agora inbox for [email protected]  

help / color / mirror / Atom feed

[PATCH 2/2] Add the "dummy_table_am" test module
268+ messages / 2 participants
[nested] [flat]

* [PATCH 2/2] Add the "dummy_table_am" test module
@ 2025-03-01 19:50 Julien Tachoires <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Julien Tachoires @ 2025-03-01 19:50 UTC (permalink / raw)

This test module is in charge of testing TAM reloptions. It's very
similar to what we do in dummy_index_am as we have to exercise the
exact same kind of feature.
---
 src/test/modules/Makefile                     |   1 +
 src/test/modules/dummy_table_am/Makefile      |  20 +
 src/test/modules/dummy_table_am/README        |  14 +
 .../dummy_table_am/dummy_table_am--1.0.sql    |  13 +
 .../modules/dummy_table_am/dummy_table_am.c   | 588 ++++++++++++++++++
 .../dummy_table_am/dummy_table_am.control     |   5 +
 .../dummy_table_am/expected/reloptions.out    | 181 ++++++
 src/test/modules/dummy_table_am/meson.build   |  33 +
 .../modules/dummy_table_am/sql/reloptions.sql |  99 +++
 src/test/modules/meson.build                  |   1 +
 10 files changed, 955 insertions(+)
 create mode 100644 src/test/modules/dummy_table_am/Makefile
 create mode 100644 src/test/modules/dummy_table_am/README
 create mode 100644 src/test/modules/dummy_table_am/dummy_table_am--1.0.sql
 create mode 100644 src/test/modules/dummy_table_am/dummy_table_am.c
 create mode 100644 src/test/modules/dummy_table_am/dummy_table_am.control
 create mode 100644 src/test/modules/dummy_table_am/expected/reloptions.out
 create mode 100644 src/test/modules/dummy_table_am/meson.build
 create mode 100644 src/test/modules/dummy_table_am/sql/reloptions.sql

