Skip to contentSkip to conversation

openblas: new package

PATCH Superseded Vicente Olivert Riera 5 participants 12 messages May 19, 2016 – May 23, 2016

v3 of 8 Superseded Latest is v8 →

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 openblas: new package

---
Changes v2 -> v3:
- Switch version to HEAD of develop branch as it contains support for MIPS32.
- Add a patch to fix build failure of P5600 and I6400 targets.
- Add support for OpenMP when toolchain has NPTL threads.

Changes v1 -> v2:
- Multi-threading is not available for static-only since it uses dlfcn.h, so modify the "ifeq" statement accordingly.

 package/Config.in              |   1 +
 package/openblas/Config.in     | 240 +++++++++++++++++++++++++++++++++++++++++
 package/openblas/openblas.hash |   3 +
 package/openblas/openblas.mk   |  54 ++++++++++
 4 files changed, 298 insertions(+)
 create mode 100644 package/openblas/Config.in
 create mode 100644 package/openblas/openblas.hash
 create mode 100644 package/openblas/openblas.mk
package/Config.in +1 -0
index 0056beb..b59c2f1 100644
@@ -1270,6 +1270,7 @@ endif
 	source "package/msgpack/Config.in"
 	source "package/mtdev2tuio/Config.in"
 	source "package/netbsd-queue/Config.in"
+	source "package/openblas/Config.in"
 	source "package/orc/Config.in"
 	source "package/p11-kit/Config.in"
 	source "package/poco/Config.in"
package/openblas/Config.in new file +240 -0

240 lines not on this page · raw message

package/openblas/openblas.hash new file +3 -0
new file mode 100644
index 0000000..c537f8f
@@ -0,0 +1,3 @@
+# Locally calculated
+sha256 745650877bc7550f52eaf1f07bc691b9fd645f0d33eaacfc7c760a3e2a9f87dd  openblas-7a190653698ecd6576653109adbae5b805278c7e.tar.gz
+sha256 c0aa38db2eee30d2c23735b55b814f77cf0d39c12c50dfe617b8ebf941573ea5  e12cff87b86615f5a4643d246a6c1963a0e81ca5.patch
package/openblas/openblas.mk new file +54 -0

54 lines not on this page · raw message

Signature · 1 line
2.7.3
#1 lore raw

PATCH v3 1/1 openblas: new package

Hi Vicente,

Quoted 6 lines · Vicente Olivert Riera

---
Changes v2 -> v3:
- Switch version to HEAD of develop branch as it contains support for MIPS32.
- Add a patch to fix build failure of P5600 and I6400 targets.

Where is this patch?

Quoted 6 lines

- Add support for OpenMP when toolchain has NPTL threads.

Changes v1 -> v2:
- Multi-threading is not available for static-only since it uses dlfcn.h, so modify the "ifeq" statement accordingly.

Quoted 8 lines
 package/Config.in              |   1 +
 package/openblas/Config.in     | 240 +++++++++++++++++++++++++++++++++++++++++
 package/openblas/openblas.hash |   3 +
 package/openblas/openblas.mk   |  54 ++++++++++
 4 files changed, 298 insertions(+)
 create mode 100644 package/openblas/Config.in
 create mode 100644 package/openblas/openblas.hash
 create mode 100644 package/openblas/openblas.mk

baruch

Signature · 3 lines
     http://baruch.siach.name/blog/                  ~. .~   Tk Open Systems
=}------------------------------------------------ooO--U--Ooo------------{=
   - baruch at tkos.co.il - tel: +972.2.679.5364, http://www.tkos.co.il -
#2 lore raw

PATCH v3 1/1 openblas: new package

Hello Baruch,

Quoted 11 lines · Baruch Siach

Hi Vicente,

Quoted 6 lines · Vicente Olivert Riera

---
Changes v2 -> v3:
- Switch version to HEAD of develop branch as it contains support for MIPS32.
- Add a patch to fix build failure of P5600 and I6400 targets.

Where is this patch?

referenced in the openblas.mk file, in OPENBLAS_PATCH variable:

OPENBLAS_PATCH =
https://github.com/vriera/OpenBLAS/commit/e12cff87b86615f5a4643d246a6c1963a0e81ca5.patch

Regards,

Vincent.

Quoted 17 lines
Quoted 6 lines

- Add support for OpenMP when toolchain has NPTL threads.

