Brutex
/
cavis
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
cavis/libnd4j/blas/cpu/NDArrayLambda.hpp

325 lines
18 KiB
C++
Raw Normal View History

Dev branch merge: dev_20190606 (#7904) * correct logsoftmax looss (#2) * Small SameDiff listener fix (#4) * Various fixes (#6) * #7839 Fix for asXMatrix and tests * #7866 EmbeddingSequenceLayer dtype fix + test * #7856 SameDiff save/load stream methods * #7859 RegressionEvaluation rank 4 fix + tests + axis configuration * EvaluationBinary 3d/4d * More evaluation 3d/4d tests * #7847 Evaluation empty checks * Small test ifx * #7848 Fix median edge case * Improve DL4J samediff layer tests * [WIP] FastText wrapper implemented (#8) * FastText implemented * Some fixes * Fix shapes for wordsNearest * Validation of input vectors * Fixes * Fixed test * Thread tagged * Some tweaks * setContextClassLoader for DeallocatorServiceThread * Numpy format tests (#1) * Various fixes (#11) * #7852 SameDiff gather fix * #7892 SameDiff placeholder to constant conversion * #7890 validate input rank for MLN/CG init methods * Fix broken permute shape calculation * Permute and gather fixes * Tests * #7850 LogSumExp fix + test * Handful of test fixes * Empty arrays with non-scalar shapes (#10) * minor rearrangements for lambdas * empty tensors with non-scalar shapes * numpy empty tensors with non-scalar shapes * few more empty tweaks * Small fixes * conv3d signature update * micro fix in batchnorm mkldnn * Import fixes * Fix * MKL-DNN update * Small fill fix * fill with empty input + test * Fixes * Small error improvement * Fix * one special test * couple of fixes for lstm * Rewrite TFGraphMapper.getNDArrayFromTensor to be maintainable and less error prone * Fixes * FP16 * Unsigned * BFloat16 * Fill op - empty tweaks * - couple of fixes for empty arrays construction - stack updated * strided slice fix * one transform test * provide method for reducing shapeInfo in case of input array is empty * Fixed reduceAlongDimensions to use empty input properly. * couple of broadcast tests * couple of tests broadcast tests + tweak to make them pass * add check of non-empty to methods producing sub-arrays * Fixed reshapeC with zeros in shape. * complete empty check in reduce_... legacy ops * Concat and cumsum/prod * Tweak to empty shape inference on import * add empty check to the rest of reduce legacy ops * one more test * correct typo in evalReduceShapeInfoEmpty * Added tests for reduce_* ops to tests with zero shapes. * few more tests for empty reductions * Fixed strided_slice op with empty case and tests. * one more empty reduction test * Fixed strided_slice test. * add empty check to NDArray::reshapei * infOrMax * empty min/max with infinity tests * made unstack working correctly with empty arrays * few IndexReduce tests + tweaks for empty shapes * add test for empty concat * few tests fixed * Validation fix for reductions on empty shapes * Reverse fix * Reduction shape calc fixes * SameDiff.generateOutputVariable: don't use shape function to determine number of outputs * Range fix * - NDArray constructor updated for scalars/empty arrays - few tests fixed * More fixes * Empty creator fixes * concat fix * concat fix * TF import tests: allow 'both all NaN' and 'both all inf' to pass * Slice, zero fraction, and reshape fixes * transpose, gather * Zero fraction * scalar cast fix * Empty reduction axis support * few more tests fixed * Fixed input checks conforming with TF for concat op and tests. * few tests fixed * matmul scalar shape fix * Fixed checkout for data type and scalarity with concat to allow non-empty scalars with vector concats. * broadcast bool fix * few more tests * few more tests * correct evalReduceShapeInfoEmpty * argmax/argmin + tests * one more empty edge case + one more test * argmax/argmin/realdiv_bp tweaks * empty reshape test + fix * Helper fixes * Small fixes * Gather test fix * Gather test fix * Small fixes * reduce scalar zero values * scalar mean workaround * Remove debug code * along dim mean workaround * one more test * - equalsTo() tweak for empty arrays - one more test * broadcast tweaks
2019-06-15 13:34:34 +02:00
template<typename T>
