From 9e169ae243ce315fb779f5d3f84c28b9dbc39612 Mon Sep 17 00:00:00 2001
From: FJ-NaokiMatsumura <98728917+FJ-NaokiMatsumura@users.noreply.github.com>
Date: Tue, 29 Mar 2022 00:16:03 +0900
Subject: batchedblas: new package (#29028)

---
 .../repos/builtin/packages/batchedblas/AVX2.patch  | 11 ++++++
 .../repos/builtin/packages/batchedblas/package.py  | 41 ++++++++++++++++++++++
 2 files changed, 52 insertions(+)
 create mode 100644 var/spack/repos/builtin/packages/batchedblas/AVX2.patch
 create mode 100644 var/spack/repos/builtin/packages/batchedblas/package.py

diff --git a/var/spack/repos/builtin/packages/batchedblas/AVX2.patch b/var/spack/repos/builtin/packages/batchedblas/AVX2.patch
new file mode 100644
index 0000000000..219aed9ad5
--- /dev/null
+++ b/var/spack/repos/builtin/packages/batchedblas/AVX2.patch
@@ -0,0 +1,11 @@
+--- spack-src/bblas_src/batched_blas_fp16.h.orig	2022-02-16 17:15:22.000000000 +0900
++++ spack-src/bblas_src/batched_blas_fp16.h	2022-02-16 17:15:31.000000000 +0900
+@@ -13,8 +13,6 @@
+ #  if BF_NMANT>7 || BF_NMANT<=1
+ #    error "too large or small mantissa for BFLIKE_FLOAT"
+ #  endif
+-#elif defined(__AVX2__)
+-#  define FP16_AVX2_EMULATION
+ #elif defined(__clang__) && __clang_major__ >= 8
+ #  define FP16_AUTO_PROMOTION
+ #else
diff --git a/var/spack/repos/builtin/packages/batchedblas/package.py b/var/spack/repos/builtin/packages/batchedblas/package.py
new file mode 100644
index 0000000000..0c92cedd72
--- /dev/null
+++ b/var/spack/repos/builtin/packages/batchedblas/package.py
@@ -0,0 +1,41 @@
+# Copyright 2013-2022 Lawrence Livermore National Security, LLC and other
+# Spack Project Developers. See the top-level COPYRIGHT file for details.
+#
+# SPDX-License-Identifier: (Apache-2.0 OR MIT)
+
+from spack import *
+
+
+class Batchedblas(MakefilePackage):
+    """Batched BLAS is one of the new approaches for a task-based BLAS
+    invocation, and it is defined as a new interface that allows users to
+    execute multiple independent BLAS operations as a single subroutine call"""
+
+    homepage = "https://www.r-ccs.riken.jp/labs/lpnctrt/projects/batchedblas/index.html"
+    url      = "https://www.r-ccs.riken.jp/labs/lpnctrt/projects/batchedblas/BatchedBLAS-1.0.tar.gz"
+
+    version('1.0', sha256='798ae4e7cc4ad5c3d5f3479f3d001da566d7d5205779103aaf10cd5b956ba433')
+
+    depends_on('blas')
+
+    patch('AVX2.patch')
+
+    def edit(self, spec, prefix):
+        CCFLAGS = [self.compiler.openmp_flag, '-I./', '-O3']
+        BLAS = ['-lm', spec['blas'].libs.ld_flags]
+        if not spec.satisfies('^mkl'):
+            CCFLAGS.append('-D_CBLAS_')
+        if spec.satisfies('%intel'):
+            CCFLAGS.extend(['-Os'])
+        elif spec.satisfies('%fj'):
+            CCFLAGS.extend(['-std=gnu11', '-Kfast,ocl', '-Nclang'])
+        makefile_src = FileFilter('bblas_src/Makefile')
+        makefile_src.filter(r'^\s*CCFLAG\s*=.*', 'CCFLAG = %s' % ' '.join(CCFLAGS))
+        makefile_src.filter(r'^\s*BLAS\s*=.*', 'BLAS = %s' % ' '.join(BLAS))
+
+    def install(self, spec, prefix):
+        mkdirp(prefix.lib)
+        mkdirp(prefix.include)
+        install(join_path('bblas_src', '*.h'), prefix.include)
+        install(join_path('bblas_src', 'libbblas.a'), prefix.lib)
+        install(join_path('bblas_src', 'libbblas.so'), prefix.lib)
-- 
cgit v1.2.3-70-g09d2