Changes v1 -> v2:
- Multi-threading is not available for static-only since it uses dlfcn.h, so modify the "ifeq" statement accordingly.

Quoted 8 lines
 package/Config.in              |   1 +
 package/openblas/Config.in     | 240 +++++++++++++++++++++++++++++++++++++++++
 package/openblas/openblas.hash |   3 +
 package/openblas/openblas.mk   |  54 ++++++++++
 4 files changed, 298 insertions(+)
 create mode 100644 package/openblas/Config.in
 create mode 100644 package/openblas/openblas.hash
 create mode 100644 package/openblas/openblas.mk

baruch

#3 lore raw

PATCH v3 1/1 openblas: new package

Hi Vicente,

Quoted 9 lines · Vicente Olivert Riera
Quoted 8 lines · Baruch Siach
Quoted 6 lines · Vicente Olivert Riera

---
Changes v2 -> v3:
- Switch version to HEAD of develop branch as it contains support for MIPS32.
- Add a patch to fix build failure of P5600 and I6400 targets.

Quoted 6 lines
Quoted 1 line

Where is this patch?

referenced in the openblas.mk file, in OPENBLAS_PATCH variable:

OPENBLAS_PATCH =
https://github.com/vriera/OpenBLAS/commit/e12cff87b86615f5a4643d246a6c1963a0e81ca5.patch

Right, I missed that. Please add a link to your pull request
(https://github.com/xianyi/OpenBLAS/pull/886) in a comment.

Thanks,
baruch

Signature · 3 lines
     http://baruch.siach.name/blog/                  ~. .~   Tk Open Systems
=}------------------------------------------------ooO--U--Ooo------------{=
   - baruch at tkos.co.il - tel: +972.2.679.5364, http://www.tkos.co.il -
#4 lore raw

PATCH v3 1/1 openblas: new package

In reply to Vicente Olivert Riera

Hello,

Quoted 154 lines · Vicente Olivert Riera
package/openblas/Config.in new file +154 -0
+choice
+	prompt "OpenBLAS target CPU"
+	help
+	  OpenBLAS target CPU
+
+config BR2_PACKAGE_OPENBLAS_TARGET_P2
+	bool "P2"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_KATMAI
+	bool "KATMAI"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_COPPERMINE
+	bool "COPPERMINE"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_NORTHWOOD
+	bool "NORTHWOOD"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_PRESCOTT
+	bool "PRESCOTT"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_BANIAS
+	bool "BANIAS"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_YONAH
+	bool "YONAH"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_CORE2
+	bool "CORE2"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_PENRYN
+	bool "PENRYN"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_DUNNINGTON
+	bool "DUNNINGTON"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_NEHALEM
+	bool "NEHALEM"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_SANDYBRIDGE
+	bool "SANDYBRIDGE"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_HASWELL
+	bool "HASWELL"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_ATOM
+	bool "ATOM"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_ATHLON
+	bool "ATHLON"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_OPTERON
+	bool "OPTERON"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_OPTERON_SSE3
+	bool "OPTERON_SSE3"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_BARCELONA
+	bool "BARCELONA"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_SHANGHAI
+	bool "SHANGHAI"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_ISTANBUL
+	bool "ISTANBUL"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_BOBCAT
+	bool "BOBCAT"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_BULLDOZER
+	bool "BULLDOZER"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_PILEDRIVER
+	bool "PILEDRIVER"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_STEAMROLLER
+	bool "STEAMROLLER"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_EXCAVATOR
+	bool "EXCAVATOR"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_SSE_GENERIC
+	bool "SSE_GENERIC"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_VIAC3
+	bool "VIAC3"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_NANO
+	bool "NANO"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_POWER4
+	bool "POWER4"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_POWER5
+	bool "POWER5"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_POWER6
+	bool "POWER6"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_POWER7
+	bool "POWER7"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_POWER8
+	bool "POWER8"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_PPCG4
+	bool "PPCG4"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_PPC970
+	bool "PPC970"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_PPC970MP
+	bool "PPC970MP"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_PPC440
+	bool "PPC440"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_PPC440FP2
+	bool "PPC440FP2"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_CELL
+	bool "CELL"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_P5600
+	bool "P5600"
+	depends on BR2_mips || BR2_mipsel
+config BR2_PACKAGE_OPENBLAS_TARGET_SICORTEX
+	bool "SICORTEX"
+	depends on BR2_mips64 || BR2_mips64el
+config BR2_PACKAGE_OPENBLAS_TARGET_LOONGSON3A
+	bool "LOONGSON3A"
+	depends on BR2_mips64 || BR2_mips64el
+config BR2_PACKAGE_OPENBLAS_TARGET_LOONGSON3B
+	bool "LOONGSON3B"
+	depends on BR2_mips64 || BR2_mips64el
+config BR2_PACKAGE_OPENBLAS_TARGET_I6400
+	bool "I6400"
+	depends on BR2_mips64 || BR2_mips64el
+config BR2_PACKAGE_OPENBLAS_TARGET_SPARC
+	bool "SPARC"
+	depends on BR2_sparc || BR2_sparc64
+config BR2_PACKAGE_OPENBLAS_TARGET_SPARCV7
+	bool "SPARCV7"
+	depends on BR2_sparc || BR2_sparc64
+config BR2_PACKAGE_OPENBLAS_TARGET_CORTEXA15
+	bool "CORTEXA15"
+	depends on BR2_arm || BR2_armeb
+config BR2_PACKAGE_OPENBLAS_TARGET_CORTEXA9
+	bool "CORTEXA9"
+	depends on BR2_arm || BR2_armeb
+config BR2_PACKAGE_OPENBLAS_TARGET_ARMV7
+	bool "ARMV7"
+	depends on BR2_arm || BR2_armeb
+config BR2_PACKAGE_OPENBLAS_TARGET_ARMV6
+	bool "ARMV6"

