Commit graph

167 commits

Author SHA1 Message Date
SChernykh
cafd868773 Fixed errors found by static analysis 2020-12-08 12:16:59 +01:00
XMRig
c8ee6f7db8
Move Profiler and more cleanup. 2020-12-04 09:23:40 +07:00
SChernykh
f16d1837f8 Optimized JIT compiler
More branch-free code
2020-11-29 14:05:50 +01:00
SChernykh
c10ec90b60 Make single thread bench cheat-resistant
Each hash is dependent on the previous hash to make multi-threaded cheating impossible.
2020-11-15 20:38:27 +01:00
cohcho
a64ff6b7c7 CompiledVm: define default constructor 2020-11-09 16:29:42 +00:00
SChernykh
6b7b3511ce Also fix RelWithDebIfno build in Visual Studio 2020-10-27 14:25:43 +01:00
SChernykh
50bdaba526 Fixed Debug build in Visual Studio 2020-10-27 14:08:36 +01:00
SChernykh
4bac3e7695 Fix 32-bit compilation 2020-10-07 18:19:35 +02:00
xmrig
59bd6d4187
Merge pull request #1878 from SChernykh/dev
Fixed ARM compilation
2020-10-07 23:11:39 +07:00
SChernykh
166c011d37 Fixed ARM compilation 2020-10-07 18:09:42 +02:00
xmrig
1289942567
Merge pull request #1876 from SChernykh/dev
RandomX: added `huge-pages-jit` config parameter
2020-10-07 22:48:57 +07:00
SChernykh
44dcded866 RandomX: added huge-pages-jit config parameter
Set to false by default, gives 0.2% boost on Ryzen 7 3700X with 16 threads, but hashrate might be unstable on Ryzen between launches. Use with caution.
2020-10-07 17:42:55 +02:00
cohcho
a705ab775b RandomX: align args
tempHash/output must be 16-byte aligned for randomx_calculate_hash{,_first,_next}
2020-10-07 14:47:18 +00:00
xmrig
116fb3d3f9
Merge pull request #1864 from cohcho/soft_aes_optimization2
soft_aes: fix previous optimization
2020-10-05 12:20:41 +07:00
cohcho
5f0f2506e8 soft_aes: fix previous optimization
Previously removed unrolled variant is faster on some CPUs
Some CPUs are faster with added unrolled variant
The best variant depends on number of threads on some CPUs
2020-10-04 14:47:58 +00:00
SChernykh
ebf259fa7c RandomX: removed rx/loki
Loki forks to PoS on October 9th.
2020-10-02 17:02:52 +02:00
XMRig
d45bb24a32
Renamed WITH_SSE to WITH_SSE4_1 and make it work on all platforms. 2020-10-01 11:00:08 +07:00
SChernykh
7b4f768114 RandomX: optimized soft AES code
Unrolled loop was 5-10% slower depending on CPU.
2020-09-29 21:22:11 +02:00
xmrig
dfab81e9fa
Merge pull request #1858 from SChernykh/dev
RandomX: removed duplicate constants in Blake2b
2020-09-27 16:51:03 +07:00
SChernykh
3025c265e8 RandomX: removed duplicate constatns in Blake2b 2020-09-27 11:50:08 +02:00
xmrig
ee603ab9e2
Merge pull request #1857 from SChernykh/dev
RandomX: isolate SSE4.1 code to fix crashes on old CPUs
2020-09-27 16:47:56 +07:00
SChernykh
84f8a0dc54 RandomX: isolate SSE4.1 code to fix crashes on old CPUs 2020-09-27 11:46:32 +02:00
cohcho
9be3b69109 soft_aes: fix previous optimization
the best order of hash/fill/prefetch depends on hw/soft AES
only hw AES is faster after previous optimization
2020-09-25 15:26:19 +00:00
SChernykh
1e26e58660 Fix for ARM compilation 2020-09-23 11:44:08 +02:00
SChernykh
9768bf65d1 RandomX improved performance of GCC compiled binaries
JIT compilator was slower compared to MSVC compiled binary. Up to +0.1% speedup on rx/wow in Linux.
