Improved gendata_h48k2 performance

Performance for the "real coordinate" case (h0) is ok, but still around 3x
slower than the gendata_h48k4 method. Performance for the intermediate
tables is not good, but for now I'll keep it like this.

This also shows that for h11 we should use the gendata_h48k4 method for h11.
For intermediate table, h8, h9 and h10 might be too slow to compute in an
acceptable time. We could restrict to h7 (largest with base = 9).

1 files changed, 2 insertions, 2 deletions

diff --git a/src/arch/arch.h b/src/arch/arch.h
index d5e0218..e5de0cf 100644
--- a/src/arch/arch.h
+++ b/src/arch/arch.h
@@ -1,4 +1,4 @@
-#if defined(CUBE_AVX2)
+#if defined(AVX2)
 
 #include <immintrin.h>
 
@@ -9,7 +9,7 @@ typedef __m256i cube_t;
 #include "avx2.h"
 #endif
 
-#elif defined(CUBE_NEON)
+#elif defined(NEON)
 
 #include <stdlib.h>
 #include <arm_neon.h>