sci-libs/openblas: add 0.3.21

I had to rebase the loongarch patch, and removed an old workaround for
FFLAGS="-fcheck=bounds" in the test suite, but other than that a
standard update.

Bug: https://bugs.gentoo.org/726474
Closes: https://bugs.gentoo.org/878883
Signed-off-by: Michael Orlitzky <mjo@gentoo.org>

1 files changed, 237 insertions, 0 deletions

diff --git a/sci-libs/openblas/files/openblas-0.3.21-fix-loong.patch b/sci-libs/openblas/files/openblas-0.3.21-fix-loong.patch
new file mode 100644
index 000000000000..ed112d62380b
--- /dev/null
+++ b/sci-libs/openblas/files/openblas-0.3.21-fix-loong.patch
@@ -0,0 +1,237 @@
+https://github.com/xianyi/OpenBLAS/pull/3626
+
+From 5af7b8638b3b972bd0e25597ec07927fffd52108 Mon Sep 17 00:00:00 2001
+From: WANG Xuerui <xen0n@gentoo.org>
+Date: Fri, 13 May 2022 15:26:33 +0800
+Subject: [PATCH 1/3] Move LoongArch make rules to Makefile.loongarch64
+
+---
+ Makefile.loongarch64 | 5 +++--
+ Makefile.system      | 5 -----
+ 2 files changed, 3 insertions(+), 7 deletions(-)
+
+diff --git a/Makefile.loongarch64 b/Makefile.loongarch64
+index 05ea9c679d..fced1a9462 100644
+--- a/Makefile.loongarch64
++++ b/Makefile.loongarch64
+@@ -1,3 +1,4 @@
+-ifdef BINARY64
+-else
++ifeq ($(CORE), LOONGSON3R5)
++CCOMMON_OPT += -march=loongarch64 -mabi=lp64
++FCOMMON_OPT += -march=loongarch64 -mabi=lp64
+ endif
+diff --git a/Makefile.system b/Makefile.system
+index 3be5efa0c2..9c91425b0f 100644
+--- a/Makefile.system
++++ b/Makefile.system
+@@ -895,11 +895,6 @@ ifeq ($(OSNAME), AIX)
+ BINARY_DEFINED = 1
+ endif
+ 
+-ifeq ($(ARCH), loongarch64)
+-CCOMMON_OPT += -march=loongarch64 -mabi=lp64
+-FCOMMON_OPT += -march=loongarch64 -mabi=lp64
+-endif
+-
+ endif
+ 
+ ifndef BINARY_DEFINED
+
+From 869061c783d60ba0b799fab643952c493b57fb0a Mon Sep 17 00:00:00 2001
+From: WANG Xuerui <xen0n@gentoo.org>
+Date: Fri, 13 May 2022 15:53:22 +0800
+Subject: [PATCH 2/3] Probe for old-world LASX flavor on LoongArch and use
+ generic DGEMM kernel otherwise
+
+See: https://bugs.gentoo.org/844013
+---
+ Makefile.system                       |  1 +
+ c_check                               | 27 ++++++++++++++++++++++++++-
+ kernel/loongarch64/KERNEL.LOONGSON3R5 |  5 +++++
+ 3 files changed, 32 insertions(+), 1 deletion(-)
+
+diff --git a/Makefile.system b/Makefile.system
+index 9c91425b0f..b86aafbddb 100644
+--- a/Makefile.system
++++ b/Makefile.system
+@@ -1702,6 +1702,7 @@ export TARGET_CORE
+ export NO_AVX512
+ export NO_AVX2
+ export BUILD_BFLOAT16
++export NO_LASX
+ 
+ export SBGEMM_UNROLL_M
+ export SBGEMM_UNROLL_N
+diff --git a/c_check b/c_check
+index 01d4f4a7cf..a37e91f82c 100755
+--- a/c_check
++++ b/c_check
+@@ -112,7 +112,7 @@ case "$architecture" in
+     	defined=1
+     	;;
+     arm|arm64) defined=1 ;;
+-    zarch|e2k|alpha|ia64|riscv64|loonarch64)
++    zarch|e2k|alpha|ia64|riscv64|loongarch64)
+     	defined=1
+     	BINARY=64
+     	;;
+@@ -240,6 +240,29 @@ if [ "$architecture" = "riscv64" ]; then
+     rm -rf "$tmpd"
+ fi
+ 
++no_lasx=0
++if [ "$architecture" = "loongarch64" ]; then
++    tmpd=`mktemp -d`
++    tmpf="$tmpd/a.c"
++    # Old-world assembly flavor: LASX registers named "$xrNN", different
++    # from the ISA manual which suggests "$xNN". This is the flavor we
++    # currently support.
