Ported Arches to macOS

.gitignore

@@ -422,4 +422,12 @@ out/
 
 datasets/
 
-ext/
+ext/
+
+# macOS Ignores
+.DS_Store
+
+# CLion Ignores
+.idea
+cmake-build*
+

CMakeLists.txt

@@ -7,5 +7,9 @@ set_property(GLOBAL PROPERTY USE_FOLDERS ON)
 set(CMAKE_CXX_STANDARD 20)
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 
+add_compile_definitions(FMT_CONSTEVAL=constexpr)
+
+set(CMAKE_POLICY_VERSION_MINIMUM 3.10)
+
 add_subdirectory(external)
 add_subdirectory("src")

include/rtm/bvh.hpp

@@ -17,6 +17,19 @@
 
 namespace rtm {
 
+#if defined __aarch64__
+inline uint64_t _pdep_u64_sw(uint64_t data, uint64_t mask)
+{
+	uint64_t result = 0;
+	for (uint64_t bit = 1; mask; bit <<= 1) {
+		if (data & bit)
+			result |= mask & (-mask);
+		mask &= mask - 1;
+	}
+	return result;
+}
+#endif
+
 union BVHPtr
 {
 	struct
@@ -241,10 +254,16 @@ class BVH
 			rtm::vec3 centroid(build_object.aabb.centroid() - aabb.min);
 			centroid = centroid * scale * max;
 
-			build_object.morton_code = 
+			build_object.morton_code =
+#if defined __aarch64__
+				_pdep_u64_sw((uint32_t)centroid.x, 0b0001001001001001001001001001001001001001001001001001001001001001ull) |
+				_pdep_u64_sw((uint32_t)centroid.y, 0b0010010010010010010010010010010010010010010010010010010010010010ull) |
+				_pdep_u64_sw((uint32_t)centroid.z, 0b0100100100100100100100100100100100100100100100100100100100100100ull);
+#else
 				_pdep_u64((uint32_t)centroid.x, 0b0001001001001001001001001001001001001001001001001001001001001001ull) |
 				_pdep_u64((uint32_t)centroid.y, 0b0010010010010010010010010010010010010010010010010010010010010010ull) |
 				_pdep_u64((uint32_t)centroid.z, 0b0100100100100100100100100100100100100100100100100100100100100100ull);
+#endif
 		}
 	}
 
@@ -300,7 +319,11 @@ class BVH
 		for(uint i = start; i < end; ++i)
 			mask |= bld_objs[i].morton_code ^ bld_objs[start].morton_code;
 
+#if defined __aarch64__
+		uint64_t prefix = mask ? __builtin_clzll(mask) : 64;
+#else
 		uint64_t prefix = _lzcnt_u64(mask);
+#endif
 		if(prefix == 64)
 			return (start + end) / 2;
 

include/rtm/float.hpp

@@ -7,13 +7,15 @@
 #include <cmath>
 	#if defined BUILD_PLATFORM_WINDOWS
 		#include <intrin.h>
+	#elif defined __aarch64__
+		#include <arm_neon.h>
 	#elif defined BUILD_PLATFORM_LINUX
 		#include <immintrin.h>
 	#endif
 #include <cfenv>
 #endif
 
