Random dataset accesses - asm only

Initial support for large pages

src/AssemblyGeneratorX86.cpp

@@ -28,6 +28,11 @@ namespace RandomX {
 	static const char* regR32[8] = { "r8d", "r9d", "r10d", "r11d", "r12d", "r13d", "r14d", "r15d" };
 	static const char* regF[8] = { "xmm8", "xmm9", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7" };
 
+	static const char* regMx = "edi";
+	static const char* regIc = "ebp";
+	static const char* regStackBeginAddr = "rbx";
+	static const char* regScratchpadAddr = "rsi";
+
 	void AssemblyGeneratorX86::generateProgram(const void* seed) {
 		asmCode.str(std::string()); //clear
 		Pcg32 gen(seed);
@@ -48,7 +53,7 @@ namespace RandomX {
 
 	void AssemblyGeneratorX86::generateCode(Instruction& instr, int i) {
 		asmCode << "rx_i_" << i << ": ;" << instr.getName() << std::endl;
-		asmCode << "\tdec edi" << std::endl;
+		asmCode << "\tdec " << regIc << std::endl;
 		asmCode << "\tjz rx_finish" << std::endl;
 		auto generator = engine[instr.opcode];
 		(this->*generator)(instr, i);
@@ -56,54 +61,34 @@ namespace RandomX {
 
 	void AssemblyGeneratorX86::genar(Instruction& instr) {
 		asmCode << "\txor " << regR[instr.rega % RegistersCount] << ", 0" << std::hex << instr.addra << "h" << std::dec << std::endl;
-		switch (instr.loca & 7)
+		asmCode << "\tmov ecx, " << regR32[instr.rega % RegistersCount] << std::endl;
+		switch (instr.loca & 3)
 		{
-		case 0:
-		case 1:
-		case 2:
-		case 3:
-			asmCode << "\tmov ecx, " << regR32[instr.rega % RegistersCount] << std::endl;
-			asmCode << "\tcall rx_read_dataset_r" << std::endl;
-			return;
-
-		case 4:
-			asmCode << "\tmov eax, " << regR32[instr.rega % RegistersCount] << std::endl;
-			asmCode << "\tand eax, " << (ScratchpadL2 - 1) << std::endl;
-			asmCode << "\tmov rax, qword ptr [rsi + rax * 8]" << std::endl;
-			return;
-
-		default:
-			asmCode << "\tmov eax, " << regR32[instr.rega % RegistersCount] << std::endl;
-			asmCode << "\tand eax, " << (ScratchpadL1 - 1) << std::endl;
-			asmCode << "\tmov rax, qword ptr [rsi + rax * 8]" << std::endl;
-			return;
+			case 0:
+			case 1:
+			case 2:
+				asmCode << "\tcall rx_readint_l1" << std::endl;
+				return;
+			default: //3
+				asmCode << "\tcall rx_readint_l2" << std::endl;
+				return;
 		}
 	}
 
 
 	void AssemblyGeneratorX86::genaf(Instruction& instr) {
 		asmCode << "\txor " << regR[instr.rega % RegistersCount] << ", 0" << std::hex << instr.addra << "h" << std::dec << std::endl;
-		switch (instr.loca & 7)
+		asmCode << "\tmov ecx, " << regR32[instr.rega % RegistersCount] << std::endl;
+		switch (instr.loca & 3)
 		{
-		case 0:
-		case 1:
-		case 2:
-		case 3:
-			asmCode << "\tmov ecx, " << regR32[instr.rega % RegistersCount] << std::endl;
-			asmCode << "\tcall rx_read_dataset_f" << std::endl;
-			return;
-
-		case 4:
-			asmCode << "\tmov eax, " << regR32[instr.rega % RegistersCount] << std::endl;
-			asmCode << "\tand eax, " << (ScratchpadL2 - 1) << std::endl;
-			asmCode << "\tcvtdq2pd xmm0, qword ptr [rsi + rax * 8]" << std::endl;
-			return;
-
-		default:
-			asmCode << "\tmov eax, " << regR32[instr.rega % RegistersCount] << std::endl;
-			asmCode << "\tand eax, " << (ScratchpadL1 - 1) << std::endl;
-			asmCode << "\tcvtdq2pd xmm0, qword ptr [rsi + rax * 8]" << std::endl;
-			return;
+			case 0:
+			case 1:
+			case 2:
+				asmCode << "\tcall rx_readfloat_l1" << std::endl;
+				return;
+			default: //3
+				asmCode << "\tcall rx_readfloat_l2" << std::endl;
+				return;
 		}
 	}
 
