Setup optimization blaze defines globally

backend/meson.build

@@ -12,6 +12,40 @@ cc = meson.get_compiler('cpp')
 py = import('python').find_installation(pure: false)
 py_dep = py.dependency()
 
+# set up global defines to optimize blaze usages
+add_global_arguments(
+    # - Enable external BLAS kernels
+    '-DBLAZE_BLAS_MODE=0',
+    # - Set default matrix storage order to column-major, since many of our
+    #   functions are implemented for column-major layout. This default reduces
+    #   conversions.
+    '-DBLAZE_DEFAULT_STORAGE_ORDER=blaze::rowMajor',
+    # - Disable SMP parallelization. This disables SMP parallelization for all
+    #   possible backends (OpenMP, C++11 threads, Boost, HPX):
+    #   https://bitbucket.org/blaze-lib/blaze/wiki/Serial%20Execution#!option-3-deactivation-of-parallel-execution
+    '-DBLAZE_USE_SHARED_MEMORY_PARALLELIZATION=0',
+    # - Disable MPI parallelization
+    '-DBLAZE_MPI_PARALLEL_MODE=0',
+    # - Using the default cache size, which may have been configured automatically
+    #   by the Blaze CMake configuration for the machine we are running on. We
+    #   could override it here explicitly to tune performance.
+    # BLAZE_CACHE_SIZE
+    '-DBLAZE_USE_PADDING=1',
+    # - Always enable non-temporal stores for cache optimization of large data
+    #   structures: https://bitbucket.org/blaze-lib/blaze/wiki/Configuration%20Files#!streaming-non-temporal-stores
+    '-DBLAZE_USE_STREAMING=1',
+    # - Initializing default-constructed structures for fundamental types
+    '-DBLAZE_USE_DEFAULT_INITIALIZATON=1',
+    # Use Sleef for vectorization of more math functions
+    '-DBLAZE_USE_SLEEF=1',
+    # Set inlining settings
+    '-DBLAZE_USE_STRONG_INLINE=1',
+    '-DBLAZE_USE_ALWAYS_INLINE=1',
+    # Set vectorization (leave to 1, else there is no use for blaze)
+    '-DBLAZE_USE_VECTORIZATION=1',
+    language: 'cpp'
+)
+
 # find dependencies and create dep objects
 pybind11_dep = declare_dependency(
     include_directories: run_command(
@@ -53,6 +87,9 @@ if not sleef_dep.found()
     )
 endif
 
+# blaze and deps commonly used together
+blaze_deps = [blaze_dep, blaze_tensor_dep, sleef_dep]
+
 brigand_dep = dependency('brigand', fallback : 'brigand')
 cxxopts_dep = dependency('cxxopts', fallback : 'cxxopts')
 

backend/src/meson.build

@@ -10,7 +10,7 @@ example_1 = py.extension_module(
 _linalg = py.extension_module(
     '_linalg',
     sources: ['_linalg.cpp'],
-    dependencies: [py_dep, pybind11_dep, blaze_dep, blaze_tensor_dep],
+    dependencies: [py_dep, pybind11_dep] + blaze_deps,
     link_language: 'cpp',
     install: true,
     subdir: package,
@@ -19,7 +19,7 @@ _linalg = py.extension_module(
 _linalg_numpy = py.extension_module(
     '_linalg_numpy',
     sources: ['_linalg_numpy.cpp'],
-    dependencies: [py_dep, pybind11_dep, blaze_dep, blaze_tensor_dep],
+    dependencies: [py_dep, pybind11_dep] + blaze_deps,
     link_language: 'cpp',
     install: true,
     subdir: package,
@@ -33,7 +33,7 @@ _PyArrays = py.extension_module(
         'Utilities/Math/Python/BlazeMatrix.cpp',
         'Utilities/Math/Python/BlazeTensor.cpp',
     ],
-    dependencies: [py_dep, pybind11_dep, blaze_dep, blaze_tensor_dep],
+    dependencies: [py_dep, pybind11_dep] + blaze_deps,
     link_language: 'cpp',
     install: true,
     subdir: package,