Extending packaged dot to support matmul ufunc

.github/workflows/ci.yml

@@ -59,17 +59,37 @@ jobs:
         run: |
           sudo apt-get update
           sudo apt-get install -y libmpfr-dev libssl-dev libfftw3-dev
+
       - name: Install SLEEF
         run: |
+          sudo apt-get update -y
+          sudo apt-get install -y cmake gcc g++ make git pkg-config
           git clone --branch 3.8 https://github.com/shibatch/sleef.git
           cd sleef
           cmake -S . -B build -DSLEEF_BUILD_QUAD:BOOL=ON -DSLEEF_BUILD_SHARED_LIBS:BOOL=ON -DCMAKE_POSITION_INDEPENDENT_CODE=ON
           cmake --build build/ --clean-first -j
-          sudo cmake --install build --prefix /usr
+          sudo cmake --install build --prefix /usr/local
+
       - name: Install quaddtype
         working-directory: quaddtype
         run: |
-          LDFLAGS="-Wl,-rpath,/usr/lib" python -m pip install . -v --no-build-isolation -Cbuilddir=build -C'compile-args=-v' -Csetup-args="-Dbuildtype=debug"
+          # Initialize submodules first
+          git submodule update --init --recursive
+          ls -la numpy_quaddtype/QBLAS/
+
+          # Set environment variables with proper export and correct paths
+          export CFLAGS="-I/usr/local/include -I$(pwd)/numpy_quaddtype/QBLAS/include"
+          export CXXFLAGS="-I/usr/local/include -I$(pwd)/numpy_quaddtype/QBLAS/include -fext-numeric-literals"
+          export LDFLAGS="-L/usr/local/lib64 -L/usr/local/lib -Wl,-rpath,/usr/local/lib64 -Wl,-rpath,/usr/local/lib -fopenmp"
+          export LD_LIBRARY_PATH="/usr/local/lib64:/usr/local/lib:$LD_LIBRARY_PATH"
+
+          # Install with meson args to ensure the C++ flags are passed through
+          python -m pip install . -v --no-build-isolation \
+            -Cbuilddir=build \
+            -C'compile-args=-v' \
+            -Csetup-args="-Dbuildtype=debug" \
+            -Csetup-args="-Dcpp_args=-fext-numeric-literals"
+
       - name: Run quaddtype tests
         working-directory: quaddtype
         run: |

quaddtype/README.md

@@ -27,7 +27,7 @@ np.array([1,2,3], dtype=QuadPrecDType("longdouble"))
 
 The code needs the quad precision pieces of the sleef library, which
 is not available on most systems by default, so we have to generate
-that first.  The below assumes one has the required pieces to build
+that first. The below assumes one has the required pieces to build
 sleef (cmake and libmpfr-dev), and that one is in the package
 directory locally.
 
@@ -40,6 +40,7 @@ cd ..
 ```
 
 Building the `numpy-quaddtype` package from locally installed sleef:
+
 ```bash
 export SLEEF_DIR=$PWD/sleef/build
 export LIBRARY_PATH=$SLEEF_DIR/lib
@@ -57,10 +58,13 @@ export LDFLAGS="-Wl,-rpath,$SLEEF_DIR/lib -fopenmp -latomic -lpthread"
 export CFLAGS="-fPIC"
 export CXXFLAGS="-fPIC"
 
+# To build without QBLAS (default for MSVC)
+# export CFLAGS="-fPIC -DDISABLE_QUADBLAS"
+# export CXXFLAGS="-fPIC -DDISABLE_QUADBLAS"
+
 python -m pip install . -v --no-build-isolation -Cbuilddir=build -C'compile-args=-v'
 
 # Run the tests
 cd ..
 python -m pytest
 ```
-

quaddtype/meson.build

@@ -50,12 +50,21 @@ srcs = [
     'numpy_quaddtype/src/scalar_ops.h',
     'numpy_quaddtype/src/scalar_ops.cpp',
     'numpy_quaddtype/src/ops.hpp',
-    'numpy_quaddtype/src/umath.h',
-    'numpy_quaddtype/src/umath.cpp',
     'numpy_quaddtype/src/dragon4.h',
     'numpy_quaddtype/src/dragon4.c',
     'numpy_quaddtype/src/quadblas_interface.h',
-    'numpy_quaddtype/src/quadblas_interface.cpp'
+    'numpy_quaddtype/src/quadblas_interface.cpp',
+    'numpy_quaddtype/src/umath/umath.h',
+    'numpy_quaddtype/src/umath/umath.cpp',
+    'numpy_quaddtype/src/umath/binary_ops.h',
+    'numpy_quaddtype/src/umath/binary_ops.cpp',
+    'numpy_quaddtype/src/umath/unary_ops.h',
+    'numpy_quaddtype/src/umath/unary_ops.cpp',
+    'numpy_quaddtype/src/umath/comparison_ops.h',
+    'numpy_quaddtype/src/umath/comparison_ops.cpp',
+    'numpy_quaddtype/src/umath/promoters.hpp',
+    'numpy_quaddtype/src/umath/matmul.h',
+    'numpy_quaddtype/src/umath/matmul.cpp',
 ]
 
 py.install_sources(

quaddtype/numpy_quaddtype/__init__.py

@@ -3,7 +3,6 @@
     QuadPrecDType,
     is_longdouble_128,
     get_sleef_constant,
-    qblas_dot as dot,
     set_num_threads,
     get_num_threads,
     get_quadblas_version
@@ -17,7 +16,7 @@
     # Constants
     'pi', 'e', 'log2e', 'log10e', 'ln2', 'ln10', 'max_value', 'min_value', 'epsilon', 
     # QuadBLAS related functions
-    'dot', 'set_num_threads', 'get_num_threads', 'get_quadblas_version'
+    'set_num_threads', 'get_num_threads', 'get_quadblas_version'
 ]
 
 def SleefQuadPrecision(value):
@@ -40,8 +39,4 @@ def LongDoubleQuadPrecDType():
 ln10 = get_sleef_constant("ln10")
 max_value = get_sleef_constant("quad_max")
 min_value = get_sleef_constant("quad_min")
-epsilon = get_sleef_constant("epsilon")
-
-num_cores = multiprocessing.cpu_count()
-# set default number of threads for QuadBLAS
-set_num_threads(num_cores)
+epsilon = get_sleef_constant("epsilon")