build(deps): bump chia from 0.42.0 to 0.42.1 in /keygen-rs#6

Open

dependabot[bot] wants to merge 208 commits intomainfrom

dependabot/cargo/keygen-rs/chia-0.42.1

Commits on Apr 20, 2026

Add streaming pipeline for low-VRAM GPUs (fits under 8 GB at k=28)

Abraham Sewill
and
claude
committed
README: document streaming (≤8 GB) pipeline and its automatic dispatch

Abraham Sewill
and
claude
committed
Parallelize compute_bucket_offsets and drop the l_count_max host fence

Abraham Sewill
and
claude
committed
README: explain CUDA autodetect and the fat-build fallback

Abraham Sewill
and
claude
committed
README: hoist Hardware compatibility, move Performance to the bottom

Abraham Sewill
and
claude
committed
Bump pinned-slot count to 3, batch channel to depth 2

Abraham Sewill
and
claude
committed
Fixed claude's typos.
Abraham Sewill
committed
Port to SYCL/AdaptiveCpp; CUDA backend opt-in via XCHPLOT2_BUILD_CUDA

Abraham Sewill
and
claude
committed
Stable parallel SYCL radix sort for non-CUDA builds + parity test

Abraham Sewill
and
claude
committed

Commits on Apr 21, 2026

GpuBufferPool: include xs_temp_bytes in pair_bytes

Abraham Sewill
and
claude
committed
Auto-detect ACPP_TARGETS in CMake and build.rs

Abraham Sewill
and
claude
committed
README: link to cuda-only branch for NVIDIA-only users

Abraham Sewill
and
claude
committed
README: refresh Performance numbers post-SYCL port

Abraham Sewill
and
claude
committed
README: list AdaptiveCpp + auto-fetched + optional runtime deps

Abraham Sewill
and
claude
committed
Add Containerfile + install-deps.sh + FetchContent fallback for AdaptiveCpp

Abraham Sewill
and
claude
committed
Containerfile: end-to-end NVIDIA build + bit-identical plot output

Abraham Sewill
and
claude
committed
README: bump pool VRAM threshold to ~17 GB free / 18 GB+ cards

Abraham Sewill
and
claude
committed
README: add WIP status note up top

Abraham Sewill
and
claude
committed
build.rs: read AdaptiveCpp lib dir from CMake instead of hardcoding

Abraham Sewill
and
claude
committed
GpuBufferPool + streaming pipeline: free-on-throw, clearer OOM message

Abraham Sewill
and
claude
committed
SortSycl: ping-pong over caller buffers, drop internal alt allocation

Abraham Sewill
and
claude
committed
SortCuda: switch to CUB DoubleBuffer mode to match SortSycl scratch profile

Abraham Sewill
and
claude
committed
Add compose.yaml + bundle parity tests in container image

Abraham Sewill
and
claude
committed
Add scripts/build-container.sh — host-side GPU autodetect for compose

Abraham Sewill
and
claude
committed
Bump version to 0.2.0

Abraham Sewill
and
claude
committed
Containerfile: parametrize LLVM root for AMD/ROCm bitcode compatibility

Abraham Sewill
and
claude
committed
scripts: install rocminfo on AMD path; better no-GPU error

Abraham Sewill
and
claude
committed
build-container.sh: capture rocminfo/nvidia-smi output before grep

Abraham Sewill
and
claude
committed
build-container.sh: SIGPIPE fix in gfx detection (was killing the script)

Abraham Sewill
and
claude
committed
gitignore docs/ — internal design notes, never user-facing

Abraham Sewill
and
claude
committed
Containerfile: skip Ubuntu llvm-18 on AMD path; add LLVM linkage diag

Abraham Sewill
and
claude
committed
compose: pin ROCm to 6.2 + drop LLVM_ROOT override

Abraham Sewill
and
claude
committed
compose: use rocm/dev-ubuntu-24.04:6.2-complete for the rocm service

Abraham Sewill
and
claude
committed
Conditionalize cuda_fp16.h via CudaHalfShim — fixes AMD/HIP build clash

