ARM64Emitter: Implement instructions to move data to/from SP

Common/Arm64Emitter.cpp

@@ -1464,6 +1464,15 @@ void ARM64XEmitter::MOV(ARM64Reg Rd, ARM64Reg Rm)
 		_assert_msg_(JIT, false, "Non-GPRs not supported in MOV");
 	}
 }
+
+void ARM64XEmitter::MOVfromSP(ARM64Reg Rd) {
+	ADD(Rd, ARM64Reg::SP, 0, false);
+}
+
+void ARM64XEmitter::MOVtoSP(ARM64Reg Rn) {
+	ADD(ARM64Reg::SP, Rn, 0, false);
+}
+
 void ARM64XEmitter::MVN(ARM64Reg Rd, ARM64Reg Rm)
 {
 	ORN(Rd, Is64Bit(Rd) ? ZR : WZR, Rm, ArithOption(Rm, ST_LSL, 0));

Common/Arm64Emitter.h

@@ -566,6 +566,10 @@ public:
 	void MOV(ARM64Reg Rd, ARM64Reg Rm);
 	void MVN(ARM64Reg Rd, ARM64Reg Rm);
 
+	// Wrapper around ADD reg, reg, imm.
+	void MOVfromSP(ARM64Reg Rd);
+	void MOVtoSP(ARM64Reg Rn);
+
 	// TODO: These are "slow" as they use arith+shift, should be replaced with UBFM/EXTR variants.
 	void LSR(ARM64Reg Rd, ARM64Reg Rm, int shift);
 	void LSL(ARM64Reg Rd, ARM64Reg Rm, int shift);

Common/x64Emitter.h

@@ -391,6 +391,7 @@ public:
 	// INC and DEC are slow on Intel Core, but not on AMD. They create a
 	// false flag dependency because they only update a subset of the flags.
 	// XCHG is SLOW and should be avoided.
+	// Actually, REP MOVSB has gotten fast again on recent CPU generations, though it's not all that useful anyway.
 
 	// Debug breakpoint
 	void INT3();
@@ -410,7 +411,6 @@ public:
 	void LAHF(); // 3 cycle vector path
 	void SAHF(); // direct path fast
 
-
 	// Stack control
 	void PUSH(X64Reg reg);
 	void POP(X64Reg reg);

Core/MemMap.cpp

@@ -253,6 +253,7 @@ void MemoryMap_Shutdown(u32 flags) {
 
 void Init() {
 	// On some 32 bit platforms, you can only map < 32 megs at a time.
+	// TODO: Wait, wtf? What platforms are those? This seems bad.
 	const static int MAX_MMAP_SIZE = 31 * 1024 * 1024;
 	_dbg_assert_msg_(MEMMAP, g_MemorySize < MAX_MMAP_SIZE * 3, "ACK - too much memory for three mmap views.");
 	for (size_t i = 0; i < ARRAY_SIZE(views); i++) {

Core/Util/DisArm64.cpp

@@ -159,11 +159,15 @@ static void DataProcessingImmediate(uint32_t w, uint64_t addr, Instruction *inst
 		// Add/subtract immediate value
 		int op = (w >> 30) & 1;
 		int imm = ((w >> 10) & 0xFFF);
-		int shift = ((w >> 22) & 0x3) * 16;
+		int shift = ((w >> 22) & 0x1) * 12;
 		int s = ((w >> 29) & 1);
 		imm <<= shift;
 		if (s && Rd == 31) {
 			snprintf(instr->text, sizeof(instr->text), "cmp %c%d, #%d", r, Rn, imm);
+		} else if (!shift && Rn == 31 && imm == 0) {
+			snprintf(instr->text, sizeof(instr->text), "mov %c%d, sp", r, Rd);
+		} else if (!shift && Rd == 31 && imm == 0) {
+			snprintf(instr->text, sizeof(instr->text), "mov sp, %c%d", r, Rn);
 		} else {
 			snprintf(instr->text, sizeof(instr->text), "%s%s %c%d, %c%d, #%d", op == 0 ? "add" : "sub", s ? "s" : "", r, Rd, r, Rn, imm);
 		}

unittest/TestArm64Emitter.cpp

@@ -41,6 +41,11 @@ bool TestArm64Emitter() {
 	ARM64XEmitter emitter((u8 *)code);
 	ARM64FloatEmitter fp(&emitter);
 
+	emitter.MOVfromSP(X3);
+	RET(CheckLast(emitter, "910003e3 mov x3, sp"));
+	emitter.MOVtoSP(X11);
+	RET(CheckLast(emitter, "9100017f mov sp, x11"));
+
 	fp.EOR(Q0, Q1, Q2);
 	RET(CheckLast(emitter, "6e221c20 eor q0, q1, q2"));
 	fp.EOR(D0, D1, D2);