++    #
++    # As the LASX ISA manual is not out yet, we cannot predict what the
++    # new-world flavor would look like, so do not probe flavor for now.
++    # The compiler flags are also unsuitable for new-world gcc.
++    code='"xvld $xr0, $a0, 0\n"'
++    printf "int main(void){ __asm__ volatile(%s); }\n" "$code" >> "$tmpf"
++    args=" -march=loongarch64 -mabi=lp64 -mlasx -c -o $tmpf.o $tmpf"
++    no_lasx=0
++    {
++        $compiler_name $flags $args >/dev/null 2>&1
++    } || {
++        no_lasx=1
++    }
++    rm -rf "$tmpd"
++fi
++
+ c11_atomics=0
+ case "$data" in
+     *HAVE_C11*)
+@@ -350,6 +373,7 @@ done
+ 
+  [ "$makefile" = "-" ] && {
+     [ "$no_rv64gv" -eq 1 ] && printf "NO_RV64GV=1\n"
++    [ "$no_lasx" -eq 1 ] && printf "NO_LASX=1\n"
+     [ "$no_avx512" -eq 1 ] && printf "NO_AVX512=1\n"
+     [ "$no_avx2" -eq 1 ] && printf "NO_AVX2=1\n"
+     [ "$oldgcc" -eq 1 ] && printf "OLDGCC=1\n"
+@@ -380,6 +404,7 @@ done
+         printf "MSA_FLAGS=%s\n" "$msa_flags"
+     }
+     [ "$no_rv64gv" -eq 1 ] && printf "NO_RV64GV=1\n"
++    [ "$no_lasx" -eq 1 ] && printf "NO_LASX=1\n"
+     [ "$no_avx512" -eq 1 ] && printf "NO_AVX512=1\n"
+     [ "$no_avx2" -eq 1 ] && printf "NO_AVX2=1\n"
+     [ "$oldgcc" -eq 1 ] && printf "OLDGCC=1\n"
+diff --git a/kernel/loongarch64/KERNEL.LOONGSON3R5 b/kernel/loongarch64/KERNEL.LOONGSON3R5
+index cda3590402..79b57d1e4d 100644
+--- a/kernel/loongarch64/KERNEL.LOONGSON3R5
++++ b/kernel/loongarch64/KERNEL.LOONGSON3R5
+@@ -1,3 +1,7 @@
++ifeq ($(NO_LASX), 1)
++# No LASX support in compiler, unable to consume the optimized version
++# leave out DGEMMKERNEL to pull in the generic version
++else
+ DGEMMKERNEL    = dgemm_kernel_16x4.S
+ DGEMMINCOPY    = dgemm_ncopy_16.S
+ DGEMMITCOPY    = dgemm_tcopy_16.S
+@@ -7,6 +11,7 @@ DGEMMINCOPYOBJ = dgemm_incopy$(TSUFFIX).$(SUFFIX)
+ DGEMMITCOPYOBJ = dgemm_itcopy$(TSUFFIX).$(SUFFIX)
+ DGEMMONCOPYOBJ = dgemm_oncopy$(TSUFFIX).$(SUFFIX)
+ DGEMMOTCOPYOBJ = dgemm_otcopy$(TSUFFIX).$(SUFFIX)
++endif
+ 
+ DTRSMKERNEL_LN  = ../generic/trsm_kernel_LN.c
+ DTRSMKERNEL_LT  = ../generic/trsm_kernel_LT.c
+
+From 699c57d3767c9a5a5a69262a4b70a1f80e0bb2e2 Mon Sep 17 00:00:00 2001
+From: WANG Xuerui <xen0n@gentoo.org>
+Date: Fri, 13 May 2022 16:23:46 +0800
+Subject: [PATCH 3/3] Fix ABI CFLAGS setting on new-world LoongArch
+
+See: https://bugs.gentoo.org/844013
+---
+ Makefile.loongarch64 | 13 +++++++++++--
+ Makefile.system      |  1 +
+ c_check              | 14 ++++++++++++++
