ShuriZma
/
dolphin
mirror of https://github.com/dolphin-emu/dolphin.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #12209 from JosJuice/frsqrte-exp-lsb 

PowerPC: Flip the order of frsqrte_expected
This commit is contained in:
Admiral H. Curtiss 2023-10-10 10:38:07 +02:00 committed by GitHub
parent 410ec83575 9e0fea8fc7
commit 1b7a590b4b
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 7 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

15
Source/Core/Common/FloatUtils.cpp
View File

@ -74,14 +74,14 @@ u32 ClassifyFloat(float fvalue)
}
const std::array<BaseAndDec, 32> frsqrte_expected = {{
{0x3ffa000, 0x7a4}, {0x3c29000, 0x700}, {0x38aa000, 0x670}, {0x3572000, 0x5f2},
{0x3279000, 0x584}, {0x2fb7000, 0x524}, {0x2d26000, 0x4cc}, {0x2ac0000, 0x47e},
{0x2881000, 0x43a}, {0x2665000, 0x3fa}, {0x2468000, 0x3c2}, {0x2287000, 0x38e},
{0x20c1000, 0x35e}, {0x1f12000, 0x332}, {0x1d79000, 0x30a}, {0x1bf4000, 0x2e6},
{0x1a7e800, 0x568}, {0x17cb800, 0x4f3}, {0x1552800, 0x48d}, {0x130c000, 0x435},
{0x10f2000, 0x3e7}, {0x0eff000, 0x3a2}, {0x0d2e000, 0x365}, {0x0b7c000, 0x32e},
{0x09e5000, 0x2fc}, {0x0867000, 0x2d0}, {0x06ff000, 0x2a8}, {0x05ab800, 0x283},
{0x046a000, 0x261}, {0x0339800, 0x243}, {0x0218800, 0x226}, {0x0105800, 0x20b},
{0x3ffa000, 0x7a4}, {0x3c29000, 0x700}, {0x38aa000, 0x670}, {0x3572000, 0x5f2},
{0x3279000, 0x584}, {0x2fb7000, 0x524}, {0x2d26000, 0x4cc}, {0x2ac0000, 0x47e},
{0x2881000, 0x43a}, {0x2665000, 0x3fa}, {0x2468000, 0x3c2}, {0x2287000, 0x38e},
{0x20c1000, 0x35e}, {0x1f12000, 0x332}, {0x1d79000, 0x30a}, {0x1bf4000, 0x2e6},
}};
double ApproximateReciprocalSquareRoot(double val)
@ -128,13 +128,12 @@ double ApproximateReciprocalSquareRoot(double val)
exponent += 1LL << 52;
}
const bool odd_exponent = !(exponent & (1LL << 52));
const s64 exponent_lsb = exponent & (1LL << 52);
exponent = ((0x3FFLL << 52) - ((exponent - (0x3FELL << 52)) / 2)) & (0x7FFLL << 52);
integral = sign | exponent;
const int i = static_cast<int>(mantissa >> 37);
const int index = i / 2048 + (odd_exponent ? 16 : 0);
const auto& entry = frsqrte_expected[index];
const int i = static_cast<int>((exponent_lsb | mantissa) >> 37);
const auto& entry = frsqrte_expected[i / 2048];
integral |= static_cast<s64>(entry.m_base - entry.m_dec * (i % 2048)) << 26;
return BitCast<double>(integral);

1
Source/Core/Core/PowerPC/Jit64Common/Jit64AsmCommon.cpp
View File

@ -141,7 +141,6 @@ void CommonAsmRoutines::GenFrsqrte()
MOV(64, R(RSCRATCH_EXTRA), R(RSCRATCH));
SHR(64, R(RSCRATCH_EXTRA), Imm8(48));
AND(32, R(RSCRATCH_EXTRA), Imm8(0x1F));
XOR(32, R(RSCRATCH_EXTRA), Imm8(0x10)); // int index = i / 2048 + (odd_exponent ? 16 : 0);
PUSH(RSCRATCH2);
MOV(64, R(RSCRATCH2), ImmPtr(GetConstantFromPool(Common::frsqrte_expected)));

1
Source/Core/Core/PowerPC/JitArm64/JitAsm.cpp
View File

@ -343,7 +343,6 @@ void JitArm64::GenerateFrsqrte()
AND(ARM64Reg::X2, ARM64Reg::X2, LogicalImm(0x10, 64));
MOVP2R(ARM64Reg::X1, &Common::frsqrte_expected);
ORR(ARM64Reg::X2, ARM64Reg::X2, ARM64Reg::X3, ArithOption(ARM64Reg::X3, ShiftType::LSR, 48));
EOR(ARM64Reg::X2, ARM64Reg::X2, LogicalImm(0x10, 64));
ADD(ARM64Reg::X2, ARM64Reg::X1, ARM64Reg::X2, ArithOption(ARM64Reg::X2, ShiftType::LSL, 3));
LDP(IndexType::Signed, ARM64Reg::W1, ARM64Reg::W2, ARM64Reg::X2, 0);
UBFX(ARM64Reg::X3, ARM64Reg::X3, 37, 11);