JitCache: use a custom bit-set class.

Unfortunately, this appears to be necessary for the sake of performance;
the standard library equivalents don't work well enough on Visual Studio
2013. vector<bool>::insert() is way too slow in debug
builds to be usable, and std::bitset generates inefficient code in release
builds.
This commit is contained in:
magumagu 2014-05-24 11:18:32 -07:00
parent 6bb08de8a0
commit 282e9bd292
2 changed files with 41 additions and 7 deletions

View File

@ -109,8 +109,7 @@ using namespace Gen;
links_to.clear(); links_to.clear();
block_map.clear(); block_map.clear();
valid_block.clear(); valid_block.ClearAll();
valid_block.resize(VALID_BLOCK_MASK_SIZE, false);
num_blocks = 0; num_blocks = 0;
memset(blockCodePointers, 0, sizeof(u8*)*MAX_NUM_BLOCKS); memset(blockCodePointers, 0, sizeof(u8*)*MAX_NUM_BLOCKS);
@ -165,7 +164,7 @@ using namespace Gen;
u32 pAddr = b.originalAddress & 0x1FFFFFFF; u32 pAddr = b.originalAddress & 0x1FFFFFFF;
for (u32 i = 0; i < (b.originalSize + 7) / 8; ++i) for (u32 i = 0; i < (b.originalSize + 7) / 8; ++i)
valid_block[pAddr / 32 + i] = true; valid_block.Set(pAddr / 32 + i);
block_map[std::make_pair(pAddr + 4 * b.originalSize - 1, pAddr)] = block_num; block_map[std::make_pair(pAddr + 4 * b.originalSize - 1, pAddr)] = block_num;
if (block_link) if (block_link)
@ -335,10 +334,10 @@ using namespace Gen;
bool destroy_block = true; bool destroy_block = true;
if (length == 32) if (length == 32)
{ {
if (!valid_block[pAddr / 32]) if (!valid_block.Test(pAddr / 32))
destroy_block = false; destroy_block = false;
else else
valid_block[pAddr / 32] = false; valid_block.Clear(pAddr / 32);
} }
// destroy JIT blocks // destroy JIT blocks

View File

@ -6,6 +6,7 @@
#include <bitset> #include <bitset>
#include <map> #include <map>
#include <memory>
#include <vector> #include <vector>
#include "Core/PowerPC/Gekko.h" #include "Core/PowerPC/Gekko.h"
@ -64,6 +65,40 @@ struct JitBlock
typedef void (*CompiledCode)(); typedef void (*CompiledCode)();
// This is essentially just an std::bitset, but Visual Studia 2013's
// implementation of std::bitset is slow.
class ValidBlockBitSet final
{
enum
{
VALID_BLOCK_MASK_SIZE = 0x20000000 / 32,
VALID_BLOCK_ALLOC_ELEMENTS = VALID_BLOCK_MASK_SIZE / 32
};
std::unique_ptr<u32[]> m_valid_block;
public:
ValidBlockBitSet()
{
m_valid_block.reset(new u32[VALID_BLOCK_ALLOC_ELEMENTS]);
ClearAll();
}
void Set(u32 bit)
{
m_valid_block[bit / 32] |= 1u << (bit % 32);
}
void Clear(u32 bit)
{
m_valid_block[bit / 32] &= ~(1u << (bit % 32));
}
void ClearAll()
{
memset(m_valid_block.get(), 0, sizeof(u32) * VALID_BLOCK_ALLOC_ELEMENTS);
}
bool Test(u32 bit)
{
return (m_valid_block[bit / 32] & (1u << (bit % 32))) != 0;
}
};
class JitBaseBlockCache class JitBaseBlockCache
{ {
@ -72,11 +107,11 @@ class JitBaseBlockCache
int num_blocks; int num_blocks;
std::multimap<u32, int> links_to; std::multimap<u32, int> links_to;
std::map<std::pair<u32,u32>, u32> block_map; // (end_addr, start_addr) -> number std::map<std::pair<u32,u32>, u32> block_map; // (end_addr, start_addr) -> number
std::vector<bool> valid_block; ValidBlockBitSet valid_block;
enum enum
{ {
MAX_NUM_BLOCKS = 65536*2, MAX_NUM_BLOCKS = 65536*2,
VALID_BLOCK_MASK_SIZE = 0x20000000 / 32,
}; };
bool RangeIntersect(int s1, int e1, int s2, int e2) const; bool RangeIntersect(int s1, int e1, int s2, int e2) const;