OSchip
/
OpenBLAS
mirror of https://github.com/xianyi/OpenBLAS.git
0
0
Fork 2
Code Issues Packages Projects Releases Wiki Activity
9,676 Commits 58 Branches 70 Tags 121 MiB
Commit Graph

378 Commits

Author SHA1 Message Date
Martin Kroeker 9bfc3612f9
Merge branch 'OpenMathLib:develop' into issue5414 2025-10-12 09:18:06 -07:00
Yichao Yu b94e9b92ad Fix compilation on ARM 
Define a dummy function if SME is not supported, following what sgemm does
 2025-10-11 20:28:59 -04:00
Martin Kroeker e40714cabd
Merge pull request #5450 from quic/topic/strmm_direct_sme1 
Support for SME1 based strmm_direct kernel for cblas_strmm level 3 API
 2025-10-11 15:20:19 -07:00
changjua 644ea07ef9 Support for SME1 based strmm_direct kernel for cblas_strmm level 3 API 2025-10-10 10:48:27 +08:00
Martin Kroeker 47a66aef0f
Update limits based on benchmarking the SME code on Apple M4 2025-10-08 14:36:17 +02:00
Martin Kroeker 20f5ed1a94
Merge branch 'OpenMathLib:develop' into issue5414 2025-10-08 05:27:28 -07:00
Chris Sidebottom 578e7dae85 Fix bf16->f32 conversion for NEOVERSEV1 and NEOVERSEN2 targets 
This fixes an issue originally introduced with the BGEMM kernel.

I've updated the tests to run with `beta=1.0` so as to test loading and
updating from C.

Alongside this, the tests now return sensible return values to reduce
the risk of them being ignored.

Also fixed a bug in `generic/gemv_t.c` resulting in weird outputs for
`bgemv`.
 2025-10-06 18:05:58 +00:00
Martin Kroeker c889558317
Rework for DYNAMIC_ARCH use and use of SGEMM functions by SSYMM 2025-10-02 07:39:24 -07:00
Martin Kroeker fc516af155
Merge branch 'develop' into issue5414 2025-10-01 14:12:59 -07:00
Rajendra Prasad Matcha 19268471cc Support for SME1 based ssymm_direct kernel for cblas_ssymm level 3 API 2025-09-30 15:05:33 +05:30
Martin Kroeker 2b5d8c789d
remove debugging printout 2025-08-24 13:50:08 -07:00
Martin Kroeker b4fc09e9e1
Add registers d8 to d15 to clobber lists as the code does not expressly save them 2025-08-23 14:39:27 -07:00
Martin Kroeker 8e50b8d525
Add d8 to d15 to clobber lists as the code does not expressly save them 2025-08-23 14:36:49 -07:00
h-motoki 855945befb Implementing SVE in [SD]AXPY Kernels for A64FX and Graviton3E 2025-08-21 20:56:58 +09:00
Martin Kroeker edaa73fd24
Hide the local 2VLx2VL symbol as static is insufficient for this with gcc 2025-08-20 06:33:28 -07:00
Martin Kroeker 501728a354
adjust register 20 accesses to 21 after moving x18 2025-08-20 06:24:38 -07:00
Martin Kroeker 05dbb54362
Delete misplaced file 2025-08-19 05:12:09 -07:00
Martin Kroeker 0bc19a1335
Update SME kernel details 2025-08-18 14:38:16 -07:00
Martin Kroeker ca542f319f
Add VORTEXM4 2025-08-18 08:41:38 -07:00
Martin Kroeker 53d3bb50cc
Get symbol name from build system; change b.first to b.mi for AppleClang compatibility 2025-08-18 01:37:50 -07:00
Martin Kroeker 08a00326a4
Build symbol name from build system variables 2025-08-18 01:35:41 -07:00
Martin Kroeker 89898fc499
Add sgemm_direct_performant for switching between direct and regular kernels 2025-08-18 01:31:40 -07:00
Martin Kroeker 22c6607db9
Use ASMNAME to get symbol name from build system; leave x18 unused as reserved on MacOS 2025-08-18 01:30:10 -07:00
Martin Kroeker ca22e28ca1
Rename sgemm_direct_sme1.S to sgemm_direct_sme1_2VLx2VL.S 2025-08-18 01:25:44 -07:00
Martin Kroeker f3b2a15fad
Merge pull request #5420 from yuanjia111/develop 
Move the value assignment of vector x in gemv_n_sve.c to the outermos…
 2025-08-16 12:06:53 -07:00
yuanjia 803e8d4838 Move the value assignment of vector x in gemv_n_sve.c to the outermost loop to reduce the repeated data retrieval. 
1.Verify correctness using BLAS-Tester
    2.Using the built-in benchmark to verify performance, the performance of float and doule type improved by about 60% and about 40% respectively.The test command is:
     export OMP_NUM_THREADS=1;numactl -C 10 -l ./sgemv.goto 3000 4000 100
     export OMP_NUM_THREADS=1;numactl -C 10 -l ./dgemv.goto 3000 4000 100
 2025-08-12 18:03:16 +08:00
Chris Sidebottom 5f47b872f1 Remove older kernels for BGEMM on NEOVERSEV1 2025-08-11 09:25:19 +00:00
Chris Sidebottom 114316f361 Optimize SBGEMM / BGEMM for NEOVERSEV1 further 
This changes the kernels to pack full SVE vectors and reduces the
overall complexity of the inner GEMM loop.
 2025-08-11 09:25:13 +00:00
Martin Kroeker f1ee61ea30
Include NEON header for the bfloat conversion functions 2025-08-04 00:21:39 -07:00
Martin Kroeker b3ffd5524a
Include NEON header for the bfloat conversion functions 2025-08-04 00:20:28 -07:00
Martin Kroeker a5e7c0e3e0
Merge pull request #5396 from abhishek-iitmadras/abhishekk_bfloat16 
ARM64: Enable bfloat16 kernels by default
 2025-07-28 13:39:08 -07:00
abhishek-fujitsu 0bc79da587 add neon header 2025-07-25 11:10:20 +05:30
Chris Sidebottom ea2faf0c9a Add optimized BGEMM for NEOVERSEN2 target 
This re-uses the existing NEOVERSEN2 8x4 `sbgemm` kernel to implement `bgemm`.
 2025-07-24 10:59:28 +00:00
Chris Sidebottom 2c3cdaf74e Optimized BGEMV for NEOVERSEV1 target 
- Adds bgemv T based off of sbgemv T kernel
- Adds bgemv N which is slightly alterated to not use Y as an
accumulator due to the output being bf16 which results in loss of
precision
- Enables BGEMM_GEMV_FORWARD to proxy BGEMM to BGEMV with new kernels
 2025-07-23 10:51:41 +01:00
Martin Kroeker 39c90f9859
Merge pull request #5380 from quic/topic/sgemm_direct_sme1_alpha_beta 
SME1 based direct kernel (with alpha and beta) for cblas_sgemm level 3
 2025-07-18 23:23:39 +02:00
Rajendra Prasad Matcha eae0abfdb6 SME1 based direct kernel with alpha and beta for cblas_sgemm level 3 API. 2025-07-17 16:14:31 +05:30
Chris Sidebottom 740efd71c4 Add optimized BGEMM kernel for NEOVERSEV1 target 
This also improves the testing and generic kernel by re-using the BF16
conversion functions.