void NDArray::applyTriplewiseLambda(NDArray* second, NDArray *third, const std::function<T(T, T, T)>& func, NDArray* target) {
if (target == nullptr)
target = this;
if (second == nullptr) {
nd4j_printf("applyTriplewiseLambda requires three operands to be valid NDArrays, but Second is NULL\n","");
throw std::runtime_error("second is null");
}
if (third == nullptr) {
nd4j_printf("applyTriplewiseLambda requires three operands to be valid NDArrays, but Third is NULL\n","");
throw std::runtime_error("third is null");
}
if(dataType() != DataTypeUtils::fromT<T>())
throw std::runtime_error("NDArray::applyTriplewiseLambda<T> method: wrong template parameter T, its type should be the same as type of this array!");
if(dataType() != second->dataType() || dataType() != third->dataType() || dataType() != target->dataType())
throw std::runtime_error("NDArray::applyTriplewiseLambda<T> method: bother four arrays (this, second, third, target) should have the same type !");
if (this->lengthOf() != second->lengthOf() || this->lengthOf() != third->lengthOf() || !this->isSameShape(second) || !this->isSameShape(third)) {
nd4j_printf("applyPairwiseLambda requires both operands to have the same shape\n","");
throw std::runtime_error("Shapes mismach");
}
auto f = this->bufferAsT<T>();
auto s = second->bufferAsT<T>();
auto t = third->bufferAsT<T>();
auto z = target->bufferAsT<T>();
if (this->ordering() == second->ordering() && this->ordering() == third->ordering() && this->ordering() == target->ordering() && (this->ews() == 1 && target->ews() == 1) && this->ews() == second->ews() && this->ews() == third->ews()) {
PRAGMA_OMP_PARALLEL_FOR_SIMD
for (Nd4jLong e = 0; e < _length; e++)
z[e] = func(f[e], s[e], t[e]);
} else {
if (f == z) {
PRAGMA_OMP_PARALLEL_FOR_SIMD
for (int e = 0; e < _length; e++) {
auto tOffset = this->getOffset(e);
auto uOffset = second->getOffset(e);
auto vOffset = third->getOffset(e);
f[tOffset] = func(f[tOffset], s[uOffset], t[vOffset]);
}
} else {
PRAGMA_OMP_PARALLEL_FOR_SIMD
for (int e = 0; e < _length; e++) {
auto tOffset = this->getOffset(e);
auto uOffset = second->getOffset(e);
auto vOffset = third->getOffset(e);
auto zOffset = target->getOffset(e);
z[zOffset] = func(f[tOffset], s[uOffset], t[vOffset]);
}
}
}
}
template void NDArray::applyTriplewiseLambda(NDArray* second, NDArray *third, const std::function<double (double, double, double)>& func, NDArray* target);
template void NDArray::applyTriplewiseLambda(NDArray* second, NDArray *third, const std::function<float (float, float, float)>& func, NDArray* target);
template void NDArray::applyTriplewiseLambda(NDArray* second, NDArray *third, const std::function<float16 (float16, float16, float16)>& func, NDArray* target);
template void NDArray::applyTriplewiseLambda(NDArray* second, NDArray *third, const std::function<bfloat16 (bfloat16, bfloat16, bfloat16)>& func, NDArray* target);
template void NDArray::applyTriplewiseLambda(NDArray* second, NDArray *third, const std::function<Nd4jLong (Nd4jLong, Nd4jLong, Nd4jLong)>& func, NDArray* target);
template void NDArray::applyTriplewiseLambda(NDArray* second, NDArray *third, const std::function<int (int, int, int)>& func, NDArray* target);
template void NDArray::applyTriplewiseLambda(NDArray* second, NDArray *third, const std::function<int16_t (int16_t, int16_t, int16_t)>& func, NDArray* target);
template void NDArray::applyTriplewiseLambda(NDArray* second, NDArray *third, const std::function<uint8_t (uint8_t, uint8_t, uint8_t)>& func, NDArray* target);
template void NDArray::applyTriplewiseLambda(NDArray* second, NDArray *third, const std::function<uint16_t (uint16_t, uint16_t, uint16_t)>& func, NDArray* target);
template void NDArray::applyTriplewiseLambda(NDArray* second, NDArray *third, const std::function<uint32_t (uint32_t, uint32_t, uint32_t)>& func, NDArray* target);
template void NDArray::applyTriplewiseLambda(NDArray* second, NDArray *third, const std::function<uint64_t (uint64_t, uint64_t, uint64_t)>& func, NDArray* target);
template void NDArray::applyTriplewiseLambda(NDArray* second, NDArray *third, const std::function<int8_t (int8_t, int8_t, int8_t)>& func, NDArray* target);
