Update compute-graph capture (#286)

* update compiler notes * suport arbitrary sized utl tuple * add type_name metafunction to get typename at compile-time * restructure compute_graph & functor files & deps * new aliasing logic using compile-time hashing * update tests * fix hip compilation * try to fix risc-v ci * temporarily allow matmul test to fail on sycl (for pocl) * reduce compilation jobs to avoid oom on gcc
alifahrri · Jul 7, 2024 · bcc2ddc · bcc2ddc
1 parent a1b6f10
commit bcc2ddc
Show file tree

Hide file tree

Showing 143 changed files with 4,510 additions and 2,790 deletions.
diff --git a/.github/workflows/gcc.yml b/.github/workflows/gcc.yml
@@ -32,7 +32,7 @@ jobs:
         run: bash scripts/download_doctest_header.sh -d include/
       - name: build
         # use single thread to minimize memory usage, Process completed with exit code 137.
-        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake ../.. && make -j2 VERBOSE=1'
+        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake ../.. && make -j1 VERBOSE=1'
       - name: run tests
         run: |
           ./dockcross-${{ matrix.name }} bash -c 'build/${{ matrix.name }}/tests/array/numeric-tests-doctest'
@@ -62,7 +62,7 @@ jobs:
         run: bash scripts/download_doctest_header.sh -d include/
       - name: build
         # use single thread to minimize memory usage, Process completed with exit code 137.
-        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_TESTING_GENERIC_NDARRAY=ON -DNMTOOLS_BUILD_VIEW_TESTS=ON ../.. && make -j2 VERBOSE=1 numeric-tests-view-doctest'
+        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_TESTING_GENERIC_NDARRAY=ON -DNMTOOLS_BUILD_VIEW_TESTS=ON ../.. && make -j1 VERBOSE=1 numeric-tests-view-doctest'
       - name: run tests
         run: |
           ./dockcross-${{ matrix.name }} bash -c 'build/${{ matrix.name }}/tests/view/numeric-tests-view-doctest'
@@ -87,7 +87,7 @@ jobs:
         run: bash scripts/download_doctest_header.sh -d include/
       - name: build
         # use single thread to minimize memory usage, Process completed with exit code 137.
-        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_BUILD_VIEW_TESTS=ON -DNMTOOLS_TESTING_NOSTL=ON ../.. && make -j2 VERBOSE=1 numeric-tests-view-doctest'
+        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_BUILD_VIEW_TESTS=ON -DNMTOOLS_TESTING_NOSTL=ON ../.. && make -j1 VERBOSE=1 numeric-tests-view-doctest'
       - name: run tests
         run: |
           ./dockcross-${{ matrix.name }} bash -c 'build/${{ matrix.name }}/tests/view/numeric-tests-view-doctest'
@@ -111,7 +111,7 @@ jobs:
       - name: setup doctest
         run: bash scripts/download_doctest_header.sh -d include/
       - name: build
-        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DCMAKE_TOOLCHAIN_FILE=${PWD}/../../cmake/toolchains/gcc-werror.cmake -DNMTOOLS_TEST_ALL=OFF -DNMTOOLS_BUILD_INDEX_TESTS=ON ../.. && make -j2 VERBOSE=1 numeric-tests-index-doctest'
+        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DCMAKE_TOOLCHAIN_FILE=${PWD}/../../cmake/toolchains/gcc-werror.cmake -DNMTOOLS_TEST_ALL=OFF -DNMTOOLS_BUILD_INDEX_TESTS=ON ../.. && make -j1 VERBOSE=1 numeric-tests-index-doctest'
