EmuCodeBlock: Prefer MOVAPS to MOVSD in ConvertDoubleToSingle

* The high half of the register is immediately masked so the value in it is irrelevant.
* MOVSD produces an unnecessary dependency on the high half of regOp.
* MOVAPS is implemented as a register rename on modern microarchitectures.
This commit is contained in:
MerryMage 2019-04-23 11:46:50 +01:00
parent 1baa8ee970
commit e06111e86f

View File

@ -894,7 +894,7 @@ alignas(16) static const __m128i double_qnan_bit = _mm_set_epi64x(0xffffffffffff
// unless the exponent is in the range of 874 to 896.
void EmuCodeBlock::ConvertDoubleToSingle(X64Reg dst, X64Reg src)
{
MOVSD(XMM1, R(src));
MOVAPD(XMM1, R(src));
// Grab Exponent
PAND(XMM1, MConst(double_exponent));
@ -914,7 +914,7 @@ void EmuCodeBlock::ConvertDoubleToSingle(X64Reg dst, X64Reg src)
PSUBQ(XMM0, R(XMM1));
// xmm1 = fraction | 0x0010000000000000
MOVSD(XMM1, R(src));
MOVAPD(XMM1, R(src));
PAND(XMM1, MConst(double_fraction));
POR(XMM1, MConst(double_explicit_top_bit));
@ -922,7 +922,7 @@ void EmuCodeBlock::ConvertDoubleToSingle(X64Reg dst, X64Reg src)
PSRLQ(XMM1, R(XMM0));
// OR the sign bit in.
MOVSD(XMM0, R(src));
MOVAPD(XMM0, R(src));
PAND(XMM0, MConst(double_sign_bit));
PSRLQ(XMM0, 32);
POR(XMM1, R(XMM0));
@ -934,12 +934,12 @@ void EmuCodeBlock::ConvertDoubleToSingle(X64Reg dst, X64Reg src)
// Don't Denormalize
// We want bits 0, 1
MOVSD(XMM1, R(src));
MOVAPD(XMM1, R(src));
PAND(XMM1, MConst(double_top_two_bits));
PSRLQ(XMM1, 32);
// And 5 through to 34
MOVSD(XMM0, R(src));
MOVAPD(XMM0, R(src));
PAND(XMM0, MConst(double_bottom_bits));
PSRLQ(XMM0, 29);