cumo 0.5.0 → 0.5.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 6c2daf87323aec6ca6a8770825e09c5b6bb57d2ab1b6f10b7405d8c1ea1aa07e
-  data.tar.gz: cf8359dbc51fdc479fddf3eb6204db7b4f531757d0da445c50b2e34b8a817a4c
+  metadata.gz: 17cb9dfdf9be41292bcd0204a67c5f919da60588d005d4441ad632767dce504c
+  data.tar.gz: 4c6e388cdb5b3b9f99d45a989a610d63e493fe53b3baad570c7bb2656f72d86c
 SHA512:
-  metadata.gz: c6a3eca253db3d854c8f7ced3ddc2f54836189ccf2065885e14de5acbaa1aced18e4341cf55e7b348f78db11afbd1be1558c3ab94881c3162ca6241b6e91c511
-  data.tar.gz: 5696875ce0d4e3d9d20e483892167b71696bfff6a2dc5594b87628e517e3fa29528f828bc63d8adbf6f0c4b43c47efe6beac31d0344320cf7c2f847d6c0d5d1b
+  metadata.gz: 917adaa087836d673a143364f88fb9ddf91ad84cbcc064b115258e32f4b22e70a63c45f3e655a2e56ad58e70d6a4330f3e63ce177b5ad9e0a0c9c11685b39503
+  data.tar.gz: c5ec5a4179266a1cf4c5f15b5bc0d7e8b0b01a5a69bb7a75ad3c1387e861666da8b571156bc90295f2cf702558f18e332ce7fd68823330f834ea8d9c4f8b6419

data/.rubocop_todo.yml

@@ -1,6 +1,6 @@
 # This configuration was generated by
 # `rubocop --auto-gen-config`
-# on 2025-10-05 08:00:13 UTC using RuboCop version 1.81.1.
+# on 2026-01-09 18:33:26 UTC using RuboCop version 1.82.1.
 # The point is for the user to remove these configuration records
 # one by one as the offenses are removed from the code base.
 # Note that changes in the inspected code, or installation of new
@@ -25,16 +25,8 @@ Bundler/OrderedGems:
   Exclude:
     - 'Gemfile'
 
-# Offense count: 1
-# Configuration parameters: EnforcedStyle, AllowedGems.
-# SupportedStyles: required, forbidden
-Gemspec/DependencyVersion:
-  Exclude:
-    - 'cumo.gemspec'
-
 # Offense count: 2
 # This cop supports safe autocorrection (--autocorrect).
-# Configuration parameters: Severity.
 Gemspec/DeprecatedAttributeAssignment:
   Exclude:
     - '3rd_party/mkmf-cu/mkmf-cu.gemspec'
@@ -49,14 +41,12 @@ Gemspec/DevelopmentDependencies:
 
 # Offense count: 2
 # This cop supports safe autocorrection (--autocorrect).
-# Configuration parameters: Severity.
 Gemspec/RequireMFA:
   Exclude:
     - '3rd_party/mkmf-cu/mkmf-cu.gemspec'
     - 'cumo.gemspec'
 
 # Offense count: 1
-# Configuration parameters: Severity.
 Gemspec/RequiredRubyVersion:
   Exclude:
     - '3rd_party/mkmf-cu/mkmf-cu.gemspec'
@@ -275,13 +265,15 @@ Layout/MultilineOperationIndentation:
     - 'lib/cumo/narray/extra.rb'
     - 'test/narray_test.rb'
 
-# Offense count: 16
+# Offense count: 27
 # This cop supports safe autocorrection (--autocorrect).
 # Configuration parameters: InspectBlocks.
 Layout/RedundantLineBreak:
   Exclude:
     - '3rd_party/mkmf-cu/lib/mkmf-cu/cli.rb'
+    - '3rd_party/mkmf-cu/mkmf-cu.gemspec'
     - '3rd_party/mkmf-cu/test/test_mkmf-cu.rb'
+    - 'ext/cumo/extconf.rb'
     - 'ext/cumo/narray/gen/narray_def.rb'
     - 'test/bit_test.rb'
     - 'test/cudnn_test.rb'
@@ -379,7 +371,7 @@ Lint/ConstantDefinitionInBlock:
   Exclude:
     - 'test/cuda/compiler_test.rb'
 
