[RELEASE] cucim v25.02

.github/copy-pr-bot.yaml

@@ -2,3 +2,4 @@
 # https://docs.gha-runners.nvidia.com/apps/copy-pr-bot/
 
 enabled: true
+auto_sync_draft: false

.github/workflows/build.yaml

@@ -28,7 +28,7 @@ concurrency:
 jobs:
   cpp-build:
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/conda-cpp-build.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/conda-cpp-build.yaml@branch-25.02
     with:
       build_type: ${{ inputs.build_type || 'branch' }}
       branch: ${{ inputs.branch }}
@@ -38,7 +38,7 @@ jobs:
     if: github.ref_type == 'branch'
     needs: [python-build]
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/custom-job.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/custom-job.yaml@branch-25.02
     with:
       arch: "amd64"
       branch: ${{ inputs.branch }}
@@ -51,7 +51,7 @@ jobs:
   python-build:
     needs: [cpp-build]
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/conda-python-build.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/conda-python-build.yaml@branch-25.02
     with:
       build_type: ${{ inputs.build_type || 'branch' }}
       branch: ${{ inputs.branch }}
@@ -60,15 +60,15 @@ jobs:
   upload-conda:
     needs: [cpp-build, python-build]
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/conda-upload-packages.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/conda-upload-packages.yaml@branch-25.02
     with:
       build_type: ${{ inputs.build_type || 'branch' }}
       branch: ${{ inputs.branch }}
       date: ${{ inputs.date }}
       sha: ${{ inputs.sha }}
   wheel-build:
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@branch-25.02
     with:
       build_type: ${{ inputs.build_type || 'branch' }}
       branch: ${{ inputs.branch }}
@@ -78,7 +78,7 @@ jobs:
   wheel-publish:
     needs: wheel-build
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/wheels-publish.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-publish.yaml@branch-25.02
     with:
       build_type: ${{ inputs.build_type || 'branch' }}
       branch: ${{ inputs.branch }}

.github/workflows/pr.yaml

@@ -20,32 +20,32 @@ jobs:
       - wheel-build
       - wheel-tests
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/pr-builder.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/pr-builder.yaml@branch-25.02
   checks:
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/checks.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/checks.yaml@branch-25.02
   conda-cpp-build:
     needs: checks
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/conda-cpp-build.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/conda-cpp-build.yaml@branch-25.02
     with:
       build_type: pull-request
   conda-python-build:
     needs: conda-cpp-build
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/conda-python-build.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/conda-python-build.yaml@branch-25.02
     with:
       build_type: pull-request
   conda-python-tests:
     needs: conda-python-build
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/conda-python-tests.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/conda-python-tests.yaml@branch-25.02
     with:
       build_type: pull-request
   docs-build:
     needs: conda-python-build
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/custom-job.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/custom-job.yaml@branch-25.02
     with:
       build_type: pull-request
       node_type: "gpu-v100-latest-1"
@@ -55,14 +55,14 @@ jobs:
   wheel-build:
     needs: checks
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-build.yaml@branch-25.02
     with:
       build_type: pull-request
       script: ci/build_wheel.sh
   wheel-tests:
     needs: wheel-build
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/wheels-test.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-test.yaml@branch-25.02
     with:
       build_type: pull-request
       script: ci/test_wheel.sh

.github/workflows/test.yaml

@@ -16,15 +16,15 @@ on:
 jobs:
   conda-python-tests:
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/conda-python-tests.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/conda-python-tests.yaml@branch-25.02
     with:
       build_type: nightly
       branch: ${{ inputs.branch }}
       date: ${{ inputs.date }}
       sha: ${{ inputs.sha }}
   wheel-tests:
     secrets: inherit
-    uses: rapidsai/shared-workflows/.github/workflows/wheels-test.yaml@branch-24.12
+    uses: rapidsai/shared-workflows/.github/workflows/wheels-test.yaml@branch-25.02
     with:
       build_type: nightly
       branch: ${{ inputs.branch }}

