mirror/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2025-02-05 06:39:52 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update to LLVM 16.0.1

Browse Source 
Fix Zen4+ AVX-512 detection
This commit is contained in:
Ivan Chikish 2023-04-08 15:21:22 +03:00 committed by Ivan
parent 6fbca1acfd
commit 06b0e35fb9
11 changed files with 87 additions and 54 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
.ci/build-freebsd.sh
View File

@ -7,7 +7,7 @@ git submodule -q update --init --depth 1 $(awk '/path/ && !/llvm/ && !/SPIRV/ {
# Prefer newer Clang than in base system (see also .ci/install-freebsd.sh) # Prefer newer Clang than in base system (see also .ci/install-freebsd.sh)
# libc++ isn't in llvm* packages, so download manually # libc++ isn't in llvm* packages, so download manually
fetch https://github.com/llvm/llvm-project/releases/download/llvmorg-16.0.0/llvm-project-16.0.0.src.tar.xz fetch https://github.com/llvm/llvm-project/releases/download/llvmorg-16.0.1/llvm-project-16.0.1.src.tar.xz
tar xf llvm*.tar.xz tar xf llvm*.tar.xz
export CC=clang16 CXX=clang++16 export CC=clang16 CXX=clang++16
cmake -B libcxx_build -G Ninja -S llvm*/libcxx \ cmake -B libcxx_build -G Ninja -S llvm*/libcxx \

2
.ci/get_keys-windows.sh
View File

@ -1,4 +1,4 @@
#!/bin/sh -ex #!/bin/sh -ex
curl -L -o "./llvm.lock" "https://github.com/RPCS3/llvm-mirror/releases/download/custom-build-win-16.0.0/llvmlibs_mt.7z.sha256" curl -L -o "./llvm.lock" "https://github.com/RPCS3/llvm-mirror/releases/download/custom-build-win-16.0.1/llvmlibs_mt.7z.sha256"
curl -L -o "./glslang.lock" "https://github.com/RPCS3/glslang/releases/download/custom-build-win/glslanglibs_mt.7z.sha256" curl -L -o "./glslang.lock" "https://github.com/RPCS3/glslang/releases/download/custom-build-win/glslanglibs_mt.7z.sha256"

2
.ci/setup-windows.sh
View File

@ -19,7 +19,7 @@ QT_DECL_URL="${QT_HOST}${QT_PREFIX}qtdeclarative${QT_SUFFIX}"
QT_TOOL_URL="${QT_HOST}${QT_PREFIX}qttools${QT_SUFFIX}" QT_TOOL_URL="${QT_HOST}${QT_PREFIX}qttools${QT_SUFFIX}"
QT_MM_URL="${QT_HOST}${QT_PREFIX}qtmultimedia${QT_SUFFIX}" QT_MM_URL="${QT_HOST}${QT_PREFIX}qtmultimedia${QT_SUFFIX}"
QT_SVG_URL="${QT_HOST}${QT_PREFIX}qtsvg${QT_SUFFIX}" QT_SVG_URL="${QT_HOST}${QT_PREFIX}qtsvg${QT_SUFFIX}"
LLVMLIBS_URL='https://github.com/RPCS3/llvm-mirror/releases/download/custom-build-win-16.0.0/llvmlibs_mt.7z' LLVMLIBS_URL='https://github.com/RPCS3/llvm-mirror/releases/download/custom-build-win-16.0.1/llvmlibs_mt.7z'
GLSLANG_URL='https://github.com/RPCS3/glslang/releases/download/custom-build-win/glslanglibs_mt.7z' GLSLANG_URL='https://github.com/RPCS3/glslang/releases/download/custom-build-win/glslanglibs_mt.7z'
VULKAN_SDK_URL="https://www.dropbox.com/s/cs77c3iv5mbo0bt/VulkanSDK-${VULKAN_VER}-Installer.exe" VULKAN_SDK_URL="https://www.dropbox.com/s/cs77c3iv5mbo0bt/VulkanSDK-${VULKAN_VER}-Installer.exe"

2
3rdparty/llvm/llvm vendored

@ -1 +1 @@
Subproject commit 08d094a0e457360ad8b94b017d2dc277e697ca76 Subproject commit cd89023f797900e4492da58b7bed36f702120011

