Ow native gpu 0.1.0 -> 0.9.9

.github/workflows/ci-build-intel.yml

@@ -1,10 +1,17 @@
 name: Build using Intel drivers
 
+# Manual-only. This workflow validates that Sibernetic still links against
+# the legacy 2018 Intel OpenCL CPU runtime (SRB5.0). It does not run the
+# binary or any tests — Intel's runtime can't expose itself as a usable
+# OpenCL device on the headless GitHub runner ("Seg faults" was the
+# original maintainer note). Kept around as a reference for anyone
+# resurrecting Intel-driver support; not run on every push/PR because it
+# adds CI minutes (sourceforge fetch of the SDK) for zero coverage beyond
+# what `Build using AMD drivers` already provides.
+#
+# Trigger via the Actions UI ("Run workflow") when explicitly needed.
 on:
-  push:
-    branches: [ master, dev*, ow* ]
-  pull_request:
-    branches: [ master, dev, ow* ]
+  workflow_dispatch:
 
 jobs:
 
@@ -14,7 +21,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        runs-on: [ ubuntu-22.04, ubuntu-24.04 ]
+        runs-on: [ ubuntu-22.04 ]
 
     steps:
     - name: Set git to use LF
@@ -56,7 +63,7 @@ jobs:
     - name: Build Sibernetic
       run: |
 
-        sudo apt install -y python3-dev freeglut3-dev libglu1-mesa-dev
+        sudo apt install -y python3-dev python3-numpy freeglut3-dev libglu1-mesa-dev
         #sudo apt install -y  --allow-downgrades libc-bin=2.27-3ubuntu1.5 # Fails with 2.27-3ubuntu1.6 for some reason...
         python -V
         ls -alt /usr/bin/python*

.github/workflows/ci-build.yml

@@ -5,6 +5,7 @@ on:
     branches: [ master, dev*, ow* ]
   pull_request:
     branches: [ master, dev, ow* ]
+  workflow_dispatch:
 
 jobs:
 
@@ -21,39 +22,20 @@ jobs:
       run: |
         git config --global core.autocrlf false
         git config --global core.eol lf
-        
+
     - uses: actions/checkout@v4
 
     - name: Install AMD OpenCL libraries needed for Sibernetic
       run: |
-
         lscpu
 
         echo "Installing OpenCL Drivers"
 
-        # Legacy install of Intel's OpenCL Drivers:
-        # Based on: https://github.com/openworm/OpenWorm/blob/master/Dockerfile
-        # mkdir intel-opencl-tmp
-        # cd intel-opencl-tmp
-        # mkdir intel-opencl
-        # wget https://github.com/openworm/OpenWorm/raw/dev_inte/SRB5.0_linux64.zip
-        # unzip SRB5.0_linux64.zip
-        # tar -C intel-opencl -Jxf intel-opencl-r5.0-63503.x86_64.tar.xz
-        # tar -C intel-opencl -Jxf intel-opencl-devel-r5.0-63503.x86_64.tar.xz
-        # tar -C intel-opencl -Jxf intel-opencl-cpu-r5.0-63503.x86_64.tar.xz
-        # sudo cp -R intel-opencl/* /
-        # sudo ldconfig
-        # cd ..
-        # sudo rm -r intel-opencl-tmp
-
-        # sudo cp -R /opt/intel/opencl/include/CL /usr/include/
-        # sudo apt install -y ocl-icd-opencl-dev vim 
-
         # Install AMD's OpenCL Drivers (AMD-APP-SDK 3.0):
         wget https://master.dl.sourceforge.net/project/nicehashsgminerv5viptools/APP%20SDK%20A%20Complete%20Development%20Platform/AMD%20APP%20SDK%203.0%20for%2064-bit%20Linux/AMD-APP-SDKInstaller-v3.0.130.136-GA-linux64.tar.bz2
         tar -xf AMD-APP-SDKInstaller-v3.0.130.136-GA-linux64.tar.bz2
         printf 'Y\n\n' | sudo ./AMD-APP-SDK-v3.0.130.136-GA-linux64.sh
-        
+
         sudo ln -s /opt/AMDAPPSDK-3.0/lib/x86_64/sdk/libOpenCL.so.1 /usr/lib/libOpenCL.so.1
         sudo ln -s /opt/AMDAPPSDK-3.0/lib/x86_64/sdk/libamdocl64.so /usr/lib/libamdocl64.so
 
