Dmitry Babokin
|
6d649e1dff
|
Enabling LLVM 5.0 and making fixes to track changes in LLVM for the past
couple months.
The changes are tested with LLVM 3.9, 4.0 and trunk on MacOS (sse4,
avx2, skx).
|
2017-03-01 11:10:34 -08:00 |
|
Dmitry Babokin
|
ff298f21b7
|
Adding SVML support to AVX512 targets
|
2016-11-30 05:27:10 +03:00 |
|
Niall Douglas (s [underscore] sourceforge {at} nedprod [dot] com)
|
7af7659ac2
|
Fix ARM NEON output not always being inlined. Also improved scope for ARM NEON optimisation by LLVM, gained about 2% on my code here.
|
2016-09-05 15:56:25 +01:00 |
|
Dmitry Babokin
|
45b306480e
|
-Adding support for LLVM 4.0
-Switching 3.9 support to branch/release_39
-Switching 3.8 support to tags/release_381
|
2016-07-20 22:16:50 +03:00 |
|
Vsevolod Livinskiy
|
f47e1d5cae
|
[AVX-512] Replace i1 mask with i8
|
2016-07-04 17:01:46 +03:00 |
|
Andrey Shishpanov
|
7691d961c1
|
Fixed ifelse in rsqrt, rcp def. for knl, skx (compfails with old LLVM).
|
2016-03-11 15:32:59 +03:00 |
|
Andrey Shishpanov
|
1324e6cdd5
|
added SKX target definition
|
2016-02-25 00:43:58 +03:00 |
|
Dmitry Babokin
|
f6dbffd58c
|
Updating copyright dates for recently modified files
|
2016-02-04 15:14:19 +03:00 |
|
Dmitry Babokin
|
536b30e12c
|
LLVM 3.9 version support
|
2016-01-14 17:26:38 +03:00 |
|
Anton Mitrokhin
|
ef51f8c648
|
[AVX512]: Make blend optimizations possible for avx512
|
2015-11-10 18:16:43 +03:00 |
|
Anton Mitrokhin
|
28b402a778
|
[AVX-512]: transcendentals: add exp() implementation, TODO: log() and pow()
|
2015-11-03 13:34:11 +03:00 |
|
Anton Mitrokhin
|
434aa30d12
|
IR change for x86 mask load/store instructions in LLVM 3.8 (r250817)
|
2015-10-21 17:41:13 +03:00 |
|
Anton Mitrokhin
|
7448ee97f2
|
Fixed AVX-512 IR incompatibility issue
|
2015-08-17 19:06:57 +03:00 |
|
Anton Mitrokhin
|
d37455925f
|
[AVX-512]: Scatters for i32/float
|
2015-08-16 21:50:44 +03:00 |
|
Anton Mitrokhin
|
d2720e2490
|
[AVX-512]: Gathers for float, fixed i32 gathers
|
2015-08-16 21:50:44 +03:00 |
|
Andrey Shishpanov
|
e11022c95a
|
replaced gathers for i32
|
2015-08-16 21:50:44 +03:00 |
|
Anton Mitrokhin
|
db1d817dee
|
Remove 'readnone' attribute away from non-readnone sincos()
|
2015-08-02 15:04:29 +03:00 |
|
Andrey Shishpanov
|
415f2e938c
|
some fixes for trunk 3.8
|
2015-07-17 14:27:28 +03:00 |
|
Anton Mitrokhin
|
f864338ce2
|
Changed GEP calls in target-avx512.ll to work with old LLVM
|
2015-07-16 15:32:50 +03:00 |
|
Vsevolod Livinskiy
|
21da408832
|
[AVX512]: bugfixing
|
2015-07-10 11:34:16 +03:00 |
|
Vsevolod Livinskiy
|
ba10b91648
|
[AVX-512]: masked_store was replaced
|
2015-07-09 15:36:03 +03:00 |
|
Vsevolod Livinskiy
|
25aeedb003
|
[AVX-512]: masked_load_float/double was replaced
|
2015-07-09 13:45:14 +03:00 |
|
Vsevolod Livinskiy
|
b6d2d8dd4c
|
[AVX-512]: rsqrt and rcp were replaced
|
2015-07-08 19:12:38 +03:00 |
|
