Adjusted macro to properly autovectorize on GCC (tested with 9.4.0)

pzelasko · Jan 30, 2023 · 5c279f4 · 5c279f4
1 parent a8e2029
commit 5c279f4
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 1 deletion.
diff --git a/extensions/reverb_rir.cpp b/extensions/reverb_rir.cpp
@@ -50,7 +50,7 @@ float inner_product_clang(const vector<float> &buffer_, const vector<float> &rir
 float inner_product_gcc(const vector<float> &buffer_, const vector<float> &rir_, int offset) {
     const auto SIZE = rir_.size();
     float sum = 0.0f;
-    #pragma GCC ivdep
+    #pragma omp simd reduction(+:sum)
     for (unsigned int i = 0; i < SIZE; ++i) {
         sum += buffer_[offset + i] * rir_[i];
     }

diff --git a/setup.py b/setup.py
@@ -39,6 +39,9 @@ def no_cythonize(extensions, **_ignore):
         "-Wno-unused-function",
         "-Wno-unused-local-typedefs",
     ])
+    if not platform.startswith("darwin"):
+        COMPILE_ARGS.append("-fopenmp")
+        COMPILE_ARGS.append("-march=native")
 
 if platform.startswith("darwin"):
     COMPILE_ARGS.append("-stdlib=libc++")