.github/workflows/trigger-breaking-change-alert.yaml

@@ -0,0 +1,26 @@
+name: Trigger Breaking Change Notifications
+
+on:
+  pull_request_target:
+    types:
+      - closed
+      - reopened
+      - labeled
+      - unlabeled
+
+jobs:
+  trigger-notifier:
+    if: contains(github.event.pull_request.labels.*.name, 'breaking')
+    secrets: inherit
+    uses: rapidsai/shared-workflows/.github/workflows/[email protected]
+    with:
+      sender_login: ${{ github.event.sender.login }}
+      sender_avatar: ${{ github.event.sender.avatar_url }}
+      repo: ${{ github.repository }}
+      pr_number: ${{ github.event.pull_request.number }}
+      pr_title: "${{ github.event.pull_request.title }}"
+      pr_body: "${{ github.event.pull_request.body || '_Empty PR description_' }}"
+      pr_base_ref: ${{ github.event.pull_request.base.ref }}
+      pr_author: ${{ github.event.pull_request.user.login }}
+      event_action: ${{ github.event.action }}
+      pr_merged: ${{ github.event.pull_request.merged }}

.pre-commit-config.yaml

@@ -37,3 +37,9 @@ repos:
     hooks:
         - id: rapids-dependency-file-generator
           args: ["--clean"]
+  - repo: https://github.com/shellcheck-py/shellcheck-py
+    rev: v0.10.0.1
+    hooks:
+      - id: shellcheck
+        args: ["--severity=warning"]
+        files: ^ci/

CHANGELOG.md

@@ -1,3 +1,24 @@
+# cucim 25.02.00 (13 Feb 2025)
+
+## 🐛 Bug Fixes
+
+- Fix primitives benchmark code ([#812](https://github.com/rapidsai/cucim/pull/812)) [@gigony](https://github.com/gigony)
+- CuPy 14.0 compatibility ([#808](https://github.com/rapidsai/cucim/pull/808)) [@grlee77](https://github.com/grlee77)
+
+## 🛠️ Improvements
+
+- Use `rapids-pip-retry` in CI jobs that might need retries ([#824](https://github.com/rapidsai/cucim/pull/824)) [@gforsyth](https://github.com/gforsyth)
+- Revert CUDA 12.8 shared workflow branch changes ([#818](https://github.com/rapidsai/cucim/pull/818)) [@vyasr](https://github.com/vyasr)
+- Build and test with CUDA 12.8.0 ([#815](https://github.com/rapidsai/cucim/pull/815)) [@bdice](https://github.com/bdice)
+- Add shellcheck to pre-commit and fix warnings ([#814](https://github.com/rapidsai/cucim/pull/814)) [@gforsyth](https://github.com/gforsyth)
+- Update vendored ndimage code with axes support ([#813](https://github.com/rapidsai/cucim/pull/813)) [@grlee77](https://github.com/grlee77)
+- Use GCC 13 in CUDA 12 conda builds. ([#811](https://github.com/rapidsai/cucim/pull/811)) [@bdice](https://github.com/bdice)
+- Improve performance of color distance calculations by kernel fusion ([#809](https://github.com/rapidsai/cucim/pull/809)) [@grlee77](https://github.com/grlee77)
+- Incorporate upstream changes from scikit-image 0.25 ([#806](https://github.com/rapidsai/cucim/pull/806)) [@grlee77](https://github.com/grlee77)
+- Update version references in workflow ([#803](https://github.com/rapidsai/cucim/pull/803)) [@AyodeAwe](https://github.com/AyodeAwe)
+- Require approval to run CI on draft PRs ([#798](https://github.com/rapidsai/cucim/pull/798)) [@bdice](https://github.com/bdice)
+- Add breaking change workflow trigger ([#795](https://github.com/rapidsai/cucim/pull/795)) [@AyodeAwe](https://github.com/AyodeAwe)
+
 # cucim 24.12.00 (11 Dec 2024)
 
