.\" Automatically generated by Pandoc 2.9.2.1
.\"
.TH "" "" "2023-05-26" "PMDK - " "PMDK Programmer's Manual"
.hy
.\" SPDX-License-Identifier: BSD-3-Clause
.\" Copyright 2020, Intel Corporation
.SH NAME
.PP
\f[B]pmem2_get_memmove_fn\f[R](), \f[B]pmem2_get_memset_fn\f[R](),
\f[B]pmem2_get_memcpy_fn\f[R]() - get a function that provides optimized
copying to persistent memory
.SH SYNOPSIS
.IP
.nf
\f[C]
#include <libpmem2.h>

typedef void *(*pmem2_memmove_fn)(void *pmemdest, const void *src, size_t len,
        unsigned flags);
typedef void *(*pmem2_memcpy_fn)(void *pmemdest, const void *src, size_t len,
        unsigned flags);
typedef void *(*pmem2_memset_fn)(void *pmemdest, int c, size_t len,
        unsigned flags);

struct pmem2_map;

pmem2_memmove_fn pmem2_get_memmove_fn(struct pmem2_map *map);
pmem2_memset_fn pmem2_get_memset_fn(struct pmem2_map *map);
pmem2_memcpy_fn pmem2_get_memcpy_fn(struct pmem2_map *map);
\f[R]
.fi
.SH DESCRIPTION
.PP
The \f[B]pmem2_get_memmove_fn\f[R](), \f[B]pmem2_get_memset_fn\f[R](),
\f[B]pmem2_get_memcpy_fn\f[R]() functions return a pointer to a function
responsible for efficient storing and flushing of data for mapping
\f[I]map\f[R].
.PP
\f[B]pmem2_memmove_fn\f[R](), \f[B]pmem2_memset_fn\f[R]() and
\f[B]pmem2_memcpy_fn\f[R]() functions provide the same memory copying
functionalities as their namesakes \f[B]memmove\f[R](3),
\f[B]memcpy\f[R](3) and \f[B]memset\f[R](3), and ensure that the result
has been flushed to persistence before returning (unless
\f[B]PMEM2_F_MEM_NOFLUSH\f[R] flag was used).
.PP
For example, the following code:
.IP
.nf
\f[C]
        memmove(dest, src, len);
        pmem2_persist_fn persist_fn = pmem2_get_persist_fn(map);
        persist_fn(dest, len);
\f[R]
.fi
.PP
is functionally equivalent to:
.IP
.nf
\f[C]
        pmem2_memmove_fn memmove_fn = pmem2_get_memmove_fn(map);
        memmove_fn(dest, src, len, 0);
\f[R]
.fi
.PP
Unlike libc implementation, \f[B]libpmem2\f[R] functions guarantee that
if destination buffer address and length are 8 byte aligned then all
stores will be performed using at least 8 byte store instructions.
This means that a series of 8 byte stores followed by
\f[I]persist_fn\f[R] can be safely replaced by a single
\f[I]memmove_fn\f[R] call.
.PP
The \f[I]flags\f[R] argument of all of the above functions has the same
meaning.
It can be 0 or a bitwise OR of one or more of the following flags:
.IP \[bu] 2
\f[B]PMEM2_F_MEM_NODRAIN\f[R] - modifies the behavior to skip the final
\f[I]pmem2_drain_fn\f[R] step.
This allows applications to optimize cases where several ranges are
being copied to persistent memory, followed by a single call to
\f[I]pmem2_drain_fn\f[R].
The following example illustrates how this flag might be used to avoid
multiple calls to \f[I]pmem2_drain_fn\f[R] when copying several ranges
of memory to pmem:
.IP
.nf
\f[C]
pmem2_memcpy_fn memcpy_fn = pmem2_get_memcpy_fn(map);
pmem2_drain_fn drain_fn = pmem2_get_drain_fn(map);

/* ... write several ranges to pmem ... */
memcpy_fn(pmemdest1, src1, len1, PMEM2_F_MEM_NODRAIN);
memcpy_fn(pmemdest2, src2, len2, PMEM2_F_MEM_NODRAIN);

/* ... */

/* wait for any pmem stores to drain from HW buffers */
drain_fn();
\f[R]
.fi
.IP \[bu] 2
\f[B]PMEM2_F_MEM_NOFLUSH\f[R] - Don\[cq]t flush anything.
This implies \f[B]PMEM2_F_MEM_NODRAIN\f[R].
Using this flag only makes sense when it\[cq]s followed by any function
that flushes data.
.PP
The remaining flags say \f[I]how\f[R] the operation should be done, and
are merely hints.
.IP \[bu] 2
\f[B]PMEM2_F_MEM_NONTEMPORAL\f[R] - Use non-temporal instructions.
This flag is mutually exclusive with \f[B]PMEM2_F_MEM_TEMPORAL\f[R].
On x86_64 this flag is mutually exclusive with
\f[B]PMEM2_F_MEM_NOFLUSH\f[R].
.IP \[bu] 2
\f[B]PMEM2_F_MEM_TEMPORAL\f[R] - Use temporal instructions.
This flag is mutually exclusive with \f[B]PMEM2_F_MEM_NONTEMPORAL\f[R].
.IP \[bu] 2
\f[B]PMEM2_F_MEM_WC\f[R] - Use write combining mode.
This flag is mutually exclusive with \f[B]PMEM2_F_MEM_WB\f[R].
On x86_64 this flag is mutually exclusive with
\f[B]PMEM2_F_MEM_NOFLUSH\f[R].
.IP \[bu] 2
\f[B]PMEM2_F_MEM_WB\f[R] - Use write back mode.
This flag is mutually exclusive with \f[B]PMEM2_F_MEM_WC\f[R].
On x86_64 this is an alias for \f[B]PMEM2_F_MEM_TEMPORAL\f[R].
.PP
Using an invalid combination of flags has undefined behavior.
.PP
Without any of the above flags \f[B]libpmem2\f[R] will try to guess the
best strategy based on the data size.
See \f[B]PMEM_MOVNT_THRESHOLD\f[R] description in \f[B]libpmem2\f[R](7)
for details.
.SH RETURN VALUE
.PP
The \f[B]pmem2_get_memmove_fn\f[R](), \f[B]pmem2_get_memset_fn\f[R](),
\f[B]pmem2_get_memcpy_fn\f[R]() functions never return NULL.
.PP
They return the same function for the same mapping.
.PP
This means that it\[cq]s safe to cache their return values.
However, these functions are very cheap (because their return values are
precomputed), so caching may not be necessary.
.PP
If two (or more) mappings share the same \f[I]pmem2_memmove_fn\f[R],
\f[I]pmem2_memset_fn\f[R], \f[I]pmem2_memcpy_fn\f[R] and they are
adjacent to each other, it is safe to call these functions for a range
spanning those mappings.
.SH SEE ALSO
.PP
\f[B]memcpy\f[R](3), \f[B]memmove\f[R](3), \f[B]memset\f[R](3),
\f[B]pmem2_get_drain_fn\f[R](3), \f[B]pmem2_get_memcpy_fn\f[R](3),
\f[B]pmem2_get_memset_fn\f[R](3), \f[B]pmem2_map_new\f[R](3),
\f[B]pmem2_get_persist_fn\f[R](3), \f[B]libpmem2\f[R](7) and
\f[B]<https://pmem.io>\f[R]