       - name: run tests
         run: |
           ./dockcross-${{ matrix.name }} bash -c 'build/${{ matrix.name }}/tests/index/numeric-tests-index-doctest'
@@ -135,7 +135,7 @@ jobs:
       - name: setup doctest
         run: bash scripts/download_doctest_header.sh -d include/
       - name: build
-        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DCMAKE_TOOLCHAIN_FILE=${PWD}/../../cmake/toolchains/gcc-werror.cmake -DNMTOOLS_TEST_ALL=OFF -DNMTOOLS_BUILD_FUNCTIONAL_TESTS=ON ../.. && make -j2 VERBOSE=1 numeric-tests-functional-doctest'
+        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DCMAKE_TOOLCHAIN_FILE=${PWD}/../../cmake/toolchains/gcc-werror.cmake -DNMTOOLS_TEST_ALL=OFF -DNMTOOLS_BUILD_FUNCTIONAL_TESTS=ON ../.. && make -j1 VERBOSE=1 numeric-tests-functional-doctest'
       - name: run tests
         run: |
           ./dockcross-${{ matrix.name }} bash -c 'build/${{ matrix.name }}/tests/functional/numeric-tests-functional-doctest'
@@ -160,7 +160,7 @@ jobs:
         run: bash scripts/download_doctest_header.sh -d include/
       - name: build
         # use single thread to minimize memory usage, Process completed with exit code 137.
-        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_TESTING_GENERIC_NDARRAY=ON -DNMTOOLS_BUILD_VIEW_TESTS=ON -DNMTOOLS_TESTING_NOSTL=ON ../.. && make -j2 VERBOSE=1 numeric-tests-view-doctest'
+        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_TESTING_GENERIC_NDARRAY=ON -DNMTOOLS_BUILD_VIEW_TESTS=ON -DNMTOOLS_TESTING_NOSTL=ON ../.. && make -j1 VERBOSE=1 numeric-tests-view-doctest'
       - name: run tests
         run: |
           ./dockcross-${{ matrix.name }} bash -c 'build/${{ matrix.name }}/tests/view/numeric-tests-view-doctest'
@@ -184,7 +184,7 @@ jobs:
       - name: setup doctest
         run: bash scripts/download_doctest_header.sh -d include/
       - name: build
-        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_BUILD_META_TESTS=OFF -DNMTOOLS_BUILD_UTL_TESTS=OFF -DNMTOOLS_TEST_ALL=OFF -DNMTOOLS_TEST_ARRAY_EVAL=ON -DNMTOOLS_TESTING_GENERIC_NDARRAY=ON -DNMTOOLS_TESTING_NOSTL=ON ../.. && make -j2 VERBOSE=1'
+        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_BUILD_META_TESTS=OFF -DNMTOOLS_BUILD_UTL_TESTS=OFF -DNMTOOLS_TEST_ALL=OFF -DNMTOOLS_TEST_ARRAY_EVAL=ON -DNMTOOLS_TESTING_GENERIC_NDARRAY=ON -DNMTOOLS_TESTING_NOSTL=ON ../.. && make -j1 VERBOSE=1'
       - name: run tests
         run: |
           ./dockcross-${{ matrix.name }} bash -c 'build/${{ matrix.name }}/tests/array/numeric-tests-doctest'
@@ -208,7 +208,7 @@ jobs:
       - name: setup doctest
         run: bash scripts/download_doctest_header.sh -d include/
       - name: build
-        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_BUILD_META_TESTS=OFF -DNMTOOLS_BUILD_UTL_TESTS=OFF -DNMTOOLS_TEST_ALL=OFF -DNMTOOLS_TEST_ARRAY_EVAL=ON -DNMTOOLS_TESTING_GENERIC_NDARRAY=ON -DNMTOOLS_TESTING_NOSTL=OFF ../.. && make -j2 VERBOSE=1'
+        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_BUILD_META_TESTS=OFF -DNMTOOLS_BUILD_UTL_TESTS=OFF -DNMTOOLS_TEST_ALL=OFF -DNMTOOLS_TEST_ARRAY_EVAL=ON -DNMTOOLS_TESTING_GENERIC_NDARRAY=ON -DNMTOOLS_TESTING_NOSTL=OFF ../.. && make -j1 VERBOSE=1'