package/openblas/Config.in line 180

Quoted 10 lines
package/openblas/Config.in new file +10 -0
+	depends on BR2_arm || BR2_armeb
+config BR2_PACKAGE_OPENBLAS_TARGET_ARMV5
+	bool "ARMV5"
+	depends on BR2_arm || BR2_armeb
+config BR2_PACKAGE_OPENBLAS_TARGET_ARMV8
+	bool "ARMV8"
+	depends on BR2_aarch64 || BR2_aarch64_be
+config BR2_PACKAGE_OPENBLAS_TARGET_CORTEXA57
+	bool "CORTEXA57"
+	depends on BR2_aarch64 || BR2_aarch64_be

Do we really need this choice? What about directly using the architecture symbols?

I.e, replace what you have below by:

config BR2_PACKAGE_OPENBLAS_TARGET

	string
	default "ATOM"	if BR2_x86_atom
	default "CORE2"	if BR2_x86_core2
	...
	default "CORTEXA15" if BR2_cortex_a15
	...
Quoted 7 lines

new file mode 100644
index 0000000..5e6f417

package/openblas/openblas.mk new file +2 -0
@@ -0,0 +1,54 @@
+################################################################################
+#

package/openblas/openblas.mk line 12

Quoted 10 lines
package/openblas/openblas.mk new file +10 -0
+# openblas
+#
+################################################################################
+
+OPENBLAS_VERSION = 7a190653698ecd6576653109adbae5b805278c7e
+OPENBLAS_SITE = $(call github,xianyi,OpenBLAS,$(OPENBLAS_VERSION))
+OPENBLAS_LICENSE = BSD-3
+OPENBLAS_LICENSE_FILES = LICENSE
+OPENBLAS_INSTALL_STAGING = YES
+OPENBLAS_PATCH = https://github.com/vriera/OpenBLAS/commit/e12cff87b86615f5a4643d246a6c1963a0e81ca5.patch

I think that for such small patches, that are not taken from upstream, my preference is to have them in Buildroot itself rather than downloaded by <pkg>_PATCH.

package/openblas/openblas.mk line 15

Quoted 3 lines
package/openblas/openblas.mk new file +3 -0
+
+# Disable fortran if the fortran compiler doesn't actually exist.
+ifeq ($(wildcard $(TARGET_FC)),)

I don't really like this way of testing if we have Fortran support. We probably want some kind of BR2_TOOLCHAIN_HAS_FORTRAN hidden config option. Talk with Samuel, I think he had some patches that were adding this kind of stuff for the external toolchains.

package/openblas/openblas.mk line 24

Quoted 9 lines
package/openblas/openblas.mk new file +9 -0
+OPENBLAS_MAKE_OPTS += ONLY_CBLAS=1
+endif
+
+# Enable/Disable multi-threading (not for static-only since it uses dlfcn.h)
+ifeq ($(BR2_TOOLCHAIN_HAS_THREADS)x$(BR2_STATIC_LIBS),yx)
+OPENBLAS_MAKE_OPTS += USE_THREAD=1
+# Take advantage of OpenMP if we have support for it
+ifeq ($(BR2_TOOLCHAIN_HAS_THREADS_NPTL),y)
+OPENBLAS_MAKE_OPTS += USE_OPENMP=1

It seems weird that OpenMP is related to NPTL thread support. OpenMP is a separate feature of the toolchain, see the option BR2_GCC_ENABLE_OPENMP for internal toolchains.