2
BUILDING.md
View File

@ -111,7 +111,7 @@ git submodule update --init
Open `rpcs3.sln`. The recommended build configuration is `Release`. (On older revisions: `Release - LLVM`) Open `rpcs3.sln`. The recommended build configuration is `Release`. (On older revisions: `Release - LLVM`)
You may want to download the precompiled [LLVM libs](https://github.com/RPCS3/llvm-mirror/releases/download/custom-build-win-16.0.0/llvmlibs_mt.7z) and extract them to `3rdparty\llvm\`, as well as download and extract the [additional libs](https://github.com/RPCS3/glslang/releases/download/custom-build-win/glslanglibs_mt.7z) to `lib\%CONFIGURATION%-x64\` to speed up compilation time (unoptimised/debug libs are currently not available precompiled). You may want to download the precompiled [LLVM libs](https://github.com/RPCS3/llvm-mirror/releases/download/custom-build-win-16.0.1/llvmlibs_mt.7z) and extract them to `3rdparty\llvm\`, as well as download and extract the [additional libs](https://github.com/RPCS3/glslang/releases/download/custom-build-win/glslanglibs_mt.7z) to `lib\%CONFIGURATION%-x64\` to speed up compilation time (unoptimised/debug libs are currently not available precompiled).
If you're not using the precompiled libs, build the following projects in *__BUILD_BEFORE* folder by right-clicking on a project > *Build*.: If you're not using the precompiled libs, build the following projects in *__BUILD_BEFORE* folder by right-clicking on a project > *Build*.:
* glslang * glslang

36
Utilities/JIT.cpp
View File

@ -1373,6 +1373,34 @@ std::string jit_compiler::cpu(const std::string& _cpu)
return m_cpu; return m_cpu;
} }
std::string jit_compiler::triple1()
{
#if defined(_WIN32)
return llvm::Triple::normalize(llvm::sys::getProcessTriple());
#elif defined(__APPLE__) && defined(ARCH_X64)
return llvm::Triple::normalize("x86_64-unknown-linux-gnu");
#elif defined(__APPLE__) && defined(ARCH_ARM64)
return llvm::Triple::normalize("aarch64-unknown-linux-gnu");
#else
return llvm::Triple::normalize(llvm::sys::getProcessTriple());
#endif
}
std::string jit_compiler::triple2()
{
#if defined(_WIN32) && defined(ARCH_X64)
return llvm::Triple::normalize("x86_64-unknown-linux-gnu");
#elif defined(_WIN32) && defined(ARCH_ARM64)
return llvm::Triple::normalize("aarch64-unknown-linux-gnu");
#elif defined(__APPLE__) && defined(ARCH_X64)
return llvm::Triple::normalize("x86_64-unknown-linux-gnu");
#elif defined(__APPLE__) && defined(ARCH_ARM64)
return llvm::Triple::normalize("aarch64-unknown-linux-gnu");
#else
return llvm::Triple::normalize(llvm::sys::getProcessTriple());
#endif
}
jit_compiler::jit_compiler(const std::unordered_map<std::string, u64>& _link, const std::string& _cpu, u32 flags) jit_compiler::jit_compiler(const std::unordered_map<std::string, u64>& _link, const std::string& _cpu, u32 flags)
: m_context(new llvm::LLVMContext) : m_context(new llvm::LLVMContext)
, m_cpu(cpu(_cpu)) , m_cpu(cpu(_cpu))
@ -1380,7 +1408,7 @@ jit_compiler::jit_compiler(const std::unordered_map<std::string, u64>& _link, co
std::string result; std::string result;
auto null_mod = std::make_unique<llvm::Module> ("null_", *m_context); auto null_mod = std::make_unique<llvm::Module> ("null_", *m_context);
null_mod->setTargetTriple(llvm::Triple::normalize(llvm::sys::getProcessTriple())); null_mod->setTargetTriple(jit_compiler::triple1());
std::unique_ptr<llvm::RTDyldMemoryManager> mem; std::unique_ptr<llvm::RTDyldMemoryManager> mem;
@ -1394,9 +1422,7 @@ jit_compiler::jit_compiler(const std::unordered_map<std::string, u64>& _link, co
else else
{ {
mem = std::make_unique<MemoryManager2>(); mem = std::make_unique<MemoryManager2>();
#if defined(_WIN32) && defined(ARCH_X64) null_mod->setTargetTriple(jit_compiler::triple2());
null_mod->setTargetTriple(llvm::Triple::normalize("x86_64-unknown-linux-gnu"));
#endif
} }
} }
else else
@ -1412,7 +1438,7 @@ jit_compiler::jit_compiler(const std::unordered_map<std::string, u64>& _link, co
.setOptLevel(llvm::CodeGenOpt::Aggressive) .setOptLevel(llvm::CodeGenOpt::Aggressive)
.setCodeModel(flags & 0x2 ? llvm::CodeModel::Large : llvm::CodeModel::Small) .setCodeModel(flags & 0x2 ? llvm::CodeModel::Large : llvm::CodeModel::Small)
#ifdef __APPLE__ #ifdef __APPLE__
.setCodeModel(llvm::CodeModel::Large) //.setCodeModel(llvm::CodeModel::Large)
#endif #endif
.setRelocationModel(llvm::Reloc::Model::PIC_) .setRelocationModel(llvm::Reloc::Model::PIC_)
.setMCPU(m_cpu) .setMCPU(m_cpu)