diff --git a/src/test/modules/Makefile b/src/test/modules/Makefile
index 4e4be3fa511..8fe2a2904d6 100644
--- a/src/test/modules/Makefile
+++ b/src/test/modules/Makefile
@@ -9,6 +9,7 @@ SUBDIRS = \
 		  commit_ts \
 		  delay_execution \
 		  dummy_index_am \
+		  dummy_table_am \
 		  dummy_seclabel \
 		  libpq_pipeline \
 		  oauth_validator \
diff --git a/src/test/modules/dummy_table_am/Makefile b/src/test/modules/dummy_table_am/Makefile
new file mode 100644
index 00000000000..94837dff392
--- /dev/null
+++ b/src/test/modules/dummy_table_am/Makefile
@@ -0,0 +1,20 @@
+# src/test/modules/dummy_table_am/Makefile
+
+MODULES = dummy_table_am
+
+EXTENSION = dummy_table_am
+DATA = dummy_table_am--1.0.sql
+PGFILEDESC = "dummy_table_am - table access method template"
+
+REGRESS = reloptions
+
+ifdef USE_PGXS
+PG_CONFIG = pg_config
+PGXS := $(shell $(PG_CONFIG) --pgxs)
+include $(PGXS)
+else
+subdir = src/test/modules/dummy_table_am
+top_builddir = ../../../..
+include $(top_builddir)/src/Makefile.global
+include $(top_srcdir)/contrib/contrib-global.mk
+endif
diff --git a/src/test/modules/dummy_table_am/README b/src/test/modules/dummy_table_am/README
new file mode 100644
index 00000000000..50cf08ee3b1
--- /dev/null
+++ b/src/test/modules/dummy_table_am/README
@@ -0,0 +1,14 @@
+Dummy Table AM
+==============
+
+Dummy table AM is a module for testing any facility usable by a table
+access method, whose code is kept a maximum simple.
+
+This includes tests for all relation option types:
+- boolean
+- enum
+- integer
+- real
+- strings (with and without NULL as default)
+
+It also includes tests related to unrecognized options.
diff --git a/src/test/modules/dummy_table_am/dummy_table_am--1.0.sql b/src/test/modules/dummy_table_am/dummy_table_am--1.0.sql
new file mode 100644
index 00000000000..12ad3ad174b
--- /dev/null
+++ b/src/test/modules/dummy_table_am/dummy_table_am--1.0.sql
@@ -0,0 +1,13 @@
+/* src/test/modules/dummy_table_am/dummy_table_am--1.0.sql */
+
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "CREATE EXTENSION dummy_table_am" to load this file. \quit
+
+CREATE FUNCTION dummy_table_am_handler(internal)
+RETURNS table_am_handler
+AS 'MODULE_PATHNAME'
+LANGUAGE C;
+
+-- Access method
+CREATE ACCESS METHOD dummy_table_am TYPE TABLE HANDLER dummy_table_am_handler;
+COMMENT ON ACCESS METHOD dummy_table_am IS 'Dummy Table Access Method';
diff --git a/src/test/modules/dummy_table_am/dummy_table_am.c b/src/test/modules/dummy_table_am/dummy_table_am.c
new file mode 100644
index 00000000000..a473bc7dd9b
--- /dev/null
+++ b/src/test/modules/dummy_table_am/dummy_table_am.c
@@ -0,0 +1,588 @@
+/*-------------------------------------------------------------------------
+ *
+ * dummy_table_am.c
+ *		Table AM templae main file
+ *
+ * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * IDENTIFICATION
+ *	  src/test/modules/dummy_table_am/dummy_table_am.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#include "fmgr.h"
+#include "miscadmin.h"
+
+#include "access/hio.h"
+#include "access/relscan.h"
+#include "access/reloptions.h"
+#include "access/tableam.h"
+#include "access/sdir.h"
+#include "access/skey.h"
+#include "executor/tuptable.h"
+#include "utils/relcache.h"
+#include "utils/snapshot.h"
+
+
+PG_MODULE_MAGIC;
+
+/* Base structures for scans */
+typedef struct DummyScanDescData
+{
+	TableScanDescData rs_base;	/* AM independent part of the descriptor */
+
+	/* Add more fields here as needed by the AM. */
+}			DummyScanDescData;
+typedef struct DummyScanDescData *DummyScanDesc;
+
+/* parse table for fillRelOptions */
+static relopt_parse_elt dt_relopt_tab[7];
+
+/* Kind of relation options for dummy index */
+static relopt_kind dt_relopt_kind;
+
+typedef enum DummyAmEnum
+{
+	DUMMY_AM_ENUM_ONE,
+	DUMMY_AM_ENUM_TWO,
+}			DummyAmEnum;
+
+/* Dummy table options */
+typedef struct DummyTableOptions
+{
+	int32		vl_len_;		/* varlena header (do not touch directly!) */
+	int			option_int;
+	double		option_real;
+	bool		option_bool;
+	DummyAmEnum option_enum;
+	int			option_string_val_offset;
+	int			option_string_null_offset;
+	int			fillfactor;
+}			DummyTableOptions;
+
+static relopt_enum_elt_def dummyAmEnumValues[] =
+{
+	{"one", DUMMY_AM_ENUM_ONE},
+	{"two", DUMMY_AM_ENUM_TWO},
+	{(const char *) NULL}		/* list terminator */
+};
+
+/* ------------------------------------------------------------------------
+ *                     Dummy Access Method Interface
+ * ------------------------------------------------------------------------
+ */
+
+static const TupleTableSlotOps *
+dummy_slot_callbacks(Relation relation)
+{
+	return &TTSOpsMinimalTuple;
+}
+
+static TableScanDesc
+dummy_scan_begin(Relation relation, Snapshot snapshot, int nkeys, ScanKey key,
+				 ParallelTableScanDesc parallel_scan, uint32 flags)
+{
+	DummyScanDesc scan;
+
+	scan = (DummyScanDesc) palloc(sizeof(DummyScanDescData));
+
+	scan->rs_base.rs_rd = relation;
+	scan->rs_base.rs_snapshot = snapshot;
+	scan->rs_base.rs_nkeys = nkeys;
+	scan->rs_base.rs_flags = flags;
+	scan->rs_base.rs_parallel = parallel_scan;
+
+	return (TableScanDesc) scan;
+}
+
+static void
+dummy_scan_end(TableScanDesc sscan)
+{
+	DummyScanDesc scan = (DummyScanDesc) sscan;
+
+	pfree(scan);
+
+	return;
+}
+
+static void
+dummy_scan_rescan(TableScanDesc sscan, ScanKey key, bool set_params,
+				  bool allow_strat, bool allow_sync, bool allow_pagemode)
+{
+	return;
+}
+
+static bool
+dummy_scan_getnextslot(TableScanDesc sscan, ScanDirection direction,
+					   TupleTableSlot *slot)
+{
+	return true;
+}
+
+static void
+dummy_scan_set_tidrange(TableScanDesc sscan, ItemPointer mintid,
+						ItemPointer maxtid)
+{
+	return;
+}
+
+static bool
+dummy_scan_getnextslot_tidrange(TableScanDesc sscan, ScanDirection direction,
+								TupleTableSlot *slot)
+{
+	return true;
+}
+
+static Size
+dummy_parallelscan_estimate(Relation rel)
+{
+	return 0;
+}
+
+static Size
+dummy_parallelscan_initialize(Relation rel, ParallelTableScanDesc pscan)
+{
+	return 0;
+}
+
+static void
+dummy_parallelscan_reinitialize(Relation rel, ParallelTableScanDesc pscan)
+{
+	return;
+}
+
+static IndexFetchTableData *
+dummy_index_fetch_begin(Relation rel)
+{
+	return NULL;
+}
+
+static void
+dummy_index_fetch_reset(IndexFetchTableData *scan)
+{
+	return;
+}
+
+static void
+dummy_index_fetch_end(IndexFetchTableData *scan)
+{
+	return;
+}
+
+static bool
+dummy_index_fetch_tuple(struct IndexFetchTableData *scan, ItemPointer tid,
+						Snapshot snapshot, TupleTableSlot *slot,
+						bool *call_again, bool *all_dead)
+{
+	return true;
+}
+
+static void
+dummy_tuple_insert(Relation relation, TupleTableSlot *slot, CommandId cid,
+				   int options, BulkInsertStateData *bistate)
+{
+	DummyTableOptions *relopts;
+
+	relopts = (DummyTableOptions *) relation->rd_options;
+
+	elog(NOTICE, "option_int=%d, option_real=%f, option_bool=%d, option_enum=%d",
+		 relopts->option_int, relopts->option_real, relopts->option_bool, relopts->option_enum);
+
+	return;
+}
+
+static void
+dummy_tuple_insert_speculative(Relation relation, TupleTableSlot *slot,
+							   CommandId cid, int options,
+							   BulkInsertStateData *bistate, uint32 specToken)
+{
+	return;
+}
+
+static void
+dummy_tuple_complete_speculative(Relation relation, TupleTableSlot *slot,
+								 uint32 specToken, bool succeeded)
+{
+	return;
+}
+
+static void
+dummy_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
+				   CommandId cid, int options, BulkInsertStateData *bistate)
+{
+	return;
+}
+
+static TM_Result
+dummy_tuple_delete(Relation relation, ItemPointer tid, CommandId cid,
+				   Snapshot snapshot, Snapshot crosscheck, bool wait,
+				   TM_FailureData *tmfd, bool changingPart)
+{
+	return TM_Ok;
+}
+
+static TM_Result
+dummy_tuple_update(Relation relation, ItemPointer otid, TupleTableSlot *slot,
+				   CommandId cid, Snapshot snapshot, Snapshot crosscheck,
+				   bool wait, TM_FailureData *tmfd,
+				   LockTupleMode *lockmode, TU_UpdateIndexes *update_indexes)
