archive/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2025-07-14 02:38:37 +12:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

SPU LLVM: disable AVX2 shift intrinsics

Browse source 
Was incorrectly checked.
This commit is contained in:
Ivan Chikish 2023-04-23 10:23:39 +03:00 • committed by Ivan
parent a6e4b77273
commit 45fecf0059
1 changed files with 0 additions and 96 deletions
Download patch file Download diff file Expand all files Collapse all files

96
rpcs3/Emu/CPU/CPUTranslator.h
View file

@ -3356,37 +3356,6 @@ public:
{
static constexpr u32 esz = llvm_value_t<CT>::esize;
if constexpr (esz == 32)
{
#if defined(ARCH_X64)
if (m_use_fma && !llvm::isa<llvm::Constant>(b.eval(m_ir)))
return eval(llvm_calli<CT, T, U>{"llvm.x86.avx2.psllv.d", {std::forward<T>(a), std::forward<U>(b)}});
#endif
}
if constexpr (esz == 16)
{
#if defined(ARCH_X64)
if (m_use_avx512 && !llvm::isa<llvm::Constant>(b.eval(m_ir)))
return eval(llvm_calli<CT, T, U>{"llvm.x86.avx512.psllv.w.128", {std::forward<T>(a), std::forward<U>(b)}});
if (m_use_fma && !llvm::isa<llvm::Constant>(b.eval(m_ir)))
{
using t32 = value_t<u32[4]>;
auto a32 = eval(bitcast<u32[4]>(std::forward<T>(a)));
auto b32 = eval(bitcast<u32[4]>(std::forward<U>(b)));
auto sizeL = eval(b32 & 0xffff);
auto sizeH = eval(b32 >> 16);
auto dataL = eval(llvm_calli<u32[4], t32, t32>{"llvm.x86.avx2.psllv.d", {a32, sizeL}});
auto dataH = eval(llvm_calli<u32[4], t32, t32>{"llvm.x86.avx2.psllv.d", {eval(a32 & 0xffff0000), sizeH}});
return eval(bitcast<CT>((dataL & 0xffff) | dataH));
}
#endif
}
return eval(select(b < esz, a << b, splat<CT>(0)));
/*
return expr(select(b < esz, a << b, splat<CT>(0)), [](llvm::Value*& value, llvm::Module* _m) -> llvm_match_tuple<T, U>
{
static const auto M = match<CT>();
@ -3405,7 +3374,6 @@ public:
value = nullptr;
return {};
});
*/
}
// Infinite-precision logical shift right (unsigned)
@ -3414,37 +3382,6 @@ public:
{
static constexpr u32 esz = llvm_value_t<CT>::esize;
if constexpr (esz == 32)
{
#if defined(ARCH_X64)
if (m_use_fma && !llvm::isa<llvm::Constant>(b.eval(m_ir)))
return eval(llvm_calli<CT, T, U>{"llvm.x86.avx2.psrlv.d", {std::forward<T>(a), std::forward<U>(b)}});
#endif
}
if constexpr (esz == 16)
{
#if defined(ARCH_X64)
if (m_use_avx512 && !llvm::isa<llvm::Constant>(b.eval(m_ir)))
return eval(llvm_calli<CT, T, U>{"llvm.x86.avx512.psrlv.w.128", {std::forward<T>(a), std::forward<U>(b)}});
if (m_use_fma && !llvm::isa<llvm::Constant>(b.eval(m_ir)))
{
using t32 = value_t<u32[4]>;
auto a32 = eval(bitcast<u32[4]>(std::forward<T>(a)));
auto b32 = eval(bitcast<u32[4]>(std::forward<U>(b)));
auto sizeL = eval(b32 & 0xffff);
auto sizeH = eval(b32 >> 16);
auto dataL = eval(llvm_calli<u32[4], t32, t32>{"llvm.x86.avx2.psrlv.d", {eval(a32 & 0xffff), sizeL}});
auto dataH = eval(llvm_calli<u32[4], t32, t32>{"llvm.x86.avx2.psrlv.d", {a32, sizeH}});
return eval(bitcast<CT>(dataL | (dataH & 0xffff0000)));
}
#endif
}
return eval(select(b < esz, a >> b, splat<CT>(0)));
/*
return expr(select(b < esz, a >> b, splat<CT>(0)), [](llvm::Value*& value, llvm::Module* _m) -> llvm_match_tuple<T, U>
{
static const auto M = match<CT>();
@ -3463,7 +3400,6 @@ public:
value = nullptr;
return {};
});
*/
}
// Infinite-precision arithmetic shift right (signed)
@ -3472,37 +3408,6 @@ public:
{
static constexpr u32 esz = llvm_value_t<CT>::esize;
if constexpr (esz == 32)
{
#if defined(ARCH_X64)
if (m_use_fma && !llvm::isa<llvm::Constant>(b.eval(m_ir)))
return eval(llvm_calli<CT, T, U>{"llvm.x86.avx2.psrav.d", {std::forward<T>(a), std::forward<U>(b)}});
#endif
}
if constexpr (esz == 16)
{
#if defined(ARCH_X64)
if (m_use_avx512 && !llvm::isa<llvm::Constant>(b.eval(m_ir)))
return eval(llvm_calli<CT, T, U>{"llvm.x86.avx512.psrav.w.128", {std::forward<T>(a), std::forward<U>(b)}});
if (m_use_fma && !llvm::isa<llvm::Constant>(b.eval(m_ir)))
{
using t32 = value_t<u32[4]>;
auto a32 = eval(bitcast<u32[4]>(std::forward<T>(a)));
auto b32 = eval(bitcast<u32[4]>(std::forward<U>(b)));
auto sizeL = eval(b32 & 0xffff);
auto sizeH = eval(b32 >> 16);
auto dataL = eval(llvm_calli<u32[4], t32, t32>{"llvm.x86.avx2.psrav.d", {eval(a32 << 16), sizeL}});
auto dataH = eval(llvm_calli<u32[4], t32, t32>{"llvm.x86.avx2.psrav.d", {a32, sizeH}});
return eval(bitcast<CT>((dataL >> 16) | (dataH & 0xffff0000)));
}
#endif
}
return eval(a >> select(b > (esz - 1), splat<CT>(esz - 1), b));
/*
return expr(a >> select(b > (esz - 1), splat<CT>(esz - 1), b), [](llvm::Value*& value, llvm::Module* _m) -> llvm_match_tuple<T, U>
{
static const auto M = match<CT>();
@ -3521,7 +3426,6 @@ public:
value = nullptr;
return {};
});
*/
}
template <typename... Types>