Chip Kerchner
eb9bbcc299
Common unroll code.
2026-03-14 16:38:12 +00:00
Chip Kerchner
fda433f25a
Unroll K for N less than or equal to 4.
2026-03-13 22:17:47 +00:00
Chip Kerchner
9c16449dfd
Add K-unrolling to M = 8. Other small changes.
2026-03-13 15:59:29 +00:00
Chip Kerchner
6d6af1dee5
Add bool types for C.
2026-03-12 13:13:34 +00:00
Chip Kerchner
376d3a138f
Fast performing edges for FP32 GEMM of RVV.
2026-03-12 12:33:25 +00:00
Chip Kerchner
548a9f3eaa
Merge remote-tracking branch 'origin/develop' into HEAD
2026-03-11 21:07:28 +00:00
Martin Kroeker
bc69f86dba
Merge pull request #5671 from martin-frbg/cpuid_x86_cache
...
Update x86 cache size decoding table from current sandpile.org data
2026-03-11 11:34:44 +01:00
Martin Kroeker
1ff3a1a13d
Merge pull request #5670 from amritahs-ibm/handle_fp16_power
...
powerpc: Bypass FP16 as BF16 on Power
2026-03-11 09:32:31 +01:00
Martin Kroeker
02f5f620de
Update cache size decoding table with sandpile.org data
2026-03-10 22:25:48 +01:00
Amrita H S
1a708bac8a
powerpc: Bypass FP16 as BF16 on Power
...
typedef FP16 as BF16 on Power as FP16 is
not yet supported
2026-03-08 23:32:42 -05:00
Martin Kroeker
55b16e5923
Merge pull request #5643 from martin-frbg/neov2_param
...
Fix SGEMM returning wrong results in multithreading on NeoverseV2
2026-03-06 11:51:08 +01:00
Martin Kroeker
37262654d9
Merge pull request #5667 from fadara01/accelerate_sve128_sbgemm
...
Accelerate SVE128 SBGEMM/BGEMM
2026-03-06 09:14:44 +01:00
Martin Kroeker
75e2f12dae
Merge pull request #5668 from martin-frbg/issue5665
...
Fix CMake DYNAMIC_ARCH compilation with old compilers on ARM64
2026-03-05 20:40:48 +01:00
Martin Kroeker
d073702cdf
Merge pull request #5661 from martin-frbg/update_readme_java
...
Fix leftover wiki links in the README and add java insights from issue #5109
2026-03-05 18:30:33 +01:00
Martin Kroeker
78fd789da0
Add compiler test for SVE support
2026-03-05 17:35:15 +01:00
Fadi Arafeh
f30202b705
Accelerate SVE128 SBGEMM/BGEMM
...
This accelerates SBGEMM/BGEMM by extending the existing 8x4 kernel to 8x8 (unrolling N by 8)
Not sure if it's a good idea to delete the previous 8x4 kernel?
Here are the speedups on single core Neoverse-V2 (SVE128) compared to prev state:
Per-shape speedup
M=N=K=64: SBGEMM 1.164x (16.42%), BGEMM 1.133x (13.30%)
M=N=K=128: SBGEMM 1.220x (22.02%), BGEMM 1.186x (18.56%)
M=N=K=256: SBGEMM 1.241x (24.08%), BGEMM 1.235x (23.54%)
M=N=K=512: SBGEMM 1.240x (23.95%), BGEMM 1.227x (22.75%)
M=N=K=1024: SBGEMM 1.251x (25.11%), BGEMM 1.232x (23.23%)
M=N=K=2048: SBGEMM 1.235x (23.47%), BGEMM 1.246x (24.64%)
Signed-off-by: Fadi Arafeh <fadi.arafeh@arm.com >
2026-03-05 13:50:07 +00:00
Martin Kroeker
22fc689fa7
Merge pull request #5666 from martin-frbg/issue5664
...
Improve processing of linker arguments in f_check
2026-03-05 14:19:43 +01:00
Martin Kroeker
91eb0a638c
Avoid splitting linker args on dashes not preceded by a space
2026-03-05 10:09:29 +01:00
Martin Kroeker
1590d8baf0
fix install.md link for cortex-m
2026-03-04 23:12:33 +01:00
Martin Kroeker
98864c7c6f
fix reintroduced typo again
2026-03-04 20:40:45 +01:00
Martin Kroeker
db6bbc7150
Merge pull request #5660 from martin-frbg/issue5658
...
Rewrite the Haswell SROT/DROT kernel tail loop with AVX2 to get consistent FMA rounding
2026-03-04 18:15:34 +01:00
Martin Kroeker
ecdabf9d74
Merge pull request #5663 from martin-frbg/issue5662
...