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
#5 lore raw

PATCH v3 1/1 openblas: new package

Hello Thomas,

Quoted 174 lines · Thomas Petazzoni

Hello,

Quoted 154 lines · Vicente Olivert Riera
package/openblas/Config.in new file +154 -0
+choice
+	prompt "OpenBLAS target CPU"
+	help
+	  OpenBLAS target CPU
+
+config BR2_PACKAGE_OPENBLAS_TARGET_P2
+	bool "P2"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_KATMAI
+	bool "KATMAI"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_COPPERMINE
+	bool "COPPERMINE"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_NORTHWOOD
+	bool "NORTHWOOD"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_PRESCOTT
+	bool "PRESCOTT"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_BANIAS
+	bool "BANIAS"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_YONAH
+	bool "YONAH"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_CORE2
+	bool "CORE2"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_PENRYN
+	bool "PENRYN"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_DUNNINGTON
+	bool "DUNNINGTON"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_NEHALEM
+	bool "NEHALEM"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_SANDYBRIDGE
+	bool "SANDYBRIDGE"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_HASWELL
+	bool "HASWELL"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_ATOM
+	bool "ATOM"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_ATHLON
+	bool "ATHLON"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_OPTERON
+	bool "OPTERON"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_OPTERON_SSE3
+	bool "OPTERON_SSE3"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_BARCELONA
+	bool "BARCELONA"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_SHANGHAI
+	bool "SHANGHAI"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_ISTANBUL
+	bool "ISTANBUL"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_BOBCAT
+	bool "BOBCAT"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_BULLDOZER
+	bool "BULLDOZER"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_PILEDRIVER
+	bool "PILEDRIVER"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_STEAMROLLER
+	bool "STEAMROLLER"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_EXCAVATOR
+	bool "EXCAVATOR"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_SSE_GENERIC
+	bool "SSE_GENERIC"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_VIAC3
+	bool "VIAC3"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_NANO
+	bool "NANO"
+	depends on BR2_i386 || BR2_x86_64
+config BR2_PACKAGE_OPENBLAS_TARGET_POWER4
+	bool "POWER4"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_POWER5
+	bool "POWER5"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_POWER6
+	bool "POWER6"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_POWER7
+	bool "POWER7"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_POWER8
+	bool "POWER8"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_PPCG4
+	bool "PPCG4"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_PPC970
+	bool "PPC970"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_PPC970MP
+	bool "PPC970MP"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_PPC440
+	bool "PPC440"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_PPC440FP2
+	bool "PPC440FP2"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_CELL
+	bool "CELL"
+	depends on BR2_powerpc || BR2_powerpc64 || BR2_powerpc64le
+config BR2_PACKAGE_OPENBLAS_TARGET_P5600
+	bool "P5600"
+	depends on BR2_mips || BR2_mipsel
+config BR2_PACKAGE_OPENBLAS_TARGET_SICORTEX
+	bool "SICORTEX"
+	depends on BR2_mips64 || BR2_mips64el
+config BR2_PACKAGE_OPENBLAS_TARGET_LOONGSON3A
+	bool "LOONGSON3A"
+	depends on BR2_mips64 || BR2_mips64el
+config BR2_PACKAGE_OPENBLAS_TARGET_LOONGSON3B
+	bool "LOONGSON3B"
+	depends on BR2_mips64 || BR2_mips64el
+config BR2_PACKAGE_OPENBLAS_TARGET_I6400
+	bool "I6400"
+	depends on BR2_mips64 || BR2_mips64el
+config BR2_PACKAGE_OPENBLAS_TARGET_SPARC
+	bool "SPARC"
+	depends on BR2_sparc || BR2_sparc64
+config BR2_PACKAGE_OPENBLAS_TARGET_SPARCV7
+	bool "SPARCV7"
+	depends on BR2_sparc || BR2_sparc64
+config BR2_PACKAGE_OPENBLAS_TARGET_CORTEXA15
+	bool "CORTEXA15"
+	depends on BR2_arm || BR2_armeb
+config BR2_PACKAGE_OPENBLAS_TARGET_CORTEXA9
+	bool "CORTEXA9"
+	depends on BR2_arm || BR2_armeb
+config BR2_PACKAGE_OPENBLAS_TARGET_ARMV7
+	bool "ARMV7"
+	depends on BR2_arm || BR2_armeb
+config BR2_PACKAGE_OPENBLAS_TARGET_ARMV6
+	bool "ARMV6"

