separate intrinsics into core

common/unified/components/range_minimum_query_kernels.cpp

@@ -6,9 +6,8 @@
 
 #include <limits>
 
-#include <ginkgo/core/base/intrinsics.hpp>
-
 #include "common/unified/base/kernel_launch.hpp"
+#include "core/base/intrinsics.hpp"
 #include "core/components/bit_packed_storage.hpp"
 #include "core/components/range_minimum_query.hpp"
 

core/base/intrinsics.hpp

@@ -0,0 +1,93 @@
+// SPDX-FileCopyrightText: 2017 - 2025 The Ginkgo authors
+//
+// SPDX-License-Identifier: BSD-3-Clause
+
+#ifndef GKO_CORE_BASE_INTRINSICS_HPP_
+#define GKO_CORE_BASE_INTRINSICS_HPP_
+
+
+#include <ginkgo/core/base/types.hpp>
+
+// MSVC needs different intrinsics
+#ifdef _MSC_VER
+#include <intrin.h>
+
+#pragma intrinsic(_BitScanForward, _BitScanForward64, _BitScanReverse, \
+                  _BitScanReverse64)
+#endif
+
+
+namespace gko {
+namespace detail {
+
+
+/**
+ * Returns the index of the highest bit set in this bitmask.
+ * The least significant bit has index 0.
+ */
+GKO_ATTRIBUTES GKO_INLINE int find_highest_bit(uint32 bitmask)
+{
+#if defined(__CUDA_ARCH__) || defined(__HIP_DEVICE_COMPILE__)
+    return 31 - __clz(static_cast<unsigned>(bitmask));
+#elif defined(_MSC_VER)
+    unsigned long index{};
+    _BitScanReverse(&index, bitmask);
+    return index;
+#else
+    return 31 - __builtin_clz(bitmask);
+#endif
+}
+
+
+/** @copydoc find_highest_bit(uint32) */
+GKO_ATTRIBUTES GKO_INLINE int find_highest_bit(uint64 bitmask)
+{
+#if defined(__CUDA_ARCH__) || defined(__HIP_DEVICE_COMPILE__)
+    return 63 - __clzll(static_cast<unsigned long long>(bitmask));
+#elif defined(_MSC_VER)
+    unsigned long index{};
+    _BitScanReverse64(&index, bitmask);
+    return index;
+#else
+    return 63 - __builtin_clzll(bitmask);
+#endif
+}
+
+
+/**
+ * Returns the index of the lowest bit set in this bitmask.
+ * The least significant bit has index 0.
+ */
+GKO_ATTRIBUTES GKO_INLINE int find_lowest_bit(uint32 bitmask)
+{
+#if defined(__CUDA_ARCH__) || defined(__HIP_DEVICE_COMPILE__)
+    return __ffs(static_cast<unsigned>(bitmask)) - 1;
+#elif defined(_MSC_VER)
+    unsigned long index{};
+    _BitScanForward(&index, bitmask);
+    return index;
+#else
+    return __builtin_ffs(bitmask) - 1;
+#endif
+}
+
+
+/** @copydoc find_lowest_bit(uint32) */
+GKO_ATTRIBUTES GKO_INLINE int find_lowest_bit(uint64 bitmask)
+{
+#if defined(__CUDA_ARCH__) || defined(__HIP_DEVICE_COMPILE__)
+    return __ffsll(static_cast<unsigned long long>(bitmask)) - 1;
+#elif defined(_MSC_VER)
+    unsigned long index{};
+    _BitScanForward64(&index, bitmask);
+    return index;
+#else
+    return __builtin_ffsll(bitmask) - 1;
+#endif
+}
+
+
+}  // namespace detail
+}  // namespace gko
+
+#endif  // GKO_CORE_BASE_INTRINSICS_HPP_

core/components/bit_packed_storage.hpp

@@ -9,9 +9,10 @@
 #include <limits>
 #include <utility>
 
-#include <ginkgo/core/base/intrinsics.hpp>
 #include <ginkgo/core/base/types.hpp>
 