+{
+	return TM_Ok;
+}
+
+static TM_Result
+dummy_tuple_lock(Relation relation, ItemPointer tid, Snapshot snapshot,
+				 TupleTableSlot *slot, CommandId cid, LockTupleMode mode,
+				 LockWaitPolicy wait_policy, uint8 flags,
+				 TM_FailureData *tmfd)
+{
+	return TM_Ok;
+}
+
+static bool
+dummy_fetch_row_version(Relation relation, ItemPointer tid,
+						Snapshot snapshot, TupleTableSlot *slot)
+{
+	return false;
+}
+
+static void
+dummy_get_latest_tid(TableScanDesc sscan, ItemPointer tid)
+{
+	return;
+}
+
+static bool
+dummy_tuple_tid_valid(TableScanDesc scan, ItemPointer tid)
+{
+	return false;
+}
+
+static bool
+dummy_tuple_satisfies_snapshot(Relation rel, TupleTableSlot *slot,
+							   Snapshot snapshot)
+{
+	return false;
+}
+
+static TransactionId
+dummy_index_delete_tuples(Relation rel, TM_IndexDeleteOp *delstate)
+{
+	return InvalidTransactionId;
+}
+
+static void
+dummy_relation_set_new_filelocator(Relation rel,
+								   const RelFileLocator *newrlocator,
+								   char persistence,
+								   TransactionId *freezeXid,
+								   MultiXactId *minmulti)
+{
+	return;
+}
+
+static void
+dummy_relation_nontransactional_truncate(Relation rel)
+{
+	return;
+}
+
+static void
+dummy_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
+{
+	return;
+}
+
+static void
+dummy_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
+								Relation OldIndex, bool use_sort,
+								TransactionId OldestXmin,
+								TransactionId *xid_cutoff,
+								MultiXactId *multi_cutoff,
+								double *num_tuples,
+								double *tups_vacuumed,
+								double *tups_recently_dead)
+{
+	return;
+}
+
+static void
+dummy_relation_vacuum(Relation rel, struct VacuumParams *params,
+					  BufferAccessStrategy bstrategy)
+{
+	return;
+}
+
+static bool
+dummy_scan_analyze_next_block(TableScanDesc scan, ReadStream *stream)
+{
+	return false;
+}
+
+static bool
+dummy_scan_analyze_next_tuple(TableScanDesc scan, TransactionId OldestXmin,
+							  double *liverows, double *deadrows,
+							  TupleTableSlot *slot)
+{
+	return false;
+}
+
+static double
+dummy_index_build_range_scan(Relation heapRelation,
+							 Relation indexRelation,
+							 struct IndexInfo *indexInfo,
+							 bool allow_sync,
+							 bool anyvisible,
+							 bool progress,
+							 BlockNumber start_blockno,
+							 BlockNumber numblocks,
+							 IndexBuildCallback callback,
+							 void *callback_state,
+							 TableScanDesc scan)
+{
+	return 0;
+}
+
+static void
+dummy_index_validate_scan(Relation heapRelation,
+						  Relation indexRelation,
+						  struct IndexInfo *indexInfo,
+						  Snapshot snapshot,
+						  struct ValidateIndexState *state)
+{
+	return;
+}
+
+static uint64
+dummy_relation_size(Relation rel, ForkNumber forkNumber)
+{
+	return 0;
+}
+
+static bool
+dummy_relation_needs_toast_table(Relation rel)
+{
+	return false;
+}
+
+static Oid
+dummy_relation_toast_am(Relation rel)
+{
+	return InvalidOid;
+}
+
+static void
+dummy_relation_fetch_toast_slice(Relation toastrel, Oid valueid, int32 attrsize,
+								 int32 sliceoffset, int32 slicelength,
+								 struct varlena *result)
+{
+	return;
+}
+
+static void
+dummy_relation_estimate_size(Relation rel, int32 *attr_widths,
+							 BlockNumber *pages, double *tuples,
+							 double *allvisfrac)
+{
+	return;
+}
+
+static bool
+dummy_scan_bitmap_next_block(TableScanDesc scan, BlockNumber *blockno,
+							 bool *recheck, uint64 *lossy_pages,
+							 uint64 *exact_pages)
+{
+	return false;
+}
+
+static bool
+dummy_scan_bitmap_next_tuple(TableScanDesc scan, TupleTableSlot *slot)
+{
+	return false;
+}
+
+static bool
+dummy_scan_sample_next_block(TableScanDesc scan, struct SampleScanState *scanstate)
+{
+	return false;
+}
+
+static bool
+dummy_scan_sample_next_tuple(TableScanDesc scan, struct SampleScanState *scanstate,
+							 TupleTableSlot *slot)
+{
+	return false;
+}
+
+static bytea *
+dummy_relation_options(char relkind, Datum reloptions, bool validate)
+{
+	return (bytea *) build_reloptions(reloptions, validate,
+									  dt_relopt_kind,
+									  sizeof(DummyTableOptions),
+									  dt_relopt_tab, lengthof(dt_relopt_tab));
+}
+
+/*
+ * Validation function for string relation options.
+ */
+static void
+validate_string_option(const char *value)
+{
+	ereport(NOTICE,
+			(errmsg("new option value for string parameter %s",
+					value ? value : "NULL")));
+}
+
+/*
+ * This function creates a full set of relation option types,
+ * with various patterns.
+ */
+static void
+create_reloptions_table(void)
+{
+	dt_relopt_kind = add_reloption_kind();
+
+	add_int_reloption(dt_relopt_kind, "option_int",
+					  "Integer option for dummy_table_am",
+					  10, -10, 100, AccessExclusiveLock);
+	dt_relopt_tab[0].optname = "option_int";
+	dt_relopt_tab[0].opttype = RELOPT_TYPE_INT;
+	dt_relopt_tab[0].offset = offsetof(DummyTableOptions, option_int);
+
+	add_real_reloption(dt_relopt_kind, "option_real",
+					   "Real option for dummy_table_am",
+					   3.1415, -10, 100, AccessExclusiveLock);
+	dt_relopt_tab[1].optname = "option_real";
+	dt_relopt_tab[1].opttype = RELOPT_TYPE_REAL;
+	dt_relopt_tab[1].offset = offsetof(DummyTableOptions, option_real);
+
+	add_bool_reloption(dt_relopt_kind, "option_bool",
+					   "Boolean option for dummy_table_am",
+					   true, AccessExclusiveLock);
+	dt_relopt_tab[2].optname = "option_bool";
+	dt_relopt_tab[2].opttype = RELOPT_TYPE_BOOL;
+	dt_relopt_tab[2].offset = offsetof(DummyTableOptions, option_bool);
+
+	add_enum_reloption(dt_relopt_kind, "option_enum",
+					   "Enum option for dummy_table_am",
+					   dummyAmEnumValues,
+					   DUMMY_AM_ENUM_ONE,
+					   "Valid values are \"one\" and \"two\".",
+					   AccessExclusiveLock);
+	dt_relopt_tab[3].optname = "option_enum";
+	dt_relopt_tab[3].opttype = RELOPT_TYPE_ENUM;
+	dt_relopt_tab[3].offset = offsetof(DummyTableOptions, option_enum);
+
+	add_string_reloption(dt_relopt_kind, "option_string_val",
+						 "String option for dummy_table_am with non-NULL default",
+						 "DefaultValue", &validate_string_option,
+						 AccessExclusiveLock);
+	dt_relopt_tab[4].optname = "option_string_val";
+	dt_relopt_tab[4].opttype = RELOPT_TYPE_STRING;
+	dt_relopt_tab[4].offset = offsetof(DummyTableOptions,
+									   option_string_val_offset);
+
+	/*
+	 * String option for dummy_table_am with NULL default, and without
+	 * description.
+	 */
+	add_string_reloption(dt_relopt_kind, "option_string_null",
+						 NULL,	/* description */
+						 NULL, &validate_string_option,
+						 AccessExclusiveLock);
+	dt_relopt_tab[5].optname = "option_string_null";
+	dt_relopt_tab[5].opttype = RELOPT_TYPE_STRING;
+	dt_relopt_tab[5].offset = offsetof(DummyTableOptions,
+									   option_string_null_offset);
+
+	/*
+	 * fillfactor will be used to check reloption conversion when changing
+	 * table access method between heap AM and dummy_table_am.
+	 */
+	add_int_reloption(dt_relopt_kind, "fillfactor",
+					  "Fillfactor option for dummy_table_am",
+					  10, 0, 90, AccessExclusiveLock);
+	dt_relopt_tab[6].optname = "fillfactor";
+	dt_relopt_tab[6].opttype = RELOPT_TYPE_INT;
+	dt_relopt_tab[6].offset = offsetof(DummyTableOptions, fillfactor);
+}
+
+
+/*
+ * Table Access Method API
+ */
+static const TableAmRoutine dummy_table_am_methods = {
+	.type = T_TableAmRoutine,
+
+	.slot_callbacks = dummy_slot_callbacks,
+	.scan_begin = dummy_scan_begin,
+	.scan_end = dummy_scan_end,
+	.scan_rescan = dummy_scan_rescan,
+	.scan_getnextslot = dummy_scan_getnextslot,
+
+	.scan_set_tidrange = dummy_scan_set_tidrange,
+	.scan_getnextslot_tidrange = dummy_scan_getnextslot_tidrange,
+
+	.parallelscan_estimate = dummy_parallelscan_estimate,
+	.parallelscan_initialize = dummy_parallelscan_initialize,
+	.parallelscan_reinitialize = dummy_parallelscan_reinitialize,
+
+	.index_fetch_begin = dummy_index_fetch_begin,
+	.index_fetch_reset = dummy_index_fetch_reset,
+	.index_fetch_end = dummy_index_fetch_end,
+	.index_fetch_tuple = dummy_index_fetch_tuple,