-#ifdef BUILD_PLATFORM_LINUX
+#if defined BUILD_PLATFORM_LINUX && !defined __aarch64__
 	// Declarations of intrinsics used that are defined but not declared in immintrin
 	extern __m128 _mm_cos_ps(__m128 __A);
 	extern __m128 _mm_sin_ps(__m128 __A);
@@ -51,6 +53,8 @@ inline float sqrt(float input)
     float output;
 	asm volatile ("fsqrt.s %0, %1\n\t" : "=f" (output) : "f" (input));
 	return output;
+    #elif defined __aarch64__
+	return std::sqrt(input);
     #else
 	return _mm_cvtss_f32(_mm_sqrt_ss(_mm_set_ps1(input)));
 	#endif
@@ -62,6 +66,8 @@ inline float rsqrt(float input)
 	float output;
 	asm volatile ("frsqrt.s %0, %1\n\t" : "=f" (output) : "f" (input));
 	return output;
+	#elif defined __aarch64__
+	return 1.0f / std::sqrt(input);
 	#else
 	return _mm_cvtss_f32(_mm_rsqrt_ss(_mm_set_ps1(input)));
 	#endif
@@ -73,6 +79,8 @@ inline float rcp(float input)
 	float output;
 	asm volatile ("frcp.s %0, %1\n\t" : "=f" (output) : "f" (input));
 	return output;
+	#elif defined __aarch64__
+	return 1.0f / input;
 	#else
 	return _mm_cvtss_f32(_mm_rcp_ss(_mm_set_ps1(input)));
 	#endif
@@ -124,6 +132,8 @@ inline float cos(float input)
 {
 	#ifdef __riscv
 	return cos_32(input);
+	#elif defined __aarch64__
+	return std::cos(input);
 	#else
 	return _mm_cvtss_f32(_mm_cos_ps(_mm_set_ps1(input)));
 	#endif
@@ -133,6 +143,8 @@ inline float sin(float input)
 {
 	#ifdef __riscv
 	return sin_32(input);
+	#elif defined __aarch64__
+	return std::sin(input);
 	#else
 	return _mm_cvtss_f32(_mm_sin_ps(_mm_set_ps1(input)));
 	#endif

include/rtm/ftb.hpp

@@ -66,9 +66,15 @@ inline bool compress(uint prim_idx, uint prim_cnt, const Mesh& mesh, FTB* block
 	}
 
 	uint max_pfx = 0;
+#if defined __aarch64__
+	uint px = min((uint)(mask_x ? __builtin_clz(mask_x) : 32), max_pfx);
+	uint py = min((uint)(mask_y ? __builtin_clz(mask_y) : 32), max_pfx);
+	uint pz = min((uint)(mask_z ? __builtin_clz(mask_z) : 32), max_pfx);
+#else
 	uint px = min(_lzcnt_u32(mask_x), max_pfx);
 	uint py = min(_lzcnt_u32(mask_y), max_pfx);
 	uint pz = min(_lzcnt_u32(mask_z), max_pfx);
+#endif
 	uint nx = 32 - px;
 	uint ny = 32 - py;
 	uint nz = 32 - pz;

include/rtm/macros.hpp

@@ -2,17 +2,24 @@
 
 #if defined __x86_64__ || defined _M_X64
 #define __x86
+#elif defined __aarch64__ || defined _M_ARM64
+#define __aarch64
 #endif
 
 #if defined _WIN16 || defined WIN32 || defined _WIN32 || defined WIN64 || defined _WIN64 || defined __WIN32__ || defined __TOS_WIN__ || defined __WINDOWS__
 	#define BUILD_PLATFORM_WINDOWS
+#elif defined __APPLE__
+	#define BUILD_PLATFORM_MACOS
 #elif defined __linux__
 	#define BUILD_PLATFORM_LINUX
 #endif
 
 //To add breakpoints for debugging at runtime
 #if defined BUILD_PLATFORM_WINDOWS
 	#define add_breakpoint() __debugbreak()
+#elif defined BUILD_PLATFORM_MACOS
+	#include <signal.h>
+	#define add_breakpoint() raise(SIGTRAP)
 #elif defined BUILD_PLATFORM_LINUX
 	#include <signal.h>
 	#define add_breakpoint() raise(SIGINT)

include/rtm/mesh.hpp

@@ -303,7 +303,7 @@ class Mesh
 				vertex_indices.emplace_back(0);
 				tex_coord_indices.emplace_back(~0x0u);
 				normal_indices.emplace_back(~0x0u);
-				material_indices.emplace_back(material_names.size() - 1u);
+				material_indices.emplace_back(material_names.empty() ? 0u : (uint)(material_names.size() - 1u));
 				read_face(line + data_start_index, vertex_indices.back(), tex_coord_indices.back(), normal_indices.back());
 				break;
 