Vsevolod Livinskiy
|
8c1bd4ec32
|
[AVX-512]: replace with avx512 intrinsics
|
2015-07-08 16:54:27 +03:00 |
|
Anton Mitrokhin
|
c50ce30b00
|
[AVX-512]: fixed a couple of tests
|
2015-07-08 10:20:57 +03:00 |
|
Vsevolod Livinskiy
|
28b49837fc
|
round2to16 was added
|
2015-05-21 16:53:18 +03:00 |
|
Anton Mitrokhin
|
5ec16356d0
|
[AVX512]: copyright update
|
2015-05-21 15:29:04 +03:00 |
|
Vsevolod Livinskiy
|
f5e7165537
|
[AVX512]: packed_load/store
|
2015-05-21 15:27:27 +03:00 |
|
Vsevolod Livinskiy
|
d7cd5986db
|
[AVX512]: disable prefetch
|
2015-05-21 14:51:51 +03:00 |
|
Vsevolod Livinskiy
|
3514e03327
|
[AVX512]: disable Transcendentals and Trigonometry
|
2015-05-21 14:51:50 +03:00 |
|
Anton Mitrokhin
|
ef9c98fba8
|
[AVX512]: uniform float/double round/ceil/floor
|
2015-05-21 14:51:50 +03:00 |
|
Vsevolod Livinskiy
|
2110708c8e
|
[AVX512]: sqrt/rsqrt/rcp
|
2015-05-21 14:51:50 +03:00 |
|
Vsevolod Livinskiy
|
82f5716362
|
[AVX512]: max/min functions
|
2015-05-21 14:51:50 +03:00 |
|
Anton Mitrokhin
|
a6b7e717f5
|
[AVX512]: gathers/scatters
|
2015-05-21 14:51:50 +03:00 |
|
Anton Mitrokhin
|
66b94fc37c
|
[AVX512]: add default -sde- wrapexe to runtests.py for knl-avx512 target; float/double varying rounding
|
2015-05-21 14:51:50 +03:00 |
|
Anton Mitrokhin
|
f2743a6dc5
|
[AVX512]: masked_load_i8/16/32/64
|
2015-05-21 14:51:50 +03:00 |
|
Anton Mitrokhin
|
28fda1a013
|
[AVX512]: movmsk/any/all/none
|
2015-05-21 14:51:43 +03:00 |
|
Vsevolod Livinskiy
|
7c9d9f6ee6
|
[AVX512]: reduce operations was added
|
2015-05-21 14:51:32 +03:00 |
|
Anton Mitrokhin
|
2549fa12c9
|
[AVX512]: masked load-store (not all loads)
|
2015-05-21 14:51:26 +03:00 |
|
Vsevolod Livinskiy
|
bea7cc9a81
|
[AVX512]: half/float conversions
|
2015-05-21 14:51:21 +03:00 |
|
Vsevolod Livinskiy
|
9a03cd3590
|
[AVX512]: definitions through util.m4 was added
|
2015-05-21 14:51:16 +03:00 |
|
Anton Mitrokhin
|
46528caa5a
|
[AVX512]: add avx-based ll file
|
2015-05-21 14:51:08 +03:00 |
|
Anton Mitrokhin
|
7628f2a6c9
|
[AVX512]: try gemeric-16 like builtins
|
2015-05-21 14:51:03 +03:00 |
|
Vsevolod Livinskiy
|
d01718aa91
|
[AVX512]: avx512 common file was added
|
2015-05-21 14:50:56 +03:00 |
|
Anton Mitrokhin
|
3eccce5e4f
|
[AVX512]: new .ll file for knl target
|
2015-05-21 14:50:51 +03:00 |
|
Vsevolod Livinskiy
|
35222694e5
|
[AVX512]: knl target was added
|
2015-05-21 14:49:43 +03:00 |
|
Dmitry Babokin
|
ad97d70a43
|
Adding check for OSXSAVE before checking xgetbv in cpu detection code.
Fix for #1026
|
2015-04-24 21:28:06 +03:00 |
|
Dmitry Babokin
|
8e47273186
|
Copyright refresh
|
2015-04-22 16:39:11 +03:00 |
|
Dmitry Babokin
|
4518bce71b
|
Merge pull request #1013 from egaburov/armfix
Added fix for ispc on ARM
|
2015-04-22 16:09:50 +03:00 |
|
Vsevolod Livinskiy
|
7729070481
|
Check for new isa for KNL and SKX was added.
|
2015-04-17 12:00:36 +03:00 |
|