@@ -169,9 +154,9 @@ namespace RandomX {
 			asmCode << "\tmov eax, " << regR32[instr.regc % RegistersCount] << std::endl;
 			asmCode << "\txor eax, 0" << std::hex << instr.addrc << "h" << std::dec << std::endl;
 			asmCode << "\tand eax, " << (ScratchpadL2 - 1) << std::endl;
-			asmCode << "\tmov qword ptr [rsi + rax * 8], rcx" << std::endl;
+			asmCode << "\tmov qword ptr [" << regScratchpadAddr << " + rax * 8], rcx" << std::endl;
 			if (trace) {
-				asmCode << "\tmov qword ptr [rsi + rdi * 8 + 262136], rcx" << std::endl;
+				asmCode << "\tmov qword ptr [" << regScratchpadAddr << " + rdi * 8 + 262136], rcx" << std::endl;
 			}
 			return;
 
@@ -182,31 +167,31 @@ namespace RandomX {
 			asmCode << "\tmov eax, " << regR32[instr.regc % RegistersCount] << std::endl;
 			asmCode << "\txor eax, 0" << std::hex << instr.addrc << "h" << std::dec << std::endl;
 			asmCode << "\tand eax, " << (ScratchpadL1 - 1) << std::endl;
-			asmCode << "\tmov qword ptr [rsi + rax * 8], rcx" << std::endl;
+			asmCode << "\tmov qword ptr [" << regScratchpadAddr << " + rax * 8], rcx" << std::endl;
 			if (trace) {
-				asmCode << "\tmov qword ptr [rsi + rdi * 8 + 262136], rcx" << std::endl;
+				asmCode << "\tmov qword ptr [" << regScratchpadAddr << " + rdi * 8 + 262136], rcx" << std::endl;
 			}
 			return;
 
 		default:
 			asmCode << "\tmov " << regR[instr.regc % RegistersCount] << ", rax" << std::endl;
 			if (trace) {
-				asmCode << "\tmov qword ptr [rsi + rdi * 8 + 262136], rax" << std::endl;
+				asmCode << "\tmov qword ptr [" << regScratchpadAddr << " + rdi * 8 + 262136], rax" << std::endl;
 			}
+			return;
 		}
 	}
 
-	void AssemblyGeneratorX86::gencf(Instruction& instr, bool alwaysLow = false) {
-		if(!alwaysLow)
-			asmCode << "\tmovaps " << regF[instr.regc % RegistersCount] << ", xmm0" << std::endl;
-		const char* store = (!alwaysLow && (instr.locc & 8)) ? "movhpd" : "movlpd";
+	void AssemblyGeneratorX86::gencf(Instruction& instr) {
+		asmCode << "\tmovaps " << regF[instr.regc % RegistersCount] << ", xmm0" << std::endl;
+		const char* store = (instr.locc & 8) ? "movhpd" : "movlpd";
 		switch (instr.locc & 7)
 		{
 			case 4:
 				asmCode << "\tmov eax, " << regR32[instr.regc % RegistersCount] << std::endl;
 				asmCode << "\txor eax, 0" << std::hex << instr.addrc << "h" << std::dec << std::endl;
 				asmCode << "\tand eax, " << (ScratchpadL2 - 1) << std::endl;
-				asmCode << "\t" << store << " qword ptr [rsi + rax * 8], " << regF[instr.regc % RegistersCount] << std::endl;
+				asmCode << "\t" << store << " qword ptr [" << regScratchpadAddr << " + rax * 8], " << regF[instr.regc % RegistersCount] << std::endl;
 				break;
 
 			case 5:
@@ -215,11 +200,11 @@ namespace RandomX {
 				asmCode << "\tmov eax, " << regR32[instr.regc % RegistersCount] << std::endl;
 				asmCode << "\txor eax, 0" << std::hex << instr.addrc << "h" << std::dec << std::endl;
 				asmCode << "\tand eax, " << (ScratchpadL1 - 1) << std::endl;
-				asmCode << "\t" << store << " qword ptr [rsi + rax * 8], " << regF[instr.regc % RegistersCount] << std::endl;
+				asmCode << "\t" << store << " qword ptr [" << regScratchpadAddr << " + rax * 8], " << regF[instr.regc % RegistersCount] << std::endl;
 				break;
 		}
 		if (trace) {
-			asmCode << "\t" << store << " qword ptr [rsi + rdi * 8 + 262136], " << regF[instr.regc % RegistersCount] << std::endl;
+			asmCode << "\t" << store << " qword ptr [" << regScratchpadAddr << " + rdi * 8 + 262136], " << regF[instr.regc % RegistersCount] << std::endl;
 		}
 	}
 
