A64/x64: Create a global_offset optimization for the page table.

Instead of looking up the page table like:
  table[addr >> 12][addr & 0xFFF]
We can use a global offset on the table to query the memory like:
  table[addr >> 12][addr]

This saves two instructions on *every* memory access within the recompiler.

Thanks at skmp for the idea.
This commit is contained in:
Markus Wick 2019-12-31 00:08:35 +01:00 committed by MerryMage
parent 7be0038186
commit 93668c24be
2 changed files with 10 additions and 0 deletions

View file

@ -163,6 +163,13 @@ struct UserConfig {
/// relevant memory callback.
/// This is only used if page_table is not nullptr.
bool silently_mirror_page_table = true;
/// Determines if the pointer in the page_table shall be offseted locally or globally.
/// 'false' will access page_table[addr >> bits][addr & mask]
/// 'true' will access page_table[addr >> bits][addr]
/// Note: page_table[addr >> bits] will still be checked to verify active pages.
/// So there might be wrongly faulted pages which maps to nullptr.
/// This can be avoided by carefully allocating the memory region.
bool absolute_offset_page_table = false;
/// This option relates to translation. Generally when we run into an unpredictable
/// instruction the ExceptionRaised callback is called. If this is true, we define

View file

@ -737,6 +737,9 @@ static Xbyak::RegExp EmitVAddrLookup(BlockOfCode& code, A64EmitContext& ctx, Xby
code.mov(page_table, qword[page_table + tmp * sizeof(void*)]);
code.test(page_table, page_table);
code.jz(abort, code.T_NEAR);
if (ctx.conf.absolute_offset_page_table) {
return page_table + vaddr;
}
code.mov(tmp, vaddr);
code.and_(tmp, static_cast<u32>(page_size - 1));
return page_table + tmp;