include/rtm/qtb.hpp

@@ -62,9 +62,15 @@ inline bool compress(const uint* prim_ids, uint num_tris, const Mesh& mesh, QTB&
 	}
 
 	uint max_pfx = 24;
+#if defined __aarch64__
+	uint px = min((uint)((mask_x << 8) ? __builtin_clz(mask_x << 8) : 32), max_pfx);
+	uint py = min((uint)((mask_y << 8) ? __builtin_clz(mask_y << 8) : 32), max_pfx);
+	uint pz = min((uint)((mask_z << 8) ? __builtin_clz(mask_z << 8) : 32), max_pfx);
+#else
 	uint px = min(_lzcnt_u32(mask_x << 8), max_pfx);
 	uint py = min(_lzcnt_u32(mask_y << 8), max_pfx);
 	uint pz = min(_lzcnt_u32(mask_z << 8), max_pfx);
+#endif
 	uint nx = 24 - px;
 	uint ny = 24 - py;
 	uint nz = 24 - pz;

include/rtm/texture.hpp

@@ -104,8 +104,9 @@ class Texture2D
 
 	rtm::vec2 get_fract_uv(const rtm::vec2& uv) const
 	{
-		rtm::vec2 fuv = uv * rtm::vec2(width, height);
-		return (fuv - rtm::vec2((int32_t)fuv[0], (int32_t)fuv[1]));
+		rtm::vec2 wrapped_uv(uv[0] - floorf(uv[0]), uv[1] - floorf(uv[1]));
+		rtm::vec2 fuv = wrapped_uv * rtm::vec2(width, height);
+		return rtm::vec2(fuv[0] - floorf(fuv[0]), fuv[1] - floorf(fuv[1]));
 	}
 
 	Texel* get_texel_addr(const rtm::uvec2& iuv) const

src/arches-v2/CMakeLists.txt

@@ -3,7 +3,7 @@ add_compile_definitions(UNICODE _UNICODE)
 
 set(PROJECT_NAME "arches-v2")
 
-if(LINUX)
+if(LINUX AND NOT APPLE)
   # enable intrinsics like _lzcnt_u64
   add_compile_options(-march=native)
 endif()
@@ -61,10 +61,18 @@ set_target_properties(${PROJECT_NAME} PROPERTIES OUTPUT_NAME ${PROJECT_NAME})
 if(WINDOWS)
   target_link_directories(${PROJECT_NAME} PUBLIC ${PROJECT_SOURCE_DIR}/libraries/tbb)
   target_link_libraries(${PROJECT_NAME} PRIVATE tbb12.lib)
+elseif(APPLE)
+  execute_process(COMMAND brew --prefix tbb OUTPUT_VARIABLE TBB_BREW_PREFIX OUTPUT_STRIP_TRAILING_WHITESPACE)
+  target_include_directories(${PROJECT_NAME} PUBLIC ${TBB_BREW_PREFIX}/include)
+  target_link_directories(${PROJECT_NAME} PUBLIC ${TBB_BREW_PREFIX}/lib)
+  target_link_libraries(${PROJECT_NAME} PRIVATE tbb)
 elseif(LINUX)
   target_link_libraries(${PROJECT_NAME} PRIVATE tbb)
 endif()
 target_link_libraries(${PROJECT_NAME} PRIVATE Ramulator)
+target_compile_definitions(${PROJECT_NAME} PRIVATE
+  ARCHES_SOURCE_DIR="${CMAKE_SOURCE_DIR}/"
+  ARCHES_BINARY_DIR="${CMAKE_CURRENT_BINARY_DIR}/")
 #set_target_properties(${PROJECT_NAME} PROPERTIES RUNTIME_OUTPUT_DIRECTORY_DEBUG ${CMAKE_CURRENT_BINARY_DIR})
 #set_target_properties(${PROJECT_NAME} PROPERTIES RUNTIME_OUTPUT_DIRECTORY_RELEASE ${CMAKE_CURRENT_BINARY_DIR})
 set_target_properties(${PROJECT_NAME} PROPERTIES FOLDER ${PROJECT_NAME})

