode : mergingNodes)
+ {
+ double merginNodeRatio = accumulatedRatios[mergingNode];
+ if (merginNodeRatio > maxRatio)
+ maxRatio = merginNodeRatio;
+ }
- std::shared_ptr<Node> scalingNode = getPreviousScalingNode(mergingNode);
+ accumulatedRatios[node] = maxRatio;
- multiplyScalingFactor(scalingNode, 1 / rescaling);
+ // Rescale the previous scaling Nodes
+ for (std::shared_ptr<Node> mergingNode : mergingNodes)
+ {
+ double mergingNodeRatio = accumulatedRatios[mergingNode];
+ double rescaling = mergingNodeRatio / maxRatio;
+
+ std::shared_ptr<Node> scalingNode = getPreviousScalingNode(mergingNode);
+
+ multiplyScalingFactor(scalingNode, 1 / rescaling);
- accumulatedRatios[mergingNode] /= rescaling; // optional ...
+ accumulatedRatios[mergingNode] /= rescaling; // optional ...
+ }
}
}
}
@@ -591,7 +655,7 @@ std::unordered_map<std::shared_ptr<Node>, double> computeRanges(std::shared_ptr<
std::set<std::shared_ptr<Node>> nodeSet = graphView->getNodes();
for (std::shared_ptr<Node> node : nodeSet)
{
- if ((scalingNodesOnly && (node->attributes()->hasAttr("quantization.ptq.isScaling"))) || (!scalingNodesOnly && (node->type() != "Producer")))
+ if ((scalingNodesOnly && hasAttr(node, "isScaling")) || (!scalingNodesOnly && (node->type() != "Producer")))
{
std::shared_ptr<Operator> nodeOperator = node->getOperator();
std::shared_ptr<Tensor> valueTensor = std::static_pointer_cast<Tensor> (nodeOperator->getRawOutput(0));
@@ -613,7 +677,7 @@ std::unordered_map<std::shared_ptr<Node>, double> computeRanges(std::shared_ptr<
// std::shared_ptr<Node> inputNode = getFirstNode(graphView);
for (std::shared_ptr<Node> node : nodeSet)
- if ((scalingNodesOnly && (node->attributes()->hasAttr("quantization.ptq.isScaling"))) || (!scalingNodesOnly && (node->type() != "Producer")))
+ if ((scalingNodesOnly && hasAttr(node, "isScaling")) || (!scalingNodesOnly && (node->type() != "Producer")))
valueRanges.insert(std::make_pair(node, 0));
if (useCuda)
@@ -640,7 +704,7 @@ std::unordered_map<std::shared_ptr<Node>, double> computeRanges(std::shared_ptr<
std::unordered_map<std::shared_ptr<Node>, double> sampleRanges;
for (std::shared_ptr<Node> node : nodeSet)
{
- if ((scalingNodesOnly && (node->attributes()->hasAttr("quantization.ptq.isScaling"))) || (!scalingNodesOnly && (node->type() != "Producer")))
+ if ((scalingNodesOnly && hasAttr(node, "isScaling")) || (!scalingNodesOnly && (node->type() != "Producer")))
{
std::shared_ptr<Operator> nodeOperator = node->getOperator();
std::shared_ptr<Tensor> valueTensor = std::static_pointer_cast<Tensor> (nodeOperator->getRawOutput(0));
@@ -662,7 +726,7 @@ std::unordered_map<std::shared_ptr<Node>, double> computeRanges(std::shared_ptr<
for (std::shared_ptr<Node> node : nodeSet)
{
- if ((scalingNodesOnly && (node->attributes()->hasAttr("quantization.ptq.isScaling"))) || (!scalingNodesOnly && (node->type() != "Producer")))
+ if ((scalingNodesOnly && hasAttr(node, "isScaling")) || (!scalingNodesOnly && (node->type() != "Producer")))
if (sampleRanges[node] > valueRanges[node])
valueRanges[node] = sampleRanges[node];
}
@@ -708,7 +772,7 @@ void normalizeActivations(std::shared_ptr<GraphView> graphView, std::unordered_m
// Use the Scaling nodes to rescale the ranges ...