-# Offense count: 655
+# Offense count: 665
 # Configuration parameters: Only, Ignore.
 Lint/ConstantResolution:
   Enabled: false
@@ -396,7 +388,7 @@ Lint/ErbNewArguments:
   Exclude:
     - 'ext/cumo/narray/gen/erbpp2.rb'
 
-# Offense count: 7
+# Offense count: 15
 Lint/FloatComparison:
   Exclude:
     - 'test/narray_test.rb'
@@ -419,18 +411,20 @@ Lint/NonAtomicFileOperation:
   Exclude:
     - 'lib/cumo/cuda/compiler.rb'
 
-# Offense count: 24
+# Offense count: 34
 # This cop supports unsafe autocorrection (--autocorrect-all).
 # Configuration parameters: AllowedMethods, AllowedPatterns, IgnoredClasses.
 # IgnoredClasses: Time, DateTime
 Lint/NumberConversion:
   Exclude:
+    - '3rd_party/mkmf-cu/lib/mkmf-cu/cli.rb'
     - 'bench/cumo_bench.rb'
     - 'bench/numo_bench.rb'
     - 'ext/cumo/narray/gen/cogen_kernel.rb'
     - 'ext/cumo/narray/gen/erbln.rb'
     - 'lib/cumo/narray/extra.rb'
     - 'test/cudnn_test.rb'
+    - 'test/narray_test.rb'
 
 # Offense count: 2
 # This cop supports safe autocorrection (--autocorrect).
@@ -522,17 +516,6 @@ Naming/MethodParameterName:
     - 'lib/cumo/narray/extra.rb'
     - 'test/ractor_test.rb'
 
-# Offense count: 1
-# Configuration parameters: NamePrefix, ForbiddenPrefixes, AllowedMethods, MethodDefinitionMacros, UseSorbetSigs.
-# NamePrefix: is_, has_, have_, does_
-# ForbiddenPrefixes: is_, has_, have_, does_
-# AllowedMethods: is_a?
-# MethodDefinitionMacros: define_method, define_singleton_method
-Naming/PredicatePrefix:
-  Exclude:
-    - 'spec/**/*'
-    - 'ext/cumo/extconf.rb'
-
 # Offense count: 1
 # Configuration parameters: EnforcedStyle, CheckMethodNames, CheckSymbols, AllowedIdentifiers, AllowedPatterns.
 # SupportedStyles: snake_case, normalcase, non_integer
@@ -719,7 +702,7 @@ Style/Documentation:
     - 'lib/cumo/linalg.rb'
     - 'lib/cumo/narray/extra.rb'
 
-# Offense count: 203
+# Offense count: 202
 # Configuration parameters: AllowedMethods, RequireForNonPublicMethods.
 Style/DocumentationMethod:
   Enabled: false
@@ -767,7 +750,7 @@ Style/FileWrite:
   Exclude:
     - 'lib/cumo/cuda/compiler.rb'
 
-# Offense count: 29
+# Offense count: 27
 # Configuration parameters: AllowedVariables.
 Style/GlobalVars:
   Exclude:
@@ -775,12 +758,11 @@ Style/GlobalVars:
     - 'ext/cumo/narray/gen/cogen.rb'
     - 'ext/cumo/narray/gen/cogen_kernel.rb'
 
-# Offense count: 12
+# Offense count: 10
 # This cop supports safe autocorrection (--autocorrect).
 # Configuration parameters: MinBodyLength, AllowConsecutiveConditionals.
 Style/GuardClause:
   Exclude:
-    - 'ext/cumo/extconf.rb'
     - 'ext/cumo/narray/gen/erbpp2.rb'
     - 'lib/cumo/cuda/link_state.rb'
     - 'lib/cumo/cuda/module.rb'
@@ -865,9 +847,9 @@ Style/InvertibleUnlessCondition:
     - 'lib/cumo/cuda/compiler.rb'
     - 'lib/cumo/cuda/device.rb'
 
