 conrad
|
cce74c0e49
|
use direct_dot() instead of direct_dot_generic() since the former can now detect integer types
|
2025-07-18 14:20:01 +10:00 |
|
|
conrad
|
0db44c0fa8
|
use forced optimisation only for fp16
|
2025-07-18 14:09:27 +10:00 |
|
|
conrad
|
fb2fd4a049
|
split direct_dot_generic() into direct_dot_generic_short() and direct_dot_generic_long()
|
2025-07-18 11:44:59 +10:00 |
|
|
conrad
|
996bb45534
|
rename direct_dot_arma() to direct_dot_generic()
|
2025-07-18 11:13:51 +10:00 |
|
|
conrad
|
0601761cf2
|
add handling of cx_fp16
|
2025-07-18 00:11:38 +10:00 |
|
|
conrad
|
084a695129
|
add TODO
|
2025-07-17 18:37:42 +10:00 |
|
|
conrad
|
344f123ac4
|
add warning message
|
2025-07-17 18:33:37 +10:00 |
|
|
conrad
|
deaf5c413a
|
don't use divide-and-conquer when matrix size is too large for LAPACK with 32 bit ints
|
2025-07-17 18:31:14 +10:00 |
|
|
conrad
|
9fcd2f3449
|
add TODOs
|
2025-07-17 17:06:22 +10:00 |
|
|
conrad
|
71eeb56cee
|
add TODOs
|
2025-07-17 17:00:20 +10:00 |
|
|
conrad
|
a135bbcb94
|
remove no longer used variable
|
2025-07-17 16:58:19 +10:00 |
|
|
conrad
|
5f334da813
|
don't use divide-and-conquer when matrix size is too large for LAPACK with 32 bit ints
|
2025-07-17 16:54:41 +10:00 |
|
|
conrad
|
ad478221e1
|
don't use divide-and-conquer when matrix size is too large for LAPACK with 32 bit ints
|
2025-07-17 14:38:58 +10:00 |
|
|
conrad
|
7dc07e12b2
|
partial revert
|
2025-07-16 17:05:18 +10:00 |
|
|
conrad
|
8c295b058e
|
use constexpr for 0
|
2025-07-16 16:58:59 +10:00 |
|
|
conrad
|
9c9a25f7f6
|
force vectorisation of direct_dot_arma() under GCC
|
2025-07-16 15:17:48 +10:00 |
|
|
conrad
|
15e03f9bac
|
remove forced inline
|
2025-07-16 14:21:52 +10:00 |
|
|
conrad
|
677bdb3bad
|
swap order to ensure consistency with previous version
|
2025-07-14 14:55:45 +10:00 |
|
|
conrad
|
79f6b87781
|
fix
|
2025-07-14 14:28:15 +10:00 |
|
|
conrad
|
39b2a52bd0
|
allow to optionally disable optimised handling of pow()
|
2025-07-14 14:02:19 +10:00 |
|
|
conrad
|
ac3fcabca8
|
simplifications
|
2025-07-14 13:29:53 +10:00 |
|
|
conrad
|
4aaf66547e
|
compile time note that ARMA_CRIPPLED_LAPACK is not supported
|
2025-07-14 13:20:12 +10:00 |
|
|
conrad
|
d676e4cfa2
|
remove handling of ARMA_CRIPPLED_LAPACK
|
2025-07-14 13:17:56 +10:00 |
|
|
conrad
|
3b1ac86a9f
|
remove auxlib::crippled_lapack()
|
2025-07-14 13:07:00 +10:00 |
|
|
conrad
|
460e13a272
|
expand handling of fp16 and cx_fp16
|
2025-07-12 23:44:53 +10:00 |
|
|
conrad
|
ef4c5a7844
|
ensure blas cx type only
|
2025-07-12 21:25:29 +10:00 |
|
|
conrad
|
d8f31aa94f
|
prevent shadow warnings
|
2025-07-12 19:38:34 +10:00 |
|
|
conrad
|
5155b0bb38
|
add ARMA_DONT_USE_FP16 option
|
2025-07-12 19:08:58 +10:00 |
|
|
conrad
|
bfb650ea5d
|
fix
|
2025-07-12 11:21:31 +10:00 |
|
|
conrad
|
7369ee1086
|
allow use of fp16 with fft functions
|
2025-07-12 00:04:15 +10:00 |
|
|
conrad
|
f06c6ab8f8
|
rename arma_is_blas_type_only to arma_is_real_or_cx_only for consistency
|
2025-07-11 23:46:23 +10:00 |
|
|
conrad
|
206606d40d
|
rename arma_fp16_only to arma_fp16_real_only
|
2025-07-11 23:24:37 +10:00 |
|
|
conrad
|
1e731455ec
|
fix
|
2025-07-11 23:21:08 +10:00 |
|
|
conrad
|
c87cf374c4
|
add support for cx_fp16
|
2025-07-11 23:14:44 +10:00 |
|
|
conrad
|
bbe5dcf225
|
add is_blas_cx trait
|
2025-07-11 20:45:49 +10:00 |
|
|
conrad
|
6fd1531d80
|
enable saving/loading sparse fp16
|
2025-07-11 20:41:59 +10:00 |
|
|
conrad
|
30ee79ee37
|
enable fp16 support for SpMat
|
2025-07-11 16:06:01 +10:00 |
|
|
conrad
|
9264db2876
|
fix
|
2025-07-10 15:34:34 +10:00 |
|
|
conrad
|
35668ad125
|
limit to supported blas types only
|
2025-07-10 15:18:46 +10:00 |
|
|
conrad
|
4e6392f0c2
|
replace is_supported_blas_type with is_blas_type
|
2025-07-10 14:57:30 +10:00 |
|
|
conrad
|
5f833e68ce
|
explicitly limit is_cx to std::complex<float> and std::complex<double>
|
2025-07-10 14:53:32 +10:00 |
|
|
conrad
|
0820f68e69
|
restrict to blas real types only
|
2025-07-09 14:05:48 +10:00 |
|
|
conrad
|
c3db1ead03
|
ensure junk args are const and are explicitly ignored
|
2025-07-09 13:48:27 +10:00 |
|
|
conrad
|
7dadab9eb1
|
add stub mat_norm_2() for fp16
|
2025-07-09 13:42:00 +10:00 |
|
|
conrad
|
46de13b363
|
use std::uniform_real_distribution with promoted type
|
2025-07-08 23:32:03 +10:00 |
|
|
conrad
|
98df75f7d7
|
simplifications
|
2025-07-07 23:45:27 +10:00 |
|
|
conrad
|
91065683da
|
minor layout fixes
|
2025-07-07 23:29:11 +10:00 |
|
|
conrad
|
c2175c86ee
|
C++23 required for fp16 support
|
2025-07-07 20:14:29 +10:00 |
|
|
conrad
|
8239b26f62
|
use only std::float16_t as fp16
|
2025-07-07 20:04:50 +10:00 |
|
|
conrad
|
a32600ae17
|
fix
|
2025-07-07 17:06:31 +10:00 |
|