piwalker/dolphin
1
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2024-11-14 21:37:52 -07:00
Code Issues Actions Packages Projects Releases Wiki Activity

Merge pull request #9751 from JosJuice/jitarm64-fcmpx-ftz

Browse Source 
JitArm64: Make fcmpX with flush-to-zero enabled less bad
This commit is contained in:
JMC47 2021-06-05 05:27:52 -04:00 committed by GitHub
commit 638909aec6
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 40 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
Source/Core/Common/ArmCPUDetect.cpp
View File

@ -63,13 +63,13 @@ CPUInfo::CPUInfo()
void CPUInfo::Detect()
{
// Set some defaults here
// When ARMv8 CPUs come out, these need to be updated.
HTT = false;
OS64bit = true;
CPU64bit = true;
Mode64bit = true;
vendor = CPUVendor::ARM;
bFlushToZero = true;
bAFP = false;
#ifdef __APPLE__
num_cores = std::thread::hardware_concurrency();

27
Source/Core/Common/ArmFPURoundMode.cpp
View File

@ -2,8 +2,10 @@
// Licensed under GPLv2+
// Refer to the license.txt file included.
#include "Common/CPUDetect.h"
#include "Common/CommonTypes.h"
#include "Common/FPURoundMode.h"
#include "Common/Logging/Log.h"
#ifdef _MSC_VER
#include <intrin.h>
@ -45,8 +47,25 @@ void SetPrecisionMode(PrecisionMode mode)
void SetSIMDMode(int rounding_mode, bool non_ieee_mode)
{
// Flush-To-Zero (non-IEEE mode: denormal outputs are set to +/- 0)
// When AH is disabled, FZ controls flush-to-zero for both inputs and outputs. When AH is enabled,
// FZ controls flush-to-zero for outputs, and FIZ controls flush-to-zero for inputs.
constexpr u32 FZ = 1 << 24;
constexpr u32 AH = 1 << 1;
constexpr u32 FIZ = 1 << 0;
constexpr u32 flush_to_zero_mask = FZ | AH | FIZ;
// On CPUs with FEAT_AFP support, setting AH = 1, FZ = 1, FIZ = 0 emulates the GC/Wii CPU's
// "non-IEEE mode". Unfortunately, FEAT_AFP didn't exist until 2020, so we can't count on setting
// AH actually doing anything. But flushing both inputs and outputs seems to cause less problems
// than flushing nothing, so let's just set FZ and AH and roll with whatever behavior we get.
const u32 flush_to_zero_bits = (non_ieee_mode ? FZ | AH : 0);
static bool afp_warning_shown = false;
if (!afp_warning_shown && !cpu_info.bAFP && non_ieee_mode)
{
afp_warning_shown = true;
WARN_LOG_FMT(POWERPC,
"Non-IEEE mode was requested, but host CPU is not known to support FEAT_AFP");
}
// lookup table for FPSCR.RN-to-FPCR.RMode translation
constexpr u32 rounding_mode_table[] = {
@ -55,9 +74,11 @@ void SetSIMDMode(int rounding_mode, bool non_ieee_mode)
(1 << 22), // +inf
(2 << 22), // -inf
};
constexpr u32 rounding_mode_mask = 3 << 22;
const u32 rounding_mode_bits = rounding_mode_table[rounding_mode];
const u64 base = default_fpcr & ~(0b111 << 22);
SetFPCR(base | rounding_mode_table[rounding_mode] | (non_ieee_mode ? FZ : 0));
const u64 base = default_fpcr & ~(flush_to_zero_mask | rounding_mode_mask);
SetFPCR(base | rounding_mode_bits | flush_to_zero_bits);
}
void SaveSIMDState()

1
Source/Core/Common/CPUDetect.h
View File

@ -64,6 +64,7 @@ struct CPUInfo
bool bCRC32 = false;
bool bSHA1 = false;
bool bSHA2 = false;
bool bAFP = false; // Alternate floating-point behavior
// Call Detect()
explicit CPUInfo();

15
Source/Core/Core/PowerPC/JitArm64/JitArm64_FloatingPoint.cpp
View File

@ -3,6 +3,7 @@
// Refer to the license.txt file included.
#include "Common/Arm64Emitter.h"
#include "Common/CPUDetect.h"
#include "Common/CommonTypes.h"
#include "Common/StringUtil.h"
@ -374,7 +375,19 @@ void JitArm64::fcmpX(UGeckoInstruction inst)
const u32 b = inst.FB;
const int crf = inst.CRFD;
const bool singles = fpr.IsSingle(a, true) && fpr.IsSingle(b, true);
// On the GC/Wii CPU, outputs are flushed to zero if FPSCR.NI is set, and inputs are never
// flushed to zero. Ideally we would emulate FPSCR.NI by setting FPCR.FZ and FPCR.AH, but
// unfortunately FPCR.AH is a very new feature that we can't rely on (as of 2021). For CPUs
// without FPCR.AH, the best we can do (without killing the performance by explicitly flushing
// outputs using bitwise operations) is to only set FPCR.FZ, which flushes both inputs and
// outputs. This may cause problems in some cases, and one such case is Pokémon Battle Revolution,
// which does not progress past the title screen if a denormal single compares equal to zero.
// Workaround: Perform the comparison using a double operation instead. This ensures that denormal
// singles behave correctly in comparisons, but we still have a problem with denormal doubles.
const bool input_ftz_workaround =
!cpu_info.bAFP && (!js.fpr_is_store_safe[a] || !js.fpr_is_store_safe[b]);
const bool singles = fpr.IsSingle(a, true) && fpr.IsSingle(b, true) && !input_ftz_workaround;
const RegType type = singles ? RegType::LowerPairSingle : RegType::LowerPair;
const auto reg_encoder = singles ? EncodeRegToSingle : EncodeRegToDouble;