From 429da867f17e2f2e820cce3927eba709ac9508a6 Mon Sep 17 00:00:00 2001
From: "yuxi.hong.research@gmail.com"
 <exouser@image-build-instance-ubuntu-24-1777879275710816896.js2local>
Date: Thu, 4 Jun 2026 02:10:53 +0000
Subject: [PATCH 1/4] use cuda128 for build

---
 .github/workflows/buildcupy.yaml | 42 ++++++++++++--------------------
 1 file changed, 16 insertions(+), 26 deletions(-)

diff --git a/.github/workflows/buildcupy.yaml b/.github/workflows/buildcupy.yaml
index 56f917f8..73966f1e 100644
--- a/.github/workflows/buildcupy.yaml
+++ b/.github/workflows/buildcupy.yaml
@@ -22,35 +22,25 @@ jobs:
           python-version: 3.11
       - name: Install dependencies and pylops
         run: |
-          srun --account=yuxilab -n 1 -N 1 --gres=gpu:L40S:1 bash -c '
-            uv sync --locked  --extra advanced \
-              --extra stat --extra gpu-cu12 \
-              --extra deep-cu128 --all-groups
-          '
+          uv sync --locked --extra advanced \
+            --extra stat --extra gpu-cu12 \
+            --extra deep-cu128 --all-groups
           echo "done!"
-      - name: Install CUDA 12.8 toolkit
+      - name: Use local CUDA 12.8 toolkit
         run: |
-          srun --account=yuxilab -n 1 -N 1 --gres=gpu:L40S:1 bash -c '
-            # Currently, the NVIDIA GPU drivers on the self-hosted runner
-            # and the CUDA installation in CUDA_HOME require different PTX versions
-            # (8.7 and 8.8, respectively). CUDA and NVCC are temporarely
-            # installed in the uv environment, and the CUDA paths are set in
-            # the test command, to work around this issue.
-            uv add nvidia-cuda-nvcc-cu12==12.8.*
-            # Numba resolves libNVVM via CUDA_HOME and needs a versioned
-            # libnvvm.so.N (the pip wheel only ships a bare libnvvm.so), so
-            # add the symlink once here against the wheel-provided toolkit.
-            ln -sf libnvvm.so .venv/lib/python3.11/site-packages/nvidia/cuda_nvcc/nvvm/lib64/libnvvm.so.4
-          '
+          if [ ! -d "$HOME/local/cuda" ]; then
+            echo "CUDA 12.8 toolkit not found at $HOME/local/cuda"
+            exit 1
+          fi
+          echo "CUDA_HOME=$HOME/local/cuda" >> "$GITHUB_ENV"
           echo "done!"
       - name: Test with pytest
         run: |
-          srun --account=yuxilab -n 1 -N 1 --gres=gpu:L40S:1 bash -c '
-            # Use the pip-installed CUDA 12.8 toolkit (matches the 12.8 driver),
-            # not any system CUDA on the runner, so numba emits PTX <= 8.7.
-            export CUDA_HOME="$PWD/.venv/lib/python3.11/site-packages/nvidia/cuda_nvcc"
-            export CUPY_PYLOPS=1
-            export TEST_CUPY_PYLOPS=1
-            uv run pytest --color=yes pytests/
-          '
+          # Use the local CUDA 12.8 toolchain from $HOME/local/cuda.
+          export CUDA_HOME="${CUDA_HOME:-$HOME/local/cuda}"
+          export CUPY_PYLOPS=1
+          export TEST_CUPY_PYLOPS=1
+          export PATH="$CUDA_HOME/bin:$PATH"
+          export LD_LIBRARY_PATH="$CUDA_HOME/lib64:$LD_LIBRARY_PATH"
+          uv run pytest --color=yes pytests/
           echo "done!"

From ed04f7fe991ea59ae5010125b31f7608ecc19077 Mon Sep 17 00:00:00 2001
From: Matteo Ravasi <matteoravasi@gmail.com>
Date: Thu, 4 Jun 2026 09:01:58 +0100
Subject: [PATCH 2/4] Update build environment to include 'exouser'

---
 .github/workflows/buildcupy.yaml | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/buildcupy.yaml b/.github/workflows/buildcupy.yaml
index 73966f1e..fc8c5e12 100644
--- a/.github/workflows/buildcupy.yaml
+++ b/.github/workflows/buildcupy.yaml
@@ -10,7 +10,9 @@ on:
 
 jobs:
   build:
-    runs-on: self-hosted
+    runs-on:
+      - self-hosted
+      - exouser
     steps:
       - name: Check out source repository
         uses: actions/checkout@v6