       - name: run tests
         run: |
           ./dockcross-${{ matrix.name }} bash -c 'build/${{ matrix.name }}/tests/array/numeric-tests-doctest'
@@ -232,7 +232,7 @@ jobs:
       - name: setup doctest
         run: bash scripts/download_doctest_header.sh -d include/
       - name: build
-        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_BUILD_CONSTEXPR_TESTS=ON -DNMTOOLS_BUILD_META_TESTS=OFF -DNMTOOLS_BUILD_UTL_TESTS=OFF -DNMTOOLS_TEST_ALL=OFF -DNMTOOLS_TEST_ARRAY_EVAL=ON ../.. && make -j2 VERBOSE=1'
+        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_BUILD_CONSTEXPR_TESTS=ON -DNMTOOLS_BUILD_META_TESTS=OFF -DNMTOOLS_BUILD_UTL_TESTS=OFF -DNMTOOLS_TEST_ALL=OFF -DNMTOOLS_TEST_ARRAY_EVAL=ON ../.. && make -j1 VERBOSE=1'
       - name: run tests
         run: |
           ./dockcross-${{ matrix.name }} bash -c 'build/${{ matrix.name }}/tests/constexpr/numeric-tests-constexpr-doctest'
@@ -256,7 +256,7 @@ jobs:
       - name: setup doctest
         run: bash scripts/download_doctest_header.sh -d include/
       - name: build
-        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_BUILD_CONSTEXPR_TESTS=ON -DNMTOOLS_BUILD_META_TESTS=OFF -DNMTOOLS_BUILD_UTL_TESTS=OFF -DNMTOOLS_TEST_ALL=OFF -DNMTOOLS_TEST_ARRAY_EVAL=ON -DNMTOOLS_TESTING_GENERIC_NDARRAY=ON -DNMTOOLS_TESTING_NOSTL=ON ../.. && make -j2 VERBOSE=1'
+        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_BUILD_CONSTEXPR_TESTS=ON -DNMTOOLS_BUILD_META_TESTS=OFF -DNMTOOLS_BUILD_UTL_TESTS=OFF -DNMTOOLS_TEST_ALL=OFF -DNMTOOLS_TEST_ARRAY_EVAL=ON -DNMTOOLS_TESTING_GENERIC_NDARRAY=ON -DNMTOOLS_TESTING_NOSTL=ON ../.. && make -j1 VERBOSE=1'
       - name: run tests
         run: |
           ./dockcross-${{ matrix.name }} bash -c 'build/${{ matrix.name }}/tests/constexpr/numeric-tests-constexpr-doctest'
@@ -280,7 +280,7 @@ jobs:
       - name: setup doctest
         run: bash scripts/download_doctest_header.sh -d include/
       - name: build
-        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_BUILD_CONSTEXPR_TESTS=ON  -DNMTOOLS_BUILD_META_TESTS=OFF -DNMTOOLS_BUILD_UTL_TESTS=OFF -DNMTOOLS_TEST_ALL=OFF -DNMTOOLS_TEST_ARRAY_EVAL=ON -DNMTOOLS_TESTING_GENERIC_NDARRAY=ON -DNMTOOLS_TESTING_NOSTL=OFF ../.. && make -j2 VERBOSE=1'
+        run: ./dockcross-${{ matrix.name }} bash -c 'mkdir -p build/${{ matrix.name }} && cd build/${{ matrix.name }}/ && cmake -DNMTOOLS_BUILD_CONSTEXPR_TESTS=ON  -DNMTOOLS_BUILD_META_TESTS=OFF -DNMTOOLS_BUILD_UTL_TESTS=OFF -DNMTOOLS_TEST_ALL=OFF -DNMTOOLS_TEST_ARRAY_EVAL=ON -DNMTOOLS_TESTING_GENERIC_NDARRAY=ON -DNMTOOLS_TESTING_NOSTL=OFF ../.. && make -j1 VERBOSE=1'
       - name: run tests
         run: |
           ./dockcross-${{ matrix.name }} bash -c 'build/${{ matrix.name }}/tests/constexpr/numeric-tests-constexpr-doctest'
diff --git a/.gitignore b/.gitignore
@@ -24,4 +24,5 @@ pocl
 *.ll
 llvm.sh
 install
-amdgpu-*
+amdgpu-*
+*.dot
diff --git a/docs/compiler-notes.txt b/docs/compiler-notes.txt
@@ -16,7 +16,7 @@ Documenting various note on behaviour difference between clang & gcc (or with so
   https://godbolt.org/z/WYxahvxTx
 
