Martin Kroeker [Sun, 14 Jun 2020 16:27:03 +0000 (18:27 +0200)]
Merge pull request #2664 from ACSimon33/exported_symbols
Add missing exported symbols.
Martin Kroeker [Sun, 14 Jun 2020 16:26:43 +0000 (18:26 +0200)]
Merge pull request #2663 from martin-frbg/issue2654
Respect predefined defaults for AR, AS, LD and RANLIB
Martin Kroeker [Sun, 14 Jun 2020 16:25:37 +0000 (18:25 +0200)]
Merge pull request #2661 from martin-frbg/issue2660
Report selected DYNAMIC_ARCH kernel rather than one of its aliases in gotoblas_corename
Martin Kroeker [Sat, 13 Jun 2020 21:21:13 +0000 (23:21 +0200)]
Respect predefined defaults for AR, AS, LD and RANLIB
Simon Märtens [Sat, 13 Jun 2020 20:37:39 +0000 (22:37 +0200)]
Added missing exported symbols.
Martin Kroeker [Sat, 13 Jun 2020 17:25:28 +0000 (19:25 +0200)]
Make gotoblas_corename report the name of the selected TARGET rather than its aliases
Martin Kroeker [Sat, 13 Jun 2020 17:14:47 +0000 (19:14 +0200)]
Merge pull request #64 from xianyi/develop
rebase
Martin Kroeker [Wed, 10 Jun 2020 10:19:33 +0000 (12:19 +0200)]
Merge pull request #2653 from craft-zhang/cortex-a53
fix INIT8x4 of SGEMM on Arm Cortex-A53
Martin Kroeker [Tue, 9 Jun 2020 18:31:06 +0000 (20:31 +0200)]
Merge pull request #2652 from martin-frbg/flang-fixes
Fixes for compilation with flang binary release
20190329
ZhangDanfeng [Tue, 9 Jun 2020 17:01:16 +0000 (01:01 +0800)]
fix INIT8x4
Signed-off-by: ZhangDanfeng <467688405@qq.com>
Martin Kroeker [Tue, 9 Jun 2020 14:11:13 +0000 (16:11 +0200)]
Limit optimization level to O1 for flang and add -frecursive
Martin Kroeker [Tue, 9 Jun 2020 14:09:34 +0000 (16:09 +0200)]
When building with flang, use it also for the final link step to get dependencies right
Martin Kroeker [Tue, 9 Jun 2020 14:07:03 +0000 (16:07 +0200)]
Apply previously AOCC-specific workaround to all versions of flang
Martin Kroeker [Tue, 9 Jun 2020 13:54:30 +0000 (15:54 +0200)]
Merge pull request #63 from xianyi/develop
rebase
Martin Kroeker [Sun, 7 Jun 2020 17:45:52 +0000 (19:45 +0200)]
Merge pull request #2648 from martin-frbg/lapack411
Break out of potentially infinite rescaling loop in LAPACK xLARGV/xLARTG/xLARTGP
Martin Kroeker [Sun, 7 Jun 2020 17:45:11 +0000 (19:45 +0200)]
Merge pull request #2647 from martin-frbg/aocc-flang
Small fixes for flang in general and the AMD AOCC version of it in particular
Martin Kroeker [Sun, 7 Jun 2020 17:44:50 +0000 (19:44 +0200)]
Merge pull request #2645 from martin-frbg/misc_fixes
Miscellaneous fixes
Martin Kroeker [Sun, 7 Jun 2020 12:30:20 +0000 (14:30 +0200)]
Break out of potentially infinite rescaling loop in LAPACK xLARGV/xLARTG/xLARTGP
Reference-LAPACK issue 411
Martin Kroeker [Sun, 7 Jun 2020 11:18:22 +0000 (13:18 +0200)]
Merge pull request #2646 from wjc404/develop
Optimize AVX512 parallel DGEMM performance
Martin Kroeker [Sat, 6 Jun 2020 22:05:02 +0000 (00:05 +0200)]