Abraham Sewill
and
claude
committed
Guard cuda_runtime.h via CudaHalfShim; symlink clang-offload-bundler on ROCm

Abraham Sewill
and
claude
committed
Drop raw cuda_fp16.h from GpuPipeline; fan-out bundler symlinks

Abraham Sewill
and
claude
committed
install-deps: drop CUDA headers from AMD paths

Abraham Sewill
and
claude
committed
build.rs: gate cudart/cudadevrt link on XCHPLOT2_BUILD_CUDA=ON

Abraham Sewill
and
claude
committed
build.rs: autodetect XCHPLOT2_BUILD_CUDA from nvcc availability

Abraham Sewill
and
claude
committed
build.rs: link libamdhip64 when ACPP_TARGETS targets HIP

Abraham Sewill
and
claude
committed
CMakeLists: move plot_file_parity out of CUDA-gated block

Abraham Sewill
and
claude
committed
compose + README: document AMD rootless seccomp/cap requirements

Abraham Sewill
and
claude
committed

Commits on Apr 22, 2026

README: mark AMD ROCm path as validated end-to-end

Abraham Sewill
and
claude
committed
install-deps: pin AdaptiveCpp to LLVM 16-20; ROCm device libs detect

Abraham Sewill
and
claude
committed
GpuBufferPool: split d_pair_a / d_pair_b sizing — saves ~2-3 GB at k=28

Abraham Sewill
and
claude
committed
GpuPipeline: replace stubbed phase timers with chrono-based wall timing

Abraham Sewill
and
claude
committed
XsKernel: per-sub-phase wall timing (gen / sort / pack) under env flag

Abraham Sewill
and
claude
committed
CMakeLists: force -O3 on SYCL TUs so AdaptiveCpp's acpp doesn't AOT at -O0

Abraham Sewill
and
claude
committed
Revert: target_compile_options(-O2/-O3) on pos2_gpu broke AMD parity

Abraham Sewill
and
claude
committed
Containerfile: install clang-18 + libclang-cpp18 + libomp-18-dev in runtime

Abraham Sewill
and
claude
committed
Containerfile: ship builder stage as runtime — pragmatic correctness fix

Abraham Sewill
and
claude
committed
compose + build script: fail loud on missing ACPP_GFX (root cause of AMD bugs)

Abraham Sewill
and
claude
committed
Containerfile: restore slim runtime stage (~1 GB image-size win)

Abraham Sewill
and
claude
committed
README: document AMD container's sudo + privileged + ACPP_GFX trifecta

Abraham Sewill
and
claude
committed
CMakeLists: re-enable -O3 for SYCL TUs (was wrongly blamed for ACPP_GFX bug)

Abraham Sewill
and
claude
committed
gpu: port bitsliced AES to SYCL for sub_group-cooperative hashing

Abraham Sewill
and
claude
committed
gpu: portable attrs on bp_sbox_circuit so SYCL TUs can include it

Abraham Sewill
and
claude
committed
gpu: per-thread coarsening for Xs gen + T1/T2/T3 match kernels

Abraham Sewill
and
claude
committed
gpu: revert per-thread coarsening (net loss at k=28 on RX 6700 XT)

Abraham Sewill
and
claude
committed
gpu: instrument streaming-path with [phase-timing] output

Abraham Sewill
and
claude
committed
gpu: drop unused d_keys_in slot in d_storage — pool now fits 12 GiB

Abraham Sewill
and
claude
committed
gpu: lazy pinned-host alloc in GpuBufferPool — single-plot saves ~1.2 s

Abraham Sewill
and
claude
committed
gpu: wire bitsliced AES through native __builtin_amdgcn_ballot_w32

Abraham Sewill
and
claude
committed
gpu: switch bs_ballot to __acpp_if_target_hip for host-pass safety

Abraham Sewill
and
claude
committed
gpu: revert bitsliced wiring (round 2) — native ballot was necessary but not sufficient

Abraham Sewill
and
claude
committed
gpu: lazy d_pair_a alloc overlapping with Xs gen — saves first-plot wall