@@ -66,9 +48,7 @@ jobs:
 
     - name: Build Sibernetic
       run: |
-
-        sudo apt install -y python3-dev freeglut3-dev libglu1-mesa-dev
-        #sudo apt install -y  --allow-downgrades libc-bin=2.27-3ubuntu1.5 # Fails with 2.27-3ubuntu1.6 for some reason...
+        sudo apt install -y python3-dev python3-numpy freeglut3-dev libglu1-mesa-dev
         python -V
         ls -alt /usr/bin/python*
         ls -alt
@@ -79,29 +59,113 @@ jobs:
       run: |
         ./Release/Sibernetic -h
 
-    - name: Run quick Sibernetic test
+    - name: Run quick Sibernetic OpenCL test
       run: |
         ldd ./Release/Sibernetic
-
         ./Release/Sibernetic -no_g timelimit=0.001
 
     - name: Run tests
       run: |
-        pip install c302
-        pip install neuron
+        pip install c302 neuron pytest pyneuroml
         pip list
         which nrniv
         export NEURON_HOME=/home/runner/.local
-        ./run_all_tests.sh 
+        ./run_all_tests.sh
 
     - name: Generated file info
       run: |
-
-        ls -alt  simulations/*
+        ls -alt simulations/*
         more simulations/C1_*/report.json
 
     - name: Final version info
       run: |
-
         python -V
         pip list
+
+
+  build-macos:
+    # Native Metal substrate exercise. The main Sibernetic binary on Apple
+    # Silicon builds with OW_NO_OPENCL (no OpenCL driver available); the
+    # actual GPU simulation path lives in src/metal_diff/sib_metal, a
+    # separate Apple-Metal-shader-based binary that implements XPBD on the
+    # Apple GPU. This job builds and smoke-tests both: the main binary as a
+    # build-system regression check, the Metal substrate as the real GPU
+    # path.
+    runs-on: macos-latest
+
+    steps:
+    - uses: actions/checkout@v6
+
+    - name: Environment info
+      run: |
+        sw_vers
+        uname -m
+        python3 -V
+        which python3.13 || true
+        brew --version
+        xcrun --version
+
+    - name: Build main Sibernetic binary via setup.sh
+      run: ./setup.sh
+
+    - name: Print Sibernetic help
+      run: ./Release/Sibernetic -h
+
+    - name: Print main-binary linkage
+      run: otool -L ./Release/Sibernetic
+
+    - name: Build native Metal substrate (sib_metal)
+      run: |
+        cd src/metal_diff
+        ./build.sh
+        ls -la sib_metal
+
+    - name: Install numpy for smoke-test driver
+      run: |
+        # The smoke test below uses load_config.py which `import numpy as np`.
+        # macos-latest's system python3 doesn't ship numpy by default; install it.
+        # `--break-system-packages` is required on Python 3.13+ runners that
+        # treat the system interpreter as PEP-668 externally-managed.
+        python3 -m pip install --break-system-packages --quiet numpy
+
+    - name: Native Metal substrate smoke test (xpbd_step on demo1)
+      run: |
+        # load_to_metal_buffers reads `configuration/<scenario>` relative
+        # to cwd, so the smoke test must run from the repo root, not from
+        # src/metal_diff/. Pre-locate the binary so we can call it after
+        # the cd.
+        export SIB_METAL=$(pwd)/src/metal_diff/sib_metal
+        $SIB_METAL --help 2>&1 | head -30 || true
+        # Exercise the per-step XPBD pipeline: load demo1, run 10 steps,
+        # check the output file exists. This catches regressions in any of
+        # the M6/M7 kernels (density solve, distance constraints, floor,
+        # spring forces, pair forces) because all of them are touched by
+        # a single xpbd_step call.
+        python3 - <<PY
+        import os, sys, subprocess
+        sys.path.insert(0, 'src/metal_diff')
+        from load_config import load_to_metal_buffers
+        info = load_to_metal_buffers('demo1', out_dir='/tmp/ci_demo1', h=3.34)
+        # Initial pos/vel
+        import numpy as np
+        pos = np.fromfile(info['paths']['pos_active'], dtype=np.float32).reshape(-1, 3)
+        vel = np.zeros_like(pos)
+        pos.astype(np.float32).tofile('/tmp/ci_pos_in.bin')
+        vel.astype(np.float32).tofile('/tmp/ci_vel_in.bin')
+        cmd = [os.environ['SIB_METAL'], 'xpbd_step',
+               str(info['n_active']), str(info['n_static']),
+               '3.34', '2e-12', '1000.0', '2e-5', '-9.8', '2.0', '1e-3', '3',
+               '/tmp/ci_pos_in.bin', '/tmp/ci_vel_in.bin',
+               info['paths']['pos_static'],
+               str(info['n_bonds']), info['paths']['bonds'],
+               '3.3e-9', '50',                    # alpha_dist, n_steps
+               '7.4e-6', '5e-5', '5500.0', '0.0'] # sim_scale, visc, K, restitution
+        r = subprocess.run(cmd, capture_output=True, text=True, timeout=60)
+        if r.returncode != 0:
+            print('STDOUT:', r.stdout); print('STDERR:', r.stderr); sys.exit(1)
+        out = np.fromfile('/tmp/xpbd_pos_out.bin', dtype=np.float32).reshape(-1, 3)
+        # Sanity: cube should still be in the box (no NaN/Inf, all y in [0, 100])
+        assert np.all(np.isfinite(out)), 'NaN/Inf in output positions'
+        assert np.all((out[:, 1] >= -10) & (out[:, 1] <= 100)), 'cube escaped box'
+        print(f'Native Metal smoke test PASSED ({info["n_active"]} active particles, 50 steps)')
+        PY