-# Offense count: 122
+# Offense count: 119
 # This cop supports safe autocorrection (--autocorrect).
-# Configuration parameters: IgnoreMacros, AllowedMethods, AllowedPatterns, IncludedMacros, AllowParenthesesInMultilineCall, AllowParenthesesInChaining, AllowParenthesesInCamelCaseMethod, AllowParenthesesInStringInterpolation, EnforcedStyle.
+# Configuration parameters: IgnoreMacros, AllowedMethods, AllowedPatterns, IncludedMacros, IncludedMacroPatterns, AllowParenthesesInMultilineCall, AllowParenthesesInChaining, AllowParenthesesInCamelCaseMethod, AllowParenthesesInStringInterpolation, EnforcedStyle.
 # SupportedStyles: require_parentheses, omit_parentheses
 Style/MethodCallWithArgsParentheses:
   Enabled: false
@@ -888,7 +870,7 @@ Style/MethodCalledOnDoEndBlock:
     - 'ext/cumo/narray/gen/cogen_kernel.rb'
     - 'lib/cumo/narray/extra.rb'
 
-# Offense count: 106
+# Offense count: 105
 # This cop supports safe autocorrection (--autocorrect).
 # Configuration parameters: EnforcedStyle.
 # SupportedStyles: if, case, both
@@ -923,13 +905,12 @@ Style/MutableConstant:
     - '3rd_party/mkmf-cu/lib/mkmf-cu/cli.rb'
     - 'test/test_helper.rb'
 
-# Offense count: 14
+# Offense count: 13
 # This cop supports safe autocorrection (--autocorrect).
 # Configuration parameters: EnforcedStyle.
 # SupportedStyles: both, prefix, postfix
 Style/NegatedIf:
   Exclude:
-    - 'ext/cumo/extconf.rb'
     - 'ext/cumo/narray/gen/erbpp2.rb'
     - 'ext/cumo/narray/gen/spec.rb'
     - 'lib/cumo/narray/extra.rb'
@@ -1190,7 +1171,7 @@ Style/StringHashKeys:
   Exclude:
     - '3rd_party/mkmf-cu/test/test_mkmf-cu.rb'
 
-# Offense count: 1376
+# Offense count: 1369
 # This cop supports safe autocorrection (--autocorrect).
 # Configuration parameters: EnforcedStyle, ConsistentQuotesInMultiline.
 # SupportedStyles: single_quotes, double_quotes
@@ -1229,7 +1210,7 @@ Style/TernaryParentheses:
     - 'ext/cumo/narray/gen/narray_def.rb'
     - 'lib/cumo/narray/extra.rb'
 
-# Offense count: 9
+# Offense count: 8
 Style/TopLevelMethodDefinition:
   Exclude:
     - 'bench/cumo_bench.rb'

data/3rd_party/mkmf-cu/lib/mkmf-cu/cli.rb

@@ -1,5 +1,6 @@
 # frozen_string_literal: true
 
+require "mkmf"
 require "open3"
 require_relative "nvcc"
 
