Cache very common instructions

[athre0z]

It might be worth a try to cache the ~10 most common instructions in the decoder for improved performance. I did a very hacky quick test and determined that caching CC and 00 00 alone results in a ~10% performance boost.

[flobernd]

Concept idea:

Add an extra layer above the ZydisDecoderDecodeBuffer function (care for the double pointer):

ZydisDecoderDecodeBufferCached(ZydisDecoder decoder, ..., 
    const ZydisCacheTable* cache, ZydisDecodedInstruction** instruction)

// The `instruction` parameter could be used as in- and output: A pointer to the cached
// instruction is returned, if it was found in the cache ...
// ... or directly fill the passed struct, if the instruction was not found in the cache.
ZydisDecodedInstruction instruction_data;
ZydisDecodedInstruction* instruction = &instruction_data;
ZydisDecoderDecodeBufferCached(..., &instruction);

// E.g.:
ZyanStatus ZydisDecoderDecodeBufferCached(..., ZydisDecodedInstruction** instruction)
{
    // ...
    if (found_in_cache)
    {
        *instruction = &cache[index];
        return ZYAN_STATUS_TRUE; // indicate, that the instruction was cached
    } else
    {
        const ZyanStatus status = ZydisDecoderDecodeBuffer(..., *instruction);
        ZYAN_CHECK(status);
        return ZYAN_STATUS_FALSE; // indicate, that the instruction was not cached
    }
}

Make sure to add good documentation about the in/out parameter and warn the user to not override data in a cached instruction (could be indicated by the return code).

Add an initializer function:

ZydisDecoderInitializeCache(ZydisDecoder decoder, ZydisCacheTable* cache)

Define the ZydisCacheTable struct:

typedef struct ZydisCacheTable_
{
    ZyanI8 index[2][256]; // <- up to 2-byte opcodes for now; easy to extend
    ZydisDecodedInstruction instructions[n];
} ZydisCacheTable;

[athre0z]

Make sure to add good documentation about the in/out parameter and warn the user to not override data in a cached instruction (could be indicated by the return code).

With a detached caching layer, we can just make the instruction argument const, thus clearly communicating the immutability.