animica: add SHA3-256 CUDA kernel + animicaHash plugin export#218
Open
animicaorg wants to merge 1 commit into
Open
animica: add SHA3-256 CUDA kernel + animicaHash plugin export#218animicaorg wants to merge 1 commit into
animicaorg wants to merge 1 commit into
Conversation
This is the xmrig-cuda counterpart to the Animica integration in ercmine/xmrig@animica-mining. With this plugin loaded, an xmrig built with -DWITH_ANIMICA=ON resolves the `animicaHash` symbol via dlsym, and `--algo animica` mines on the GPU via CUDA. What's in this commit: - src/Animica/Animica.cu — the kernel. One CUDA thread per nonce: builds the single 136-byte rate block (32-byte prefix + 8-byte LE nonce + 0x06 ... 0x80 SHA3 padding), runs Keccak-f[1600] for 24 rounds inlined in registers, squeezes the first 32 bytes, and big-endian-compares against the 256-bit target. Atomic-reserves a slot in a small results buffer when the digest is ≤ target. - src/Animica/Animica.h — single-function dispatcher signature matching the CudaLib::animicaHash() symbol xmrig dlsym's. - src/xmrig-cuda.h / src/xmrig-cuda.cpp — `XMRIG_EXPORT bool animicaHash(...)` exposed at the plugin boundary, gated by XMRIG_ALGO_ANIMICA. setJob() for the ANIMICA family is a deliberate no-op — the kernel reads job_blob directly per call, no device-side scratch carried between turns. - src/cryptonight.h — forward declaration of Animica::hash() so the rest of the plugin compiles without including the .cu header. - src/crypto/common/Algorithm.h / .cpp — adds ANIMICA_SHA3 (0x41010000) to the algorithm enum and the parse table, mirroring the encoding chosen in ercmine/xmrig. - CMakeLists.txt + cmake/CUDA.cmake — `WITH_ANIMICA=ON` build flag (defaults on), CUDA_ANIMICA_SOURCES list, and the XMRIG_ALGO_ANIMICA preprocessor define plumbed alongside KAWPOW / RANDOMX. What the host xmrig sees: - On plugin dlopen, the optional `uv_dlsym(animicaHash)` in CudaLib succeeds (was a no-op against stock xmrig-cuda). - CudaLib::hasAnimicaSupport() returns true. - CudaAnimicaRunner::run() dispatches into this kernel and reports any winning nonce back to the worker submit path. Not in this commit (intentional): - No GPU build verified on this host — the dev box has no CUDA toolkit installed. The .cu compiles against nvcc on a CUDA-equipped builder is the next-step manual test, after which the dlopen + Animica end-to-end against pool.animica.org:3333 is the same exercise the CPU/OpenCL paths already pass.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds the CUDA SHA3-256 kernel for Animica (Layer-1 hashshare PoW) — companion to xmrig/xmrig#3817.
With this plugin loaded, an xmrig built with
-DWITH_ANIMICA=ONresolves theanimicaHashsymbol viadlsym, andxmrig --algo animica --url pool.animica.org:3333mines on the GPU via CUDA. Without it (e.g. stock upstream xmrig-cuda), xmrig'sCudaLib::hasAnimicaSupport()probe returns false, the runner emits one yellow warning, and mining falls back cleanly to OpenCL/CPU — no segfault, no broken state.What's in this PR
Kernel (
src/Animica/Animica.cu)0x06 ... 0x80SHA3 padding), runs Keccak-f[1600] for 24 rounds inlined in registers, squeezes the first 32 bytes, big-endian-compares against the 256-bit target.AnimicaHash.cpp::sha3_256(which produces bit-identical output by construction; same round constants, same0x06/0x80padding).device_id(capped at 16 GPUs/host).Plugin export (
src/xmrig-cuda.{h,cpp})XMRIG_EXPORT bool animicaHash(nvid_ctx*, uint8_t*, uint64_t, uint32_t, uint32_t*, uint32_t*, uint32_t*)matches the signature xmrig'sCudaLibuv_dlsym's.setJob()for the ANIMICA family is a deliberate no-op — the kernel readsjob_blobdirectly per call, no device-side scratch carried between turns (in contrast to KawPow's DAG or RandomX's dataset).Algorithm registration (
src/crypto/common/Algorithm.{h,cpp})Algorithm::ANIMICA_SHA3 = 0x41010000+Family::ANIMICA = 0x41000000, mirroring the encoding in the companion xmrig PR.XMRIG_ALGO_ANIMICA.Build glue (
CMakeLists.txt,cmake/CUDA.cmake)option(WITH_ANIMICA "..." ON).XMRIG_ALGO_ANIMICAdefine plumbed alongsideXMRIG_ALGO_KAWPOW.CUDA_ANIMICA_SOURCESlist with the .cu + .h.Footprint
When
WITH_ANIMICA=OFFthe binary is byte-identical to upstream. Animica adds ~300 lines including the kernel; no new dependencies (uses only the CUDA Runtime API and the existing xmrig-cudanvid_ctx).Companion PR
CPU + OpenCL paths, algorithm enum, Stratum + AICF dispatch in xmrig: xmrig/xmrig#3817.
Build verification
The kernel hasn't been benchmarked on hardware yet — the dev box has no CUDA toolkit installed. Compilation against
nvccand a real-GPU run againstpool.animica.org:3333is the next-step exercise, after which the xmrig + xmrig-cuda + Animica pool end-to-end is the same one the CPU and OpenCL paths already pass.