@@ -454,15 +439,14 @@ namespace RandomX {
 
 	void AssemblyGeneratorX86::h_FPROUND(Instruction& instr, int i) {
 		genar(instr);
-		asmCode << "\tmov rcx, rax" << std::endl;
+		//asmCode << "\tmov rcx, rax" << std::endl;
 		asmCode << "\tshl eax, 13" << std::endl;
-		asmCode << "\tand rcx, -2048" << std::endl;
+		//asmCode << "\tand rcx, -2048" << std::endl;
 		asmCode << "\tand eax, 24576" << std::endl;
-		asmCode << "\tcvtsi2sd " << regF[instr.regc % RegistersCount] << ", rcx" << std::endl;
+		//asmCode << "\tmovaps " << regF[instr.regc % RegistersCount] << ", xmm0" << std::endl;
 		asmCode << "\tor eax, 40896" << std::endl;
 		asmCode << "\tmov dword ptr [rsp - 8], eax" << std::endl;
 		asmCode << "\tldmxcsr dword ptr [rsp - 8]" << std::endl;
-		gencf(instr, true);
 	}
 
 	static inline const char* jumpCondition(Instruction& instr, bool invert = false) {
@@ -496,7 +480,7 @@ namespace RandomX {
 		asmCode << "\tjmp rx_i_" << wrapInstr(i + 1) << std::endl;
 		asmCode << "taken_call_" << i << ":" << std::endl;
 		if (trace) {
-			asmCode << "\tmov qword ptr [rsi + rdi * 8 + 262136], rax" << std::endl;
+			asmCode << "\tmov qword ptr [" << regScratchpadAddr << " + rdi * 8 + 262136], rax" << std::endl;
 		}
 		asmCode << "\tpush rax" << std::endl;
 		asmCode << "\tcall rx_i_" << wrapInstr(i + (instr.imm8 & 127) + 2) << std::endl;
@@ -504,7 +488,7 @@ namespace RandomX {
 
 	void AssemblyGeneratorX86::h_RET(Instruction& instr, int i) {
 		genar(instr);
-		asmCode << "\tcmp rsp, rbp" << std::endl;
+		asmCode << "\tcmp rsp, " << regStackBeginAddr << std::endl;
 		asmCode << "\tje short not_taken_ret_" << i << std::endl;
 		asmCode << "\txor rax, qword ptr [rsp + 8]" << std::endl;
 		gencr(instr);

src/AssemblyGeneratorX86.hpp

@@ -45,7 +45,7 @@ namespace RandomX {
 		void genbr132(Instruction&);
 		void genbf(Instruction&, const char*);
 		void gencr(Instruction&);
-		void gencf(Instruction&, bool);
+		void gencf(Instruction&);
 
 		void generateCode(Instruction&, int);
 

src/CompiledVirtualMachine.cpp

@@ -47,8 +47,8 @@ namespace RandomX {
 	}
 
 	void CompiledVirtualMachine::execute() {
-		//executeProgram(reg, mem, scratchpad, readDataset);
-		compiler.getProgramFunc()(reg, mem, scratchpad);
+		executeProgram(reg, mem, scratchpad, readDataset);
+		//compiler.getProgramFunc()(reg, mem, scratchpad);
 #ifdef TRACEVM
 		for (int32_t i = InstructionCount - 1; i >= 0; --i) {
 			std::cout << std::hex << tracepad[i].u64 << std::endl;

src/VirtualMachine.hpp

@@ -40,6 +40,6 @@ namespace RandomX {
 		DatasetReadFunc readDataset;
 		alignas(16) RegisterFile reg;
 		MemoryRegisters mem;
-		alignas(16) convertible_t scratchpad[ScratchpadLength];
+		alignas(64) convertible_t scratchpad[ScratchpadLength];
 	};
 }

src/dataset.cpp

@@ -26,6 +26,7 @@ along with RandomX.  If not, see<http://www.gnu.org/licenses/>.
 #include "dataset.hpp"
 #include "Pcg32.hpp"
 #include "Cache.hpp"
+#include "virtualMemory.hpp"
 