Use only -O1 with AMD AOCC version of flang
to prevent miscompilation of LAPACK codes and tests on Ryzen
Martin Kroeker [Sat, 6 Jun 2020 13:22:27 +0000 (15:22 +0200)]
Update "cosmetic fixes for non-C99 compilers"
wjc404 [Sat, 6 Jun 2020 06:56:57 +0000 (14:56 +0800)]
Add files via upload
Martin Kroeker [Fri, 5 Jun 2020 08:05:34 +0000 (10:05 +0200)]
Cosmetic fixes for non-C99 compilers
Martin Kroeker [Fri, 5 Jun 2020 08:04:16 +0000 (10:04 +0200)]
Delete spurious copy of common_param.h
Martin Kroeker [Fri, 5 Jun 2020 08:03:36 +0000 (10:03 +0200)]
make blas_quickdivide unsigned and guard against miscompilation
Martin Kroeker [Fri, 5 Jun 2020 07:51:06 +0000 (09:51 +0200)]
Merge pull request #62 from xianyi/develop
rebase
Martin Kroeker [Fri, 5 Jun 2020 06:33:48 +0000 (08:33 +0200)]
Merge pull request #2644 from martin-frbg/cmake-maxstack
Add CMAKE support for MAX_STACK_ALLOC setting
Martin Kroeker [Thu, 4 Jun 2020 17:07:27 +0000 (19:07 +0200)]
Limit MAX_STACK_ALLOC availability to non-Wndows
Martin Kroeker [Thu, 4 Jun 2020 17:03:46 +0000 (19:03 +0200)]
Update azure-pipelines.yml
Martin Kroeker [Thu, 4 Jun 2020 15:58:13 +0000 (17:58 +0200)]
Update azure-pipelines.yml
Martin Kroeker [Thu, 4 Jun 2020 12:45:31 +0000 (14:45 +0200)]
Add CMAKE support for MAX_STACK_ALLOC setting
Martin Kroeker [Thu, 4 Jun 2020 05:58:45 +0000 (07:58 +0200)]
Merge pull request #2643 from craft-zhang/cortex-a53
Improve performance of SGEMM on Arm Cortex-A53
Martin Kroeker [Wed, 3 Jun 2020 22:02:37 +0000 (00:02 +0200)]
Merge pull request #2638 from leezu/actions
Add Github Actions test for DYNAMIC_ARCH builds on Linux and macOS
ZhangDanfeng [Wed, 3 Jun 2020 18:09:38 +0000 (02:09 +0800)]
sgemm copy source init
Signed-off-by: ZhangDanfeng <467688405@qq.com>
ZhangDanfeng [Wed, 3 Jun 2020 18:08:48 +0000 (02:08 +0800)]
Insert prefetch
Signed-off-by: ZhangDanfeng <467688405@qq.com>
Martin Kroeker [Wed, 3 Jun 2020 14:43:46 +0000 (16:43 +0200)]
Merge pull request #2641 from martin-frbg/ppcg4
Work around PPC G4 test failures
Martin Kroeker [Wed, 3 Jun 2020 13:28:59 +0000 (15:28 +0200)]
set cmake build type to Release
Martin Kroeker [Wed, 3 Jun 2020 13:16:27 +0000 (15:16 +0200)]
revert clang pragma
Martin Kroeker [Wed, 3 Jun 2020 13:15:41 +0000 (15:15 +0200)]
Update sgemm_kernel_16x4_skylakex_3.c
Martin Kroeker [Wed, 3 Jun 2020 13:07:25 +0000 (15:07 +0200)]
preset optimization level for apple clang
Martin Kroeker [Wed, 3 Jun 2020 11:46:43 +0000 (13:46 +0200)]
Update dynamic_arch.yml
Martin Kroeker [Wed, 3 Jun 2020 10:18:15 +0000 (12:18 +0200)]
Make cmake build verbose to see options in use
Martin Kroeker [Wed, 3 Jun 2020 08:22:15 +0000 (10:22 +0200)]
Revert "keep Apple Clang from optimizing this"
Martin Kroeker [Wed, 3 Jun 2020 07:15:29 +0000 (09:15 +0200)]
Change PPCG4 CGEMM_M to match kernel change
Martin Kroeker [Wed, 3 Jun 2020 07:13:29 +0000 (09:13 +0200)]