src/arches-v2/isa/registers.cpp

@@ -1,5 +1,5 @@
 #include "registers.hpp"
-
+#include <cfenv>
 
 namespace Arches { namespace ISA { namespace RISCV {
 
@@ -19,6 +19,22 @@ FloatingPointRegisterFile::FloatingPointRegisterFile()
 {
 	fcsr.data = 0u;
 	//Make rounding mode match simulator rounding mode
+#if defined BUILD_ARCH_aarch64
+	switch (fegetround()) {
+	case FE_TONEAREST:
+		fcsr.frm = 0b000;
+		break;
+	case FE_DOWNWARD:
+		fcsr.frm = 0b010;
+		break;
+	case FE_UPWARD:
+		fcsr.frm = 0b011;
+		break;
+	case FE_TOWARDZERO:
+		fcsr.frm = 0b001;
+		break;
+	}
+#else
 	switch ((_mm_getcsr() >> 13) & 0b11) {
 	case 0b00: //nearest (even)
 		fcsr.frm = 0b000;
@@ -33,6 +49,7 @@ FloatingPointRegisterFile::FloatingPointRegisterFile()
 		fcsr.frm = 0b001;
 		break;
 	}
+#endif
 
 	for (int i = 0; i < sizeof(valid); ++i) valid[i] = true;
 }

src/arches-v2/isa/riscv.cpp

@@ -2,6 +2,8 @@
 
 #if defined BUILD_PLATFORM_WINDOWS
 	#include <intrin.h>
+#elif defined __aarch64__
+	#include <cmath>
 #elif defined BUILD_PLATFORM_LINUX
 	#include <immintrin.h>
 #endif
@@ -695,15 +697,27 @@ InstructionInfo const isa_OP_FP[32] = //r.funct5
 	InstructionInfo(0b010'10, IMPL_NONE),
 	InstructionInfo(0b010'11, "fsqrt.s", InstrType::FSQRT, Encoding::R, RegFile::FLOAT, EXEC_DECL 
 	{
+#if defined __aarch64__
+		unit->float_regs->registers[instr.r.rd].f32 = std::sqrt(unit->float_regs->registers[instr.r.rs1].f32);
+#else
 		unit->float_regs->registers[instr.r.rd].f32 = _mm_cvtss_f32(_mm_sqrt_ss(_mm_set_ps1(unit->float_regs->registers[instr.r.rs1].f32)));
+#endif
 	}),
 	InstructionInfo(0b011'00, "fisqrt.s", InstrType::FSQRT, Encoding::R, RegFile::FLOAT, EXEC_DECL
 	{
+#if defined __aarch64__
+		unit->float_regs->registers[instr.r.rd].f32 = 1.0f / std::sqrt(unit->float_regs->registers[instr.r.rs1].f32);
+#else
 		unit->float_regs->registers[instr.r.rd].f32 = _mm_cvtss_f32(_mm_rsqrt_ps(_mm_set_ps1(unit->float_regs->registers[instr.r.rs1].f32)));
+#endif
 	}),
 	InstructionInfo(0b011'01, "frcp.s", InstrType::FRCP, Encoding::R, RegFile::FLOAT, EXEC_DECL
 	{
+#if defined __aarch64__
+		unit->float_regs->registers[instr.r.rd].f32 = 1.0f / unit->float_regs->registers[instr.r.rs1].f32;
+#else
 		unit->float_regs->registers[instr.r.rd].f32 = _mm_cvtss_f32(_mm_rcp_ss(_mm_set_ps1(unit->float_regs->registers[instr.r.rs1].f32)));
+#endif
 	}),
 	InstructionInfo(0b011'10, IMPL_NONE),
 	InstructionInfo(0b011'11, IMPL_NONE),

src/arches-v2/main.cpp

@@ -219,14 +219,15 @@ static TRaXKernelArgs initilize_buffers(Units::UnitMainMemoryBase** drams, const
 	for(uint32_t i = 0; i < mesh.materials.size(); ++i)
 		mesh.materials[i].albedo_texture.texels = nullptr;  // to not free device memory textures
 
