diff --git a/include/aidge/backend/cpu/operator/FCImpl_kernels.hpp b/include/aidge/backend/cpu/operator/FCImpl_kernels.hpp
index ab114dda2f617b73ae9213daffd551c2dfac0f31..0d2d9e450142a91173381937d31e1ccecf2c779c 100644
--- a/include/aidge/backend/cpu/operator/FCImpl_kernels.hpp
+++ b/include/aidge/backend/cpu/operator/FCImpl_kernels.hpp
@@ -126,7 +126,6 @@ void FCImpl_cpu_backward_kernel(const DimSize_t batchSize,
void* weightGrad_,
void* biasesGrad_)
{
- Log::info("\t\t Call to FCImpl_cpu_backward_kernel");
// FIXME: missing FC attributes as arguments
const I* input = static_cast<const I*>(input_);
const I* originalInput = static_cast<const I*>(originalInput_);
@@ -136,7 +135,6 @@ void FCImpl_cpu_backward_kernel(const DimSize_t batchSize,
B* biasesGrad = static_cast<B*>(biasesGrad_);
- Log::info("\t\t Call to FCImpl_cpu_backward_kernel - Doing Bias grad");
// bias grad
if (biasesGrad == nullptr) { // no bias
// Do nothing
@@ -151,7 +149,6 @@ void FCImpl_cpu_backward_kernel(const DimSize_t batchSize,
}
}
- Log::info("\t\t Call to FCImpl_cpu_backward_kernel - Doing Weight grad");
// weight grad
for (std::size_t o = 0; o < outputFeatureSize; ++o) {
for (std::size_t c = 0; c < inputFeatureSize; ++c) {
@@ -163,7 +160,6 @@ void FCImpl_cpu_backward_kernel(const DimSize_t batchSize,
}
}
- Log::info("\t\t Call to FCImpl_cpu_backward_kernel - Doing Input grad");
// input grad
for (std::size_t b = 0; b < batchSize; ++b) {
for (std::size_t c = 0; c < inputFeatureSize; ++c) {
@@ -174,8 +170,6 @@ void FCImpl_cpu_backward_kernel(const DimSize_t batchSize,
output[b*inputFeatureSize + c]+= sum;
}
}
-
- Log::info("\t\t Call to FCImpl_cpu_backward_kernel - Done");
}
// Kernels registration to implementation entry point
diff --git a/src/operator/AddImpl.cpp b/src/operator/AddImpl.cpp
index cff6128741db657136aca1006c0f273ce64aa87a..6fb0ab3d8762a03e840f157cb9a90d9bad177891 100644
--- a/src/operator/AddImpl.cpp
+++ b/src/operator/AddImpl.cpp
@@ -77,4 +77,8 @@ void Aidge::AddImpl_cpu::backward() {
getCPUPtr(in0grad),
getCPUPtr(in1grad));
+ Log::info(" ");
+ Log::info("(AddImpl.cpp) Gradient of output 0 : {}", out0grad->toString());
+ Log::info("(AddImpl.cpp) Gradient of input 0 : {}", in0grad->toString());
+ Log::info("(AddImpl.cpp) Gradient of input 1 : {}", in1grad->toString());
}
diff --git a/src/operator/FCImpl.cpp b/src/operator/FCImpl.cpp
index 53502be1525d19ac105cd2afa046b68c75835a02..841d3aee1f2a1b922441baf4c72387ff46073105 100644
--- a/src/operator/FCImpl.cpp
+++ b/src/operator/FCImpl.cpp
@@ -55,7 +55,6 @@ void Aidge::FCImpl_cpu::forward()
template <>
void Aidge::FCImpl_cpu::backward()
{
- Log::notice("\t Call to Aidge::FCIMpl_cpu::backward()");
const FC_Op& op_ = dynamic_cast<const FC_Op&>(mOp);
const auto& fc_grad = op_.getOutput(0)->grad();
AIDGE_ASSERT(fc_grad, "missing ouput #0 gradient");
@@ -63,7 +62,6 @@ void Aidge::FCImpl_cpu::backward()
AIDGE_ASSERT(op_.getInput(1)->grad(), "missing input #1 gradient");
const auto impl = Registrar<FCImpl_cpu>::create(getBestMatch(getRequiredSpec()));
- Log::notice("\t Call to Aidge::FCIMpl_cpu::backward() - Found impl in registrar");
// Convert input data (no overhead if not needed!)
