public inbox for [email protected]  

help / color / mirror / Atom feed

[PATCH 1/2] Allow TAP test to excecise tablespace.
10+ messages / 3 participants
[nested] [flat]

* [PATCH 1/2] Allow TAP test to excecise tablespace.
@ 2019-04-22 11:10 Kyotaro Horiguchi <[email protected]>
  0 siblings, 0 replies; 10+ messages in thread

From: Kyotaro Horiguchi @ 2019-04-22 11:10 UTC (permalink / raw)

To perform tablespace related checks, this patch lets
PostgresNode::backup have a new parameter "tablespace_mapping", and
make init_from_backup handle capable to handle a backup created using
tablespace_mapping.
---
 src/test/perl/PostgresNode.pm  | 10 ++++++++--
 src/test/perl/RecursiveCopy.pm | 33 +++++++++++++++++++++++++++++----
 2 files changed, 37 insertions(+), 6 deletions(-)

diff --git a/src/test/perl/PostgresNode.pm b/src/test/perl/PostgresNode.pm
index 76874141c5..59a939821d 100644
--- a/src/test/perl/PostgresNode.pm
+++ b/src/test/perl/PostgresNode.pm
@@ -540,13 +540,19 @@ target server since it isn't done by default.
 
 sub backup
 {
-	my ($self, $backup_name) = @_;
+	my ($self, $backup_name, %params) = @_;
 	my $backup_path = $self->backup_dir . '/' . $backup_name;
 	my $name        = $self->name;
+	my @rest = ();
+
+	if (defined $params{tablespace_mapping})
+	{
+		push(@rest, "--tablespace-mapping=$params{tablespace_mapping}");
+	}
 
 	print "# Taking pg_basebackup $backup_name from node \"$name\"\n";
 	TestLib::system_or_bail('pg_basebackup', '-D', $backup_path, '-h',
-		$self->host, '-p', $self->port, '--no-sync');
+		$self->host, '-p', $self->port, '--no-sync', @rest);
 	print "# Backup finished\n";
 	return;
 }
diff --git a/src/test/perl/RecursiveCopy.pm b/src/test/perl/RecursiveCopy.pm
index baf5d0ac63..c912ce412d 100644
--- a/src/test/perl/RecursiveCopy.pm
+++ b/src/test/perl/RecursiveCopy.pm
@@ -22,6 +22,7 @@ use warnings;
 use Carp;
 use File::Basename;
 use File::Copy;
+use TestLib;
 
 =pod
 
@@ -97,14 +98,38 @@ sub _copypath_recurse
 	# invoke the filter and skip all further operation if it returns false
 	return 1 unless &$filterfn($curr_path);
 
-	# Check for symlink -- needed only on source dir
-	# (note: this will fall through quietly if file is already gone)
-	croak "Cannot operate on symlink \"$srcpath\"" if -l $srcpath;
-
 	# Abort if destination path already exists.  Should we allow directories
 	# to exist already?
 	croak "Destination path \"$destpath\" already exists" if -e $destpath;
 
+	# Check for symlink -- needed only on source dir
+	# (note: this will fall through quietly if file is already gone)
+	if (-l $srcpath)
+	{
+		croak "Cannot operate on symlink \"$srcpath\""
+		  if ($srcpath !~ /\/(pg_tblspc\/[0-9]+)$/);
+
+		# We have mapped tablespaces. Copy them individually
+		my $linkname = $1;
+		my $tmpdir = TestLib::tempdir;
+		my $dstrealdir = TestLib::real_dir($tmpdir);
+		my $srcrealdir = readlink($srcpath);
+
+		opendir(my $dh, $srcrealdir);
+		while (readdir $dh)
+		{
+			next if (/^\.\.?$/);
+			my $spath = "$srcrealdir/$_";
+			my $dpath = "$dstrealdir/$_";
+
+			copypath($spath, $dpath);
+		}
+		closedir $dh;
+
+		symlink $dstrealdir, $destpath;
+		return 1;
+	}
+
 	# If this source path is a file, simply copy it to destination with the
 	# same name and we're done.
 	if (-f $srcpath)
-- 
2.16.3


----Next_Part(Mon_Apr_22_21_19_33_2019_510)--
Content-Type: Text/X-Patch; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
 filename="0002-Add-check-for-recovery-failure-caused-by-tablespace.patch"



^ permalink  raw  reply  [nested|flat] 10+ messages in thread

* [PATCH 2/5] Allow TAP test to excecise tablespace.
@ 2019-04-22 11:58 Kyotaro Horiguchi <[email protected]>
  0 siblings, 0 replies; 10+ messages in thread

From: Kyotaro Horiguchi @ 2019-04-22 11:58 UTC (permalink / raw)

To perform tablespace related checks, this patch lets
PostgresNode::backup have a new parameter "tablespace_mapping", and
make init_from_backup handle capable to handle a backup created using
tablespace_mapping.
---
 src/test/perl/PostgresNode.pm  | 45 +++++++++++++++++++++++++++++++++++++-----
 src/test/perl/RecursiveCopy.pm | 38 +++++++++++++++++++++++++++++++----
 2 files changed, 74 insertions(+), 9 deletions(-)

diff --git a/src/test/perl/PostgresNode.pm b/src/test/perl/PostgresNode.pm
index 76874141c5..e951acc461 100644
--- a/src/test/perl/PostgresNode.pm
+++ b/src/test/perl/PostgresNode.pm
@@ -157,6 +157,7 @@ sub new
 		_host    => $pghost,
 		_basedir => "$TestLib::tmp_check/t_${testname}_${name}_data",
 		_name    => $name,
+		_tablespaces => [],
 		_logfile_generation => 0,
 		_logfile_base       => "$TestLib::log_path/${testname}_${name}",
 		_logfile            => "$TestLib::log_path/${testname}_${name}.log"
@@ -342,6 +343,26 @@ sub backup_dir
 