6
Utilities/JIT.h
View File

@ -546,6 +546,12 @@ public:
// Get CPU info // Get CPU info
static std::string cpu(const std::string& _cpu); static std::string cpu(const std::string& _cpu);
// Get system triple (PPU)
static std::string triple1();
// Get system triple (SPU)
static std::string triple2();
}; };
#endif #endif

7
rpcs3/Emu/CPU/CPUTranslator.cpp
View File

@ -125,7 +125,9 @@ void cpu_translator::initialize(llvm::LLVMContext& context, llvm::ExecutionEngin
cpu == "bdver2" || cpu == "bdver2" ||
cpu == "bdver3" || cpu == "bdver3" ||
cpu == "bdver4" || cpu == "bdver4" ||
cpu.startswith("znver")) cpu == "znver1" ||
cpu == "znver2" ||
cpu == "znver3")
{ {
m_use_fma = true; m_use_fma = true;
m_use_avx = true; m_use_avx = true;
@ -158,7 +160,8 @@ void cpu_translator::initialize(llvm::LLVMContext& context, llvm::ExecutionEngin
cpu == "icelake-server" || cpu == "icelake-server" ||
cpu == "tigerlake" || cpu == "tigerlake" ||
cpu == "rocketlake" || cpu == "rocketlake" ||
cpu == "sapphirerapids") cpu == "sapphirerapids" ||
(cpu.startswith("znver") && cpu != "znver1" && cpu != "znver2" && cpu != "znver3"))
{ {
m_use_avx = true; m_use_avx = true;
m_use_fma = true; m_use_fma = true;

1
rpcs3/Emu/CPU/CPUTranslator.h
View File

@ -13,6 +13,7 @@
#pragma GCC diagnostic ignored "-Wstrict-aliasing" #pragma GCC diagnostic ignored "-Wstrict-aliasing"
#pragma GCC diagnostic ignored "-Weffc++" #pragma GCC diagnostic ignored "-Weffc++"
#pragma GCC diagnostic ignored "-Wmissing-noreturn" #pragma GCC diagnostic ignored "-Wmissing-noreturn"
#pragma GCC diagnostic ignored "-Wredundant-decls"
#endif #endif
#include "llvm/IR/LLVMContext.h" #include "llvm/IR/LLVMContext.h"
#include "llvm/ExecutionEngine/ExecutionEngine.h" #include "llvm/ExecutionEngine/ExecutionEngine.h"

