Merge rapidsai/release/26.02 into rapidsai/main

Author: jakirkham

.github/workflows/build.yaml

@@ -4,6 +4,7 @@ on:
   push:
     branches:
       - "main"
+      - "release/*"
     tags:
       - v[0-9][0-9].[0-9][0-9].[0-9][0-9]
   workflow_dispatch:

README.md

@@ -99,6 +99,42 @@ docker cp $tmp_id:/input notebooks
 docker rm -v ${tmp_id}
 ```
 
+## cuslide2 Plugin (Experimental)
+
+cuCIM includes an experimental `cuslide2` plugin that uses [nvImageCodec](https://developer.nvidia.com/nvimagecodec) for GPU-accelerated TIFF decoding. This plugin provides enhanced performance for batch ROI decoding operations.
+
+### Enabling cuslide2
+
+To enable the cuslide2 plugin, set the `ENABLE_CUSLIDE2` environment variable:
+
+```bash
+ENABLE_CUSLIDE2=1 python your_script.py
+```
+
+### Test Scripts
+
+cuCIM provides test scripts to verify cuslide2 functionality with different TIFF formats:
+
+**Aperio SVS files:**
+
+```bash
+# Download a sample SVS file and run the test
+python scripts/test_aperio_svs.py --download
+
+# Or test with your own SVS file
+ENABLE_CUSLIDE2=1 python scripts/test_aperio_svs.py /path/to/your/file.svs
+```
+
+**Philips TIFF files:**
+
+```bash
+# Test with a Philips TIFF file
+ENABLE_CUSLIDE2=1 python scripts/test_philips_tiff.py /path/to/your/philips.tiff
+
+# List available test options
+python scripts/test_philips_tiff.py --help
+```
+
 ## Build/Install from Source
 
 See build [instructions](CONTRIBUTING.md#setting-up-your-build-environment).
@@ -119,4 +155,4 @@ is used in this project.
 
 Apache-2.0 License (see [LICENSE](LICENSE) file).
 
-Copyright (c) 2020-2025, NVIDIA CORPORATION.
+Copyright (c) 2020-2026, NVIDIA CORPORATION.

benchmarks/skimage/_image_bench.py

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2021-2025, NVIDIA CORPORATION. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2021-2026, NVIDIA CORPORATION. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
 import itertools
@@ -127,6 +127,17 @@ def _index(self, name, var_kwargs, dtype=None, shape=None):
             index += ", " + self.index_str
         return index
 
+    def _prep_kwargs_string(self, kwargs):
+        params = []
+        for k, v in kwargs.items():
+            if isinstance(v, types.FunctionType):
+                params.append(f"{k}={v.__name__}")
+            elif isinstance(v, (np.ndarray, cp.ndarray)):
+                params.append(f"{k}=array,shape={v.shape},dtype={v.dtype.name}")
+            else:
+                params.append(f"{k}={v}")
+        return ", ".join(params)
+
     def get_reps(self, func, args, kwargs, target_duration=5, cpu=True):
         if not cpu:
             # dry run
@@ -165,19 +176,22 @@ def run_benchmark(self, duration=3, verbose=True):
                 if "brute_force" in var_kwargs_gpu:
                     var_kwargs_gpu["brute_force"] = True
 
-                kw_cpu = {**self.fixed_kwargs_cpu, **var_kwargs_cpu}
                 kw_gpu = {**self.fixed_kwargs_gpu, **var_kwargs_gpu}
-                rep_kwargs_cpu = self.get_reps(
-                    self.func_cpu, self.args_cpu, kw_cpu, duration, cpu=True
-                )
                 rep_kwargs_gpu = self.get_reps(
                     self.func_gpu, self.args_gpu, kw_gpu, duration, cpu=False
                 )
-                print("Number of Repetitions : ", rep_kwargs_gpu)
+                print("Number of Repetitions (GPU): ", rep_kwargs_gpu)
+
+                if self.run_cpu is True:
+                    kw_cpu = {**self.fixed_kwargs_cpu, **var_kwargs_cpu}
+                    rep_kwargs_cpu = self.get_reps(
+                        self.func_cpu, self.args_cpu, kw_cpu, duration, cpu=True
+                    )
                 perf_gpu = repeat(
                     self.func_gpu, self.args_gpu, kw_gpu, **rep_kwargs_gpu
                 )
 
+                df.at[index, "GPU: kwargs"] = self._prep_kwargs_string(kw_gpu)
                 df.at[index, "shape"] = f"{self.shape}"
                 # df.at[index,  "description"] = index
                 df.at[index, "function_name"] = self.function_name
@@ -194,6 +208,7 @@ def run_benchmark(self, duration=3, verbose=True):
                     df.at[index, "CPU: host (mean)"] = perf.cpu_times.mean()
                     df.at[index, "CPU: host (std)"] = perf.cpu_times.std()
 
+                    df.at[index, "CPU: kwargs"] = self._prep_kwargs_string(kw_cpu)
                 df.at[index, "GPU: host (mean)"] = perf_gpu.cpu_times.mean()
                 df.at[index, "GPU: host (std)"] = perf_gpu.cpu_times.std()
                 df.at[index, "GPU: device (mean)"] = perf_gpu.gpu_times.mean()