 =pod
 
+=item $node->make_tablespace_dir()
+
+make a tablespace directory 
+
+=cut
+
+sub make_tablespace_dir
+{
+	my ($self, $name) = @_;
+	my $basedir = $self->basedir;
+
+	die "tablespace name contains '/'" if ($name =~ m#/#);
+	my $reldir = "../$name";
+	mkdir $self->data_dir() . "/". $reldir;
+	push($self->{_tablespaces}, $reldir);
+	return $reldir;
+}
+
+=pod
+
 =item $node->info()
 
 Return a string containing human-readable diagnostic information (paths, etc)
@@ -540,13 +561,27 @@ target server since it isn't done by default.
 
 sub backup
 {
-	my ($self, $backup_name) = @_;
-	my $backup_path = $self->backup_dir . '/' . $backup_name;
+	my ($self, $backup_name, %params) = @_;
+	my $backup_path = $self->backup_dir . '/' . $backup_name . '/' . "pgdata";
 	my $name        = $self->name;
+	my @rest = ();
+
+	if (defined $params{tablespace_mapping})
+	{
+		foreach my $p (split /,/, $params{tablespace_mapping})
+		{
+			push(@rest, "--tablespace-mapping=$p");
+		}
+	}
+
+	foreach my $p ($self->{_tablespaces})
+	{
+		mkdir "$backup_path/$p";
+	}
 
 	print "# Taking pg_basebackup $backup_name from node \"$name\"\n";
 	TestLib::system_or_bail('pg_basebackup', '-D', $backup_path, '-h',
-		$self->host, '-p', $self->port, '--no-sync');
+		$self->host, '-p', $self->port, '--no-sync', @rest);
 	print "# Backup finished\n";
 	return;
 }
@@ -592,7 +627,7 @@ sub backup_fs_cold
 sub _backup_fs
 {
 	my ($self, $backup_name, $hot) = @_;
-	my $backup_path = $self->backup_dir . '/' . $backup_name;
+	my $backup_path = $self->backup_dir . '/' . $backup_name . '/' . "pgdata";
 	my $port        = $self->port;
 	my $name        = $self->name;
 
@@ -655,7 +690,7 @@ unconditionally set to enable replication connections.
 sub init_from_backup
 {
 	my ($self, $root_node, $backup_name, %params) = @_;
-	my $backup_path = $root_node->backup_dir . '/' . $backup_name;
+	my $backup_path = $root_node->backup_dir . '/' . $backup_name . '/' . "pgdata";
 	my $host        = $self->host;
 	my $port        = $self->port;
 	my $node_name   = $self->name;
diff --git a/src/test/perl/RecursiveCopy.pm b/src/test/perl/RecursiveCopy.pm
index baf5d0ac63..f165db7348 100644
--- a/src/test/perl/RecursiveCopy.pm
+++ b/src/test/perl/RecursiveCopy.pm
@@ -22,6 +22,7 @@ use warnings;
 use Carp;
 use File::Basename;
 use File::Copy;
+use TestLib;
 
 =pod
 
@@ -97,14 +98,43 @@ sub _copypath_recurse
 	# invoke the filter and skip all further operation if it returns false
 	return 1 unless &$filterfn($curr_path);
 
-	# Check for symlink -- needed only on source dir
-	# (note: this will fall through quietly if file is already gone)
-	croak "Cannot operate on symlink \"$srcpath\"" if -l $srcpath;
-
 	# Abort if destination path already exists.  Should we allow directories
 	# to exist already?
 	croak "Destination path \"$destpath\" already exists" if -e $destpath;
 
+	# Check for symlink -- needed only on source dir
+	# (note: this will fall through quietly if file is already gone)
+	if (-l $srcpath)
+	{
+		croak "Cannot operate on symlink \"$srcpath\""
+		  if ($srcpath !~ /\/(pg_tblspc\/[0-9]+)$/);
+
+		# We have mapped tablespaces. Copy them individually
+		my $linkname = $1;
+		my $rpath = my $target = readlink($srcpath);
+
+		#convert to pgdata-based if relative
+		$rpath =~ s#^\.\./##; 
+
+		my $srcrealdir = "$base_src_dir/$rpath";
+		my $dstrealdir = "$base_dest_dir/$rpath";
+
+		mkdir $dstrealdir;
+		opendir(my $dh, $srcrealdir);
+		while (readdir $dh)
+		{
+			next if (/^\.\.?$/);
+			my $spath = "$srcrealdir/$_";
+			my $dpath = "$dstrealdir/$_";
+
+			copypath($spath, $dpath);
+		}
+		closedir $dh;
+
+		symlink $target, $destpath;
+		return 1;
+	}
+
 	# If this source path is a file, simply copy it to destination with the
 	# same name and we're done.
 	if (-f $srcpath)
-- 
2.16.3


