It appears that __Float128 is not supported on AArch64 systems (at le…

…ast, not on the OS/X Clang compilation used by GitHub actions; see the failure message for the previous commit). There is probably some way to get at 128-bit floats on AArch64, but for the nonce, we're just going with plain long doubles (presumably 80-bit) on all ARM platforms. The advantages of 128-bit floats for this purpose are unclear anyway.
Bill-Gray · Apr 28, 2024 · 3caa731 · 3caa731
1 parent 3bca05b
commit 3caa731
Showing 1 changed file with 1 addition and 3 deletions.
diff --git a/lsquare.cpp b/lsquare.cpp
@@ -44,9 +44,7 @@ idea for nearly-singular matrices.  */
       112 bits of precision.  Either is better than "ordinary" doubles
       with a mere 52 bits of precision.   */
 
-#if defined( __aarch64__)
-   #define ldouble _Float128
-#elif ((__GNUC__ * 100) + __GNUC_MINOR__) >= 406 && !defined( __arm__)
+#if ((__GNUC__ * 100) + __GNUC_MINOR__) >= 406 && !defined( __arm__) && !defined( __aarch64_)
    #define ldouble    __float128
 #else
    #define ldouble    long double