-	size_t temp = TRAX_KERNEL_ARGS_ADDRESS;
+	paddr_t temp = TRAX_KERNEL_ARGS_ADDRESS;
 	write_array(drams, xbar, 256, (uint8_t*)&args, sizeof(TRaXKernelArgs), temp);
 	return args;
 }
 
 static void run_sim_trax(SimulationConfig& sim_config)
 {
 	std::string project_folder_path = get_project_folder_path();
+	std::string project_binary_path = get_project_binary_path();
 
 #if 0 //RTX 4090 ish
 	//Compute
@@ -244,7 +245,7 @@ static void run_sim_trax(SimulationConfig& sim_config)
 
 	//DRAM
 	UnitDRAM::Configuration dram_config;
-	dram_config.config_path = project_folder_path + "build\\src\\arches-v2\\config-files\\gddr6x_21000_config.yaml";
+	dram_config.config_path = project_binary_path + "config-files/gddr6x_21000_config.yaml";
 	dram_config.size = 1ull << 30; //1GB per partition
 	dram_config.clock_ratio = dram_clock / core_clock;
 	dram_config.latency = 254;
@@ -305,7 +306,7 @@ static void run_sim_trax(SimulationConfig& sim_config)
 
 	//DRAM
 	UnitDRAM::Configuration dram_config;
-	dram_config.config_path = project_folder_path + "build\\src\\arches-v2\\config-files\\gddr6_14000_config.yaml";
+	dram_config.config_path = project_binary_path + "config-files/gddr6_14000_config.yaml";
 	dram_config.size = 1ull << 30; //1GB per partition
 	dram_config.clock_ratio = dram_clock / core_clock;
 	dram_config.latency = 254;
@@ -375,7 +376,7 @@ static void run_sim_trax(SimulationConfig& sim_config)
 
 	//DRAM
 	UnitDRAM::Configuration dram_config;
-	dram_config.config_path = project_folder_path + "build/src/arches-v2/config-files/gddr6_14000_config.yaml";
+	dram_config.config_path = project_binary_path + "config-files/gddr6_14000_config.yaml";
 	dram_config.size = 1ull << 30; //1GB
 	dram_config.clock_ratio = dram_clock / core_clock;
 	dram_config.latency = 92;
@@ -433,7 +434,7 @@ static void run_sim_trax(SimulationConfig& sim_config)
 
 	//DRAM
 	UnitDRAM::Configuration dram_config;
-	dram_config.config_path = project_folder_path + "build\\src\\arches-v2\\config-files\\gddr6_pch_config.yaml";
+	dram_config.config_path = project_binary_path + "config-files/gddr6_pch_config.yaml";
 	dram_config.size = 1ull << 30; //1GB
 	dram_config.clock_ratio = dram_clock / core_clock;
 	dram_config.latency = 1;

src/arches-v2/shared-utils.hpp

@@ -19,6 +19,9 @@
 #if defined BUILD_PLATFORM_WINDOWS
 	#include <Windows.h>
 	#define MAX_FILENAME_LENGTH MAX_PATH
+#elif defined BUILD_PLATFORM_MACOS
+	#include <climits>
+	#define MAX_FILENAME_LENGTH PATH_MAX
 #elif defined BUILD_PLATFORM_LINUX
 	#include <linux/limits.h>
 	#define MAX_FILENAME_LENGTH FILENAME_MAX
@@ -33,10 +36,12 @@ void set_full_exe_name(const char *name) {
 
 std::string get_project_folder_path()
 {
-	// CHAR path[MAX_PATH];
-	// GetModuleFileNameA(NULL, path, MAX_PATH);
-	std::string executable_path(full_exe_name);
-	return executable_path.substr(0, executable_path.rfind("build"));
+	return ARCHES_SOURCE_DIR;
+}
+
+std::string get_project_binary_path()
+{
+	return ARCHES_BINARY_DIR;
 }
 
 template <typename T>