From 029e74e55073c1f0dd77f9dc96f9f0329cf3c358 Mon Sep 17 00:00:00 2001
From: yuxihong <yuxi.hong.research@gmail.com>
Date: Thu, 4 Jun 2026 17:37:10 +0000
Subject: [PATCH 3/4] add a new exouser runner

---
 .github/workflows/buildcupy-exouser.yaml | 77 ++++++++++++++++++++++++
 1 file changed, 77 insertions(+)
 create mode 100644 .github/workflows/buildcupy-exouser.yaml

diff --git a/.github/workflows/buildcupy-exouser.yaml b/.github/workflows/buildcupy-exouser.yaml
new file mode 100644
index 00000000..5ac5185a
--- /dev/null
+++ b/.github/workflows/buildcupy-exouser.yaml
@@ -0,0 +1,77 @@
+name: PyLops Testing (CuPy Exouser)
+
+on:
+  pull_request:
+    types: [opened, synchronize, reopened]
+  push:
+    branches:
+      - main
+      - dev
+
+jobs:
+  build:
+    runs-on:
+      - self-hosted
+      - exouser
+    steps:
+      - name: Check out source repository
+        uses: actions/checkout@v6
+        with:
+          fetch-depth: 0
+      - name: Install uv with Python 3.11
+        uses: astral-sh/setup-uv@v6
+        with:
+          python-version: 3.11
+      - name: Install dependencies and pylops
+        run: |
+          uv sync --extra advanced \
+            --extra stat --extra gpu-cu12 \
+            --group dev
+          uv pip install \
+            --default-index https://download.pytorch.org/whl/cu128 \
+            torch==2.11.0+cu128
+          uv pip install \
+            nvidia-cuda-nvcc-cu12==12.2.128 \
+            nvidia-cuda-nvrtc-cu12==12.8.93 \
+            nvidia-cuda-runtime-cu12==12.8.90
+          echo "done!"
+
+      - name: Configure CUDA NVVM from uv wheels
+        run: |
+          CUDA_HOME="$PWD/.venv/lib/python3.11/site-packages/nvidia/cuda_nvcc"
+          CUDA_RUNTIME_ROOT="$PWD/.venv/lib/python3.11/site-packages/nvidia/cuda_runtime"
+          CUDA_NVRTC_ROOT="$PWD/.venv/lib/python3.11/site-packages/nvidia/cuda_nvrtc"
+
+          if [ ! -f "$CUDA_HOME/nvvm/lib64/libnvvm.so" ]; then
+            echo "libnvvm.so not found in $CUDA_HOME/nvvm/lib64"
+            exit 1
+          fi
+          if [ ! -f "$CUDA_RUNTIME_ROOT/lib/libcudart.so.12" ]; then
+            echo "libcudart.so.12 not found in $CUDA_RUNTIME_ROOT/lib"
+            exit 1
+          fi
+          if [ ! -f "$CUDA_NVRTC_ROOT/lib/libnvrtc.so.12" ]; then
+            echo "libnvrtc.so.12 not found in $CUDA_NVRTC_ROOT/lib"
+            exit 1
+          fi
+
+          mkdir -p "$CUDA_HOME/lib64"
+          ln -sf "$CUDA_HOME/nvvm/lib64/libnvvm.so" "$CUDA_HOME/nvvm/lib64/libnvvm.so.4"
+          ln -sf "$CUDA_HOME/nvvm/lib64/libnvvm.so" "$CUDA_HOME/lib64/libnvvm.so"
+          ln -sf "$CUDA_RUNTIME_ROOT/lib/libcudart.so.12" "$CUDA_HOME/lib64/libcudart.so.12"
+          ln -sf "$CUDA_RUNTIME_ROOT/lib/libcudart.so.12" "$CUDA_HOME/lib64/libcudart.so"
+          ln -sf "$CUDA_NVRTC_ROOT/lib/libnvrtc.so.12" "$CUDA_HOME/lib64/libnvrtc.so.12"
+          ln -sf "$CUDA_HOME/lib64/libnvrtc.so.12" "$CUDA_HOME/lib64/libnvrtc.so"
+
+          echo "CUDA_HOME=$CUDA_HOME" >> "$GITHUB_ENV"
+          echo "CUDA_NVVM_LIB=$CUDA_HOME/nvvm/lib64" >> "$GITHUB_ENV"
+          echo "done!"
+      - name: Test with pytest
+        run: |
+          export CUDA_HOME="${CUDA_HOME:-$PWD/.venv/lib/python3.11/site-packages/nvidia/cuda_nvcc}"
+          export CUPY_PYLOPS=1
+          export TEST_CUPY_PYLOPS=1
+          export LD_LIBRARY_PATH="$CUDA_HOME/lib64:$CUDA_NVVM_LIB:$LD_LIBRARY_PATH"
+          export NUMBA_CUDA_NVVM="$CUDA_NVVM_LIB/libnvvm.so"
+          uv run --no-sync pytest --color=yes pytests/
+          echo "done!"