 ## 🚨 Breaking Changes

VERSION

@@ -1 +1 @@
-24.12.00
+25.02.00

benchmarks/primitives.cpp

@@ -89,10 +89,10 @@ static void string_memcpy(benchmark::State& state)
         char * c_str = (char*) malloc(size + 1);
         memcpy(c_str, data.data(), size);
         c_str[size] = '\0';
-        free(c_str);
         // Make sure the variable is not optimized away by compiler
         benchmark::DoNotOptimize(c_str);
         benchmark::DoNotOptimize(size);
+        free(c_str);
     }
 }
 BENCHMARK(string_memcpy);
@@ -105,9 +105,9 @@ static void string_strcpy(benchmark::State& state)
         std::string data = "#########################################################################################################################################################################################";
         char * c_str = (char*) malloc(data.size() + 1);
         strcpy(c_str, data.data());
-        free(c_str);
         // Make sure the variable is not optimized away by compiler
         benchmark::DoNotOptimize(c_str);
+        free(c_str);
     }
 }
 BENCHMARK(string_strcpy);
@@ -120,9 +120,9 @@ static void string_strdup(benchmark::State& state)
     {
         std::string data = "#########################################################################################################################################################################################";
         char * c_str = strdup(data.data());
-        free(c_str);
         // Make sure the variable is not optimized away by compiler
         benchmark::DoNotOptimize(c_str);
+        free(c_str);
     }
 }
 BENCHMARK(string_strdup);
@@ -140,12 +140,12 @@ static void alloc_malloc(benchmark::State& state)
             arr[i] = (char*)malloc(10);
             arr[i][0] = i;
         }
+        // Make sure the variable is not optimized away by compiler
+        benchmark::DoNotOptimize(arr);
         for (int i = 0; i < 30000; i++)
         {
             free(arr[i]);
         }
-        // Make sure the variable is not optimized away by compiler
-        benchmark::DoNotOptimize(arr);
     }
 }
 BENCHMARK(alloc_malloc);//->Iterations(100);
@@ -163,12 +163,12 @@ static void alloc_pmr(benchmark::State& state)
             arr[i] = static_cast<char*>(cucim_malloc(10));
             arr[i][0] = i;
         }
+        // Make sure the variable is not optimized away by compiler
+        benchmark::DoNotOptimize(arr);
         for (int i = 0; i < 30000; i++)
         {
             cucim_free(arr[i]);
         }
-        // Make sure the variable is not optimized away by compiler
-        benchmark::DoNotOptimize(arr);
     }
 }
 BENCHMARK(alloc_pmr);//->Iterations(100);

benchmarks/skimage/cucim_color_bench.py

@@ -1,4 +1,5 @@
 import argparse
+import math
 import os
 import pickle
 
@@ -25,6 +26,10 @@
     "lab2xyz",
     "rgba2rgb",
     "label2rgb",
+    "deltaE_cie76",
+    "deltaE_ciede94",
+    "deltaE_ciede2000",
+    "deltaE_cmc",
 ]
 
 
@@ -40,6 +45,32 @@ def set_args(self, dtype):
         self.args_gpu = (imaged,)
 
 
+class DeltaEBench(ImageBench):
+    def set_args(self, dtype):
+        from skimage import color, data
+
+        # create synthetic lab image pair
+        rgb1 = data.astronaut()
+        lab1 = color.rgb2lab(rgb1)
+        lab2 = color.rgb2lab(np.roll(rgb1, (1, 1), axis=(0, 1)))
+
+        # change to desired dtype
+        lab1 = lab1.astype(dtype, copy=False)
+        lab2 = lab2.astype(dtype, copy=False)
+
+        # tile then crop as needed to get the expected size
+        n_tile0 = math.ceil(self.shape[0] / lab1.shape[0])
+        n_tile1 = math.ceil(self.shape[1] / lab1.shape[1])
+        lab1 = np.tile(lab1, (n_tile0, n_tile1, 1))
+        lab1 = lab1[: self.shape[0], : self.shape[1], :]
+        lab2 = np.tile(lab2, (n_tile0, n_tile1, 1))
+        lab2 = lab2[: self.shape[0], : self.shape[1], :]
+
+        print(f"{lab1.shape=}")
+        self.args_cpu = (lab1, lab2)
+        self.args_gpu = (cp.asarray(lab1), cp.asarray(lab2))
+
+
 class RGBABench(ImageBench):
     def set_args(self, dtype):
         if self.shape[-1] != 4:
@@ -162,6 +193,23 @@ def main(args):
                     results = B.run_benchmark(duration=args.duration)
                     all_results = pd.concat([all_results, results["full"]])
 