- if (node->attributes()->hasAttr("quantization.ptq.isScaling"))
+ if (hasAttr(node, "isScaling"))
{
std::shared_ptr<Node> prevNode = node->getParent(0);
@@ -732,26 +796,35 @@ void normalizeActivations(std::shared_ptr<GraphView> graphView, std::unordered_m
if (isMerging(node))
{
- std::vector<std::shared_ptr<Node>> mergingNodes = node->getParents();
-
- // Compute the max ratio ...
-
- double maxRatio = 0;
- for (std::shared_ptr<Node> mergingNode : mergingNodes)
+ if (node->type() == "MatMul")
{
- double mergingNodeRatio = accumulatedRatios[mergingNode];
- if (mergingNodeRatio > maxRatio)
- maxRatio = mergingNodeRatio;
+ double leftRatio = accumulatedRatios[node->getParent(0)];
+ double rightRatio = accumulatedRatios[node->getParent(1)];
+ accumulatedRatios[node] = leftRatio * rightRatio;
}
+ else
+ {
+ std::vector<std::shared_ptr<Node>> mergingNodes = node->getParents();
- accumulatedRatios[node] = maxRatio;
+ // Compute the max ratio ...
- for (std::shared_ptr<Node> mergingNode : mergingNodes)
- {
- double mergingNodeRatio = accumulatedRatios[mergingNode];
- std::shared_ptr<Node> scalingNode = getPreviousScalingNode(mergingNode);
- multiplyScalingFactor(scalingNode, mergingNodeRatio / maxRatio);
- // Log::notice(" SCALING NODE : {} {}", scalingNode->type(), scalingNode->name());
+ double maxRatio = 0;
+ for (std::shared_ptr<Node> mergingNode : mergingNodes)
+ {
+ double mergingNodeRatio = accumulatedRatios[mergingNode];
+ if (mergingNodeRatio > maxRatio)
+ maxRatio = mergingNodeRatio;
+ }
+
+ accumulatedRatios[node] = maxRatio;
+
+ for (std::shared_ptr<Node> mergingNode : mergingNodes)
+ {
+ double mergingNodeRatio = accumulatedRatios[mergingNode];
+ std::shared_ptr<Node> scalingNode = getPreviousScalingNode(mergingNode);
+ multiplyScalingFactor(scalingNode, mergingNodeRatio / maxRatio);
+ // Log::notice(" SCALING NODE : {} {}", scalingNode->type(), scalingNode->name());
+ }
}
}
@@ -801,7 +874,7 @@ std::unordered_map<std::shared_ptr<Node>, std::pair<bool, bool>> computeSignMap(
signMap[node].second = false;
}
- if (node->attributes()->hasAttr("quantization.ptq.isScaling"))
+ if (hasAttr(node, "isScaling"))
{
signMap[node].second = false;
@@ -848,7 +921,7 @@ std::unordered_map<std::shared_ptr<Node>, std::pair<bool, bool>> computeSignMap(
// Arbitration : Signed type wins !
for(std::shared_ptr<Node> parent : parentNodes)
{
- while (!parent->attributes()->hasAttr("quantization.ptq.isScaling"))
+ while (!hasAttr(parent, "isScaling"))
{
signMap[parent] = std::make_pair(false, false);
// We are on a branch so nodes always have 1 parent ...
@@ -956,7 +1029,7 @@ void quantizeNormalizedNetwork(std::shared_ptr<GraphView> graphView, std::uint8_
rescaling /= inputIsUnsigned ? unsignedMax : signedMax;
rescaling *= outputIsUnsigned ? unsignedMax : signedMax;
- std::shared_ptr<Node> scalingNode = getUniqueChildren(node); // TODO : assert if scalingNode is a Scaling ...
