Memory API

This API provides special instruction intrinsics, cache operations and memcpy()-type routines. shz_memcpy() is a generic drop-in memcpy() replacement; when you know and control the alignment and batch sizes, you can micro-optimize by calling the most specific specialization instead. sq_ routines write through the SH4 store queues, and _xmtrx routines use the XMTRX register bank as a scratch buffer, clobbering it. memcpy()-like routines will typically always check for proper alignment and size increments of parameters using assert(), so make sure to build a release build (-DNDEBUG) for maximal gainz when not debugging. There is no shz_memset() yet; it is listed as a TODO in the header.

• C header: #include <sh4zam/shz_mem.h>
• C++ header: #include <sh4zam/shz_mem.hpp>
• Symbols: 23

Suffixes

Suffixes used in this module. The full list is in Naming and Suffixes.

• sq_: Writes through the SH4 store queues (burst writes to memory or the PVR).
• _xmtrx: Uses the XMTRX register bank as a 64-byte scratch buffer. Clobbers XMTRX.
• memcpyN: A numeric suffix on memcpy/memset names the element granularity or count, e.g. shz_memcpy32_1 copies exactly one 32-byte line.

C stdlib Replacements

Routine replacing the C standard library copy/set API.

Symbol	Description
shz_memcpy	Generic drop-in fast memcpy() replacement.
shz_memmove	Generic drop-in fast memmove() replacement.

Specializations

Specialized routines for specific sizes + alignments.

Symbol	Description
shz_memcpy1	Copies an unaligned buffer to another one byte at a time.
shz_memcpy2	Copies from one 2-byte aligned buffer to another two bytes at a time.
shz_memcpy4	Copies a from one 4-byte aligned buffer to another 4 bytes at a time.
shz_memcpy8	Copies a from one 8-byte aligned buffer to another 8 bytes at a time.
shz_memset8	Assigns the given 8-byte value to the bytes in dst.
shz_memcpy32	Copies bytes from the src to the dst buffer in 32-byte chunks.
shz_sq_memcpy32	Copies bytes from src to dst in 32-byte chunks, using the Store Queues.
shz_sq_memcpy32_xmtrx	Copies bytes from src to dst in 32-byte chunks, using the Store Queues and XMTRX.
shz_memcpy64	Specialized memcpy() variant for copying multiples of 64-bytes.
shz_memcpy128	Specialized memcpy() variant for copying multiples of 128 bytes.

Constant-sized Operations

Specialized routines for operating on statically sized buffers.

Symbol	Description
shz_memcpy2_8	Copies 8 shorts from src to dst.
shz_memcpy2_16	Copies 16 shorts from src to dst.
shz_memset2_16	Sets the values of the 16 shorts pointed to by dst to the given value.
shz_memcpy4_16	Copies 16 4-byte, long values from src to dst.
shz_memcpy32_1	Copies 32 bytes from p1 to p2 as a single chunk.
shz_memswap32_1	Swaps the values within the given 32-byte buffers.
shz_memswap32_1_xmtrx	Swaps the values within the given 32-byte buffers, using XMTRX.
shz_sq_memcpy32_1	Copies src to dst in a single 32-byte transaction using the Store Queues.
shz_sq_memcpy32_1_xmtrx	Copies src to dst in a single 32-byte transaction using the Store Queues and XMTRX.
shz_dcache_alloc_line	Intrinsic around the SH4 MOVCA.L instruction.

Macros

Symbol	Description
SHZ_PREFETCH	Used to prepopulate the cache line mapping to the given address, before it is used.