simpler ARM detection

CMakeLists.txt

@@ -1,4 +1,4 @@
-cmake_minimum_required(VERSION 3.3)
+cmake_minimum_required(VERSION 3.5)
 project(STREAMVBYTE VERSION "2.0.0")
 
 set(STREAMVBYTE_LIB_VERSION "2.0.0" CACHE STRING "streamvbyte library version")
@@ -97,11 +97,6 @@ if(MSVC)
     add_definitions("-D__restrict__=__restrict")
 endif()
 
-# test for arm
-if(CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64.*|AARCH64.*)")
-    set(BASE_FLAGS ${BASE_FLAGS} "-D__ARM_NEON__")
-endif()
-
 set(STREAMVBYTE_SRCS
     ${PROJECT_SOURCE_DIR}/src/streamvbyte_encode.c
     ${PROJECT_SOURCE_DIR}/src/streamvbyte_decode.c

Makefile

@@ -5,12 +5,8 @@
 
 PROCESSOR:=$(shell uname -m)
 
-ifeq ($(PROCESSOR), aarch64)
-# for 64-bit ARM processors (e.g., Linux), they may fail to defined __ARM_NEON__
-CFLAGS = -fPIC -std=c99 -O3 -Wall -Wextra -pedantic -Wshadow -D__ARM_NEON__
-else
+
 CFLAGS = -fPIC -std=c99 -O3 -Wall -Wextra -pedantic -Wshadow
-endif
 LDFLAGS = -shared
 LIBNAME=libstreamvbyte.so.0.0.1
 LNLIBNAME=libstreamvbyte.so

src/streamvbyte_decode.c

@@ -7,7 +7,7 @@
 #pragma clang diagnostic ignored "-Wdeclaration-after-statement"
 #endif
 
-#ifdef __ARM_NEON__
+#ifdef STREAMVBYTE_IS_ARM64
 #include "streamvbyte_arm_decode.c"
 #endif
 
@@ -76,7 +76,7 @@ size_t streamvbyte_decode(const uint8_t *in, uint32_t *out, uint32_t count) {
     keyPtr += (count/4) & ~ 7U;
     count &= 31;
   }
-#elif defined(__ARM_NEON__)
+#elif defined(STREAMVBYTE_IS_ARM64)
   dataPtr = svb_decode_vector(out, keyPtr, dataPtr, count);
   out += count - (count & 3);
   keyPtr += count/4;
@@ -105,7 +105,7 @@ bool streamvbyte_validate_stream(const uint8_t *in, size_t inCount,
   const uint8_t *keyPtr = in;
   uint64_t encodedSize = 0;
 
-#if defined(__ARM_NEON__)
+#if defined(STREAMVBYTE_IS_ARM64)
   encodedSize = svb_validate_vector(&keyPtr, &outCount);
 #endif
 

src/streamvbyte_encode.c

@@ -62,7 +62,7 @@ static uint8_t *svb_encode_scalar(const uint32_t *in,
 }
 
 
-#ifdef __ARM_NEON__
+#ifdef STREAMVBYTE_IS_ARM64
 #include "streamvbyte_arm_encode.c"
 #endif
 
@@ -120,7 +120,7 @@ size_t streamvbyte_encode(const uint32_t *in, uint32_t count, uint8_t *out) {
   uint32_t keyLen = (count + 3) / 4;  // 2-bits rounded to full byte
   uint8_t *dataPtr = keyPtr + keyLen; // variable byte data after all keys
 
-#if defined(__ARM_NEON__)
+#if defined(STREAMVBYTE_IS_ARM64)
 
   uint32_t count_quads = count / 4;
   count -= 4 * count_quads;

src/streamvbyte_isadetection.h

@@ -43,7 +43,9 @@ POSSIBILITY OF SUCH DAMAGE.
 #include <stdbool.h>
 #include <stdlib.h>
 
-
+#if defined(__aarch64__) || defined(_M_ARM64) || defined(_M_ARM64EC)
+#define STREAMVBYTE_IS_ARM64 1
+#endif // defined(__aarch64__) || defined(_M_ARM64) || defined(_M_ARM64EC)
 #if defined(_MSC_VER)
 /* Microsoft C/C++-compatible compiler */
 #include <intrin.h>
@@ -52,7 +54,7 @@ POSSIBILITY OF SUCH DAMAGE.
 #elif defined(__GNUC__) && (defined(__x86_64__) || defined(__i386__))
 /* GCC-compatible compiler, targeting x86/x86-64 */
 #include <x86intrin.h>
-#elif defined(__GNUC__) && defined(__ARM_NEON__)
+#elif defined(__GNUC__) && defined(STREAMVBYTE_IS_ARM64)
 /* GCC-compatible compiler, targeting ARM with NEON */
 #include <arm_neon.h>
 #elif defined(__GNUC__) && defined(__IWMMXT__)
@@ -263,9 +265,9 @@ static inline uint32_t streamvbyte_detect_supported_architectures(void) {
 }
 #endif
 
-#ifdef __ARM_NEON__
+#ifdef STREAMVBYTE_IS_ARM64
 #define STREAMVBYTE_ARM
-#endif 
+#endif // STREAMVBYTE_IS_ARM64
 
 #ifdef STREAMVBYTE_X64
 // this is almost standard?

tests/unit.c

@@ -2827,7 +2827,7 @@ int main(void) {
   } else {
     printf("Code was not vectorized (x64).\n");
   }
-#elif defined(__ARM_NEON__)
+#elif defined(STREAMVBYTE_IS_ARM64)
   printf("Code was vectorized (ARM NEON).\n");
 #else
   printf("Warning: you tested non-vectorized code.\n");