Types and Definitions SilcBool TRUE FALSE SilcUInt8 SilcInt8 SilcUInt16 SilcInt16 SilcUInt32 SilcInt32 SilcUInt64 SilcInt64 SilcFloat32 SilcFloat64 SilcSocket SilcParam SilcCompareValue SilcCompare SILC_MAX SILC_MIN SILC_GET16_MSB SILC_GET32_MSB SILC_GET64_MSB SILC_GET16_LSB SILC_GET32_LSB SILC_PUT16_MSB SILC_PUT32_MSB SILC_PUT64_MSB SILC_PUT16_LSB SILC_PUT32_LSB SILC_SWAB_16 SILC_SWAB_32 SILC_PTR_TO_32 SILC_PTR_TO_64 SILC_32_TO_PTR SILC_64_TO_PTR SILC_ASSERT SILC_VERIFY silc_rol silc_rolc silc_ror silc_rorc silc_rol64 silc_rolc64 silc_ror64 silc_rorc64 silc_offsetof silc_attribute silc_likely silc_unlikely silc_prefetch silc_prefetch_block

silc_prefetch_block

NAME

    static inline void silc_prefetch_block(void *addr,
                                           int prefetch_length,
                                           const int cache_line_length)

DESCRIPTION

Enforced block prefetch. This function loads the specified amount `prefetch_length' of memory from the specified address `addr' to CPU cache with each loaded cache line being the size of `cache_line_length'. If you don't know the cache line size use 64 bytes. Note that, the `cache_line_length' is a const int. In this context this mean its value must not come from a variable but must be a constant (the code won't compile if it comes from a variable).

The `prefetch_length' must be multiple of twice of the `cache_line_length' or 128 if you don't know the cache line size, hence the minimum length for `prefetch_length' is 128 bytes when the `cache_line_length' is 64 bytes. Shorter cache line length (32 bytes) can be used too.

You should use the correct `cache_line_length' value for your CPU or the value of the CPU for which you want to optimize your code. Intel CPUs usually have cache size of 32 or 64 bytes. The most optimal prefetch is achieved if the `cache_line_length' is the actual CPU cache line size. Always do performance testing with and without prefetching to make sure the prefetch actually helps. If used improperly, it may slow down your program.

The difference to silc_prefetch is that this function always performs the prefetch and has the ability to prefetch more than one cache line worth of memory, whereas silc_prefetch can prefetch only one cache line and may not do the prefetch at all.