// TODO: right now, if needed, memory will be allocated/deallocated at each
@@ -74,9 +72,13 @@ void Aidge::FCImpl_cpu::backward()
const auto& input1grad = op_.getInput(1)->grad()->refCastFrom(input1gradFallback, *(op_.getOutput(0)));
const auto& input2grad = (op_.getInput(2)) ? op_.getInput(2)->grad()->refCastFrom(input2gradFallback, *(op_.getOutput(0))) : Tensor();
+ Log::info(" ");
+ Log::info("(FCImpl.cpp) Gradient of output 0 : {}", fc_grad->toString());
+ Log::info("(FCImpl.cpp) Gradient of input 0 : {}", input0grad.toString());
+ Log::info("(FCImpl.cpp) Gradient of input 1 : {}", input1grad.toString());
+
// Call kernel
const auto batchSize = (input0grad.dims().size() > 1) ? input0grad.dims()[0] : 1;
- Log::notice("\t Call to Aidge::FCIMpl_cpu::backward() - Retrieved grad pointers");
impl.backward(batchSize,
input1grad.dims()[1], // nb input features
input1grad.dims()[0], // nb output features
@@ -87,5 +89,7 @@ void Aidge::FCImpl_cpu::backward()
input1grad.getImpl()->rawPtr(),
(op_.getInput(2)) ? input2grad.getImpl()->rawPtr() : nullptr);
- Log::notice("\t Call to Aidge::FCIMpl_cpu::backward() - Done");
+ Log::info("(FCImpl.cpp) Gradient of output 0 : {}", fc_grad->toString());
+ Log::info("(FCImpl.cpp) Gradient of input 0 : {}", input0grad.toString());
+ Log::info("(FCImpl.cpp) Gradient of input 1 : {}", input1grad.toString());
}
diff --git a/src/operator/MulImpl.cpp b/src/operator/MulImpl.cpp
index b8bdd161af25a87937b066fc2264ecf1a9816da0..b09165633fa28dd5bbaf7080e0f18cc6ee63b966 100644
--- a/src/operator/MulImpl.cpp
+++ b/src/operator/MulImpl.cpp
@@ -41,7 +41,6 @@ void Aidge::MulImpl_cpu::forward() {
template <>
void Aidge::MulImpl_cpu::backward() {
- Log::info("MulImpl_cpu::backward()");
const Mul_Op& op_ = dynamic_cast<const Mul_Op&>(mOp);
auto in0 = op_.getInput(0);
@@ -59,17 +58,15 @@ void Aidge::MulImpl_cpu::backward() {
/* grad0Length */ out0grad->size(),
/* input0Dims */ in0->dims(),
/* input1Dims */ in1->dims(),
- out0grad->dims(),
- getCPUPtr(in0),
- getCPUPtr(in1),
- getCPUPtr(out0grad),
+ /* outputDims */ out0grad->dims(),
+ /* input0_ */ getCPUPtr(in0),
+ /* input1_ */ getCPUPtr(in1),
+ /* grad_output_ */ getCPUPtr(out0grad),
getCPUPtr(in0grad),
getCPUPtr(in1grad));
- Log::info("MulImpl_cpu::backward() : Gradient of output 0");
- out0grad->print();
- Log::info("MulImpl_cpu::backward() : Gradient of input 0");
- in0grad->print();
- Log::info("MulImpl_cpu::backward() : Gradient of input 1");
- in1grad->print();
+ Log::info(" ");
+ Log::info("(MulImpl_cpu::backward()) After Mul Backward pass.");
+ Log::info("(MulImpl_cpu::backward()) Gradient of input 0 : {}", in0grad->toString());
+ Log::info("(MulImpl_cpu::backward()) Gradient of input 1 : {}", in1grad->toString());
}
diff --git a/src/operator/SubImpl.cpp b/src/operator/SubImpl.cpp
index 3601675206e5a86cf9c35316b5dd3daa102c59bd..8dc32f1638c936e368e58341f352ad9072293e9f 100644
--- a/src/operator/SubImpl.cpp
+++ b/src/operator/SubImpl.cpp
@@ -43,10 +43,10 @@ template <>
void Aidge::SubImpl_cpu::backward() {
const Sub_Op& op_ = dynamic_cast<const Sub_Op&>(mOp);
- //Log::info("SubImpl_cpu::backward() : Node {}", op_.name());
auto in0 = op_.getInput(0);
auto in1 = op_.getInput(1);
+ auto out0 = op_.getOutput(0);
auto in0grad = op_.getInput(0)->grad();
auto in1grad = op_.getInput(1)->grad();
