Fix error in SYCL's basic data parallel kernel (only manifested in multi-GPU setups): the grid_x_offset and grid_y_offset was swapped.

Author: breyerml

include/plssvm/backends/SYCL/kernel/cg_explicit/basic/blas.hpp

@@ -380,8 +380,8 @@ class device_kernel_inplace_matrix_scale {
         constexpr auto PADDING_SIZE_uz = static_cast<std::size_t>(PADDING_SIZE);
 
         // calculate the indices used in the current work-item
-        const auto i_idx = (idx.get_id(1) + grid_x_offset_ * THREAD_BLOCK_SIZE_uz) * INTERNAL_BLOCK_SIZE_uz;  // num_rows
-        const auto j_idx = (idx.get_id(0) + grid_y_offset_ * THREAD_BLOCK_SIZE_uz) * INTERNAL_BLOCK_SIZE_uz;  // num_rhs
+        const auto i_idx = (idx.get_id(1) + grid_y_offset_ * THREAD_BLOCK_SIZE_uz) * INTERNAL_BLOCK_SIZE_uz;  // num_rows
+        const auto j_idx = (idx.get_id(0) + grid_x_offset_ * THREAD_BLOCK_SIZE_uz) * INTERNAL_BLOCK_SIZE_uz;  // num_rhs
 
         for (unsigned internal_i = 0; internal_i < INTERNAL_BLOCK_SIZE; ++internal_i) {
             for (unsigned internal_j = 0; internal_j < INTERNAL_BLOCK_SIZE; ++internal_j) {