Abraham Sewill
and
claude
committed
gpu: free d_pair_a between plots in batch — smaller-card pool compat

Abraham Sewill
and
claude
committed

Commits on Apr 23, 2026

gpu: split xs-sort keys_a to d_storage tail — drops pool VRAM min ~1.3 GB

Abraham Sewill
and
claude
committed
docs: tighten streaming peak (~7.3 GB measured), add AMD row, fix VRAM query note

Abraham Sewill
and
claude
committed
added a donate section to the readme.
Abraham Sewill
committed
ci: add GitHub Actions workflow (shellcheck, actionlint, Rust)

Abraham Sewill
and
claude
committed
plot-write: atomic .partial + rename; SIGINT/SIGTERM cooperative stop; verify via pos2-chip Prover

Abraham Sewill
and
claude
committed
batch+cli: skip-existing / continue-on-error / disk preflight / verify / env-var help

Abraham Sewill
and
claude
committed
docs: CONTRIBUTING + SECURITY; README env-vars table + new flags

Abraham Sewill
and
claude
committed
sycl: install async_handler on the persistent queue — clean exit on async errors

Abraham Sewill
and
claude
committed
batch: preflight streaming-path VRAM before pinned-host alloc

Abraham Sewill
and
claude
committed
batch: widen streaming preflight margin to 1 GiB — sidestep AdaptiveCpp post-OOM double-free

Abraham Sewill
and
claude
committed

Commits on Apr 24, 2026

batch: revert streaming preflight margin to 256 MB

Abraham Sewill
and
claude
committed
T2 match: plumb bucket_begin/bucket_end params (stage 1 of N)

Abraham Sewill
and
claude
committed
T2 match: streaming-path N=2 tiling (stage 2 of N)

Abraham Sewill
and
claude
committed
T2 match: half-cap device staging + D2H per pass (stage 3 of N)

Abraham Sewill
and
claude
committed
T2 match: JIT H2D d_t2_meta/xbits only for their gather calls (stage 4a of N)

Abraham Sewill
and
claude
committed
CMakeLists: set CMAKE_POSITION_INDEPENDENT_CODE ON globally

Abraham Sewill
and
claude
committed
T1 sort: park d_t1_meta on pinned host across sort phase (stage 4b of N)

Abraham Sewill
and
claude
committed
batch: update streaming peak anchor to 6240 MB, trim preflight margin to 128 MB (stage 5)

Abraham Sewill
and
claude
committed
T1+T2 sort: park *_keys_merged on pinned host across gather peaks (stage 4c of N)

Abraham Sewill
and
claude
committed
T3 match: plumb bucket_begin/bucket_end params (stage 4d.1 of N)

Abraham Sewill
and
claude
committed
T3 match: streaming-path N=2 tiling + prepare/range refactor (stage 4d.2 of N)

Abraham Sewill
and
claude
committed
T3 match: half-cap d_t3 staging + D2H per pass (stage 4d.3 of N)

Abraham Sewill
and
claude
committed
Xs phase: inline gen+sort+pack, free keys_a/vals_a after sort (stage 4e of N)

Abraham Sewill
and
claude
committed
batch: re-anchor streaming_peak to 5200 MB after stage 4e (stage 5 redux)

Abraham Sewill
and
claude
committed
batch: amortise streaming pinned-host scratch across plots (stage 4f of N)

Abraham Sewill
and
claude
committed
streaming: add plain tier (skip parks + single-pass T2 match)

Abraham Sewill
and
claude
committed
readme: document 3-tier streaming dispatch (pool | plain | compact)

Abraham Sewill
and
claude
committed
T3 match: one-shot full-cap in plain tier (skip N=2 staging + h_t3)

