saulteafarmer/XenonRecomp
1
0
Fork 0
mirror of https://github.com/hedge-dev/XenonRecomp.git synced 2025-06-06 18:31:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

Implement __rdtsc and FPCSR for ARM64.

Browse Source
This commit is contained in:
squidbus 2025-03-03 17:37:59 -08:00 committed by Isaac Marovitz
parent 60e421da46
commit 8237abe22e
No known key found for this signature in database
GPG Key ID: 97250B2B09A132E1
1 changed files with 60 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

71
XenonUtils/ppc_context.h
View File

@ -221,34 +221,71 @@ struct PPCFPSCRRegister
{ {
uint32_t csr; uint32_t csr;
static constexpr size_t GuestToHost[] = { SIMDE_MM_ROUND_NEAREST, SIMDE_MM_ROUND_TOWARD_ZERO, SIMDE_MM_ROUND_UP, SIMDE_MM_ROUND_DOWN };
static constexpr size_t HostToGuest[] = { PPC_ROUND_NEAREST, PPC_ROUND_DOWN, PPC_ROUND_UP, PPC_ROUND_TOWARD_ZERO }; static constexpr size_t HostToGuest[] = { PPC_ROUND_NEAREST, PPC_ROUND_DOWN, PPC_ROUND_UP, PPC_ROUND_TOWARD_ZERO };
// simde does not handle denormal flags, so we need to implement per-arch.
#if defined(__x86_64__) || defined(_M_X64)
static constexpr size_t RoundShift = 13;
static constexpr size_t RoundMask = SIMDE_MM_ROUND_MASK;
static constexpr size_t FlushMask = SIMDE_MM_FLUSH_ZERO_MASK | _MM_DENORMALS_ZERO_MASK;
static constexpr size_t GuestToHost[] = { SIMDE_MM_ROUND_NEAREST, SIMDE_MM_ROUND_TOWARD_ZERO, SIMDE_MM_ROUND_UP, SIMDE_MM_ROUND_DOWN };
inline uint32_t getcsr() noexcept
{
return simde_mm_getcsr();
}
inline void setcsr(uint32_t csr) noexcept
{
simde_mm_setcsr(csr);
}
#elif defined(__aarch64__) || defined(_M_ARM64)
// RMode
static constexpr size_t RoundShift = 22;
static constexpr size_t RoundMask = 3 << RoundShift;
// FZ and FZ16
static constexpr size_t FlushMask = (1 << 19) | (1 << 24);
// Nearest, Zero, -Infinity, -Infinity
static constexpr size_t GuestToHost[] = { 0 << RoundShift, 3 << RoundShift, 1 << RoundShift, 2 << RoundShift };
inline uint32_t getcsr() noexcept
{
uint64_t csr;
__asm__ __volatile__("mrs %0, fpcr" : "=r"(csr));
return csr;
}
inline void setcsr(uint32_t csr) noexcept
{
__asm__ __volatile__("msr fpcr, %0" : : "r"(csr));
}
#else
# error "Missing implementation for FPSCR."
#endif
inline uint32_t loadFromHost() noexcept inline uint32_t loadFromHost() noexcept
{ {
csr = simde_mm_getcsr(); csr = getcsr();
return HostToGuest[(csr & SIMDE_MM_ROUND_MASK) >> 13]; return HostToGuest[(csr & RoundMask) >> RoundShift];
} }
inline void storeFromGuest(uint32_t value) noexcept inline void storeFromGuest(uint32_t value) noexcept
{ {
csr &= ~SIMDE_MM_ROUND_MASK; csr &= ~RoundMask;
csr |= GuestToHost[value & PPC_ROUND_MASK]; csr |= GuestToHost[value & PPC_ROUND_MASK];
simde_mm_setcsr(csr); setcsr(csr);
} }
static constexpr size_t FlushMask = SIMDE_MM_FLUSH_ZERO_MASK | _MM_DENORMALS_ZERO_MASK;
inline void enableFlushModeUnconditional() noexcept inline void enableFlushModeUnconditional() noexcept
{ {
csr |= FlushMask; csr |= FlushMask;
simde_mm_setcsr(csr); setcsr(csr);
} }
inline void disableFlushModeUnconditional() noexcept inline void disableFlushModeUnconditional() noexcept
{ {
csr &= ~FlushMask; csr &= ~FlushMask;
simde_mm_setcsr(csr); setcsr(csr);
} }
inline void enableFlushMode() noexcept inline void enableFlushMode() noexcept
@ -256,7 +293,7 @@ struct PPCFPSCRRegister
if ((csr & FlushMask) != FlushMask) [[unlikely]] if ((csr & FlushMask) != FlushMask) [[unlikely]]
{ {
csr |= FlushMask; csr |= FlushMask;
simde_mm_setcsr(csr); setcsr(csr);
} }
} }
@ -265,7 +302,7 @@ struct PPCFPSCRRegister
if ((csr & FlushMask) != 0) [[unlikely]] if ((csr & FlushMask) != 0) [[unlikely]]
{ {
csr &= ~FlushMask; csr &= ~FlushMask;
simde_mm_setcsr(csr); setcsr(csr);
} }
} }
}; };
@ -657,4 +694,16 @@ inline simde__m128i simde_mm_vsr(simde__m128i a, simde__m128i b)
return simde_mm_castps_si128(simde_mm_insert_ps(simde_mm_castsi128_ps(simde_mm_srl_epi64(a, b)), simde_mm_castsi128_ps(simde_mm_srl_epi64(simde_mm_srli_si128(a, 4), b)), 0x10)); return simde_mm_castps_si128(simde_mm_insert_ps(simde_mm_castsi128_ps(simde_mm_srl_epi64(a, b)), simde_mm_castsi128_ps(simde_mm_srl_epi64(simde_mm_srli_si128(a, 4), b)), 0x10));
} }
#if defined(__aarch64__) || defined(_M_ARM64)
inline uint64_t __rdtsc()
{
uint64_t ret;
asm volatile("mrs %0, cntvct_el0\n\t"
: "=r"(ret)::"memory");
return ret;
}
#elif !defined(__x86_64__) && !defined(_M_X64)
# error "Missing implementation for __rdtsc()"
#endif
#endif #endif