----Next_Part(Wed_Apr_24_17_02_28_2019_644)--
Content-Type: Text/X-Patch; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
 filename="v2-0003-Add-check-for-recovery-failure-caused-by-tablespace.patch"



^ permalink  raw  reply  [nested|flat] 10+ messages in thread

* [PATCH 2/4] Allow TAP test to excecise tablespace.
@ 2019-04-22 11:58 Kyotaro Horiguchi <[email protected]>
  0 siblings, 0 replies; 10+ messages in thread

From: Kyotaro Horiguchi @ 2019-04-22 11:58 UTC (permalink / raw)

To perform tablespace related checks, this patch lets
PostgresNode::backup have a new parameter "tablespace_mapping", and
make init_from_backup handle capable to handle a backup created using
tablespace_mapping.
---
 src/test/perl/PostgresNode.pm  | 45 +++++++++++++++++++++++++++++++++++++-----
 src/test/perl/RecursiveCopy.pm | 38 +++++++++++++++++++++++++++++++----
 2 files changed, 74 insertions(+), 9 deletions(-)

diff --git a/src/test/perl/PostgresNode.pm b/src/test/perl/PostgresNode.pm
index 76874141c5..e951acc461 100644
--- a/src/test/perl/PostgresNode.pm
+++ b/src/test/perl/PostgresNode.pm
@@ -157,6 +157,7 @@ sub new
 		_host    => $pghost,
 		_basedir => "$TestLib::tmp_check/t_${testname}_${name}_data",
 		_name    => $name,
+		_tablespaces => [],
 		_logfile_generation => 0,
 		_logfile_base       => "$TestLib::log_path/${testname}_${name}",
 		_logfile            => "$TestLib::log_path/${testname}_${name}.log"
@@ -342,6 +343,26 @@ sub backup_dir
 
 =pod
 
+=item $node->make_tablespace_dir()
+
+make a tablespace directory 
+
+=cut
+
+sub make_tablespace_dir
+{
+	my ($self, $name) = @_;
+	my $basedir = $self->basedir;
+
+	die "tablespace name contains '/'" if ($name =~ m#/#);
+	my $reldir = "../$name";
+	mkdir $self->data_dir() . "/". $reldir;
+	push($self->{_tablespaces}, $reldir);
+	return $reldir;
+}
+
+=pod
+
 =item $node->info()
 
 Return a string containing human-readable diagnostic information (paths, etc)
@@ -540,13 +561,27 @@ target server since it isn't done by default.
 
 sub backup
 {
-	my ($self, $backup_name) = @_;
-	my $backup_path = $self->backup_dir . '/' . $backup_name;
+	my ($self, $backup_name, %params) = @_;
+	my $backup_path = $self->backup_dir . '/' . $backup_name . '/' . "pgdata";
 	my $name        = $self->name;
+	my @rest = ();
+
+	if (defined $params{tablespace_mapping})
+	{
+		foreach my $p (split /,/, $params{tablespace_mapping})
+		{
+			push(@rest, "--tablespace-mapping=$p");
+		}
+	}
+
+	foreach my $p ($self->{_tablespaces})
+	{
+		mkdir "$backup_path/$p";
+	}
 
 	print "# Taking pg_basebackup $backup_name from node \"$name\"\n";
 	TestLib::system_or_bail('pg_basebackup', '-D', $backup_path, '-h',
-		$self->host, '-p', $self->port, '--no-sync');
+		$self->host, '-p', $self->port, '--no-sync', @rest);
 	print "# Backup finished\n";
 	return;
 }
@@ -592,7 +627,7 @@ sub backup_fs_cold
 sub _backup_fs
 {
 	my ($self, $backup_name, $hot) = @_;
-	my $backup_path = $self->backup_dir . '/' . $backup_name;
+	my $backup_path = $self->backup_dir . '/' . $backup_name . '/' . "pgdata";
 	my $port        = $self->port;
 	my $name        = $self->name;
 
@@ -655,7 +690,7 @@ unconditionally set to enable replication connections.
 sub init_from_backup
 {
 	my ($self, $root_node, $backup_name, %params) = @_;
-	my $backup_path = $root_node->backup_dir . '/' . $backup_name;
+	my $backup_path = $root_node->backup_dir . '/' . $backup_name . '/' . "pgdata";
 	my $host        = $self->host;
 	my $port        = $self->port;
 	my $node_name   = $self->name;
diff --git a/src/test/perl/RecursiveCopy.pm b/src/test/perl/RecursiveCopy.pm
index baf5d0ac63..f165db7348 100644
--- a/src/test/perl/RecursiveCopy.pm
+++ b/src/test/perl/RecursiveCopy.pm
@@ -22,6 +22,7 @@ use warnings;
 use Carp;
 use File::Basename;
 use File::Copy;
+use TestLib;
 
 =pod
 
@@ -97,14 +98,43 @@ sub _copypath_recurse
 	# invoke the filter and skip all further operation if it returns false
 	return 1 unless &$filterfn($curr_path);
 
-	# Check for symlink -- needed only on source dir
-	# (note: this will fall through quietly if file is already gone)
-	croak "Cannot operate on symlink \"$srcpath\"" if -l $srcpath;
-
 	# Abort if destination path already exists.  Should we allow directories
 	# to exist already?
 	croak "Destination path \"$destpath\" already exists" if -e $destpath;
 