 #if defined(__SSE2__)
 #include <wmmintrin.h>
@@ -161,12 +162,17 @@ namespace RandomX {
 	template
 		convertible_t datasetReadLight<true>(addr_t addr, MemoryRegisters& memory);
 
-	void datasetAlloc(dataset_t& ds) {
+	void datasetAlloc(dataset_t& ds, bool largePages) {
 		if (sizeof(size_t) <= 4)
 			throw std::runtime_error("Platform doesn't support enough memory for the dataset");
-		ds.dataset = (uint8_t*)_mm_malloc(DatasetSize, /*sizeof(__m128i)*/ 64);
-		if (ds.dataset == nullptr) {
-			throw std::runtime_error("Dataset memory allocation failed. >4 GiB of free virtual memory is needed.");
+		if (largePages) {
+			ds.dataset = (uint8_t*)allocLargePagesMemory(DatasetSize);
+		}
+		else {
+			ds.dataset = (uint8_t*)_mm_malloc(DatasetSize, 64);
+			if (ds.dataset == nullptr) {
+				throw std::runtime_error("Dataset memory allocation failed. >4 GiB of free virtual memory is needed.");
+			}
 		}
 	}
 

src/dataset.hpp

@@ -35,7 +35,7 @@ namespace RandomX {
 	template<bool softAes>
 	void initBlock(const uint8_t* cache, uint8_t* block, uint32_t blockNumber, const KeysContainer& keys);
 
-	void datasetAlloc(dataset_t& ds);
+	void datasetAlloc(dataset_t& ds, bool largePages);
 
 	template<bool softAes>
 	void datasetInit(Cache* cache, dataset_t ds, uint32_t startBlock, uint32_t blockCount);

src/executeProgram-win64.asm

@@ -15,19 +15,19 @@
 ;# You should have received a copy of the GNU General Public License
 ;# along with RandomX.  If not, see<http://www.gnu.org/licenses/>.
 
-PUBLIC executeProgram
+_RANDOMX_EXECUTE_PROGRAM SEGMENT PAGE READ EXECUTE
 
-.code
+PUBLIC executeProgram
 
 executeProgram PROC
 	; REGISTER ALLOCATION:
 	; rax -> temporary
-	; rbx -> MemoryRegisters& memory
+	; rbx -> beginning of VM stack
 	; rcx -> temporary
 	; rdx -> temporary
 	; rsi -> convertible_t& scratchpad
-	; rdi -> "ic" (instruction counter)
-	; rbp	-> beginning of VM stack
+	; rdi -> "mx"
+	; rbp	-> "ic"
 	; rsp -> end of VM stack
 	; r8 	-> "r0"
 	; r9 	-> "r1"
@@ -55,7 +55,8 @@ executeProgram PROC
 	;   | saved registers
 	;   |
 	;   v
-	; [rbp] RegisterFile& registerFile
+	; [rbx+8] RegisterFile& registerFile
+	; [rbx+0] uint8_t* dataset
 	;   |
 	;   |
 	;   | VM stack
@@ -80,17 +81,18 @@ executeProgram PROC
 	movdqu xmmword ptr [rsp+0], xmm10
 
 	; function arguments
-	push rcx				; RegisterFile& registerFile
-	mov rbx, rdx		; MemoryRegisters& memory
-	mov rsi, r8			; convertible_t& scratchpad
-	push r9
+	push rcx                    ; RegisterFile& registerFile
+	mov edi, dword ptr [rdx]    ; "mx"
+	mov rax, qword ptr [rdx+8]  ; uint8_t* dataset
+	push rax
+	mov rsi, r8                 ; convertible_t* scratchpad
 
-	mov rbp, rsp			; beginning of VM stack
-	mov rdi, 1048577	; number of VM instructions to execute + 1
+	mov rbx, rsp                ; beginning of VM stack
+	mov ebp, 524289             ; number of VM instructions to execute + 1
 
 	xorps xmm10, xmm10
 	cmpeqpd xmm10, xmm10
-	psrlq xmm10, 1		; mask for absolute value = 0x7fffffffffffffff7fffffffffffffff
+	psrlq xmm10, 1		          ; mask for absolute value = 0x7fffffffffffffff7fffffffffffffff
 
