re-enable one-work-group radix sort kernels

ndgrigorian · ndgrigorian · commit f8e7040b3649 · 2024-12-20T15:02:29.000-08:00
diff --git a/dpctl/tensor/libtensor/include/kernels/sorting/radix_sort.hpp b/dpctl/tensor/libtensor/include/kernels/sorting/radix_sort.hpp
@@ -1483,7 +1483,7 @@ sycl::event parallel_radix_sort_impl(sycl::queue &exec_q,
         dev.template get_info<sycl::info::device::max_work_group_size>();
 
     constexpr std::uint16_t ref_wg_size = 64;
-    constexpr bool enable_one_wg_radix_sort = false;
+    constexpr bool enable_one_wg_radix_sort = true;
     if (enable_one_wg_radix_sort && n_to_sort <= 16384 &&
         ref_wg_size * 8 <= max_wg_size)
     {

Original file line number	Diff line number	Diff line change
`@@ -1483,7 +1483,7 @@ sycl::event parallel_radix_sort_impl(sycl::queue &exec_q,`
`1483`	`1483`	`dev.template get_info<sycl::info::device::max_work_group_size>();`
`1484`	`1484`
`1485`	`1485`	`constexpr std::uint16_t ref_wg_size = 64;`
`1486`		`- constexpr bool enable_one_wg_radix_sort = false;`
	`1486`	`+ constexpr bool enable_one_wg_radix_sort = true;`
`1487`	`1487`	`if (enable_one_wg_radix_sort && n_to_sort <= 16384 &&`
`1488`	`1488`	`ref_wg_size * 8 <= max_wg_size)`
`1489`	`1489`	`{`