minor perf improvements

src/backend/core/registers/Registers.cpp

@@ -35,48 +35,48 @@ void Registers::SetPC32(s32 val) {
 }
 
 template <>
-u64 Registers::Read<u64>(size_t idx) {
+__attribute__((always_inline)) u64 Registers::Read<u64>(size_t idx) {
-  return idx == 0 ? 0 : gpr[idx];
+  return gpr[idx];
 }
 
 template <>
-s64 Registers::Read<s64>(const size_t idx) {
+__attribute__((always_inline)) s64 Registers::Read<s64>(const size_t idx) {
   return static_cast<s64>(Read<u64>(idx));
 }
 
 template <>
-u32 Registers::Read<u32>(size_t idx) {
+__attribute__((always_inline)) u32 Registers::Read<u32>(size_t idx) {
-  return idx == 0 ? 0 : gpr[idx];
+  return gpr[idx];
 }
 
 template <>
-s32 Registers::Read<s32>(size_t idx) {
+__attribute__((always_inline)) s32 Registers::Read<s32>(size_t idx) {
   return static_cast<s32>(Read<u32>(idx));
 }
 
 template <>
-u16 Registers::Read<u16>(size_t idx) {
+__attribute__((always_inline)) u16 Registers::Read<u16>(size_t idx) {
-  return idx == 0 ? 0 : gpr[idx];
+  return gpr[idx];
 }
 
 template <>
-s16 Registers::Read<s16>(size_t idx) {
+__attribute__((always_inline)) s16 Registers::Read<s16>(size_t idx) {
   return static_cast<s16>(Read<u16>(idx));
 }
 
 template <>
-u8 Registers::Read<u8>(size_t idx) {
+__attribute__((always_inline)) u8 Registers::Read<u8>(size_t idx) {
-  return idx == 0 ? 0 : gpr[idx];
+  return gpr[idx];
 }
 
 template <>
-s8 Registers::Read<s8>(size_t idx) {
+__attribute__((always_inline)) s8 Registers::Read<s8>(size_t idx) {
   return static_cast<s8>(Read<u8>(idx));
 }
 
 #ifndef __aarch64__
 template <>
-void Registers::Read<u64>(size_t idx, Xbyak::Reg reg) {
+__attribute__((always_inline)) void Registers::Read<u64>(size_t idx, Xbyak::Reg reg) {
   if(IsRegConstant(idx)) {
     jit->code.mov(reg.cvt64(), Read<u64>(idx));
     return;
@@ -86,7 +86,7 @@ void Registers::Read<u64>(size_t idx, Xbyak::Reg reg) {
 }
 
 template <>
-void Registers::Read<s64>(size_t idx, Xbyak::Reg reg) {
+__attribute__((always_inline)) void Registers::Read<s64>(size_t idx, Xbyak::Reg reg) {
   if(IsRegConstant(idx)) {
     jit->code.mov(reg.cvt64(), Read<s64>(idx));
     return;
@@ -96,7 +96,7 @@ void Registers::Read<s64>(size_t idx, Xbyak::Reg reg) {
 }
 
 template <>
-void Registers::Read<u32>(size_t idx, Xbyak::Reg reg) {
+__attribute__((always_inline)) void Registers::Read<u32>(size_t idx, Xbyak::Reg reg) {
   if(IsRegConstant(idx)) {
     jit->code.mov(reg.cvt32(), Read<u32>(idx));
     return;
@@ -106,7 +106,7 @@ void Registers::Read<u32>(size_t idx, Xbyak::Reg reg) {
 }
 
 template <>
-void Registers::Read<s32>(size_t idx, Xbyak::Reg reg) {
+__attribute__((always_inline)) void Registers::Read<s32>(size_t idx, Xbyak::Reg reg) {
   if(IsRegConstant(idx)) {
     jit->code.mov(reg.cvt32(), Read<s32>(idx));
     return;
@@ -116,7 +116,7 @@ void Registers::Read<s32>(size_t idx, Xbyak::Reg reg) {
 }
 
 template <>
-void Registers::Read<u16>(size_t idx, Xbyak::Reg reg) {
+__attribute__((always_inline)) void Registers::Read<u16>(size_t idx, Xbyak::Reg reg) {
   if(IsRegConstant(idx)) {
     jit->code.mov(reg.cvt16(), Read<u16>(idx));
     return;
@@ -126,7 +126,7 @@ void Registers::Read<u16>(size_t idx, Xbyak::Reg reg) {
 }
 