 1. constexpr math  
-  error: constexpr function never produces a constant expression
+  error: constexpr function never produces a constant expression  
   https://godbolt.org/z/dxqq3T
 
 1. Constexpr if with a non-bool condition  
@@ -42,7 +42,7 @@ Documenting various note on behaviour difference between clang & gcc (or with so
   https://godbolt.org/z/asrh7q
 
 1. clang and gcc disagree on if constexpr inside lambda  
-  https://godbolt.org/z/8qc13s
+  https://godbolt.org/z/8qc13s  
   https://stackoverflow.com/questions/59647445/if-constexpr-with-static-assert-in-lambda-which-compiler-is-correct
 
 1. clang and gcc disagree on partial specialization & sfinae, makes ambiguous instantiation on gcc but no issue on clang  
@@ -97,4 +97,8 @@ Documenting various note on behaviour difference between clang & gcc (or with so
 
 1. clang vs gcc disagree on constexpr evaluation  
   (cumbersome to minimally reproduce, document as issue instead)  
-  https://github.com/alifahrri/nmtools/issues/282
+  https://github.com/alifahrri/nmtools/issues/282
+
+1. clang vs gcc disagree on capturing constexpr value in lambda expression  
+  clang ok, gcc not ok  
+  https://godbolt.org/z/a1o8P9957
diff --git a/include/nmtools/array/array/atleast_nd.hpp b/include/nmtools/array/array/atleast_nd.hpp
@@ -0,0 +1,34 @@
+#ifndef NMTOOLS_ARRAY_ARRAY_ATLEAST_ND_HPP
+#define NMTOOLS_ARRAY_ARRAY_ATLEAST_ND_HPP
+
+#include "nmtools/array/view/atleast_nd.hpp"
+#include "nmtools/array/eval.hpp"
+
+namespace nmtools::array
+{
+    /**
+     * @brief Eagerly compute atleast_nd.
+     * 
+     * @tparam output_t 
+     * @tparam context_t 
+     * @tparam array_t 
+     * @param array   Input array
+     * @param context Evaluation context
+     * @param output 
+     * @return constexpr auto 
+     */
+    template <typename output_t=none_t, typename context_t=none_t, typename resolver_t=eval_result_t<>,
+        typename array_t, typename nd_t>
+    constexpr auto atleast_nd(const array_t& array, nd_t nd
+        , context_t&& context=context_t{}, output_t&& output=output_t{},meta::as_value<resolver_t> resolver=meta::as_value_v<resolver_t>)
+    {
+        auto viewed = view::atleast_nd(array,nd);
+        return eval(viewed
+            ,nmtools::forward<context_t>(context)
+            ,nmtools::forward<output_t>(output)
+            ,resolver
+        );
+    } // atleast_nd
+} // namespace nmtools::array
+
+#endif // NMTOOLS_ARRAY_ARRAY_ATLEAST_ND_HPP
diff --git a/include/nmtools/array/eval.hpp b/include/nmtools/array/eval.hpp
@@ -452,7 +452,7 @@ namespace nmtools::meta
                 // default impl of make_fixed_ndarray only support integral constant for now
                 using stype = ct<s>;
                 if constexpr (is_void_v<type>) {
-                    using type = make_tuple_type_t<stype>;
+                    using type = nmtools_tuple<stype>;
                     return as_value_v<type>;
                 } else {
                     using type = append_type_t<type,stype>;
@@ -496,7 +496,7 @@ namespace nmtools::meta
                 // default impl of make_fixed_ndarray only support integral constant for now
                 using stype = ct<s>;
                 if constexpr (is_void_v<type>) {
-                    using type = make_tuple_type_t<stype>;
+                    using type = nmtools_tuple<stype>;
                     return as_value_v<type>;
                 } else {
                     using type = append_type_t<type,stype>;

diff --git a/include/nmtools/array/functional.hpp b/include/nmtools/array/functional.hpp
@@ -1,6 +1,8 @@
 #ifndef NMTOOLS_ARRAY_FUNCTIONAL_HPP
 #define NMTOOLS_ARRAY_FUNCTIONAL_HPP
 
+#include "nmtools/array/functional/indexing.hpp"
+
 #include "nmtools/array/functional/activations/celu.hpp"
 #include "nmtools/array/functional/activations/elu.hpp"
 #include "nmtools/array/functional/activations/hardshrink.hpp"
@@ -78,6 +80,5 @@
 #include "nmtools/array/functional/squeeze.hpp"
 #include "nmtools/array/functional/where.hpp"
 #include "nmtools/array/functional/zeros.hpp"
-#include "nmtools/array/functional/indexing.hpp"
 
 #endif // NMTOOLS_ARRAY_FUNCTIONAL_HPP
diff --git a/include/nmtools/array/functional/batch_norm.hpp b/include/nmtools/array/functional/batch_norm.hpp
@@ -2,14 +2,21 @@
 #define NMTOOLS_ARRAY_FUNCTIONAL_BATCH_NORM_HPP
 
 #include "nmtools/array/functional/functor.hpp"
+#include "nmtools/array/functional/moveaxis.hpp"
+#include "nmtools/array/functional/ufuncs/add.hpp"
+#include "nmtools/array/functional/ufuncs/multiply.hpp"
+#include "nmtools/array/functional/ufuncs/subtract.hpp"
+#include "nmtools/array/functional/ufuncs/divide.hpp"
+#include "nmtools/array/functional/ufuncs/sqrt.hpp"
 #include "nmtools/array/view/batch_norm.hpp"
 
 namespace nmtools::functional
 {
-    constexpr inline auto batch_norm = functor_t{quinary_fmap_t{
-        [](const auto&...args){
-            return view::batch_norm(args...);
-    }}};
+    constexpr inline auto batch_norm_fun = [](const auto&...args){
+        return view::batch_norm(args...);
+    };
+
+    constexpr inline auto batch_norm = functor_t{quinary_fmap_t<decltype(batch_norm_fun)>{batch_norm_fun}};
 } // namespace nmtools::functional
 
 #endif // NMTOOLS_ARRAY_FUNCTIONAL_BATCH_NORM_HPP