[NFC][ESIMD] Move slm_allocator out of experimental namespace (#13901)

The slm_allocator class was not in experimental namespace in the file
containing definitions for experimental functions, likely by mistake, so
moved it to non experimental include file and moved intrinsics to avoid
confusion

Author: fineg74

sycl/include/sycl/ext/intel/esimd/detail/memory_intrin.hpp

@@ -1160,4 +1160,8 @@ __ESIMD_INTRIN void
 __esimd_nbarrier_arrive(uint8_t id, uint8_t thread_role, uint8_t num_producers,
                         uint8_t num_consumers) __ESIMD_INTRIN_END;
 
+__ESIMD_INTRIN uint32_t __esimd_slm_alloc(uint32_t size) __ESIMD_INTRIN_END;
+
+__ESIMD_INTRIN void __esimd_slm_free(uint32_t id) __ESIMD_INTRIN_END;
+
 /// @endcond ESIMD_DETAIL

sycl/include/sycl/ext/intel/esimd/memory.hpp

@@ -4502,6 +4502,37 @@ constexpr void check_atomic() {
 /// @addtogroup sycl_esimd_memory_slm
 /// @{
 
+/// RAII-style class used to implement "semi-dynamic" SLM allocation.
+/// SLM is allocated in the constructor and released in the destructor, that's
+/// why it is "dynamic", as opposed to fully static allocation style of
+/// 'slm_init'. Actual offset of SLM chunk allocated by the call is calculated
+/// at compile time, that's why it is "semi-". To calculate SLM usage by a
+/// kernel, compiler finds a path in a callgraph with the largest amount of SLM
+/// "locked" by slm_allocator objects live along the paths. slm_init call also
+/// participates in calculating SLM budget. It can be modelled as
+/// \c slm_allocator object declared at the very beginning of a kernel and live
+/// till its the very end.
+/// Only compile-time constant SLM amount is supported for now, it is provided
+/// as a class' template argument.
+///
+/// Since a call graph is used, function pointers and recursion is not
+/// supported.
+///
+/// @tparam SLMAmount The amount allocated in bytes
+template <int SLMAmount> class slm_allocator {
+  int offset;
+
+public:
+  /// Allocates the amount of SLM which is class' template parameter.
+  slm_allocator() { offset = __esimd_slm_alloc(SLMAmount); }
+
+  /// @return The allocated chunk's offset in bytes.
+  ESIMD_INLINE int get_offset() const { return offset; }
+
+  /// Releases the SLM chunk allocated in the constructor.
+  ~slm_allocator() { __esimd_slm_free(offset); }
+};
+
 /// Declare per-work-group slm size.
 /// GPU RT/driver requires this function to be called in the beginning
 /// of the kernel using SLM. There must be only 1 call site of slm_init()

sycl/include/sycl/ext/intel/experimental/esimd/detail/memory_intrin.hpp

@@ -38,9 +38,6 @@ template <__ESIMD_ENS::lsc_memory_kind Kind, __ESIMD_ENS::lsc_fence_op FenceOp,
 __ESIMD_INTRIN void
 __esimd_lsc_fence(__ESIMD_DNS::simd_mask_storage_t<N> pred) __ESIMD_INTRIN_END;
 
-__ESIMD_INTRIN uint32_t __esimd_slm_alloc(uint32_t size) __ESIMD_INTRIN_END;
-
-__ESIMD_INTRIN void __esimd_slm_free(uint32_t id) __ESIMD_INTRIN_END;
 __ESIMD_INTRIN uint8_t __esimd_named_barrier_allocate(uint8_t NbarCount)
     __ESIMD_INTRIN_END;
 

sycl/include/sycl/ext/intel/experimental/esimd/memory.hpp

@@ -2938,37 +2938,6 @@ atomic_update(AccessorTy acc, Toffset offset, simd<T, N> src0, simd<T, N> src1,
       acc, offset, src1, src0, mask);
 }
 
-/// RAII-style class used to implement "semi-dynamic" SLM allocation.
-/// SLM is allocated in the constructor and released in the destructor, that's
-/// why it is "dynamic", as opposed to fully static allocation style of
-/// 'slm_init'. Actual offset of SLM chunk allocated by the call is calculated
-/// at compile time, that's why it is "semi-". To calculate SLM usage by a
-/// kernel, compiler finds a path in a callgraph with the largest amount of SLM
-/// "locked" by slm_allocator objects live along the paths. slm_init call also
-/// participates in calculating SLM budget. It can be modelled as
-/// \c slm_allocator object declared at the very beginning of a kernel and live
-/// till its the very end.
-/// Only compile-time constant SLM amount is supported for now, it is provided
-/// as a class' template argument.
-///
-/// Since a call graph is used, function pointers and recursion is not
-/// supported.
-///
-/// @tparam SLMAmount The amount allocated in bytes
-template <int SLMAmount> class slm_allocator {
-  int offset;
-
-public:
-  /// Allocates the amount of SLM which is class' template parameter.
-  slm_allocator() { offset = __esimd_slm_alloc(SLMAmount); }
-
-  /// @return The allocated chunk's offset in bytes.
-  ESIMD_INLINE int get_offset() const { return offset; }
-
-  /// Releases the SLM chunk allocated in the constructor.
-  ~slm_allocator() { __esimd_slm_free(offset); }
-};
-
 } // namespace esimd
 } // namespace ext::intel
 } // namespace _V1