Fix ARM compile

include/simdjson/arm64/bitmask.h

@@ -33,9 +33,8 @@ simdjson_really_inline uint64_t prefix_xor(uint64_t bitmask) {
   return bitmask;
 }
 
+} // unnamed namespace
 } // namespace arm64
 } // namespace simdjson
-} // namespace simdjson
-SIMDJSON_UNTARGET_IMPLEMENTATION
 
 #endif

include/simdjson/arm64/simd.h

@@ -256,6 +256,9 @@ simdjson_really_inline int8x16_t make_int8x16_t(int8_t x1,  int8_t x2,  int8_t x
     // sensible, but the AVX ISA makes this kind of approach difficult.
     template<typename L>
     simdjson_really_inline void compress(uint16_t mask, L * output) const {
+      using internal::thintable_epi8;
+      using internal::BitsSetTable256mul2;
+      using internal::pshufb_combine_table;
       // this particular implementation was inspired by work done by @animetosho
       // we do it in two steps, first 8 bytes and then second 8 bytes
       uint8_t mask1 = uint8_t(mask); // least significant 8 bits

include/simdjson/haswell/intrinsics.h

@@ -41,11 +41,11 @@
 // has it as a macro.
 #ifndef _blsr_u64
 // we roll our own
-SIMDJSON_TARGET_IMPLEMENTATION
+SIMDJSON_TARGET_HASWELL
 static simdjson_really_inline uint64_t _blsr_u64(uint64_t n) {
   return (n - 1) & n;
 }
-SIMDJSON_UNTARGET_IMPLEMENTATION
+SIMDJSON_UNTARGET_REGION
 #endif //  _blsr_u64
 #endif // SIMDJSON_CLANG_VISUAL_STUDIO