10
rpcs3/Emu/Cell/PPUThread.cpp
View File

@ -2132,7 +2132,7 @@ static void ppu_check(ppu_thread& ppu, u64 addr)
// ppu_check() shall not return directly // ppu_check() shall not return directly
if (ppu.test_stopped()) if (ppu.test_stopped())
; {}
ppu_escape(&ppu); ppu_escape(&ppu);
} }
@ -3690,7 +3690,7 @@ bool ppu_initialize(const ppu_module& info, bool check_only)
// Settings: should be populated by settings which affect codegen (TODO) // Settings: should be populated by settings which affect codegen (TODO)
enum class ppu_settings : u32 enum class ppu_settings : u32
{ {
non_win32, platform_bit,
accurate_dfma, accurate_dfma,
fixup_vnan, fixup_vnan,
fixup_nj_denormals, fixup_nj_denormals,
@ -3707,8 +3707,8 @@ bool ppu_initialize(const ppu_module& info, bool check_only)
be_t<bs_t<ppu_settings>> settings{}; be_t<bs_t<ppu_settings>> settings{};
#ifndef _WIN32 #if !defined(_WIN32) && !defined(__APPLE__)
settings += ppu_settings::non_win32; settings += ppu_settings::platform_bit;
#endif #endif
if (g_cfg.core.use_accurate_dfma) if (g_cfg.core.use_accurate_dfma)
settings += ppu_settings::accurate_dfma; settings += ppu_settings::accurate_dfma;
@ -3937,7 +3937,7 @@ static void ppu_initialize2(jit_compiler& jit, const ppu_module& module_part, co
std::unique_ptr<Module> _module = std::make_unique<Module>(obj_name, jit.get_context()); std::unique_ptr<Module> _module = std::make_unique<Module>(obj_name, jit.get_context());
// Initialize target // Initialize target
_module->setTargetTriple(Triple::normalize(sys::getProcessTriple())); _module->setTargetTriple(jit_compiler::triple1());
_module->setDataLayout(jit.get_engine().getTargetMachine()->createDataLayout()); _module->setDataLayout(jit.get_engine().getTargetMachine()->createDataLayout());
// Initialize translator // Initialize translator