+ std::shared_ptr<Node> scalingNode = getUniqueChild(node); // TODO : assert if scalingNode is a Scaling ...
multiplyScalingFactor(scalingNode,rescaling) ;
}
@@ -971,8 +1044,12 @@ void quantizeNormalizedNetwork(std::shared_ptr<GraphView> graphView, std::uint8_
rescaling /= inputIsUnsigned ? unsignedMax : signedMax;
rescaling *= outputIsUnsigned ? unsignedMax : signedMax;
- std::shared_ptr<Node> scalingNode = getUniqueChildren(node); // TODO : assert if scalingNode is a Scaling ...
+ std::shared_ptr<Node> scalingNode = getUniqueChild(node); // TODO : assert if scalingNode is a Scaling ...
+ // TODO : double check this ...
+ if (node->type() == "MatMul")
+ rescaling /= inputIsUnsigned ? unsignedMax : signedMax;
+
multiplyScalingFactor(scalingNode, rescaling) ;
}
@@ -983,19 +1060,19 @@ void quantizeNormalizedNetwork(std::shared_ptr<GraphView> graphView, std::uint8_
std::shared_ptr<Node> prevScalingNode = node->getParent(0);
multiplyScalingFactor(prevScalingNode, rescaling);
- std::shared_ptr<Node> nextScalingNode = getUniqueChildren(node);
+ std::shared_ptr<Node> nextScalingNode = getUniqueChild(node);
multiplyScalingFactor(nextScalingNode, 1 / rescaling);
}
// Handle the Scaling Nodes ...
- if (node->attributes()->hasAttr("quantization.ptq.isScaling"))
+ if (hasAttr(node, "isScaling"))
{
// Don't touch the scalings that precede non-linearities ...
bool precedesNonLinearNode = false;
if (node->getChildren().size() == 1)
- if (isNotQuantized(getUniqueChildren(node)))
+ if (isNotQuantized(getUniqueChild(node)))
precedesNonLinearNode = true;
if (!noQuant && !precedesNonLinearNode)
@@ -1061,7 +1138,8 @@ static void insertCompensationNodes(std::shared_ptr<GraphView> graphView, std::u
std::string mulNodeName = makeUniqueName(node->name() + "_Mul", graphView);
std::shared_ptr<Node> mulNode = Mul(mulNodeName);
- mulNode->attributes()->addAttr("quantization.ptq.isCompensation", 0.0);
+ addAttr(mulNode, "isCompensation");
+
mulNode->getOperator()->setDataType(DataType::Float64); // getDataType(parentNode)
mulNode->getOperator()->setBackend(node->getOperator()->backend());
@@ -1123,7 +1201,7 @@ void performSingleShiftApproximation(std::shared_ptr<GraphView> graphView, bool
static void printScalingFactors(std::shared_ptr<GraphView> graphView)
{
for (auto node : retrieveNodeVector(graphView))
- if (node->attributes()->hasAttr("quantization.ptq.isScaling") || node->type() == "Quantizer")
+ if (hasAttr(node, "isScaling") || node->type() == "Quantizer")
{
double scalingFactor = getScalingFactor(node);
Log::info(" {:.6f} ({})", scalingFactor, node->name());
@@ -1163,6 +1241,7 @@ void quantizeNetwork(std::shared_ptr<GraphView> graphView, std::uint8_t nbBits,
Log::notice(" Inserting the scaling nodes ...");
insertScalingNodes(graphView);
+ // TODO : double check this !