@@ -37,29 +38,35 @@ module MakeMakefileCuda
       cmd = "nvcc #{s}"
       if ENV['CUMO_NVCC_GENERATE_CODE']
         cmd << " --generate-code=#{ENV['CUMO_NVCC_GENERATE_CODE']}"
-      elsif ENV['DEBUG']
-        cmd << " -arch=sm_35"
       else
-        # Ref. https://en.wikipedia.org/wiki/CUDA
-        if cuda_version >= Gem::Version.new("13.0")
-          # CUDA 13.0
-          capability = [75, 87, 89, 90, 121]
-        elsif cuda_version >= Gem::Version.new("12.9")
-          # CUDA 12.9
-          capability = [50, 60, 70, 75, 87, 89, 90, 121]
-        elsif cuda_version >= Gem::Version.new("12.8")
-          # CUDA 12.8
-          capability = [50, 60, 70, 75, 87, 89, 90, 120]
-        elsif cuda_version >= Gem::Version.new("12.0")
-          # CUDA 12.0 – 12.6
-          capability = [50, 60, 70, 75, 87, 89, 90]
-        elsif cuda_version >= Gem::Version.new("11.8")
-          # CUDA 11.8
-          capability = [35, 50, 60, 70, 75, 87, 89, 90]
-        else
-          # CUDA 11.0
-          capability = [35, 50, 60, 70, 75, 80]
+        capability = nil
+        if find_executable('nvidia-smi')
+          arch_version = `nvidia-smi --query-gpu=compute_cap --format=csv,noheader`.strip
+          capability = [(arch_version.to_f * 10).to_i] unless arch_version.empty?
         end
+        unless capability
+          # Ref. https://en.wikipedia.org/wiki/CUDA
+          if cuda_version >= Gem::Version.new("13.0")
+            # CUDA 13.0
+            capability = [75, 80, 86, 87, 89, 90, 100, 103, 110, 120, 121]
+          elsif cuda_version >= Gem::Version.new("12.9")
+            # CUDA 12.9
+            capability = [50, 60, 61, 62, 70, 72, 75, 80, 86, 87, 89, 90, 100, 103, 110, 120, 121]
+          elsif cuda_version >= Gem::Version.new("12.8")
+            # CUDA 12.8
+            capability = [50, 60, 61, 62, 70, 72, 75, 80, 86, 87, 89, 90, 100, 103, 110, 120]
+          elsif cuda_version >= Gem::Version.new("12.0")
+            # CUDA 12.0 – 12.6
+            capability = [50, 60, 61, 62, 70, 72, 75, 80, 86, 87, 89, 90]
+          elsif cuda_version >= Gem::Version.new("11.8")
+            # CUDA 11.8
+            capability = [35, 50, 60, 61, 62, 70, 72, 75, 80, 86, 87, 89, 90]
+          else
+            # CUDA 11.0
+            capability = [35, 50, 60, 61, 62, 70, 72, 75, 80]
+          end
+        end
+
         capability.each do |arch|
           cmd << " --generate-code=arch=compute_#{arch},code=sm_#{arch}"
         end

data/CHANGELOG.md

@@ -1,3 +1,31 @@
+# 0.5.2 (2025/01/25)
+
+Fixes:
+
+* Backport: Add support for copy on write with store_binary and frozen string
+* Remove unnecessary debug code
+* Fix capability list
+* Build only with supported capabilities to reduce compilation time
+* Fix SEGV when calling {mean, var, stddev, rms} on a single-element array (#154)
+* Suppress warning message for deprecated declarations
+* Fix variable typo in complex log2 and log10 functions (#152)
+
+# 0.5.1 (2025/12/30)
+
+Enhancements:
+
+* Add CUDA 13 support (#153)
+* Add cuDNN 9 support
+
+Fixes:
+
+* Backport: fix example code
+* Backport: fix example code
+* Backport: fix doc
+* Backport: fix documents
+* Backport: fix document of logseq
+* Backport: trim comment out
+
 # 0.5.0 (2025/11/01)
 
 Fixes:

data/Dockerfile

@@ -0,0 +1,34 @@
+FROM nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04
+
+ARG RUBY_VERSION=3.4.7
+
+ENV DEBIAN_FRONTEND=noninteractive
+ENV RBENV_ROOT="/root/.rbenv"
+ENV PATH="${RBENV_ROOT}/bin:${RBENV_ROOT}/shims:${PATH}"
+
+ENV CUDA_PATH=/usr/local/cuda
+ENV LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH}
+ENV CPATH=/usr/local/cuda/include:${CPATH}
+ENV LIBRARY_PATH=/usr/local/cuda/lib64:${LIBRARY_PATH}
+
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    git \
+    build-essential \
+    wget \
+    curl \
+    vim \
+    ca-certificates \
+    libssl-dev \
+    libreadline-dev \
+    zlib1g-dev \
+    libyaml-dev \
+    libffi-dev \
+    && rm -rf /var/lib/apt/lists/*
+
+RUN git clone --depth 1 https://github.com/rbenv/ruby-build.git && \
+    cd ruby-build/bin && ./ruby-build ${RUBY_VERSION} /usr && \
+    git config --global --add safe.directory /workspace
+
+WORKDIR /workspace
+
+CMD ["/bin/bash"]

data/cumo.gemspec

@@ -19,7 +19,7 @@ Gem::Specification.new do |spec|
   spec.required_ruby_version = ">= 3.0.0"
 