package/openblas/Config.in line 180

Quoted 10 lines
package/openblas/Config.in new file +10 -0
+	depends on BR2_arm || BR2_armeb
+config BR2_PACKAGE_OPENBLAS_TARGET_ARMV5
+	bool "ARMV5"
+	depends on BR2_arm || BR2_armeb
+config BR2_PACKAGE_OPENBLAS_TARGET_ARMV8
+	bool "ARMV8"
+	depends on BR2_aarch64 || BR2_aarch64_be
+config BR2_PACKAGE_OPENBLAS_TARGET_CORTEXA57
+	bool "CORTEXA57"
+	depends on BR2_aarch64 || BR2_aarch64_be

Do we really need this choice? What about directly using the architecture symbols?

I.e, replace what you have below by:

Quoted 7 lines

config BR2_PACKAGE_OPENBLAS_TARGET

	string
	default "ATOM"	if BR2_x86_atom
	default "CORE2"	if BR2_x86_core2
	...
	default "CORTEXA15" if BR2_cortex_a15
	...

No, we cannot do that because there isn't only one OpenBLAS target per BR arch symbol. For instance, for mips64 there are 3 OpenBLAS targets available. For x86 there are lots of them. Same for x86_64. Plust the user may want to build a generic buildroot (because it doesn't have specific support for certain core) and build OpenBLAS for that core.

Quoted 13 lines
Quoted 13 lines

new file mode 100644
index 0000000..5e6f417

package/openblas/openblas.mk new file +8 -0
@@ -0,0 +1,54 @@
+################################################################################
+#
+# openblas
+#
+################################################################################
+
+OPENBLAS_VERSION = 7a190653698ecd6576653109adbae5b805278c7e
+OPENBLAS_SITE = $(call github,xianyi,OpenBLAS,$(OPENBLAS_VERSION))
Quoted 8 lines

package/openblas/openblas.mk line 12

Quoted 4 lines
package/openblas/openblas.mk new file +4 -0
+OPENBLAS_LICENSE = BSD-3
+OPENBLAS_LICENSE_FILES = LICENSE
+OPENBLAS_INSTALL_STAGING = YES
+OPENBLAS_PATCH = https://github.com/vriera/OpenBLAS/commit/e12cff87b86615f5a4643d246a6c1963a0e81ca5.patch

I think that for such small patches, that are not taken from upstream, my preference is to have them in Buildroot itself rather than downloaded by <pkg>_PATCH.

I remember someone told me that we prefer this method when possible... Anyway my pull request has been merged so if I have to respin this patch I will just bump the version and get rid of the patch.

Quoted 8 lines

package/openblas/openblas.mk line 15

Quoted 3 lines
package/openblas/openblas.mk new file +3 -0
+
+# Disable fortran if the fortran compiler doesn't actually exist.
+ifeq ($(wildcard $(TARGET_FC)),)

I don't really like this way of testing if we have Fortran support. We probably want some kind of BR2_TOOLCHAIN_HAS_FORTRAN hidden config option. Talk with Samuel, I think he had some patches that were adding this kind of stuff for the external toolchains.

Yeah, that would be great, but we don't have it right now. I wrote a patch for SSP toolchain support and it's taking ages to get applied. I don't know if I want to wait for another patch for the BR2_TOOLCHAIN_HAS_FORTRAN stuff to get applied before adding this package to Buildroot. Why don't do that in the future? Add the package now. When we have that toolchain-has-fortran thing available, then we modify the package to use it.

