better handling of static translation cache allocation, handle nested

execution paths from the cpu core, cleanups for KPX_MAX_CPUS == 1.

SheepShaver/src/kpx_cpu/sheepshaver_glue.cpp

@@ -118,9 +118,6 @@ public:
 	uint32 get_cr() const		{ return cr().get(); }
 	void set_cr(uint32 v)		{ cr().set(v); }
 
-	// Execution loop
-	void execute(uint32 entry, bool enable_cache = false);
-
 	// Execute 68k routine
 	void execute_68k(uint32 entry, M68kRegisters *r);
 
@@ -244,12 +241,6 @@ void sheepshaver_cpu::execute_sheep(uint32 opcode)
 	}
 }
 
-// Execution loop
-void sheepshaver_cpu::execute(uint32 entry, bool enable_cache)
-{
-	powerpc_cpu::execute(entry, enable_cache);
-}
-
 // Handle MacOS interrupt
 void sheepshaver_cpu::interrupt(uint32 entry)
 {
@@ -686,7 +677,7 @@ void emul_ppc(uint32 entry)
 	current_cpu->start_log();
 #endif
 	// start emulation loop and enable code translation or caching
-	current_cpu->execute(entry, true);
+	current_cpu->execute(entry);
 }
 
 /*

SheepShaver/src/kpx_cpu/src/cpu/jit/basic-dyngen.cpp

@@ -28,8 +28,8 @@ int __op_jmp0, __op_jmp1;
 #define DEFINE_GEN(NAME,ARGS) void basic_dyngen::NAME ARGS
 #include "basic-dyngen-ops.hpp"
 
-basic_dyngen::basic_dyngen(dyngen_cpu_base cpu)
-	: parent_cpu(cpu)
+basic_dyngen::basic_dyngen(dyngen_cpu_base cpu, int cache_size)
+	: parent_cpu(cpu), basic_jit_cache(cache_size)
 {
 	execute_func = gen_start();
 	gen_op_execute();

SheepShaver/src/kpx_cpu/src/cpu/jit/basic-dyngen.hpp

@@ -52,7 +52,7 @@ class basic_dyngen
 public:
 
 	// Constructor, parent CPU required
-	basic_dyngen(dyngen_cpu_base cpu);
+	basic_dyngen(dyngen_cpu_base cpu, int cache_size = -1);
 
 	// Return CPU context associated to this code generator
 	dyngen_cpu_base cpu() const

SheepShaver/src/kpx_cpu/src/cpu/jit/jit-cache.cpp

@@ -30,10 +30,13 @@
 #endif
 
 #if STATIC_ICACHE_ALLOC
-static uint8 g_translation_cache[basic_jit_cache::JIT_CACHE_SIZE];
+const int G_TRANSLATION_CACHE_SIZE = 3 * 1024 * 1024; // 3 MB
+static uint8 g_translation_cache[G_TRANSLATION_CACHE_SIZE];
+static uint8 *g_translation_cache_p;
+static uint8 *g_translation_cache_end_p;
 #endif
 
-basic_jit_cache::basic_jit_cache(uint32 init_cache_size)
+basic_jit_cache::basic_jit_cache(int init_cache_size)
 	: tcode_start(NULL), code_start(NULL), code_p(NULL), code_end(NULL)
 {
 	init_translation_cache(init_cache_size);
@@ -45,16 +48,27 @@ basic_jit_cache::~basic_jit_cache()
 }
 
 bool
-basic_jit_cache::init_translation_cache(uint32 size)
+basic_jit_cache::init_translation_cache(int size)
 {
-	// Round up translation cache size to next 16 KB boundaries
-	const uint32 roundup = 16 * 1024;
-	uint32 effective_cache_size = (size - JIT_CACHE_SIZE_GUARD) & -roundup;
-	cache_size = size & -roundup;
+#if STATIC_ICACHE_ALLOC
+	if (g_translation_cache_p == 0) {
+		g_translation_cache_p = g_translation_cache;
+		g_translation_cache_end_p = g_translation_cache_p + G_TRANSLATION_CACHE_SIZE;
+	}
+#endif
+
+	if (size == -1)
+		size = JIT_CACHE_SIZE;
+
+	// Round up translation cache size to next guard size boundaries boundaries
+	const uint32 roundup = JIT_CACHE_SIZE_GUARD;
+	cache_size = (size + roundup - 1) & -roundup;
+	assert(cache_size > 0);
 