From e17548dd681ccd9370c7c4fb649602dc62acd5b4 Mon Sep 17 00:00:00 2001
From: mrava87 <matteoravasi@gmail.com>
Date: Thu, 4 Jun 2026 20:31:28 +0100
Subject: [PATCH 4/4] ci: reorganized buildcupy actions

---
 .github/workflows/buildcupy-exouser.yaml | 77 ------------------------
 .github/workflows/buildcupy.yaml         | 53 ++++++++++++----
 .github/workflows/buildcupy1.yaml        | 58 ++++++++++++++++++
 3 files changed, 99 insertions(+), 89 deletions(-)
 delete mode 100644 .github/workflows/buildcupy-exouser.yaml
 create mode 100644 .github/workflows/buildcupy1.yaml

diff --git a/.github/workflows/buildcupy-exouser.yaml b/.github/workflows/buildcupy-exouser.yaml
deleted file mode 100644
index 5ac5185a..00000000
--- a/.github/workflows/buildcupy-exouser.yaml
+++ /dev/null
@@ -1,77 +0,0 @@
-name: PyLops Testing (CuPy Exouser)
-
-on:
-  pull_request:
-    types: [opened, synchronize, reopened]
-  push:
-    branches:
-      - main
-      - dev
-
-jobs:
-  build:
-    runs-on:
-      - self-hosted
-      - exouser
-    steps:
-      - name: Check out source repository
-        uses: actions/checkout@v6
-        with:
-          fetch-depth: 0
-      - name: Install uv with Python 3.11
-        uses: astral-sh/setup-uv@v6
-        with:
-          python-version: 3.11
-      - name: Install dependencies and pylops
-        run: |
-          uv sync --extra advanced \
-            --extra stat --extra gpu-cu12 \
-            --group dev
-          uv pip install \
-            --default-index https://download.pytorch.org/whl/cu128 \
-            torch==2.11.0+cu128
-          uv pip install \
-            nvidia-cuda-nvcc-cu12==12.2.128 \
-            nvidia-cuda-nvrtc-cu12==12.8.93 \
-            nvidia-cuda-runtime-cu12==12.8.90
-          echo "done!"
-
-      - name: Configure CUDA NVVM from uv wheels
-        run: |
-          CUDA_HOME="$PWD/.venv/lib/python3.11/site-packages/nvidia/cuda_nvcc"
-          CUDA_RUNTIME_ROOT="$PWD/.venv/lib/python3.11/site-packages/nvidia/cuda_runtime"
-          CUDA_NVRTC_ROOT="$PWD/.venv/lib/python3.11/site-packages/nvidia/cuda_nvrtc"
-
-          if [ ! -f "$CUDA_HOME/nvvm/lib64/libnvvm.so" ]; then
-            echo "libnvvm.so not found in $CUDA_HOME/nvvm/lib64"
-            exit 1
-          fi
-          if [ ! -f "$CUDA_RUNTIME_ROOT/lib/libcudart.so.12" ]; then
-            echo "libcudart.so.12 not found in $CUDA_RUNTIME_ROOT/lib"
-            exit 1
-          fi
-          if [ ! -f "$CUDA_NVRTC_ROOT/lib/libnvrtc.so.12" ]; then
-            echo "libnvrtc.so.12 not found in $CUDA_NVRTC_ROOT/lib"
-            exit 1
-          fi
-
-          mkdir -p "$CUDA_HOME/lib64"
-          ln -sf "$CUDA_HOME/nvvm/lib64/libnvvm.so" "$CUDA_HOME/nvvm/lib64/libnvvm.so.4"
-          ln -sf "$CUDA_HOME/nvvm/lib64/libnvvm.so" "$CUDA_HOME/lib64/libnvvm.so"
-          ln -sf "$CUDA_RUNTIME_ROOT/lib/libcudart.so.12" "$CUDA_HOME/lib64/libcudart.so.12"
-          ln -sf "$CUDA_RUNTIME_ROOT/lib/libcudart.so.12" "$CUDA_HOME/lib64/libcudart.so"
-          ln -sf "$CUDA_NVRTC_ROOT/lib/libnvrtc.so.12" "$CUDA_HOME/lib64/libnvrtc.so.12"
-          ln -sf "$CUDA_HOME/lib64/libnvrtc.so.12" "$CUDA_HOME/lib64/libnvrtc.so"
-
-          echo "CUDA_HOME=$CUDA_HOME" >> "$GITHUB_ENV"
-          echo "CUDA_NVVM_LIB=$CUDA_HOME/nvvm/lib64" >> "$GITHUB_ENV"
-          echo "done!"
-      - name: Test with pytest
-        run: |
-          export CUDA_HOME="${CUDA_HOME:-$PWD/.venv/lib/python3.11/site-packages/nvidia/cuda_nvcc}"
-          export CUPY_PYLOPS=1
-          export TEST_CUPY_PYLOPS=1
-          export LD_LIBRARY_PATH="$CUDA_HOME/lib64:$CUDA_NVVM_LIB:$LD_LIBRARY_PATH"
-          export NUMBA_CUDA_NVVM="$CUDA_NVVM_LIB/libnvvm.so"
-          uv run --no-sync pytest --color=yes pytests/
-          echo "done!"
diff --git a/.github/workflows/buildcupy.yaml b/.github/workflows/buildcupy.yaml
index fc8c5e12..5ebb952d 100644
--- a/.github/workflows/buildcupy.yaml
+++ b/.github/workflows/buildcupy.yaml
@@ -1,4 +1,4 @@
-name: PyLops Testing (CuPy)
+name: PyLops Testing with CuPy
 