Move the early exit in ?GESV for NRHS=0 after the GETRF call
2026-03-04 16:14:06 +01:00
Martin Kroeker
dc8b16c57c
Move the early exit for NRHS=0 after the GETRF call
2026-03-04 12:21:27 +01:00
Martin Kroeker
754ad2ad4f
Fix leftover wiki links and add java insights from issue 5109
2026-03-04 11:44:36 +01:00
Martin Kroeker
3166fffcec
Merge pull request #5659 from lindsayad/handle-emerald-rapids
...
Handle Intel's emerald rapids and do some formatting in the cpuid_x86 file
2026-03-03 20:06:49 +01:00
Martin Kroeker
df29cc0205
Use AVX2 in the tail loop too for consistent FMA rounding
2026-03-03 15:51:51 +01:00
Alex Lindsay
692023e364
Switch case ordering for exmodel 12 to be sorted
2026-03-02 16:23:38 -07:00
Alex Lindsay
5a534a63e8
clang-format and make cpuid_x86.c more readable
2026-03-02 15:51:46 -07:00
Alex Lindsay
303903e29c
Handle emerald rapids model
2026-03-02 15:39:41 -07:00
Martin Kroeker
18638c70ef
Merge pull request #5656 from martin-frbg/issue5653
...
Add pragma to limit optimization in POWER10 DGEMV kernel
2026-02-22 15:54:59 +01:00
Martin Kroeker
ef27ec6bed
Add pragma to limit optimization level
2026-02-22 13:42:41 +01:00
Martin Kroeker
da0e066c9e
Merge pull request #5655 from martin-frbg/intel-default-cpuid
...
Add feature-based fallbacks for unknown/future Intel CPUIDs
2026-02-20 22:19:25 +01:00
Martin Kroeker
1d0ca19457
Add feature-based fallbacks for unknown/future Intel cpus
2026-02-20 16:45:08 +01:00
Martin Kroeker
30cf14c548
Merge pull request #5640 from ChipKerchner/RVV_Narrow_Accumulate_FP16_GEMM
...
Added ability to accumulate in FP16. Convert BF16 to FP32. For FP16 and BF16 GEMM in RISC-V (BF16 now works for pre-RVA23)
2026-02-20 14:22:27 +01:00
Martin Kroeker
b4db4a1713
Merge pull request #5654 from martin-frbg/issue5627-2
...
Use generic SCAL kernels for PPC970 running FreeBSD
2026-02-20 12:45:42 +01:00
Martin Kroeker
43728ade59
Merge pull request #5651 from martin-frbg/issue5650
...
Fix gmake build with only a subset of precision types
2026-02-20 08:02:50 +01:00
Martin Kroeker
46b963b9a0
Use generic C kernels for SCAL on FreeBSD
2026-02-19 22:46:03 +01:00
Martin Kroeker
822aae6cab
Merge pull request #5652 from martin-frbg/issue5649
...
Fix passing of C/ZDOTC results in C-converted LAPACK on non-Windows systems
2026-02-19 22:05:10 +01:00
Martin Kroeker
dccbf18c1f
fix storing of ZDOTC result on non-Windows
2026-02-19 19:45:20 +01:00
Martin Kroeker
0cfb587fde
fix storing of ZDOTU result on non-Windows
2026-02-19 18:46:20 +01:00
Martin Kroeker
92fcffff54
fix storing of CDOTC result on non-Windows
2026-02-19 18:45:07 +01:00
Martin Kroeker
5a07c1b61c
Delete misplaced lapack-netlib/chpgst.c
2026-02-19 18:42:30 +01:00
Martin Kroeker
11986454b3
fix storing of CDOTU result on non-Windows
2026-02-19 18:05:59 +01:00
Martin Kroeker
946a2cffec
fix storing of CDOTC result on non-Windows
2026-02-19 16:53:18 +01:00
Martin Kroeker
ef1c06f5eb
fix storing of CDOTC result on non-Windows
2026-02-19 14:08:59 +01:00
Martin Kroeker
b7542ffb3d
fix storing of CDOTC result on non-Windows
2026-02-19 13:34:48 +01:00
Martin Kroeker
5d29f88fed
fix storing of CDOTC result on non-Windows
2026-02-19 13:18:20 +01:00
Martin Kroeker
61db4e8191
fix storing of CDOTC result on non-Windows
2026-02-19 13:07:15 +01:00
Martin Kroeker
bf0d7eaacc
fix storing of CDOTC result on non-Windows
2026-02-19 12:59:46 +01:00
Martin Kroeker
1da181dac6
fix storing of CDOTC result on non-Windows
2026-02-19 12:05:27 +01:00