crossLayerEqualization(graphView);
Log::notice(" Normalizing the parameters ...");
@@ -1171,13 +1250,9 @@ void quantizeNetwork(std::shared_ptr<GraphView> graphView, std::uint8_t nbBits,
Log::notice(" Computing the value ranges ...");
std::unordered_map<std::shared_ptr<Node>, double> valueRanges = computeRanges(graphView, inputDataSet, true, useCuda);
- //Log::info(" === RANGES (BEFORE ADJUST) ===");
-
Log::notice(" Optimizing the clipping values ...");
valueRanges = adjustRanges(clippingMode, valueRanges, nbBits, graphView, inputDataSet, useCuda, verbose);
- //Log:debug("=== RANGES (AFTER ADJUST) ===");
- //printRanges(graphView, valueRanges);
Log::notice(" Normalizing the activations ...");
normalizeActivations(graphView, valueRanges);
@@ -1229,7 +1304,7 @@ void clearBiases(std::shared_ptr<GraphView> graphView)
if (node->type() == "FC" || node->type() == "Conv2D") {
std::shared_ptr<Tensor> biasTensor = std::static_pointer_cast<OperatorTensor>(node->getOperator())->getInput(2);
//rescaleTensor(biasTensor, 0);
- insertScalingBelowProducer(node->getParent(2),0,graphView);
+ insertScalingBelowProducer(node->getParent(2), 0, graphView);
}
}
}
diff --git a/src/QAT/QAT_FixedQ.cpp b/src/QAT/QAT_FixedQ.cpp
index c182d6cc5b8402dabbf33c706ba8f406d4e6a162..8e9adb220930c4633900fff6b54a1d951eb023af 100644
--- a/src/QAT/QAT_FixedQ.cpp
+++ b/src/QAT/QAT_FixedQ.cpp
@@ -154,8 +154,8 @@ void QuantFixedQ::devQAT(std::shared_ptr<GraphView> graphView)
{
SequentialScheduler scheduler(graphView);
scheduler.generateScheduling();
- auto s = scheduler.getSequentialStaticScheduling();
- for (std::shared_ptr<Node> node : s)
+ auto nodeVector = graphView->getOrderedNodes();
+ for (std::shared_ptr<Node> node : nodeVector)
Log::info(" name : {} ", node->name());
}
diff --git a/src/QAT/QAT_LSQ.cpp b/src/QAT/QAT_LSQ.cpp
index dcac6819365e134d777be7479a95d6b8e4093b5e..6eae077b060027eb4029f6b59f55376a1674df70 100644
--- a/src/QAT/QAT_LSQ.cpp
+++ b/src/QAT/QAT_LSQ.cpp
@@ -9,164 +9,164 @@
*
********************************************************************************/
- #include "aidge/quantization/QAT/QAT_LSQ.hpp"
- #include "aidge/operator/LSQ.hpp"
- #include "aidge/operator/ReLU.hpp"
-
-
- #include "aidge/data/Tensor.hpp"
- #include "aidge/graph/GraphView.hpp"
- #include "aidge/scheduler/SequentialScheduler.hpp"
- #include "aidge/scheduler/Scheduler.hpp"
- #include "aidge/graph/Matching.hpp"
- #include "aidge/recipes/QuantRecipes.hpp"
-
-
- namespace Aidge
- {
-
- static float getTensorAbsMean(std::shared_ptr<Tensor> tensor)
- {
- auto valueTensor = (*tensor).abs().mean();
- std::shared_ptr<Tensor> fallback;
- const Tensor& localTensor = valueTensor.refCastFrom(fallback, DataType::Float32, "cpu");
- return localTensor.get<float>(0);
- }
-
- static float getTensorStd(std::shared_ptr<Tensor> tensor)
- {
- auto valueTensor = (*tensor);
-
- auto skewedTensor = valueTensor - valueTensor.mean();
- auto squaredTensor = skewedTensor * skewedTensor;
- auto varianceTensor = squaredTensor.mean();
-
- std::shared_ptr<Tensor> fallback;
- auto localTensor = varianceTensor.refCastFrom(fallback, DataType::Float32, "cpu");
-
- float variance = localTensor.get<float>(0);
- return std::sqrt(variance);
- }
-
-
- // INIT THE STEP SIZE OF A QUANTIZER NODE
-
- static bool initStepSize(std::shared_ptr<Node> quantizer)
- {
- const auto quantizerOp = std::static_pointer_cast<LSQ_Op>(quantizer->getOperator());
-
- // This formula is the one proposed in the paper ...
