Useful inline functions for assembler-optimised common operations

---

Detailed Description

There are little operations which one needs from time to time which can be relatively expensive or impossible to do in C. For these situations, TnFOX provides a series of small inline functions capable of providing them written in assembler on x86 (i486 or later only) and x64 but with fallback generic implementations in C. For x86 and x64 architectures, the relevent intrinsics for SSE and SSE2 are defined if the compiler is not compiling for a non-supporting processor.

Using these functions can seriously improve the speed of your code. You should also check out FX::Maths for SIMD optimised maths.

Functions
void	FX::fxprefetchmemT (const void *ptr) throw ()
void	FX::fxprefetchmemNT (const void *ptr) throw ()
FXuint	FX::fxbitscan (FXuint x) throw ()
FXuint	FX::fxbitscanrev (FXuint x) throw ()
void	FX::fxendianswap (FXushort &v) throw ()
void	FX::fxendianswap (FXuint &v) throw ()
void	FX::fxendianswap (FXulong &v) throw ()

---

Function Documentation

void FX::fxprefetchmemT

(

const void *

ptr

)

throw () [inline]

Pretches a cache line into the processor cache temporally (ie; it will be used multiple times)

void FX::fxprefetchmemNT

(

const void *

ptr

)

throw () [inline]

Pretches a cache line into the processor cache non-temporally (ie; it will be used only once)

FXuint FX::fxbitscan

(

FXuint

x

)

throw () [inline]

Forward scans an unsigned integer, returning the index of the first set bit. Compiles into 21 x86 cycles with no branching, though on x86 and x64 it directly uses the bsl instruction

References FX::fxbitscan().

Referenced by FX::TnFXSQLDBColumn::effectiveType(), and FX::fxbitscan().

FXuint FX::fxbitscanrev

(

FXuint

x

)

throw () [inline]

Backward scans an unsigned integer, returning the index of the first set bit. Compiles into roughly 24 x86 cycles with no branching, though on x86 and x64 it directly uses the bsr instruction. You should note that this implementation uses illegal C++ which may fail with aggressive enough optimisation - in this situation, enable the alternative 36 x86 cycle implementation in the source code.

References FX::fxbitscanrev().

Referenced by FX::fxbitscanrev().

void FX::fxendianswap

(

FXushort &

v

)

throw () [inline]

Endian swaps the two bytes at v

void FX::fxendianswap

(

FXuint &

v

)

throw () [inline]

Endian swaps the four bytes at v

void FX::fxendianswap

(

FXulong &

v

)

throw () [inline]

Endian swaps the eight bytes at v