@@ -202,14 +217,14 @@ def run_benchmark(self, duration=3, verbose=True):
                     props = cp.cuda.runtime.getDeviceProperties(device.id)
                     gpu_name = props["name"].decode()
 
-                df.at[index, "GPU: DEV Name"] = [gpu_name for i in range(len(df))]
+                df.at[index, "GPU: DEV Name"] = gpu_name
                 cmd = "cat /proc/cpuinfo"
                 cpuinfo = subprocess.check_output(cmd, shell=True).strip()
                 cpu_name = (
                     re.search("\nmodel name.*\n", cpuinfo.decode()).group(0).strip("\n")
                 )
                 cpu_name = cpu_name.replace("model name\t: ", "")
-                df.at[index, "CPU: DEV Name"] = [cpu_name for i in range(len(df))]
+                df.at[index, "CPU: DEV Name"] = cpu_name
 
                 # accelerations[arr_index] = df.at[index,  "GPU accel"]
                 if verbose:

benchmarks/skimage/cucim_restoration_bench.py

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2021-2025, NVIDIA CORPORATION. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2021-2026, NVIDIA CORPORATION. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
 import argparse
@@ -157,8 +157,6 @@ def main(args):
                 module_gpu=cucim.skimage.restoration,
                 run_cpu=run_cpu,
             )
-            results = B.run_benchmark(duration=args.duration)
-            all_results = pd.concat([all_results, results["full"]])
 
         elif function_name in [
             "wiener",
@@ -175,8 +173,8 @@ def main(args):
                 module_gpu=cucim.skimage.restoration,
                 run_cpu=run_cpu,
             )
-            results = B.run_benchmark(duration=args.duration)
-            all_results = pd.concat([all_results, results["full"]])
+        results = B.run_benchmark(duration=args.duration)
+        all_results = pd.concat([all_results, results["full"]])
 
     fbase = os.path.splitext(pfile)[0]
     all_results.to_csv(fbase + ".csv")

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

@@ -1,14 +1,17 @@
 #!/bin/bash
-# SPDX-FileCopyrightText: Copyright (c) 2022-2025, NVIDIA CORPORATION. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2022-2026, NVIDIA CORPORATION. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
+# Use env var if set/non-empty, otherwise default to 10
+MAX_DURATION="${CUCIM_BENCHMARK_MAX_DURATION:-10}"
+
 param_shape=("512,512" "3840,2160" "192,192,192")
 param_filt=(convert_colorspace rgb2hed hed2rgb lab2lch lch2lab xyz2lab lab2xyz rgba2rgb label2rgb)
 param_dt=(float32 uint8)
 for shape in "${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 "$dt" -t 10
+            python cucim_color_bench.py -f "$filt" -i "$shape" -d "$dt" -t "$MAX_DURATION"
         done
     done
 done
@@ -19,7 +22,7 @@ param_dt=("float32" "float64")
 for shape in "${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 "$dt" -t 10
+            python cucim_color_bench.py -f "$filt" -i "$shape" -d "$dt" -t "$MAX_DURATION"
         done
     done
 done

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

@@ -1,14 +1,17 @@
 #!/bin/bash
-# SPDX-FileCopyrightText: Copyright (c) 2022-2025, NVIDIA CORPORATION. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2022-2026, NVIDIA CORPORATION. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
+# Use env var if set/non-empty, otherwise default to 10
+MAX_DURATION="${CUCIM_BENCHMARK_MAX_DURATION:-10}"
+
 param_shape=("512,512" "3840,2160" "3840,2160,3" "192,192,192")
 param_filt=(equalize_adapthist cumulative_distribution equalize_hist rescale_intensity adjust_gamma adjust_log adjust_sigmoid is_low_contrast match_histograms)
 param_dt=(float32 uint8)
 for shape in "${param_shape[@]}"; do
     for filt in "${param_filt[@]}"; do
         for dt in "${param_dt[@]}"; do
-            python cucim_exposure_bench.py -f "$filt" -i "$shape" -d "$dt" -t 10
+            python cucim_exposure_bench.py -f "$filt" -i "$shape" -d "$dt" -t "$MAX_DURATION"
         done
     done
 done

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

@@ -1,14 +1,17 @@
 #!/bin/bash
-# SPDX-FileCopyrightText: Copyright (c) 2022-2025, NVIDIA CORPORATION. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2022-2026, NVIDIA CORPORATION. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
+# Use env var if set/non-empty, otherwise default to 10
+MAX_DURATION="${CUCIM_BENCHMARK_MAX_DURATION:-10}"
+
 param_shape=("512,512" "3840,2160" "3840,2160,3" "192,192,192")
 param_filt=(multiscale_basic_features canny daisy structure_tensor hessian_matrix hessian_matrix_det shape_index corner_kitchen_rosenfeld corner_harris corner_shi_tomasi corner_foerstner corner_peaks match_template blob_dog blob_log blob_doh)
