($INBOX_DIR/description missing)
help / color / mirror / Atom feed[PATCH v7 3/3] Optimize hash function
49+ messages / 3 participants
[nested] [flat]
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v8 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.2
--------------846F221A8A016B7AA8E567B3--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v7 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index d62b1495dd3..6189d051072 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.1
--------------C6A03312D0F99F9183A0DF46--
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v6 3/3] Optimize hash function
@ 2021-01-20 22:06 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-01-20 22:06 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 1db1e7cf14b..72e4d8a784c 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -163,15 +163,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers (Files
+ * and Buffers as of this writing), so we want to incorporate the 'kind'
+ * in the hash too, otherwise those resources will collide a lot. But
+ * because there are only a few resource kinds like that - and only a few
+ * resource kinds to begin with - we don't need to work too hard to mix
+ * 'kind' into the hash. Just add it with hash_combine(), it perturbs the
+ * result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
static void
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.29.2
--------------F30A5443A5D524A5CBC51AE7
Content-Type: text/x-patch; charset=UTF-8;
name="v2-0001-resownerbench.patch"
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment;
filename="v2-0001-resownerbench.patch"
^ permalink raw reply [nested|flat] 49+ messages in thread
* [PATCH v10 3/3] Optimize hash function
@ 2021-10-18 10:59 Heikki Linnakangas <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Heikki Linnakangas @ 2021-10-18 10:59 UTC (permalink / raw)
---
src/backend/utils/resowner/resowner.c | 24 ++++++++++++++++++------
src/include/common/hashfn.h | 15 +++++++++++++++
2 files changed, 33 insertions(+), 6 deletions(-)
diff --git a/src/backend/utils/resowner/resowner.c b/src/backend/utils/resowner/resowner.c
index 35e824d4534..118bd2b34c9 100644
--- a/src/backend/utils/resowner/resowner.c
+++ b/src/backend/utils/resowner/resowner.c
@@ -169,15 +169,27 @@ static void ReleaseAuxProcessResourcesCallback(int code, Datum arg);
* INTERNAL ROUTINES *
*****************************************************************************/
+/*
+ * Hash function for value+kind combination.
+ */
static inline uint32
hash_resource_elem(Datum value, ResourceOwnerFuncs *kind)
{
- Datum data[2];
-
- data[0] = value;
- data[1] = PointerGetDatum(kind);
-
- return hash_bytes((unsigned char *) &data, 2 * SIZEOF_DATUM);
+ /*
+ * Most resources types store a pointer in 'value', and pointers are
+ * unique all on their own. But some resources store plain integers
+ * (Files and Buffers as of this writing), so we want to incorporate the
+ * 'kind' in the hash too, otherwise those resources will collide a lot.
+ * But because there are only a few resource kinds like that - and only a
+ * few resource kinds to begin with - we don't need to work too hard to
+ * mix 'kind' into the hash. Just add it with hash_combine(), it perturbs
+ * the result enough for our purposes.
+ */
+#if SIZEOF_DATUM == 8
+ return hash_combine64(murmurhash64((uint64) value), (uint64) kind);
+#else
+ return hash_combine(murmurhash32((uint32) value), (uint32) kind);
+#endif
}
/*
diff --git a/src/include/common/hashfn.h b/src/include/common/hashfn.h
index c634cc067a1..009ffbbdd38 100644
--- a/src/include/common/hashfn.h
+++ b/src/include/common/hashfn.h
@@ -101,4 +101,19 @@ murmurhash32(uint32 data)
return h;
}
+/* 64-bit variant */
+static inline uint64
+murmurhash64(uint64 data)
+{
+ uint64 h = data;
+
+ h ^= h >> 33;
+ h *= 0xff51afd7ed558ccd;
+ h ^= h >> 33;
+ h *= 0xc4ceb9fe1a85ec53;
+ h ^= h >> 33;
+
+ return h;
+}
+
#endif /* HASHFN_H */
--
2.30.2
--------------AF226508B1039DE40796DE51--
^ permalink raw reply [nested|flat] 49+ messages in thread
* Re: Use IsA() macro instead of nodeTag comparison
@ 2026-01-09 00:56 Chao Li <[email protected]>
0 siblings, 1 reply; 49+ messages in thread
From: Chao Li @ 2026-01-09 00:56 UTC (permalink / raw)
To: Fujii Masao <[email protected]>; +Cc: Heikki Linnakangas <[email protected]>; Kirill Reshke <[email protected]>; Shinya Kato <[email protected]>; PostgreSQL Hackers <[email protected]>
> On Jan 9, 2026, at 08:32, Fujii Masao <[email protected]> wrote:
>
> On Fri, Jan 9, 2026 at 2:11 AM Heikki Linnakangas <[email protected]> wrote:
>>
>> On 08/01/2026 15:10, Kirill Reshke wrote:
>>> On Thu, 8 Jan 2026 at 17:31, Shinya Kato <[email protected]> wrote:
>>>> In copy.c, nodeTag was being compared directly, so I replaced it with
>>>> the IsA() predicate macro for consistency.