-
- // float inputAbsMean = getTensorAbsMean(quantizerOp->getInput(0));
- // float stepSize = 2.0f * (inputAbsMean / std::sqrt(quantizerOp->range().second));
-
- // .. but this formula seems to work better !!!
-
- float inputStd = getTensorStd(quantizerOp->getInput(0));
- float stepSize = 8.0f * (inputStd / (quantizerOp->range().second));
-
- // TODO : use the scalar constructor
- auto stepSizeTensor = std::make_shared<Tensor>(Array1D<float, 1>({{stepSize}}));
-
- // XXX Manage backend here ?
- stepSizeTensor->setBackend(quantizerOp->getInput(0)->backend());
- stepSizeTensor->setDataType(quantizerOp->getInput(0)->dataType());
-
- auto stepSizeProducer = quantizer->getParent(1);
-
- stepSizeProducer->getOperator()->setOutput(0, stepSizeTensor);
-
- Log::notice(" [ INIT STEP SIZE = {} ] ", stepSize);
-
- return false;
- }
-
- static void setupInputQuantizers(std::shared_ptr<GraphView> graphView, size_t nbBits)
- {
- const auto matches = SinglePassGraphMatching(graphView).match("(Conv2D#|PaddedConv2D#|FC#)");
-
- for (const auto& match : matches)
- {
- auto linearNode = match.graph->rootNode();
-
- // Log::notice(" SET INPUT QUANTIZER : {} ", linearNode->type());
-
- std::pair<int, int> signedRange = {-std::pow(2, nbBits - 1), std::pow(2, nbBits - 1) - 1};
- std::pair<int, int> unsignedRange = {0, std::pow(2, nbBits) - 1};
-
- // Create the input quantizer node
-
- auto quantizerName = makeUniqueName(linearNode->name() + "_lsq_i", graphView);
- auto quantizerNode = LSQ(signedRange, quantizerName);
-
- // Init the step-size using the node call stack
-
- quantizerNode->addBeforeForward([quantizerNode](){ return initStepSize(quantizerNode); });
-
- // Absorb the ReLU when possible ...
-
- bool nodeHasParent = static_cast<bool> (linearNode->getParents()[0]); // XXX is this safe ?
-
- if (nodeHasParent)
- {
- bool allParentsAreReLU = true;
- for (auto parentNode : linearNode->getParents())
- if (parentNode->type() != "ReLU")
- allParentsAreReLU = false;
-
- if (allParentsAreReLU) {
- auto quantizerOp = std::static_pointer_cast<LSQ_Op> (quantizerNode->getOperator());
- quantizerOp->range() = unsignedRange;
- }
-
- // TODO : remove the ReLUs when possible
- }
-
- // Insert the quantizer in the graphView ...
- // (We need to handle the case where the linear node is the first one)
-
- if (nodeHasParent) {
- graphView->insertParent(linearNode, quantizerNode, 0, 0, 0);
- } else {
- quantizerNode->addChild(graphView);
- graphView->add(quantizerNode);
- }
- }
- }
-
- // PARAM QUANTIZERS INSERTION
-
- static void setupParamQuantizers(std::shared_ptr<GraphView> graphView, size_t nbBits)
- {
- const auto matches = SinglePassGraphMatching(graphView).match("(Conv2D#|PaddedConv2D#|FC#)");
-
- std::pair<int, int> signedRange = {-std::pow(2, nbBits - 1), std::pow(2, nbBits - 1) - 1};
-
- for (const auto& match : matches)
- {
- auto linearNode = match.graph->rootNode();
-
- // Log::notice(" SET PARAM QUANTIZER : {} ", linearNode->type());
-
- // TODO : double check this, and use createUniqueName()
- auto quantizerName = makeUniqueName(linearNode->name() + "_lsq_p", graphView);
- auto quantizerNode = LSQ(signedRange, quantizerName);
-
- // Init the step-size using the node call stack
-
- quantizerNode->addBeforeForward([quantizerNode](){ return initStepSize(quantizerNode); });
-
- // Insert the quantizer in the graphView
-