Abraham Sewill
and
claude
committed
readme: document experimental Windows build path (NVIDIA/MSVC)
Abraham Sewill
committed
ci: fix actionlint deprecation + shellcheck warnings
Abraham Sewill
committed
Bump version to 0.3.0
Abraham Sewill
committed
batch: multi-GPU via --devices flag (thread-per-device)
Abraham Sewill
committed
scripts: add test-multi-gpu.sh smoke test for --devices
Abraham Sewill
committed
readme: document --devices multi-GPU flag + test-multi-gpu.sh
Abraham Sewill
committed
Bump version to 0.4.0
Abraham Sewill
committed
scripts: fix shellcheck SC2002 in test-multi-gpu.sh (useless cat)
Abraham Sewill
committed
readme: note multi-GPU scaling in perf + add XCHPLOT2_TEST_GPU_COUNT env var
Abraham Sewill
committed
readme: add Quick start + VRAM → Multi-GPU cross-link
Abraham Sewill
committed
readme: tighten status + branches blockquotes for scannability
Abraham Sewill
committed
readme: document Radeon Pro W5700 / RDNA1 gfx1013 spoof workaround
Abraham Sewill
committed
amd: autodetect RDNA1 (gfx1010/1011/1012) → gfx1013 spoof
Abraham Sewill
committed
scripts: test-multi-gpu.sh — bypass keygen via batch manifest
Abraham Sewill
committed
cli: xchplot2 parity-check subcommand
Abraham Sewill
committed
parity: extract derive_plot_id + Stats/compare to ParityCommon.hpp
Abraham Sewill
committed
build.rs: prefer GPU-vendor detection over nvcc-presence
Abraham Sewill
committed
cmake: force-include cuda_fp16.h in every CUDA TU
Abraham Sewill
committed
install-deps: skip AdaptiveCpp's CUDA probe on --gpu amd builds
Abraham Sewill
committed
readme: Windows → cuda-only branch + VS SDK + LIB troubleshooting
Abraham Sewill
committed
readme: collapse Windows section to a 2-path note
Abraham Sewill
committed
readme: fix OS bullet — match collapsed Windows section (anchor + text)
Abraham Sewill
committed
readme: add Native Windows build walkthrough under the viable-paths note
Abraham Sewill
committed
build.rs: gate NVIDIA auto-detect on sm_61 minimum (our README floor)
Abraham Sewill
committed
readme: add native Windows SYCL build section (adventurous path)
Abraham Sewill
committed
Bump version to 0.5.0
Abraham Sewill
committed
readme: micro-polish — Windows-section nudge + sm_61 auto-detect note
Abraham Sewill
committed
scripts: cross-device parity test in test-multi-gpu.sh
Abraham Sewill
committed

Commits on Apr 25, 2026

cmake: auto-find /opt/adaptivecpp + auto-probe ld.lld for FetchContent
Abraham Sewill
committed
install-deps: add lld to apt + dnf package lists
Abraham Sewill
committed
Bump version to 0.5.1
Abraham Sewill
committed

Commits on Apr 26, 2026

install-deps: two-tier GPU detection + fail-fast on no GPU
Abraham Sewill
committed
readme: install-deps.sh + LLVM rows reflect recent behaviour changes
Abraham Sewill
committed
build.rs: preflight critical system deps before invoking cmake
Abraham Sewill
committed
readme: four small clarifications from the latest audit
Abraham Sewill
committed
sort: split nvcc/SYCL boundary so .cu files don't reach sycl.hpp
Abraham Sewill
committed
Bump version to 0.5.2
Abraham Sewill
committed
cmake: stub cli_devlink.cu to fix cargo install device-link
Abraham Sewill
committed
notice: add AdaptiveCpp, AMD ROCm/HIP, Intel oneAPI sections
Abraham Sewill
committed
container: pin CUDA 12.9 base for pre-Turing GPUs (Pascal/Volta)

Abraham Sewill
and
claude
committed
container: fat binary for mixed-GPU rigs (1070 + 3060, etc.)

