 mlemanczyk
|
c413ee1162
|
Revert "Stream handling simplified to avoid dictionary lookups in favor of array lookups with known indexes."
This reverts commit a84445ef2a.
|
2025-11-16 17:14:48 +01:00 |
|
|
mlemanczyk
|
a84445ef2a
|
Stream handling simplified to avoid dictionary lookups in favor of array lookups with known indexes.
|
2025-11-16 15:01:51 +01:00 |
|
|
mlemanczyk
|
0540411cad
|
Accelerator index added for dictionary-less use-cases
|
2025-11-15 02:21:45 +01:00 |
|
|
mlemanczyk
|
0d2a88c45f
|
GpuPrimeWorkLimiter.Acquire() / .Release() removed to save time on the hot-path with billions of calls. We'll rely on AcceleratorStreamPool limits and synchronization.
|
2025-11-13 17:52:13 +01:00 |
|
|
mlemanczyk
|
43863d03fb
|
Performance improvements
|
2025-11-13 00:51:32 +01:00 |
|
|
mlemanczyk
|
d0b1639a5d
|
GPU stabilization improvements to prevent calculation crashes
|
2025-11-09 21:52:16 +01:00 |
|
|
mlemanczyk
|
8e779c5c9e
|
GpuKernelLease removed
|
2025-11-07 15:36:45 +01:00 |
|
|
mlemanczyk
|
0410fd2c91
|
PooledContext removal - continuation. GpuKernelLease prepared for removal.
|
2025-11-07 14:13:11 +01:00 |
|
|
mlemanczyk
|
f4e365194d
|
GpuContextPool removed, because we share context and accelerator between threads, using streams to separate.
|
2025-11-07 12:52:20 +01:00 |
|
|
mlemanczyk
|
119690086a
|
ConcurrentDictionary removed. Kernel made dependent on the shared GPU context.
|
2025-11-07 10:16:33 +01:00 |
|
|
mlemanczyk
|
9537addf38
|
ArrayPool<T> replaced with ThreadStaticPools
|
2025-11-07 00:46:16 +01:00 |
|
|
mlemanczyk
|
de2a78b27b
|
Streams added to all kernels and GPU operations to separate threads from each other.
|
2025-11-06 18:53:55 +01:00 |
|
|
mlemanczyk
|
64d615c402
|
Various optimizations to separate non-static data between threads and resolve read/write access issue.
|
2025-11-04 20:45:56 +01:00 |
|
|
Marek Lemańczyk
|
de7ddd8db3
|
Remove GPU context creation lock
|
2025-10-28 13:26:20 +01:00 |
|
|
Marek Lemańczyk
|
d20362a2fc
|
Switch GPU by-divisor stepper to standard modular arithmetic
|
2025-10-24 09:20:17 +02:00 |
|
|
Marek Lemańczyk
|
8acccce471
|
Add cycle cache hit tracking instrumentation
|
2025-10-20 19:28:06 +02:00 |
|
|
Marek Lemańczyk
|
6217ea7888
|
Document GPU kernel audit progress
|
2025-10-18 17:42:33 +02:00 |
|
|
Marek Lemańczyk
|
8428e6c5bd
|
Clarify skip heuristic guard and tighten order factor init
|
2025-10-17 15:05:11 +02:00 |
|
|
mlemanczyk
|
56e413064a
|
Using cleanup
|
2025-10-17 02:40:51 +02:00 |
|
|
Marek Lemańczyk
|
cd1debe9d6
|
Various micro- and medium- optimizations
|
2025-10-07 22:14:34 +02:00 |
|
|
Marek Lemańczyk
|
26cc574020
|
Treat Pollard Rho success as the GPU primality check
|
2025-10-05 06:26:09 +02:00 |
|
|
Marek Lemańczyk
|
56e3a7695f
|
Tiny cleanup
|
2025-10-03 19:36:32 +02:00 |
|
|
Marek Lemańczyk
|
ebe48d5f2f
|
Divisor-Cycle Cache Rework
|
2025-10-02 19:53:09 +02:00 |
|
|
Marek Lemańczyk
|
b8aa845c5c
|
Clarify cycle stepper reset inlining
|
2025-10-02 13:30:56 +02:00 |
|
|
Marek Lemańczyk
|
9b9c6959be
|
Reuse cycle remainder stepper in by-divisor testers
|
2025-09-25 18:51:47 +02:00 |
|
|
Marek Lemańczyk
|
8b95a42f27
|
Divisor cycle cache will now respect --divisor-cycles-batch parameter (renamed from --divisor-cycles-batch-size)
|
2025-09-25 07:02:44 +02:00 |
|
|
Marek Lemańczyk
|
49fcd93747
|
GPU logic updated to use prime % divisor result precomputed on CPU instead of re-calculating it on GPU side. Cycle lease removed.
|
2025-09-25 05:28:23 +02:00 |
|
|
Marek Lemańczyk
|
b58e2be2b0
|
GPU steps adjusted to prevent overloading it
|
2025-09-25 03:32:07 +02:00 |
|
|
Marek Lemańczyk
|
6717005529
|
Chunk GPU divisor cycle generation to avoid CPU fallback
|
2025-09-25 03:16:49 +02:00 |
|
|
Marek Lemańczyk
|
ad3effa5fa
|
Throttle GPU cycle generation batches
|
2025-09-25 02:48:14 +02:00 |
|
|
Marek Lemańczyk
|
b798970ce9
|
Restore cycle-aware divisor scans and background generation
|
2025-09-25 02:12:40 +02:00 |
|
|
Marek Lemańczyk
|
d72cfc9748
|
TODOs added
|
2025-09-25 00:02:19 +02:00 |
|
|
Marek Lemańczyk
|
a8588d0ec5
|
Formatting
|
2025-09-24 23:33:37 +02:00 |
|
|
Marek Lemańczyk
|
d4dcd0a534
|
Implement multi-tier divisor cycle caching and per-thread reuse
|
2025-09-24 19:04:07 +02:00 |
|