+        elif function_name.startswith("deltaE"):
+            # only run these functions for floating point data types
+            float_dtypes = [t for t in dtypes if np.dtype(t).kind == "f"]
+
+            B = DeltaEBench(
+                function_name=function_name,
+                shape=shape + (3,),
+                dtypes=float_dtypes,
+                fixed_kwargs={},
+                var_kwargs={},
+                # index_str=f"{fromspace.lower()}2{tospace.lower()}",
+                module_cpu=skimage.color,
+                module_gpu=cucim.skimage.color,
+                run_cpu=run_cpu,
+            )
+            results = B.run_benchmark(duration=args.duration)
+            all_results = pd.concat([all_results, results["full"]])
         elif function_name == "rgba2rgb":
             B = RGBABench(
                 function_name="rgba2rgb",

benchmarks/skimage/run-nv-bench-color.sh

@@ -9,3 +9,14 @@ for shape in "${param_shape[@]}"; do
         done
     done
 done
+
+param_shape=(512,512 3840,2160)
+param_filt=(deltaE_cie76 deltaE_ciede94 deltaE_ciede2000 deltaE_cmc)
+param_dt=(float32, float64)
+for shape in "${param_shape[@]}"; do
+    for filt in "${param_filt[@]}"; do
+        for dt in "${param_dt[@]}"; do
+            python cucim_color_bench.py -f $filt -i $shape -d $dt -t 10
+        done
+    done
+done

ci/build_docs.sh

@@ -27,7 +27,8 @@ rapids-mamba-retry install \
     "cucim=${RAPIDS_VERSION}" \
     "libcucim=${RAPIDS_VERSION}"
 
-export RAPIDS_DOCS_DIR="$(mktemp -d)"
+RAPIDS_DOCS_DIR="$(mktemp -d)"
+export RAPIDS_DOCS_DIR
 
 rapids-logger "Build Python docs"
 pushd docs

ci/build_wheel.sh

@@ -13,7 +13,7 @@ source rapids-date-string
 
 rapids-generate-version > ./VERSION
 
-RAPIDS_PY_CUDA_SUFFIX="$(rapids-wheel-ctk-name-gen ${RAPIDS_CUDA_VERSION})"
+RAPIDS_PY_CUDA_SUFFIX="$(rapids-wheel-ctk-name-gen "${RAPIDS_CUDA_VERSION}")"
 
 rapids-logger "Generating build requirements"
 
@@ -25,7 +25,7 @@ rapids-dependency-file-generator \
 | tee /tmp/requirements-build.txt
 
 rapids-logger "Installing build requirements"
-python -m pip install \
+rapids-pip-retry install \
     -v \
     --prefer-binary \
     -r /tmp/requirements-build.txt
@@ -42,7 +42,7 @@ cd "${package_dir}"
 sccache --zero-stats
 
 rapids-logger "Building '${package_name}' wheel"
-python -m pip wheel \
+rapids-pip-retry wheel \
     -w dist \
     -v \
     --no-build-isolation \
@@ -54,6 +54,7 @@ sccache --show-adv-stats
 
 mkdir -p final_dist
 python -m auditwheel repair -w final_dist dist/*
+# shellcheck disable=SC2010
 ls -1 final_dist | grep -vqz 'none'
 
 ../../ci/validate_wheel.sh final_dist