+	# Check for symlink -- needed only on source dir
+	# (note: this will fall through quietly if file is already gone)
+	if (-l $srcpath)
+	{
+		croak "Cannot operate on symlink \"$srcpath\""
+		  if ($srcpath !~ /\/(pg_tblspc\/[0-9]+)$/);
+
+		# We have mapped tablespaces. Copy them individually
+		my $linkname = $1;
+		my $rpath = my $target = readlink($srcpath);
+
+		#convert to pgdata-based if relative
+		$rpath =~ s#^\.\./##; 
+
+		my $srcrealdir = "$base_src_dir/$rpath";
+		my $dstrealdir = "$base_dest_dir/$rpath";
+
+		mkdir $dstrealdir;
+		opendir(my $dh, $srcrealdir);
+		while (readdir $dh)
+		{
+			next if (/^\.\.?$/);
+			my $spath = "$srcrealdir/$_";
+			my $dpath = "$dstrealdir/$_";
+
+			copypath($spath, $dpath);
+		}
+		closedir $dh;
+
+		symlink $target, $destpath;
+		return 1;
+	}
+
 	# If this source path is a file, simply copy it to destination with the
 	# same name and we're done.
 	if (-f $srcpath)
-- 
2.16.3


----Next_Part(Wed_Apr_24_13_18_45_2019_938)--
Content-Type: Text/X-Patch; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
 filename="0003-Add-check-for-recovery-failure-caused-by-tablespace.patch"



^ permalink  raw  reply  [nested|flat] 10+ messages in thread

* Re: Improve tab completion for ALTER FUNCTION/PROCEDURE/ROUTINE
@ 2023-01-07 12:26 vignesh C <[email protected]>
  0 siblings, 0 replies; 10+ messages in thread

From: vignesh C @ 2023-01-07 12:26 UTC (permalink / raw)
  To: Dean Rasheed <[email protected]>; +Cc: Melih Mutlu <[email protected]>; Michael Paquier <[email protected]>; Dong Wook Lee <[email protected]>; pgsql-hackers

On Fri, 6 Jan 2023 at 15:33, Dean Rasheed <[email protected]> wrote:
>
> On Fri, 6 Jan 2023 at 02:38, vignesh C <[email protected]> wrote:
> >
> > On Thu, 5 Jan 2023 at 18:22, Dean Rasheed <[email protected]> wrote:
> > >
> > > That leads to the attached, which barring objections, I'll push shortly.
> >
> > The changes look good to me.
> >
>
> Pushed.

Thanks for pushing this.

Regards,
Vignesh






^ permalink  raw  reply  [nested|flat] 10+ messages in thread

* [PATCH v2 2/3] add avx2 support in simd.h
@ 2024-03-15 17:26 Nathan Bossart <[email protected]>
  0 siblings, 0 replies; 10+ messages in thread

From: Nathan Bossart @ 2024-03-15 17:26 UTC (permalink / raw)

---
 src/include/port/simd.h | 58 ++++++++++++++++++++++++++++++++---------
 1 file changed, 45 insertions(+), 13 deletions(-)

diff --git a/src/include/port/simd.h b/src/include/port/simd.h
index 597496f2fb..767127b85c 100644
--- a/src/include/port/simd.h
+++ b/src/include/port/simd.h
@@ -18,7 +18,15 @@
 #ifndef SIMD_H
 #define SIMD_H
 
-#if (defined(__x86_64__) || defined(_M_AMD64))
+#if defined(__AVX2__)
+
+#include <immintrin.h>
+#define USE_AVX2
+typedef __m256i Vector8;
+typedef __m256i Vector32;
+
+#elif (defined(__x86_64__) || defined(_M_AMD64))
+
 /*
  * SSE2 instructions are part of the spec for the 64-bit x86 ISA. We assume
  * that compilers targeting this architecture understand SSE2 intrinsics.
@@ -107,7 +115,9 @@ static inline Vector32 vector32_eq(const Vector32 v1, const Vector32 v2);
 static inline void
 vector8_load(Vector8 *v, const uint8 *s)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u8(s);
@@ -120,7 +130,9 @@ vector8_load(Vector8 *v, const uint8 *s)
 static inline void
 vector32_load(Vector32 *v, const uint32 *s)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u32(s);
@@ -134,7 +146,9 @@ vector32_load(Vector32 *v, const uint32 *s)
 static inline Vector8
 vector8_broadcast(const uint8 c)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	return _mm256_set1_epi8(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi8(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u8(c);
@@ -147,7 +161,9 @@ vector8_broadcast(const uint8 c)
 static inline Vector32
 vector32_broadcast(const uint32 c)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_set1_epi32(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi32(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u32(c);
@@ -270,7 +286,9 @@ vector8_has_le(const Vector8 v, const uint8 c)
 static inline bool
 vector8_is_highbit_set(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_movemask_epi8(v) != 0;
+#elif defined(USE_SSE2)
 	return _mm_movemask_epi8(v) != 0;
 #elif defined(USE_NEON)
 	return vmaxvq_u8(v) > 0x7F;
@@ -308,7 +326,9 @@ vector32_is_highbit_set(const Vector32 v)
 static inline uint32
 vector8_highbit_mask(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return (uint32) _mm256_movemask_epi8(v);
+#elif defined(USE_SSE2)
 	return (uint32) _mm_movemask_epi8(v);
 #elif defined(USE_NEON)
 	/*
@@ -337,7 +357,9 @@ vector8_highbit_mask(const Vector8 v)
 static inline Vector8
 vector8_or(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u8(v1, v2);
@@ -350,7 +372,9 @@ vector8_or(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_or(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u32(v1, v2);
@@ -368,7 +392,9 @@ vector32_or(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_ssub(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_subs_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_subs_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vqsubq_u8(v1, v2);
@@ -384,7 +410,9 @@ vector8_ssub(const Vector8 v1, const Vector8 v2)
 static inline Vector8
 vector8_eq(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi8(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u8(v1, v2);
@@ -396,7 +424,9 @@ vector8_eq(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_eq(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi32(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi32(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u32(v1, v2);
@@ -411,7 +441,9 @@ vector32_eq(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_min(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_min_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_min_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vminq_u8(v1, v2);
-- 
2.25.1