+
+	.tuple_insert = dummy_tuple_insert,
+	.tuple_insert_speculative = dummy_tuple_insert_speculative,
+	.tuple_complete_speculative = dummy_tuple_complete_speculative,
+	.multi_insert = dummy_multi_insert,
+	.tuple_delete = dummy_tuple_delete,
+	.tuple_update = dummy_tuple_update,
+	.tuple_lock = dummy_tuple_lock,
+
+	.tuple_fetch_row_version = dummy_fetch_row_version,
+	.tuple_get_latest_tid = dummy_get_latest_tid,
+	.tuple_tid_valid = dummy_tuple_tid_valid,
+	.tuple_satisfies_snapshot = dummy_tuple_satisfies_snapshot,
+	.index_delete_tuples = dummy_index_delete_tuples,
+
+	.relation_set_new_filelocator = dummy_relation_set_new_filelocator,
+	.relation_nontransactional_truncate = dummy_relation_nontransactional_truncate,
+	.relation_copy_data = dummy_relation_copy_data,
+	.relation_copy_for_cluster = dummy_relation_copy_for_cluster,
+	.relation_vacuum = dummy_relation_vacuum,
+	.scan_analyze_next_block = dummy_scan_analyze_next_block,
+	.scan_analyze_next_tuple = dummy_scan_analyze_next_tuple,
+	.index_build_range_scan = dummy_index_build_range_scan,
+	.index_validate_scan = dummy_index_validate_scan,
+
+	.relation_size = dummy_relation_size,
+	.relation_needs_toast_table = dummy_relation_needs_toast_table,
+	.relation_toast_am = dummy_relation_toast_am,
+	.relation_fetch_toast_slice = dummy_relation_fetch_toast_slice,
+	.relation_estimate_size = dummy_relation_estimate_size,
+	.relation_options = dummy_relation_options,
+
+	.scan_bitmap_next_block = dummy_scan_bitmap_next_block,
+	.scan_bitmap_next_tuple = dummy_scan_bitmap_next_tuple,
+	.scan_sample_next_block = dummy_scan_sample_next_block,
+	.scan_sample_next_tuple = dummy_scan_sample_next_tuple
+};
+
+PG_FUNCTION_INFO_V1(dummy_table_am_handler);
+
+Datum
+dummy_table_am_handler(PG_FUNCTION_ARGS)
+{
+	PG_RETURN_POINTER(&dummy_table_am_methods);
+}
+
+void
+_PG_init(void)
+{
+	create_reloptions_table();
+}
diff --git a/src/test/modules/dummy_table_am/dummy_table_am.control b/src/test/modules/dummy_table_am/dummy_table_am.control
new file mode 100644
index 00000000000..08f2f868d49
--- /dev/null
+++ b/src/test/modules/dummy_table_am/dummy_table_am.control
@@ -0,0 +1,5 @@
+# dummy_table_am extension
+comment = 'dummy_table_am - table access method template'
+default_version = '1.0'
+module_pathname = '$libdir/dummy_table_am'
+relocatable = true
diff --git a/src/test/modules/dummy_table_am/expected/reloptions.out b/src/test/modules/dummy_table_am/expected/reloptions.out
new file mode 100644
index 00000000000..4c08ac4e3ac
--- /dev/null
+++ b/src/test/modules/dummy_table_am/expected/reloptions.out
@@ -0,0 +1,181 @@
+-- Tests for relation options
+CREATE EXTENSION dummy_table_am;
+CREATE TABLE dummy_test_tab (i int4) USING dummy_table_am;
+-- Silence validation checks for strings
+SET client_min_messages TO 'warning';
+-- Test with default values.
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+ unnest 
+--------
+(0 rows)
+
+DROP TABLE dummy_test_tab;
+-- Test with full set of options.
+-- Allow validation checks for strings
+SET client_min_messages TO 'notice';
+CREATE TABLE dummy_test_tab (i int4)
+  USING dummy_table_am WITH (
+  option_bool = false,
+  option_int = 5,
+  option_real = 3.1,
+  option_enum = 'two',
+  option_string_val = NULL,
+  option_string_null = 'val');
+NOTICE:  new option value for string parameter null
+NOTICE:  new option value for string parameter val
+-- Silence again validation checks for strings until the end of the test.
+SET client_min_messages TO 'warning';
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+         unnest         
+------------------------
+ option_bool=false
+ option_int=5
+ option_real=3.1
+ option_enum=two
+ option_string_val=null
+ option_string_null=val
+(6 rows)
+
+-- ALTER TABLE .. SET
+ALTER TABLE dummy_test_tab SET (option_int = 10);
+ALTER TABLE dummy_test_tab SET (option_bool = true);
+ALTER TABLE dummy_test_tab SET (option_real = 3.2);
+ALTER TABLE dummy_test_tab SET (option_string_val = 'val2');
+ALTER TABLE dummy_test_tab SET (option_string_null = NULL);
+ALTER TABLE dummy_test_tab SET (option_enum = 'one');
+ALTER TABLE dummy_test_tab SET (option_enum = 'three');
+ERROR:  invalid value for enum option "option_enum": three
+DETAIL:  Valid values are "one" and "two".
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+         unnest          
+-------------------------
+ option_int=10
+ option_bool=true
+ option_real=3.2
+ option_string_val=val2
+ option_string_null=null
+ option_enum=one
+(6 rows)
+
+-- ALTER TABLE .. RESET
+ALTER TABLE dummy_test_tab RESET (option_int);
+ALTER TABLE dummy_test_tab RESET (option_bool);
+ALTER TABLE dummy_test_tab RESET (option_real);
+ALTER TABLE dummy_test_tab RESET (option_enum);
+ALTER TABLE dummy_test_tab RESET (option_string_val);
+ALTER TABLE dummy_test_tab RESET (option_string_null);
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+ unnest 
+--------
+(0 rows)
+
+-- Cross-type checks for reloption values
+-- Integer
+ALTER TABLE dummy_test_tab SET (option_int = 3.3); -- ok
+ALTER TABLE dummy_test_tab SET (option_int = true); -- error
+ERROR:  invalid value for integer option "option_int": true
+ALTER TABLE dummy_test_tab SET (option_int = 'val3'); -- error
+ERROR:  invalid value for integer option "option_int": val3
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+     unnest     
+----------------
+ option_int=3.3
+(1 row)
+
+ALTER TABLE dummy_test_tab RESET (option_int);
+-- Boolean
+ALTER TABLE dummy_test_tab SET (option_bool = 4); -- error
+ERROR:  invalid value for boolean option "option_bool": 4
+ALTER TABLE dummy_test_tab SET (option_bool = 1); -- ok, as true
+ALTER TABLE dummy_test_tab SET (option_bool = 3.4); -- error
+ERROR:  invalid value for boolean option "option_bool": 3.4
+ALTER TABLE dummy_test_tab SET (option_bool = 'val4'); -- error
+ERROR:  invalid value for boolean option "option_bool": val4
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+    unnest     
+---------------
+ option_bool=1
+(1 row)
+
+ALTER TABLE dummy_test_tab RESET (option_bool);
+-- Float
+ALTER TABLE dummy_test_tab SET (option_real = 4); -- ok
+ALTER TABLE dummy_test_tab SET (option_real = true); -- error
+ERROR:  invalid value for floating point option "option_real": true
+ALTER TABLE dummy_test_tab SET (option_real = 'val5'); -- error
+ERROR:  invalid value for floating point option "option_real": val5
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+    unnest     
+---------------
+ option_real=4
+(1 row)
+
+ALTER TABLE dummy_test_tab RESET (option_real);
+-- Enum
+ALTER TABLE dummy_test_tab SET (option_enum = 'one'); -- ok
+ALTER TABLE dummy_test_tab SET (option_enum = 0); -- error
+ERROR:  invalid value for enum option "option_enum": 0
+DETAIL:  Valid values are "one" and "two".
+ALTER TABLE dummy_test_tab SET (option_enum = true); -- error
+ERROR:  invalid value for enum option "option_enum": true
+DETAIL:  Valid values are "one" and "two".
+ALTER TABLE dummy_test_tab SET (option_enum = 'three'); -- error
+ERROR:  invalid value for enum option "option_enum": three
+DETAIL:  Valid values are "one" and "two".
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+     unnest      
+-----------------
+ option_enum=one
+(1 row)
+
+ALTER TABLE dummy_test_tab RESET (option_enum);
+-- String
+ALTER TABLE dummy_test_tab SET (option_string_val = 4); -- ok
+ALTER TABLE dummy_test_tab SET (option_string_val = 3.5); -- ok
+ALTER TABLE dummy_test_tab SET (option_string_val = true); -- ok, as "true"
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+         unnest         
+------------------------
+ option_string_val=true
+(1 row)
+
+ALTER TABLE dummy_test_tab RESET (option_string_val);
+DROP TABLE dummy_test_tab;
+-- ALTER TABLE SET ACCESS METHOD OPTIONS
+CREATE TABLE heap_tab (i INT4) WITH (fillfactor=100, toast_tuple_target=1000);
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'heap_tab';
+         unnest          
+-------------------------