   spec.files = `git ls-files -z`.split("\x0").reject do |f|
-    f.match(%r{^(test|spec|features)/})
+    f.match(%r{^(test|spec|features|docker)/})
   end
   spec.test_files    = `git ls-files -- {test,spec,features}/*`.split("\n")
   spec.bindir        = "exe"

data/docker-build.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+
+script_dir=$(cd $(dirname ${BASH_SOURCE:-$0}); pwd)
+docker build -t cumo-dev ${script_dir}

data/docker-launch.sh

@@ -0,0 +1,4 @@
+#!/bin/bash
+
+project_dir="$(cd $(dirname ${BASH_SOURCE:-$0}); pwd)/"
+docker run --gpus all -v $project_dir:/workspace -it cumo-dev bash

data/docs/src-tree.md

@@ -6,7 +6,7 @@
     * Technically, it is not possible to use CRuby API such as `VALUE` in .cu files.
         * CRuby API is not callable from CUDA kernel because they do not have `__device__` modifier.
         * nvcc does not support `#include RUBY_EXTCONF_H`, so can not include `ruby.h`.
-    * (RULE) It is allowed to use C++14 codes in .cu files.
+    * (RULE) It is allowed to use C++17 codes in .cu files.
 * Rest of `*.{h,c}` files are for host (CPU).
     * Call C wrapper functions defined in .cu files.
     * It can use CRuby API.

data/ext/cumo/cuda/cudnn_impl.cpp

@@ -74,6 +74,25 @@ cumo_cuda_cudnn_CreateTensorDescriptor(
         status = cudnnSetTensor4dDescriptor(
                 *desc, CUDNN_TENSOR_NCHW, cudnn_dtype, shape[0], shape[1], shape[2], shape[3]);
     }
+    else if (ndim < 4) {
+        // cuDNN 9 fix: Force 4D (N, C, H, W)
+        int pad_shape[4] = {1, 1, 1, 1};
+
+        if (ndim == 1) {
+            // 1D: arrays are treated as "Channel" (1, C, 1, 1)
+            pad_shape[1] = (int)(shape[0]);
+        } else {
+            // 2D: [N, C] -> [N, C, 1, 1]
+            // 3D: [N, C, H] -> [N, C, H, 1]
+            for (int idim = 0; idim < ndim; ++idim) {
+                pad_shape[idim] = (int)(shape[idim]);
+            }
+        }
+
+        status = cudnnSetTensor4dDescriptor(
+                *desc, CUDNN_TENSOR_NCHW, cudnn_dtype,
+                pad_shape[0], pad_shape[1], pad_shape[2], pad_shape[3]);
+    }
     else {
         int int_shape[CUMO_NA_MAX_DIMENSION];
         for (int idim = 0; idim < ndim; ++idim) {
@@ -514,8 +533,11 @@ cumo_cuda_cudnn_FindConvolutionBackwardFilterAlgorithm(
 // TODO(sonots): Support other than 4, 5 dimensional arrays by reshaping into 4-dimensional arrays as Chainer does.
 cudnnBatchNormMode_t
 cumo_cuda_cudnn_GetBatchNormMode(size_t ndim, int* axis) {
-    if (ndim == 1 && axis[0] == 0) {  // (1, channels, (depth, )height, width)
-        return CUDNN_BATCHNORM_PER_ACTIVATION;
+    if (ndim == 1) {
+        return CUDNN_BATCHNORM_SPATIAL;
+    }
+    if (ndim == 2) {
+        return CUDNN_BATCHNORM_SPATIAL;
     }
     if ((ndim == 3 && axis[0] == 0 && axis[1] == 2 && axis[2] == 3) ||
         (ndim == 4 && axis[0] == 0 && axis[1] == 2 && axis[2] == 3 && axis[3] == 4)) {  // (1, channels, (1, )1, 1)
@@ -533,7 +555,7 @@ cumo_cuda_cudnn_CreateBNTensorDescriptor(
 {
     cudnnStatus_t status = CUDNN_STATUS_SUCCESS;
     status = cudnnCreateTensorDescriptor(desc);
-    if (status = CUDNN_STATUS_SUCCESS) return status;
+    if (status == CUDNN_STATUS_SUCCESS) return status;
 
     status = cudnnDeriveBNTensorDescriptor(*desc, x_desc, mode);
     return status;

data/ext/cumo/cuda/driver.c

@@ -33,7 +33,11 @@ rb_cuCtxCreate(VALUE self, VALUE flags, VALUE dev)
     CUcontext _pctx;
     CUresult status;
 