Quoted 5 lines
Quoted 5 lines
package/openblas/openblas.mk new file +5 -0
+OPENBLAS_MAKE_OPTS += ONLY_CBLAS=1
+endif
+
+# Enable/Disable multi-threading (not for static-only since it uses dlfcn.h)
+ifeq ($(BR2_TOOLCHAIN_HAS_THREADS)x$(BR2_STATIC_LIBS),yx)
Quoted 8 lines

package/openblas/openblas.mk line 24

Quoted 4 lines
package/openblas/openblas.mk new file +4 -0
+OPENBLAS_MAKE_OPTS += USE_THREAD=1
+# Take advantage of OpenMP if we have support for it
+ifeq ($(BR2_TOOLCHAIN_HAS_THREADS_NPTL),y)
+OPENBLAS_MAKE_OPTS += USE_OPENMP=1

It seems weird that OpenMP is related to NPTL thread support. OpenMP is a separate feature of the toolchain, see the option BR2_GCC_ENABLE_OPENMP for internal toolchains.

Well, I've read here [1] that "OpenMP is available for most platforms that support POSIX threads".

1: https://wiki.linaro.org/WorkingGroups/ToolChain/OptimizationDescriptions

Regards,

Vincent.

Quoted 2 lines

Thomas

#6 lore raw

PATCH v3 1/1 openblas: new package

Hello,

Quoted 8 lines · Vicente Olivert Riera
Quoted 7 lines

config BR2_PACKAGE_OPENBLAS_TARGET

	string
	default "ATOM"	if BR2_x86_atom
	default "CORE2"	if BR2_x86_core2
	...
	default "CORTEXA15" if BR2_cortex_a15
	...  
Quoted 5 lines

No, we cannot do that because there isn't only one OpenBLAS target per BR arch symbol. For instance, for mips64 there are 3 OpenBLAS targets available. For x86 there are lots of them. Same for x86_64. Plust the user may want to build a generic buildroot (because it doesn't have specific support for certain core) and build OpenBLAS for that core.

For x86 and x86_64, we have sub-options that specify which CPU is being used. Look at my example above, which shows that ATOM and CORE2 can be found by looking at the BR2_x86_<foo> options.

Regarding MIPS64, what are the practical differences between the following options?

+config BR2_PACKAGE_OPENBLAS_TARGET_SICORTEX
+ bool "SICORTEX"
+ depends on BR2_mips64 || BR2_mips64el
+config BR2_PACKAGE_OPENBLAS_TARGET_LOONGSON3A
+ bool "LOONGSON3A"
+ depends on BR2_mips64 || BR2_mips64el
+config BR2_PACKAGE_OPENBLAS_TARGET_LOONGSON3B
+ bool "LOONGSON3B"
+ depends on BR2_mips64 || BR2_mips64el
+config BR2_PACKAGE_OPENBLAS_TARGET_I6400
+ bool "I6400"
+ depends on BR2_mips64 || BR2_mips64el

If there are some practical differences in terms of gcc options being used, then it probably means that we need to add additional sub-architectures in arch/Config.in.mips.

Quoted 5 lines
Quoted 3 lines

I think that for such small patches, that are not taken from upstream, my preference is to have them in Buildroot itself rather than downloaded by <pkg>_PATCH.

I remember someone told me that we prefer this method when possible...

Yeah, we don't have a very fixed policy on this. My own thinking is that it's fine to use <pkg>_PATCH to fetch patches that are already otherwise available (from another distribution or from upstream), but when it's done by "us", it's better to have the patch in Buildroot itself. But I agree that the line between both solutions is thin.

Quoted 2 lines

Anyway my pull request has been merged so if I have to respin this patch I will just bump the version and get rid of the patch.

Good!

Quoted 8 lines
Quoted 7 lines

package/openblas/openblas.mk line 15

Quoted 2 lines
package/openblas/openblas.mk new file +2 -0
+# Disable fortran if the fortran compiler doesn't actually exist.
+ifeq ($(wildcard $(TARGET_FC)),)  

I don't really like this way of testing if we have Fortran support. We probably want some kind of BR2_TOOLCHAIN_HAS_FORTRAN hidden config option. Talk with Samuel, I think he had some patches that were adding this kind of stuff for the external toolchains.

Quoted 7 lines

Yeah, that would be great, but we don't have it right now. I wrote a patch for SSP toolchain support and it's taking ages to get applied. I don't know if I want to wait for another patch for the BR2_TOOLCHAIN_HAS_FORTRAN stuff to get applied before adding this package to Buildroot. Why don't do that in the future? Add the package now. When we have that toolchain-has-fortran thing available, then we modify the package to use it.

So, for the moment, just disable Fortran support unconditionally, and in a follow-up patch you can enable Fortran support.

