Permalink
Browse files

Enable Accurate Double to Single Conversion

  • Loading branch information...
JMC47 committed May 31, 2018
1 parent 5f48c9f commit 2795376b61371069ef2949eb31dbbddd7d9e209c
Showing with 6 additions and 70 deletions.
  1. +6 −70 Source/Core/Core/PowerPC/Jit64Common/EmuCodeBlock.cpp
@@ -864,20 +864,15 @@ void EmuCodeBlock::Force25BitPrecision(X64Reg output, const OpArg& input, X64Reg
}
}
// Since the following float conversion functions are used in non-arithmetic PPC float instructions,
// they must convert floats bitexact and never flush denormals to zero or turn SNaNs into QNaNs.
// This means we can't use CVTSS2SD/CVTSD2SS :(
// The x87 FPU doesn't even support flush-to-zero so we can use FLD+FSTP even on denormals.
// Since the following float conversion functions are used in non-arithmetic PPC float
// instructions, they must convert floats bitexact and never flush denormals to zero or turn SNaNs
// into QNaNs. This means we can't use CVTSS2SD/CVTSD2SS. The x87 FPU doesn't even support
// flush-to-zero so we can use FLD+FSTP even on denormals.
// If the number is a NaN, make sure to set the QNaN bit back to its original value.
// Another problem is that officially, converting doubles to single format results in undefined
// behavior.
// Relying on undefined behavior is a bug so no software should ever do this.
// In case it does happen, phire's more accurate implementation of ConvertDoubleToSingle() is
// reproduced below.
//#define MORE_ACCURATE_DOUBLETOSINGLE
#ifdef MORE_ACCURATE_DOUBLETOSINGLE
// behavior. Relying on undefined behavior is a bug so no software should ever do this.
// Super Mario 64 (on Wii VC) accidentally relies on this behavior. See issue #11173
alignas(16) static const __m128i double_exponent = _mm_set_epi64x(0, 0x7ff0000000000000);
alignas(16) static const __m128i double_fraction = _mm_set_epi64x(0, 0x000fffffffffffff);
@@ -954,65 +949,6 @@ void EmuCodeBlock::ConvertDoubleToSingle(X64Reg dst, X64Reg src)
MOVDDUP(dst, R(XMM1));
}
#else // MORE_ACCURATE_DOUBLETOSINGLE
alignas(16) static const __m128i double_sign_bit = _mm_set_epi64x(0xffffffffffffffff,
0x7fffffffffffffff);
alignas(16) static const __m128i single_qnan_bit = _mm_set_epi64x(0xffffffffffffffff,
0xffffffffffbfffff);
alignas(16) static const __m128i double_qnan_bit = _mm_set_epi64x(0xffffffffffffffff,
0xfff7ffffffffffff);
// Smallest positive double that results in a normalized single.
alignas(16) static const double min_norm_single = std::numeric_limits<float>::min();
void EmuCodeBlock::ConvertDoubleToSingle(X64Reg dst, X64Reg src)
{
// Most games have flush-to-zero enabled, which causes the single -> double -> single process here
// to be lossy.
// This is a problem when games use float operations to copy non-float data.
// Changing the FPU mode is very expensive, so we can't do that.
// Here, check to see if the source is small enough that it will result in a denormal, and pass it
// to the x87 unit
// if it is.
avx_op(&XEmitter::VPAND, &XEmitter::PAND, XMM0, R(src), MConst(double_sign_bit), true, true);
UCOMISD(XMM0, MConst(min_norm_single));
FixupBranch nanConversion = J_CC(CC_P, true);
FixupBranch denormalConversion = J_CC(CC_B, true);
CVTSD2SS(dst, R(src));
SwitchToFarCode();
SetJumpTarget(nanConversion);
MOVQ_xmm(R(RSCRATCH), src);
// Put the quiet bit into CF.
BT(64, R(RSCRATCH), Imm8(51));
CVTSD2SS(dst, R(src));
FixupBranch continue1 = J_CC(CC_C, true);
// Clear the quiet bit of the SNaN, which was 0 (signalling) but got set to 1 (quiet) by
// conversion.
ANDPS(dst, MConst(single_qnan_bit));
FixupBranch continue2 = J(true);
SetJumpTarget(denormalConversion);
// We're using 8 bytes on the stack
SUB(64, R(RSP), Imm8(8));
MOVSD(MatR(RSP), src);
FLD(64, MatR(RSP));
FSTP(32, MatR(RSP));
MOVSS(dst, MatR(RSP));
ADD(64, R(RSP), Imm8(8));
FixupBranch continue3 = J(true);
SwitchToNearCode();
SetJumpTarget(continue1);
SetJumpTarget(continue2);
SetJumpTarget(continue3);
// We'd normally need to MOVDDUP here to put the single in the top half of the output register
// too, but
// this function is only used to go directly to a following store, so we omit the MOVDDUP here.
}
#endif // MORE_ACCURATE_DOUBLETOSINGLE
// Converting single->double is a bit easier because all single denormals are double normals.
void EmuCodeBlock::ConvertSingleToDouble(X64Reg dst, X64Reg src, bool src_is_gpr)
{

0 comments on commit 2795376

Please sign in to comment.