+ fillfactor=100
+ toast_tuple_target=1000
+(2 rows)
+
+-- error: fillfactor is out of bounds: maximum value from the new table am is 90
+ALTER TABLE heap_tab SET ACCESS METHOD dummy_table_am;
+ERROR:  value 100 out of bounds for option "fillfactor"
+DETAIL:  Valid values are between "0" and "90".
+-- error: toast_tuple_target does not exist in the new table AM
+ALTER TABLE heap_tab SET ACCESS METHOD dummy_table_am OPTIONS (SET fillfactor '50');
+ERROR:  unrecognized parameter "toast_tuple_target"
+-- error: adding is not possible when the parameter is already defined in source reloptions
+ALTER TABLE heap_tab SET ACCESS METHOD dummy_table_am OPTIONS (ADD fillfactor '50');
+ERROR:  option "fillfactor" provided more than once
+-- error: the specified option we want to drop does not exist 
+ALTER TABLE heap_tab SET ACCESS METHOD dummy_table_am OPTIONS (DROP does_not_exist);
+ERROR:  option "does_not_exist" not found
+-- error: adding unrecognized parameter
+ALTER TABLE heap_tab SET ACCESS METHOD dummy_table_am OPTIONS (SET fillfactor '50', DROP toast_tuple_target, ADD unrecognized 'foo');
+ERROR:  unrecognized parameter "unrecognized"
+-- ok
+ALTER TABLE heap_tab SET ACCESS METHOD dummy_table_am OPTIONS (DROP fillfactor, DROP toast_tuple_target, option_int '1', option_bool 'true', option_real '0.001', option_enum 'one', option_string_val 'hello');
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'heap_tab';
+         unnest          
+-------------------------
+ option_int=1
+ option_bool=true
+ option_real=0.001
+ option_enum=one
+ option_string_val=hello
+(5 rows)
+
+DROP TABLE heap_tab;
diff --git a/src/test/modules/dummy_table_am/meson.build b/src/test/modules/dummy_table_am/meson.build
new file mode 100644
index 00000000000..6b197b15ffa
--- /dev/null
+++ b/src/test/modules/dummy_table_am/meson.build
@@ -0,0 +1,33 @@
+# Copyright (c) 2022-2025, PostgreSQL Global Development Group
+
+dummy_table_am_sources = files(
+  'dummy_table_am.c',
+)
+
+if host_system == 'windows'
+  dummy_table_am_sources += rc_lib_gen.process(win32ver_rc, extra_args: [
+    '--NAME', 'dummy_table_am',
+    '--FILEDESC', 'dummy_table_am - table access method template',])
+endif
+
+dummy_table_am = shared_module('dummy_table_am',
+  dummy_table_am_sources,
+  kwargs: pg_test_mod_args,
+)
+test_install_libs += dummy_table_am
+
+test_install_data += files(
+  'dummy_table_am.control',
+  'dummy_table_am--1.0.sql',
+)
+
+tests += {
+  'name': 'dummy_table_am',
+  'sd': meson.current_source_dir(),
+  'bd': meson.current_build_dir(),
+  'regress': {
+    'sql': [
+      'reloptions',
+    ],
+  },
+}
diff --git a/src/test/modules/dummy_table_am/sql/reloptions.sql b/src/test/modules/dummy_table_am/sql/reloptions.sql
new file mode 100644
index 00000000000..ce02533d42e
--- /dev/null
+++ b/src/test/modules/dummy_table_am/sql/reloptions.sql
@@ -0,0 +1,99 @@
+-- Tests for relation options
+CREATE EXTENSION dummy_table_am;
+
+CREATE TABLE dummy_test_tab (i int4) USING dummy_table_am;
+
+-- Silence validation checks for strings
+SET client_min_messages TO 'warning';
+
+-- Test with default values.
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+DROP TABLE dummy_test_tab;
+
+-- Test with full set of options.
+-- Allow validation checks for strings
+SET client_min_messages TO 'notice';
+CREATE TABLE dummy_test_tab (i int4)
+  USING dummy_table_am WITH (
+  option_bool = false,
+  option_int = 5,
+  option_real = 3.1,
+  option_enum = 'two',
+  option_string_val = NULL,
+  option_string_null = 'val');
+-- Silence again validation checks for strings until the end of the test.
+SET client_min_messages TO 'warning';
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+
+-- ALTER TABLE .. SET
+ALTER TABLE dummy_test_tab SET (option_int = 10);
+ALTER TABLE dummy_test_tab SET (option_bool = true);
+ALTER TABLE dummy_test_tab SET (option_real = 3.2);
+ALTER TABLE dummy_test_tab SET (option_string_val = 'val2');
+ALTER TABLE dummy_test_tab SET (option_string_null = NULL);
+ALTER TABLE dummy_test_tab SET (option_enum = 'one');
+ALTER TABLE dummy_test_tab SET (option_enum = 'three');
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+
+-- ALTER TABLE .. RESET
+ALTER TABLE dummy_test_tab RESET (option_int);
+ALTER TABLE dummy_test_tab RESET (option_bool);
+ALTER TABLE dummy_test_tab RESET (option_real);
+ALTER TABLE dummy_test_tab RESET (option_enum);
+ALTER TABLE dummy_test_tab RESET (option_string_val);
+ALTER TABLE dummy_test_tab RESET (option_string_null);
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+
+-- Cross-type checks for reloption values
+-- Integer
+ALTER TABLE dummy_test_tab SET (option_int = 3.3); -- ok
+ALTER TABLE dummy_test_tab SET (option_int = true); -- error
+ALTER TABLE dummy_test_tab SET (option_int = 'val3'); -- error
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+ALTER TABLE dummy_test_tab RESET (option_int);
+-- Boolean
+ALTER TABLE dummy_test_tab SET (option_bool = 4); -- error
+ALTER TABLE dummy_test_tab SET (option_bool = 1); -- ok, as true
+ALTER TABLE dummy_test_tab SET (option_bool = 3.4); -- error
+ALTER TABLE dummy_test_tab SET (option_bool = 'val4'); -- error
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+ALTER TABLE dummy_test_tab RESET (option_bool);
+-- Float
+ALTER TABLE dummy_test_tab SET (option_real = 4); -- ok
+ALTER TABLE dummy_test_tab SET (option_real = true); -- error
+ALTER TABLE dummy_test_tab SET (option_real = 'val5'); -- error
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+ALTER TABLE dummy_test_tab RESET (option_real);
+-- Enum
+ALTER TABLE dummy_test_tab SET (option_enum = 'one'); -- ok
+ALTER TABLE dummy_test_tab SET (option_enum = 0); -- error
+ALTER TABLE dummy_test_tab SET (option_enum = true); -- error
+ALTER TABLE dummy_test_tab SET (option_enum = 'three'); -- error
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+ALTER TABLE dummy_test_tab RESET (option_enum);
+-- String
+ALTER TABLE dummy_test_tab SET (option_string_val = 4); -- ok
+ALTER TABLE dummy_test_tab SET (option_string_val = 3.5); -- ok
+ALTER TABLE dummy_test_tab SET (option_string_val = true); -- ok, as "true"
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'dummy_test_tab';
+ALTER TABLE dummy_test_tab RESET (option_string_val);
+
+DROP TABLE dummy_test_tab;
+
+-- ALTER TABLE SET ACCESS METHOD OPTIONS
+CREATE TABLE heap_tab (i INT4) WITH (fillfactor=100, toast_tuple_target=1000);
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'heap_tab';
+-- error: fillfactor is out of bounds: maximum value from the new table am is 90
+ALTER TABLE heap_tab SET ACCESS METHOD dummy_table_am;
+-- error: toast_tuple_target does not exist in the new table AM
+ALTER TABLE heap_tab SET ACCESS METHOD dummy_table_am OPTIONS (SET fillfactor '50');
+-- error: adding is not possible when the parameter is already defined in source reloptions
+ALTER TABLE heap_tab SET ACCESS METHOD dummy_table_am OPTIONS (ADD fillfactor '50');
+-- error: the specified option we want to drop does not exist 
+ALTER TABLE heap_tab SET ACCESS METHOD dummy_table_am OPTIONS (DROP does_not_exist);
+-- error: adding unrecognized parameter
+ALTER TABLE heap_tab SET ACCESS METHOD dummy_table_am OPTIONS (SET fillfactor '50', DROP toast_tuple_target, ADD unrecognized 'foo');
+-- ok
+ALTER TABLE heap_tab SET ACCESS METHOD dummy_table_am OPTIONS (DROP fillfactor, DROP toast_tuple_target, option_int '1', option_bool 'true', option_real '0.001', option_enum 'one', option_string_val 'hello');
+SELECT unnest(reloptions) FROM pg_class WHERE relname = 'heap_tab';
+DROP TABLE heap_tab;
diff --git a/src/test/modules/meson.build b/src/test/modules/meson.build
index 2b057451473..28398254df7 100644
--- a/src/test/modules/meson.build
+++ b/src/test/modules/meson.build
@@ -4,6 +4,7 @@ subdir('brin')
 subdir('commit_ts')
 subdir('delay_execution')
 subdir('dummy_index_am')
+subdir('dummy_table_am')
 subdir('dummy_seclabel')
 subdir('gin')
 subdir('injection_points')
-- 
2.39.5