>>>
>>> Oops. Looks like oversight in 6c8f670. This is indeed case where we
>>> should use IsA()
>>>
>>>> I verified that there are no other direct comparisons left by running
>>>> the following command:
>>>> $ git grep "nodeTag(.*) =="
>>>
>>> Yep, look like this is the only case in copy.c
>
> I found a similar issue in define.c. Should we fix it there as well?
>
> diff --git a/src/backend/commands/define.c b/src/backend/commands/define.c
> index 63601a2c0b4..8313431397f 100644
> --- a/src/backend/commands/define.c
> +++ b/src/backend/commands/define.c
> @@ -349,7 +349,7 @@ defGetStringList(DefElem *def)
> (errcode(ERRCODE_SYNTAX_ERROR),
> errmsg("%s requires a parameter",
> def->defname)));
> - if (nodeTag(def->arg) != T_List)
> + if (!IsA(def->arg, List))
> elog(ERROR, "unrecognized node type: %d", (int)
> nodeTag(def->arg));
>
> foreach(cell, (List *) def->arg)
>
> Regards,
>
> --
> Fujii Masao
Yep, I did a search with `nodeTag\(.*\)\s+(?:!=|==).*`, and this is the only finding.
Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/
^ permalink raw reply [nested|flat] 49+ messages in thread
* Re: Use IsA() macro instead of nodeTag comparison
@ 2026-01-09 02:19 Shinya Kato <[email protected]>
parent: Chao Li <[email protected]>
0 siblings, 0 replies; 49+ messages in thread
From: Shinya Kato @ 2026-01-09 02:19 UTC (permalink / raw)
To: Chao Li <[email protected]>; +Cc: Fujii Masao <[email protected]>; Heikki Linnakangas <[email protected]>; Kirill Reshke <[email protected]>; PostgreSQL Hackers <[email protected]>
On Fri, Jan 9, 2026 at 9:56 AM Chao Li <[email protected]> wrote:
>
> > I found a similar issue in define.c. Should we fix it there as well?
> >
> > diff --git a/src/backend/commands/define.c b/src/backend/commands/define.c
> > index 63601a2c0b4..8313431397f 100644
> > --- a/src/backend/commands/define.c
> > +++ b/src/backend/commands/define.c
> > @@ -349,7 +349,7 @@ defGetStringList(DefElem *def)
> > (errcode(ERRCODE_SYNTAX_ERROR),
> > errmsg("%s requires a parameter",
> > def->defname)));
> > - if (nodeTag(def->arg) != T_List)
> > + if (!IsA(def->arg, List))
> > elog(ERROR, "unrecognized node type: %d", (int)
> > nodeTag(def->arg));
> >
> > foreach(cell, (List *) def->arg)
> >
>
> Yep, I did a search with `nodeTag\(.*\)\s+(?:!=|==).*`, and this is the only finding.
Oh, I missed that, sorry. Thanks, LGTM.
--
Best regards,
Shinya Kato
NTT OSS Center
^ permalink raw reply [nested|flat] 49+ messages in thread
end of thread, other threads:[~2026-01-09 02:19 UTC | newest]
Thread overview: 49+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v8 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v6 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-01-20 22:06 [PATCH v7 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2021-10-18 10:59 [PATCH v10 3/3] Optimize hash function Heikki Linnakangas <[email protected]>
2026-01-09 00:56 Re: Use IsA() macro instead of nodeTag comparison Chao Li <[email protected]>
2026-01-09 02:19 ` Re: Use IsA() macro instead of nodeTag comparison Shinya Kato <[email protected]>
This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox