rpms/openblas
3
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Compare commits

base: rpms:108e16551f564386b1f7f52963b7c82ed895e8fd
Branches Tags
rpms:master
rpms:f41-riscv64
rpms:f38-riscv64
rpms:f38
rpms:main
rpms:rawhide
rpms:f37-riscv64
rpms:f37
rpms:f36
rpms:f35
rpms:f33
rpms:master-riscv64
rpms:f29-riscv64
rpms:el6
rpms:epel7
rpms:f27
rpms:f28
rpms:f25
rpms:f26
rpms:f23
rpms:f24
rpms:el5
rpms:f22
rpms:f21
rpms:f20
rpms:f19
rpms:f18
rpms:f17
..
compare: rpms:e68b99ed3bf080bb05959f8c02365409e23c4a97
Branches Tags
rpms:f41-riscv64
rpms:f38-riscv64
rpms:f38
rpms:main
rpms:rawhide
rpms:f37-riscv64
rpms:f37
rpms:f36
rpms:f35
rpms:f33
rpms:master-riscv64
rpms:master
rpms:f29-riscv64
rpms:el6
rpms:epel7
rpms:f27
rpms:f28
rpms:f25
rpms:f26
rpms:f23
rpms:f24
rpms:el5
rpms:f22
rpms:f21
rpms:f20
rpms:f19
rpms:f18
rpms:f17

9 Commits
108e16551f ... e68b99ed3b

Author SHA1 Message Date
David Abdurachmanov
e68b99ed3b
Add support for riscv64 
Signed-off-by: David Abdurachmanov <davidlt@rivosinc.com>
 2024-11-21 17:04:56 +02:00
Fedora Release Engineering
0dd29cbca1 Rebuilt for https://fedoraproject.org/wiki/Fedora_41_Mass_Rebuild 2024-07-18 20:46:52 +00:00
Honza Horak
4c0dcdd8a4 Fix FTBFS on ppc64le and s390x 
Upstream issue that points to several commits:
https://github.com/OpenMathLib/OpenBLAS/issues/4475

Downstream trackers:
https://bugzilla.redhat.com/show_bug.cgi?id=2261415
https://issues.redhat.com/browse/RHEL-24745
 2024-02-09 22:43:30 +01:00
Fedora Release Engineering
b316bfef29 Rebuilt for https://fedoraproject.org/wiki/Fedora_40_Mass_Rebuild 2024-01-25 11:07:56 +00:00
Fedora Release Engineering
fc952d7fda Rebuilt for https://fedoraproject.org/wiki/Fedora_40_Mass_Rebuild 2024-01-21 11:04:55 +00:00
İsmail Dönmez
8f143edfab Update to version 0.3.26 
Signed-off-by: İsmail Dönmez <ismail@i10z.com>
 2024-01-08 10:08:57 +00:00
Ismail Doenmez
54c877216e Update to 0.3.25 
Fix homepage URL and tarball name

Signed-off-by: Ismail Doenmez <ismail@i10z.com>
 2023-11-30 09:17:24 +00:00
Ali Erdinc Koroglu
5123a10ecd Update to 0.3.24 2023-10-27 19:59:34 +03:00
Fedora Release Engineering
529f54963b Rebuilt for https://fedoraproject.org/wiki/Fedora_39_Mass_Rebuild 
Signed-off-by: Fedora Release Engineering <releng@fedoraproject.org>
 2023-07-20 17:58:09 +00:00
10 changed files with 808 additions and 771 deletions
Show Stats Expand all files Collapse all files

3
.gitignore vendored
View File

@ -1,2 +1 @@
/v*.tar.gz
/openblas-*.tar.gz
/OpenBLAS-*.tar.gz

64
openblas-0.3.21-c99-1.patch
View File

@ -1,64 +0,0 @@
From f703846ad9400a8ea175cb8dd43e18c152aeab93 Mon Sep 17 00:00:00 2001
From: Martin Kroeker <martin@ruby.chemie.uni-freiburg.de>
Date: Sat, 13 Aug 2022 11:38:27 +0200
Subject: [PATCH] Add function prototypes
Content-type: text/plain
---
exports/gensymbol | 16 ++++++++++++++++
exports/gensymbol.pl | 12 ++++++++++++
2 files changed, 28 insertions(+)
diff --git a/exports/gensymbol b/exports/gensymbol
index 83222a21..f05de626 100755
--- a/exports/gensymbol
+++ b/exports/gensymbol
@@ -4000,6 +4000,22 @@ case "$p1" in
no_underscore_objs="$no_underscore_objs $misc_common_objs"
printf 'int main(void){\n'
+ for obj in $underscore_objs; do
+ [ "$obj" != "xerbla" ] && printf 'extern void %s%s%s%s();\n' \
+ "$symbolprefix" "$obj" "$bu" "$symbolsuffix"
+ done
+
+ for obj in $need_2underscore_objs; do
+ printf 'extern void %s%s%s%s%s();\n' \
+ "$symbolprefix" "$obj" "$bu" "$bu" "$symbolsuffix"
+ done
+
+ for obj in $no_underscore_objs; do
+ printf 'extern void %s%s%s();\n' \
+ "$symbolprefix" "$obj" "$symbolsuffix"
+ done
+
+ printf '\n'
for obj in $underscore_objs; do
[ "$obj" != "xerbla" ] && printf '%s%s%s%s();\n' \
"$symbolprefix" "$obj" "$bu" "$symbolsuffix"
diff --git a/exports/gensymbol.pl b/exports/gensymbol.pl
index ac62bc05..e38a3cc8 100644
--- a/exports/gensymbol.pl
+++ b/exports/gensymbol.pl
@@ -3955,6 +3955,18 @@ if ($ARGV[0] eq "linktest") {
@no_underscore_objs = (@no_underscore_objs, @misc_common_objs);
print "int main(void){\n";
+ foreach $objs (@underscore_objs) {
+ print "extern void ", $symbolprefix, $objs, $bu, $symbolsuffix, "();\n" if $objs ne "xerbla";
+ }
+
+ foreach $objs (@need_2underscore_objs) {
+ print "extern void ", $symbolprefix, $objs, $bu, $bu, $symbolsuffix, "();\n";
+ }
+
+ foreach $objs (@no_underscore_objs) {
+ print "extern void ", $symbolprefix, $objs, $symbolsuffix, "();\n";
+ }
+
foreach $objs (@underscore_objs) {
print $symbolprefix, $objs, $bu, $symbolsuffix, "();\n" if $objs ne "xerbla";
}
--
2.37.3

510
openblas-0.3.21-c99-2.patch
View File

@ -1,510 +0,0 @@
From 515cf269291bec0d43651fe7bf99a71fb074a0ad Mon Sep 17 00:00:00 2001
From: Martin Kroeker <martin@ruby.chemie.uni-freiburg.de>
Date: Wed, 14 Sep 2022 11:48:36 +0200
Subject: [PATCH] Fix pointer/integer argument mismatch in calls to pow()
Content-type: text/plain
---
lapack-netlib/SRC/claed0.c | 4 ++--
lapack-netlib/SRC/claed7.c | 4 ++--
lapack-netlib/SRC/clalsa.c | 6 +++---
lapack-netlib/SRC/cstedc.c | 4 ++--
lapack-netlib/SRC/dlaed0.c | 4 ++--
lapack-netlib/SRC/dlaed7.c | 4 ++--
lapack-netlib/SRC/dlaeda.c | 8 ++++----
lapack-netlib/SRC/dlalsa.c | 6 +++---
lapack-netlib/SRC/dlasd0.c | 2 +-
lapack-netlib/SRC/dlasda.c | 4 ++--
lapack-netlib/SRC/dstedc.c | 4 ++--
lapack-netlib/SRC/slaed0.c | 4 ++--
lapack-netlib/SRC/slaed7.c | 4 ++--
lapack-netlib/SRC/slaeda.c | 8 ++++----
lapack-netlib/SRC/slalsa.c | 6 +++---
lapack-netlib/SRC/slasd0.c | 2 +-
lapack-netlib/SRC/slasda.c | 4 ++--
lapack-netlib/SRC/sstedc.c | 4 ++--
lapack-netlib/SRC/zlaed0.c | 4 ++--
lapack-netlib/SRC/zlaed7.c | 4 ++--
lapack-netlib/SRC/zlalsa.c | 6 +++---
lapack-netlib/SRC/zstedc.c | 4 ++--
22 files changed, 50 insertions(+), 50 deletions(-)
diff --git a/lapack-netlib/SRC/claed0.c b/lapack-netlib/SRC/claed0.c
index 21e40839..2b696508 100644
--- a/lapack-netlib/SRC/claed0.c
+++ b/lapack-netlib/SRC/claed0.c
@@ -796,10 +796,10 @@ L10:
temp = log((real) (*n)) / log(2.f);
lgn = (integer) temp;
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
iprmpt = indxq + *n + 1;
diff --git a/lapack-netlib/SRC/claed7.c b/lapack-netlib/SRC/claed7.c
index 49fc9ed4..1eaa7e9c 100644
--- a/lapack-netlib/SRC/claed7.c
+++ b/lapack-netlib/SRC/claed7.c
@@ -864,11 +864,11 @@ f"> */
/* Form the z-vector which consists of the last row of Q_1 and the */
/* first row of Q_2. */
- ptr = pow_ii(&c__2, tlvls) + 1;
+ ptr = pow_ii(c__2, *tlvls) + 1;
i__1 = *curlvl - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = *tlvls - i__;
- ptr += pow_ii(&c__2, &i__2);
+ ptr += pow_ii(c__2, i__2);
/* L10: */
}
curr = ptr + *curpbm;
diff --git a/lapack-netlib/SRC/clalsa.c b/lapack-netlib/SRC/clalsa.c
index 4bc3830a..2ef3e123 100644
--- a/lapack-netlib/SRC/clalsa.c
+++ b/lapack-netlib/SRC/clalsa.c
@@ -1051,7 +1051,7 @@ f"> */
/* Finally go through the left singular vector matrices of all */
/* the other subproblems bottom-up on the tree. */
- j = pow_ii(&c__2, &nlvl);
+ j = pow_ii(c__2, nlvl);
sqre = 0;
for (lvl = nlvl; lvl >= 1; --lvl) {
@@ -1065,7 +1065,7 @@ f"> */
ll = 1;
} else {
i__1 = lvl - 1;
- lf = pow_ii(&c__2, &i__1);
+ lf = pow_ii(c__2, i__1);
ll = (lf << 1) - 1;
}
i__1 = ll;
@@ -1110,7 +1110,7 @@ L170:
ll = 1;
} else {
i__2 = lvl - 1;
- lf = pow_ii(&c__2, &i__2);
+ lf = pow_ii(c__2, i__2);
ll = (lf << 1) - 1;
}
i__2 = lf;
diff --git a/lapack-netlib/SRC/cstedc.c b/lapack-netlib/SRC/cstedc.c
index 437c39e9..8f047d1c 100644
--- a/lapack-netlib/SRC/cstedc.c
+++ b/lapack-netlib/SRC/cstedc.c
@@ -836,10 +836,10 @@ f"> */
lrwmin = *n - 1 << 1;
} else if (icompz == 1) {
lgn = (integer) (log((real) (*n)) / log(2.f));
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
lwmin = *n * *n;
diff --git a/lapack-netlib/SRC/dlaed0.c b/lapack-netlib/SRC/dlaed0.c
index 95e39b0d..74e58dd2 100644
--- a/lapack-netlib/SRC/dlaed0.c
+++ b/lapack-netlib/SRC/dlaed0.c
@@ -827,10 +827,10 @@ L10:
temp = log((doublereal) (*n)) / log(2.);
lgn = (integer) temp;
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
iprmpt = indxq + *n + 1;
diff --git a/lapack-netlib/SRC/dlaed7.c b/lapack-netlib/SRC/dlaed7.c
index fd851526..d23a72be 100644
--- a/lapack-netlib/SRC/dlaed7.c
+++ b/lapack-netlib/SRC/dlaed7.c
@@ -885,11 +885,11 @@ f"> */
/* Form the z-vector which consists of the last row of Q_1 and the */
/* first row of Q_2. */
- ptr = pow_ii(&c__2, tlvls) + 1;
+ ptr = pow_ii(c__2, *tlvls) + 1;
i__1 = *curlvl - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = *tlvls - i__;
- ptr += pow_ii(&c__2, &i__2);
+ ptr += pow_ii(c__2, i__2);
/* L10: */
}
curr = ptr + *curpbm;
diff --git a/lapack-netlib/SRC/dlaeda.c b/lapack-netlib/SRC/dlaeda.c
index f4bb214d..202e1b63 100644
--- a/lapack-netlib/SRC/dlaeda.c
+++ b/lapack-netlib/SRC/dlaeda.c
@@ -754,7 +754,7 @@ f"> */
/* scheme */
i__1 = *curlvl - 1;
- curr = ptr + *curpbm * pow_ii(&c__2, curlvl) + pow_ii(&c__2, &i__1) - 1;
+ curr = ptr + *curpbm * pow_ii(c__2, *curlvl) + pow_ii(c__2, i__1) - 1;
/* Determine size of these matrices. We add HALF to the value of */
/* the SQRT in case the machine underestimates one of these square */
@@ -781,12 +781,12 @@ f"> */
/* rotations and permutation and then multiplying the center matrices */
/* against the current Z. */
- ptr = pow_ii(&c__2, tlvls) + 1;
+ ptr = pow_ii(c__2, *tlvls) + 1;
i__1 = *curlvl - 1;
for (k = 1; k <= i__1; ++k) {
i__2 = *curlvl - k;
i__3 = *curlvl - k - 1;
- curr = ptr + *curpbm * pow_ii(&c__2, &i__2) + pow_ii(&c__2, &i__3) -
+ curr = ptr + *curpbm * pow_ii(c__2, i__2) + pow_ii(c__2, i__3) -
1;
psiz1 = prmptr[curr + 1] - prmptr[curr];
psiz2 = prmptr[curr + 2] - prmptr[curr + 1];
@@ -847,7 +847,7 @@ f"> */
c__1);
i__2 = *tlvls - k;
- ptr += pow_ii(&c__2, &i__2);
+ ptr += pow_ii(c__2, i__2);
/* L70: */
}
diff --git a/lapack-netlib/SRC/dlalsa.c b/lapack-netlib/SRC/dlalsa.c
index 891ed66a..4d5c347c 100644
--- a/lapack-netlib/SRC/dlalsa.c
+++ b/lapack-netlib/SRC/dlalsa.c
@@ -951,7 +951,7 @@ f"> */
/* Finally go through the left singular vector matrices of all */
/* the other subproblems bottom-up on the tree. */
- j = pow_ii(&c__2, &nlvl);
+ j = pow_ii(c__2, nlvl);
sqre = 0;
for (lvl = nlvl; lvl >= 1; --lvl) {
@@ -965,7 +965,7 @@ f"> */
ll = 1;
} else {
i__1 = lvl - 1;
- lf = pow_ii(&c__2, &i__1);
+ lf = pow_ii(c__2, i__1);
ll = (lf << 1) - 1;
}
i__1 = ll;
@@ -1010,7 +1010,7 @@ L50:
ll = 1;
} else {
i__2 = lvl - 1;
- lf = pow_ii(&c__2, &i__2);
+ lf = pow_ii(c__2, i__2);
ll = (lf << 1) - 1;
}
i__2 = lf;
diff --git a/lapack-netlib/SRC/dlasd0.c b/lapack-netlib/SRC/dlasd0.c
index c702665b..0f88527e 100644
--- a/lapack-netlib/SRC/dlasd0.c
+++ b/lapack-netlib/SRC/dlasd0.c
@@ -824,7 +824,7 @@ f"> */
ll = 1;
} else {
i__1 = lvl - 1;
- lf = pow_ii(&c__2, &i__1);
+ lf = pow_ii(c__2, i__1);
ll = (lf << 1) - 1;
}
i__1 = ll;
diff --git a/lapack-netlib/SRC/dlasda.c b/lapack-netlib/SRC/dlasda.c
index 72f9d55f..a9190f80 100644
--- a/lapack-netlib/SRC/dlasda.c
+++ b/lapack-netlib/SRC/dlasda.c
@@ -1027,7 +1027,7 @@ f"> */
/* Now conquer each subproblem bottom-up. */
- j = pow_ii(&c__2, &nlvl);
+ j = pow_ii(c__2, nlvl);
for (lvl = nlvl; lvl >= 1; --lvl) {
lvl2 = (lvl << 1) - 1;
@@ -1039,7 +1039,7 @@ f"> */
ll = 1;
} else {
i__1 = lvl - 1;
- lf = pow_ii(&c__2, &i__1);
+ lf = pow_ii(c__2, i__1);
ll = (lf << 1) - 1;
}
i__1 = ll;
diff --git a/lapack-netlib/SRC/dstedc.c b/lapack-netlib/SRC/dstedc.c
index ef2eeabe..56511d6c 100644
--- a/lapack-netlib/SRC/dstedc.c
+++ b/lapack-netlib/SRC/dstedc.c
@@ -806,10 +806,10 @@ f"> */
lwmin = *n - 1 << 1;
} else {
lgn = (integer) (log((doublereal) (*n)) / log(2.));
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
if (icompz == 1) {
diff --git a/lapack-netlib/SRC/slaed0.c b/lapack-netlib/SRC/slaed0.c
index 33f7134c..4c523090 100644
--- a/lapack-netlib/SRC/slaed0.c
+++ b/lapack-netlib/SRC/slaed0.c
@@ -823,10 +823,10 @@ L10:
temp = log((real) (*n)) / log(2.f);
lgn = (integer) temp;
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
iprmpt = indxq + *n + 1;
diff --git a/lapack-netlib/SRC/slaed7.c b/lapack-netlib/SRC/slaed7.c
index 210d796d..22fcaf76 100644
--- a/lapack-netlib/SRC/slaed7.c
+++ b/lapack-netlib/SRC/slaed7.c
@@ -883,11 +883,11 @@ f"> */
/* Form the z-vector which consists of the last row of Q_1 and the */
/* first row of Q_2. */
- ptr = pow_ii(&c__2, tlvls) + 1;
+ ptr = pow_ii(c__2, *tlvls) + 1;
i__1 = *curlvl - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = *tlvls - i__;
- ptr += pow_ii(&c__2, &i__2);
+ ptr += pow_ii(c__2, i__2);
/* L10: */
}
curr = ptr + *curpbm;
diff --git a/lapack-netlib/SRC/slaeda.c b/lapack-netlib/SRC/slaeda.c
index 7edaf8a7..3806427c 100644
--- a/lapack-netlib/SRC/slaeda.c
+++ b/lapack-netlib/SRC/slaeda.c
@@ -753,7 +753,7 @@ f"> */
/* scheme */
i__1 = *curlvl - 1;
- curr = ptr + *curpbm * pow_ii(&c__2, curlvl) + pow_ii(&c__2, &i__1) - 1;
+ curr = ptr + *curpbm * pow_ii(c__2, *curlvl) + pow_ii(c__2, i__1) - 1;
/* Determine size of these matrices. We add HALF to the value of */
/* the SQRT in case the machine underestimates one of these square */
@@ -779,12 +779,12 @@ f"> */
/* rotations and permutation and then multiplying the center matrices */
/* against the current Z. */
- ptr = pow_ii(&c__2, tlvls) + 1;
+ ptr = pow_ii(c__2, *tlvls) + 1;
i__1 = *curlvl - 1;
for (k = 1; k <= i__1; ++k) {
i__2 = *curlvl - k;
i__3 = *curlvl - k - 1;
- curr = ptr + *curpbm * pow_ii(&c__2, &i__2) + pow_ii(&c__2, &i__3) -
+ curr = ptr + *curpbm * pow_ii(c__2, i__2) + pow_ii(c__2, i__3) -
1;
psiz1 = prmptr[curr + 1] - prmptr[curr];
psiz2 = prmptr[curr + 2] - prmptr[curr + 1];
@@ -844,7 +844,7 @@ f"> */
c__1);
i__2 = *tlvls - k;
- ptr += pow_ii(&c__2, &i__2);
+ ptr += pow_ii(c__2, i__2);
/* L70: */
}
diff --git a/lapack-netlib/SRC/slalsa.c b/lapack-netlib/SRC/slalsa.c
index 53da2c7b..77a79b80 100644
--- a/lapack-netlib/SRC/slalsa.c
+++ b/lapack-netlib/SRC/slalsa.c
@@ -946,7 +946,7 @@ f"> */
/* Finally go through the left singular vector matrices of all */
/* the other subproblems bottom-up on the tree. */
- j = pow_ii(&c__2, &nlvl);
+ j = pow_ii(c__2, nlvl);
sqre = 0;
for (lvl = nlvl; lvl >= 1; --lvl) {
@@ -960,7 +960,7 @@ f"> */
ll = 1;
} else {
i__1 = lvl - 1;
- lf = pow_ii(&c__2, &i__1);
+ lf = pow_ii(c__2, i__1);
ll = (lf << 1) - 1;
}
i__1 = ll;
@@ -1005,7 +1005,7 @@ L50:
ll = 1;
} else {
i__2 = lvl - 1;
- lf = pow_ii(&c__2, &i__2);
+ lf = pow_ii(c__2, i__2);
ll = (lf << 1) - 1;
}
i__2 = lf;
diff --git a/lapack-netlib/SRC/slasd0.c b/lapack-netlib/SRC/slasd0.c
index aa553579..be1a7419 100644
--- a/lapack-netlib/SRC/slasd0.c
+++ b/lapack-netlib/SRC/slasd0.c
@@ -821,7 +821,7 @@ f"> */
ll = 1;
} else {
i__1 = lvl - 1;
- lf = pow_ii(&c__2, &i__1);
+ lf = pow_ii(c__2, i__1);
ll = (lf << 1) - 1;
}
i__1 = ll;
diff --git a/lapack-netlib/SRC/slasda.c b/lapack-netlib/SRC/slasda.c
index 71424c3f..1d336d1c 100644
--- a/lapack-netlib/SRC/slasda.c
+++ b/lapack-netlib/SRC/slasda.c
@@ -1023,7 +1023,7 @@ f"> */
/* Now conquer each subproblem bottom-up. */
- j = pow_ii(&c__2, &nlvl);
+ j = pow_ii(c__2, nlvl);
for (lvl = nlvl; lvl >= 1; --lvl) {
lvl2 = (lvl << 1) - 1;
@@ -1035,7 +1035,7 @@ f"> */
ll = 1;
} else {
i__1 = lvl - 1;
- lf = pow_ii(&c__2, &i__1);
+ lf = pow_ii(c__2, i__1);
ll = (lf << 1) - 1;
}
i__1 = ll;
diff --git a/lapack-netlib/SRC/sstedc.c b/lapack-netlib/SRC/sstedc.c
index 46ed15a1..61ad3dd3 100644
--- a/lapack-netlib/SRC/sstedc.c
+++ b/lapack-netlib/SRC/sstedc.c
@@ -804,10 +804,10 @@ f"> */
lwmin = *n - 1 << 1;
} else {
lgn = (integer) (log((real) (*n)) / log(2.f));
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
if (icompz == 1) {
diff --git a/lapack-netlib/SRC/zlaed0.c b/lapack-netlib/SRC/zlaed0.c
index 37bd12b0..2b25f6e4 100644
--- a/lapack-netlib/SRC/zlaed0.c
+++ b/lapack-netlib/SRC/zlaed0.c
@@ -793,10 +793,10 @@ L10:
temp = log((doublereal) (*n)) / log(2.);
lgn = (integer) temp;
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
iprmpt = indxq + *n + 1;
diff --git a/lapack-netlib/SRC/zlaed7.c b/lapack-netlib/SRC/zlaed7.c
index 09305191..8665ee12 100644
--- a/lapack-netlib/SRC/zlaed7.c
+++ b/lapack-netlib/SRC/zlaed7.c
@@ -864,11 +864,11 @@ f"> */
/* Form the z-vector which consists of the last row of Q_1 and the */
/* first row of Q_2. */
- ptr = pow_ii(&c__2, tlvls) + 1;
+ ptr = pow_ii(c__2, *tlvls) + 1;
i__1 = *curlvl - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = *tlvls - i__;
- ptr += pow_ii(&c__2, &i__2);
+ ptr += pow_ii(c__2, i__2);
/* L10: */
}
curr = ptr + *curpbm;
diff --git a/lapack-netlib/SRC/zlalsa.c b/lapack-netlib/SRC/zlalsa.c
index d17016e7..cd0819c3 100644
--- a/lapack-netlib/SRC/zlalsa.c
+++ b/lapack-netlib/SRC/zlalsa.c
@@ -1051,7 +1051,7 @@ f"> */
/* Finally go through the left singular vector matrices of all */
/* the other subproblems bottom-up on the tree. */
- j = pow_ii(&c__2, &nlvl);
+ j = pow_ii(c__2, nlvl);
sqre = 0;
for (lvl = nlvl; lvl >= 1; --lvl) {
@@ -1065,7 +1065,7 @@ f"> */
ll = 1;
} else {
i__1 = lvl - 1;
- lf = pow_ii(&c__2, &i__1);
+ lf = pow_ii(c__2, i__1);
ll = (lf << 1) - 1;
}
i__1 = ll;
@@ -1110,7 +1110,7 @@ L170:
ll = 1;
} else {
i__2 = lvl - 1;
- lf = pow_ii(&c__2, &i__2);
+ lf = pow_ii(c__2, i__2);
ll = (lf << 1) - 1;
}
i__2 = lf;
diff --git a/lapack-netlib/SRC/zstedc.c b/lapack-netlib/SRC/zstedc.c
index 4cfc4184..55baba2d 100644
--- a/lapack-netlib/SRC/zstedc.c
+++ b/lapack-netlib/SRC/zstedc.c
@@ -836,10 +836,10 @@ f"> */
lrwmin = *n - 1 << 1;
} else if (icompz == 1) {
lgn = (integer) (log((doublereal) (*n)) / log(2.));
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
- if (pow_ii(&c__2, &lgn) < *n) {
+ if (pow_ii(c__2, lgn) < *n) {
++lgn;
}
lwmin = *n * *n;
--
2.37.3

26
openblas-0.3.21-c99-3.patch
View File

@ -1,26 +0,0 @@
From 91110f92d218492d0efbdc1fdf34277ca45f4b36 Mon Sep 17 00:00:00 2001
From: Martin Kroeker <martin@ruby.chemie.uni-freiburg.de>
Date: Wed, 14 Sep 2022 14:03:31 +0200
Subject: [PATCH] fix missing return type in function declaration
Content-type: text/plain
---
ctest/c_sblat1c.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/ctest/c_sblat1c.c b/ctest/c_sblat1c.c
index 4993d31b..57e4707a 100644
--- a/ctest/c_sblat1c.c
+++ b/ctest/c_sblat1c.c
@@ -969,7 +969,7 @@ real *sfac;
1.17 };
/* Local variables */
- extern /* Subroutine */ srottest_();
+ extern /* Subroutine */ void srottest_();
static integer i__, k, ksize;
extern /* Subroutine */ int stest_(), srotmtest_();
static integer ki, kn;
--
2.37.3

26
openblas-0.3.21-c99-4.patch
View File

@ -1,26 +0,0 @@
From 9402df5604e69f86f58953e3883f33f98c930baf Mon Sep 17 00:00:00 2001
From: Martin Kroeker <martin@ruby.chemie.uni-freiburg.de>
Date: Wed, 14 Sep 2022 21:44:34 +0200
Subject: [PATCH] Fix missing external declaration
Content-type: text/plain
---
driver/others/blas_server_omp.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/driver/others/blas_server_omp.c b/driver/others/blas_server_omp.c
index 1a5fd06a..c158f92e 100644
--- a/driver/others/blas_server_omp.c
+++ b/driver/others/blas_server_omp.c
@@ -69,6 +69,8 @@
int blas_server_avail = 0;
+extern int openblas_omp_adaptive_env();
+
static void * blas_thread_buffer[MAX_PARALLEL_NUMBER][MAX_CPU_NUMBER];
#ifdef HAVE_C11
static atomic_bool blas_buffer_inuse[MAX_PARALLEL_NUMBER];
--
2.37.3

62
openblas-0.3.21-c99-5.patch
View File

@ -1,62 +0,0 @@
From 101a2c77c3f3610933f450cefca3e312edab2186 Mon Sep 17 00:00:00 2001
From: Martin Kroeker <martin@ruby.chemie.uni-freiburg.de>
Date: Thu, 15 Sep 2022 09:19:19 +0200
Subject: [PATCH] Fix warnings
Content-type: text/plain
---
kernel/x86_64/dgemm_ncopy_8_skylakex.c | 24 ++++++++++++------------
kernel/x86_64/omatcopy_rt.c | 2 +-
2 files changed, 13 insertions(+), 13 deletions(-)
diff --git a/kernel/x86_64/dgemm_ncopy_8_skylakex.c b/kernel/x86_64/dgemm_ncopy_8_skylakex.c
index 74b336f3..874ef68d 100644
--- a/kernel/x86_64/dgemm_ncopy_8_skylakex.c
+++ b/kernel/x86_64/dgemm_ncopy_8_skylakex.c
@@ -52,18 +52,18 @@ int CNAME(BLASLONG m, BLASLONG n, FLOAT * __restrict a, BLASLONG lda, FLOAT * __
FLOAT ctemp05, ctemp06, ctemp07, ctemp08;
FLOAT ctemp09, ctemp10, ctemp11, ctemp12;
FLOAT ctemp13, ctemp14, ctemp15, ctemp16;
- FLOAT ctemp17, ctemp18, ctemp19, ctemp20;
- FLOAT ctemp21, ctemp22, ctemp23, ctemp24;
- FLOAT ctemp25, ctemp26, ctemp27, ctemp28;
- FLOAT ctemp29, ctemp30, ctemp31, ctemp32;
- FLOAT ctemp33, ctemp34, ctemp35, ctemp36;
- FLOAT ctemp37, ctemp38, ctemp39, ctemp40;
- FLOAT ctemp41, ctemp42, ctemp43, ctemp44;
- FLOAT ctemp45, ctemp46, ctemp47, ctemp48;
- FLOAT ctemp49, ctemp50, ctemp51, ctemp52;
- FLOAT ctemp53, ctemp54, ctemp55, ctemp56;
- FLOAT ctemp57, ctemp58, ctemp59, ctemp60;
- FLOAT ctemp61, ctemp62, ctemp63, ctemp64;
+ FLOAT ctemp17 /*, ctemp18, ctemp19, ctemp20*/ ;
+ FLOAT /*ctemp21, ctemp22,*/ ctemp23, ctemp24;
+ FLOAT ctemp25 /*, ctemp26, ctemp27, ctemp28*/ ;
+ FLOAT /*ctemp29, ctemp30,*/ ctemp31, ctemp32;
+ FLOAT ctemp33 /*, ctemp34, ctemp35, ctemp36*/ ;
+ FLOAT /*ctemp37, ctemp38,*/ ctemp39, ctemp40;
+ FLOAT ctemp41 /*, ctemp42, ctemp43, ctemp44*/ ;
+ FLOAT /*ctemp45, ctemp46,*/ ctemp47, ctemp48;
+ FLOAT ctemp49 /*, ctemp50, ctemp51, ctemp52*/ ;
+ FLOAT /*ctemp53, ctemp54,*/ ctemp55, ctemp56;
+ FLOAT ctemp57 /*, ctemp58, ctemp59, ctemp60*/ ;
+ FLOAT /*ctemp61, ctemp62,*/ ctemp63, ctemp64;
aoffset = a;
diff --git a/kernel/x86_64/omatcopy_rt.c b/kernel/x86_64/omatcopy_rt.c
index e695f00c..b11893f5 100644
--- a/kernel/x86_64/omatcopy_rt.c
+++ b/kernel/x86_64/omatcopy_rt.c
@@ -142,7 +142,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
,"xmm0","xmm1","xmm2","xmm3","xmm4","xmm5","xmm6","xmm7","xmm8","xmm9","xmm10","xmm11","xmm12","xmm13","xmm14","xmm15");\
}
int CNAME(BLASLONG rows, BLASLONG cols, FLOAT alpha, FLOAT *a, BLASLONG lda, FLOAT *b, BLASLONG ldb){
- float *src, *dst, *dst_tmp, *src_base, *dst_base;
+ float *src, *dst, *dst_tmp=0, *src_base, *dst_base;
uint64_t src_ld_bytes = (uint64_t)lda * sizeof(float), dst_ld_bytes = (uint64_t)ldb * sizeof(float), num_rows = 0;
BLASLONG cols_left, rows_done; float ALPHA = alpha;
if(ALPHA==0.0){
--
2.37.3

55
openblas-0.3.21-sbgemm-test.patch
View File

@ -1,55 +0,0 @@
Fixing FTBFS on power:
https://bugzilla.redhat.com/show_bug.cgi?id=2120974
Upstream issue: https://github.com/xianyi/OpenBLAS/issues/3738
Upstream fix: https://github.com/xianyi/OpenBLAS/pull/3718
commit d9dc015cfc78fc32f555995a89d6957ef0184ea2
Author: Martin Kroeker <martin@ruby.chemie.uni-freiburg.de>
Date: Mon Aug 8 14:52:10 2022 +0200
Use blasint for INTERFACE64 compatibility
diff --git a/test/compare_sgemm_sbgemm.c b/test/compare_sgemm_sbgemm.c
index a2c358cf..d4b59145 100644
--- a/test/compare_sgemm_sbgemm.c
+++ b/test/compare_sgemm_sbgemm.c
@@ -76,9 +76,9 @@ float16to32 (bfloat16_bits f16)
int
main (int argc, char *argv[])
{
- int m, n, k;
+ blasint m, n, k;
int i, j, l;
- int x;
+ blasint x;
int ret = 0;
int loop = 100;
char transA = 'N', transB = 'N';
commit 3d338b57de1837f1e2264a1262a9ee9203f31c6c
Author: Martin Kroeker <martin@ruby.chemie.uni-freiburg.de>
Date: Mon Aug 8 17:09:45 2022 +0200
remove spurious loops
diff --git a/test/compare_sgemm_sbgemm.c b/test/compare_sgemm_sbgemm.c
index d4b59145..276fecae 100644
--- a/test/compare_sgemm_sbgemm.c
+++ b/test/compare_sgemm_sbgemm.c
@@ -112,7 +112,6 @@ main (int argc, char *argv[])
&m, BB, &k, &beta, CC, &m);
for (i = 0; i < n; i++)
for (j = 0; j < m; j++)
- for (l = 0; l < k; l++)
if (fabs (CC[i * m + j] - C[i * m + j]) > 1.0)
ret++;
if (transA == 'N' && transB == 'N')
@@ -126,7 +125,6 @@ main (int argc, char *argv[])
}
for (i = 0; i < n; i++)
for (j = 0; j < m; j++)
- for (l = 0; l < k; l++)
if (CC[i * m + j] != DD[i * m + j])
ret++;
}

752
openblas-0.3.26-incompatibletypes.patch Normal file
View File

@ -0,0 +1,752 @@
This is a compilation of more upstream commits related to:
https://github.com/OpenMathLib/OpenBLAS/issues/4475
From 63004fa5f76ef1058975271314bc4591e7878726 Mon Sep 17 00:00:00 2001
From: Honza Horak <hhorak@redhat.com>
Date: Fri, 9 Feb 2024 09:49:41 +0100
Subject: [PATCH 1/6] Fix incompatible pointer type in BFLOAT16 mode
Upstream commit:
commit 68d354814f9f846338e1988c4f609c8add419012
Author: Martin Kroeker <martin@ruby.chemie.uni-freiburg.de>
Date: Sun Feb 4 01:14:22 2024 +0100
Fix incompatible pointer type in BFLOAT16 mode
---
interface/gemmt.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/interface/gemmt.c b/interface/gemmt.c
index 046432670..2fb9954ad 100644
--- a/interface/gemmt.c
+++ b/interface/gemmt.c
@@ -478,7 +478,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
#endif
// for alignment
buffer_size = (buffer_size + 3) & ~3;
- STACK_ALLOC(buffer_size, FLOAT, buffer);
+ STACK_ALLOC(buffer_size, IFLOAT, buffer);
#ifdef SMP
@@ -567,7 +567,7 @@ void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
#endif
// for alignment
buffer_size = (buffer_size + 3) & ~3;
- STACK_ALLOC(buffer_size, FLOAT, buffer);
+ STACK_ALLOC(buffer_size, IFLOAT, buffer);
#ifdef SMP
--
2.41.0
From edfd4f52f3f22344863c233411ae792fb12aa81b Mon Sep 17 00:00:00 2001
From: Honza Horak <hhorak@redhat.com>
Date: Fri, 9 Feb 2024 09:53:40 +0100
Subject: [PATCH 2/6] Separate the interface for SBGEMMT from GEMMT due to
differences in GEMV arguments
Upstream commit:
commit d4db6a9f16a5c82bbe1860f591cc731c4d83d7c8
Author: Martin Kroeker <martin@ruby.chemie.uni-freiburg.de>
Date: Tue Feb 6 22:23:47 2024 +0100
Separate the interface for SBGEMMT from GEMMT due to differences in GEMV arguments
---
interface/CMakeLists.txt | 1 +
interface/Makefile | 4 +-
interface/sbgemmt.c | 447 +++++++++++++++++++++++++++++++++++++++
3 files changed, 450 insertions(+), 2 deletions(-)
create mode 100644 interface/sbgemmt.c
diff --git a/interface/CMakeLists.txt b/interface/CMakeLists.txt
index 4e082928b..3110f2e90 100644
--- a/interface/CMakeLists.txt
+++ b/interface/CMakeLists.txt
@@ -119,6 +119,7 @@ endif ()
if (BUILD_BFLOAT16)
GenerateNamedObjects("bf16dot.c" "" "sbdot" ${CBLAS_FLAG} "" "" true "BFLOAT16")
GenerateNamedObjects("gemm.c" "" "sbgemm" ${CBLAS_FLAG} "" "" true "BFLOAT16")
+ GenerateNamedObjects("gemmt.c" "" "sbgemmt" ${CBLAS_FLAG} "" "" true "BFLOAT16")
GenerateNamedObjects("sbgemv.c" "" "sbgemv" ${CBLAS_FLAG} "" "" true "BFLOAT16")
GenerateNamedObjects("tobf16.c" "SINGLE_PREC" "sbstobf16" ${CBLAS_FLAG} "" "" true "BFLOAT16")
GenerateNamedObjects("tobf16.c" "DOUBLE_PREC" "sbdtobf16" ${CBLAS_FLAG} "" "" true "BFLOAT16")
diff --git a/interface/Makefile b/interface/Makefile
index 78335357b..d106ca568 100644
--- a/interface/Makefile
+++ b/interface/Makefile
@@ -1301,7 +1301,7 @@ xhpr2.$(SUFFIX) xhpr2.$(PSUFFIX) : zhpr2.c
ifeq ($(BUILD_BFLOAT16),1)
sbgemm.$(SUFFIX) sbgemm.$(PSUFFIX) : gemm.c ../param.h
$(CC) -c $(CFLAGS) $< -o $(@F)
-sbgemmt.$(SUFFIX) sbgemmt.$(PSUFFIX) : gemmt.c ../param.h
+sbgemmt.$(SUFFIX) sbgemmt.$(PSUFFIX) : sbgemmt.c ../param.h
$(CC) -c $(CFLAGS) $< -o $(@F)
endif
@@ -1932,7 +1932,7 @@ cblas_sgemmt.$(SUFFIX) cblas_sgemmt.$(PSUFFIX) : gemmt.c ../param.h
$(CC) -DCBLAS -c $(CFLAGS) $< -o $(@F)
ifeq ($(BUILD_BFLOAT16),1)
-cblas_sbgemmt.$(SUFFIX) cblas_sbgemmt.$(PSUFFIX) : gemmt.c ../param.h
+cblas_sbgemmt.$(SUFFIX) cblas_sbgemmt.$(PSUFFIX) : sbgemmt.c ../param.h
$(CC) -DCBLAS -c $(CFLAGS) $< -o $(@F)
endif
diff --git a/interface/sbgemmt.c b/interface/sbgemmt.c
new file mode 100644
index 000000000..759af4bfb
--- /dev/null
+++ b/interface/sbgemmt.c
@@ -0,0 +1,447 @@
+/*********************************************************************/
+/* Copyright 2024, The OpenBLAS Project. */
+/* All rights reserved. */
+/* */
+/* Redistribution and use in source and binary forms, with or */
+/* without modification, are permitted provided that the following */
+/* conditions are met: */
+/* */
+/* 1. Redistributions of source code must retain the above */
+/* copyright notice, this list of conditions and the following */
+/* disclaimer. */
+/* */
+/* 2. Redistributions in binary form must reproduce the above */
+/* copyright notice, this list of conditions and the following */
+/* disclaimer in the documentation and/or other materials */
+/* provided with the distribution. */
+/* */
+/* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */
+/* AUSTIN ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, */
+/* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */
+/* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */
+/* DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY OF TEXAS AT */
+/* AUSTIN OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */
+/* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES */
+/* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE */
+/* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR */
+/* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF */
+/* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT */
+/* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT */
+/* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */
+/* POSSIBILITY OF SUCH DAMAGE. */
+/* */
+/*********************************************************************/
+
+#include <stdio.h>
+#include <stdlib.h>
+#include "common.h"
+
+#define SMP_THRESHOLD_MIN 65536.0
+#define ERROR_NAME "SBGEMMT "
+
+#ifndef GEMM_MULTITHREAD_THRESHOLD
+#define GEMM_MULTITHREAD_THRESHOLD 4
+#endif
+
+#ifndef CBLAS
+
+void NAME(char *UPLO, char *TRANSA, char *TRANSB,
+ blasint * M, blasint * K,
+ FLOAT * Alpha,
+ IFLOAT * a, blasint * ldA,
+ IFLOAT * b, blasint * ldB, FLOAT * Beta, FLOAT * c, blasint * ldC)
+{
+
+ blasint m, k;
+ blasint lda, ldb, ldc;
+ int transa, transb, uplo;
+ blasint info;
+
+ char transA, transB, Uplo;
+ blasint nrowa, nrowb;
+ IFLOAT *buffer;
+ IFLOAT *aa, *bb;
+ FLOAT *cc;
+ FLOAT alpha, beta;
+
+ PRINT_DEBUG_NAME;
+
+ m = *M;
+ k = *K;
+
+ alpha = *Alpha;
+ beta = *Beta;
+
+ lda = *ldA;
+ ldb = *ldB;
+ ldc = *ldC;
+
+ transA = *TRANSA;
+ transB = *TRANSB;
+ Uplo = *UPLO;
+ TOUPPER(transA);
+ TOUPPER(transB);
+ TOUPPER(Uplo);
+
+ transa = -1;
+ transb = -1;
+ uplo = -1;
+
+ if (transA == 'N')
+ transa = 0;
+ if (transA == 'T')
+ transa = 1;
+
+ if (transA == 'R')
+ transa = 0;
+ if (transA == 'C')
+ transa = 1;
+
+ if (transB == 'N')
+ transb = 0;
+ if (transB == 'T')
+ transb = 1;
+
+ if (transB == 'R')
+ transb = 0;
+ if (transB == 'C')
+ transb = 1;
+
+ if (Uplo == 'U')
+ uplo = 0;
+ if (Uplo == 'L')
+ uplo = 1;
+ nrowa = m;
+ if (transa & 1) nrowa = k;
+ nrowb = k;
+ if (transb & 1) nrowb = m;
+
+ info = 0;
+
+ if (ldc < MAX(1, m))
+ info = 13;
+ if (ldb < MAX(1, nrowb))
+ info = 10;
+ if (lda < MAX(1, nrowa))
+ info = 8;
+ if (k < 0)
+ info = 5;
+ if (m < 0)
+ info = 4;
+ if (transb < 0)
+ info = 3;
+ if (transa < 0)
+ info = 2;
+ if (uplo < 0)
+ info = 1;
+
+ if (info != 0) {
+ BLASFUNC(xerbla) (ERROR_NAME, &info, sizeof(ERROR_NAME));
+ return;
+ }
+#else
+
+void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo,
+ enum CBLAS_TRANSPOSE TransA, enum CBLAS_TRANSPOSE TransB, blasint m,
+ blasint k,
+ FLOAT alpha,
+ IFLOAT * A, blasint LDA,
+ IFLOAT * B, blasint LDB, FLOAT beta, FLOAT * c, blasint ldc)
+{
+ IFLOAT *aa, *bb;
+ FLOAT *cc;
+
+ int transa, transb, uplo;
+ blasint info;
+ blasint lda, ldb;
+ IFLOAT *a, *b;
+ XFLOAT *buffer;
+
+ PRINT_DEBUG_CNAME;
+
+ uplo = -1;
+ transa = -1;
+ transb = -1;
+ info = 0;
+
+ if (order == CblasColMajor) {
+ if (Uplo == CblasUpper) uplo = 0;
+ if (Uplo == CblasLower) uplo = 1;
+
+ if (TransA == CblasNoTrans)
+ transa = 0;
+ if (TransA == CblasTrans)
+ transa = 1;
+
+ if (TransA == CblasConjNoTrans)
+ transa = 0;
+ if (TransA == CblasConjTrans)
+ transa = 1;
+
+ if (TransB == CblasNoTrans)
+ transb = 0;
+ if (TransB == CblasTrans)
+ transb = 1;
+
+ if (TransB == CblasConjNoTrans)
+ transb = 0;
+ if (TransB == CblasConjTrans)
+ transb = 1;
+
+ a = (void *)A;
+ b = (void *)B;
+ lda = LDA;
+ ldb = LDB;
+
+ info = -1;
+
+ blasint nrowa;
+ blasint nrowb;
+ nrowa = m;
+ if (transa & 1) nrowa = k;
+ nrowb = k;
+ if (transb & 1) nrowb = m;
+
+ if (ldc < MAX(1, m))
+ info = 13;
+ if (ldb < MAX(1, nrowb))
+ info = 10;
+ if (lda < MAX(1, nrowa))
+ info = 8;
+ if (k < 0)
+ info = 5;
+ if (m < 0)
+ info = 4;
+ if (transb < 0)
+ info = 3;
+ if (transa < 0)
+ info = 2;
+ if (uplo < 0)
+ info = 1;
+ }
+
+ if (order == CblasRowMajor) {
+
+ a = (void *)B;
+ b = (void *)A;
+
+ lda = LDB;
+ ldb = LDA;
+
+ if (Uplo == CblasUpper) uplo = 0;
+ if (Uplo == CblasLower) uplo = 1;
+
+ if (TransB == CblasNoTrans)
+ transa = 0;
+ if (TransB == CblasTrans)
+ transa = 1;
+
+ if (TransB == CblasConjNoTrans)
+ transa = 0;
+ if (TransB == CblasConjTrans)
+ transa = 1;
+
+ if (TransA == CblasNoTrans)
+ transb = 0;
+ if (TransA == CblasTrans)
+ transb = 1;
+
+ if (TransA == CblasConjNoTrans)
+ transb = 0;
+ if (TransA == CblasConjTrans)
+ transb = 1;
+
+ info = -1;
+
+ blasint ncola;
+ blasint ncolb;
+
+ ncola = m;
+ if (transa & 1) ncola = k;
+ ncolb = k;
+
