RPCS3/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2026-10-11 22:49:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

PPU LLVM: Estimate a whole vector per call

Browse Source
This commit is contained in:
izcarti
2026-10-10 22:06:19 +03:00
committed by Elad
parent cb450d7e8f
commit 63ad856afa
2 changed files with 18 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rpcs3/Emu/Cell/PPUThread.cpp
+13 -4
View File
@@ -4888,6 +4888,15 @@ extern void ppu_initialize()
}
}
template <u32 (*Estimate)(u32, bool)>
static void ppu_vec_estimate(v128& v, u32 nj)
{
for (u32 i = 0; i < 4; i++)
{
v._u32[i] = Estimate(v._u32[i], nj != 0);
}
}
bool ppu_initialize(const ppu_module<lv2_obj>& info, bool check_only, u64 file_size)
{
ppu_log.notice("Entering ppu_initialize(const ppu_module&..)");
@@ -4968,10 +4977,10 @@ bool ppu_initialize(const ppu_module<lv2_obj>& info, bool check_only, u64 file_s
{ "__escape", reinterpret_cast<u64>(+ppu_escape) },
{ "__read_maybe_mmio32", reinterpret_cast<u64>(+ppu_read_mmio_aware_u32) },
{ "__write_maybe_mmio32", reinterpret_cast<u64>(+ppu_write_mmio_aware_u32) },
{ "__vrefp", reinterpret_cast<u64>(+[](u32 x, u32 nj) { return ppu_vrefp(x, nj != 0); }) },
{ "__vrsqrtefp", reinterpret_cast<u64>(+[](u32 x, u32 nj) { return ppu_vrsqrtefp(x, nj != 0); }) },
{ "__vexptefp", reinterpret_cast<u64>(+[](u32 x, u32 nj) { return ppu_vexptefp(x, nj != 0); }) },
{ "__vlogefp", reinterpret_cast<u64>(+[](u32 x, u32 nj) { return ppu_vlogefp(x, nj != 0); }) },
{ "__vrefp", reinterpret_cast<u64>(&ppu_vec_estimate<ppu_vrefp>) },
{ "__vrsqrtefp", reinterpret_cast<u64>(&ppu_vec_estimate<ppu_vrsqrtefp>) },
{ "__vexptefp", reinterpret_cast<u64>(&ppu_vec_estimate<ppu_vexptefp>) },
{ "__vlogefp", reinterpret_cast<u64>(&ppu_vec_estimate<ppu_vlogefp>) },
};
for (u64 index = 0; index < 1024; index++)
rpcs3/Emu/Cell/PPUTranslator.cpp
+5 -9
View File
@@ -491,16 +491,12 @@ Value* PPUTranslator::VecHandleResult(Value* val, bool flush_denormals_manually)
Value* PPUTranslator::VecEstimate(StringRef name, u32 vr)
{
const auto b = GetVr(vr, VrType::vi32);
const auto nj = ZExt(RegLoad(m_nj), GetType<u32>());
Value* result = GetUndef<u32[4]>();
auto& entry = m_function->getEntryBlock();
const auto data = IRBuilder<>(&entry, entry.getFirstInsertionPt()).CreateAlloca(GetType<u32[4]>());
for (u32 i = 0; i < 4; i++)
{
result = m_ir->CreateInsertElement(result, Call(GetType<u32>(), m_pure_attr, name, m_ir->CreateExtractElement(b, i), nj), i);
}
return result;
m_ir->CreateStore(GetVr(vr, VrType::vi32), data);
Call(GetType<void>(), name, data, ZExt(RegLoad(m_nj), GetType<u32>()));
return m_ir->CreateLoad(GetType<u32[4]>(), data);
}
Value* PPUTranslator::GetAddr(u64 _add)