auto out0grad = op_.getOutput(0)->grad();
@@ -55,9 +55,11 @@ void Aidge::SubImpl_cpu::backward() {
const auto impl = Registrar<SubImpl_cpu>::create(getBestMatch(getRequiredSpec()));
- in0grad->print();
- in1grad->print();
- out0grad->print();
+ Log::info(" ");
+ Log::info("(SubImpl.cpp) Gradient of input 0 : {}", in0grad->toString());
+ Log::info("(SubImpl.cpp) Gradient of input 1 : {}", in1grad->toString());
+ Log::info("(SubImpl.cpp) Gradient of output 0 : {}", out0grad->toString());
+
// Call kernel
impl.backward(/* input0Length */ in0grad->size(),
@@ -71,10 +73,7 @@ void Aidge::SubImpl_cpu::backward() {
/* gradInput1 */ getCPUPtr(in1grad));
- Log::info("SubImpl_cpu::backward() : Gradient of output 0");
- out0grad->print();
- Log::info("SubImpl_cpu::backward() : Gradient of input 0");
- in0grad->print();
- Log::info("SubImpl_cpu::backward() : Gradient of input 1");
- in1grad->print();
+ Log::info("(SubImpl.cpp) After backward.");
+ Log::info("(SubImpl.cpp) Gradient of input 0 : {}", in0grad->toString());
+ Log::info("(SubImpl.cpp) Gradient of input 1 : {}", in1grad->toString());
}
diff --git a/unit_tests/operator/Test_Memorize.cpp b/unit_tests/operator/Test_Memorize.cpp
index 8f6b30420aa84393280c08422c00b7162fa15da8..c3e1576ab258edbbec68e0aa63f844ed5a7ba801 100644
--- a/unit_tests/operator/Test_Memorize.cpp
+++ b/unit_tests/operator/Test_Memorize.cpp
@@ -25,7 +25,6 @@
#include "aidge/graph/OpArgs.hpp"
#include "aidge/operator/Add.hpp"
#include "aidge/operator/Memorize.hpp"
-#include "aidge/operator/PerMemorize.hpp"
#include "aidge/operator/Producer.hpp"
#include "aidge/recipes/GraphViewHelper.hpp"
#include "aidge/scheduler/SequentialScheduler.hpp"
@@ -169,86 +168,4 @@ TEST_CASE("[cpu/operator] Memorize(backward)", "[Memorize][CPU]") {
}
}
-TEST_CASE("[cpu/operator] Memorize(forward)", "[Memorize][Periodic]") {
- auto mul = Mul();
-
- //auto mem = Memorize(/*endStep=*/3);
- //auto mem = Identity();
- auto mem = PerMemorize(2);
-
- auto add = Add();
- auto pop = Pop();
- auto stack = Stack(3);
-
- // Initialization tensor for Memorize
- auto input = Producer(std::make_shared<Tensor>(Array1D<float,2>({1.0f, 1.0f})), "input");
- auto decay = Producer(std::make_shared<Tensor>(Array1D<float,2>({0.9f, 0.9f})), "decay");
- auto init = Producer(std::make_shared<Tensor>(Array1D<float,2>({0.0f, 0.0f})), "init");
- auto back = std::make_shared<Tensor>(Array1D<float,2>({1.0f, 1.0f}));
-
- auto initTensor = std::make_shared<Tensor>(Array2D<float,3,2>({
- {
- {1,1},
- {1,1},
- {1,1},
- }
- }));
-
- std::static_pointer_cast<OperatorTensor>(pop->getOperator())->setInput(0, initTensor);
-
- auto memOp = std::static_pointer_cast<OperatorTensor>(mem->getOperator());
- memOp->setOutput(1, std::make_shared<Tensor>(Array1D<float,2>({0.0f, 0.0f})));
- memOp->setOutput(0, std::make_shared<Tensor>(Array1D<float,2>({0.0f, 0.0f})));
-
- //init->addChild(mem,0,0);
- mem->addChild(mul,1,0);
- decay->addChild(mul,0,1);
- mul->addChild(add,0,1);
- pop->addChild(add,0,0);
- add->addChild(mem,0,0);
- mem->addChild(stack,1,0);
-
- auto graphView = getConnectedGraphView(mem);
- graphView->compile();
-
- Log::info("GraphView output nodes : {}", graphView->outputNodes().size());
- for(auto node : graphView->outputNodes())
- {
- Log::info("output node type : {}", node->type());
- }
- // TODO: Set ordered outputs for this node.