Abraham Sewill
and
claude
committed
build: preflight nvcc/arch compatibility (CUDA 13 + Pascal/Volta)

Abraham Sewill
and
claude
committed
cmake: share pos2_gpu CUDA objects via OBJECT lib for cargo install

Abraham Sewill
and
claude
committed
cmake: avoid duplicate CUDA .o in pos2_gpu + xchplot2_cli (nvlink fix)

Abraham Sewill
and
claude
committed
container: silence rocm ACPP_GFX:? check on non-rocm builds

Abraham Sewill
and
claude
committed
cmake: pull CUDA OBJECT lib into sycl_sort_parity (CUB-adapter fix)

Abraham Sewill
and
claude
committed
container: add CPU build path (AdaptiveCpp OpenMP backend)

Abraham Sewill
and
claude
committed

Commits on Apr 27, 2026

container: in-container preflight message + --no-cache build flag

Abraham Sewill
and
claude
committed
build: link libomp when ACPP_TARGETS=omp (CPU backend)

Abraham Sewill
and
claude
committed
cpu: --cpu flag + SyclBackend dispatch (commit 2 of CPU support)

Abraham Sewill
and
claude
committed
readme: document --cpu and the cpu container service

Abraham Sewill
and
claude
committed
container: close the script-vs-compose UX gap

Abraham Sewill
and
claude
committed
readme: tldr "Pick a path" guide at top of Build section

Abraham Sewill
and
claude
committed
build: friendlier preflight when host deps are missing

Abraham Sewill
and
claude
committed
cpu: route --cpu through pos2-chip's Plotter (replaces SYCL→OMP)

Abraham Sewill
and
claude
committed
container: cuda service GPU pass-through works under Docker too

Abraham Sewill
and
claude
committed
batch: --tier plain|compact|auto CLI flag for streaming pipeline

Abraham Sewill
and
claude
committed
sycl: filter to CUDA-backend devices on CUB builds (mixed-vendor host)

Abraham Sewill
and
claude
committed
cmake: rescan link group + allow-multiple-definition on xchplot2 exe

Abraham Sewill
and
claude
committed
batch: minimal streaming tier (~3.83 GiB floor) for 4 GiB cards

Abraham Sewill
and
claude
committed
Bump version to 0.6.0

Abraham Sewill
and
claude
committed
batch: XCHPLOT2_SYCL_CPU_BENCH=1 routes --cpu through SYCL pipeline

Abraham Sewill
and
claude
committed
docs: AMD 4 GiB targets in build-container example list

Abraham Sewill
and
claude
committed
ci: split CUDA_ARCH assignment from export (shellcheck SC2155)

Abraham Sewill
and
claude
committed
cpu: accept pool-PK 128-byte memos (not just pool-PH 112-byte)

Abraham Sewill
and
claude
committed
scripts: split container host bootstrap into install-container-deps.sh

Abraham Sewill
and
claude
committed
scripts: install-container-deps.sh --dry-run + CDI-WARN explanation

Abraham Sewill
and
claude
committed
ci: install-container-deps.sh dry-run fixtures + container smoke

Abraham Sewill
and
claude
committed
ci: harden install-container-deps run.sh against CWD-dependent ROOT

Abraham Sewill
and
claude
committed
keygen-rs: cargo fmt

Abraham Sewill
and
claude
committed
ci: add dependabot + cargo-fmt + typos + markdownlint + hadolint + compose-config

Abraham Sewill
and
claude
committed
build(deps): bump actions/checkout from 5 to 6
dependabot[bot]
authored
build(deps): bump sha2 from 0.10.9 to 0.11.0 in /keygen-rs
dependabot[bot]
authored
Merge pull request #4 from Jsewill/dependabot/github_actions/actions/checkout-6
Jsewill
authored
build(deps): bump hadolint/hadolint-action from 3.1.0 to 3.3.0
dependabot[bot]
authored
build(deps): bump DavidAnson/markdownlint-cli2-action from 18 to 23
dependabot[bot]
authored
Merge pull request #5 from Jsewill/dependabot/cargo/keygen-rs/sha2-0.11.0
Jsewill
authored
Merge pull request #3 from Jsewill/dependabot/github_actions/hadolint/hadolint-action-3.3.0
Jsewill
authored
Merge pull request #2 from Jsewill/dependabot/github_actions/DavidAnson/markdownlint-cli2-action-23
Jsewill
authored

Commits on Apr 28, 2026

docs(Containerfile): correct stale LLVM_ROOT override claim

Abraham Sewill
and
claude
committed
streaming: print exact bytes (not truncated MB) on alloc failures