 template <>
-void Registers::Read<s16>(size_t idx, Xbyak::Reg reg) {
+__attribute__((always_inline)) void Registers::Read<s16>(size_t idx, Xbyak::Reg reg) {
   if(IsRegConstant(idx)) {
     jit->code.mov(reg.cvt16(), Read<s16>(idx));
     return;
@@ -136,7 +136,7 @@ void Registers::Read<s16>(size_t idx, Xbyak::Reg reg) {
 }
 
 template <>
-void Registers::Read<u8>(size_t idx, Xbyak::Reg reg) {
+__attribute__((always_inline)) void Registers::Read<u8>(size_t idx, Xbyak::Reg reg) {
   if(IsRegConstant(idx)) {
     jit->code.mov(reg.cvt8(), Read<u8>(idx));
     return;
@@ -146,7 +146,7 @@ void Registers::Read<u8>(size_t idx, Xbyak::Reg reg) {
 }
 
 template <>
-void Registers::Read<s8>(size_t idx, Xbyak::Reg reg) {
+__attribute__((always_inline)) void Registers::Read<s8>(size_t idx, Xbyak::Reg reg) {
   if(IsRegConstant(idx)) {
     jit->code.mov(reg.cvt8(), Read<s8>(idx));
     return;
@@ -157,199 +157,208 @@ void Registers::Read<s8>(size_t idx, Xbyak::Reg reg) {
 #endif
 
 template <>
-void Registers::Write<bool>(size_t idx, bool v) {
+__attribute__((always_inline)) void Registers::Write<bool>(size_t idx, bool v) {
   if (idx == 0)
     return;
 
+  if (jit) [[unlikely]]
+    regIsConstant |= (1 << idx);
+
+  gpr[idx] = v;
+}
+
+template <>
+__attribute__((always_inline)) void Registers::Write<u64>(size_t idx, u64 v) {
+  if (idx == 0)
+    return;
+
+  if (jit) [[unlikely]]
     regIsConstant |= (1 << idx);
   gpr[idx] = v;
 }
 
 template <>
-void Registers::Write<u64>(size_t idx, u64 v) {
+__attribute__((always_inline)) void Registers::Write<s64>(size_t idx, s64 v) {
-  if (idx == 0)
-    return;
-
-  regIsConstant |= (1 << idx);
-  gpr[idx] = v;
-}
-
-template <>
-void Registers::Write<s64>(size_t idx, s64 v) {
   Write<u64>(idx, v);
 }
 
 template <>
-void Registers::Write<u32>(size_t idx, u32 v) {
+__attribute__((always_inline)) void Registers::Write<u32>(size_t idx, u32 v) {
   if (idx == 0)
     return;
 
+  if (jit) [[unlikely]]
     regIsConstant |= (1 << idx);
   gpr[idx] = v;
 }
 
 
 template <>
-void Registers::Write<s32>(size_t idx, s32 v) {
+__attribute__((always_inline)) void Registers::Write<s32>(size_t idx, s32 v) {
   if (idx == 0)
     return;
 
+  if (jit) [[unlikely]]
     regIsConstant |= (1 << idx);
   gpr[idx] = v;
 }
 
 template <>
-void Registers::Write<u16>(size_t idx, u16 v) {
+__attribute__((always_inline)) void Registers::Write<u16>(size_t idx, u16 v) {
   if (idx == 0)
     return;
 
+  if (jit) [[unlikely]]
     regIsConstant |= (1 << idx);
   gpr[idx] = v;
 }
 
 
 template <>
-void Registers::Write<s16>(size_t idx, s16 v) {
+__attribute__((always_inline)) void Registers::Write<s16>(size_t idx, s16 v) {
   if (idx == 0)
     return;
 
+  if (jit) [[unlikely]]
     regIsConstant |= (1 << idx);
   gpr[idx] = v;
 }
 
 template <>
-void Registers::Write<u8>(size_t idx, u8 v) {
+__attribute__((always_inline)) void Registers::Write<u8>(size_t idx, u8 v) {
   if (idx == 0)
     return;
   
