Skip to content

[enhancement] Compiler hints for arm #3141

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 14 commits into from
May 15, 2025
Merged

[enhancement] Compiler hints for arm #3141

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
14 commits
Select commit Hold shift + click to select a range
a49109f
compiler hints for arm
rakshithgb-fujitsu Mar 6, 2025
7ac52c8
clang format
rakshithgb-fujitsu Mar 25, 2025
49bfff3
Update cpp/daal/src/algorithms/dtrees/forest/classification/df_classi…
rakshithgb-fujitsu Apr 10, 2025
42fd1b6
Update cpp/daal/src/algorithms/dtrees/forest/classification/df_classi…
rakshithgb-fujitsu Apr 10, 2025
096a3d1
Linting
Alexsandruss Apr 27, 2025
a9694bc
removed pragmas with no impact on optimisation
rakshithgb-fujitsu Apr 30, 2025
fadfa62
Update multiclassclassifier_train_oneagainstone_impl.i
rakshithgb-fujitsu Apr 30, 2025
8b749ed
updates to svd and qr
rakshithgb-fujitsu Apr 30, 2025
b59b892
removed comments
rakshithgb-fujitsu Apr 30, 2025
d303974
change macro name
rakshithgb-fujitsu Apr 30, 2025
fbe767e
build fix
rakshithgb-fujitsu May 1, 2025
b437ff4
clang fix
rakshithgb-fujitsu May 2, 2025
9dac232
Fix build error
Vika-F May 14, 2025
7a9be67
clang fix
rakshithgb-fujitsu May 15, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 8 additions & 8 deletions cpp/daal/src/algorithms/cholesky/cholesky_impl.i
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -151,13 +151,13 @@ bool CholeskyKernel<algorithmFPType, method, cpu>::copyToFullMatrix(NumericTable

for (size_t i = iBlock * blockSize; i < endBlock; i++)
{
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t j = 0; j <= i; j++)
{
pL[i * dim + j] = pA[i * dim + j];
}
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t j = (i + 1); j < dim; j++)
{
Expand All @@ -176,13 +176,13 @@ bool CholeskyKernel<algorithmFPType, method, cpu>::copyToFullMatrix(NumericTable
{
const size_t ind = (i + 1) * i / 2;

PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t j = 0; j <= i; j++)
{
pL[i * dim + j] = pA[ind + j];
}
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t j = (i + 1); j < dim; j++)
{
Expand All @@ -201,13 +201,13 @@ bool CholeskyKernel<algorithmFPType, method, cpu>::copyToFullMatrix(NumericTable
{
const size_t ind = (2 * dim - j + 1) * j / 2;

PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t i = 0; i < j; i++)
{
pL[i * dim + j] = algorithmFPType(0);
}
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t i = j; i < dim; i++)
{
Expand Down Expand Up @@ -247,7 +247,7 @@ services::Status CholeskyKernel<algorithmFPType, method, cpu>::copyToLowerTriang
{
const size_t ind = (i + 1) * i / 2;

PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t j = 0; j <= i; j++)
{
Expand Down Expand Up @@ -276,7 +276,7 @@ services::Status CholeskyKernel<algorithmFPType, method, cpu>::copyToLowerTriang
{
const size_t ind = (j + 1) * j / 2;

PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t i = 0; i <= j; i++)
{
Expand Down
12 changes: 6 additions & 6 deletions cpp/daal/src/algorithms/covariance/covariance_impl.i
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -227,7 +227,7 @@ public:
/* Sum input array elements in case of non-normalized data */
for (DAAL_INT i = 0; i < nRows; i++)
{
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (DAAL_INT j = 0; j < _nFeatures; j++)
{
Expand Down Expand Up @@ -269,7 +269,7 @@ public:
}

/// It is safe to use aligned loads and stores because the data in TArrayScalableCalloc data structures is aligned
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
PRAGMA_VECTOR_ALIGNED
for (size_t i = 0; i < (_nFeatures * _nFeatures); i++)
Expand All @@ -286,7 +286,7 @@ public:
return;
}
/// It is safe to use aligned loads and stores because the data is aligned
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
PRAGMA_VECTOR_ALIGNED
for (size_t i = 0; i < _nFeatures; i++)
Expand Down Expand Up @@ -407,7 +407,7 @@ services::Status updateDenseCrossProductAndSums(bool isNormalized, size_t nFeatu
}
for (size_t i = 0; i < nFeatures; i++)
{
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t j = 0; j < nFeatures; j++)
{
Expand Down Expand Up @@ -501,7 +501,7 @@ void mergeCrossProductAndSums(size_t nFeatures, const algorithmFPType * partialC
if (nObsValue == 0)
{
daal::threader_for(nFeatures, nFeatures, [=](size_t i) {
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t j = 0; j <= i; j++)
{
Expand All @@ -517,7 +517,7 @@ void mergeCrossProductAndSums(size_t nFeatures, const algorithmFPType * partialC
algorithmFPType invNewNObs = 1.0 / (nObsValue + partialNObsValue);