+ if (transb & 1) {
+ ncolb = m;
+ }
+
+ if (ldc < MAX(1,m))
+ info = 13;
+ if (ldb < MAX(1, ncolb))
+ info = 8;
+ if (lda < MAX(1, ncola))
+ info = 10;
+ if (k < 0)
+ info = 5;
+ if (m < 0)
+ info = 4;
+ if (transb < 0)
+ info = 2;
+ if (transa < 0)
+ info = 3;
+ if (uplo < 0)
+ info = 1;
+ }
+
+ if (info >= 0) {
+ BLASFUNC(xerbla) (ERROR_NAME, &info, sizeof(ERROR_NAME));
+ return;
+ }
+
+#endif
+ int buffer_size;
+ blasint i, j;
+
+#ifdef SMP
+ int nthreads;
+#endif
+
+
+#ifdef SMP
+ static int (*gemv_thread[]) (BLASLONG, BLASLONG, FLOAT, IFLOAT *,
+ BLASLONG, IFLOAT *, BLASLONG, FLOAT,
+ FLOAT *, BLASLONG, int) = {
+ sbgemv_thread_n, sbgemv_thread_t,
+ };
+#endif
+ int (*gemv[]) (BLASLONG, BLASLONG, FLOAT, IFLOAT *, BLASLONG,
+ IFLOAT *, BLASLONG, FLOAT, FLOAT *, BLASLONG) = {
+ SBGEMV_N, SBGEMV_T,};
+
+
+ if (m == 0)
+ return;
+
+ IDEBUG_START;
+
+ const blasint incb = ((transb & 1) == 0) ? 1 : ldb;
+
+ if (uplo == 1) {
+ for (i = 0; i < m; i++) {
+ j = m - i;
+
+ aa = a + i;
+ bb = b + i * ldb;
+ if (transa & 1) {
+ aa = a + lda * i;
+ }
+ if (transb & 1)
+ bb = b + i;
+ cc = c + i * ldc + i;
+
+#if 0
+ if (beta != ONE)
+ SCAL_K(l, 0, 0, beta, cc, 1, NULL, 0, NULL, 0);
+
+ if (alpha == ZERO)
+ continue;
+#endif
+
+ IDEBUG_START;
+
+ buffer_size = j + k + 128 / sizeof(FLOAT);
+#ifdef WINDOWS_ABI
+ buffer_size += 160 / sizeof(FLOAT);
+#endif
+ // for alignment
+ buffer_size = (buffer_size + 3) & ~3;
+ STACK_ALLOC(buffer_size, IFLOAT, buffer);
+
+#ifdef SMP
+
+ if (1L * j * k < 2304L * GEMM_MULTITHREAD_THRESHOLD)
+ nthreads = 1;
+ else
+ nthreads = num_cpu_avail(2);
+
+ if (nthreads == 1) {
+#endif
+
+ if (!(transa & 1))
+ (gemv[(int)transa]) (j, k, alpha, aa, lda,
+ bb, incb, beta, cc, 1);
+ else
+ (gemv[(int)transa]) (k, j, alpha, aa, lda,
+ bb, incb, beta, cc, 1);
+
+#ifdef SMP
+ } else {
+ if (!(transa & 1))
+ (gemv_thread[(int)transa]) (j, k, alpha, aa,
+ lda, bb, incb, beta, cc,
+ 1, nthreads);
+ else
+ (gemv_thread[(int)transa]) (k, j, alpha, aa,
+ lda, bb, incb, beta, cc,
+ 1, nthreads);
+
+ }
+#endif
+
+ STACK_FREE(buffer);
+ }
+ } else {
+
+ for (i = 0; i < m; i++) {
+ j = i + 1;
+
+ bb = b + i * ldb;
+ if (transb & 1) {
+ bb = b + i;
+ }
+ cc = c + i * ldc;
+
+#if 0
+ if (beta != ONE)
+ SCAL_K(l, 0, 0, beta, cc, 1, NULL, 0, NULL, 0);
+
+ if (alpha == ZERO)
+ continue;
+#endif
+ IDEBUG_START;
+
+ buffer_size = j + k + 128 / sizeof(FLOAT);
+#ifdef WINDOWS_ABI
+ buffer_size += 160 / sizeof(FLOAT);
+#endif
+ // for alignment
+ buffer_size = (buffer_size + 3) & ~3;
+ STACK_ALLOC(buffer_size, IFLOAT, buffer);
+
+#ifdef SMP
+
+ if (1L * j * k < 2304L * GEMM_MULTITHREAD_THRESHOLD)
+ nthreads = 1;
+ else
+ nthreads = num_cpu_avail(2);
+
+ if (nthreads == 1) {
+#endif
+
+ if (!(transa & 1))
+ (gemv[(int)transa]) (j, k, alpha, a, lda, bb,
+ incb, beta, cc, 1);
+ else
+ (gemv[(int)transa]) (k, j, alpha, a, lda, bb,
+ incb, beta, cc, 1);
+
+#ifdef SMP
+ } else {
+ if (!(transa & 1))
+ (gemv_thread[(int)transa]) (j, k, alpha, a, lda,
+ bb, incb, beta, cc, 1,
+ nthreads);
+ else
+ (gemv_thread[(int)transa]) (k, j, alpha, a, lda,
+ bb, incb, beta, cc, 1,
+ nthreads);
+ }
+#endif
+
+ STACK_FREE(buffer);
+ }
+ }
+
+ IDEBUG_END;
+
+ return;
+}
--
2.41.0
From 9a4c2d61a345866e4540f9d6da87eb881419b411 Mon Sep 17 00:00:00 2001
From: Honza Horak <hhorak@redhat.com>
Date: Fri, 9 Feb 2024 09:54:52 +0100
Subject: [PATCH 3/6] fix type conversion warnings
upstream commit:
commit fb99fc2e6e4ec8ecdcfffe1ca1aeb787464d2825
Author: Martin Kroeker <martin@ruby.chemie.uni-freiburg.de>
Date: Wed Feb 7 13:42:08 2024 +0100
fix type conversion warnings
---
test/compare_sgemm_sbgemm.c | 18 ++++++++++++++----
1 file changed, 14 insertions(+), 4 deletions(-)
diff --git a/test/compare_sgemm_sbgemm.c b/test/compare_sgemm_sbgemm.c
index cf808b56d..4afa8bf93 100644
--- a/test/compare_sgemm_sbgemm.c
+++ b/test/compare_sgemm_sbgemm.c
@@ -81,6 +81,16 @@ float16to32 (bfloat16_bits f16)
return f32.v;
}
+float
+float32to16 (float32_bits f32)
+{
+ bfloat16_bits f16;
+ f16.bits.s = f32.bits.s;
+ f16.bits.e = f32.bits.e;
+ f16.bits.m = (uint32_t) f32.bits.m >> 16;
+ return f32.v;
+}
+
int
main (int argc, char *argv[])
{
@@ -108,16 +118,16 @@ main (int argc, char *argv[])
A[j * k + i] = ((FLOAT) rand () / (FLOAT) RAND_MAX) + 0.5;
B[j * k + i] = ((FLOAT) rand () / (FLOAT) RAND_MAX) + 0.5;
C[j * k + i] = 0;
- AA[j * k + i].v = *(uint32_t *) & A[j * k + i] >> 16;
- BB[j * k + i].v = *(uint32_t *) & B[j * k + i] >> 16;
+ AA[j * k + i].v = float32to16( A[j * k + i] );
+ BB[j * k + i].v = float32to16( B[j * k + i] );
CC[j * k + i] = 0;
DD[j * k + i] = 0;
}
}
SGEMM (&transA, &transB, &m, &n, &k, &alpha, A,
&m, B, &k, &beta, C, &m);
- SBGEMM (&transA, &transB, &m, &n, &k, &alpha, AA,
- &m, BB, &k, &beta, CC, &m);
+ SBGEMM (&transA, &transB, &m, &n, &k, &alpha, (bfloat16*) AA,
+ &m, (bfloat16*)BB, &k, &beta, CC, &m);
for (i = 0; i < n; i++)
for (j = 0; j < m; j++)
if (fabs (CC[i * m + j] - C[i * m + j]) > 1.0)
--
2.41.0
From 5593507ddbd5d35d088cd4db6285de6b9d84a405 Mon Sep 17 00:00:00 2001
From: Honza Horak <hhorak@redhat.com>
Date: Fri, 9 Feb 2024 09:56:11 +0100
Subject: [PATCH 4/6] fix prototype for c/zaxpby
Upstream commit:
commit b3fa16345d83b723b8984b78dc6a2bb5d9f3d479
Author: Martin Kroeker <martin@ruby.chemie.uni-freiburg.de>
Date: Thu Feb 8 13:15:34 2024 +0100
fix prototype for c/zaxpby
---
common_interface.h | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/common_interface.h b/common_interface.h
index 318827920..1f6cb5f6d 100644
--- a/common_interface.h
+++ b/common_interface.h
@@ -764,8 +764,8 @@ xdouble BLASFUNC(qlamc3)(xdouble *, xdouble *);
void BLASFUNC(saxpby) (blasint *, float *, float *, blasint *, float *, float *, blasint *);
void BLASFUNC(daxpby) (blasint *, double *, double *, blasint *, double *, double *, blasint *);
-void BLASFUNC(caxpby) (blasint *, float *, float *, blasint *, float *, float *, blasint *);
-void BLASFUNC(zaxpby) (blasint *, double *, double *, blasint *, double *, double *, blasint *);
+void BLASFUNC(caxpby) (blasint *, void *, float *, blasint *, void *, float *, blasint *);
+void BLASFUNC(zaxpby) (blasint *, void *, double *, blasint *, void *, double *, blasint *);
void BLASFUNC(somatcopy) (char *, char *, blasint *, blasint *, float *, float *, blasint *, float *, blasint *);
void BLASFUNC(domatcopy) (char *, char *, blasint *, blasint *, double *, double *, blasint *, double *, blasint *);
--
2.41.0
From 42b30ed2c54034b2b1dbb15bb9e3e705e704b6a9 Mon Sep 17 00:00:00 2001
From: Honza Horak <hhorak@redhat.com>
Date: Fri, 9 Feb 2024 09:56:46 +0100
Subject: [PATCH 5/6] fix incompatible pointer types
Upstream commit:
commit 500ac4de5e20596d5cd797d745db97dd0a62ff86
Author: Martin Kroeker <martin@ruby.chemie.uni-freiburg.de>
Date: Thu Feb 8 13:18:34 2024 +0100
fix incompatible pointer types
---
interface/zaxpby.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/interface/zaxpby.c b/interface/zaxpby.c
index 3a4db7403..e5065270d 100644
--- a/interface/zaxpby.c
+++ b/interface/zaxpby.c
@@ -39,12 +39,14 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#ifndef CBLAS
-void NAME(blasint *N, FLOAT *ALPHA, FLOAT *x, blasint *INCX, FLOAT *BETA, FLOAT *y, blasint *INCY)
+void NAME(blasint *N, void *VALPHA, FLOAT *x, blasint *INCX, void *VBETA, FLOAT *y, blasint *INCY)
{
blasint n = *N;
blasint incx = *INCX;
blasint incy = *INCY;
+ FLOAT* ALPHA = (FLOAT*) VALPHA;
+ FLOAT* BETA = (FLOAT*) VBETA;
#else
--
2.41.0
From 1c525b6e704523912a04fbd026300a2ff95341f3 Mon Sep 17 00:00:00 2001
From: Honza Horak <hhorak@redhat.com>
Date: Fri, 9 Feb 2024 15:29:17 +0100
Subject: [PATCH 6/6] fix sbgemm bfloat16 conversion errors introduced in PR
4488
Upstream commit:
commit e9f480111e1d5b6f69c8053f79375b0a4242712f
Author: Martin Kroeker <martin@ruby.chemie.uni-freiburg.de>
Date: Wed Feb 7 19:57:18 2024 +0100
fix sbgemm bfloat16 conversion errors introduced in PR 4488
---
test/compare_sgemm_sbgemm.c | 18 ++++++------------
1 file changed, 6 insertions(+), 12 deletions(-)
diff --git a/test/compare_sgemm_sbgemm.c b/test/compare_sgemm_sbgemm.c
index 4afa8bf93..bc74233ab 100644
--- a/test/compare_sgemm_sbgemm.c
+++ b/test/compare_sgemm_sbgemm.c
@@ -81,16 +81,6 @@ float16to32 (bfloat16_bits f16)
return f32.v;
}
-float
-float32to16 (float32_bits f32)
-{
- bfloat16_bits f16;
- f16.bits.s = f32.bits.s;
- f16.bits.e = f32.bits.e;
- f16.bits.m = (uint32_t) f32.bits.m >> 16;
- return f32.v;
-}
-
int
main (int argc, char *argv[])
{
@@ -110,6 +100,8 @@ main (int argc, char *argv[])
float C[m * n];
bfloat16_bits AA[m * k], BB[k * n];
float DD[m * n], CC[m * n];
+ bfloat16 atmp,btmp;
+ blasint one=1;
for (j = 0; j < m; j++)
{
@@ -118,8 +110,10 @@ main (int argc, char *argv[])
A[j * k + i] = ((FLOAT) rand () / (FLOAT) RAND_MAX) + 0.5;
B[j * k + i] = ((FLOAT) rand () / (FLOAT) RAND_MAX) + 0.5;
C[j * k + i] = 0;
- AA[j * k + i].v = float32to16( A[j * k + i] );
- BB[j * k + i].v = float32to16( B[j * k + i] );
+ sbstobf16_(&one, &A[j*k+i], &one, &atmp, &one);
+ sbstobf16_(&one, &B[j*k+i], &one, &btmp, &one);
+ AA[j * k + i].v = atmp;
+ BB[j * k + i].v = btmp;
CC[j * k + i] = 0;
DD[j * k + i] = 0;
}
--
2.41.0

