Highway POC

numpy · Mousius · Jul 7, 2023 · Jul 8, 2023 · Aug 10, 2023 · Aug 11, 2023
commit 085dff14178c2601ee07db7eafb7da204586c12b
diff --git a/.gitmodules b/.gitmodules
@@ -13,3 +13,6 @@
 [submodule "vendored-meson/meson-python"]
 	path = vendored-meson/meson-python
 	url = https://github.com/numpy/meson-python.git
+[submodule "numpy/core/src/highway"]
+	path = numpy/core/src/highway
+	url = https://github.com/google/highway.git
diff --git a/numpy/core/meson.build b/numpy/core/meson.build
@@ -79,6 +79,10 @@ endif
 # important, because code in NumPy typically does not check the value but only
 # whether the symbol is defined. So `#define HAVE_SOMETHING 0` is wrong.
 
+#cmake = import('cmake')
+#hwy = cmake.subproject('highway')
+
+
 cdata = configuration_data()
 
 cdata.set('NPY_ABI_VERSION', C_ABI_VERSION)
@@ -1077,6 +1081,7 @@ src_umath = umath_gen_headers + [
   src_file.process('src/umath/matmul.c.src'),
   src_file.process('src/umath/matmul.h.src'),
   'src/umath/ufunc_type_resolution.c',
+  'src/umath/absolute.cpp',
   'src/umath/clip.cpp',
   'src/umath/clip.h',
   'src/umath/dispatching.c',
@@ -1141,6 +1146,7 @@ py.extension_module('_multiarray_umath',
     'src/multiarray',
     'src/npymath',
     'src/umath',
+    'src/highway',
   ],
   dependencies: blas_dep,
   link_with: [npymath_lib, multiarray_umath_mtargets.static_lib('_multiarray_umath_mtargets')],

diff --git a/numpy/core/src/highway b/numpy/core/src/highway
diff --git a/numpy/core/src/umath/absolute.cpp b/numpy/core/src/umath/absolute.cpp
@@ -0,0 +1,66 @@
+#define _UMATHMODULE
+#define _MULTIARRAYMODULE
+#define NPY_NO_DEPRECATED_API NPY_API_VERSION
+
+#define PY_SSIZE_T_CLEAN
+#include <Python.h>
+
+#undef HWY_TARGET_INCLUDE
+#define HWY_TARGET_INCLUDE "absolute.cpp"  // this file
+#include <hwy/foreach_target.h>  // must come before highway.h
+#include <hwy/highway.h>
+
8000#include "numpy/ndarraytypes.h"
+#include "numpy/npy_common.h"
+#include "numpy/npy_math.h"
+#include "numpy/utils.h"
+
+namespace numpy {
+namespace HWY_NAMESPACE {  // required: unique per target
+
+// Can skip hn:: prefixes if already inside hwy::HWY_NAMESPACE.
+namespace hn = hwy::HWY_NAMESPACE;
+using T = npy_int;
+
+// Alternative to per-function HWY_ATTR: see HWY_BEFORE_NAMESPACE
+HWY_ATTR void SuperAbsolute(const T* HWY_RESTRICT input_array,
+                T* HWY_RESTRICT output_array,
+                const size_t size) {
+  const hn::ScalableTag<T> d;
+  for (size_t i = 0; i < size; i += hn::Lanes(d)) {
+    const auto in = hn::Load(d, input_array + i);
+    auto x = hn::Abs(in);
+    hn::Store(x, d, output_array + i);
+  }
+}
+
+}
+}
+
+#if HWY_ONCE
+namespace numpy {
+
+HWY_EXPORT(SuperAbsolute);
+
+extern "C" {
+NPY_NO_EXPORT void
+INT_absolute(char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
+{
+    npy_int *ip1 = (npy_int*) args[0];
+    npy_int *op1 = (npy_int*) args[1];
+    // npy_intp is1 = steps[0];
+    // npy_intp os1 = steps[1];
+    npy_intp n = dimensions[0];
+    // npy_intp i;
+  // This must reside outside of HWY_NAMESPACE because it references (calls the
+  // appropriate one from) the per-target implementations there.
+  // For static dispatch, use HWY_STATIC_DISPATCH.
+  static auto dispatcher = HWY_DYNAMIC_DISPATCH(SuperAbsolute);
+  return dispatcher(ip1, op1, n);
+}
+}
+
+}
+#endif
+
+
diff --git a/numpy/core/src/umath/loops_autovec.dispatch.c.src b/numpy/core/src/umath/loops_autovec.dispatch.c.src
@@ -213,13 +213,16 @@ NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_invert)
  * #TYPE = BYTE, SHORT, INT, LONG, LONGLONG#
  * #type = npy_byte, npy_short, npy_int, npy_long, npy_longlong#
  * #c    = ,,,l,ll#
+ * #abs  = 1,1,0,1,1#
  */
 
+#if @abs@
 NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_absolute)
 (char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))
 {
     UNARY_LOOP_FAST(@type@, @type@, *out = (in >= 0) ? in : -in);
 }
+#endif
 
 NPY_NO_EXPORT void NPY_CPU_DISPATCH_CURFX(@TYPE@_sign)
 (char **args, npy_intp const *dimensions, npy_intp const *steps, void *NPY_UNUSED(func))