Ryujinx/ARMeilleure/Instructions/SoftFloat.cs

2740 lines
83 KiB
C#
Raw Normal View History

Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 14:56:22 -04:00
using ARMeilleure.State;
using System;
using System.Diagnostics;
namespace ARMeilleure.Instructions
{
static class SoftFloat
{
static SoftFloat()
{
RecipEstimateTable = BuildRecipEstimateTable();
RecipSqrtEstimateTable = BuildRecipSqrtEstimateTable();
}
internal static readonly byte[] RecipEstimateTable;
internal static readonly byte[] RecipSqrtEstimateTable;
private static byte[] BuildRecipEstimateTable()
{
byte[] tbl = new byte[256];
for (int idx = 0; idx < 256; idx++)
{
uint src = (uint)idx + 256u;
Debug.Assert(256u <= src && src < 512u);
src = (src << 1) + 1u;
uint aux = (1u << 19) / src;
uint dst = (aux + 1u) >> 1;
Debug.Assert(256u <= dst && dst < 512u);
tbl[idx] = (byte)(dst - 256u);
}
return tbl;
}
private static byte[] BuildRecipSqrtEstimateTable()
{
byte[] tbl = new byte[384];
for (int idx = 0; idx < 384; idx++)
{
uint src = (uint)idx + 128u;
Debug.Assert(128u <= src && src < 512u);
if (src < 256u)
{
src = (src << 1) + 1u;
}
else
{
src = (src >> 1) << 1;
src = (src + 1u) << 1;
}
uint aux = 512u;
while (src * (aux + 1u) * (aux + 1u) < (1u << 28))
{
aux = aux + 1u;
}
uint dst = (aux + 1u) >> 1;
Debug.Assert(256u <= dst && dst < 512u);
tbl[idx] = (byte)(dst - 256u);
}
return tbl;
}
}
static class SoftFloat16_32
{
public static float FPConvert(ushort valueBits)
{
ExecutionContext context = NativeInterface.GetContext();
double real = valueBits.FPUnpackCv(out FPType type, out bool sign, context);
float result;
if (type == FPType.SNaN || type == FPType.QNaN)
{
if ((context.Fpcr & FPCR.Dn) != 0)
{
result = FPDefaultNaN();
}
else
{
result = FPConvertNaN(valueBits);
}
if (type == FPType.SNaN)
{
FPProcessException(FPException.InvalidOp, context);
}
}
else if (type == FPType.Infinity)
{
result = FPInfinity(sign);
}
else if (type == FPType.Zero)
{
result = FPZero(sign);
}
else
{
result = FPRoundCv(real, context);
}
return result;
}
private static float FPDefaultNaN()
{
return -float.NaN;
}
private static float FPInfinity(bool sign)
{
return sign ? float.NegativeInfinity : float.PositiveInfinity;
}
private static float FPZero(bool sign)
{
return sign ? -0f : +0f;
}
private static float FPMaxNormal(bool sign)
{
return sign ? float.MinValue : float.MaxValue;
}
private static double FPUnpackCv(
this ushort valueBits,
out FPType type,
out bool sign,
ExecutionContext context)
{
sign = (~(uint)valueBits & 0x8000u) == 0u;
uint exp16 = ((uint)valueBits & 0x7C00u) >> 10;
uint frac16 = (uint)valueBits & 0x03FFu;
double real;
if (exp16 == 0u)
{
if (frac16 == 0u)
{
type = FPType.Zero;
real = 0d;
}
else
{
type = FPType.Nonzero; // Subnormal.
real = Math.Pow(2d, -14) * ((double)frac16 * Math.Pow(2d, -10));
}
}
else if (exp16 == 0x1Fu && (context.Fpcr & FPCR.Ahp) == 0)
{
if (frac16 == 0u)
{
type = FPType.Infinity;
real = Math.Pow(2d, 1000);
}
else
{
type = (~frac16 & 0x0200u) == 0u ? FPType.QNaN : FPType.SNaN;
real = 0d;
}
}
else
{
type = FPType.Nonzero; // Normal.
