archive/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2025-07-04 14:01:25 +12:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

SPU LLVM: Replace xorsum with simple checksum

Browse source 
- Fixes crashes in GTAV
This commit is contained in:
Malcolm Jestadt 2025-02-01 11:24:43 -05:00 • committed by Elad
parent 394fc8eb79
commit 911f0928cf
1 changed files with 7 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

16
rpcs3/Emu/Cell/SPULLVMRecompiler.cpp
View file

@ -1677,9 +1677,9 @@ public:
llvm::Value* acc = nullptr; llvm::Value* acc = nullptr;
// Use 512bit xorsum to verify integrity if size is atleast 512b * 3 // Use a 512bit simple checksum to verify integrity if size is atleast 512b * 3
// This code uses a 512bit vector for all hardware to ensure behavior matches. // This code uses a 512bit vector for all hardware to ensure behavior matches.
// The xorsum path is still faster even on narrow hardware. // The checksum path is still faster even on narrow hardware.
if ((end - starta) >= 192 && !g_cfg.core.precise_spu_verification) if ((end - starta) >= 192 && !g_cfg.core.precise_spu_verification)
{ {
for (u32 j = starta; j < end; j += 64) for (u32 j = starta; j < end; j += 64)
@ -1721,12 +1721,12 @@ public:
vls = m_ir->CreateShuffleVector(vls, ConstantAggregateZero::get(vls->getType()), llvm::ArrayRef(indices, 16)); vls = m_ir->CreateShuffleVector(vls, ConstantAggregateZero::get(vls->getType()), llvm::ArrayRef(indices, 16));
} }
acc = acc ? m_ir->CreateXor(acc, vls) : vls; acc = acc ? m_ir->CreateAdd(acc, vls) : vls;
check_iterations++; check_iterations++;
} }
// Create the Xorsum // Create the checksum
u32 xorsum[16] = {0}; u32 checksum[16] = {0};
for (u32 j = 0; j < func.data.size(); j += 16) // Process 16 elements per iteration for (u32 j = 0; j < func.data.size(); j += 16) // Process 16 elements per iteration
{ {
@ -1734,12 +1734,12 @@ public:
{ {
if (j + i < func.data.size()) if (j + i < func.data.size())
{ {
xorsum[i] ^= func.data[j + i]; checksum[i] += func.data[j + i];
} }
} }
} }
auto* const_vector = ConstantDataVector::get(m_context, llvm::ArrayRef(xorsum, 16)); auto* const_vector = ConstantDataVector::get(m_context, llvm::ArrayRef(checksum, 16));
acc = m_ir->CreateXor(acc, const_vector); acc = m_ir->CreateXor(acc, const_vector);
// Pattern for PTEST // Pattern for PTEST
@ -1752,8 +1752,6 @@ public:
elem = m_ir->CreateOr(elem, m_ir->CreateExtractElement(acc, i)); elem = m_ir->CreateOr(elem, m_ir->CreateExtractElement(acc, i));
} }
spu_log.error("end");
// Compare result with zero // Compare result with zero
const auto cond = m_ir->CreateICmpNE(elem, m_ir->getInt64(0)); const auto cond = m_ir->CreateICmpNE(elem, m_ir->getInt64(0));
m_ir->CreateCondBr(cond, label_diff, label_body, m_md_unlikely); m_ir->CreateCondBr(cond, label_diff, label_body, m_md_unlikely);