- graphView->insertParent(linearNode, quantizerNode, 1, 0, 0);
- }
- }
-
- void QuantLSQ::setupQuantizers(std::shared_ptr<GraphView> graphView, size_t nbBits)
- {
- sanitizeNodeNames(graphView);
- setupInputQuantizers(graphView, nbBits);
- setupParamQuantizers(graphView, nbBits);
- }
-
- }
\ No newline at end of file
+#include "aidge/quantization/QAT/QAT_LSQ.hpp"
+#include "aidge/operator/LSQ.hpp"
+#include "aidge/operator/ReLU.hpp"
+
+
+#include "aidge/data/Tensor.hpp"
+#include "aidge/graph/GraphView.hpp"
+#include "aidge/scheduler/SequentialScheduler.hpp"
+#include "aidge/scheduler/Scheduler.hpp"
+#include "aidge/graph/Matching.hpp"
+#include "aidge/recipes/QuantRecipes.hpp"
+
+
+namespace Aidge
+{
+
+static float getTensorAbsMean(std::shared_ptr<Tensor> tensor)
+{
+ auto valueTensor = (*tensor).abs().mean();
+ std::shared_ptr<Tensor> fallback;
+ const Tensor& localTensor = valueTensor.refCastFrom(fallback, DataType::Float32, "cpu");
+ return localTensor.get<float>(0);
+}
+
+static float getTensorStd(std::shared_ptr<Tensor> tensor)
+{
+ auto valueTensor = (*tensor);
+
+ auto skewedTensor = valueTensor - valueTensor.mean();
+ auto squaredTensor = skewedTensor * skewedTensor;
+ auto varianceTensor = squaredTensor.mean();
+
+ std::shared_ptr<Tensor> fallback;
+ auto localTensor = varianceTensor.refCastFrom(fallback, DataType::Float32, "cpu");
+
+ float variance = localTensor.get<float>(0);
+ return std::sqrt(variance);
+}
+
+
+// INIT THE STEP SIZE OF A QUANTIZER NODE
+
+static bool initStepSize(std::shared_ptr<Node> quantizer)
+{
+ const auto quantizerOp = std::static_pointer_cast<LSQ_Op>(quantizer->getOperator());
+
+ // This formula is the one proposed in the paper ...
+
+ // float inputAbsMean = getTensorAbsMean(quantizerOp->getInput(0));
+ // float stepSize = 2.0f * (inputAbsMean / std::sqrt(quantizerOp->range().second));
+
+ // .. but this formula seems to work better !!!
+
+ float inputStd = getTensorStd(quantizerOp->getInput(0));
+ float stepSize = 8.0f * (inputStd / (quantizerOp->range().second));
+
+ // TODO : use the scalar constructor
+ auto stepSizeTensor = std::make_shared<Tensor>(Array1D<float, 1>({{stepSize}}));
+
+ // XXX Manage backend here ?
+ stepSizeTensor->setBackend(quantizerOp->getInput(0)->backend());
+ stepSizeTensor->setDataType(quantizerOp->getInput(0)->dataType());
+
+ auto stepSizeProducer = quantizer->getParent(1);
+
+ stepSizeProducer->getOperator()->setOutput(0, stepSizeTensor);
+
+ Log::notice(" [ INIT STEP SIZE = {} ] ", stepSize);
+
+ return false;
+}
+
+static void setupInputQuantizers(std::shared_ptr<GraphView> graphView, size_t nbBits)
+{
+ const auto matches = SinglePassGraphMatching(graphView).match("(Conv2D#|PaddedConv2D#|FC#)");
+
+ for (const auto& match : matches)
+ {
+ auto linearNode = match.graph->rootNode();
+
+ // Log::notice(" SET INPUT QUANTIZER : {} ", linearNode->type());
+
+ std::pair<int, int> signedRange = {-std::pow(2, nbBits - 1), std::pow(2, nbBits - 1) - 1};
+ std::pair<int, int> unsignedRange = {0, std::pow(2, nbBits) - 1};
+
+ // Create the input quantizer node
+
+ auto quantizerName = makeUniqueName(linearNode->name() + "_lsq_i", graphView);
+ auto quantizerNode = LSQ(signedRange, quantizerName);
+
+ // Init the step-size using the node call stack
+
+ quantizerNode->addBeforeForward([quantizerNode](){ return initStepSize(quantizerNode); });
+
+ // Absorb the ReLU when possible ...