template void NDArray::applyTriplewiseLambda(NDArray* second, NDArray *third, const std::function<bool (bool, bool, bool)>& func, NDArray* target);
//////////////////////////////////////////////////////////////////////////
template<typename T>
void NDArray::applyPairwiseLambda(const NDArray* other, const std::function<T(T, T)>& func, NDArray* target) {
if (target == nullptr)
target = this;
if (other == nullptr) {
nd4j_printf("applyPairwiseLambda requires both operands to be valid NDArrays, but Y is NULL\n","");
throw std::runtime_error("Other is null");
}
if(dataType() != DataTypeUtils::fromT<T>())
throw std::runtime_error("NDArray::applyPairwiseLambda<T> method: wrong template parameter T, its type should be the same as type of this array!");
if(dataType() != other->dataType() || dataType() != target->dataType())
throw std::runtime_error("NDArray::applyPairwiseLambda<T> method: all three arrays (this, other, target) must have the same type !");
if (this->lengthOf() != other->lengthOf()) {
nd4j_printf("applyPairwiseLambda requires both operands to have the same shape\n","");
throw std::runtime_error("Shapes mismach");
}
auto f = this->bufferAsT<T>();
auto s = other->bufferAsT<T>();
auto z = target->bufferAsT<T>();
if (this->ordering() == other->ordering() && this->ordering() == target->ordering() && (this->ews() == 1 && target->ews() == 1) && this->ews() == other->ews()) {
PRAGMA_OMP_PARALLEL_FOR_SIMD
for (int e = 0; e < _length; e++)
z[e] = func(f[e], s[e]);
} else {
if (f == z) {
PRAGMA_OMP_PARALLEL_FOR_SIMD
for (int e = 0; e < _length; e++) {
auto xOffset = this->getOffset(e);
auto yOffset = other->getOffset(e);
f[xOffset] = func(f[xOffset], s[yOffset]);
}
} else {
PRAGMA_OMP_PARALLEL_FOR_SIMD
for (int e = 0; e < _length; e++) {
auto xOffset = this->getOffset(e);
auto yOffset = other->getOffset(e);
auto zOffset = target->getOffset(e);
z[zOffset] = func(f[xOffset], s[yOffset]);
}
}
}
}
template void NDArray::applyPairwiseLambda(const NDArray* other, const std::function<double (double, double)>& func, NDArray* target);
template void NDArray::applyPairwiseLambda(const NDArray* other, const std::function<float (float, float)>& func, NDArray* target);
template void NDArray::applyPairwiseLambda(const NDArray* other, const std::function<float16 (float16, float16)>& func, NDArray* target);
template void NDArray::applyPairwiseLambda(const NDArray* other, const std::function<bfloat16 (bfloat16, bfloat16)>& func, NDArray* target);
template void NDArray::applyPairwiseLambda(const NDArray* other, const std::function<Nd4jLong (Nd4jLong, Nd4jLong)>& func, NDArray* target);
template void NDArray::applyPairwiseLambda(const NDArray* other, const std::function<int (int, int)>& func, NDArray* target);
template void NDArray::applyPairwiseLambda(const NDArray* other, const std::function<int16_t (int16_t, int16_t)>& func, NDArray* target);
template void NDArray::applyPairwiseLambda(const NDArray* other, const std::function<uint8_t (uint8_t, uint8_t)>& func, NDArray* target);
template void NDArray::applyPairwiseLambda(const NDArray* other, const std::function<uint16_t (uint16_t, uint16_t)>& func, NDArray* target);
template void NDArray::applyPairwiseLambda(const NDArray* other, const std::function<uint32_t (uint32_t, uint32_t)>& func, NDArray* target);
template void NDArray::applyPairwiseLambda(const NDArray* other, const std::function<uint64_t (uint64_t, uint64_t)>& func, NDArray* target);
template void NDArray::applyPairwiseLambda(const NDArray* other, const std::function<int8_t (int8_t, int8_t)>& func, NDArray* target);
template void NDArray::applyPairwiseLambda(const NDArray* other, const std::function<bool (bool, bool)>& func, NDArray* target);
//////////////////////////////////////////////////////////////////////////
template<typename T>
void NDArray::applyLambda(const std::function<T(T)>& func, NDArray* target) {
if (target == nullptr)
target = this;
if(dataType() != DataTypeUtils::fromT<T>())
throw std::runtime_error("NDArray::applyLambda<T> method: wrong template parameter T, its type should be the same as type of this array!");
if(dataType() != target->dataType())
