conrad
|
efc2bcf316
|
rearrange
|
2025-08-11 18:35:11 +10:00 |
|
conrad
|
e38cbaab8e
|
handle corner case where tic() is reused
|
2025-08-11 18:33:09 +10:00 |
|
conrad
|
a30ad16ebe
|
rename warm() to unfreeze()
|
2025-08-11 18:25:11 +10:00 |
|
conrad
|
d0da492c26
|
use type promotion for fp16 version of mat_norm_2()
|
2025-08-11 17:11:17 +10:00 |
|
conrad
|
3f5d4784fc
|
add freeze() and warm()
|
2025-08-11 17:03:00 +10:00 |
|
conrad
|
7a15d37a99
|
simplifications
|
2025-08-11 14:32:48 +10:00 |
|
conrad
|
1e40f043ff
|
use constexpr
|
2025-08-11 01:09:38 +10:00 |
|
conrad
|
21f2bf28d2
|
simplification
|
2025-08-11 00:51:52 +10:00 |
|
conrad
|
4a5c3d8f27
|
rearrange
|
2025-08-11 00:48:59 +10:00 |
|
conrad
|
2b49392e7c
|
remove no longer needed macros
|
2025-08-09 23:32:25 +10:00 |
|
conrad
|
26dcdbbf0a
|
add explanation
|
2025-08-09 23:28:59 +10:00 |
|
conrad
|
16b0912a6f
|
use C++14 deprecation attributes directly
|
2025-08-09 23:22:55 +10:00 |
|
conrad
|
6991c9aa25
|
update to https
|
2025-08-08 20:20:21 +10:00 |
|
conrad
|
e8027c3e14
|
update to https
|
2025-08-08 20:17:18 +10:00 |
|
conrad
|
b2489b2d87
|
rename variable for overall consistency
|
2025-08-08 20:05:37 +10:00 |
|
conrad
|
3de20fe5cb
|
fix
|
2025-08-07 22:53:15 +10:00 |
|
conrad
|
dd1be88826
|
use forced optimisation for fp16 dot products
|
2025-08-07 16:35:02 +10:00 |
|
conrad
|
0c8e97e10e
|
use type promotion only where necessary
|
2025-08-07 16:02:57 +10:00 |
|
conrad
|
fa4b928937
|
remove no longer needed function
|
2025-08-07 14:57:44 +10:00 |
|
conrad
|
4001efe359
|
strip out type promotion
|
2025-08-07 14:56:29 +10:00 |
|
conrad
|
c294a16d84
|
strip out type promotion
|
2025-08-07 14:52:13 +10:00 |
|
conrad
|
28670b8854
|
minor cleanup
|
2025-08-07 14:18:09 +10:00 |
|
conrad
|
c5613778a8
|
strip out type promotion
|
2025-08-07 14:17:32 +10:00 |
|
conrad
|
2c60189b8f
|
use namespace prefix
|
2025-08-06 23:14:19 +10:00 |
|
conrad
|
3cfd3f90b4
|
rearrange
|
2025-08-06 23:02:41 +10:00 |
|
conrad
|
db49fc5bb5
|
add handling of cx_fp16
|
2025-08-06 22:58:17 +10:00 |
|
conrad
|
2eb334e0ab
|
increase accuracy for fp16
|
2025-08-06 14:29:11 +10:00 |
|
conrad
|
52cd25c365
|
increase accuracy for fp16
|
2025-08-06 14:12:15 +10:00 |
|
conrad
|
43786efafe
|
better handling of fp16 via type promotion in vec_norm_1
|
2025-08-06 13:30:41 +10:00 |
|
conrad
|
624da4833a
|
better handling of fp16 via type promotion in op_norm_dot
|
2025-08-06 13:01:02 +10:00 |
|
conrad
|
5bf5668272
|
better handling of fp16 via type promotion in vec_norm_2
|
2025-08-06 12:41:52 +10:00 |
|
conrad
|
c94b17927c
|
use type promotion for fp16 elements
|
2025-08-06 10:36:07 +10:00 |
|
conrad
|
d5e8bae784
|
implement mat_norm_2() for fp16 via type promoted matrix
|
2025-08-05 11:30:13 +10:00 |
|
conrad
|
567d6ccd2c
|
rearrange
|
2025-08-04 23:55:51 +10:00 |
|
conrad
|
7eca200236
|
add promote variant of direct_mean_robust()
|
2025-08-04 23:52:48 +10:00 |
|
conrad
|
e7b39cf480
|
add type promotion variant for optimised handling of sum(square(X))
|
2025-08-04 21:31:23 +10:00 |
|
conrad
|
51d4f2e898
|
more efficient row-wise summation with promotion
|
2025-08-04 20:54:54 +10:00 |
|
conrad
|
8047a5f342
|
use type promotion for fp16
|
2025-08-04 17:06:11 +10:00 |
|
conrad
|
600642f829
|
simplifications
|
2025-08-04 15:52:34 +10:00 |
|
conrad
|
9d68301372
|
use promoted type instead of specialising for fp16
|
2025-08-04 15:19:43 +10:00 |
|
conrad
|
20ca20b31b
|
expand specialised handling of fp16
|
2025-08-01 13:01:09 +10:00 |
|
conrad
|
972f2f648f
|
simplification
|
2025-07-30 12:16:24 +10:00 |
|
conrad
|
273b91efab
|
rename op_accu_fp16mat to op_accu_mat_promote
|
2025-07-30 12:12:45 +10:00 |
|
conrad
|
c3b3a42f46
|
add specialisations for Proxy
|
2025-07-30 12:09:45 +10:00 |
|
conrad
|
8bf1eb2af8
|
reduce compiler processing when fp16 is not available
|
2025-07-30 11:39:36 +10:00 |
|
conrad
|
9072a2f7c8
|
place-holder implementation of specialised accu() for fp16
|
2025-07-30 00:19:41 +10:00 |
|
conrad
|
8abea07d35
|
simplify handling of openmp
|
2025-07-29 23:28:51 +10:00 |
|
conrad
|
2fe2b5fa36
|
fix group names
|
2025-07-29 23:07:37 +10:00 |
|
conrad
|
aba91e9610
|
refactor accu() into op_accu and spop_accu
|
2025-07-29 22:43:11 +10:00 |
|
conrad
|
89e7703875
|
specialisation for fp16
|
2025-07-28 12:56:45 +10:00 |
|