2020-09-22 13:48:11 +02:00
SChernykh
891a46382e RandomX: AES improvements
- A bit faster hardware AES code when compiled with MSVC
- More reliable software AES benchmark
2020-09-21 17:51:08 +02:00
SChernykh
c7476e076b RandomX refactoring, moved more stuff to compile time
Small x86 JIT compiler speedup.
2020-09-18 20:51:25 +02:00
SChernykh
8d1168385a RandomX: returned old soft AES impl and auto-select between the two 2020-09-15 20:48:27 +02:00
SChernykh
a05393727c RandomX: added performance profiler (for developers)
Also optimized Blake2b SSE4.1 code size to avoid code cache pollution.
2020-09-12 23:07:52 +02:00
SChernykh
4a9db89527 RandomX: added SSE4.1-optimized Blake2b
+0.15% on `rx/0`
+0.3% on `rx/wow`
2020-09-10 14:28:40 +02:00
SChernykh
a84b45b1bb RandomX: added parameter for scratchpad prefetch mode
`scratchpad_prefetch_mode` can have 4 values:
0: off
1: use `prefetcht0` instruction (default, same as previous XMRig versions)
2: use `prefetchnta` instruction (faster on Coffee Lake and a few other CPUs)
3: use `mov` instruction
2020-09-04 16:16:07 +02:00
XMRig
72c8404d18
Fix compile warnings. 2020-08-24 10:04:46 +07:00
XMRig
3e4bf8cd6c
Fix compile warning 2020-08-17 06:08:14 +07:00
XMRig
00b4ae9c36
Fixed compile warning and updated build.uv.sh. 2020-08-16 16:03:27 +07:00
SChernykh
5926dee354 RandomX JIT: optimized address mask calculation 2020-08-12 16:45:16 +02:00
SChernykh
5bc89fdc8b Fixed RandomX initialization for VS debug builds 2020-07-21 10:10:07 +02:00
SChernykh
3d740e81a2 RandomX: tweaked Ryzen code
Very small speedup
2020-07-05 16:06:59 +02:00
XMRig
b34e3e1a7b
Remove unused code. 2020-05-04 02:07:38 +07:00
SChernykh
80d944bf82 Optimized RandomX dataset initialization
- Use single Argon2 implemenation
- Auto-select the fastest Argon2 implementation for RandomX
2020-05-03 20:44:59 +02:00
XMRig
c18478a6b4
Small cleanups. 2020-05-03 13:38:34 +07:00
SChernykh
bfd017d064 Refactored CFROUND 2020-04-21 15:44:04 +02:00
SChernykh
abb3340cc7 RandomX JIT refactoring
- Smaller memory footprint
- A bit faster overall
2020-04-09 14:24:54 +02:00
SChernykh
92810ad761 Fixed VM destruction 2020-04-08 08:31:53 +02:00
SChernykh
39bd3ca1da Fix off-by-one error 2020-04-07 18:53:08 +02:00
SChernykh
4d0edde66d Fixed pool lock 2020-04-07 18:48:02 +02:00
SChernykh
69cbfd682a Use node number instead of affinity 2020-04-07 18:46:22 +02:00
SChernykh
6ae37a9519 Pooled allocation of RandomX VMs
+0.5% speedup on Zen2 when the whole L3 cache is used (16 threads on 3700X/3800X, 32 threads on 3950X).
2020-04-07 18:31:35 +02:00
kevacoin
0528ccd01e Added Keva. 2020-03-04 16:23:33 -08:00
XMRig
616c52f266
#1572 Fix compile warning. 2020-03-01 11:59:53 +07:00
SChernykh
131085be80 Optimized CFROUND
Shorter version using BMI2 instructionns
2020-02-21 19:00:58 +01:00