throw std::runtime_error("NDArray::applyLambda<T> method: types of this and target array should match !");
auto f = this->bufferAsT<T>();
auto z = target->bufferAsT<T>();
if (this->ordering() == target->ordering() && (this->ews() == 1 && target->ews() == 1)) {
PRAGMA_OMP_PARALLEL_FOR_SIMD
for (int e = 0; e < _length; e++)
z[e] = func(f[e]);
} else {
if (f == z) {
PRAGMA_OMP_PARALLEL_FOR_SIMD
for (int e = 0; e < _length; e++) {
auto xOffset = this->getOffset(e);
f[xOffset] = func(f[xOffset]);
}
} else {
PRAGMA_OMP_PARALLEL_FOR_SIMD
for (int e = 0; e < _length; e++) {
auto xOffset = this->getOffset(e);
auto zOffset = target->getOffset(e);
z[zOffset] = func(f[xOffset]);
}
}
}
}
template void NDArray::applyLambda(const std::function<double(double)>& func, NDArray* target);
template void NDArray::applyLambda(const std::function<float(float)>& func, NDArray* target);
template void NDArray::applyLambda(const std::function<float16(float16)>& func, NDArray* target);
template void NDArray::applyLambda(const std::function<bfloat16(bfloat16)>& func, NDArray* target);
template void NDArray::applyLambda(const std::function<Nd4jLong(Nd4jLong)>& func, NDArray* target);
template void NDArray::applyLambda(const std::function<int16_t(int16_t)>& func, NDArray* target);
template void NDArray::applyLambda(const std::function<int32_t(int32_t)>& func, NDArray* target);
template void NDArray::applyLambda(const std::function<uint8_t(uint8_t)>& func, NDArray* target);
template void NDArray::applyLambda(const std::function<uint16_t(uint16_t)>& func, NDArray* target);
template void NDArray::applyLambda(const std::function<uint32_t(uint32_t)>& func, NDArray* target);
template void NDArray::applyLambda(const std::function<uint64_t(uint64_t)>& func, NDArray* target);
template void NDArray::applyLambda(const std::function<int8_t(int8_t)>& func, NDArray* target);
template void NDArray::applyLambda(const std::function<bool(bool)>& func, NDArray* target);
//////////////////////////////////////////////////////////////////////////
template<typename T>
void NDArray::applyIndexedLambda(const std::function<T(Nd4jLong, T)>& func, NDArray* target) {
if (target == nullptr)
target = this;
if(dataType() != DataTypeUtils::fromT<T>())
throw std::runtime_error("NDArray::applyIndexedLambda<T> method: wrong template parameter T, its type should be the same as type of this array!");
if(dataType() != target->dataType())
throw std::runtime_error("NDArray::applyIndexedLambda<T> method: types of this and target array should match !");
auto f = this->bufferAsT<T>();
auto z = target->bufferAsT<T>();
if (this->ordering() == target->ordering() && (this->ews() == 1 && target->ews() == 1)) {
PRAGMA_OMP_PARALLEL_FOR_SIMD
for (Nd4jLong e = 0; e < _length; e++)
z[e] = func(e, f[e]);
} else {
if (f == z) {
PRAGMA_OMP_PARALLEL_FOR_SIMD
for (Nd4jLong e = 0; e < _length; e++) {
auto xOffset = this->getOffset(e);
f[xOffset] = func(e, f[xOffset]);
}
} else {
PRAGMA_OMP_PARALLEL_FOR_SIMD
for (Nd4jLong e = 0; e < _length; e++) {
auto xOffset = this->getOffset(e);
auto zOffset = target->getOffset(e);
z[zOffset] = func(e, f[xOffset]);
}
}
}
}
template void NDArray::applyIndexedLambda(const std::function<double(Nd4jLong, double)>& func, NDArray* target);
template void NDArray::applyIndexedLambda(const std::function<float(Nd4jLong, float)>& func, NDArray* target);
template void NDArray::applyIndexedLambda(const std::function<float16(Nd4jLong, float16)>& func, NDArray* target);
template void NDArray::applyIndexedLambda(const std::function<bfloat16(Nd4jLong, bfloat16)>& func, NDArray* target);
template void NDArray::applyIndexedLambda(const std::function<Nd4jLong(Nd4jLong, Nd4jLong)>& func, NDArray* target);
template void NDArray::applyIndexedLambda(const std::function<int(Nd4jLong, int)>& func, NDArray* target);
template void NDArray::applyIndexedLambda(const std::function<int16_t(Nd4jLong, int16_t)>& func, NDArray* target);
template void NDArray::applyIndexedLambda(const std::function<uint8_t (Nd4jLong, uint8_t)>& func, NDArray* target);
template void NDArray::applyIndexedLambda(const std::function<uint16_t (Nd4jLong, uint16_t)>& func, NDArray* target);