--y52row5sua5l45qt--





^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread

* [PATCH v12 2/3] Use time stamp counter to measure time on Linux/x86
@ 2025-07-26 00:57 Lukas Fittl <[email protected]>
  0 siblings, 0 replies; 268+ messages in thread

From: Lukas Fittl @ 2025-07-26 00:57 UTC (permalink / raw)

We switch to using the time stamp counter (TSC) instead of clock_gettime()
to reduce overhead of EXPLAIN (ANALYZE, TIME ON). Tests showed that runtime
is reduced by around 10% for queries moving lots of rows through the plan.

For now this is only enabled on Linux/x86, in case the system clocksource is
reported as TSC. Relying on the Linux kernel simplifies the logic to detect
if the present TSC is usable (frequency invariant, synchronized between
sockets, etc.). In all other cases we fallback to clock_gettime().

Note, that we intentionally use RDTSC in the fast paths, rather than RDTSCP.
RDTSCP waits for outstanding instructions to retire on out-of-order CPUs.
This adds noticably for little benefit in the typical InstrStartNode() /
InstrStopNode() use case. The macro to be used in such cases is called
INSTR_TIME_SET_CURRENT_FAST(). The original macro INSTR_TIME_SET_CURRENT()
uses RDTSCP and is supposed to be used when precision is more important
than performance.

Author: David Geier <[email protected]>
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Reviewed-by:
Discussion: https://www.postgresql.org/message-id/flat/20200612232810.f46nbqkdhbutzqdg%40alap3.anarazel.de
---
 src/backend/access/heap/vacuumlazy.c |   4 +-
 src/backend/executor/instrument.c    |  12 +-
 src/backend/utils/init/postinit.c    |   3 +
 src/bin/pgbench/pgbench.c            |   3 +
 src/bin/psql/startup.c               |   4 +
 src/common/Makefile                  |   1 +
 src/common/instr_time.c              | 206 +++++++++++++++++++++++++++
 src/common/meson.build               |   1 +
 src/include/portability/instr_time.h | 136 +++++++++++++++---
 9 files changed, 348 insertions(+), 22 deletions(-)
 create mode 100644 src/common/instr_time.c

diff --git a/src/backend/access/heap/vacuumlazy.c b/src/backend/access/heap/vacuumlazy.c
index d2b031fdd06..5027048cac4 100644
--- a/src/backend/access/heap/vacuumlazy.c
+++ b/src/backend/access/heap/vacuumlazy.c
@@ -3409,8 +3409,8 @@ count_nondeletable_pages(LVRelState *vacrel, bool *lock_waiter_detected)
 			INSTR_TIME_SET_CURRENT(currenttime);
 			elapsed = currenttime;
 			INSTR_TIME_SUBTRACT(elapsed, starttime);
-			if ((INSTR_TIME_GET_MICROSEC(elapsed) / 1000)
-				>= VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
+			if (INSTR_TIME_GET_MILLISEC(elapsed) >=
+				VACUUM_TRUNCATE_LOCK_CHECK_INTERVAL)
 			{
 				if (LockHasWaitersRelation(vacrel->rel, AccessExclusiveLock))
 				{
diff --git a/src/backend/executor/instrument.c b/src/backend/executor/instrument.c
index 56e635f4700..01f67c5d972 100644
--- a/src/backend/executor/instrument.c
+++ b/src/backend/executor/instrument.c
@@ -67,9 +67,13 @@ InstrInit(Instrumentation *instr, int instrument_options)
 void
 InstrStartNode(Instrumentation *instr)
 {
-	if (instr->need_timer &&
-		!INSTR_TIME_SET_CURRENT_LAZY(instr->starttime))
-		elog(ERROR, "InstrStartNode called twice in a row");
+	if (instr->need_timer)
+	{
+		if (!INSTR_TIME_IS_ZERO(instr->starttime))
+			elog(ERROR, "InstrStartNode called twice in a row");
+		else
+			INSTR_TIME_SET_CURRENT_FAST(instr->starttime);
+	}
 
 	/* save buffer usage totals at node entry, if needed */
 	if (instr->need_bufusage)
@@ -95,7 +99,7 @@ InstrStopNode(Instrumentation *instr, double nTuples)
 		if (INSTR_TIME_IS_ZERO(instr->starttime))
 			elog(ERROR, "InstrStopNode called without start");
 
-		INSTR_TIME_SET_CURRENT(endtime);
+		INSTR_TIME_SET_CURRENT_FAST(endtime);
 		INSTR_TIME_ACCUM_DIFF(instr->counter, endtime, instr->starttime);
 
 		INSTR_TIME_SET_ZERO(instr->starttime);
diff --git a/src/backend/utils/init/postinit.c b/src/backend/utils/init/postinit.c
index 641e535a73c..d573409903b 100644
--- a/src/backend/utils/init/postinit.c
+++ b/src/backend/utils/init/postinit.c
@@ -810,6 +810,9 @@ InitPostgres(const char *in_dbname, Oid dboid,
 	/* Initialize portal manager */
 	EnablePortalManager();
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/*
 	 * Load relcache entries for the shared system catalogs.  This must create
 	 * at least entries for pg_database and catalogs used for authentication.
diff --git a/src/bin/pgbench/pgbench.c b/src/bin/pgbench/pgbench.c
index 1515ed405ba..79bef2d2aec 100644
--- a/src/bin/pgbench/pgbench.c
+++ b/src/bin/pgbench/pgbench.c
@@ -7290,6 +7290,9 @@ main(int argc, char **argv)
 		initRandomState(&state[i].cs_func_rs);
 	}
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	/* opening connection... */
 	con = doConnect();
 	if (con == NULL)
diff --git a/src/bin/psql/startup.c b/src/bin/psql/startup.c
index 249b6aa5169..d615df593c7 100644
--- a/src/bin/psql/startup.c
+++ b/src/bin/psql/startup.c
@@ -24,6 +24,7 @@
 #include "help.h"
 #include "input.h"
 #include "mainloop.h"
+#include "portability/instr_time.h"
 #include "settings.h"
 
 /*
@@ -327,6 +328,9 @@ main(int argc, char *argv[])
 