 	; reset rounding mode
 	mov dword ptr [rsp-8], 40896
@@ -162,7 +164,7 @@ executeProgram PROC
 
 rx_finish:
 	; unroll the stack
-	mov rsp, rbp
+	mov rsp, rbx
 
 	; save VM register values
 	pop rcx
@@ -202,57 +204,103 @@ rx_finish:
 	pop rbx
 
 	; return
-	ret	0
+	ret
+	
+TransformAddress MACRO reg32, reg64
+;# Transforms the address in the register so that the transformed address
+;# lies in a different cache line than the original address (mod 2^N).
+;# This is done to prevent a load-store dependency.
+;# There are 3 different transformations that can be used: x -> 9*x+C, x -> x+C, x -> x^C
+	lea reg32, [reg64+reg64*8+127]  ;# C = -119 -110 -101 -92 -83 -74 -65 -55 -46 -37 -28 -19 -10 -1 9 18 27 36 45 54 63 73 82 91 100 109 118 127
+	;lea reg32, [reg64-128]         ;# C = all except -7 to +7
+	;xor reg32, -8                  ;# C = all except 0 to 7
+ENDM
+
+ReadMemoryRandom MACRO spmask, float
+;# IN     ecx = random 32-bit address
+;# OUT    rax = 64-bit integer return value
+;# OUT   xmm0 = 128-bit floating point return value
+;# GLOBAL rbp = "ic" number of instructions until the end of the program
+;# GLOBAL rbx = address of the dataset address
+;# GLOBAL rsi = address of the scratchpad
+;# GLOBAL rdi = "mx" random 32-bit dataset address
+;# MODIFY rcx, rdx
+LOCAL L_prefetch, L_read, L_return
+	mov eax, ebp
+	and al, 63
+	jz short L_prefetch     ;# "ic" divisible by 64 -> prefetch
+	xor edx, edx
+	cmp al, 14
+	je short L_read         ;# "ic" = 14 (mod 64) -> random read
+	cmovb edx, ecx          ;# "ic" < 14 (mod 64) -> modify random read address
+	xor edi, edx
+L_return:
+	and ecx, spmask                 ;# limit address to the specified scratchpad size
+IF float
+	cvtdq2pd xmm0, qword ptr [rsi+rcx*8]
+ELSE
+	mov rax, qword ptr [rsi+rcx*8]
+ENDIF
+	ret	
+L_prefetch:
+	mov rax, qword ptr [rbx]        ;# load the dataset address
+	and edi, -64                    ;# align "mx" to the start of a cache line
+	prefetchnta byte ptr [rax+rdi]
+	jmp short L_return
+L_read:
+	push rcx
+	TransformAddress ecx, rcx       ;# TransformAddress function
+	and ecx, spmask-7               ;# limit address to the specified scratchpad size aligned to multiple of 8
+	call rx_read_dataset
+	pop rcx
+	jmp short L_return
+ENDM
+
+ALIGN 64
+rx_readint_l1:
+ReadMemoryRandom 2047, 0
 