--6c2NcOVqGQ03X4Wi
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
	filename="v2-0003-test_lfind32-benchmark.patch"



^ permalink  raw  reply  [nested|flat] 10+ messages in thread

* [PATCH v2 2/3] add avx2 support in simd.h
@ 2024-03-15 17:26 Nathan Bossart <[email protected]>
  0 siblings, 0 replies; 10+ messages in thread

From: Nathan Bossart @ 2024-03-15 17:26 UTC (permalink / raw)

---
 src/include/port/simd.h | 58 ++++++++++++++++++++++++++++++++---------
 1 file changed, 45 insertions(+), 13 deletions(-)

diff --git a/src/include/port/simd.h b/src/include/port/simd.h
index 597496f2fb..767127b85c 100644
--- a/src/include/port/simd.h
+++ b/src/include/port/simd.h
@@ -18,7 +18,15 @@
 #ifndef SIMD_H
 #define SIMD_H
 
-#if (defined(__x86_64__) || defined(_M_AMD64))
+#if defined(__AVX2__)
+
+#include <immintrin.h>
+#define USE_AVX2
+typedef __m256i Vector8;
+typedef __m256i Vector32;
+
+#elif (defined(__x86_64__) || defined(_M_AMD64))
+
 /*
  * SSE2 instructions are part of the spec for the 64-bit x86 ISA. We assume
  * that compilers targeting this architecture understand SSE2 intrinsics.
@@ -107,7 +115,9 @@ static inline Vector32 vector32_eq(const Vector32 v1, const Vector32 v2);
 static inline void
 vector8_load(Vector8 *v, const uint8 *s)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u8(s);
@@ -120,7 +130,9 @@ vector8_load(Vector8 *v, const uint8 *s)
 static inline void
 vector32_load(Vector32 *v, const uint32 *s)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u32(s);
@@ -134,7 +146,9 @@ vector32_load(Vector32 *v, const uint32 *s)
 static inline Vector8
 vector8_broadcast(const uint8 c)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	return _mm256_set1_epi8(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi8(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u8(c);
@@ -147,7 +161,9 @@ vector8_broadcast(const uint8 c)
 static inline Vector32
 vector32_broadcast(const uint32 c)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_set1_epi32(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi32(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u32(c);
@@ -270,7 +286,9 @@ vector8_has_le(const Vector8 v, const uint8 c)
 static inline bool
 vector8_is_highbit_set(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_movemask_epi8(v) != 0;
+#elif defined(USE_SSE2)
 	return _mm_movemask_epi8(v) != 0;
 #elif defined(USE_NEON)
 	return vmaxvq_u8(v) > 0x7F;
@@ -308,7 +326,9 @@ vector32_is_highbit_set(const Vector32 v)
 static inline uint32
 vector8_highbit_mask(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return (uint32) _mm256_movemask_epi8(v);
+#elif defined(USE_SSE2)
 	return (uint32) _mm_movemask_epi8(v);
 #elif defined(USE_NEON)
 	/*
@@ -337,7 +357,9 @@ vector8_highbit_mask(const Vector8 v)
 static inline Vector8
 vector8_or(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u8(v1, v2);
@@ -350,7 +372,9 @@ vector8_or(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_or(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u32(v1, v2);
@@ -368,7 +392,9 @@ vector32_or(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_ssub(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_subs_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_subs_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vqsubq_u8(v1, v2);
@@ -384,7 +410,9 @@ vector8_ssub(const Vector8 v1, const Vector8 v2)
 static inline Vector8
 vector8_eq(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi8(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u8(v1, v2);
@@ -396,7 +424,9 @@ vector8_eq(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_eq(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi32(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi32(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u32(v1, v2);
@@ -411,7 +441,9 @@ vector32_eq(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_min(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_min_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_min_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vminq_u8(v1, v2);
-- 
2.25.1


--6c2NcOVqGQ03X4Wi
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
	filename="v2-0003-test_lfind32-benchmark.patch"



^ permalink  raw  reply  [nested|flat] 10+ messages in thread

* [PATCH v3 2/3] add avx2 support in simd.h
@ 2024-03-15 17:26 Nathan Bossart <[email protected]>
  0 siblings, 0 replies; 10+ messages in thread

From: Nathan Bossart @ 2024-03-15 17:26 UTC (permalink / raw)

---
 src/include/port/simd.h | 58 ++++++++++++++++++++++++++++++++---------
 1 file changed, 45 insertions(+), 13 deletions(-)

diff --git a/src/include/port/simd.h b/src/include/port/simd.h
index 597496f2fb..767127b85c 100644
--- a/src/include/port/simd.h
+++ b/src/include/port/simd.h
@@ -18,7 +18,15 @@
 #ifndef SIMD_H
 #define SIMD_H
 
