PowerPC: Negate m_dec values in frsqrte table

This value is used in a multiplication. The result of this
multiplication is then subtracted from m_base. By negating m_dec, we are
free to use an addition instead.

On x64, this saves an instruction.
This commit is contained in:
Sintendo
2023-10-10 22:47:31 +02:00
parent 60e3b4c093
commit 3c3168706c
3 changed files with 14 additions and 15 deletions

View File

@ -74,14 +74,14 @@ u32 ClassifyFloat(float fvalue)
}
const std::array<BaseAndDec, 32> frsqrte_expected = {{
{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},
{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)
@ -134,7 +134,7 @@ double ApproximateReciprocalSquareRoot(double val)
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;
integral |= static_cast<s64>(entry.m_base + entry.m_dec * (i % 2048)) << 26;
return BitCast<double>(integral);
}