73
openblas.spec
View File

@ -1,6 +1,6 @@
%bcond_with system_lapack
# Version of bundled lapack
%global lapackver 3.9.1
%global lapackver 3.11.0
# DO NOT "CLEAN UP" OR MODIFY THIS SPEC FILE WITHOUT ASKING THE
# MAINTAINER FIRST!
@ -14,28 +14,24 @@
# "obsoleted" features are still kept in the spec.
Name: openblas
Version: 0.3.21
Release: 5%{?dist}
Version: 0.3.26
Release: 5.0.riscv64%{?dist}
Summary: An optimized BLAS library based on GotoBLAS2
License: BSD
URL: https://github.com/xianyi/OpenBLAS/
Source0: https://github.com/xianyi/OpenBLAS/archive/v%{version}/openblas-%{version}.tar.gz
License: BSD-3-Clause
URL: https://github.com/OpenMathLib/OpenBLAS
Source0: %url/archive/v%{version}/OpenBLAS-%{version}.tar.gz
# Use system lapack
Patch0: openblas-0.2.15-system_lapack.patch
# Drop extra p from threaded library name
Patch1: openblas-0.2.5-libname.patch
# Don't use constructor priorities on too old architectures
Patch2: openblas-0.2.15-constructor.patch
# Fix SBGEMM test to work with INTERFACE64
Patch3: openblas-0.3.21-sbgemm-test.patch
# Supply the proper flags to the test makefile
Patch4: openblas-0.3.11-tests.patch
# C99 porting.
Patch5: openblas-0.3.21-c99-1.patch
Patch6: openblas-0.3.21-c99-2.patch
Patch7: openblas-0.3.21-c99-3.patch
Patch8: openblas-0.3.21-c99-4.patch
Patch9: openblas-0.3.21-c99-5.patch
Patch3: openblas-0.3.11-tests.patch
# Fix incompatible pointer types (causes FTBFS on ppc64le)
Patch4: openblas-0.3.26-incompatibletypes.patch
BuildRequires: make
BuildRequires: gcc
@ -248,13 +244,8 @@ cd OpenBLAS-%{version}
%if 0%{?rhel} == 5
%patch 2 -p1 -b .constructor
%endif
%patch3 -p1 -b .sbgemm
%patch4 -p1 -b .tests
%patch5 -p1 -b .c99-1
%patch6 -p1 -b .c99-2
%patch7 -p1 -b .c99-3
%patch8 -p1 -b .c99-4
%patch9 -p1 -b .c99-5
%patch 3 -p1 -b .tests
%patch 4 -p1 -b .incompatibletypes
# Fix source permissions
find -name \*.f -exec chmod 644 {} \;
@ -392,6 +383,9 @@ TARGET="TARGET=ARMV8 DYNAMIC_ARCH=1 DYNAMIC_OLDER=1"
%ifarch s390x
TARGET="TARGET=ZARCH_GENERIC DYNAMIC_ARCH=1 DYNAMIC_OLDER=1"
%endif
%ifarch riscv64
TARGET="TARGET=RISCV64_GENERIC DYNAMIC_ARCH=0"
%endif
%if 0%{?rhel} == 5
# Gfortran too old to recognize -frecursive
@ -454,6 +448,9 @@ suffix=""
%ifarch armv7hl
suffix="_armv7"
%endif
%ifarch riscv64
suffix="_riscv64_generic"
%endif
slibname=`basename %{buildroot}%{_libdir}/libopenblas${suffix}-*.so .so`
mv %{buildroot}%{_libdir}/${slibname}.a %{buildroot}%{_libdir}/lib%{name}.a
if [[ "$suffix" != "" ]]; then
@ -660,6 +657,38 @@ rm -rf %{buildroot}%{_libdir}/pkgconfig
%endif
%changelog
* Thu Nov 21 2024 David Abdurachmanov <davidlt@rivosinc.com> - 0.3.26-5.0.riscv64
- Add support for riscv64
* Thu Jul 18 2024 Fedora Release Engineering <releng@fedoraproject.org> - 0.3.26-5
- Rebuilt for https://fedoraproject.org/wiki/Fedora_41_Mass_Rebuild
* Fri Feb 09 2024 Honza Horak <hhorak@redhat.com> - 0.3.26-4
- Fix FTBFS on ppc64le and s390x
Resolves: BZ#2261415
Resolves: RHEL-24745
* Thu Jan 25 2024 Fedora Release Engineering <releng@fedoraproject.org> - 0.3.26-3
- Rebuilt for https://fedoraproject.org/wiki/Fedora_40_Mass_Rebuild
* Sun Jan 21 2024 Fedora Release Engineering <releng@fedoraproject.org> - 0.3.26-2
- Rebuilt for https://fedoraproject.org/wiki/Fedora_40_Mass_Rebuild
* Sun Jan 07 2024 İsmail Dönmez <ismail@i10z.com> - 0.3.26-1
- Update to 0.3.26
* Thu Nov 30 2023 İsmail Dönmez <ismail@i10z.com> - 0.3.25-1
- Update to 0.3.25
* Fri Oct 27 2023 Ali Erdinc Koroglu <aekoroglu@fedoraproject.org> - 0.3.24-1
- Update to 0.3.24
* Tue Aug 01 2023 Pavel Šimovec <psimovec@redhat.com> - 0.3.23-1
- Update to 0.3.23 (RHBZ #2182038)
* Thu Jul 20 2023 Fedora Release Engineering <releng@fedoraproject.org> - 0.3.21-6
- Rebuilt for https://fedoraproject.org/wiki/Fedora_39_Mass_Rebuild
* Thu Jan 19 2023 Fedora Release Engineering <releng@fedoraproject.org> - 0.3.21-5
- Rebuilt for https://fedoraproject.org/wiki/Fedora_38_Mass_Rebuild

2
sources
View File

@ -1 +1 @@
SHA512 (openblas-0.3.21.tar.gz) = 4625c8e6ccfa9120281fd714d3f6b7c3ba2265470c1be76121d6b25dc3dacb899d26e5d9a417ddc616d23909f1411495aa995ef8d8d6df8511cd5cefbabcb1c5
SHA512 (OpenBLAS-0.3.26.tar.gz) = 01d3a536fbfa62f276fd6b1ad0e218fb3d91f41545fc83ddc74979fa26372d8389f0baa20334badfe0adacd77bd944c50a47ac920577373fcc1d495553084373