cuda.forward.normalization.BackwardNormalizationKernel.cu Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of komputation Show documentation
Show all versions of komputation Show documentation
Komputation is a neural network framework for the JVM written in the Kotlin programming language.
#include "reduction/SumReduction.cuh"
#include "symbols/NaN.cuh"
__global__ void backwardNormalizationKernel (
int batchSize,
int numberRows,
int numberEntriesPerInstance,
int numberIterations,
float* chain,
float* forward,
float* sums,
float* result) {
extern __shared__ float sharedData[];
int indexInstance = blockIdx.x;
int indexColumn = blockIdx.y;
int startInstance = indexInstance * numberEntriesPerInstance;
int startNextInstance = startInstance + numberEntriesPerInstance;
int startColumnWithinInstance = indexColumn * numberRows;
int startEntryWithinColumn = threadIdx.x * numberIterations;
int firstEntryWithinBatch = startInstance + startColumnWithinInstance + startEntryWithinColumn;
if(firstEntryWithinBatch < startNextInstance) {
int lastEntryWithinBatch = min(firstEntryWithinBatch + numberIterations, startNextInstance);
if(indexInstance < batchSize) {
float thisValue = 0.0f;
for(int index = firstEntryWithinBatch; index < lastEntryWithinBatch; index++) {
thisValue += chain[index] * forward[index];
}
int warpId = threadIdx.x / warpSize;
int laneId = threadIdx.x % warpSize;
reduceToSum(thisValue, warpId, laneId, sharedData);
int indexColumnInBatch = indexInstance * gridDim.y + indexColumn;
for(int indexEntry = firstEntryWithinBatch; indexEntry < firstEntryWithinBatch + numberIterations; indexEntry++) {
result[indexEntry] = (-sharedData[0] + chain[indexEntry]) / sums[indexColumnInBatch];
}
}
else {
setToNan(result, firstEntryWithinBatch, lastEntryWithinBatch);
}
}
}
© 2015 - 2025 Weber Informatics LLC | Privacy Policy