Merge pull request #3341 from juliantaylor/sse2-sqrt

vectorize sqrt ufunc with SSE2

1 files changed, 8 insertions, 2 deletions

diff --git a/numpy/core/src/scalarmathmodule.c.src b/numpy/core/src/scalarmathmodule.c.src
index 11716dd46..0f5f195f4 100644
--- a/numpy/core/src/scalarmathmodule.c.src
+++ b/numpy/core/src/scalarmathmodule.c.src
@@ -1766,8 +1766,14 @@ get_functions(void)
     }
     funcdata = ((PyUFuncObject *)obj)->data;
     signatures = ((PyUFuncObject *)obj)->types;
-    i = 0;
-    j = 0;
+    /*
+     * sqrt ufunc is specialized for double and float loops in
+     * generate_umath.py, the first to go into FLOAT/DOUBLE_sqrt
+     * they have the same signature as the scalar variants so we need to skip
+     * over them
+     */
+    i = 4;
+    j = 2;
     while (signatures[i] != NPY_FLOAT) {
         i += 2; j++;
     }