.gitignore

@@ -2,6 +2,8 @@
 /src/main_sim2.py
 /src/main_sim.py~
 /Release/Sibernetic
+/src/metal_diff/sib_metal
+__pycache__/
 *~
 # Compiled Object files
 *.slo
@@ -69,3 +71,6 @@ simulations
 .vscode/*
 .settings/*
 .idea/*
+
+# Project-local Python venv created by setup.sh
+/.venv/

AGENTS.md

@@ -0,0 +1,42 @@
+# Repository Guide
+
+This repository contains the **Sibernetic** simulator along with Python
+bindings and tests.  The code base mixes C++ (in `src/` and `inc/`), OpenCL
+kernels and a number of helper Python scripts.
+
+## Layout
+- `src/` – main C++ sources and OpenCL kernels (`sphFluid.cl`).
+- `inc/` – C++ headers with physics constants, solver classes and helpers.
+- `configuration/` – example configuration files for the simulator.
+- `buffers/` – output data (created at runtime).
+- `tests/` – Python tests; `run_all_tests.sh` wraps `sibernetic_c302.py`
+  for automated runs.
+- Python utilities such as `main_sim.py`, `sibernetic_c302.py` and
+  `plot_positions.py` can drive the simulator or analyse its output.
+
+## Building and Testing
+Before building you should install dependencies via `./setup.sh`.
+To compile the C++ code use `make`.  A convenience script `test.sh`
+runs code formatting, static checks via **ruff**, builds the simulator
+and executes the test suite:
+
+```bash
+./test.sh
+```
+
+`test.sh` in turn calls `run_all_tests.sh` which performs multiple
+runs of `sibernetic_c302.py` and verifies the produced output files.
+
+For Apple Silicon GPU simulation, build the native Metal substrate
+separately via `cd src/metal_diff && ./build.sh`; the resulting
+`sib_metal` binary exposes the differentiable XPBD pipeline (see
+README.md "Differentiable physics with native Metal").
+
+## Contributing Notes
+- Keep C++ headers and sources under `inc/` and `src/` respectively.
+- OpenCL kernels reside in `src/*.cl`.
+- When adding Python scripts ensure they pass `ruff format` and
+  `ruff check`.
+- Test additions should be placed under `tests/` and runnable via
+  `test.sh`.
+