+#if defined(CUDA_VERSION) && CUDA_VERSION >= 13000
+    status = cuCtxCreate(&_pctx, NULL, _flags, _dev);
+#else
     status = cuCtxCreate(&_pctx, _flags, _dev);
+#endif
 
     check_status(status);
     return SIZET2NUM((size_t)_pctx);
@@ -418,5 +422,9 @@ Init_cumo_cuda_driver()
 
     cuInit(0);
     cuDeviceGet(&cuDevice, 0);
+#if defined(CUDA_VERSION) && CUDA_VERSION >= 13000
+    cuCtxCreate(&context, NULL, 0, cuDevice);
+#else
     cuCtxCreate(&context, 0, cuDevice);
+#endif
 }

data/ext/cumo/depend.erb

@@ -55,6 +55,6 @@ run-ctest : <%= __dir__ %>/cuda/memory_pool_impl_test.exe
 	./$<
 
 <%= __dir__ %>/cuda/memory_pool_impl_test.exe: <%= __dir__ %>/cuda/memory_pool_impl_test.cpp <%= __dir__ %>/cuda/memory_pool_impl.cpp <%= __dir__ %>/cuda/memory_pool_impl.hpp
-	nvcc -std=c++14 <%= ENV['DEBUG'] ? '-g -O0 --compiler-options -Wall' : '' %> -L. -L$(libdir) -I. $(INCFLAGS) -o $@ $< <%= __dir__ %>/cuda/memory_pool_impl.cpp
+	nvcc -std=c++17 <%= ENV['DEBUG'] ? '-g -O0 --compiler-options -Wall' : '' %> -L. -L$(libdir) -I. $(INCFLAGS) -o $@ $< <%= __dir__ %>/cuda/memory_pool_impl.cpp
 
 CLEANOBJS = <%= __dir__ %>/*.o <%= __dir__ %>/*/*.o <%= __dir__ %>/*/*/*.o <%= __dir__ %>/*.bak <%= __dir__ %>/narray/types/*.c <%= __dir__ %>/narray/types/*_kernel.cu <%= __dir__ %>/*.exe <%= __dir__ %>/*/*.exe

data/ext/cumo/extconf.rb

@@ -29,7 +29,7 @@ MakeMakefileCuda.install!(cxx: true)
 if ENV['DEBUG']
   $CFLAGS << " -g -O0 -Wall"
 end
-$CXXFLAGS << " -std=c++14"
+$CXXFLAGS << " -std=c++17"
 #$CFLAGS=" $(cflags) -O3 -m64 -msse2 -funroll-loops"
 #$CFLAGS=" $(cflags) -O3"
 $INCFLAGS = "-I$(srcdir)/include -I$(srcdir)/narray -I$(srcdir)/cuda #{$INCFLAGS}"

data/ext/cumo/include/cumo/cuda/cumo_thrust.hpp

@@ -26,10 +26,12 @@ class cumo_thrust_strided_range
 {
     public:
 
-    typedef typename thrust::iterator_difference<Iterator>::type difference_type;
+    typedef typename thrust::iterator_traits<Iterator>::difference_type difference_type;
 
-    struct stride_functor : public thrust::unary_function<difference_type,difference_type>
+    struct stride_functor
     {
+        using argument_type = difference_type;
+        using result_type   = difference_type;
         difference_type stride;
 