+#include "core/base/intrinsics.hpp"
+
 namespace gko {
 
 

core/components/range_minimum_query.hpp

@@ -10,11 +10,11 @@
 #include <utility>
 
 #include <ginkgo/core/base/array.hpp>
-#include <ginkgo/core/base/intrinsics.hpp>
 #include <ginkgo/core/base/math.hpp>
 #include <ginkgo/core/base/types.hpp>
 
 #include "core/base/index_range.hpp"
+#include "core/base/intrinsics.hpp"
 #include "core/components/bit_packed_storage.hpp"
 
 namespace gko {

include/ginkgo/core/base/intrinsics.hpp

@@ -10,14 +10,6 @@
 
 #include <ginkgo/core/base/types.hpp>
 
-// MSVC needs different intrinsics
-#ifdef _MSC_VER
-#include <intrin.h>
-
-#pragma intrinsic(_BitScanForward, _BitScanForward64, _BitScanReverse, \
-                  _BitScanReverse64)
-#endif
-
 
 namespace gko {
 namespace detail {
@@ -49,72 +41,6 @@ GKO_ATTRIBUTES GKO_INLINE int popcount(uint64 bitmask)
 }
 
 
-/**
- * Returns the index of the highest bit set in this bitmask.
- * The least significant bit has index 0.
- */
-GKO_ATTRIBUTES GKO_INLINE int find_highest_bit(uint32 bitmask)
-{
-#if defined(__CUDA_ARCH__) || defined(__HIP_DEVICE_COMPILE__)
-    return 31 - __clz(static_cast<unsigned>(bitmask));
-#elif defined(_MSC_VER)
-    unsigned long index{};
-    _BitScanReverse(&index, bitmask);
-    return index;
-#else
-    return 31 - __builtin_clz(bitmask);
-#endif
-}
-
-
-/** @copydoc find_highest_bit(uint32) */
-GKO_ATTRIBUTES GKO_INLINE int find_highest_bit(uint64 bitmask)
-{
-#if defined(__CUDA_ARCH__) || defined(__HIP_DEVICE_COMPILE__)
-    return 63 - __clzll(static_cast<unsigned long long>(bitmask));
-#elif defined(_MSC_VER)
-    unsigned long index{};
-    _BitScanReverse64(&index, bitmask);
-    return index;
-#else
-    return 63 - __builtin_clzll(bitmask);
-#endif
-}
-
-
-/**
- * Returns the index of the lowest bit set in this bitmask.
- * The least significant bit has index 0.
- */
-GKO_ATTRIBUTES GKO_INLINE int find_lowest_bit(uint32 bitmask)
-{
-#if defined(__CUDA_ARCH__) || defined(__HIP_DEVICE_COMPILE__)
-    return __ffs(static_cast<unsigned>(bitmask)) - 1;
-#elif defined(_MSC_VER)
-    unsigned long index{};
-    _BitScanForward(&index, bitmask);
-    return index;
-#else
-    return __builtin_ffs(bitmask) - 1;
-#endif
-}
-
-
-/** @copydoc find_lowest_bit(uint32) */
-GKO_ATTRIBUTES GKO_INLINE int find_lowest_bit(uint64 bitmask)
-{
-#if defined(__CUDA_ARCH__) || defined(__HIP_DEVICE_COMPILE__)
-    return __ffsll(static_cast<unsigned long long>(bitmask)) - 1;
-#elif defined(_MSC_VER)
-    unsigned long index{};
-    _BitScanForward64(&index, bitmask);
-    return index;
-#else
-    return __builtin_ffsll(bitmask) - 1;
-#endif
-}
-
-
 }  // namespace detail
 }  // namespace gko
 

reference/components/range_minimum_query_kernels.cpp

@@ -6,8 +6,7 @@
 
 #include <limits>
 
-#include <ginkgo/core/base/intrinsics.hpp>
-
+#include "core/base/intrinsics.hpp"
 #include "core/components/bit_packed_storage.hpp"
 #include "core/components/range_minimum_query.hpp"
 

test/base/intrinsics.cpp

@@ -2,6 +2,8 @@
 //
 // SPDX-License-Identifier: BSD-3-Clause
 
+#include "core/base/intrinsics.hpp"
+
 #include <memory>
 
 #include <gtest/gtest.h>