71
rpcs3/Emu/Cell/SPURecompiler.cpp
View File

@ -4290,7 +4290,7 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
const auto fail = llvm::BasicBlock::Create(m_context, "", m_function); const auto fail = llvm::BasicBlock::Create(m_context, "", m_function);
m_ir->CreateCondBr(m_ir->CreateICmpEQ(m_base_pc, m_ir->getInt32(m_base)), next, fail); m_ir->CreateCondBr(m_ir->CreateICmpEQ(m_base_pc, m_ir->getInt32(m_base)), next, fail);
m_ir->SetInsertPoint(fail); m_ir->SetInsertPoint(fail);
m_ir->CreateStore(m_ir->getInt32(target), spu_ptr<u32>(&spu_thread::pc), true); m_ir->CreateStore(m_ir->getInt32(target), spu_ptr<u32>(&spu_thread::pc));
tail_chunk(nullptr); tail_chunk(nullptr);
m_ir->SetInsertPoint(next); m_ir->SetInsertPoint(next);
} }
@ -4328,7 +4328,7 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
{ {
ensure(!m_finfo->fn); ensure(!m_finfo->fn);
m_ir->CreateStore(m_ir->getInt32(target), spu_ptr<u32>(&spu_thread::pc), true); m_ir->CreateStore(m_ir->getInt32(target), spu_ptr<u32>(&spu_thread::pc));
} }
else else
{ {
@ -4766,7 +4766,7 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
} }
template <typename T, uint I> template <typename T, uint I>
void set_vr(const bf_t<u32, I, 7>& index, T expr, bool fixup = true) void set_vr(const bf_t<u32, I, 7>& index, T expr, std::function<llvm::KnownBits()> vr_assume = nullptr, bool fixup = true)
{ {
// Process expression // Process expression
const auto value = expr.eval(m_ir); const auto value = expr.eval(m_ir);
@ -4790,6 +4790,10 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
return; return;
} }
if (vr_assume)
{
}
set_reg_fixed(index, value, fixup); set_reg_fixed(index, value, fixup);
} }
@ -4869,7 +4873,7 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
// Update PC for current or explicitly specified instruction address // Update PC for current or explicitly specified instruction address
void update_pc(u32 target = -1) void update_pc(u32 target = -1)
{ {
m_ir->CreateStore(m_ir->CreateAnd(get_pc(target + 1 ? target : m_pos), 0x3fffc), spu_ptr<u32>(&spu_thread::pc), true); m_ir->CreateStore(m_ir->CreateAnd(get_pc(target + 1 ? target : m_pos), 0x3fffc), spu_ptr<u32>(&spu_thread::pc));
} }
// Call cpu_thread::check_state if necessary and return or continue (full check) // Call cpu_thread::check_state if necessary and return or continue (full check)
@ -4889,14 +4893,14 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
if (may_be_unsafe_for_savestate) if (may_be_unsafe_for_savestate)
{ {
m_ir->CreateStore(m_ir->getInt8(1), spu_ptr<u8>(&spu_thread::unsavable), true); m_ir->CreateStore(m_ir->getInt8(1), spu_ptr<u8>(&spu_thread::unsavable));
} }
m_ir->CreateStore(m_ir->getFalse(), m_fake_global1, true); m_ir->CreateStore(m_ir->getFalse(), m_fake_global1);
if (may_be_unsafe_for_savestate) if (may_be_unsafe_for_savestate)
{ {
m_ir->CreateStore(m_ir->getInt8(0), spu_ptr<u8>(&spu_thread::unsavable), true); m_ir->CreateStore(m_ir->getInt8(0), spu_ptr<u8>(&spu_thread::unsavable));
} }
m_ir->CreateBr(_body); m_ir->CreateBr(_body);
@ -5003,11 +5007,7 @@ public:
// Create LLVM module // Create LLVM module
std::unique_ptr<Module> _module = std::make_unique<Module>(m_hash + ".obj", m_context); std::unique_ptr<Module> _module = std::make_unique<Module>(m_hash + ".obj", m_context);
#if defined(_WIN32) && defined(ARCH_X64) _module->setTargetTriple(jit_compiler::triple2());
_module->setTargetTriple(Triple::normalize("x86_64-unknown-linux-gnu"));
#else
_module->setTargetTriple(Triple::normalize(sys::getProcessTriple()));
#endif
_module->setDataLayout(m_jit.get_engine().getTargetMachine()->createDataLayout()); _module->setDataLayout(m_jit.get_engine().getTargetMachine()->createDataLayout());
m_module = _module.get(); m_module = _module.get();
@ -5035,7 +5035,7 @@ public:
// Emit state check // Emit state check
const auto pstate = spu_ptr<u32>(&spu_thread::state); const auto pstate = spu_ptr<u32>(&spu_thread::state);