Quoted 6 lines
Quoted 3 lines

It seems weird that OpenMP is related to NPTL thread support. OpenMP is a separate feature of the toolchain, see the option BR2_GCC_ENABLE_OPENMP for internal toolchains.

Well, I've read here [1] that "OpenMP is available for most platforms that support POSIX threads".

If the support has been enabled in the compiler, which is what I'm talking about. Have you tried your packages with a Buildroot internal toolchain, with OpenMP support disabled?

Best regards,

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
#7 lore raw

PATCH v3 1/1 openblas: new package

Quoted 3 lines · Thomas Petazzoni

Hello,

On Thu, 19 May 2016 14:42:46 +0100, Vicente Olivert Riera wrote:

[snip]

Quoted 11 lines
Quoted 5 lines
Quoted 5 lines
Quoted 3 lines
Quoted 3 lines

I think that for such small patches, that are not taken from upstream, my preference is to have them in Buildroot itself rather than downloaded by <pkg>_PATCH.

I remember someone told me that we prefer this method when possible...

Yeah, we don't have a very fixed policy on this. My own thinking is that it's fine to use <pkg>_PATCH to fetch patches that are already otherwise available (from another distribution or from upstream), but when it's done by "us", it's better to have the patch in Buildroot itself. But I agree that the line between both solutions is thin.

I indeed once wrote "we prefer to download patches through FOO_PATCH rather
than including it". I think in that particular case it was a patch from a different distro. Anyway, I didn't really think it through at the time. I certainly agree that it should only be for upstream patches, and maybe also patches from different distros.

Regards,
Arnout

Signature · 6 lines
Arnout Vandecappelle                          arnout at mind be
Senior Embedded Software Architect            +32-16-286500
Essensium/Mind                                http://www.mind.be
G.Geenslaan 9, 3001 Leuven, Belgium           BE 872 984 063 RPR Leuven
LinkedIn profile: http://www.linkedin.com/in/arnoutvandecappelle
GPG fingerprint:  7493 020B C7E3 8618 8DEC 222C 82EB F404 F9AC 0DDF
#8 lore raw

PATCH v3 1/1 openblas: new package

In reply to Thomas Petazzoni

Hello Samuel,

Quoted 7 lines · Thomas Petazzoni
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 2 lines

package/openblas/openblas.mk line 15

Quoted 2 lines
package/openblas/openblas.mk new file +2 -0
+# Disable fortran if the fortran compiler doesn't actually exist.
+ifeq ($(wildcard $(TARGET_FC)),)  

I don't really like this way of testing if we have Fortran support. We probably want some kind of BR2_TOOLCHAIN_HAS_FORTRAN hidden config option. Talk with Samuel, I think he had some patches that were adding this kind of stuff for the external toolchains.

do you have those patches Thomas is talking about? If so, do you plan to send them to the ML?

Regards,

Vincent.

#10 lore raw

PATCH v3 1/1 openblas: new package

Hi Vicente, all,

Quoted 8 lines · Thomas Petazzoni

Hello,

Quoted 2 lines · Vicente Olivert Riera

do you have those patches Thomas is talking about? If so, do you plan to send them to the ML?

See: https://github.com/tSed/buildroot/commits/sma/ext-toolchain-options

Yes, I will send them at some point ;-)
In its current state the branch should be easily rebased on master I think. But there is some part of this work that could benefit from the shell module work from this series [1], so I'm waiting for this series [1] to be merged before.
I plan to poke people about [1] as soon as the 2016.05 release is out, and continue works based on it.

[1] http://patchwork.ozlabs.org/bundle/smartin/rpath_v9_part01/

Regards,

Signature · 1 line
Samuel
#11 lore raw

PATCH v3 1/1 openblas: new package

Thanks Samuel.

Regards,

Vincent.

Quoted 26 lines · Samuel Martin

Hi Vicente, all,

Quoted 8 lines · Thomas Petazzoni

Hello,

Quoted 2 lines · Vicente Olivert Riera

do you have those patches Thomas is talking about? If so, do you plan to send them to the ML?

See: https://github.com/tSed/buildroot/commits/sma/ext-toolchain-options

Yes, I will send them at some point ;-)
In its current state the branch should be easily rebased on master I think. But there is some part of this work that could benefit from the shell module work from this series [1], so I'm waiting for this series [1] to be merged before.
I plan to poke people about [1] as soon as the 2016.05 release is out, and continue works based on it.

[1] http://patchwork.ozlabs.org/bundle/smartin/rpath_v9_part01/

Regards,