-
- auto scheduler = SequentialScheduler(graphView);
- scheduler.forward();
-
- //std::static_pointer_cast<OperatorTensor>(mem->getOperator())->getOutput(0)->print();
- std::static_pointer_cast<OperatorTensor>(stack->getOperator())->getOutput(0)->print();
- REQUIRE(true);
-}
-
-TEST_CASE("[cpu/operator] Memorize(forward)", "[Memorize][Periodic2]") {
-
-
- auto input = Producer(std::make_shared<Tensor>(1.0f));
- auto init = Producer(std::make_shared<Tensor>(1.0f));
- auto add = Add();
-
- //auto mem = PerMemorize(3);
- auto mem = Memorize(3);
-
- input->addChild(add,0,0);
- add->addChild(mem,0,0);
- mem->addChild(add, 1,1);
- init->addChild(mem, 0, 1);
-
- auto gv = getConnectedGraphView(mem);
- gv->compile();
-
- auto scheduler = SequentialScheduler(gv);
- scheduler.forward();
-
- std::static_pointer_cast<OperatorTensor>(mem->getOperator())->getOutput(0)->print();
-
- REQUIRE(true);
-}
} // namespace Aidge
diff --git a/unit_tests/operator/Test_MetaOperator.cpp b/unit_tests/operator/Test_MetaOperator.cpp
index 4a4dc22dafe6cbb5a880a985df3c7690a67e77a9..3a3930fb793f6f3f10f038fbf4368a8f9bddc4e4 100644
--- a/unit_tests/operator/Test_MetaOperator.cpp
+++ b/unit_tests/operator/Test_MetaOperator.cpp
@@ -708,7 +708,8 @@ TEST_CASE("[cpu/operator] MetaOperator", "[MetaOperator][CPU]") {
auto fc2 = FC(outChannels, inChannels, true, "fc2");
// NOTE: Account for init step by adding 1 to the max timestep
// parameter.
- auto lif1 = Leaky(nbTimeSteps + 1, beta, threshold, LeakyReset::Subtraction, "leaky");
+ //auto lif1 = Leaky(nbTimeSteps + 1, beta, threshold, LeakyReset::Subtraction, "leaky");
+ auto lif1 = Leaky(nbTimeSteps, beta, threshold, LeakyReset::Subtraction, "leaky");
// associateInput() does not work
fc1->input(1).first->getOperator()->setOutput(0, myWeights);
@@ -858,7 +859,7 @@ TEST_CASE("[cpu/operator] MetaOperator", "[Leaky][CPU][Simple]") {
constexpr auto inChannels = 4;
constexpr auto outChannels = 4;
- constexpr auto beta = 1.0;
+ constexpr auto beta = 0.8;
constexpr auto threshold = 1.0;
constexpr auto nbTimeSteps = 2;
@@ -892,6 +893,7 @@ TEST_CASE("[cpu/operator] MetaOperator", "[Leaky][CPU][Simple]") {
auto lif1 = Leaky(nbTimeSteps, beta, threshold, LeakyReset::Subtraction, "leaky");
auto lif1Op = std::static_pointer_cast<OperatorTensor>(lif1->getOperator());
+
auto fc1Op = std::static_pointer_cast<OperatorTensor>(fc1->getOperator());
//fc1Op->associateInput(0, input);
@@ -913,10 +915,24 @@ TEST_CASE("[cpu/operator] MetaOperator", "[Leaky][CPU][Simple]") {
graph->compile("cpu", DataType::Float32);
graph->forwardDims();
+ Log::info("Lif 1 op tensors : \
+ \n Input 0 : {} \
+ \n Input 1 : {} \
+ \n Output 0 : {} \
+ \n Output 1 : {}",
+ static_cast<void*>(lif1Op->getInput(0).get()),
+ static_cast<void*>(lif1Op->getInput(1).get()),
+ static_cast<void*>(lif1Op->getOutput(0).get()),
+ static_cast<void*>(lif1Op->getOutput(1).get())
+ );
+
auto scheduler = SequentialScheduler(graph);
REQUIRE_NOTHROW(scheduler.generateScheduling());
REQUIRE_NOTHROW(scheduler.forward(true));
+ // Modify graph here -- COuld also be checked by type
+ auto mGraph = graph->getNode("leaky");
+
// Print output
//Log::notice("FC1 op output 0");
//fc1Op->getOutput(0)->print();