 #if STATIC_ICACHE_ALLOC
-	if (cache_size <= JIT_CACHE_SIZE) {
-		tcode_start = g_translation_cache;
+	if (cache_size <= (g_translation_cache_end_p - g_translation_cache_p)) {
+		tcode_start = g_translation_cache_p;
+		g_translation_cache_p += cache_size;
 		goto done;
 	}
 #endif
@@ -75,7 +89,7 @@ basic_jit_cache::init_translation_cache(uint32 size)
 	D(bug("basic_jit_cache: Translation cache: %d KB at %p\n", cache_size / 1024, tcode_start));
 	code_start = tcode_start;
 	code_p = code_start;
-	code_end = code_p + effective_cache_size;
+	code_end = code_p + size;
 	return true;
 }
 
@@ -84,8 +98,16 @@ basic_jit_cache::kill_translation_cache()
 {
 	if (tcode_start) {
 #if STATIC_ICACHE_ALLOC
-		if (cache_size > JIT_CACHE_SIZE)
+		if ((tcode_start - g_translation_cache) <= G_TRANSLATION_CACHE_SIZE) {
+			if (tcode_start == g_translation_cache_p - cache_size) {
+				D(bug("basic_jit_cache: Merge back free translation cache: %d KB at %p\n",
+					  cache_size / 1024, tcode_start));
+				g_translation_cache_p -= cache_size;
+			}
+			return;
+		}
 #endif
+		D(bug("basic_jit_cache: Release translation cache\n"));
 		vm_release(tcode_start, cache_size);
 	}
 }

SheepShaver/src/kpx_cpu/src/cpu/jit/jit-cache.hpp

@@ -28,10 +28,8 @@
 class basic_jit_cache
 {
 	// Default cache size (2 MB)
-public:
 	static const uint32 JIT_CACHE_SIZE = 2 * 1024 * 1024;
 	static const uint32 JIT_CACHE_SIZE_GUARD = 4096;
-private:
 	uint32 cache_size;
 
 	// Translation cache (allocated base, current pointer, end pointer)
@@ -43,7 +41,7 @@ private:
 protected:
 
 	// Initialize translation cache
-	bool init_translation_cache(uint32 size);
+	bool init_translation_cache(int size);
 	void kill_translation_cache();
 
 	// Initialize user code start
@@ -56,8 +54,8 @@ public:
 
 public:
 
-	// Default constructor & destructor
-	basic_jit_cache(uint32 init_cache_size = JIT_CACHE_SIZE);
+	// Default constructor & destructor (use default JIT_CACHE_SIZE)
+	basic_jit_cache(int init_cache_size = -1);
 	~basic_jit_cache();
 
 	// Invalidate translation cache

SheepShaver/src/kpx_cpu/src/cpu/ppc/ppc-config.hpp

@@ -37,20 +37,6 @@
 #endif
 
 
-/**
- *	PPC_NO_BASIC_CPU_BASE
- *
- *		Define to not inherit from basic_cpu, thus removing two
- *		vtables. Otherwise, access to registers require an extra
- *		offset from "this" because vtables are stored before other
- *		regular members.
- **/
-
-#ifndef PPC_NO_BASIC_CPU_BASE
-#undef  PPC_NO_BASIC_CPU_BASE
-#endif
-
-
 /**
  *	PPC_NO_STATIC_II_INDEX_TABLE
  *
@@ -144,8 +130,7 @@
  *		Sanity checks and features enforcements
  **/
 
-#ifdef SHEEPSHAVER
-#define PPC_NO_BASIC_CPU_BASE
+#if KPX_MAX_CPUS == 1
 #undef PPC_NO_STATIC_II_INDEX_TABLE
 #endif
 

SheepShaver/src/kpx_cpu/src/cpu/ppc/ppc-cpu.cpp

@@ -225,6 +225,7 @@ void powerpc_cpu::initialize()
 	init_decoder();
 	init_registers();
 	init_decode_cache();
+	execute_depth = 0;
 