Abraham Sewill
and
claude
committed
streaming: add SYCL-radix scratch overhead to peak predictions

Abraham Sewill
and
claude
committed
streaming: validate t1_count + reject zero-byte allocs early

Abraham Sewill
and
claude
committed
cpu-bench: fix --cpu + XCHPLOT2_SYCL_CPU_BENCH=1 device selection

Abraham Sewill
and
claude
committed
streaming: fix 4^k scaling in peak predictions

Abraham Sewill
and
claude
committed
streaming minimal: 5200 → 3754 MB peak at k=28 (fits 4 GiB cap)

Abraham Sewill
and
claude
committed
pool: include alloc bytes + underlying err in OOM diagnostics

Abraham Sewill
and
claude
committed

Commits on Apr 29, 2026

parity: add sycl_t1_parity, broaden 0-entries diag for generic JIT

Abraham Sewill
and
claude
committed
diag: replace mis-lowering with miscompile (typos CI)

Abraham Sewill
and
claude
committed

Commits on May 4, 2026

diag: POS2GPU_T1_DEBUG=1 logs d_xs sample + t1_count around T1 match

Abraham Sewill
and
claude
committed
build: skip CUDA runtime/fp16 includes when ROCm is also present

Abraham Sewill
and
claude
committed
diag: POS2GPU_T1_DEBUG=1 — sample Xs gen/sort outputs at head/mid/tail

Abraham Sewill
and
claude
committed
diag: trivial-kernel + d_aes_tables sanity in POS2GPU_T1_DEBUG=1

Abraham Sewill
and
claude
committed
fix(amd): provide __half via hip_fp16.h fallback in CudaHalfShim

Abraham Sewill
and
claude
committed
build: embed AdaptiveCpp + ROCm rpath for plain-cmake binaries

Abraham Sewill
and
claude
committed
build: kernel-dispatch self-test at first SYCL queue construction

Abraham Sewill
and
claude
committed
build: XCHPLOT2_NO_GFX_SPOOF=1 opts out of the gfx1013 RDNA1 workaround

Abraham Sewill
and
claude
committed
tools: add hellosycl — minimal SYCL kernel-dispatch sanity check

Abraham Sewill
and
claude
committed
build: link libamdhip64 directly so AdaptiveCpp HIP backend loads

Abraham Sewill
and
claude
committed
build: default RDNA1 to ACPP_TARGETS=generic; gfx1013 spoof now opt-in

Abraham Sewill
and
claude
committed
build: link libamdhip64 whenever ROCm is reachable, not just hip:* targets

Abraham Sewill
and
claude
committed
build: add amd_gpu_present() — separate AMD detection from gfx target

Abraham Sewill
and
claude
committed
build: lower NVIDIA arch floor to sm_50 (Maxwell); fix wrong half-intrinsic claim

Abraham Sewill
and
claude
committed
docs: README hardware section reflects sm_50 floor + RDNA1 generic default

Abraham Sewill
and
claude
committed
diag: fix wrong "should be 0xC66363A5" hint in d_aes_tables dump

Abraham Sewill
and
claude
committed
docs: add Troubleshooting section covering AMD + spurious BUILD_CUDA failure modes

Abraham Sewill
and
claude
committed
docs: troubleshooting CUB entry now mentions the queue-init selftest catch

Abraham Sewill
and
claude
committed
fix(build): amd_gpu_present() falls back to /sys/class/drm vendor probe

Abraham Sewill
and
claude
committed
feat(sort): runtime backend dispatch — single binary handles NVIDIA + AMD/Intel concurrently

Abraham Sewill
and
claude
committed
feat(cli): `xchplot2 devices` lists visible GPUs + sort routing

Abraham Sewill
and
claude
committed
feat(cli): --devices all means everything; --devices gpu means all-GPUs-only

Abraham Sewill
and
claude
committed
fix(batch): work-queue dispatch — fast workers keep pulling instead of idling

Abraham Sewill
and
claude
committed
build(deps): bump chia from 0.42.0 to 0.42.1 in /keygen-rs
dependabot[bot]
authored