-#if (defined(__x86_64__) || defined(_M_AMD64))
+#if defined(__AVX2__)
+
+#include <immintrin.h>
+#define USE_AVX2
+typedef __m256i Vector8;
+typedef __m256i Vector32;
+
+#elif (defined(__x86_64__) || defined(_M_AMD64))
+
 /*
  * SSE2 instructions are part of the spec for the 64-bit x86 ISA. We assume
  * that compilers targeting this architecture understand SSE2 intrinsics.
@@ -107,7 +115,9 @@ static inline Vector32 vector32_eq(const Vector32 v1, const Vector32 v2);
 static inline void
 vector8_load(Vector8 *v, const uint8 *s)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u8(s);
@@ -120,7 +130,9 @@ vector8_load(Vector8 *v, const uint8 *s)
 static inline void
 vector32_load(Vector32 *v, const uint32 *s)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u32(s);
@@ -134,7 +146,9 @@ vector32_load(Vector32 *v, const uint32 *s)
 static inline Vector8
 vector8_broadcast(const uint8 c)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	return _mm256_set1_epi8(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi8(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u8(c);
@@ -147,7 +161,9 @@ vector8_broadcast(const uint8 c)
 static inline Vector32
 vector32_broadcast(const uint32 c)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_set1_epi32(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi32(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u32(c);
@@ -270,7 +286,9 @@ vector8_has_le(const Vector8 v, const uint8 c)
 static inline bool
 vector8_is_highbit_set(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_movemask_epi8(v) != 0;
+#elif defined(USE_SSE2)
 	return _mm_movemask_epi8(v) != 0;
 #elif defined(USE_NEON)
 	return vmaxvq_u8(v) > 0x7F;
@@ -308,7 +326,9 @@ vector32_is_highbit_set(const Vector32 v)
 static inline uint32
 vector8_highbit_mask(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return (uint32) _mm256_movemask_epi8(v);
+#elif defined(USE_SSE2)
 	return (uint32) _mm_movemask_epi8(v);
 #elif defined(USE_NEON)
 	/*
@@ -337,7 +357,9 @@ vector8_highbit_mask(const Vector8 v)
 static inline Vector8
 vector8_or(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u8(v1, v2);
@@ -350,7 +372,9 @@ vector8_or(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_or(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u32(v1, v2);
@@ -368,7 +392,9 @@ vector32_or(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_ssub(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_subs_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_subs_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vqsubq_u8(v1, v2);
@@ -384,7 +410,9 @@ vector8_ssub(const Vector8 v1, const Vector8 v2)
 static inline Vector8
 vector8_eq(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi8(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u8(v1, v2);
@@ -396,7 +424,9 @@ vector8_eq(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_eq(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi32(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi32(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u32(v1, v2);
@@ -411,7 +441,9 @@ vector32_eq(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_min(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_min_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_min_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vminq_u8(v1, v2);
-- 
2.25.1


--gBBFr7Ir9EOA20Yy
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
	filename="v3-0003-optimize-pg_lfind32-by-processing-tail-with-fewer.patch"



^ permalink  raw  reply  [nested|flat] 10+ messages in thread

* [PATCH v3 2/3] add avx2 support in simd.h
@ 2024-03-15 17:26 Nathan Bossart <[email protected]>
  0 siblings, 0 replies; 10+ messages in thread

From: Nathan Bossart @ 2024-03-15 17:26 UTC (permalink / raw)

---
 src/include/port/simd.h | 58 ++++++++++++++++++++++++++++++++---------
 1 file changed, 45 insertions(+), 13 deletions(-)

diff --git a/src/include/port/simd.h b/src/include/port/simd.h
index 597496f2fb..767127b85c 100644
--- a/src/include/port/simd.h
+++ b/src/include/port/simd.h
@@ -18,7 +18,15 @@
 #ifndef SIMD_H
 #define SIMD_H
 
-#if (defined(__x86_64__) || defined(_M_AMD64))
+#if defined(__AVX2__)
+
+#include <immintrin.h>
+#define USE_AVX2
+typedef __m256i Vector8;
+typedef __m256i Vector32;
+
+#elif (defined(__x86_64__) || defined(_M_AMD64))
+
 /*
  * SSE2 instructions are part of the spec for the 64-bit x86 ISA. We assume
  * that compilers targeting this architecture understand SSE2 intrinsics.
@@ -107,7 +115,9 @@ static inline Vector32 vector32_eq(const Vector32 v1, const Vector32 v2);
 static inline void
 vector8_load(Vector8 *v, const uint8 *s)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u8(s);
@@ -120,7 +130,9 @@ vector8_load(Vector8 *v, const uint8 *s)
 static inline void
 vector32_load(Vector32 *v, const uint32 *s)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u32(s);
@@ -134,7 +146,9 @@ vector32_load(Vector32 *v, const uint32 *s)
 static inline Vector8
 vector8_broadcast(const uint8 c)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	return _mm256_set1_epi8(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi8(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u8(c);
@@ -147,7 +161,9 @@ vector8_broadcast(const uint8 c)
 static inline Vector32
 vector32_broadcast(const uint32 c)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_set1_epi32(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi32(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u32(c);
@@ -270,7 +286,9 @@ vector8_has_le(const Vector8 v, const uint8 c)
 static inline bool
 vector8_is_highbit_set(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_movemask_epi8(v) != 0;
+#elif defined(USE_SSE2)
 	return _mm_movemask_epi8(v) != 0;
 #elif defined(USE_NEON)
 	return vmaxvq_u8(v) > 0x7F;
@@ -308,7 +326,9 @@ vector32_is_highbit_set(const Vector32 v)
 static inline uint32
 vector8_highbit_mask(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return (uint32) _mm256_movemask_epi8(v);
+#elif defined(USE_SSE2)
 	return (uint32) _mm_movemask_epi8(v);
 #elif defined(USE_NEON)
 	/*
@@ -337,7 +357,9 @@ vector8_highbit_mask(const Vector8 v)
 static inline Vector8
 vector8_or(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u8(v1, v2);
@@ -350,7 +372,9 @@ vector8_or(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_or(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u32(v1, v2);
@@ -368,7 +392,9 @@ vector32_or(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_ssub(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_subs_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_subs_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vqsubq_u8(v1, v2);
@@ -384,7 +410,9 @@ vector8_ssub(const Vector8 v1, const Vector8 v2)
 static inline Vector8
 vector8_eq(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi8(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u8(v1, v2);
@@ -396,7 +424,9 @@ vector8_eq(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_eq(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi32(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi32(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u32(v1, v2);
@@ -411,7 +441,9 @@ vector32_eq(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_min(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_min_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_min_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vminq_u8(v1, v2);
-- 
2.25.1