Change complex DOT and ROT to generic kernels and switch CGEMM
in response to test failures seen in #2628 and BLAS-Tester
Martin Kroeker [Wed, 3 Jun 2020 06:52:53 +0000 (08:52 +0200)]
Keep Apple Clang from optimizing this
Martin Kroeker [Tue, 2 Jun 2020 15:31:45 +0000 (17:31 +0200)]
Keep apple clang 11.0.3 from trying to optimize this (and running out of registers)
Leonard Lausen [Mon, 1 Jun 2020 20:51:11 +0000 (20:51 +0000)]
Print CPU info in output
Leonard Lausen [Sun, 31 May 2020 01:17:05 +0000 (01:17 +0000)]
Add Github Actions test for DYNAMIC_ARCH builds
Martin Kroeker [Sun, 31 May 2020 13:22:12 +0000 (15:22 +0200)]
Update f_check
Martin Kroeker [Sun, 31 May 2020 13:16:09 +0000 (15:16 +0200)]
Merge pull request #2636 from martin-frbg/issue2634
Fix CMAKE build Issues on OS X
Martin Kroeker [Sun, 31 May 2020 10:39:36 +0000 (12:39 +0200)]
Disable EXPRECISION and add -lm on OSX (same as the BSDs and Linux)
Martin Kroeker [Sun, 31 May 2020 10:37:57 +0000 (12:37 +0200)]
Document option USE_LOCKING
Martin Kroeker [Sat, 30 May 2020 12:37:12 +0000 (14:37 +0200)]
Merge pull request #2635 from ilayn/patch-1
BUG: Fix the loop range in ZHEEQUB.f
Ilhan Polat [Sat, 30 May 2020 12:11:11 +0000 (14:11 +0200)]
BUG: Fix the loop range in ZHEEQUB.f
Martin Kroeker [Fri, 29 May 2020 11:35:51 +0000 (13:35 +0200)]
Use the fortran compiler to link LAPACK-related benchmarks
to fix linking problems with (at least) the AMD version of flang that creates dependencies on more than just the fortran runtime.
Martin Kroeker [Fri, 29 May 2020 11:29:10 +0000 (13:29 +0200)]
add support for RETURN_BY_STACK semantics, e.g. clang
Martin Kroeker [Fri, 29 May 2020 11:23:51 +0000 (13:23 +0200)]
remove libomp from link list only for pgfortran
at least the AMD (aocc) flavor of flang wants to link to a (real or dummy) libomp by default
Martin Kroeker [Fri, 29 May 2020 11:06:25 +0000 (13:06 +0200)]
Merge pull request #59 from xianyi/develop
rebase
Martin Kroeker [Tue, 26 May 2020 06:36:24 +0000 (08:36 +0200)]
Merge pull request #2627 from pkubaj/patch-1
Add powerpc (32-bit)
Martin Kroeker [Tue, 26 May 2020 06:35:58 +0000 (08:35 +0200)]
Merge pull request #2626 from docularxu/working-gcc-version-detections
make GCC version detection OS-independent
pkubaj [Mon, 25 May 2020 11:14:09 +0000 (13:14 +0200)]
Add powerpc (32-bit)
Only powerpc64 is present.
Guodong Xu [Mon, 25 May 2020 10:40:12 +0000 (10:40 +0000)]
make GCC version detection OS-independent
Previous design put GCC version detection inside of OSNAME 'WINNT'.
However, such detections are required for 'Linux' and possibly other
OS'es as well. For example, there is usage of the GCC versions
in Makefile.arm64. When compiling on Linux machine, in the previous
design, Markfile.arm64 will not know the correct GCC version.
The fix is to move GCC version detection into common part, not
wrapped by anything.
