diff --git a/src/dsp/msa_macro.h b/src/dsp/msa_macro.h
index de026a1d..51f6c643 100644
--- a/src/dsp/msa_macro.h
+++ b/src/dsp/msa_macro.h
@@ -14,6 +14,10 @@
 #ifndef WEBP_DSP_MSA_MACRO_H_
 #define WEBP_DSP_MSA_MACRO_H_
 
+#include "src/dsp/dsp.h"
+
+#if defined(WEBP_USE_MSA)
+
 #include <stdint.h>
 #include <msa.h>
 
@@ -1389,4 +1393,5 @@ static WEBP_INLINE uint32_t func_hadd_uh_u32(v8u16 in) {
 } while (0)
 #define AVER_UB2_UB(...) AVER_UB2(v16u8, __VA_ARGS__)
 
+#endif  // WEBP_USE_MSA
 #endif  // WEBP_DSP_MSA_MACRO_H_
diff --git a/src/dsp/neon.h b/src/dsp/neon.h
index aa1dea13..c591f9b9 100644
--- a/src/dsp/neon.h
+++ b/src/dsp/neon.h
@@ -12,10 +12,12 @@
 #ifndef WEBP_DSP_NEON_H_
 #define WEBP_DSP_NEON_H_
 
-#include <arm_neon.h>
-
 #include "src/dsp/dsp.h"
 
+#if defined(WEBP_USE_NEON)
+
+#include <arm_neon.h>
+
 // Right now, some intrinsics functions seem slower, so we disable them
 // everywhere except newer clang/gcc or aarch64 where the inline assembly is
 // incompatible.
@@ -98,4 +100,5 @@ static WEBP_INLINE int32x4x4_t Transpose4x4_NEON(const int32x4x4_t rows) {
 } while (0)
 #endif
 
+#endif  // WEBP_USE_NEON
 #endif  // WEBP_DSP_NEON_H_