--gBBFr7Ir9EOA20Yy
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
	filename="v3-0003-optimize-pg_lfind32-by-processing-tail-with-fewer.patch"



^ permalink  raw  reply  [nested|flat] 10+ messages in thread

* [PATCH v3 2/3] add avx2 support in simd.h
@ 2024-03-15 17:26 Nathan Bossart <[email protected]>
  0 siblings, 0 replies; 10+ messages in thread

From: Nathan Bossart @ 2024-03-15 17:26 UTC (permalink / raw)

---
 src/include/port/simd.h | 58 ++++++++++++++++++++++++++++++++---------
 1 file changed, 45 insertions(+), 13 deletions(-)

diff --git a/src/include/port/simd.h b/src/include/port/simd.h
index 597496f2fb..767127b85c 100644
--- a/src/include/port/simd.h
+++ b/src/include/port/simd.h
@@ -18,7 +18,15 @@
 #ifndef SIMD_H
 #define SIMD_H
 
-#if (defined(__x86_64__) || defined(_M_AMD64))
+#if defined(__AVX2__)
+
+#include <immintrin.h>
+#define USE_AVX2
+typedef __m256i Vector8;
+typedef __m256i Vector32;
+
+#elif (defined(__x86_64__) || defined(_M_AMD64))
+
 /*
  * SSE2 instructions are part of the spec for the 64-bit x86 ISA. We assume
  * that compilers targeting this architecture understand SSE2 intrinsics.
@@ -107,7 +115,9 @@ static inline Vector32 vector32_eq(const Vector32 v1, const Vector32 v2);
 static inline void
 vector8_load(Vector8 *v, const uint8 *s)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u8(s);
@@ -120,7 +130,9 @@ vector8_load(Vector8 *v, const uint8 *s)
 static inline void
 vector32_load(Vector32 *v, const uint32 *s)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u32(s);
@@ -134,7 +146,9 @@ vector32_load(Vector32 *v, const uint32 *s)
 static inline Vector8
 vector8_broadcast(const uint8 c)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	return _mm256_set1_epi8(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi8(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u8(c);
@@ -147,7 +161,9 @@ vector8_broadcast(const uint8 c)
 static inline Vector32
 vector32_broadcast(const uint32 c)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_set1_epi32(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi32(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u32(c);
@@ -270,7 +286,9 @@ vector8_has_le(const Vector8 v, const uint8 c)
 static inline bool
 vector8_is_highbit_set(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_movemask_epi8(v) != 0;
+#elif defined(USE_SSE2)
 	return _mm_movemask_epi8(v) != 0;
 #elif defined(USE_NEON)
 	return vmaxvq_u8(v) > 0x7F;
@@ -308,7 +326,9 @@ vector32_is_highbit_set(const Vector32 v)
 static inline uint32
 vector8_highbit_mask(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return (uint32) _mm256_movemask_epi8(v);
+#elif defined(USE_SSE2)
 	return (uint32) _mm_movemask_epi8(v);
 #elif defined(USE_NEON)
 	/*
@@ -337,7 +357,9 @@ vector8_highbit_mask(const Vector8 v)
 static inline Vector8
 vector8_or(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u8(v1, v2);
@@ -350,7 +372,9 @@ vector8_or(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_or(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u32(v1, v2);
@@ -368,7 +392,9 @@ vector32_or(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_ssub(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_subs_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_subs_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vqsubq_u8(v1, v2);
@@ -384,7 +410,9 @@ vector8_ssub(const Vector8 v1, const Vector8 v2)
 static inline Vector8
 vector8_eq(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi8(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u8(v1, v2);
@@ -396,7 +424,9 @@ vector8_eq(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_eq(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi32(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi32(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u32(v1, v2);
@@ -411,7 +441,9 @@ vector32_eq(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_min(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_min_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_min_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vminq_u8(v1, v2);
-- 
2.25.1


--gBBFr7Ir9EOA20Yy
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
	filename="v3-0003-optimize-pg_lfind32-by-processing-tail-with-fewer.patch"