m_ir->CreateCondBr(m_ir->CreateICmpNE(m_ir->CreateLoad(get_type<u32>(), pstate, true), m_ir->getInt32(0)), label_stop, label_test, m_md_unlikely); m_ir->CreateCondBr(m_ir->CreateICmpNE(m_ir->CreateLoad(get_type<u32>(), pstate), m_ir->getInt32(0)), label_stop, label_test, m_md_unlikely);
// Emit code check // Emit code check
u32 check_iterations = 0; u32 check_iterations = 0;
@ -5043,7 +5043,7 @@ public:
// Set block hash for profiling (if enabled) // Set block hash for profiling (if enabled)
if (g_cfg.core.spu_prof && g_cfg.core.spu_verification) if (g_cfg.core.spu_prof && g_cfg.core.spu_verification)
m_ir->CreateStore(m_ir->getInt64((m_hash_start & -65536)), spu_ptr<u64>(&spu_thread::block_hash), true); m_ir->CreateStore(m_ir->getInt64((m_hash_start & -65536)), spu_ptr<u64>(&spu_thread::block_hash));
if (!g_cfg.core.spu_verification) if (!g_cfg.core.spu_verification)
{ {
@ -5294,7 +5294,7 @@ public:
// Set block hash for profiling (if enabled) // Set block hash for profiling (if enabled)
if (g_cfg.core.spu_prof) if (g_cfg.core.spu_prof)
m_ir->CreateStore(m_ir->getInt64((m_hash_start & -65536) | (m_entry >> 2)), spu_ptr<u64>(&spu_thread::block_hash), true); m_ir->CreateStore(m_ir->getInt64((m_hash_start & -65536) | (m_entry >> 2)), spu_ptr<u64>(&spu_thread::block_hash));
m_finfo = &m_functions[m_entry]; m_finfo = &m_functions[m_entry];
m_ir->CreateBr(add_block(m_entry)); m_ir->CreateBr(add_block(m_entry));
@ -5535,7 +5535,7 @@ public:
{ {
for (auto& i : bb) for (auto& i : bb)
{ {
// Replace volatile fake store with spu_test_state call // Replace fake store with spu_test_state call
if (auto si = dyn_cast<StoreInst>(&i); si && si->getOperand(1) == m_fake_global1) if (auto si = dyn_cast<StoreInst>(&i); si && si->getOperand(1) == m_fake_global1)
{ {
m_ir->SetInsertPoint(si); m_ir->SetInsertPoint(si);
@ -5681,11 +5681,7 @@ public:
// Create LLVM module // Create LLVM module
std::unique_ptr<Module> _module = std::make_unique<Module>("spu_interpreter.obj", m_context); std::unique_ptr<Module> _module = std::make_unique<Module>("spu_interpreter.obj", m_context);
#if defined(_WIN32) && defined(ARCH_X64) _module->setTargetTriple(jit_compiler::triple2());
_module->setTargetTriple(Triple::normalize("x86_64-unknown-linux-gnu"));
#else
_module->setTargetTriple(Triple::normalize(sys::getProcessTriple()));
#endif
_module->setDataLayout(m_jit.get_engine().getTargetMachine()->createDataLayout()); _module->setDataLayout(m_jit.get_engine().getTargetMachine()->createDataLayout());
m_module = _module.get(); m_module = _module.get();
@ -5985,7 +5981,7 @@ public:
ncall->setTailCall(); ncall->setTailCall();
m_ir->CreateRetVoid(); m_ir->CreateRetVoid();
m_ir->SetInsertPoint(_stop); m_ir->SetInsertPoint(_stop);
m_ir->CreateStore(m_interp_pc, spu_ptr<u32>(&spu_thread::pc), true); m_ir->CreateStore(m_interp_pc, spu_ptr<u32>(&spu_thread::pc));
call("spu_escape", spu_runtime::g_escape, m_thread)->setTailCall(); call("spu_escape", spu_runtime::g_escape, m_thread)->setTailCall();
m_ir->CreateRetVoid(); m_ir->CreateRetVoid();
} }
@ -6233,8 +6229,9 @@ public:
} }
else else
{ {
const auto val = m_ir->CreateLoad(get_type<u64>(), ptr, true); const auto val = m_ir->CreateLoad(get_type<u64>(), ptr);
m_ir->CreateStore(m_ir->getInt64(0), ptr, true); val->setAtomic(llvm::AtomicOrdering::Unordered);
m_ir->CreateStore(m_ir->getInt64(0), ptr)->setAtomic(llvm::AtomicOrdering::Unordered);
val0 = val; val0 = val;
} }
@ -6319,7 +6316,7 @@ public:
} }
case SPU_RdEventMask: case SPU_RdEventMask:
{ {
res.value = m_ir->CreateTrunc(m_ir->CreateLShr(m_ir->CreateLoad(get_type<u64>(), spu_ptr<u64>(&spu_thread::ch_events), true), 32), get_type<u32>()); res.value = m_ir->CreateTrunc(m_ir->CreateLShr(m_ir->CreateLoad(get_type<u64>(), spu_ptr<u64>(&spu_thread::ch_events)), 32), get_type<u32>());
break; break;
} }
case SPU_RdEventStat: case SPU_RdEventStat:
@ -6575,7 +6572,7 @@ public:
// Illegal update, access violate with special address // Illegal update, access violate with special address
m_ir->SetInsertPoint(fail); m_ir->SetInsertPoint(fail);