         stride_functor(difference_type stride)
@@ -86,8 +88,10 @@ struct cumo_thrust_minmax_pair
 // returns a cumo_thrust_minmax_pair whose minimum and maximum values
 // are initialized to x.
 template <typename T>
-struct cumo_thrust_minmax_unary_op : public thrust::unary_function< T, cumo_thrust_minmax_pair<T> >
+struct cumo_thrust_minmax_unary_op
 {
+    using argument_type = T;
+    using result_type   = cumo_thrust_minmax_pair<T>;
     __host__ __device__ cumo_thrust_minmax_pair<T> operator()(const T& x) const
     {
         cumo_thrust_minmax_pair<T> result;
@@ -102,8 +106,11 @@ struct cumo_thrust_minmax_unary_op : public thrust::unary_function< T, cumo_thru
 // maximum values are the min() and max() respectively of
 // the minimums and maximums of the input pairs
 template <typename T>
-struct cumo_thrust_minmax_binary_op : public thrust::binary_function< cumo_thrust_minmax_pair<T>, cumo_thrust_minmax_pair<T>, cumo_thrust_minmax_pair<T> >
+struct cumo_thrust_minmax_binary_op
 {
+    using first_argument_type  = cumo_thrust_minmax_pair<T>;
+    using second_argument_type = cumo_thrust_minmax_pair<T>;
+    using result_type          = cumo_thrust_minmax_pair<T>;
     __host__ __device__ cumo_thrust_minmax_pair<T> operator()(const cumo_thrust_minmax_pair<T>& x, const cumo_thrust_minmax_pair<T>& y) const
     {
         cumo_thrust_minmax_pair<T> result;
@@ -157,10 +164,10 @@ struct cumo_thrust_variance_unary_op
 // all values that have been agregated so far
 template <typename T>
 struct cumo_thrust_variance_binary_op
-    : public thrust::binary_function<const cumo_thrust_variance_data<T>&,
-                                     const cumo_thrust_variance_data<T>&,
-                                           cumo_thrust_variance_data<T> >
 {
+    using first_argument_type  = const cumo_thrust_variance_data<T>&;
+    using second_argument_type = const cumo_thrust_variance_data<T>&;
+    using result_type          = cumo_thrust_variance_data<T>;
     __host__ __device__
     cumo_thrust_variance_data<T> operator()(const cumo_thrust_variance_data<T>& x, const cumo_thrust_variance_data <T>& y) const
     {

data/ext/cumo/include/cumo/cuda/cumo_thrust_complex.hpp

@@ -49,10 +49,10 @@ struct cumo_thrust_complex_variance_unary_op
 // all values that have been agregated so far
 template <typename T, typename R>
 struct cumo_thrust_complex_variance_binary_op
-    : public thrust::binary_function<const cumo_thrust_complex_variance_data<T,R>&,
-                                     const cumo_thrust_complex_variance_data<T,R>&,
-                                           cumo_thrust_complex_variance_data<T,R> >
 {
+    using first_argument_type  = const cumo_thrust_complex_variance_data<T,R>&;
+    using second_argument_type = const cumo_thrust_complex_variance_data<T,R>&;
+    using result_type          = cumo_thrust_complex_variance_data<T,R>;
     __host__ __device__
     cumo_thrust_complex_variance_data<T,R> operator()(const cumo_thrust_complex_variance_data<T,R>& x, const cumo_thrust_complex_variance_data<T,R>& y) const
     {

data/ext/cumo/include/cumo/narray.h

@@ -226,6 +226,7 @@ typedef struct {
 typedef struct {
     cumo_narray_t base;
     char    *ptr;
+    bool     owned;
 } cumo_narray_data_t;
 
 
@@ -360,6 +361,7 @@ _cumo_na_get_narray_t(VALUE obj, unsigned char cumo_na_type)
 #define CUMO_NA_DATA(na)             ((cumo_narray_data_t*)(na))
 #define CUMO_NA_VIEW(na)             ((cumo_narray_view_t*)(na))
 #define CUMO_NA_DATA_PTR(na)         (CUMO_NA_DATA(na)->ptr)
+#define CUMO_NA_DATA_OWNED(na)       (CUMO_NA_DATA(na)->owned)
 #define CUMO_NA_VIEW_DATA(na)        (CUMO_NA_VIEW(na)->data)
 #define CUMO_NA_VIEW_OFFSET(na)      (CUMO_NA_VIEW(na)->offset)
 #define CUMO_NA_VIEW_STRIDX(na)      (CUMO_NA_VIEW(na)->stridx)

data/ext/cumo/include/cumo/types/complex.h

@@ -166,14 +166,14 @@ static inline dtype c_log(dtype x) {
 static inline dtype c_log2(dtype x) {
     dtype z;
     z = c_log(x);
-    z = c_mul_r(x,M_LOG2E);
+    z = c_mul_r(z,M_LOG2E);
     return z;
 }
 