 on:
   pull_request:
@@ -24,25 +24,54 @@ jobs:
           python-version: 3.11
       - name: Install dependencies and pylops
         run: |
-          uv sync --locked --extra advanced \
+          uv sync --extra advanced \
             --extra stat --extra gpu-cu12 \
-            --extra deep-cu128 --all-groups
+            --group dev
+          uv pip install \
+            --default-index https://download.pytorch.org/whl/cu128 \
+            torch==2.11.0+cu128
+          uv pip install \
+            nvidia-cuda-nvcc-cu12==12.2.128 \
+            nvidia-cuda-nvrtc-cu12==12.8.93 \
+            nvidia-cuda-runtime-cu12==12.8.90
           echo "done!"
-      - name: Use local CUDA 12.8 toolkit
+
+      - name: Configure CUDA NVVM from uv wheels
         run: |
-          if [ ! -d "$HOME/local/cuda" ]; then
-            echo "CUDA 12.8 toolkit not found at $HOME/local/cuda"
+          CUDA_HOME="$PWD/.venv/lib/python3.11/site-packages/nvidia/cuda_nvcc"
+          CUDA_RUNTIME_ROOT="$PWD/.venv/lib/python3.11/site-packages/nvidia/cuda_runtime"
+          CUDA_NVRTC_ROOT="$PWD/.venv/lib/python3.11/site-packages/nvidia/cuda_nvrtc"
+
+          if [ ! -f "$CUDA_HOME/nvvm/lib64/libnvvm.so" ]; then
+            echo "libnvvm.so not found in $CUDA_HOME/nvvm/lib64"
+            exit 1
+          fi
+          if [ ! -f "$CUDA_RUNTIME_ROOT/lib/libcudart.so.12" ]; then
+            echo "libcudart.so.12 not found in $CUDA_RUNTIME_ROOT/lib"
             exit 1
           fi
-          echo "CUDA_HOME=$HOME/local/cuda" >> "$GITHUB_ENV"
+          if [ ! -f "$CUDA_NVRTC_ROOT/lib/libnvrtc.so.12" ]; then
+            echo "libnvrtc.so.12 not found in $CUDA_NVRTC_ROOT/lib"
+            exit 1
+          fi
+
+          mkdir -p "$CUDA_HOME/lib64"
+          ln -sf "$CUDA_HOME/nvvm/lib64/libnvvm.so" "$CUDA_HOME/nvvm/lib64/libnvvm.so.4"
+          ln -sf "$CUDA_HOME/nvvm/lib64/libnvvm.so" "$CUDA_HOME/lib64/libnvvm.so"
+          ln -sf "$CUDA_RUNTIME_ROOT/lib/libcudart.so.12" "$CUDA_HOME/lib64/libcudart.so.12"
+          ln -sf "$CUDA_RUNTIME_ROOT/lib/libcudart.so.12" "$CUDA_HOME/lib64/libcudart.so"
+          ln -sf "$CUDA_NVRTC_ROOT/lib/libnvrtc.so.12" "$CUDA_HOME/lib64/libnvrtc.so.12"
+          ln -sf "$CUDA_HOME/lib64/libnvrtc.so.12" "$CUDA_HOME/lib64/libnvrtc.so"
+
+          echo "CUDA_HOME=$CUDA_HOME" >> "$GITHUB_ENV"
+          echo "CUDA_NVVM_LIB=$CUDA_HOME/nvvm/lib64" >> "$GITHUB_ENV"
           echo "done!"
       - name: Test with pytest
         run: |
-          # Use the local CUDA 12.8 toolchain from $HOME/local/cuda.
-          export CUDA_HOME="${CUDA_HOME:-$HOME/local/cuda}"
+          export CUDA_HOME="${CUDA_HOME:-$PWD/.venv/lib/python3.11/site-packages/nvidia/cuda_nvcc}"
           export CUPY_PYLOPS=1
           export TEST_CUPY_PYLOPS=1
