JitIL: Improved the speed of JitIL time profiler.

git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@6444 8ced0084-cf51-0410-be5f-012b33b47a6e
This commit is contained in:
nodchip 2010-11-19 15:17:26 +00:00
parent 96cdb1ad07
commit b0c20fe643

View file

@ -156,34 +156,39 @@ ps_adds1
#ifdef _WIN32
// For profiling
// FIXME: This is currently for windows only.
#include <unordered_map>
#include <windows.h>
#include <intrin.h>
namespace JitILProfiler {
struct Performance {
struct Block {
u32 index;
u64 codeHash;
u64 toalElapsed;
u64 numberOfCalls;
Performance() : toalElapsed(0), numberOfCalls(0) { }
Block() : index(0), codeHash(0), toalElapsed(0), numberOfCalls(0) { }
};
static u64 hash;
static std::vector<Block> blocks;
static u32 blockIndex;
static u64 beginTime;
// std::map is a bit slow for this purpose.
static std::tr1::unordered_map<u64, Performance> profiledData;
static Block& Add(u64 codeHash) {
const u32 blockIndex = blocks.size();
blocks.push_back(Block());
Block& block = blocks.back();
block.index = blockIndex;
block.codeHash = codeHash;
return block;
}
// These functions need to be static function
// because they are called with ABI_CallFunction().
static void Begin(u32 higher, u32 lower) {
hash = (((u64)higher) << 32) | lower;
LARGE_INTEGER largeInteger;
QueryPerformanceCounter(&largeInteger);
beginTime = largeInteger.QuadPart;
static void Begin(u32 index) {
blockIndex = index;
beginTime = __rdtsc();
}
static void End() {
LARGE_INTEGER largeInteger;
QueryPerformanceCounter(&largeInteger);
const u64 endTime = largeInteger.QuadPart;
const u64 endTime = __rdtsc();
const u64 duration = endTime - beginTime;
Performance& performance = profiledData[hash];
performance.toalElapsed += duration;
++performance.numberOfCalls;
Block& block = blocks[blockIndex];
block.toalElapsed += duration;
++block.numberOfCalls;
}
struct JitILProfilerFinalizer {
virtual ~JitILProfilerFinalizer() {
@ -192,10 +197,10 @@ namespace JitILProfiler {
FILE* file = fopen(buffer, "w");
setvbuf(file, NULL, _IOFBF, 1024 * 1024);
fprintf(file, "code hash,total elapsed,number of calls,elapsed per call\n");
for (std::tr1::unordered_map<u64, Performance>::iterator it = profiledData.begin(), itEnd = profiledData.end(); it != itEnd; ++it) {
const u64 codeHash = it->first;
const u64 totalElapsed = it->second.toalElapsed;
const u64 numberOfCalls = it->second.numberOfCalls;
for (std::vector<Block>::iterator it = blocks.begin(), itEnd = blocks.end(); it != itEnd; ++it) {
const u64 codeHash = it->codeHash;
const u64 totalElapsed = it->toalElapsed;
const u64 numberOfCalls = it->numberOfCalls;
const double elapsedPerCall = totalElapsed / (double)numberOfCalls;
fprintf(file, "%016llx,%lld,%lld,%f\n", codeHash, totalElapsed, numberOfCalls, elapsedPerCall);
}
@ -214,7 +219,8 @@ namespace JitILProfiler {
#else
namespace JitILProfiler {
// FIXME: Dummy functions for linux. Please implement them.
static void Begin(u32 higher, u32 lower) { }
static Block& Add(u64 codeHash) { }
static void Begin(u32 index) { }
static void End() { }
static void Init() { }
static void Shutdown() { }
@ -565,7 +571,8 @@ const u8* JitIL::DoJit(u32 em_address, PPCAnalyst::CodeBuffer *code_buf, JitBloc
// Ported from boost::hash
codeHash ^= inst + (codeHash << 6) + (codeHash >> 2);
}
ABI_CallFunctionCC(JitILProfiler::Begin, (u32)(codeHash >> 32), (u32)codeHash);
JitILProfiler::Block& block = JitILProfiler::Add(codeHash);
ABI_CallFunctionC(JitILProfiler::Begin, block.index);
}
// Start up IR builder (structure that collects the