 static inline dtype c_log10(dtype x) {
     dtype z;
     z = c_log(x);
-    z = c_mul_r(x,M_LOG10E);
+    z = c_mul_r(z,M_LOG10E);
     return z;
 }
 

data/ext/cumo/include/cumo/types/complex_macro_kernel.h

@@ -157,18 +157,27 @@ __host__ __device__ static inline dtype f_seq(dtype x, dtype y, double c)
 /* --------- thrust ----------------- */
 #include "cumo/cuda/cumo_thrust_complex.hpp"
 
-struct cumo_thrust_plus : public thrust::binary_function<dtype, dtype, dtype>
+struct cumo_thrust_plus
 {
+    using first_argument_type  = dtype;
+    using second_argument_type = dtype;
+    using result_type          = dtype;
     __host__ __device__ dtype operator()(dtype x, dtype y) { return m_add(x,y); }
 };
 
-struct cumo_thrust_multiplies : public thrust::binary_function<dtype, dtype, dtype>
+struct cumo_thrust_multiplies
 {
+    using first_argument_type  = dtype;
+    using second_argument_type = dtype;
+    using result_type          = dtype;
     __host__ __device__ dtype operator()(dtype x, dtype y) { return m_mul(x,y); }
 };
 
-struct cumo_thrust_multiplies_mulsum_nan : public thrust::binary_function<dtype, dtype, dtype>
+struct cumo_thrust_multiplies_mulsum_nan
 {
+    using first_argument_type  = dtype;
+    using second_argument_type = dtype;
+    using result_type          = dtype;
     __host__ __device__ dtype operator()(dtype x, dtype y) {
         if (not_nan(x) && not_nan(y)) {
             return m_mul(x, y);
@@ -178,8 +187,10 @@ struct cumo_thrust_multiplies_mulsum_nan : public thrust::binary_function<dtype,
     }
 };
 
-struct cumo_thrust_square : public thrust::unary_function<dtype, dtype>
+struct cumo_thrust_square
 {
+    using argument_type = dtype;
+    using result_type   = dtype;
     __host__ __device__ rtype operator()(const dtype& x) const { return c_abs_square(x); }
 };
 

data/ext/cumo/include/cumo/types/real_accum_kernel.h

@@ -72,18 +72,27 @@ __host__ __device__ static inline dtype f_minimum_nan(dtype x, dtype y)
 /* --------- thrust ----------------- */
 #include "cumo/cuda/cumo_thrust.hpp"
 
-struct cumo_thrust_plus : public thrust::binary_function<dtype, dtype, dtype>
+struct cumo_thrust_plus
 {
+    using first_argument_type  = dtype;
+    using second_argument_type = dtype;
+    using result_type          = dtype;
     __host__ __device__ dtype operator()(dtype x, dtype y) { return m_add(x,y); }
 };
 
-struct cumo_thrust_multiplies : public thrust::binary_function<dtype, dtype, dtype>
+struct cumo_thrust_multiplies
 {
+    using first_argument_type  = dtype;
+    using second_argument_type = dtype;
+    using result_type          = dtype;
     __host__ __device__ dtype operator()(dtype x, dtype y) { return m_mul(x,y); }
 };
 
-struct cumo_thrust_multiplies_mulsum_nan : public thrust::binary_function<dtype, dtype, dtype>
+struct cumo_thrust_multiplies_mulsum_nan
 {
+    using first_argument_type  = dtype;
+    using second_argument_type = dtype;
+    using result_type          = dtype;
     __host__ __device__ dtype operator()(dtype x, dtype y) {
         if (not_nan(x) && not_nan(y)) {
             return m_mul(x, y);
@@ -93,8 +102,10 @@ struct cumo_thrust_multiplies_mulsum_nan : public thrust::binary_function<dtype,
     }
 };
 
-struct cumo_thrust_square : public thrust::unary_function<dtype, dtype>
+struct cumo_thrust_square
 {
+    using argument_type = dtype;
+    using result_type   = dtype;
     __host__ __device__ rtype operator()(const dtype& x) const { return m_square(x); }
 };