+
+ bool nodeHasParent = static_cast<bool> (linearNode->getParents()[0]); // XXX is this safe ?
+
+ if (nodeHasParent)
+ {
+ bool allParentsAreReLU = true;
+ for (auto parentNode : linearNode->getParents())
+ if (parentNode->type() != "ReLU")
+ allParentsAreReLU = false;
+
+ if (allParentsAreReLU) {
+ auto quantizerOp = std::static_pointer_cast<LSQ_Op> (quantizerNode->getOperator());
+ quantizerOp->range() = unsignedRange;
+ }
+
+ // TODO : remove the ReLUs when possible
+ }
+
+ // Insert the quantizer in the graphView ...
+ // (We need to handle the case where the linear node is the first one)
+
+ if (nodeHasParent) {
+ graphView->insertParent(linearNode, quantizerNode, 0, 0, 0);
+ } else {
+ quantizerNode->addChild(graphView);
+ graphView->add(quantizerNode);
+ }
+ }
+}
+
+// PARAM QUANTIZERS INSERTION
+
+static void setupParamQuantizers(std::shared_ptr<GraphView> graphView, size_t nbBits)
+{
+ const auto matches = SinglePassGraphMatching(graphView).match("(Conv2D#|PaddedConv2D#|FC#)");
+
+ std::pair<int, int> signedRange = {-std::pow(2, nbBits - 1), std::pow(2, nbBits - 1) - 1};
+
+ for (const auto& match : matches)
+ {
+ auto linearNode = match.graph->rootNode();
+
+ // Log::notice(" SET PARAM QUANTIZER : {} ", linearNode->type());
+
+ // TODO : double check this, and use createUniqueName()
+ auto quantizerName = makeUniqueName(linearNode->name() + "_lsq_p", graphView);
+ auto quantizerNode = LSQ(signedRange, quantizerName);
+
+ // Init the step-size using the node call stack
+
+ quantizerNode->addBeforeForward([quantizerNode](){ return initStepSize(quantizerNode); });
+
+ // Insert the quantizer in the graphView
+
+ graphView->insertParent(linearNode, quantizerNode, 1, 0, 0);
+ }
+}
+
+void QuantLSQ::setupQuantizers(std::shared_ptr<GraphView> graphView, size_t nbBits)
+{
+ sanitizeNodeNames(graphView);
+ setupInputQuantizers(graphView, nbBits);
+ setupParamQuantizers(graphView, nbBits);
+}
+
+}
\ No newline at end of file
diff --git a/src/backend/cuda/operator/LSQImpl.cpp b/src/backend/cuda/operator/LSQImpl.cpp
index fa45f211e72f6742b72584aadf2a109c3bdca594..2ef9ce09ebaafc3606f91a4ae76443a79ad9240f 100644
--- a/src/backend/cuda/operator/LSQImpl.cpp
+++ b/src/backend/cuda/operator/LSQImpl.cpp
@@ -57,7 +57,7 @@ void Aidge::LSQImpl_cuda::backward() {
if (mWorkspace != nullptr) {
cudaFree(mWorkspace);
}
- CHECK_CUDA_STATUS(cudaMalloc(&mWorkspace, 8 * gra_int0->size())); // XXX This must be changed !!!