const auto ptr = _ptr<u32>(m_memptr, 0xffdead04); const auto ptr = _ptr<u32>(m_memptr, 0xffdead04);
m_ir->CreateStore(m_ir->getInt32("TAG\0"_u32), ptr, true); m_ir->CreateStore(m_ir->getInt32("TAG\0"_u32), ptr);
m_ir->CreateBr(next); m_ir->CreateBr(next);
m_ir->SetInsertPoint(any); m_ir->SetInsertPoint(any);
@ -7097,7 +7094,7 @@ public:
minusb = eval(x); minusb = eval(x);
} }
if (auto k = get_known_bits(minusb); (k & kbc<u32>(32)).isZero()) if (auto k = get_known_bits(minusb); !!(k.Zero & 32))
{ {
set_vr(op.rt, a >> (minusb & 31)); set_vr(op.rt, a >> (minusb & 31));
return; return;
@ -7116,7 +7113,7 @@ public:
minusb = eval(x); minusb = eval(x);
} }
if (auto k = get_known_bits(minusb); (k & kbc<u32>(32)).isZero()) if (auto k = get_known_bits(minusb); !!(k.Zero & 32))
{ {
set_vr(op.rt, a >> (minusb & 31)); set_vr(op.rt, a >> (minusb & 31));
return; return;
@ -7129,7 +7126,7 @@ public:
{ {
const auto [a, b] = get_vrs<u32[4]>(op.ra, op.rb); const auto [a, b] = get_vrs<u32[4]>(op.ra, op.rb);
if (auto k = get_known_bits(b); (k & kbc<u32>(32)).isZero()) if (auto k = get_known_bits(b); !!(k.Zero & 32))
{ {
set_vr(op.rt, a << (b & 31)); set_vr(op.rt, a << (b & 31));
return; return;
@ -7154,7 +7151,7 @@ public:
minusb = eval(x); minusb = eval(x);
} }
if (auto k = get_known_bits(minusb); (k & kbc<u16>(16)).isZero()) if (auto k = get_known_bits(minusb); !!(k.Zero & 16))
{ {
set_vr(op.rt, a >> (minusb & 15)); set_vr(op.rt, a >> (minusb & 15));
return; return;
@ -7173,7 +7170,7 @@ public:
minusb = eval(x); minusb = eval(x);
} }
if (auto k = get_known_bits(minusb); (k & kbc<u16>(16)).isZero()) if (auto k = get_known_bits(minusb); !!(k.Zero & 16))
{ {
set_vr(op.rt, a >> (minusb & 15)); set_vr(op.rt, a >> (minusb & 15));
return; return;
@ -7186,7 +7183,7 @@ public:
{ {
const auto [a, b] = get_vrs<u16[8]>(op.ra, op.rb); const auto [a, b] = get_vrs<u16[8]>(op.ra, op.rb);
if (auto k = get_known_bits(b); (k & kbc<u16>(16)).isZero()) if (auto k = get_known_bits(b); !!(k.Zero & 16))
{ {
set_vr(op.rt, a << (b & 15)); set_vr(op.rt, a << (b & 15));
return; return;
@ -8992,7 +8989,7 @@ public:
const auto i = select(a > 0x47f0000000000000, eval(s | 0x47f0000000000000), d); const auto i = select(a > 0x47f0000000000000, eval(s | 0x47f0000000000000), d);
const auto n = select(a > 0x7ff0000000000000, splat<s64[2]>(0x47f8000000000000), i); const auto n = select(a > 0x7ff0000000000000, splat<s64[2]>(0x47f8000000000000), i);
const auto z = select(a < 0x3810000000000000, s, n); const auto z = select(a < 0x3810000000000000, s, n);
set_vr(op.rt, zshuffle(bitcast<f64[2]>(z), 2, 0, 3, 1), false); set_vr(op.rt, zshuffle(bitcast<f64[2]>(z), 2, 0, 3, 1), nullptr, false);
return; return;
} }
@ -9808,11 +9805,11 @@ public:
m_ir->CreateCondBr(cond.value, halt, next, m_md_unlikely); m_ir->CreateCondBr(cond.value, halt, next, m_md_unlikely);
m_ir->SetInsertPoint(halt); m_ir->SetInsertPoint(halt);
if (m_interp_magn) if (m_interp_magn)
m_ir->CreateStore(m_function->getArg(2), spu_ptr<u32>(&spu_thread::pc))->setVolatile(true); m_ir->CreateStore(m_function->getArg(2), spu_ptr<u32>(&spu_thread::pc));
else else
update_pc(); update_pc();
const auto ptr = _ptr<u32>(m_memptr, 0xffdead00); const auto ptr = _ptr<u32>(m_memptr, 0xffdead00);
m_ir->CreateStore(m_ir->getInt32("HALT"_u32), ptr)->setVolatile(true); m_ir->CreateStore(m_ir->getInt32("HALT"_u32), ptr);
m_ir->CreateBr(next); m_ir->CreateBr(next);
m_ir->SetInsertPoint(next); m_ir->SetInsertPoint(next);
} }
@ -10292,8 +10289,8 @@ public:
if (m_finfo && m_finfo->fn) if (m_finfo && m_finfo->fn)
{ {
// Can't afford external tail call in true functions // Can't afford external tail call in true functions
m_ir->CreateStore(m_ir->getInt32("BIJT"_u32), _ptr<u32>(m_memptr, 0xffdead20))->setVolatile(true); m_ir->CreateStore(m_ir->getInt32("BIJT"_u32), _ptr<u32>(m_memptr, 0xffdead20));
m_ir->CreateStore(m_ir->getFalse(), m_fake_global1, true); m_ir->CreateStore(m_ir->getFalse(), m_fake_global1);
m_ir->CreateBr(sw->getDefaultDest()); m_ir->CreateBr(sw->getDefaultDest());
} }
else else