 	PQsetNoticeProcessor(pset.db, NoticeProcessor, NULL);
 
+	/* initialize high-precision interval timing */
+	INSTR_TIME_INITIALIZE();
+
 	SyncVariables();
 
 	if (options.list_dbs)
diff --git a/src/common/Makefile b/src/common/Makefile
index 2c720caa509..1a2fbbe887f 100644
--- a/src/common/Makefile
+++ b/src/common/Makefile
@@ -59,6 +59,7 @@ OBJS_COMMON = \
 	file_perm.o \
 	file_utils.o \
 	hashfn.o \
+	instr_time.o \
 	ip.o \
 	jsonapi.o \
 	keywords.o \
diff --git a/src/common/instr_time.c b/src/common/instr_time.c
new file mode 100644
index 00000000000..fdf47699f20
--- /dev/null
+++ b/src/common/instr_time.c
@@ -0,0 +1,206 @@
+/*-------------------------------------------------------------------------
+ *
+ * instr_time.c
+ *	   Non-inline parts of the portable high-precision interval timing
+ *	 implementation
+ *
+ * Portions Copyright (c) 2025, PostgreSQL Global Development Group
+ *
+ *
+ * IDENTIFICATION
+ *	  src/backend/port/instr_time.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#if defined(HAVE__GET_CPUID) || (defined(HAVE__CPUIDEX) && !defined(_MSC_VER))
+#include <cpuid.h>
+#endif
+
+#if defined(HAVE__CPUID) || (defined(HAVE__CPUIDEX) && defined(_MSC_VER))
+#include <intrin.h>
+#endif
+
+#include "portability/instr_time.h"
+
+#ifndef WIN32
+/*
+ * Stores what the number of cycles needs to be multiplied with to end up
+ * with nanoseconds using integer math. See comment in pg_initialize_rdtsc()
+ * for more details.
+ *
+ * By default assume we are using clock_gettime() as a fallback which uses
+ * nanoseconds as ticks. Hence, we set the multiplier to the precision scalar
+ * so that the division in INSTR_TIME_GET_NANOSEC() won't change the nanoseconds.
+ *
+ * When using the RDTSC instruction directly this is filled in during initialization
+ * based on the relevant CPUID fields.
+ */
+int64		ticks_per_ns_scaled = TICKS_TO_NS_PRECISION;
+int64		ticks_per_sec = NS_PER_S;
+int64		max_ticks_no_overflow = PG_INT64_MAX / TICKS_TO_NS_PRECISION;
+
+#if defined(__x86_64__) && defined(__linux__)
+/*
+ * Indicates if RDTSC can be used (Linux/x86 only, when OS uses TSC clocksource)
+ */
+bool		has_rdtsc = false;
+
+/*
+ * Indicates if RDTSCP can be used. True if RDTSC can be used and RDTSCP is available.
+ */
+bool		has_rdtscp = false;
+
+#define CPUID_HYPERVISOR_VMWARE(words) (words[1] == 0x61774d56 && words[2] == 0x4d566572 && words[3] == 0x65726177) /* VMwareVMware */
+#define CPUID_HYPERVISOR_KVM(words) (words[1] == 0x4b4d564b && words[2] == 0x564b4d56 && words[3] == 0x0000004d)	/* KVMKVMKVM */
+
+static bool
+get_tsc_frequency_khz(uint32 *tsc_freq)
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x15, &r[0] /* denominator */ , &r[1] /* numerator */ , &r[2] /* hz */ , &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x15);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[2] > 0)
+	{
+		if (r[0] == 0 || r[1] == 0)
+			return false;
+
+		*tsc_freq = r[2] / 1000 * r[1] / r[0];
+		return true;
+	}
+
+	/* Some CPUs only report frequency in 16H */
+
+#if defined(HAVE__GET_CPUID)
+	__get_cpuid(0x16, &r[0] /* base_mhz */ , &r[1], &r[2], &r[3]);
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x16);
+#else
+#error cpuid instruction not available
+#endif
+
+	if (r[0] > 0)
+	{
+		*tsc_freq = r[0] * 1000;
+		return true;
+	}
+
+	/*
+	 * Check if we have a KVM or VMware Hypervisor passing down TSC frequency
+	 * to us in a guest VM
+	 *
+	 * Note that accessing the 0x40000000 leaf for Hypervisor info requires
+	 * use of __cpuidex to set ECX to 0. The similar __get_cpuid_count
+	 * function does not work as expected since it contains a check for
+	 * __get_cpuid_max, which has been observed to be lower than the special
+	 * Hypervisor leaf.
+	 */
+#if defined(HAVE__CPUIDEX)
+	__cpuidex((int32 *) r, 0x40000000, 0);
+	if (r[0] >= 0x40000010 && (CPUID_HYPERVISOR_VMWARE(r) || CPUID_HYPERVISOR_KVM(r)))
+	{
+		__cpuidex((int32 *) r, 0x40000010, 0);
+		if (r[0] > 0)
+		{
+			*tsc_freq = r[0];
+			return true;
+		}
+	}
+#endif
+
+	return false;
+}
+
+static bool
+is_rdtscp_available()
+{
+	uint32		r[4] = {0, 0, 0, 0};
+
+#if defined(HAVE__GET_CPUID)
+	if (!__get_cpuid(0x80000001, &r[0], &r[1], &r[2], &r[3]))
+		return false;
+#elif defined(HAVE__CPUID)
+	__cpuid(r, 0x80000001);
+#else
+#error cpuid instruction not available
+#endif
+
+	return (r[3] & (1 << 27)) != 0;
+}
+
+/*
+ * Decide whether we use the RDTSC instruction at runtime, for Linux/x86,
+ * instead of incurring the overhead of a full clock_gettime() call.
+ *
+ * This can't be reliably determined at compile time, since the
+ * availability of an "invariant" TSC (that is not affected by CPU
+ * frequency changes) is dependent on the CPU architecture. Additionally,
+ * there are cases where TSC availability is impacted by virtualization,
+ * where a simple cpuid feature check would not be enough.
+ *
+ * Since Linux already does a significant amount of work to determine
+ * whether TSC is a viable clock source, decide based on that.
+ */
+void
+pg_initialize_rdtsc(void)
+{
+	FILE	   *fp = fopen("/sys/devices/system/clocksource/clocksource0/current_clocksource", "r");
+
+	if (fp)
+	{
+		char		buf[128];
+
+		if (fgets(buf, sizeof(buf), fp) != NULL && strcmp(buf, "tsc\n") == 0)
+		{
+			/*
+			 * Compute baseline CPU peformance, determines speed at which
+			 * RDTSC advances.
+			 */
+			uint32		tsc_freq;
+
+			if (get_tsc_frequency_khz(&tsc_freq))
+			{
+				/*
+				 * Ticks to nanoseconds conversion requires floating point
+				 * math because because:
+				 *
+				 * sec = ticks / frequency_hz ns  = ticks / frequency_hz *
+				 * 1,000,000,000 ns  = ticks * (1,000,000,000 / frequency_hz)
+				 * ns  = ticks * (1,000,000 / frequency_khz) <-- now in
+				 * kilohertz
+				 *
+				 * Here, 'ns' is usually a floating number. For example for a
+				 * 2.5 GHz CPU the scaling factor becomes 1,000,000 /
+				 * 2,500,000 = 1.2.
+				 *
+				 * To be able to use integer math we work around the lack of
+				 * precision. We first scale the integer up and after the
+				 * multiplication by the number of ticks in
+				 * INSTR_TIME_GET_NANOSEC() we divide again by the same value.
+				 * We picked the scaler such that it provides enough precision
+				 * and is a power-of-two which allows for shifting instead of
+				 * doing an integer division.
+				 */
+				ticks_per_ns_scaled = INT64CONST(1000000) * TICKS_TO_NS_PRECISION / tsc_freq;
+				ticks_per_sec = tsc_freq * 1000;	/* KHz->Hz */
+				max_ticks_no_overflow = PG_INT64_MAX / ticks_per_ns_scaled;
+
+				has_rdtsc = true;
+				has_rdtscp = is_rdtscp_available();
+			}
+		}
+
+		fclose(fp);
+	}
+}
+#endif							/* defined(__x86_64__) && defined(__linux__) */
+
+#endif							/* WIN32 */
diff --git a/src/common/meson.build b/src/common/meson.build
index 1540ba67cca..62b90b3e609 100644
--- a/src/common/meson.build
+++ b/src/common/meson.build
@@ -13,6 +13,7 @@ common_sources = files(
   'file_perm.c',
   'file_utils.c',
   'hashfn.c',
+  'instr_time.c',
   'ip.c',
   'jsonapi.c',
   'keywords.c',
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index f71a851b18d..e2e339a0c4f 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -4,9 +4,11 @@
  *	  portable high-precision interval timing
  *
  * This file provides an abstraction layer to hide portability issues in
- * interval timing.  On Unix we use clock_gettime(), and on Windows we use
- * QueryPerformanceCounter().  These macros also give some breathing room to
- * use other high-precision-timing APIs.
+ * interval timing. On Linux/x86 we use the rdtsc instruction when a TSC
+ * clocksource is also used on the host OS.  Otherwise, and on other
+ * Unix-like systems we use clock_gettime() and on Windows we use
+ * QueryPerformanceCounter(). These macros also give some breathing
+ * room to use other high-precision-timing APIs.
  *
  * The basic data type is instr_time, which all callers should treat as an
  * opaque typedef.  instr_time can store either an absolute time (of
@@ -17,10 +19,11 @@
  *
  * INSTR_TIME_SET_ZERO(t)			set t to zero (memset is acceptable too)
  *
- * INSTR_TIME_SET_CURRENT(t)		set t to current time
+ * INSTR_TIME_SET_CURRENT_FAST(t)	set t to current time without waiting
+ * 									for instructions in out-of-order window
  *
- * INSTR_TIME_SET_CURRENT_LAZY(t)	set t to current time if t is zero,
- *									evaluates to whether t changed
+ * INSTR_TIME_SET_CURRENT(t)		set t to current time while waiting for
+ * 									instructions in OOO to retire
  *
  * INSTR_TIME_ADD(x, y)				x += y
  *
@@ -81,6 +84,15 @@ typedef struct instr_time
 