+ALIGN 64
+rx_readint_l2:
+ReadMemoryRandom 32767, 0
+
+ALIGN 64
+rx_readfloat_l1:
+ReadMemoryRandom 2047, 1
+
+ALIGN 64
+rx_readfloat_l2:
+ReadMemoryRandom 32767, 1
+
+ALIGN 64
 rx_read_dataset:
-	push r8
-	push r9
-	push r10
-	push r11
-	mov rdx, rbx
-	movd qword ptr [rsp - 8], xmm1
-	movd qword ptr [rsp - 16], xmm2
-	sub rsp, 48
-	call qword ptr [rbp]
-	add rsp, 48
-	movd xmm2, qword ptr [rsp - 16]
-	movd xmm1, qword ptr [rsp - 8]
-	pop r11
-	pop r10
-	pop r9
-	pop r8
-	ret 0
-
-rx_read_dataset_r:
-	mov edx, dword ptr [rbx]	; ma
-	mov rax, qword ptr [rbx+8]	; dataset
-	mov rax, qword ptr [rax+rdx]
-	add dword ptr [rbx], 8
-	xor ecx, dword ptr [rbx+4]	; mx
-	mov dword ptr [rbx+4], ecx
-	test ecx, 0FFF8h
-	jne short rx_read_dataset_r_ret
-	and ecx, -8
-	mov dword ptr [rbx], ecx
-	mov rdx, qword ptr [rbx+8]
-	prefetcht0 byte ptr [rdx+rcx]
-rx_read_dataset_r_ret:
-	ret 0
-
-rx_read_dataset_f:
-	mov edx, dword ptr [rbx]	; ma
-	mov rax, qword ptr [rbx+8]	; dataset
-	cvtdq2pd xmm0, qword ptr [rax+rdx]
-	add dword ptr [rbx], 8
-	xor ecx, dword ptr [rbx+4]	; mx
-	mov dword ptr [rbx+4], ecx
-	test ecx, 0FFF8h
-	jne short rx_read_dataset_f_ret
-	and ecx, -8
-	mov dword ptr [rbx], ecx
-	prefetcht0 byte ptr [rax+rcx]
-rx_read_dataset_f_ret:
-	ret 0
+;# IN     rcx = scratchpad index - must be divisible by 8
+;# GLOBAL rbx = address of the dataset address
+;# GLOBAL rsi = address of the scratchpad
+;# GLOBAL rdi = "mx" random 32-bit dataset address
+;# MODIFY rax, rcx, rdx
+	mov rax, qword ptr [rbx]        ;# load the dataset address
+	lea rcx, [rsi+rcx*8]            ;# scratchpad cache line
+	lea rax, [rax+rdi]              ;# dataset cache line	              
+	mov rdx, qword ptr [rax+0]      ;# load first dataset quadword (prefetched into the cache by now)
+	xor qword ptr [rcx+0], rdx      ;# XOR the dataset item with a scratchpad item, repeat for the rest of the cacheline
+	mov rdx, qword ptr [rax+8]
+	xor qword ptr [rcx+8], rdx
+	mov rdx, qword ptr [rax+16]
+	xor qword ptr [rcx+16], rdx
+	mov rdx, qword ptr [rax+24]
+	xor qword ptr [rcx+24], rdx
+	mov rdx, qword ptr [rax+32]
+	xor qword ptr [rcx+32], rdx
+	mov rdx, qword ptr [rax+40]
+	xor qword ptr [rcx+40], rdx
+	mov rdx, qword ptr [rax+48]
+	xor qword ptr [rcx+48], rdx
+	mov rdx, qword ptr [rax+56]
+	xor qword ptr [rcx+56], rdx
+	ret
 executeProgram ENDP
 
+_RANDOMX_EXECUTE_PROGRAM ENDS
+
 END

src/main.cpp

@@ -162,7 +162,7 @@ void mine(RandomX::VirtualMachine* vm, std::atomic<int>& atomicNonce, AtomicHash
 }
 
 int main(int argc, char** argv) {
-	bool softAes, lightClient, genAsm, compiled, help;
+	bool softAes, lightClient, genAsm, compiled, help, largePages;
 	int programCount, threadCount;
 	readOption("--help", argc, argv, help);
 
@@ -177,6 +177,7 @@ int main(int argc, char** argv) {
 	readOption("--compiled", argc, argv, compiled);
 	readIntOption("--threads", argc, argv, threadCount, 1);
 	readIntOption("--nonces", argc, argv, programCount, 1000);
+	readOption("--largePages", argc, argv, largePages);
 
 	if (genAsm) {
 		generateAsm(programCount);
@@ -216,7 +217,7 @@ int main(int argc, char** argv) {
 		}
 		else {
 			RandomX::Cache* cache = dataset.cache;
-			RandomX::datasetAlloc(dataset);
+			RandomX::datasetAlloc(dataset, largePages);
 			if (threadCount > 1) {
 				auto perThread = RandomX::DatasetBlockCount / threadCount;
 				auto remainder = RandomX::DatasetBlockCount % threadCount;