-          export PATH="$CUDA_HOME/bin:$PATH"
-          export LD_LIBRARY_PATH="$CUDA_HOME/lib64:$LD_LIBRARY_PATH"
-          uv run pytest --color=yes pytests/
+          export LD_LIBRARY_PATH="$CUDA_HOME/lib64:$CUDA_NVVM_LIB:$LD_LIBRARY_PATH"
+          export NUMBA_CUDA_NVVM="$CUDA_NVVM_LIB/libnvvm.so"
+          uv run --no-sync pytest --color=yes pytests/
           echo "done!"
diff --git a/.github/workflows/buildcupy1.yaml b/.github/workflows/buildcupy1.yaml
new file mode 100644
index 00000000..71c4fbff
--- /dev/null
+++ b/.github/workflows/buildcupy1.yaml
@@ -0,0 +1,58 @@
+name: PyLops Testing (CuPy)
+
+on:
+  pull_request:
+    types: [opened, synchronize, reopened]
+  push:
+    branches:
+      - main
+      - dev
+
+jobs:
+  build:
+    runs-on:
+      - self-hosted
+      - uni
+    steps:
+      - name: Check out source repository
+        uses: actions/checkout@v6
+        with:
+          fetch-depth: 0
+      - name: Install uv with Python 3.11
+        uses: astral-sh/setup-uv@v6
+        with:
+          python-version: 3.11
+      - name: Install dependencies and pylops
+        run: |
+          srun --account=yuxilab -n 1 -N 1 --gres=gpu:L40S:1 bash -c '
+            uv sync --locked  --extra advanced \
+              --extra stat --extra gpu-cu12 \
+              --extra deep-cu128 --all-groups
+          '
+          echo "done!"
+      - name: Install CUDA 12.8 toolkit
+        run: |
+          srun --account=yuxilab -n 1 -N 1 --gres=gpu:L40S:1 bash -c '
+            # Currently, the NVIDIA GPU drivers on the self-hosted runner
+            # and the CUDA installation in CUDA_HOME require different PTX versions
+            # (8.7 and 8.8, respectively). CUDA and NVCC are temporarely
+            # installed in the uv environment, and the CUDA paths are set in
+            # the test command, to work around this issue.
+            uv add nvidia-cuda-nvcc-cu12==12.8.*
+            # Numba resolves libNVVM via CUDA_HOME and needs a versioned
+            # libnvvm.so.N (the pip wheel only ships a bare libnvvm.so), so
+            # add the symlink once here against the wheel-provided toolkit.
+            ln -sf libnvvm.so .venv/lib/python3.11/site-packages/nvidia/cuda_nvcc/nvvm/lib64/libnvvm.so.4
+          '
+          echo "done!"
+      - name: Test with pytest
+        run: |
+          srun --account=yuxilab -n 1 -N 1 --gres=gpu:L40S:1 bash -c '
+            # Use the pip-installed CUDA 12.8 toolkit (matches the 12.8 driver),
+            # not any system CUDA on the runner, so numba emits PTX <= 8.7.
+            export CUDA_HOME="$PWD/.venv/lib/python3.11/site-packages/nvidia/cuda_nvcc"
+            export CUPY_PYLOPS=1
+            export TEST_CUPY_PYLOPS=1
+            uv run pytest --color=yes pytests/
+          '
+          echo "done!"