^ permalink  raw  reply  [nested|flat] 10+ messages in thread

* [PATCH v2 2/3] add avx2 support in simd.h
@ 2024-03-15 17:26 Nathan Bossart <[email protected]>
  0 siblings, 0 replies; 10+ messages in thread

From: Nathan Bossart @ 2024-03-15 17:26 UTC (permalink / raw)

---
 src/include/port/simd.h | 58 ++++++++++++++++++++++++++++++++---------
 1 file changed, 45 insertions(+), 13 deletions(-)

diff --git a/src/include/port/simd.h b/src/include/port/simd.h
index 597496f2fb..767127b85c 100644
--- a/src/include/port/simd.h
+++ b/src/include/port/simd.h
@@ -18,7 +18,15 @@
 #ifndef SIMD_H
 #define SIMD_H
 
-#if (defined(__x86_64__) || defined(_M_AMD64))
+#if defined(__AVX2__)
+
+#include <immintrin.h>
+#define USE_AVX2
+typedef __m256i Vector8;
+typedef __m256i Vector32;
+
+#elif (defined(__x86_64__) || defined(_M_AMD64))
+
 /*
  * SSE2 instructions are part of the spec for the 64-bit x86 ISA. We assume
  * that compilers targeting this architecture understand SSE2 intrinsics.
@@ -107,7 +115,9 @@ static inline Vector32 vector32_eq(const Vector32 v1, const Vector32 v2);
 static inline void
 vector8_load(Vector8 *v, const uint8 *s)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u8(s);
@@ -120,7 +130,9 @@ vector8_load(Vector8 *v, const uint8 *s)
 static inline void
 vector32_load(Vector32 *v, const uint32 *s)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u32(s);
@@ -134,7 +146,9 @@ vector32_load(Vector32 *v, const uint32 *s)
 static inline Vector8
 vector8_broadcast(const uint8 c)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	return _mm256_set1_epi8(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi8(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u8(c);
@@ -147,7 +161,9 @@ vector8_broadcast(const uint8 c)
 static inline Vector32
 vector32_broadcast(const uint32 c)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_set1_epi32(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi32(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u32(c);
@@ -270,7 +286,9 @@ vector8_has_le(const Vector8 v, const uint8 c)
 static inline bool
 vector8_is_highbit_set(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_movemask_epi8(v) != 0;
+#elif defined(USE_SSE2)
 	return _mm_movemask_epi8(v) != 0;
 #elif defined(USE_NEON)
 	return vmaxvq_u8(v) > 0x7F;
@@ -308,7 +326,9 @@ vector32_is_highbit_set(const Vector32 v)
 static inline uint32
 vector8_highbit_mask(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return (uint32) _mm256_movemask_epi8(v);
+#elif defined(USE_SSE2)
 	return (uint32) _mm_movemask_epi8(v);
 #elif defined(USE_NEON)
 	/*
@@ -337,7 +357,9 @@ vector8_highbit_mask(const Vector8 v)
 static inline Vector8
 vector8_or(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u8(v1, v2);
@@ -350,7 +372,9 @@ vector8_or(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_or(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u32(v1, v2);
@@ -368,7 +392,9 @@ vector32_or(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_ssub(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_subs_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_subs_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vqsubq_u8(v1, v2);
@@ -384,7 +410,9 @@ vector8_ssub(const Vector8 v1, const Vector8 v2)
 static inline Vector8
 vector8_eq(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi8(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u8(v1, v2);
@@ -396,7 +424,9 @@ vector8_eq(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_eq(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi32(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi32(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u32(v1, v2);
@@ -411,7 +441,9 @@ vector32_eq(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_min(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_min_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_min_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vminq_u8(v1, v2);
-- 
2.25.1


--6c2NcOVqGQ03X4Wi
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
	filename="v2-0003-test_lfind32-benchmark.patch"



^ permalink  raw  reply  [nested|flat] 10+ messages in thread

end of thread, other threads:[~2024-03-15 17:26 UTC | newest]

Thread overview: 10+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2019-04-22 11:10 [PATCH 1/2] Allow TAP test to excecise tablespace. Kyotaro Horiguchi <[email protected]>
2019-04-22 11:58 [PATCH 2/5] Allow TAP test to excecise tablespace. Kyotaro Horiguchi <[email protected]>
2019-04-22 11:58 [PATCH 2/4] Allow TAP test to excecise tablespace. Kyotaro Horiguchi <[email protected]>
2023-01-07 12:26 Re: Improve tab completion for ALTER FUNCTION/PROCEDURE/ROUTINE vignesh C <[email protected]>
2024-03-15 17:26 [PATCH v2 2/3] add avx2 support in simd.h Nathan Bossart <[email protected]>
2024-03-15 17:26 [PATCH v2 2/3] add avx2 support in simd.h Nathan Bossart <[email protected]>
2024-03-15 17:26 [PATCH v3 2/3] add avx2 support in simd.h Nathan Bossart <[email protected]>
2024-03-15 17:26 [PATCH v3 2/3] add avx2 support in simd.h Nathan Bossart <[email protected]>
2024-03-15 17:26 [PATCH v3 2/3] add avx2 support in simd.h Nathan Bossart <[email protected]>
2024-03-15 17:26 [PATCH v2 2/3] add avx2 support in simd.h Nathan Bossart <[email protected]>

This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox