Merge remote-tracking branch 'EclipseRepo/dev' into feat/operator_globalAveragePooling

include/aidge/backend/cpu/operator/ReLUImpl_forward_kernels.hpp

@@ -25,7 +25,7 @@ void ReLUImpl_cpu_forward_kernel(std::size_t inputLenght,
     const I* input = static_cast<const I*>(input_);
     O* output = static_cast<O*>(output_);
 
-#pragma omp parallel for if (inputLenght > 1024)
+//#pragma omp parallel for if (inputLenght > 1024)
     for (std::size_t i = 0; i < inputLenght; ++i) {
         output[i] = input[i] > 0 ? input[i] : 0;
     }

include/aidge/backend/cpu/operator/SigmoidImpl_forward_kernels.hpp

@@ -25,7 +25,7 @@ void SigmoidImpl_cpu_forward_kernel(std::size_t inputLenght,
     const I* input = static_cast<const I*>(input_);
     O* output = static_cast<O*>(output_);
 
-#pragma omp parallel for if (inputLenght > 1024)
+//#pragma omp parallel for if (inputLenght > 1024)
     for (std::size_t i = 0; i < inputLenght; ++i) {
         output[i] = static_cast<O>(1.0) / (static_cast<O>(1.0) + std::exp(-input[i]));
     }

include/aidge/backend/cpu/operator/TanhImpl_forward_kernels.hpp

@@ -25,7 +25,7 @@ void TanhImpl_cpu_forward_kernel(std::size_t inputLenght,
     const I* input = static_cast<const I*>(input_);
     O* output = static_cast<O*>(output_);
 
-#pragma omp parallel for if (inputLenght > 1024)
+//#pragma omp parallel for if (inputLenght > 1024)
     for (std::size_t i = 0; i < inputLenght; ++i) {
         output[i] = std::tanh(input[i]);
     }

src/operator/DivImpl.cpp

@@ -91,16 +91,10 @@ void Aidge::DivImpl_cpu::forward() {
     std::size_t contiguousIdx = nbDims - 1;
     for (; contiguousIdx+1 > 0; --contiguousIdx) {
         if (dims0[contiguousIdx] != dims1[contiguousIdx]) {
-            if (contiguousIdx == (nbDims -1)) {
-                if (dims0[contiguousIdx] == 1) {
-                    while ((dims0[contiguousIdx] == 1) && (contiguousIdx+1 > 0)) {
-                        --contiguousIdx;
-                    }
-                }
-                else {
-                    while ((dims1[contiguousIdx] == 1) && (contiguousIdx+1 > 0)) {
-                        --contiguousIdx;
-                    }
+            if (contiguousIdx == (nbDims -1)) { // last dimensions of one of the input Tensor are of size 1
+                const std::vector<std::size_t>& dims = (dims0[contiguousIdx] == 1) ? dims0 : dims1;
+                while ((contiguousIdx+1 > 0) && (dims[contiguousIdx] == 1)) {
+                    --contiguousIdx;
                 }
             }
             break;

src/operator/FCImpl.cpp

@@ -57,9 +57,10 @@ void Aidge::FCImpl_cpu::forward()
     const auto& input2 = std::static_pointer_cast<Tensor>(mOp.getRawInput(2))->refCastFrom(input2Fallback, *std::static_pointer_cast<Tensor>(mOp.getRawOutput(0)));
 
     // Call kernel
+    const auto batchSize = (input0.dims().size() > 1) ? input0.dims()[0] : 1;
     kernelFunc(dynamic_cast<const FC_Op&>(mOp).getStaticAttributes(),
-        input0.dims()[0],
-        input0.size() / input0.dims()[0],
+        batchSize,
+        input0.size() / batchSize,
         input0.getImpl()->rawPtr(), input1.getImpl()->rawPtr(), input2.getImpl()->rawPtr(),
         getCPUPtr(mOp.getRawOutput(0)));
 }