jgrusewski 1fae917c22 perf(cuda): H100 kernel optimizations — nvcc pipeline, kernel fusion, GPU-only training ...

- Migrate from NVRTC JIT to cached nvcc -O3 for all CUDA kernels
- Fuse guard kernels, increase prefetch chunk, eliminate per-step GPU alloc
- H100-specific: fused Adam, warp reductions, shmem tiling, PPO occupancy
- Vectorize gather_states with __ldg() and 4x unroll
- sincosf() Box-Muller + paired Gaussian generation in noisy nets
- Shared-memory tiled branching DQN forward pass for sm_<90
- GPU-resident training guard kernel replacing Candle tensor ops
- Eliminate all to_vec1/to_vec2 CPU roundtrips, DtoD weight copy
- GPU PER mandatory everywhere — kill CPU replay path on CUDA
- Full GPU action masking — eliminate CPU fallback path
- Fix cuBLAS handle sharing via OnceLock (root cause of 49 cascade failures)
- Fix ILLEGAL_ADDRESS: scratch1_dist buffer overflow, stack sizing, curand determinism
- Fix CudaStream lifetime: bind before .context() to extend lifetime
- Keep raw cudarc buffers alive across epochs
- Add gpu-hotpath-guard.sh (37 patterns) and ptx-cache-invalidate.sh

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

2026-03-15 11:58:40 +01:00

..

argo-compile-deploy.sh

feat(infra): Argo workflow templates, drop cuDNN, GPU hotpath fixes

2026-03-12 01:44:03 +01:00

argo-test.sh

scripts: add argo-test.sh CLI wrapper for GPU test workflow

2026-03-13 11:56:56 +01:00

argo-train.sh

feat(infra): Argo workflow templates, drop cuDNN, GPU hotpath fixes

2026-03-12 01:44:03 +01:00

build_docker_images.sh

…

check-gpu-availability.sh

fix(ml): cast Bellman target to F32 before TD error sub on BF16 GPUs

2026-03-04 15:29:15 +01:00

deploy-secrets.sh

scripts: add deploy-secrets.sh for Scaleway Secrets Manager integration

2026-03-01 23:41:13 +01:00

devpod-setup.sh

…

docker-init-migrations.sh

…

enforce_coverage.sh

refactor: update K8s, CI, Docker, Prometheus, scripts, and FXT CLI for api rename

2026-03-04 23:46:36 +01:00

entrypoint-generic.sh

…

generate_dev_certs.sh

refactor: update K8s, CI, Docker, Prometheus, scripts, and FXT CLI for api rename

2026-03-04 23:46:36 +01:00

generate-training-pipeline.sh

feat(ml): add manually-triggered GitLab CI training pipeline

2026-02-26 09:04:58 +01:00

gpu-hotpath-guard.sh

perf(cuda): H100 kernel optimizations — nvcc pipeline, kernel fusion, GPU-only training

2026-03-15 11:58:40 +01:00

gpu-hotpath-hook.sh

feat(dqn): enable Branching DQN with 45 factored actions (5×3×3)

2026-03-11 22:00:13 +01:00

health-check.sh

…

init-db-dev.sql

perf(ci): compile once with PVC sccache, package with Kaniko

2026-02-26 00:50:25 +01:00

init-db.sql

perf(ci): compile once with PVC sccache, package with Kaniko

2026-02-26 00:50:25 +01:00

pre-commit-hook.sh

feat(dqn): enable Branching DQN with 45 factored actions (5×3×3)

2026-03-11 22:00:13 +01:00

prepare-sqlx-offline.sh

perf(ci): compile once with PVC sccache, package with Kaniko

2026-02-26 00:50:25 +01:00

ptx-cache-invalidate.sh

perf(cuda): H100 kernel optimizations — nvcc pipeline, kernel fusion, GPU-only training

2026-03-15 11:58:40 +01:00

refresh-test-data.sh

infra: add populate-test-data job and refresh script

2026-03-13 11:54:22 +01:00

setup-database.sh

refactor: update scripts for kebab-case service binary names

2026-03-04 22:08:31 +01:00

start-fxt.sh

…

start.sh

refactor: update scripts for kebab-case service binary names

2026-03-04 22:08:31 +01:00

stop.sh

refactor: update scripts for kebab-case service binary names

2026-03-04 22:08:31 +01:00

validate_data_quality.sh

refactor: update scripts for kebab-case service binary names

2026-03-04 22:08:31 +01:00