Built on top of https://github.com/OpenMathLib/OpenBLAS/pull/5357 and derived from https://github.com/OpenMathLib/OpenBLAS/pull/5287

Co-authored-by: Ye Tao <ye.tao@arm.com>
 2025-07-10 23:23:27 +00:00
Martin Kroeker fd37406817
Merge branch 'develop' into optimized_gemv_n_1x3 2025-07-08 21:05:30 +02:00
Iha, Taisei f7ad906b49 Performance improvements of [SD]DOT with loop-unrolling on A64FX 2025-07-04 22:57:44 +09:00
Martin Kroeker ee26caffb3
Merge pull request #5309 from davidz-ampere/dev-ampereone 
Add support for Ampere AmpereOne processors
 2025-06-24 12:27:08 +02:00
davidz-ampere aa90ab4142 Add support for Ampere AmpereOne processors 2025-06-24 00:12:34 -04:00
Ian McInerney badef1d32e Update sbgemm_tcopy_4_neoversev1 kernel to use standard C types 2025-06-19 14:26:16 +01:00
davidz-ampere 84730068af reduce duplicate kernel code 2025-06-17 03:05:34 -04:00
davidz-ampere be68ef03b4 Add support for Ampere processors 2025-06-15 22:00:40 -04:00
Martin Kroeker 58eeb9041c
fix handling of dummy2 2025-06-12 03:03:01 -07:00
Martin Kroeker 1589d0b21e
Merge pull request #5281 from martin-frbg/zscal_arm64 
kernel/arm64: fixed cscal and zscal
 2025-06-12 01:04:18 -07:00
Sharif Inamdar 8279e68805 Optimize gemv_n_sve_v1x3 kernel 
- Calculate predicate outside the loop
- Divide matrix in blocks of 3
 2025-06-11 10:16:56 +00:00
Arne Juul 5442aff218 Accumulate results in output register explicitly 2025-06-09 19:03:22 +00:00
Martin Kroeker 28f8fdaf0f
support flag for NaN/Inf handling and fix scaling of NaN/Inf values 2025-05-23 14:59:59 +02:00
Martin Kroeker 5141a90993
Fix ARMV9SME target in DYNAMIC_ARCH and add SME query code for MacOS (#5222) 
* Fix ARMV9SME target and add support_sme1 code for MacOS
* make sgemm_direct unconditionally available on all arm64
* build a (dummy) sgemm_direct kernel on all arm64





* Update dynamic_arm64.c
 2025-05-10 22:39:32 +02:00