Signed-off-by: Guodong Xu <guodong.xu@linaro.com>
Martin Kroeker [Mon, 25 May 2020 10:14:46 +0000 (12:14 +0200)]
Merge pull request #2618 from craft-zhang/cortex-A53
Improve performance of SGEMM and STRMM on Arm Cortex-A53
张丹枫 [Wed, 20 May 2020 14:35:26 +0000 (22:35 +0800)]
update CONTRIBUTORS.md, adding myself
张丹枫 [Wed, 20 May 2020 14:34:47 +0000 (22:34 +0800)]
split cortex-a53 param to match 8x8 kernel
张丹枫 [Wed, 20 May 2020 13:57:12 +0000 (21:57 +0800)]
update sgemm and strmm kernel selecting strategy
张丹枫 [Wed, 20 May 2020 13:55:32 +0000 (21:55 +0800)]
use general register to speedup
张丹枫 [Wed, 20 May 2020 13:52:49 +0000 (21:52 +0800)]
align general register using to strmm_kernel_8x8
zhangdanfeng [Mon, 18 May 2020 08:51:33 +0000 (16:51 +0800)]
sgemm kernel use sgemm_kernel_8x8_cortexa53
Signed-off-by: zhangdanfeng <zhangdanfeng@cloudwalk.cn>
zhangdanfeng [Mon, 18 May 2020 08:47:33 +0000 (16:47 +0800)]
optimized for cortex-a53
Signed-off-by: zhangdanfeng <zhangdanfeng@cloudwalk.cn>
Martin Kroeker [Wed, 20 May 2020 12:51:04 +0000 (14:51 +0200)]
Merge pull request #2623 from mhillenibm/zarch_dgemm_z14
s390x: Use new sgemm kernel also for DGEMM and DTRMM on Z14 (+ small cleanup)
Marius Hillenbrand [Tue, 19 May 2020 12:56:34 +0000 (14:56 +0200)]
s390x: Use new sgemm kernel also for DGEMM and DTRMM on Z14
Apply our new GEMM kernel implementation, written in C with vector intrinsics,
also for DGEMM and DTRMM on Z14 and newer (i.e., architectures with FP32 SIMD
instructions). As a result, we gain around 10% in performance on z15, in
addition to improving maintainability.
Signed-off-by: Marius Hillenbrand <mhillen@linux.ibm.com>
Marius Hillenbrand [Tue, 19 May 2020 12:30:44 +0000 (14:30 +0200)]
s390x/GEMM: replace 0-init with peeled first iteration
... since it gains another ~2% of SGEMM and DGEMM performance on z15;
also, the code just called for that cleanup.
Signed-off-by: Marius Hillenbrand <mhillen@linux.ibm.com>
Martin Kroeker [Tue, 19 May 2020 21:07:22 +0000 (23:07 +0200)]
Merge pull request #2622 from martin-frbg/issue2619
Improve declaration of LAPACKE_get_nancheck
Martin Kroeker [Tue, 19 May 2020 15:59:31 +0000 (17:59 +0200)]
Improve declaration of LAPACKE_get_nancheck
Martin Kroeker [Mon, 18 May 2020 11:23:58 +0000 (13:23 +0200)]
Merge pull request #2617 from martin-frbg/issue2616
Add workaround for unhandled gmake jobserver flags in c_check/f_check
Martin Kroeker [Sun, 17 May 2020 13:28:14 +0000 (15:28 +0200)]
Ignore spurious all-numeric library names derived from mishandled jobserver flags
Martin Kroeker [Sun, 17 May 2020 13:26:57 +0000 (15:26 +0200)]
Ignore spurious all-numeric library names derived from mishandled jobserver flags
Martin Kroeker [Sun, 17 May 2020 13:23:32 +0000 (15:23 +0200)]
Merge pull request #58 from xianyi/develop
rebase
Martin Kroeker [Thu, 14 May 2020 19:06:34 +0000 (21:06 +0200)]
Merge pull request #2615 from mhillenibm/z14_alignment_hints
s390x: improvise vector alignment hints for older compilers
Marius Hillenbrand [Wed, 13 May 2020 15:48:50 +0000 (17:48 +0200)]
s390x: improvise vector alignment hints for older compilers
Introduce inline assembly so that we can employ vector loads with
alignment hints on older compilers (pre gcc-9), since these are still
used in distributions such as RHEL 8 and Ubuntu 18.04 LTS.