 	// Init cache range invalidate recorder
 	cache_range.start = cache_range.end = 0;
@@ -365,14 +366,16 @@ bool powerpc_cpu::check_spcflags()
 	return true;
 }
 
-void powerpc_cpu::execute(uint32 entry, bool enable_cache)
+
+void powerpc_cpu::execute(uint32 entry)
 {
 	pc() = entry;
 #if PPC_EXECUTE_DUMP_STATE
 	const bool dump_state = true;
 #endif
+	execute_depth++;
 #if PPC_ENABLE_JIT
-	if (enable_cache) {
+	if (execute_depth == 1) {
 		for (;;) {
 			block_info *bi = compile_block(pc());
 
@@ -382,7 +385,7 @@ void powerpc_cpu::execute(uint32 entry, bool enable_cache)
 
 				if (!spcflags().empty()) {
 					if (!check_spcflags())
-						return;
+						goto return_site;
 
 					// Force redecoding if cache was invalidated
 					if (spcflags().test(SPCFLAG_JIT_EXEC_RETURN)) {
@@ -395,11 +398,11 @@ void powerpc_cpu::execute(uint32 entry, bool enable_cache)
 					break;
 			}
 		}
-		return;
+		goto return_site;
 	}
 #endif
 #if PPC_DECODE_CACHE
-	if (enable_cache) {
+	if (execute_depth == 1) {
 		for (;;) {
 #if PPC_PROFILE_COMPILE_TIME
 			compile_count++;
@@ -474,7 +477,7 @@ void powerpc_cpu::execute(uint32 entry, bool enable_cache)
 
 				if (!spcflags().empty()) {
 					if (!check_spcflags())
-						return;
+						goto return_site;
 
 					// Force redecoding if cache was invalidated
 					if (spcflags().test(SPCFLAG_JIT_EXEC_RETURN)) {
@@ -487,7 +490,7 @@ void powerpc_cpu::execute(uint32 entry, bool enable_cache)
 					break;
 			}
 		}
-		return;
+		goto return_site;
 	}
 #endif
 	for (;;) {
@@ -508,8 +511,10 @@ void powerpc_cpu::execute(uint32 entry, bool enable_cache)
 			dump_registers();
 #endif
 		if (!spcflags().empty() && !check_spcflags())
-			return;
+			goto return_site;
 	}
+  return_site:
+	--execute_depth;
 }
 
 void powerpc_cpu::execute()

SheepShaver/src/kpx_cpu/src/cpu/ppc/ppc-cpu.hpp

@@ -33,7 +33,7 @@
 #include <vector>
 
 class powerpc_cpu
-#ifndef PPC_NO_BASIC_CPU_BASE
+#ifndef SHEEPSHAVER
 	: public basic_cpu
 #endif
 {
@@ -200,10 +200,13 @@ private:
 	// Check special CPU flags
 	bool check_spcflags();
 
+	// Current execute() nested level
+	int execute_depth;
+
 public:
 
 	// Initialization & finalization
-#ifdef PPC_NO_BASIC_CPU_BASE
+#ifdef SHEEPSHAVER
 	powerpc_cpu()
 #if PPC_ENABLE_JIT
 		: codegen(this)
@@ -238,7 +241,7 @@ public:
 	void fake_dump_registers(uint32);
 
 	// Start emulation loop
-	void execute(uint32 entry, bool enable_cache = true);
+	void execute(uint32 entry);
 	void execute();
 
 	// Interrupts handling
@@ -311,7 +314,6 @@ private:
 	friend class powerpc_dyngen;
 	powerpc_dyngen codegen;
 	block_info *compile_block(uint32 entry);
-	powerpc_dyngen *codegen_ptr() { return &codegen; }
 #endif
 
 	// Semantic action templates

SheepShaver/src/kpx_cpu/src/cpu/ppc/ppc-dyngen.hpp

@@ -73,8 +73,8 @@ public:
 	friend class powerpc_dyngen_helper;
 
 	// Default constructor
-	powerpc_dyngen(dyngen_cpu_base cpu)
-		: basic_dyngen(cpu)
+	powerpc_dyngen(dyngen_cpu_base cpu, int cache_size = -1)
+		: basic_dyngen(cpu, cache_size)
 		{ }
 
 	// Load/store registers