feat: Allow ARM architecture to use SIMD trit validation

A new project sse2neon is added as git submodule to allow dcurl running
on ARM architecture to use SIMD acceleration without writing NEON
intrinsic functions.

Close #39.

.gitmodules

@@ -4,3 +4,6 @@
 [submodule "deps/rabbitmq-c"]
 	path = deps/rabbitmq-c
 	url = https://github.com/alanxz/rabbitmq-c.git
+[submodule "deps/sse2neon"]
+	path = deps/sse2neon
+	url = https://github.com/DLTcollab/sse2neon.git

mk/submodule.mk

@@ -54,11 +54,21 @@ endif
 	cd $(LIBRABBITMQ_PATH)/build && \
          cmake --build .
 
+# sse2neon related variables
+SSE2NEON_PATH = deps/sse2neon
+SSE2NEON_INCLUDE := -I $(SSE2NEON_PATH)
+
+$(SSE2NEON_PATH)/sse2neon.h:
+	git submodule update --init $(SSE2NEON_PATH)
+
 # Submodules
 SUBS := $(LIBTUV_PATH)/include
 ifeq ($(BUILD_REMOTE),1)
     SUBS += $(LIBRABBITMQ_PATH)/librabbitmq
 endif
+ifeq ($(UNAME_M),$(filter $(UNAME_M),arm aarch64))
+    SUBS += $(SSE2NEON_PATH)/sse2neon.h
+endif
 # Submodule related objects
 SUB_OBJS := $(LIBTUV_OBJS)
 ifeq ($(BUILD_REMOTE),1)
@@ -69,3 +79,6 @@ SUB_INCLUDE := $(LIBTUV_INCLUDE)
 ifeq ($(BUILD_REMOTE),1)
     SUB_INCLUDE += $(LIBRABBITMQ_INCLUDE)
 endif
+ifeq ($(UNAME_M),$(filter $(UNAME_M),arm aarch64))
+    SUB_INCLUDE += $(SSE2NEON_INCLUDE)
+endif

src/trinary.c

@@ -7,7 +7,7 @@
  */
 
 #include "trinary.h"
-#if defined(__SSE4_2__)
+#if defined(__SSE4_2__) || defined(__ARM_NEON)
 #include "trinary_sse42.h"
 #endif
 #include <stdint.h>
@@ -35,7 +35,7 @@ static bool validateTrits(Trobject_t *trits)
     if (trits->type != TYPE_TRITS)
         return false;
 
-#if defined(__SSE4_2__)
+#if defined(__SSE4_2__) || defined(__ARM_NEON)
     return validateTrits_sse42(trits);
 #endif
     for (int i = 0; i < trits->len; i++)

src/trinary_sse42.h

@@ -9,8 +9,16 @@
 #ifndef TRINARY_SSE42_H_
 #define TRINARY_SSE42_H_
 
+#if defined(__SSE4_2__)
 #include <nmmintrin.h>
+#endif
 #include "constants.h"
+#if defined(__ARM_NEON)
+#include "sse2neon.h"
+#endif
+#if !defined(__SSE4_2__) && !defined(__ARM_NEON)
+#error "The hardware architecture should support SSE4.2 or NEON instruction."
+#endif
 
 #define BLOCK_8BIT(type) (sizeof(type) / sizeof(int8_t))
 #define BYTE_OF_128BIT 16
@@ -66,6 +74,7 @@ static inline bool validateTrits_sse42(Trobject_t *trits)
     return true;
 }
 
+#if defined(__SSE4_2__)
 static inline bool validateTrytes_sse42(Trobject_t *trytes)
 {
     const int block_8bit = BLOCK_8BIT(__m128i);
@@ -452,5 +461,6 @@ static inline Trobject_t *trits_from_trytes_sse42(Trobject_t *trytes)
 
     return trits;
 }
+#endif  // #if defined(__SSE4_2__)
 
 #endif