+  if (jit) [[unlikely]]
     regIsConstant |= (1 << idx);
   gpr[idx] = v;
 }
 
 
 template <>
-void Registers::Write<s8>(size_t idx, s8 v) {
+__attribute__((always_inline)) void Registers::Write<s8>(size_t idx, s8 v) {
   if (idx == 0)
     return;
 
+  if (jit) [[unlikely]]
     regIsConstant |= (1 << idx);
   gpr[idx] = v;
 }
 
 #ifndef __aarch64__
 template <>
-void Registers::Write<bool>(size_t idx, Xbyak::Reg v) {
+__attribute__((always_inline)) void Registers::Write<bool>(size_t idx, Xbyak::Reg v) {
   if (idx == 0)
     return;
 
-  regIsConstant &= ~(1 << idx);
-
   if (!jit)
     panic("Did you try to call Registers::Write(size_t, *Xbyak::Reg*) from the interpreter?");
   
+  regIsConstant &= ~(1 << idx);
+
   jit->code.movsx(v.cvt64(), v.cvt8());
   jit->code.mov(jit->GPR<u64>(idx), v);
 }
 
 template <>
-void Registers::Write<s8>(size_t idx, Xbyak::Reg v) {
+__attribute__((always_inline)) void Registers::Write<s8>(size_t idx, Xbyak::Reg v) {
   if (idx == 0)
     return;
 
-  regIsConstant &= ~(1 << idx);
-
   if (!jit)
     panic("Did you try to call Registers::Write(size_t, *Xbyak::Reg*) from the interpreter?");
 
+  regIsConstant &= ~(1 << idx);
+
   jit->code.movsx(v.cvt64(), v.cvt8());
   jit->code.mov(jit->GPR<u64>(idx), v);
 }
 
 template <>
-void Registers::Write<u8>(size_t idx, Xbyak::Reg v) {
+__attribute__((always_inline)) void Registers::Write<u8>(size_t idx, Xbyak::Reg v) {
   if (idx == 0)
     return;
 
-  regIsConstant &= ~(1 << idx);
-
   if (!jit)
     panic("Did you try to call Registers::Write(size_t, *Xbyak::Reg*) from the interpreter?");
 
+  regIsConstant &= ~(1 << idx);
+
   jit->code.movzx(v.cvt64(), v.cvt8());
   jit->code.mov(jit->GPR<u64>(idx), v.cvt64());
 }
 
 template <>
-void Registers::Write<s16>(size_t idx, Xbyak::Reg v) {
+__attribute__((always_inline)) void Registers::Write<s16>(size_t idx, Xbyak::Reg v) {
   if (idx == 0)
     return;
 
-  regIsConstant &= ~(1 << idx);
-
   if (!jit)
     panic("Did you try to call Registers::Write(size_t, *Xbyak::Reg*) from the interpreter?");
 
+  regIsConstant &= ~(1 << idx);
+
   jit->code.movsx(v.cvt64(), v.cvt16());
   jit->code.mov(jit->GPR<u64>(idx), v.cvt64());
 }
 
 template <>
-void Registers::Write<u16>(size_t idx, Xbyak::Reg v) {
+__attribute__((always_inline)) void Registers::Write<u16>(size_t idx, Xbyak::Reg v) {
   if (idx == 0)
     return;
 
-  regIsConstant &= ~(1 << idx);
-
   if (!jit)
     panic("Did you try to call Registers::Write(size_t, *Xbyak::Reg*) from the interpreter?");
 
+  regIsConstant &= ~(1 << idx);
+
   jit->code.movzx(v.cvt64(), v.cvt16());
   jit->code.mov(jit->GPR<u64>(idx), v.cvt64());
 }
 
 template <>
-void Registers::Write<s32>(size_t idx, Xbyak::Reg v) {
+__attribute__((always_inline)) void Registers::Write<s32>(size_t idx, Xbyak::Reg v) {
   if (idx == 0)
     return;
 
-  regIsConstant &= ~(1 << idx);
-
   if (!jit)
     panic("Did you try to call Registers::Write(size_t, *Xbyak::Reg*) from the interpreter?");
 
+  regIsConstant &= ~(1 << idx);
+
   jit->code.movsxd(v.cvt64(), v.cvt32());
   jit->code.mov(jit->GPR<u64>(idx), v.cvt64());
 }
 
 template <>
-void Registers::Write<u32>(size_t idx, Xbyak::Reg v) {
+__attribute__((always_inline)) void Registers::Write<u32>(size_t idx, Xbyak::Reg v) {
   if (idx == 0)
     return;
 
