西风逍遥游
811eff1c3a
ci(quality): warnings-as-errors gate, sanitizers, smoke, Python lint/tests
...
- EVENGINE_ENABLE_STRICT_WARNINGS option: MSVC /W4 /WX, GCC/Clang
-Wall -Wextra -Werror, with documented noise suppressions and
-isystem/-external:I for third-party headers; enabled on the
Windows/Linux debug CI jobs.
- Fix warnings across engine, modules and tests exposed by the gate,
including real bugs (incomplete-type deletes, Timespan microsecond
overflow, overload hiding, enum switch gaps, dead code).
- Add Linux ASan+UBSan job, example smoke step, python-tools test
job, ruff config (scripts+tools), weekly Linux coverage job and a
CodeQL workflow.
2026-08-22 02:27:37 +08:00
西风逍遥游
cff9bb4d0c
tensor: AITemplate-style compile pipeline, gpgpu Sequence, mini-LLM demo and weight-only quantization
...
Refactor the tensor module around AITemplate-style static compilation:
- Optimizer: DCE, constant folding, elementwise chain fusion, matmul/conv
bias+activation epilogue fusion, liveness-based static memory planning,
topological group execution order.
- KernelGen: shape-baked GLSL per fused group; two-pass softmax/layernorm;
shared-memory SDPA; naive + 16x16 tiled matmul templates.
- GpuBackend: arena buffer allocation per memory plan, matmul autotuning,
single command-buffer submission via the new gpgpu::Sequence (uploads,
dispatches and downloads recorded once per run, Kompute-style).
- shaderc in-process GLSL->SPIR-V on Windows (no glslc.exe dependency),
with glslc spawn fallback.
- CPU reference kernels shared by eager and the compiled interpreter;
no-Vulkan fallback keeps compiled graphs runnable headless.
Neural-network coverage: softmax/logsoftmax, layernorm/rmsnorm (with
weight/bias), gelu/silu, conv1d/conv2d, pool, embedding, concat/slice,
axis reduces, argmax, cast, fused SDPA (+masked), resize2d, batched
matmul, seeded random.
Weight-only quantization (activations stay fp32):
- fp16 (IEEE half), fp8 (e4m3), fp4 (e2m1), int8, int4 with per-group
block scales (MXFP-style: scale = maxAbs / format max magnitude).
- Quantized weights capture into graph Const nodes; CPU interpreter
dequantizes at materialization; GPU matmul/embedding kernels dequantize
inline from packed uint buffers (naive matmul variant).
- TF::quantizeWeight API + script binding.
Mini-LLM validation (TinyStories-gpt2-3M, 8 layers d=64):
- test/tensor_llm.cpp builds the full GPT-2 graph from engine ops and
matches numpy reference: probs within 3e-7, greedy generation 24/24
identical (~100 ms/token CPU), compiled path exact.
- tensor.llm.quantizedDtypes: fp16/int8/fp8/int4 all keep 16/16 greedy
top-1 (fp4 16/16 with proper block scaling), weights 27.6 MB -> 3.3-13 MB.
- scripts/export_tiny_gpt2.py converts a PyTorch .bin checkpoint (no torch
dependency) plus GPT-2 tokenizer into engine assets + numpy reference.
Tests: gpgpu 8/8, tensor 31/31 (original suite) + tensor.llm.* + tensor.quant.*
Assets in test/assets/mini_llm are gitignored (regenerate via the script).
2026-08-18 17:42:38 +08:00
西风逍遥游