Informing the hardware about alignment can speed up vector loads. For
that purpose, we can encode hints about 8-byte or 16-byte alignment of
the memory operand into the opcodes. gcc-9 and newer automatically emit
such hints, where applicable. Add a bit of inline assembly that achieves
the same for older compilers. Since an older binutils may not know about
the additional operand for the hints, we explicitly encode the opcode in
hex.
Signed-off-by: Marius Hillenbrand <mhillen@linux.ibm.com>
Martin Kroeker [Wed, 13 May 2020 13:09:23 +0000 (15:09 +0200)]
Merge pull request #2614 from mhillenibm/gemm_vec_z14
s390x: Improve performance of SGEMM and STRMM on z14 and newer
Marius Hillenbrand [Tue, 12 May 2020 14:14:00 +0000 (16:14 +0200)]
Update CONTRIBUTORS.md
Signed-off-by: Marius Hillenbrand <mhillen@linux.ibm.com>
Marius Hillenbrand [Tue, 12 May 2020 13:06:38 +0000 (15:06 +0200)]
s390x/Z14: Change register blocking for SGEMM to 16x4
Change register blocking for SGEMM (and STRMM) on z14 from 8x4 to 16x4
by adjusting SGEMM_DEFAULT_UNROLL_M and choosing the appropriate copy
implementations. Actually make KERNEL.Z14 more flexible, so that the
change in param.h suffices. As a result, performance for SGEMM improves
by around 30% on z15.
On z14, FP SIMD instructions can operate on float-sized scalars in
vector registers, while z13 could do that for double-sized scalars only.
Thus, we can double the amount of elements of C that are held in
registers in an SGEMM kernel.
Signed-off-by: Marius Hillenbrand <mhillen@linux.ibm.com>
Marius Hillenbrand [Tue, 12 May 2020 12:40:30 +0000 (14:40 +0200)]
s390x: Use new sgemm kernel also for strmm on Z14 and newer
Employ the newly added GEMM kernel also for STRMM on Z14. The
implementation in C with vector intrinsics exploits FP32 SIMD operations
and thereby gains performance over the existing assembly code. Extend
the implementation for handling triangular matrix multiplication,
accordingly. As added benefit, the more flexible C code enables us to
adjust register blocking in the subsequent commit.
Tested via make -C test / ctest / utest and by a couple of additional
unit tests that exercise blocking.
Signed-off-by: Marius Hillenbrand <mhillen@linux.ibm.com>
Marius Hillenbrand [Tue, 12 May 2020 12:13:54 +0000 (14:13 +0200)]
s390x: Add vectorized sgemm kernel for Z14 and newer
Add a new GEMM kernel implementation to exploit the FP32 SIMD
operations introduced with z14 and employ it for SGEMM on z14 and newer
architectures.
The SIMD extensions introduced with z13 support operations on
double-sized scalars in vector registers. Thus, the existing SGEMM code
would extend floats to doubles before operating on them. z14 extended
SIMD support to operations on 32-bit floats. By employing these
instructions, we can operate on twice the number of scalars per
instruction (four floats in each vector registers) and avoid the
conversion operations.
The code is written in C with explicit vectorization. In experiments,
this kernel improves performance on z14 and z15 by around 2x over the
current implementation in assembly. The flexibilty of the C code paves
the way for adjustments in subsequent commits.
Tested via make -C test / ctest / utest and by a couple of additional
unit tests that exercise blocking (e.g., partial register blocks with
fewer than UNROLL_M rows and/or fewer than UNROLL_N columns).