+ CHECK_CUDA_STATUS(cudaMalloc(&mWorkspace, 4 * gra_int0->size())); // XXX This must be changed !!!
mWorkspaceSize = gra_int0->size();
}
diff --git a/src/operator/PTQMetaOps.cpp b/src/operator/PTQMetaOps.cpp
index f86d454245a7fe088edd027732a91f5775cd2acf..c70a7726c143ed4cd028099f849de25a16ab11d3 100644
--- a/src/operator/PTQMetaOps.cpp
+++ b/src/operator/PTQMetaOps.cpp
@@ -70,8 +70,6 @@ static std::shared_ptr<Node> getSubNode(std::shared_ptr<GraphView> graphView, st
return mulNode;
}
-
-
void updateScalingFactor(std::shared_ptr<Node> metaOpNode, double scalingFactor)
{
if(metaOpNode->type() != "Scaling" && metaOpNode->type() != "Quantizer")
diff --git a/src/recipes/QuantRecipes.cpp b/src/recipes/QuantRecipes.cpp
index f03eb462088b16645fe600769e2a5e2c990f21b6..c1848828dd3aef121819547d1e82174185decdab 100644
--- a/src/recipes/QuantRecipes.cpp
+++ b/src/recipes/QuantRecipes.cpp
@@ -9,12 +9,17 @@
*
********************************************************************************/
-
+#include "aidge/graph/OpArgs.hpp"
+#include "aidge/operator/Producer.hpp"
#include "aidge/operator/Conv.hpp"
+#include "aidge/operator/Transpose.hpp"
+#include "aidge/operator/MatMul.hpp"
#include "aidge/operator/BatchNorm.hpp"
//#include "aidge/quantization/PTQ/PTQ.hpp"
#include "aidge/recipes/QuantRecipes.hpp"
#include "aidge/graph/Node.hpp"
+#include "aidge/operator/FC.hpp"
+#include "aidge/graph/Matching.hpp"
namespace Aidge
@@ -121,4 +126,63 @@ void sanitizeNodeNames(std::shared_ptr<GraphView> graphView)
}
}
-}
\ No newline at end of file
+void reorderMatMulInputs(std::shared_ptr<GraphView> graphView)
+{
+ const auto matches = SinglePassGraphMatching(graphView).match("(MatMul#)");
+
+ for (auto match : matches)
+ {
+ auto node = match.graph->rootNode();
+
+ // Check if the MatMul inputs have to be permuted
+
+ bool permuteInputs = false;
+
+ if (node->getParent(0))
+ if (node->getParent(0)->type() == "Producer")
+ permuteInputs = true;
+
+ if (node->getParent(1))
+ if (node->getParent(1)->type() == "Producer")
+ permuteInputs = false;
+
+ // Perform the permutation of the inputs ...
+
+ if (permuteInputs)
+ {
+ auto prevMatMul = node;
+ auto prevTensor = (std::static_pointer_cast<OperatorTensor> (node->getOperator()))->getInput(0);
+
+ // Create the new MatMul op and it's Producer
+
+ auto newMatMul = MatMul();
+
+ auto newDims = prevTensor->dims();
+ std::swap(newDims[0], newDims[1]);
+ auto newTensor = std::make_shared<Tensor>(newDims);
+
+ newTensor->setDataType(prevTensor->dataType());
+ newTensor->setBackend(prevTensor->backend());
+ newTensor->copyTranspose(*prevTensor, std::vector<Aidge::DimSize_t>({1, 0}));
+
+ auto newProducer = Producer(newTensor, "");
+ newProducer->addChild(newMatMul, 0, 1);
+
+ // Replace the node by a micrograph
+
+ auto prevMicroGraph = Sequential({prevMatMul});
+ prevMicroGraph->add(prevMatMul->getParent(0));
+
+ auto newMicroGraph = Sequential({Transpose({1, 0}), newMatMul, Transpose({1, 0})});
+ newMicroGraph->add(newMatMul->getParent(1));
+
+ graphView->replace(prevMicroGraph, newMicroGraph);
+ }
+ }
+
+ // TODO : fold the Transpose operators when possible ...
+
+ // USE REGEXPS !!!
+}
+
+}