Allow building on ARM64. Only Mac tested, and deps not tested.

.gitmodules

@@ -34,3 +34,6 @@
 [submodule "dep/rtmidi"]
 	path = dep/rtmidi
 	url = https://github.com/VCVRack/rtmidi.git
+[submodule "dep/sse2neon"]
+	path = dep/sse2neon
+	url = https://github.com/DLTcollab/sse2neon.git

arch.mk

@@ -3,6 +3,9 @@ MACHINE = $(shell $(CC) -dumpmachine)
 ifneq (,$(findstring x86_64-,$(MACHINE)))
 	ARCH_X64 := 1
 	ARCH_NAME := x64
+else ifneq (,$(findstring arm64-,$(MACHINE)))
+	ARCH_ARM64 := 1
+	ARCH_NAME := arm64
 else
 $(error Could not determine CPU architecture of $(MACHINE). Try hacking around in arch.mk)
 endif

compile.mk

@@ -14,7 +14,7 @@ FLAGS += -MMD -MP
 # Debugger symbols. These are removed with `strip`.
 FLAGS += -g
 # Optimization
-FLAGS += -O3 -march=nehalem -funsafe-math-optimizations -fno-omit-frame-pointer
+FLAGS += -O3 -funsafe-math-optimizations -fno-omit-frame-pointer
 # Warnings
 FLAGS += -Wall -Wextra -Wno-unused-parameter
 # C++ standard
@@ -23,6 +23,11 @@ CXXFLAGS += -std=c++11
 # Architecture-independent flags
 ifdef ARCH_X64
 	FLAGS += -DARCH_X64
+	FLAGS += -march=nehalem
+endif
+ifdef ARCH_ARM64
+	FLAGS += -DARCH_ARM64
+	FLAGS += -march=armv8-a+fp+simd
 endif
 
 ifdef ARCH_LIN

dep/Makefile

@@ -55,6 +55,7 @@ osdialog = include/osdialog.h
 pffft = include/pffft.h
 fuzzysearchdatabase = include/FuzzySearchDatabase.hpp
 ghcfilesystem = include/ghc/filesystem.hpp
+sse2neon = include/sse2neon/sse2neon.h
 
 DEPS += $(glew)
 DEPS += $(glfw)
@@ -72,6 +73,7 @@ DEPS += $(osdialog)
 DEPS += $(pffft)
 DEPS += $(fuzzysearchdatabase)
 DEPS += $(ghcfilesystem)
+DEPS += $(sse2neon)
 
 
 DEP_LOCAL := .
@@ -255,6 +257,10 @@ $(ghcfilesystem): filesystem/include/ghc
 	mkdir -p include
 	cp -r $^ include/
 
+$(sse2neon): sse2neon/sse2neon.h
+	mkdir -p include
+	cp $^ include/
+
 # Helpers
 
 src: glew-2.1.0 glfw jansson-2.12 speexdsp-SpeexDSP-1.2rc3 libsamplerate-0.1.9 openssl-1.1.1k curl-7.79.1 zstd-1.4.5 libarchive-3.4.3 rtaudio nanovg nanosvg oui-blendish osdialog

dep/sse2neon

@@ -0,0 +1 @@
+Subproject commit 988782cbadf95c2072b4b1b2b8fa0afa81b01c36

include/simd/Vector.hpp

@@ -1,6 +1,6 @@
 #pragma once
 #include <cstring>
-#include <pmmintrin.h>
+#include "common.hpp"
 
 
 namespace rack {

include/simd/common.hpp

@@ -0,0 +1,9 @@
+#pragma once
+
+#if defined ARCH_X64
+	// Intel intrinsics header
+	#include <x86intrin.h>
+#elif defined ARCH_ARM64
+	// Translation header for using SSE3 intrinsics on ARM64 NEON
+	#include <sse2neon.h>
+#endif

include/simd/sse_mathfun.h

@@ -43,7 +43,7 @@ This derived source file is released under the zlib license.
   (this is the zlib license)
 */
 #pragma once
-#include <pmmintrin.h>
+#include "common.hpp"
 
 
 /** Generate 1.f without accessing memory */

src/common.cpp

@@ -30,6 +30,9 @@ const std::string APP_VERSION = TOSTRING(_APP_VERSION);
 #if defined ARCH_X64
 	const std::string APP_ARCH = "x64";
 #endif
+#if defined ARCH_ARM64
+	const std::string APP_ARCH = "arm64";
+#endif
 const std::string API_URL = "https://api.vcvrack.com";
 
 

src/engine/Engine.cpp

@@ -5,7 +5,9 @@
 #include <mutex>
 #include <atomic>
 #include <tuple>
-#include <pmmintrin.h>
+#if defined ARCH_X64
+	#include <pmmintrin.h>
+#endif
 
 #include <engine/Engine.hpp>
 #include <settings.hpp>
@@ -21,6 +23,7 @@ namespace rack {
 namespace engine {
 
 
+#if defined ARCH_X64
 static void initMXCSR() {
 	// Set CPU to flush-to-zero (FTZ) and denormals-are-zero (DAZ) mode
 	// https://software.intel.com/en-us/node/682949
@@ -29,6 +32,7 @@ static void initMXCSR() {
 	// Reset other flags
 	_MM_SET_ROUNDING_MODE(_MM_ROUND_NEAREST);
 }
+#endif
 
 
 /** Barrier based on mutexes.
@@ -92,7 +96,9 @@ struct SpinBarrier {
 		while (true) {
 			if (step.load(std::memory_order_relaxed) != s)
 				return;
+#if defined ARCH_X64
 			__builtin_ia32_pause();
+#endif
 		}
 	}
 };
@@ -139,7 +145,9 @@ struct HybridBarrier {
 		while (!yielded.load(std::memory_order_relaxed)) {
 			if (step.load(std::memory_order_relaxed) != s)
 				return;
+#if defined ARCH_X64
 			__builtin_ia32_pause();
+#endif
 		}
 
 		// Wait on mutex CV
@@ -529,8 +537,10 @@ void Engine::stepBlock(int frames) {
 	std::lock_guard<std::mutex> stepLock(internal->blockMutex);
 	SharedLock<SharedMutex> lock(internal->mutex);
 	// Configure thread
+#if defined ARCH_X64
 	uint32_t csr = _mm_getcsr();
 	initMXCSR();
+#endif
 	random::init();
 
 	internal->blockFrame = internal->frame;
@@ -573,8 +583,10 @@ void Engine::stepBlock(int frames) {
 		internal->meterMax = 0.0;
 	}
 
+#if defined ARCH_X64
 	// Reset MXCSR back to original value
 	_mm_setcsr(csr);
+#endif
 }
 
 
@@ -1299,7 +1311,9 @@ void EngineWorker::run() {
 	// Configure thread
 	contextSet(engine->internal->context);
 	system::setThreadName(string::f("Worker %d", id));
+#if defined ARCH_X64
 	initMXCSR();
+#endif
 	random::init();
 
 	while (true) {