review updates

* remove unused variables * add documentation to highest_precision helper Co-authored-by: Terry Cojean <terry.cojean@kit.edu>
ginkgo-project · Oct 1, 2021 · c3f5fa3 · c3f5fa3
1 parent 9bd1f32
commit c3f5fa3
Show file tree

Hide file tree

Showing 3 changed files with 14 additions and 11 deletions.
diff --git a/include/ginkgo/core/base/math.hpp b/include/ginkgo/core/base/math.hpp
@@ -508,6 +508,17 @@ template <typename T>
 using increase_precision = typename detail::increase_precision_impl<T>::type;
 
 
+/**
+ * Obtains the smallest arithmetic type that is able to store elements of all
+ * template parameter types exactly. All template type parameters need to be
+ * either real or complex types, mixing them is not possible.
+ *
+ * Formally, it computes a right-fold over the type list, with the highest
+ * precision of a pair of real arithmetic types T1, T2 computed as
+ * `decltype(T1{} + T2{})`, or
+ * `std::complex<highest_precision<remove_complex<T1>, remove_complex<T2>>>` for
+ * complex types.
+ */
 template <typename... Ts>
 using highest_precision =
     typename detail::highest_precision_variadic<Ts...>::type;

diff --git a/omp/matrix/coo_kernels.cpp b/omp/matrix/coo_kernels.cpp
@@ -131,7 +131,6 @@ void spmv2_blocked(std::shared_ptr<const OmpExecutor> exec,
                     for (auto local_nz = nz;
                          local_nz < end && coo_row[local_nz] == first;
                          local_nz++) {
-                        const auto row = first;
                         const auto col = coo_col[local_nz];
 #pragma unroll
                         for (size_type i = 0; i < block_size; i++) {
@@ -190,7 +189,6 @@ void spmv2_blocked(std::shared_ptr<const OmpExecutor> exec,
                     // sums
                     partial_sum.fill(zero<ValueType>());
                     for (auto local_nz = nz; local_nz < end; local_nz++) {
-                        const auto row = last;
                         const auto col = coo_col[local_nz];
 #pragma unroll
                         for (size_type i = 0; i < block_size; i++) {
@@ -205,13 +203,12 @@ void spmv2_blocked(std::shared_ptr<const OmpExecutor> exec,
                     for (size_type i = 0; i < block_size; i++) {
                         const auto rhs = i + rhs_base;
                         const auto row = last;
-                        atomic_add(c->at(last, rhs), partial_sum[i]);
+                        atomic_add(c->at(row, rhs), partial_sum[i]);
                     }
                 }
                 // handle row overlap with following thread: block partial sums
                 partial_sum.fill(zero<ValueType>());
                 for (; nz < end; nz++) {
-                    const auto row = last;
                     const auto col = coo_col[nz];
                     for (size_type rhs = rounded_rhs; rhs < num_rhs; rhs++) {
                         partial_sum[rhs - rounded_rhs] +=
@@ -221,8 +218,7 @@ void spmv2_blocked(std::shared_ptr<const OmpExecutor> exec,
                 // handle row overlap with following thread: block add to memory
                 for (size_type rhs = rounded_rhs; rhs < num_rhs; rhs++) {
                     const auto row = last;
-                    atomic_add(c->at(last, rhs),
-                               partial_sum[rhs - rounded_rhs]);
+                    atomic_add(c->at(row, rhs), partial_sum[rhs - rounded_rhs]);
                 }
             }
         }
@@ -260,7 +256,6 @@ void spmv2_small_rhs(std::shared_ptr<const OmpExecutor> exec,
                 // handle row overlap with previous thread: partial sums
                 partial_sum.fill(zero<ValueType>());
                 for (; nz < end && coo_row[nz] == first; nz++) {
-                    const auto row = first;
                     const auto col = coo_col[nz];
 #pragma unroll
                     for (size_type rhs = 0; rhs < num_rhs; rhs++) {
@@ -287,7 +282,6 @@ void spmv2_small_rhs(std::shared_ptr<const OmpExecutor> exec,
                 // handle row overlap with following thread: partial sums
                 partial_sum.fill(zero<ValueType>());
                 for (; nz < end; nz++) {
-                    const auto row = last;
                     const auto col = coo_col[nz];
 #pragma unroll
                     for (size_type rhs = 0; rhs < num_rhs; rhs++) {
@@ -299,7 +293,7 @@ void spmv2_small_rhs(std::shared_ptr<const OmpExecutor> exec,
 #pragma unroll
                 for (size_type rhs = 0; rhs < num_rhs; rhs++) {
                     const auto row = last;
-                    atomic_add(c->at(last, rhs), partial_sum[rhs]);
+                    atomic_add(c->at(row, rhs), partial_sum[rhs]);
                 }
             }
         }

diff --git a/omp/matrix/sellp_kernels.cpp b/omp/matrix/sellp_kernels.cpp
@@ -60,7 +60,6 @@ void spmv_small_rhs(std::shared_ptr<const OmpExecutor> exec,
                     matrix::Dense<ValueType>* c, OutFn out)
 {
     GKO_ASSERT(b->get_size()[1] == num_rhs);
-    auto col_idxs = a->get_const_col_idxs();
     auto slice_lengths = a->get_const_slice_lengths();
     auto slice_sets = a->get_const_slice_sets();
     auto slice_size = a->get_slice_size();
@@ -100,7 +99,6 @@ void spmv_blocked(std::shared_ptr<const OmpExecutor> exec,
                   const matrix::Dense<ValueType>* b,
                   matrix::Dense<ValueType>* c, OutFn out)
 {
-    auto col_idxs = a->get_const_col_idxs();
     auto slice_lengths = a->get_const_slice_lengths();
     auto slice_sets = a->get_const_slice_sets();
     auto slice_size = a->get_slice_size();