Changes to ARM build. It can now do 64-bit and 128-bit operations. Still need to do a Windows ARM build and validate generated assembly code.

Doxyfile

@@ -38,7 +38,7 @@ PROJECT_NAME           = X-Mem
 # could be handy for archiving the generated documentation or if some version
 # control system is used.
 
-PROJECT_NUMBER         = 2.1.15
+PROJECT_NUMBER         = 2.1.16
 
 # Using the PROJECT_BRIEF tag one can provide an optional one line description
 # for a project that appears at the top of each page and should give viewer a

README.md

@@ -1,7 +1,7 @@
 README
 ------------------------------------------------------------------------------------------------------------
 
-X-Mem: Extensible Memory Benchmarking Tool v2.1.15
+X-Mem: Extensible Memory Benchmarking Tool v2.1.16
 ------------------------------------------------------------------------------------------------------------
 
 The flexible open-source research tool for characterizing memory hierarchy throughput, latency, and power. 
@@ -10,7 +10,7 @@ Originally authored by Mark Gottscho (Email: <mgottscho@ucla.edu>) as a Summer 2
 
 This project is under active development. Stay tuned for more updates.
 
-PROJECT REVISION DATE: April 23, 2015.
+PROJECT REVISION DATE: April 24, 2015.
 
 ------------------------------------------------------------------------------------------------------------
 LICENSE

SConscript_linux_arm

@@ -9,7 +9,7 @@ env = Environment(CXX = "arm-linux-gnueabihf-g++-4.8")
 # Customize build settings
 
 # LINUX
-env.Append(CPPFLAGS = '-Wall -Wno-unused-but-set-variable -Wno-unused-variable -g -O3 -std=c++11 -fabi-version=6')
+env.Append(CPPFLAGS = '-Wall -Wno-unused-but-set-variable -Wno-unused-variable -g -O3 -std=c++11 -fabi-version=6 -mfloat-abi=hard -mfpu=vfpv3 -mfpu=neon')
 env.Append(CPPPATH = ['src/include'])
 env.Append(CPPPATH = ['src/include/ext/DelayInjectedLoadedLatencyBenchmark']) # Extension: Delay-injected loaded latency benchmark
 env.Append(CPPPATH = ['src/include/ext/StreamBenchmark']) # Extension: Stream benchmark

src/README.md

@@ -1,7 +1,7 @@
 README
 ------------------------------------------------------------------------------------------------------------
 
-X-Mem: Extensible Memory Benchmarking Tool v2.1.15
+X-Mem: Extensible Memory Benchmarking Tool v2.1.16
 ------------------------------------------------------------------------------------------------------------
 
 The flexible open-source research tool for characterizing memory hierarchy throughput, latency, and power. 
@@ -10,7 +10,7 @@ Originally authored by Mark Gottscho (Email: <mgottscho@ucla.edu>) as a Summer 2
 
 This project is under active development. Stay tuned for more updates.
 
-PROJECT REVISION DATE: April 23, 2015.
+PROJECT REVISION DATE: April 24, 2015.
 
 ------------------------------------------------------------------------------------------------------------
 LICENSE

src/benchmark_kernels.cpp

@@ -51,6 +51,10 @@
 #include <smmintrin.h>
 #endif
 
+#if defined(ARCH_ARM) && defined(ARCH_ARM_NEON)
+#include <arm_neon.h>
+#endif
+
 using namespace xmem;
 
 #if defined(__gnu_linux__) && defined(ARCH_INTEL_X86_64) && (defined(HAS_WORD_128) || defined(HAS_WORD_256))
@@ -65,7 +69,12 @@ using namespace xmem;
 #define my_64b_extractLSB_256b(w) _mm256_extract_epi64(w, 0) //AVX intrinsic, corresponds to ??? instruction. Header: immintrin.h
 #endif
 
-//TODO: ARM intrinsics?
+#if defined(ARCH_ARM) && defined(ARCH_ARM_NEON)
+#define my_64b_set_128b_word(a, b) vcombine_u64(a, b)
+
+#define my_32b_extractLSB_128b(w) vget_low_u32(w) //NEON intrinsic, corresponds to "vmov" instruction. Header: arm_neon.h
+#define my_64b_extractLSB_128b(w) vget_low_u64(w) //NEON intrinsic, corresponds to "vmov" instruction. Header: arm_neon.h
+#endif
 
 #if defined(_WIN32) && defined(ARCH_INTEL_X86_64)
 /* Hand-coded assembly functions for the 128-bit and 256-bit benchmark kernels on Windows x86-64 where applicable.

src/include/common.h

@@ -43,9 +43,13 @@
 #include <immintrin.h> //for Intel __m256i datatype
 #endif
 
+#if defined(__arm__) || defined(_M_ARM)
+#include <arm_neon.h> //For ARM uint64x2_t datatype
+#endif
+
 namespace xmem {
 
-#define VERSION "2.1.15"
+#define VERSION "2.1.16"
 
 #if !defined(_WIN32) && !defined(__gnu_linux__)
 #error Neither Windows/GNULinux build environments were detected!
@@ -331,10 +335,8 @@ namespace xmem {
 	extern double g_ns_per_tick;
 
 	//Typedef the platform specific stuff to word sizes to match 4 different chunk options
-	//TODO: better way to detect 64-bit ARM other than NEON extensions?
-#if defined(ARCH_64BIT) || defined(ARCH_ARM_NEON)
-#define HAS_WORD_64
-#endif
+#define HAS_WORD_64 //For now, assume this is always available, even on native 32-bit architectures. The compiler will emulate 64-bit operations.
+
 #if defined(ARCH_INTEL_AVX) || defined(ARCH_ARM_NEON)
 #define HAS_WORD_128
 #endif
@@ -350,8 +352,8 @@ namespace xmem {
 #ifdef ARCH_INTEL
 	typedef __m128i Word128_t;
 #endif
-#ifdef ARCH_ARM
-	#error TODO: Implement for ARM
+#if defined(ARCH_ARM) && defined(ARCH_ARM_NEON)
+	typedef uint64x2_t Word128_t;
 #endif
 #endif
 #ifdef HAS_WORD_256