 param_dt=(float64 float32)
 for shape in "${param_shape[@]}"; do
     for filt in "${param_filt[@]}"; do
         for dt in "${param_dt[@]}"; do
-            python cucim_feature_bench.py -f "$filt" -i "$shape" -d "$dt" -t 3
+            python cucim_feature_bench.py -f "$filt" -i "$shape" -d "$dt" -t "$MAX_DURATION"
         done
     done
 done

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

@@ -1,15 +1,18 @@
 #!/bin/bash
-# SPDX-FileCopyrightText: Copyright (c) 2022-2025, NVIDIA CORPORATION. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2022-2026, NVIDIA CORPORATION. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
+# Use env var if set/non-empty, otherwise default to 10
+MAX_DURATION="${CUCIM_BENCHMARK_MAX_DURATION:-10}"
+
 param_shape=("512,512" "3840,2160" "3840,2160,3" "192,192,192")
 param_filt=(gabor gaussian median rank_order unsharp_mask sobel prewitt scharr roberts roberts_pos_diag roberts_neg_diag farid laplace meijering sato frangi hessian threshold_isodata threshold_otsu threshold_yen threshold_local threshold_li threshold_minimum threshold_mean threshold_triangle threshold_niblack threshold_sauvola apply_hysteresis_threshold threshold_multiotsu)
 # param_filt=(rank_order )
 param_dt=(float64 float32 float16)
 for shape in "${param_shape[@]}"; do
     for filt in "${param_filt[@]}"; do
         for dt in "${param_dt[@]}"; do
-            python cucim_filters_bench.py -f "$filt" -i "$shape" -d "$dt" -t 10
+            python cucim_filters_bench.py -f "$filt" -i "$shape" -d "$dt" -t "$MAX_DURATION"
         done
     done
 done

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

@@ -1,14 +1,17 @@
 #!/bin/bash
-# SPDX-FileCopyrightText: Copyright (c) 2022-2025, NVIDIA CORPORATION. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2022-2026, NVIDIA CORPORATION. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
+# Use env var if set/non-empty, otherwise default to 10
+MAX_DURATION="${CUCIM_BENCHMARK_MAX_DURATION:-10}"
+
 param_shape=("512,512" "3840,2160" "3840,2160,3" "192,192,192")
 param_filt=(label regionprops moments moments_central centroid inertia_tensor inertia_tensor_eigvals block_reduce shannon_entropy profile_line)
 param_dt=(float32 uint8)
 for shape in "${param_shape[@]}"; do
     for filt in "${param_filt[@]}"; do
         for dt in "${param_dt[@]}"; do
-            python cucim_measure_bench.py -f "$filt" -i "$shape" -d "$dt" -t 10
+            python cucim_measure_bench.py -f "$filt" -i "$shape" -d "$dt" -t "$MAX_DURATION"
         done
     done
 done
@@ -21,7 +24,7 @@ done
 # for shape in "${param_shape[@]}"; do
 #     for filt in "${param_filt[@]}"; do
 #         for dt in "${param_dt[@]}"; do
-#             python cucim_measure_bench.py -f $filt -i $shape -d $dt -t 10
+#             python cucim_measure_bench.py -f $filt -i $shape -d $dt -t "$MAX_DURATION"
 #         done
 #     done
 # done
@@ -33,7 +36,7 @@ done
 # for shape in "${param_shape[@]}"; do
 #     for filt in "${param_filt[@]}"; do
 #         for dt in "${param_dt[@]}"; do
-#             python cucim_measure_bench.py -f $filt -i $shape -d $dt -t 10
+#             python cucim_measure_bench.py -f $filt -i $shape -d $dt -t "$MAX_DURATION"
 #         done
 #     done
 # done

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

@@ -1,14 +1,17 @@
 #!/bin/bash
-# SPDX-FileCopyrightText: Copyright (c) 2022-2025, NVIDIA CORPORATION. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2022-2026, NVIDIA CORPORATION. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
+# Use env var if set/non-empty, otherwise default to 10
+MAX_DURATION="${CUCIM_BENCHMARK_MAX_DURATION:-10}"
+
 param_shape=("512,512" "3840,2160" "3840,2160,3" "192,192,192")
 param_filt=(structural_similarity mean_squared_error normalized_root_mse peak_signal_noise_ratio normalized_mutual_information)
 param_dt=(float32 uint8)
 for shape in "${param_shape[@]}"; do
     for filt in "${param_filt[@]}"; do
         for dt in "${param_dt[@]}"; do
-            python cucim_metrics_bench.py -f "$filt" -i "$shape" -d "$dt" -t 4
+            python cucim_metrics_bench.py -f "$filt" -i "$shape" -d "$dt" -t "$MAX_DURATION"
         done
     done
 done
@@ -20,7 +23,7 @@ param_dt=(uint8)
 for shape in "${param_shape[@]}"; do
     for filt in "${param_filt[@]}"; do
         for dt in "${param_dt[@]}"; do
-            python cucim_metrics_bench.py -f "$filt" -i "$shape" -d "$dt" -t 4
+            python cucim_metrics_bench.py -f "$filt" -i "$shape" -d "$dt" -t "$MAX_DURATION"
         done
     done
 done