src/virtualMemory.cpp

@@ -0,0 +1,108 @@
+/*
+Copyright (c) 2018 tevador
+
+This file is part of RandomX.
+
+RandomX is free software: you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation, either version 3 of the License, or
+(at your option) any later version.
+
+RandomX is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License
+along with RandomX.  If not, see<http://www.gnu.org/licenses/>.
+*/
+
+#include "virtualMemory.hpp"
+
+#include <stdexcept>
+
+#ifdef _WIN32
+#include <windows.h>
+#else
+#ifdef __APPLE__
+#include <mach/vm_statistics.h>
+#endif
+#include <sys/types.h>
+#include <sys/mman.h>
+#ifndef MAP_ANONYMOUS
+#define MAP_ANONYMOUS MAP_ANON
+#endif
+#endif
+
+#ifdef _WIN32
+std::string getErrorMessage(const char* function) {
+	LPSTR messageBuffer = nullptr;
+	size_t size = FormatMessageA(FORMAT_MESSAGE_ALLOCATE_BUFFER | FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_IGNORE_INSERTS,
+		NULL, GetLastError(), MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT), (LPSTR)&messageBuffer, 0, NULL);
+	std::string message(messageBuffer, size);
+	LocalFree(messageBuffer);
+	return std::string(function) + std::string(": ") + message;
+}
+
+void setPrivilege(const char* pszPrivilege, BOOL bEnable) {
+	HANDLE           hToken;
+	TOKEN_PRIVILEGES tp;
+	BOOL             status;
+	DWORD            error;
+
+	if (!OpenProcessToken(GetCurrentProcess(), TOKEN_ADJUST_PRIVILEGES | TOKEN_QUERY, &hToken))
+		throw std::runtime_error(getErrorMessage("OpenProcessToken"));
+
+	if (!LookupPrivilegeValue(NULL, pszPrivilege, &tp.Privileges[0].Luid))
+		throw std::runtime_error(getErrorMessage("LookupPrivilegeValue"));
+
+	tp.PrivilegeCount = 1;
+
+	if (bEnable)
+		tp.Privileges[0].Attributes = SE_PRIVILEGE_ENABLED;
+	else
+		tp.Privileges[0].Attributes = 0;
+
+	status = AdjustTokenPrivileges(hToken, FALSE, &tp, 0, (PTOKEN_PRIVILEGES)NULL, 0);
+
+	error = GetLastError();
+	if (!status || (error != ERROR_SUCCESS))
+		throw std::runtime_error(getErrorMessage("AdjustTokenPrivileges"));
+
+	if (!CloseHandle(hToken))
+		throw std::runtime_error(getErrorMessage("CloseHandle"));
+}
+#endif
+
+void* allocExecutableMemory(size_t bytes) {
+	void* mem;
+#ifdef _WIN32
+	mem = VirtualAlloc(nullptr, bytes, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
+	if (mem == nullptr)
+		throw std::runtime_error(getErrorMessage("allocExecutableMemory - VirtualAlloc"));
+#else
+	mem = mmap(nullptr, CodeSize, PROT_READ | PROT_WRITE | PROT_EXEC, MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
+	if (mem == MAP_FAILED)
+		throw std::runtime_error("allocExecutableMemory - mmap failed");
+#endif
+	return mem;
+}
+
+void* allocLargePagesMemory(size_t bytes) {
+	void* mem;
+#ifdef _WIN32
+	setPrivilege("SeLockMemoryPrivilege", 1);
+	mem = VirtualAlloc(NULL, bytes, MEM_COMMIT | MEM_RESERVE | MEM_LARGE_PAGES, PAGE_READWRITE);
+	if (mem == nullptr)
+		throw std::runtime_error(getErrorMessage("allocLargePagesMemory - VirtualAlloc"));
+#else
+#ifdef __APPLE__
+	mem = mmap(nullptr, bytes, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, VM_FLAGS_SUPERPAGE_SIZE_2MB, 0);
+#else
+	mem = mmap(nullptr, bytes, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS | MAP_HUGETLB | MAP_POPULATE, -1, 0);
+#endif
+	if (mem == MAP_FAILED)
+		throw std::runtime_error("allocLargePagesMemory - mmap failed");
+#endif
+	return mem;
+}

src/virtualMemory.hpp

@@ -0,0 +1,23 @@
+/*
+Copyright (c) 2018 tevador
+
+This file is part of RandomX.
+
+RandomX is free software: you can redistribute it and/or modify
+it under the terms of the GNU General Public License as published by
+the Free Software Foundation, either version 3 of the License, or
+(at your option) any later version.
+
+RandomX is distributed in the hope that it will be useful,
+but WITHOUT ANY WARRANTY; without even the implied warranty of
+MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+GNU General Public License for more details.
+
+You should have received a copy of the GNU General Public License
+along with RandomX.  If not, see<http://www.gnu.org/licenses/>.
+*/
+
+#pragma once
+
+void* allocExecutableMemory(size_t);
+void* allocLargePagesMemory(size_t);