jit: Add more reasonable estimates for RX protect.

Common/CodeBlock.h

@@ -80,7 +80,7 @@ public:
 		// If not WX Exclusive, no need to call ProtectMemoryPages because we never change the protection from RWX.
 		PoisonMemory(offset);
 		ResetCodePtr(offset);
-		if (PlatformIsWXExclusive()) {
+		if (PlatformIsWXExclusive() && offset != 0) {
 			// Need to re-protect the part we didn't clear.
 			ProtectMemoryPages(region, offset, MEM_PROT_READ | MEM_PROT_EXEC);
 		}

Common/Thunk.cpp

@@ -43,7 +43,7 @@ void ThunkManager::Init()
 #endif
 
 	AllocCodeSpace(THUNK_ARENA_SIZE);
-	BeginWrite();
+	BeginWrite(512);
 	save_regs = GetCodePtr();
 #if PPSSPP_ARCH(AMD64)
 	for (int i = 2; i < ABI_GetNumXMMRegs(); i++)
@@ -151,7 +151,7 @@ const void *ThunkManager::ProtectFunction(const void *function, int num_params)
 
 	_assert_msg_(region != nullptr, "Can't protect functions before the emu is started.");
 
-	BeginWrite();
+	BeginWrite(128);
 	const u8 *call_point = GetCodePtr();
 	Enter(this, true);
 

Core/MIPS/ARM/ArmAsm.cpp

@@ -71,8 +71,8 @@ namespace MIPSComp {
 using namespace ArmJitConstants;
 
 void ArmJit::GenerateFixedCode() {
+	BeginWrite(GetMemoryProtectPageSize());
 	const u8 *start = AlignCodePage();
-	BeginWrite();
 
 	// LR == SCRATCHREG2 on ARM32 so it needs to be pushed.
 	restoreRoundingMode = AlignCode16(); {

Core/MIPS/ARM/ArmJit.cpp

@@ -234,7 +234,7 @@ void ArmJit::Compile(u32 em_address) {
 		ClearCache();
 	}
 
-	BeginWrite();
+	BeginWrite(JitBlockCache::MAX_BLOCK_INSTRUCTIONS * 16);
 
 	int block_num = blocks.AllocateBlock(em_address);
 	JitBlock *b = blocks.GetBlock(block_num);

Core/MIPS/ARM64/Arm64Asm.cpp

@@ -97,8 +97,8 @@ namespace MIPSComp {
 using namespace Arm64JitConstants;
 
 void Arm64Jit::GenerateFixedCode(const JitOptions &jo) {
+	BeginWrite(GetMemoryProtectPageSize());
 	const u8 *start = AlignCodePage();
-	BeginWrite();
 
 	if (jo.useStaticAlloc) {
 		saveStaticRegisters = AlignCode16();

Core/MIPS/ARM64/Arm64Jit.cpp

@@ -230,7 +230,7 @@ void Arm64Jit::Compile(u32 em_address) {
 		ClearCache();
 	}
 
-	BeginWrite(4);
+	BeginWrite(JitBlockCache::MAX_BLOCK_INSTRUCTIONS * 16);
 
 	int block_num = blocks.AllocateBlock(em_address);
 	JitBlock *b = blocks.GetBlock(block_num);

Core/MIPS/x86/Asm.cpp

@@ -66,8 +66,8 @@ void ImHere() {
 }
 
 void Jit::GenerateFixedCode(JitOptions &jo) {
+	BeginWrite(GetMemoryProtectPageSize());
 	AlignCodePage();
-	BeginWrite();
 
 	restoreRoundingMode = AlignCode16(); {
 		STMXCSR(MIPSSTATE_VAR(temp));

Core/MIPS/x86/Jit.cpp

@@ -281,7 +281,8 @@ void Jit::Compile(u32 em_address) {
 		return;
 	}
 
-	BeginWrite();
+	// Sometimes we compile fairly large blocks, although it's uncommon.
+	BeginWrite(JitBlockCache::MAX_BLOCK_INSTRUCTIONS * 16);
 
 	int block_num = blocks.AllocateBlock(em_address);
 	JitBlock *b = blocks.GetBlock(block_num);

