Help xcode produce fmla instructions

ggerganov · Apr 21, 2024 · d3d40bf · d3d40bf
1 parent 9d4d14c
commit d3d40bf
Showing 1 changed file with 14 additions and 0 deletions.
diff --git a/sgemm.cpp b/sgemm.cpp
@@ -213,6 +213,20 @@ inline U madd(T a, T b, U c) {
     return add(mul(a, b), c);
 }
 
+// xcode needs a little help to produce fmla instructions
+#if defined(__ARM_FEATURE_FMA)
+template <>
+inline float32x4_t madd(float32x4_t a, float32x4_t b, float32x4_t c) {
+    return vfmaq_f32(c, b, a);
+}
+#if defined(__ARM_FEATURE_FP16_VECTOR_ARITHMETIC) && !defined(_MSC_VER)
+template <>
+inline float16x8_t madd(float16x8_t a, float16x8_t b, float16x8_t c) {
+    return vfmaq_f16(c, b, a);
+}
+#endif
+#endif
+
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 // FLOATING POINT MATRIX MULTIPLICATION