Signed-off-by: Marius Hillenbrand <mhillen@linux.ibm.com>
Marius Hillenbrand [Tue, 12 May 2020 09:09:28 +0000 (11:09 +0200)]
Update CONTRIBUTORS.md, adding myself
Signed-off-by: Marius Hillenbrand <mhillen@linux.ibm.com>
Marius Hillenbrand [Mon, 11 May 2020 11:00:10 +0000 (13:00 +0200)]
s390x: choose SIMD kernels at run-time based on OS and compiler support
Extend and simplify the run-time detection for dynamic architecture support for z
to check HW_CAP and only use SIMD features if advertised by the OS.
While at it, also honor the env variable LD_HWCAP_MASK and do not use
the CPU features masked there.
Note that we can only use the SIMD features on z13 or newer (i.e.,
Vector Facility or Vector-Enhancements Facilities) when the operating
system supports properly context-switching the vector registers. The OS
advertises that support as a bit in the HW_CAP value in the auxiliary
vector. While all recent Linux kernels have that support, we should
maintain compatibility with older versions that may still be in use.
Signed-off-by: Marius Hillenbrand <mhillen@linux.ibm.com>
Marius Hillenbrand [Mon, 11 May 2020 16:37:04 +0000 (18:37 +0200)]
s390x: only build kernels supported by gcc with dynamic arch support
When building with dynamic arch support, only build kernels for
architectures that are supported by the gcc we are building with.
Signed-off-by: Marius Hillenbrand <mhillen@linux.ibm.com>
Marius Hillenbrand [Mon, 11 May 2020 10:37:21 +0000 (12:37 +0200)]
s390x: gate dynamic arch detection on gcc version and add generic
When building OpenBLAS with DYNAMIC_ARCH=1 on s390x (aka zarch), make
sure to include support for systems without the facilities introduced
with z13 (i.e., zarch_generic). Adjust runtime detection to fallback to
that generic code when running on a unknown platform other than Z13
through Z15.
When detecting a Z13 or newer system, add a check for gcc support for
the architecture-specific features before selecting the respective
kernel. Fallback to Z13 or generic code, in case.
Signed-off-by: Marius Hillenbrand <mhillen@linux.ibm.com>
Martin Kroeker [Tue, 12 May 2020 06:34:02 +0000 (08:34 +0200)]
Merge pull request #2612 from RajalakshmiSR/testshgemm
Improve shgemm test
Rajalakshmi Srinivasaraghavan [Mon, 11 May 2020 22:15:10 +0000 (17:15 -0500)]
Improve shgemm test
This patch adds another check to test shgemm results.
Martin Kroeker [Mon, 11 May 2020 19:08:41 +0000 (21:08 +0200)]
Merge pull request #2611 from RajalakshmiSR/bench_half
Include shgemm in benchtest
Rajalakshmi Srinivasaraghavan [Mon, 11 May 2020 14:57:46 +0000 (09:57 -0500)]
Include shgemm in benchtest
This patch is to enable benchtest for half precision gemm
when BUILD_HALF is set during make.
Martin Kroeker [Sun, 10 May 2020 11:10:31 +0000 (13:10 +0200)]
Merge pull request #2610 from martin-frbg/issue2552-3
Temporary workaround for excessive LAPACK test failures with COMPLEX on Skylake-X
Martin Kroeker [Sat, 9 May 2020 21:49:18 +0000 (23:49 +0200)]
Work around excessive LAPACK test failures on Skylake-X
Something in the plain C parts of x86_64 cscal.c and zscal.c appears to be miscompiled by both gfortran9 and ifort when compiling for skylakex-avx512, even when the optimized Haswell microkernel is not in use.
Martin Kroeker [Sat, 9 May 2020 19:33:02 +0000 (21:33 +0200)]
Merge pull request #2609 from martin-frbg/issue2552-2
Correct ifort options
Martin Kroeker [Sat, 9 May 2020 15:15:36 +0000 (17:15 +0200)]
Correct ifort options
to same as suggested by reference-lapack
Martin Kroeker [Sat, 9 May 2020 14:36:14 +0000 (16:36 +0200)]
Merge pull request #2608 from martin-frbg/issue2604
Handle trailing whitespace and empty variables in KERNEL files