+ 3 files changed, 26 insertions(+), 2 deletions(-)
+
+diff --git a/Makefile.loongarch64 b/Makefile.loongarch64
+index fced1a9462..d6cbe848cd 100644
+--- a/Makefile.loongarch64
++++ b/Makefile.loongarch64
+@@ -1,4 +1,13 @@
+ ifeq ($(CORE), LOONGSON3R5)
+-CCOMMON_OPT += -march=loongarch64 -mabi=lp64
+-FCOMMON_OPT += -march=loongarch64 -mabi=lp64
++CCOMMON_OPT += -march=loongarch64
++FCOMMON_OPT += -march=loongarch64
++
++ifeq ($(LOONGARCH_CC_FLAVOR), NEW)
++CCOMMON_OPT += -mabi=lp64d
++FCOMMON_OPT += -mabi=lp64d
++endif
++ifeq ($(LOONGARCH_CC_FLAVOR), OLD)
++CCOMMON_OPT += -mabi=lp64
++FCOMMON_OPT += -mabi=lp64
++endif
+ endif
+diff --git a/Makefile.system b/Makefile.system
+index b86aafbddb..4360a861cb 100644
+--- a/Makefile.system
++++ b/Makefile.system
+@@ -1702,6 +1702,7 @@ export TARGET_CORE
+ export NO_AVX512
+ export NO_AVX2
+ export BUILD_BFLOAT16
++export LOONGARCH_CC_FLAVOR
+ export NO_LASX
+ 
+ export SBGEMM_UNROLL_M
+diff --git a/c_check b/c_check
+index a37e91f82c..573f0e52fd 100755
+--- a/c_check
++++ b/c_check
+@@ -240,8 +240,19 @@ if [ "$architecture" = "riscv64" ]; then
+     rm -rf "$tmpd"
+ fi
+ 
++loongarch_cc_flavor=""
+ no_lasx=0
+ if [ "$architecture" = "loongarch64" ]; then
++    # Currently there are only 2 flavors: new-world and old-world.
++    # They differ in ABI names accepted, so check it to avoid creating temp
++    # files.
++    loongarch_cc_flavor="NEW"
++    {
++        $compiler_name -mabi=lp64d -E -o - /dev/null >/dev/null 2>&1
++    } || {
++        loongarch_cc_flavor="OLD"
++    }
++
+     tmpd=`mktemp -d`
+     tmpf="$tmpd/a.c"
+     # Old-world assembly flavor: LASX registers named "$xrNN", different
+@@ -260,6 +271,7 @@ if [ "$architecture" = "loongarch64" ]; then
+     } || {
+         no_lasx=1
+     }
++
+     rm -rf "$tmpd"
+ fi
+ 
+@@ -377,6 +389,7 @@ done
+     [ "$no_avx512" -eq 1 ] && printf "NO_AVX512=1\n"
+     [ "$no_avx2" -eq 1 ] && printf "NO_AVX2=1\n"
+     [ "$oldgcc" -eq 1 ] && printf "OLDGCC=1\n"
++    [ -n "$loongarch_cc_flavor" ] && printf "LOONGARCH_CC_FLAVOR=%s\n" "$loongarch_cc_flavor"
+     exit 0
+ }
+ 
+@@ -408,6 +421,7 @@ done
+     [ "$no_avx512" -eq 1 ] && printf "NO_AVX512=1\n"
+     [ "$no_avx2" -eq 1 ] && printf "NO_AVX2=1\n"
+     [ "$oldgcc" -eq 1 ] && printf "OLDGCC=1\n"
++    [ -n "$loongarch_cc_flavor" ] && printf "LOONGARCH_CC_FLAVOR=%s\n" "$loongarch_cc_flavor"
+ } >> "$makefile"
+ 
+ os=`echo "$os" | tr '[[:lower:]]' '[[:upper:]]'/ `