Merge branch 'branch-25.02' into perf-batched-memcpy-orc-stats

rapidsai · Dec 11, 2024 · 6fe45b8 · 6fe45b8
2 parents 06b8276 + 3801e74
commit 6fe45b8
Show file tree

Hide file tree

Showing 3 changed files with 13 additions and 10 deletions.
diff --git a/cpp/CMakeLists.txt b/cpp/CMakeLists.txt
@@ -277,7 +277,7 @@ rapids_cpm_init()
 # Not using rapids-cmake since we never want to find, always download.
 CPMAddPackage(
   NAME rapids_logger GITHUB_REPOSITORY rapidsai/rapids-logger GIT_SHALLOW TRUE GIT_TAG
-  14bb233d2420f7187a690f0bb528ec0420c70d48
+  c510947ae9d3a67530cfe3e5eaccb5a3b8ea0e55 VERSION c510947ae9d3a67530cfe3e5eaccb5a3b8ea0e55
 )
 rapids_make_logger(cudf EXPORT_SET cudf-exports)
 

diff --git a/cpp/include/cudf/detail/get_value.cuh b/cpp/include/cudf/detail/get_value.cuh
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020-2023, NVIDIA CORPORATION.
+ * Copyright (c) 2020-2024, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -17,6 +17,7 @@
 #pragma once
 
 #include <cudf/column/column_view.hpp>
+#include <cudf/detail/utilities/vector_factories.hpp>
 #include <cudf/utilities/error.hpp>
 #include <cudf/utilities/traits.hpp>
 #include <cudf/utilities/type_dispatcher.hpp>
@@ -48,11 +49,9 @@ T get_value(column_view const& col_view, size_type element_index, rmm::cuda_stre
   CUDF_EXPECTS(data_type(type_to_id<T>()) == col_view.type(), "get_value data type mismatch");
   CUDF_EXPECTS(element_index >= 0 && element_index < col_view.size(),
                "invalid element_index value");
-  T result;
-  CUDF_CUDA_TRY(cudaMemcpyAsync(
-    &result, col_view.data<T>() + element_index, sizeof(T), cudaMemcpyDefault, stream.value()));
-  stream.synchronize();
-  return result;
+  return cudf::detail::make_host_vector_sync(
+           device_span<T const>{col_view.data<T>() + element_index, 1}, stream)
+    .front();
 }
 
 }  // namespace detail

diff --git a/cpp/include/cudf/table/table_device_view.cuh b/cpp/include/cudf/table/table_device_view.cuh
@@ -16,6 +16,8 @@
 #pragma once
 
 #include <cudf/column/column_device_view.cuh>
+#include <cudf/detail/utilities/cuda_memcpy.hpp>
+#include <cudf/detail/utilities/vector_factories.hpp>
 #include <cudf/table/table_view.hpp>
 #include <cudf/types.hpp>
 #include <cudf/utilities/default_stream.hpp>
@@ -251,7 +253,7 @@ auto contiguous_copy_column_device_views(HostTableView source_view, rmm::cuda_st
   // A buffer of CPU memory is allocated to hold the ColumnDeviceView
   // objects. Once filled, the CPU memory is then copied to device memory
   // and the pointer is set in the d_columns member.
-  std::vector<int8_t> h_buffer(padded_views_size_bytes);
+  auto h_buffer = cudf::detail::make_host_vector<int8_t>(padded_views_size_bytes, stream);
   // Each ColumnDeviceView instance may have child objects which may
   // require setting some internal device pointers before being copied
   // from CPU to device.
@@ -266,8 +268,10 @@ auto contiguous_copy_column_device_views(HostTableView source_view, rmm::cuda_st
   auto d_columns = detail::child_columns_to_device_array<ColumnDeviceView>(
     source_view.begin(), source_view.end(), h_ptr, d_ptr);
 
-  CUDF_CUDA_TRY(cudaMemcpyAsync(d_ptr, h_ptr, views_size_bytes, cudaMemcpyDefault, stream.value()));
-  stream.synchronize();
+  auto const h_span = host_span<int8_t const>{h_buffer}.subspan(
+    static_cast<int8_t const*>(h_ptr) - h_buffer.data(), views_size_bytes);
+  auto const d_span = device_span<int8_t>{static_cast<int8_t*>(d_ptr), views_size_bytes};
+  cudf::detail::cuda_memcpy(d_span, h_span, stream);
   return std::make_tuple(std::move(descendant_storage), d_columns);
 }