 #ifndef WIN32
 
+/*
+ * Make sure this is a power-of-two, so that the compiler can turn the
+ * multiplications and divisions into shifts.
+ */
+#define TICKS_TO_NS_PRECISION (1<<14)
+
+extern int64 ticks_per_ns_scaled;
+extern int64 ticks_per_sec;
+extern int64 max_ticks_no_overflow;
 
 /* Use clock_gettime() */
 
@@ -106,9 +118,18 @@ typedef struct instr_time
 #define PG_INSTR_CLOCK	CLOCK_REALTIME
 #endif
 
-/* helper for INSTR_TIME_SET_CURRENT */
+#if defined(__x86_64__) && defined(__linux__)
+#include <x86intrin.h>
+#include <cpuid.h>
+
+extern bool has_rdtsc;
+extern bool has_rdtscp;
+
+extern void pg_initialize_rdtsc(void);
+#endif
+
 static inline instr_time
-pg_clock_gettime_ns(void)
+pg_clock_gettime(void)
 {
 	instr_time	now;
 	struct timespec tmp;
@@ -119,11 +140,94 @@ pg_clock_gettime_ns(void)
 	return now;
 }
 
+static inline instr_time
+pg_get_ticks_fast(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtsc)
+	{
+		instr_time	now;
+
+		now.ticks = __rdtsc();
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline instr_time
+pg_get_ticks(void)
+{
+#if defined(__x86_64__) && defined(__linux__)
+	if (has_rdtscp)
+	{
+		instr_time	now;
+		uint32		unused;
+
+		now.ticks = __rdtscp(&unused);
+		return now;
+	}
+#endif
+
+	return pg_clock_gettime();
+}
+
+static inline int64_t
+pg_ticks_to_ns(instr_time t)
+{
+	/*
+	 * Would multiplication overflow? If so perform computation in two parts.
+	 * Check overflow without actually overflowing via: a * b > max <=> a >
+	 * max / b
+	 */
+	int64		ns = 0;
+
+	if (unlikely(t.ticks > max_ticks_no_overflow))
+	{
+		/*
+		 * Compute how often the maximum number of ticks fits completely into
+		 * the number of elapsed ticks and convert that number into
+		 * nanoseconds. Then multiply by the count to arrive at the final
+		 * value. In a 2nd step we adjust the number of elapsed ticks and
+		 * convert the remaining ticks.
+		 */
+		int64		count = t.ticks / max_ticks_no_overflow;
+		int64		max_ns = max_ticks_no_overflow * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+
+		ns = max_ns * count;
+
+		/*
+		 * Subtract the ticks that we now already accounted for, so that they
+		 * don't get counted twice.
+		 */
+		t.ticks -= count * max_ticks_no_overflow;
+		Assert(t.ticks >= 0);
+	}
+
+	ns += t.ticks * ticks_per_ns_scaled / TICKS_TO_NS_PRECISION;
+	return ns;
+}
+
+static inline void
+pg_initialize_get_ticks()
+{
+#if defined(__x86_64__) && defined(__linux__)
+	pg_initialize_rdtsc();
+#endif
+}
+
+#define INSTR_TIME_INITIALIZE() \
+	pg_initialize_get_ticks()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_get_ticks_fast())
+
 #define INSTR_TIME_SET_CURRENT(t) \
-	((t) = pg_clock_gettime_ns())
+	((t) = pg_get_ticks())
 
 #define INSTR_TIME_GET_NANOSEC(t) \
-	((int64) (t).ticks)
+	pg_ticks_to_ns(t)
 
 
 #else							/* WIN32 */
@@ -131,7 +235,7 @@ pg_clock_gettime_ns(void)
 
 /* Use QueryPerformanceCounter() */
 
-/* helper for INSTR_TIME_SET_CURRENT */
+/* helper for INSTR_TIME_SET_CURRENT / INSTR_TIME_SET_CURRENT_FAST */
 static inline instr_time
 pg_query_performance_counter(void)
 {
@@ -153,6 +257,11 @@ GetTimerFrequency(void)
 	return (double) f.QuadPart;
 }
 
+#define INSTR_TIME_INITIALIZE()
+
+#define INSTR_TIME_SET_CURRENT_FAST(t) \
+	((t) = pg_query_performance_counter())
+
 #define INSTR_TIME_SET_CURRENT(t) \
 	((t) = pg_query_performance_counter())
 
@@ -168,13 +277,8 @@ GetTimerFrequency(void)
 
 #define INSTR_TIME_IS_ZERO(t)	((t).ticks == 0)
 
-
 #define INSTR_TIME_SET_ZERO(t)	((t).ticks = 0)
 
-#define INSTR_TIME_SET_CURRENT_LAZY(t) \
-	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
-
-
 #define INSTR_TIME_ADD(x,y) \
 	((x).ticks += (y).ticks)
 
-- 
2.47.3


--vtqqtrpooseurzip
Content-Type: text/x-diff; charset=utf-8
Content-Disposition: attachment;
	filename="v12-0003-pg_test_timing-Add-fast-flag-to-test-fast-timing.patch"



^ permalink  raw  reply  [nested|flat] 268+ messages in thread