Core/MIPS/x86/JitSafeMem.cpp

@@ -462,7 +462,7 @@ void JitSafeMemFuncs::Init(ThunkManager *thunks) {
 	AllocCodeSpace(FUNCS_ARENA_SIZE);
 	thunks_ = thunks;
 
-	BeginWrite();
+	BeginWrite(1024);
 	readU32 = GetCodePtr();
 	CreateReadFunc(32, (const void *)&Memory::Read_U32);
 	readU16 = GetCodePtr();

GPU/Common/VertexDecoderArm.cpp

@@ -161,7 +161,7 @@ static const JitLookup jitLookup[] = {
 
 JittedVertexDecoder VertexDecoderJitCache::Compile(const VertexDecoder &dec, int32_t *jittedSize) {
 	dec_ = &dec;
-	BeginWrite();
+	BeginWrite(4096);
 	const u8 *start = AlignCode16();
 
 	bool prescaleStep = false;

GPU/Common/VertexDecoderArm64.cpp

@@ -143,7 +143,7 @@ static const JitLookup jitLookup[] = {
 JittedVertexDecoder VertexDecoderJitCache::Compile(const VertexDecoder &dec, int32_t *jittedSize) {
 	dec_ = &dec;
 
-	BeginWrite();
+	BeginWrite(4096);
 	const u8 *start = AlignCode16();
 
 	bool prescaleStep = false;

GPU/Common/VertexDecoderX86.cpp

@@ -164,7 +164,7 @@ static const JitLookup jitLookup[] = {
 
 JittedVertexDecoder VertexDecoderJitCache::Compile(const VertexDecoder &dec, int32_t *jittedSize) {
 	dec_ = &dec;
-	BeginWrite();
+	BeginWrite(4096);
 	const u8 *start = this->AlignCode16();
 
 #if PPSSPP_ARCH(X86)

GPU/Software/DrawPixelX86.cpp

@@ -42,11 +42,12 @@ SingleFunc PixelJitCache::CompileSingle(const PixelFuncID &id) {
 		RegCache::GEN_ARG_ID,
 	});
 
-	BeginWrite();
+	BeginWrite(64);
 	Describe("Init");
 	WriteConstantPool(id);
 
 	const u8 *resetPos = AlignCode16();
+	EndWrite();
 	bool success = true;
 
 #if PPSSPP_PLATFORM(WINDOWS)

GPU/Software/RasterizerRegCache.cpp

@@ -420,7 +420,7 @@ int CodeBlock::WriteProlog(int extraStack, const std::vector<RegCache::Reg> &vec
 #if PPSSPP_ARCH(X86) || PPSSPP_ARCH(AMD64)
 	using namespace Gen;
 
-	BeginWrite();
+	BeginWrite(32768);
 	AlignCode16();
 	lastPrologStart_ = GetWritableCodePtr();
 

GPU/Software/SamplerX86.cpp

@@ -45,7 +45,7 @@ FetchFunc SamplerJitCache::CompileFetch(const SamplerID &id) {
 	regCache_.ForceRetain(RegCache::GEN_RESULT);
 	regCache_.ChangeReg(XMM0, RegCache::VEC_RESULT);
 
-	BeginWrite();
+	BeginWrite(2048);
 	Describe("Init");
 	const u8 *start = AlignCode16();
 
@@ -122,7 +122,7 @@ FetchFunc SamplerJitCache::CompileFetch(const SamplerID &id) {
 
 NearestFunc SamplerJitCache::CompileNearest(const SamplerID &id) {
 	_assert_msg_(!id.fetch && !id.linear, "Fetch and linear should be cleared on sampler id");
-	BeginWrite();
+	BeginWrite(2048);
 	Describe("Init");
 
 	// Let's drop some helpful constants here.
@@ -438,7 +438,7 @@ NearestFunc SamplerJitCache::CompileNearest(const SamplerID &id) {
 
 LinearFunc SamplerJitCache::CompileLinear(const SamplerID &id) {
 	_assert_msg_(id.linear && !id.fetch, "Only linear should be set on sampler id");
-	BeginWrite();
+	BeginWrite(2048);
 	Describe("Init");
 
 	// We don't use stackArgPos_ here, this is just for DXT.