-  regIsConstant &= ~(1 << idx);
-
   if (!jit)
     panic("Did you try to call Registers::Write(size_t, *Xbyak::Reg*) from the interpreter?");
 
+  regIsConstant &= ~(1 << idx);
+
   jit->code.movzx(v.cvt64(), v.cvt32());
   jit->code.mov(jit->GPR<u64>(idx), v.cvt64());
 }
 
 template <>
-void Registers::Write<u64>(size_t idx, Xbyak::Reg v) {
+__attribute__((always_inline)) void Registers::Write<u64>(size_t idx, Xbyak::Reg v) {
   if (idx == 0)
     return;
 
-  regIsConstant &= ~(1 << idx);
-
   if (!jit)
     panic("Did you try to call Registers::Write(size_t, *Xbyak::Reg*) from the interpreter?");
 
+  regIsConstant &= ~(1 << idx);
+
   jit->code.mov(jit->GPR<u64>(idx), v.cvt64());
 }
 
 template <>
-void Registers::Write<s64>(size_t idx, Xbyak::Reg v) {
+__attribute__((always_inline)) void Registers::Write<s64>(size_t idx, Xbyak::Reg v) {
   Write<u64>(idx, v);
 }
 #endif

src/backend/core/registers/Registers.hpp

@@ -66,13 +66,13 @@ struct Registers {
   }
 
   template <typename T>
-  T Read(size_t);
+  __attribute__((always_inline)) T Read(size_t);
   template <typename T>
-  void Read(size_t, Xbyak::Reg);
+  __attribute__((always_inline)) void Read(size_t, Xbyak::Reg);
   template <typename T>
-  void Write(size_t, T);
+  __attribute__((always_inline)) void Write(size_t, T);
   template <typename T>
-  void Write(size_t, Xbyak::Reg);
+  __attribute__((always_inline)) void Write(size_t, Xbyak::Reg);
 
   std::array<s64, 32> gpr{};
 };

src/utils/log.hpp

@@ -28,33 +28,11 @@ void print(const std::format_string<Args...> fmt, Args... args) {
   }
 }
 
-#define panic(fmt, ...) do \
+#define panic(fmt, ...) do { Util::print<Util::Error>("[FATAL] " fmt, __VA_ARGS__); exit(-1); } while(0)
-  { \
+#define error(fmt, ...) do { Util::print<Util::Error>("[ERROR] " fmt, __VA_ARGS__); } while(0)
-    Util::print<Util::Error>("[FATAL] " fmt, __VA_ARGS__); \
+#define warn(fmt, ...) do { Util::print<Util::Warn>("[WARN] " fmt, __VA_ARGS__); } while(0)
-    exit(-1); \
+#define info(fmt, ...) do { Util::print<Util::Info>("[INFO] " fmt, __VA_ARGS__); } while(0)
-  } while(0)
+#define debug(fmt, ...) do { Util::print<Util::Debug>("[DEBUG] " fmt, __VA_ARGS__); } while(0)
-#define error(fmt, ...) do \
+#define trace(fmt, ...) do { Util::print<Util::Trace>("[TRACE] " fmt, __VA_ARGS__); } while(0)
-  { \
+#define always(fmt, ...) do { Util::print<Util::Always>(fmt, __VA_ARGS__); } while(0)
-    Util::print<Util::Error>("[ERROR] " fmt, __VA_ARGS__); \
-  } while(0)
-#define warn(fmt, ...) do \
-  { \
-    Util::print<Util::Warn>("[WARN] " fmt, __VA_ARGS__); \
-  } while(0)
-#define info(fmt, ...) do \
-  { \
-    Util::print<Util::Info>("[INFO] " fmt, __VA_ARGS__); \
-  } while(0)
-#define debug(fmt, ...) do \
-  { \
-    Util::print<Util::Debug>("[DEBUG] " fmt, __VA_ARGS__); \
-  } while(0)
-#define trace(fmt, ...) do \
-  { \
-    Util::print<Util::Trace>("[TRACE] " fmt, __VA_ARGS__); \
-  } while(0)
-#define always(fmt, ...) do \
-  { \
-    Util::print<Util::Always>(fmt, __VA_ARGS__); \
-  } while(0)
 } // namespace Util