eden-emu/opus
17
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
5187 commits 285 branches 59 tags 26 MiB
Commit graph

44 commits

Author SHA1 Message Date
Timothy B. Terriberry
59dc75fa97
Rework 32-bit SSE loads yet again. 
The existing code in vec_avx.h produced
  warning: dereferencing type-punned pointer will break
   strict-aliasing rules
 with gcc 6.4.0.
We already had a macro to work around this within the rules of the
 C standard, but trying to use that here does not get optimized
 into a single MOVD like we were hoping.
Replacing it with memcpy() instead does get optimized correctly,
 but requires switching from a macro to an inline function in order
 to be able to declare a local variable and return a value.
We already have such an inline function in NSQ_del_dec_avx2.c, so
 hoist that out and use it everywhere, and then convert vec_avx.h
 to use it also.
 2024-02-23 02:23:37 -05:00
Jean-Marc Valin
2e034f6f31
Adding RTCD for DNN code 
Starting with compute_linear()
 2023-11-15 23:45:32 -05:00
Jean-Marc Valin
58923f61c2
Fix non-AVX builds 2023-11-11 03:24:21 -05:00
Jean-Marc Valin
1ada7d4d6f
Vectorizing sgemv for multiples of 4 with SSE 2023-11-03 02:48:38 -04:00
Jean-Marc Valin
62b546436f
Speed up general case for float matrix multiply 2023-10-30 00:08:53 -04:00
Jean-Marc Valin
88c58cfaf3
nnet.h no longer needs to #include "vec.h" 2023-10-20 17:25:27 -04:00
Jean-Marc Valin
81624caf9c
Silencing alignment warnings on x86 intrinsics 
Those intrinsics don't actually require alignment so we're OK
 2023-10-07 17:45:39 -04:00
Michael Klingbeil
d431c321f1
Fixes vnni macro redefinition with clang 2023-09-01 23:18:21 -04:00
Jean-Marc Valin
e9f8402a71
Handle float matrices with multiple of 8 rows 2023-08-01 19:16:27 -04:00
Jean-Marc Valin
8f7c72a662
Always define USE_SU_BIAS in vec_avx.h 2023-07-22 14:56:05 -04:00
Jean-Marc Valin
4710bdf712
Add SSE2 support 
Not so much for old machines, as for getting decent performance
when not setting -march= (SSE2 is part of the amd64 ABI).
 2023-07-22 14:56:05 -04:00
Jean-Marc Valin
9261eb5c37
Refactoring to make VNNI and SSE2 easier 2023-07-22 14:56:04 -04:00
Jean-Marc Valin
62cd1c963b
Transition to LinearLayer and remove unused code 2023-07-20 01:01:34 -04:00
Jean-Marc Valin
f5a68a41b0
Add generic linear layer 
Should be able to handle all previous GRU variants and more.
 2023-07-20 01:01:32 -04:00
xnorpx
7122abde59 Rename celt_exp to lpcnet_exp 
Depending on what defines are set there is collisions with the ones
in Opus. To avoid these errors we rename the exp functions and
macros.

Signed-off-by: Jean-Marc Valin <jmvalin@amazon.com>
 2023-05-24 00:46:20 -04:00
xnorpx
879084f6f0 Fix some of C4244 double to float warnings 2023-05-24 00:30:19 -04:00
xnorpx
702fffb70a Include math.h to make header self-contained. 
Signed-off-by: Jean-Marc Valin <jmvalin@amazon.com>
 2023-05-23 11:24:35 -04:00
xnorpx
5b96946277 Use pragma message instead of warning on MSVC 
Signed-off-by: Jean-Marc Valin <jmvalin@amazon.com>
 2023-05-23 02:31:09 -04:00
Jan Buethe
d80f99f78b added void to shut up missing prototype warning 2022-10-21 15:33:41 +00:00
Jean-Marc Valin
60a009b457 Making codebase C90-compliant 2022-01-19 18:10:44 -05:00
Jean-Marc Valin
4298f2f9e1 Adding support for SSE2 and SSSE3 2021-07-11 03:36:20 -04:00
Jean-Marc Valin
116bcb38fb Adding SSE 4.1 for older platforms 
AVX without AVX2 should now work again too.
 2021-07-10 14:08:01 -04:00
Jean-Marc Valin
e8f70128d5 same conversion cleanup as 3206cec for sgemv_accum8x4() 2021-07-10 01:59:49 -04:00
Jean-Marc Valin
714380e71b More manual unrolling 2021-07-10 01:59:49 -04:00
Jean-Marc Valin
44fe055682 cleanup float<->int conversions 2021-07-10 01:59:49 -04:00
Jean-Marc Valin
60d6eab63d Doing a bit of unrolling to speed things up 2021-07-10 01:59:49 -04:00
Jean-Marc Valin
54abdb6f5d Sparse matrix indexing optimization 
The 4* is now stored in the table to avoid computing it in the loop
 2021-07-10 01:59:49 -04:00
Jean-Marc Valin
d332100808 Representing output pdf as binary probability tree 
Saves on the MDense/softmax computation since we only need to compute
8 values instead of 256.
 2021-07-10 01:59:49 -04:00
Jean-Marc Valin
e35441f2cc Faster activation functions for AVX 
Using rational function approximation for tanh() and sigmoid.
 2021-06-29 04:05:48 -04:00
Jean-Marc Valin
c1535c8ccf Adding option to disable int8 dot products 2021-06-24 17:31:05 -04:00
Jean-Marc Valin
0b9f6bab81 Remove unnecessary mask in exp() approximation 
This isn't necessary since valid exponents can't flip the sign bit
 2021-06-21 01:34:38 -04:00
Jean-Marc Valin
ae2ae5ead6 Remove useless multiply by one 
See bffdcee95 (commitcomment-46372726)
 2021-06-21 01:30:51 -04:00
Jean-Marc Valin
8c3fe6f31d Cleaning up float version 2021-01-16 02:11:21 -05:00
Jean-Marc Valin
83657d0e43 Dot product AVX2 code for non-sparse multiply 2021-01-16 02:11:21 -05:00
Jean-Marc Valin
e695355ba5 some cleanup 2021-01-16 02:11:20 -05:00
Jean-Marc Valin
d87f974431 Vectorizing conversion 2021-01-16 02:11:20 -05:00
Jean-Marc Valin
6b582edbed WIP: remove scalar code from AVX2 code 2021-01-16 02:11:20 -05:00
Jean-Marc Valin
be392e3857 WIP: Got some AVX2 code working 2021-01-16 02:11:20 -05:00
Jean-Marc Valin
2b4652f9f6 WIP: cleanup 2021-01-16 02:11:20 -05:00
Jean-Marc Valin
bce779886d WIP: signed*unsigned arithmetic 2021-01-16 02:11:20 -05:00
Jean-Marc Valin
11736ca9e3 WIP: 8-bit mul 2021-01-16 02:11:19 -05:00
Jean-Marc Valin
c045702e51 Add non-dot-product AVX code 2021-01-16 02:11:19 -05:00
Jean-Marc Valin
8e405b44e0 Improve accuracy of AVX sigmoid 
Reciprocal approximation could cause the sigmoid output to be
greater than 1.0.
 2021-01-16 01:51:39 -05:00
David Rowe
7dc696b9a4 refactored for different machines, sgemv_accum16 using NEON intrisics 
Signed-off-by: Jean-Marc Valin <jmvalin@jmvalin.ca>
 2018-12-10 21:28:29 -05:00