daal::threader_for(nFeatures, nFeatures, [=](size_t i) {
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t j = 0; j <= i; j++)
{
Expand Down
6 changes: 3 additions & 3 deletions cpp/daal/src/algorithms/dtrees/dtrees_train_data_helper.i
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -419,7 +419,7 @@ int doPartition(SizeType n, const IndexType * aIdx, const ResponseType * aRespon
SizeType iRight = 0;
int iRowSplitVal = -1;

PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (SizeType i = 0; i < n; ++i)
{
Expand Down Expand Up @@ -457,7 +457,7 @@ int doPartitionIdx(SizeType n, const IndexType * aIdx, const IndexType * aIdx2,

if (aIdx2)
{
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (SizeType i = 0; i < n; ++i)
{
Expand All @@ -478,7 +478,7 @@ int doPartitionIdx(SizeType n, const IndexType * aIdx, const IndexType * aIdx2,
}
else
{
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (SizeType i = 0; i < n; ++i)
{
Expand Down
20 changes: 10 additions & 10 deletions ...orithms/dtrees/forest/classification/df_classification_predict_dense_default_batch_impl.i
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -68,7 +68,7 @@ DAAL_FORCEINLINE void fillResults(const size_t nClasses, const enum VotingMethod
{
if (votingMethod == VotingMethod::unweighted || probas == nullptr)
{
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t i = 0; i < blockSize; ++i)
{
Expand All @@ -80,7 +80,7 @@ DAAL_FORCEINLINE void fillResults(const size_t nClasses, const enum VotingMethod
{
for (size_t i = 0; i < blockSize; ++i)
{
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t j = 0; j < nClasses; ++j)
{
Expand Down Expand Up @@ -340,7 +340,7 @@ Status PredictClassificationTask<algorithmFPType, cpu>::predictByTrees(const siz
}
else if (_votingMethod == VotingMethod::weighted)
{
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t i = 0; i < _nClasses; ++i)
{
Expand All @@ -357,7 +357,7 @@ Status PredictClassificationTask<algorithmFPType, cpu>::predictByTrees(const siz
sum += resPtr[i];
}

PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t i = 0; i < _nClasses; ++i)
{
Expand Down Expand Up @@ -403,7 +403,7 @@ Status PredictClassificationTask<algorithmFPType, cpu>::predictByTreesWithoutCon
}
else if (_votingMethod == VotingMethod::weighted)
{
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t i = 0; i < _nClasses; ++i)
{
Expand All @@ -420,7 +420,7 @@ Status PredictClassificationTask<algorithmFPType, cpu>::predictByTreesWithoutCon
sum += resPtr[i];
}

PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t i = 0; i < _nClasses; ++i)
{
Expand Down Expand Up @@ -449,7 +449,7 @@ Status PredictClassificationTask<algorithmFPType, cpu>::parallelPredict(const al

SafeStatus safeStat;

PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t i = 0; i < treeSize; ++i)
{
Expand Down Expand Up @@ -769,7 +769,7 @@ DAAL_FORCEINLINE Status PredictClassificationTask<algorithmFPType, cpu>::predict
}
if (probPtr != nullptr)
{
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t j = 0; j < _nClasses; ++j)
{
Expand Down Expand Up @@ -891,7 +891,7 @@ DAAL_FORCEINLINE Status PredictClassificationTask<float, avx512>::predictOneRowB
{
const size_t treeSize = _aTree[iTree + i]->getNumberOfRows();
const DecisionTreeNode * aNode = (const DecisionTreeNode *)(*_aTree[iTree + i]).getArray();
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t j = 0; j < treeSize; ++j)
{
Expand Down Expand Up @@ -1002,7 +1002,7 @@ DAAL_FORCEINLINE Status PredictClassificationTask<float, avx512>::predictOneRowB
}
if (probPtr != nullptr)
{
PRAGMA_IVDEP
PRAGMA_FORCE_SIMD
PRAGMA_VECTOR_ALWAYS
for (size_t j = 0; j < _nClasses; ++j)
{
Expand Down
Loading
Loading