Fix meson based compilation on arm32

Author: ankith26

meson.build

@@ -293,8 +293,8 @@ if host_machine.cpu_family() == 'arm'
     # first check if compiler supports the flag, and use it then. Needed only
     # on 32-bit armv7.
     flag = '-mfpu=neon'
-    if cc.has_argument(flag)
-        simd_sse2_neon_flags += flag
+    if cc.has_argument(flag) and host_machine.cpu() == 'armv7l'
+        simd_sse2_neon_flags += [flag, '-march=armv7-a']
         simd_sse2_neon = true
     endif
 elif host_machine.cpu_family() == 'aarch64'

src_c/simd_fill.h

@@ -21,11 +21,6 @@
 #endif
 #endif
 
-#if PG_ENABLE_ARM_NEON
-// sse2neon.h is from here: https://github.com/DLTcollab/sse2neon
-#include "include/sse2neon.h"
-#endif /* PG_ENABLE_ARM_NEON */
-
 #if defined(__SSE2__)
 #define PG_ENABLE_SSE_NEON 1
 #elif PG_ENABLE_ARM_NEON

src_c/simd_shared.h

@@ -33,11 +33,6 @@ pg_has_avx2();
 #endif
 #endif
 
-#if PG_ENABLE_ARM_NEON
-// sse2neon.h is from here: https://github.com/DLTcollab/sse2neon
-#include "include/sse2neon.h"
-#endif /* PG_ENABLE_ARM_NEON */
-
 /* This defines PG_ENABLE_SSE_NEON as True if either SSE or NEON is available
  * at compile time. Since we do compile time translation of SSE2->NEON, they
  * have the same code paths, so this reduces code duplication of those paths.

src_c/simd_surface_fill_sse2.c

@@ -1,5 +1,10 @@
 #include "simd_fill.h"
 
+#if PG_ENABLE_ARM_NEON
+// sse2neon.h is from here: https://github.com/DLTcollab/sse2neon
+#include "include/sse2neon.h"
+#endif /* PG_ENABLE_ARM_NEON */
+
 #define BAD_SSE2_FUNCTION_CALL                                               \
     printf(                                                                  \
         "Fatal Error: Attempted calling an SSE2 function when both compile " \