template void NDArray::applyIndexedLambda(const std::function<uint32_t (Nd4jLong, uint32_t)>& func, NDArray* target);
template void NDArray::applyIndexedLambda(const std::function<uint64_t (Nd4jLong, uint64_t)>& func, NDArray* target);
template void NDArray::applyIndexedLambda(const std::function<int8_t(Nd4jLong, int8_t)>& func, NDArray* target);
template void NDArray::applyIndexedLambda(const std::function<bool(Nd4jLong, bool)>& func, NDArray* target);
//////////////////////////////////////////////////////////////////////////
template<typename T>
void NDArray::applyIndexedPairwiseLambda(NDArray* other, const std::function<T(Nd4jLong, T, T)>& func, NDArray* target) {
if (target == nullptr)
target = this;
if (other == nullptr) {
nd4j_printf("applyIndexedPairwiseLambda requires both operands to be valid NDArrays, but Y is NULL\n","");
throw std::runtime_error("Other is null");
}
if(dataType() != DataTypeUtils::fromT<T>())
throw std::runtime_error("NDArray::applyIndexedPairwiseLambda<T> method: wrong template parameter T, its type should be the same as type of this array!");
if(dataType() != target->dataType())
throw std::runtime_error("NDArray::applyIndexedPairwiseLambda<T> method: types of this and target array should match !");
if (this->lengthOf() != other->lengthOf()) {
nd4j_printf("applyIndexedPairwiseLambda requires both operands to have the same shape\n","");
throw std::runtime_error("Shapes mismach");
}
auto f = this->bufferAsT<T>();
auto s = other->bufferAsT<T>();
auto z = target->bufferAsT<T>();
if (this->ordering() == other->ordering() && this->ordering() == target->ordering() && (this->ews() == 1 && target->ews() == 1) && this->ews() == other->ews()) {
PRAGMA_OMP_PARALLEL_FOR_SIMD
for (Nd4jLong e = 0; e < _length; e++)
z[e] = func((Nd4jLong) e, f[e], s[e]);
} else {
if (f == z) {
PRAGMA_OMP_PARALLEL_FOR_SIMD
for (int e = 0; e < _length; e++) {
auto xOffset = this->getOffset(e);
auto yOffset = other->getOffset(e);
f[xOffset] = func((Nd4jLong) e, f[xOffset], s[yOffset]);
}
} else {
PRAGMA_OMP_PARALLEL_FOR_SIMD
for (int e = 0; e < _length; e++) {
auto xOffset = this->getOffset(e);
auto yOffset = other->getOffset(e);
auto zOffset = target->getOffset(e);
z[zOffset] = func((Nd4jLong) e, f[xOffset], s[yOffset]);
}
}
}
}
template void NDArray::applyIndexedPairwiseLambda(NDArray* other, const std::function<double (Nd4jLong, double, double)>& func, NDArray* target);
template void NDArray::applyIndexedPairwiseLambda(NDArray* other, const std::function<float (Nd4jLong, float, float)>& func, NDArray* target);
template void NDArray::applyIndexedPairwiseLambda(NDArray* other, const std::function<float16 (Nd4jLong, float16, float16)>& func, NDArray* target);
template void NDArray::applyIndexedPairwiseLambda(NDArray* other, const std::function<bfloat16 (Nd4jLong, bfloat16, bfloat16)>& func, NDArray* target);
template void NDArray::applyIndexedPairwiseLambda(NDArray* other, const std::function<Nd4jLong (Nd4jLong, Nd4jLong, Nd4jLong)>& func, NDArray* target);
template void NDArray::applyIndexedPairwiseLambda(NDArray* other, const std::function<int (Nd4jLong, int, int)>& func, NDArray* target);
template void NDArray::applyIndexedPairwiseLambda(NDArray* other, const std::function<int16_t (Nd4jLong, int16_t, int16_t)>& func, NDArray* target);
template void NDArray::applyIndexedPairwiseLambda(NDArray* other, const std::function<uint8_t (Nd4jLong, uint8_t, uint8_t)>& func, NDArray* target);
template void NDArray::applyIndexedPairwiseLambda(NDArray* other, const std::function<uint16_t (Nd4jLong, uint16_t, uint16_t)>& func, NDArray* target);
template void NDArray::applyIndexedPairwiseLambda(NDArray* other, const std::function<uint32_t (Nd4jLong, uint32_t, uint32_t)>& func, NDArray* target);
template void NDArray::applyIndexedPairwiseLambda(NDArray* other, const std::function<uint64_t (Nd4jLong, uint64_t, uint64_t)>& func, NDArray* target);
template void NDArray::applyIndexedPairwiseLambda(NDArray* other, const std::function<int8_t (Nd4jLong, int8_t, int8_t)>& func, NDArray* target);
template void NDArray::applyIndexedPairwiseLambda(NDArray* other, const std::function<bool (Nd4jLong, bool, bool)>& func, NDArray* target);