real = Math.Pow(2d, (int)exp16 - 15) * (1d + (double)frac16 * Math.Pow(2d, -10));
}
return sign ? -real : real;
}
private static float FPRoundCv(double real, ExecutionContext context)
{
const int minimumExp = -126;
const int e = 8;
const int f = 23;
bool sign;
double mantissa;
if (real < 0d)
{
sign = true;
mantissa = -real;
}
else
{
sign = false;
mantissa = real;
}
int exponent = 0;
while (mantissa < 1d)
{
mantissa *= 2d;
exponent--;
}
while (mantissa >= 2d)
{
mantissa /= 2d;
exponent++;
}
if ((context.Fpcr & FPCR.Fz) != 0 && exponent < minimumExp)
{
context.Fpsr |= FPSR.Ufc;
return FPZero(sign);
}
uint biasedExp = (uint)Math.Max(exponent - minimumExp + 1, 0);
if (biasedExp == 0u)
{
mantissa /= Math.Pow(2d, minimumExp - exponent);
}
uint intMant = (uint)Math.Floor(mantissa * Math.Pow(2d, f));
double error = mantissa * Math.Pow(2d, f) - (double)intMant;
if (biasedExp == 0u && (error != 0d || (context.Fpcr & FPCR.Ufe) != 0))
{
FPProcessException(FPException.Underflow, context);
}
bool overflowToInf;
bool roundUp;
switch (context.Fpcr.GetRoundingMode())
{
default:
case FPRoundingMode.ToNearest:
roundUp = (error > 0.5d || (error == 0.5d && (intMant & 1u) == 1u));
overflowToInf = true;
break;
case FPRoundingMode.TowardsPlusInfinity:
roundUp = (error != 0d && !sign);
overflowToInf = !sign;
break;
case FPRoundingMode.TowardsMinusInfinity:
roundUp = (error != 0d && sign);
overflowToInf = sign;
break;
case FPRoundingMode.TowardsZero:
roundUp = false;
overflowToInf = false;
break;
}
if (roundUp)
{
intMant++;
if (intMant == 1u << f)
{
biasedExp = 1u;
}
if (intMant == 1u << (f + 1))
{
biasedExp++;
intMant >>= 1;
}
}
float result;
if (biasedExp >= (1u << e) - 1u)
{
result = overflowToInf ? FPInfinity(sign) : FPMaxNormal(sign);
FPProcessException(FPException.Overflow, context);
error = 1d;
}
else
{
result = BitConverter.Int32BitsToSingle(
(int)((sign ? 1u : 0u) << 31 | (biasedExp & 0xFFu) << 23 | (intMant & 0x007FFFFFu)));
}
if (error != 0d)
{
FPProcessException(FPException.Inexact, context);
}
return result;
}
private static float FPConvertNaN(ushort valueBits)
{
return BitConverter.Int32BitsToSingle(
(int)(((uint)valueBits & 0x8000u) << 16 | 0x7FC00000u | ((uint)valueBits & 0x01FFu) << 13));
}
private static void FPProcessException(FPException exc, ExecutionContext context)
{
int enable = (int)exc + 8;
if ((context.Fpcr & (FPCR)(1 << enable)) != 0)
{
throw new NotImplementedException("Floating-point trap handling.");
}
else
{
context.Fpsr |= (FPSR)(1 << (int)exc);
}
}
}
static class SoftFloat32_16
{
public static ushort FPConvert(float value)
{
ExecutionContext context = NativeInterface.GetContext();
double real = value.FPUnpackCv(out FPType type, out bool sign, out uint valueBits, context);
bool altHp = (context.Fpcr & FPCR.Ahp) != 0;
ushort resultBits;
if (type == FPType.SNaN || type == FPType.QNaN)
{
if (altHp)
{
resultBits = FPZero(sign);
}
else if ((context.Fpcr & FPCR.Dn) != 0)
{
resultBits = FPDefaultNaN();
}
else
{
resultBits = FPConvertNaN(valueBits);
}
if (type == FPType.SNaN || altHp)
{
FPProcessException(FPException.InvalidOp, context);
}
}
else if (type == FPType.Infinity)
{
if (altHp)
{
resultBits = (ushort)((sign ? 1u : 0u) << 15 | 0x7FFFu);
FPProcessException(FPException.InvalidOp, context);
}
else
{
resultBits = FPInfinity(sign);
}
}
else if (type == FPType.Zero)
{
resultBits = FPZero(sign);
}
else
{
resultBits = FPRoundCv(real, context);
}
return resultBits;
}
private static ushort FPDefaultNaN()
{
return (ushort)0x7E00u;
}
private static ushort FPInfinity(bool sign)
{
return sign ? (ushort)0xFC00u : (ushort)0x7C00u;
}
private static ushort FPZero(bool sign)
{
return sign ? (ushort)0x8000u : (ushort)0x0000u;
}
private static ushort FPMaxNormal(bool sign)
{
return sign ? (ushort)0xFBFFu : (ushort)0x7BFFu;
}
private static double FPUnpackCv(
this float value,
out FPType type,
out bool sign,
out uint valueBits,
ExecutionContext context)
{
valueBits = (uint)BitConverter.SingleToInt32Bits(value);
sign = (~valueBits & 0x80000000u) == 0u;
uint exp32 = (valueBits & 0x7F800000u) >> 23;
uint frac32 = valueBits & 0x007FFFFFu;
double real;
if (exp32 == 0u)
{
if (frac32 == 0u || (context.Fpcr & FPCR.Fz) != 0)
{
type = FPType.Zero;
real = 0d;
if (frac32 != 0u)
{
FPProcessException(FPException.InputDenorm, context);
}
}
else
{
type = FPType.Nonzero; // Subnormal.
real = Math.Pow(2d, -126) * ((double)frac32 * Math.Pow(2d, -23));
}
}
else if (exp32 == 0xFFu)
{
if (frac32 == 0u)
{
type = FPType.Infinity;
real = Math.Pow(2d, 1000);
}
else
{
type = (~frac32 & 0x00400000u) == 0u ? FPType.QNaN : FPType.SNaN;
real = 0d;
}
}
else
{
type = FPType.Nonzero; // Normal.
real = Math.Pow(2d, (int)exp32 - 127) * (1d + (double)frac32 * Math.Pow(2d, -23));
}
return sign ? -real : real;
}
private static ushort FPRoundCv(double real, ExecutionContext context)
{
const int minimumExp = -14;
const int e = 5;
const int f = 10;
bool sign;
double mantissa;
if (real < 0d)
{
sign = true;
mantissa = -real;
}
else
{
sign = false;
mantissa = real;
}
int exponent = 0;
while (mantissa < 1d)
{
mantissa *= 2d;
exponent--;
}
while (mantissa >= 2d)
{
mantissa /= 2d;
exponent++;
}
uint biasedExp = (uint)Math.Max(exponent - minimumExp + 1, 0);
if (biasedExp == 0u)
{
mantissa /= Math.Pow(2d, minimumExp - exponent);
}
uint intMant = (uint)Math.Floor(mantissa * Math.Pow(2d, f));
double error = mantissa * Math.Pow(2d, f) - (double)intMant;
if (biasedExp == 0u && (error != 0d || (context.Fpcr & FPCR.Ufe) != 0))
{
FPProcessException(FPException.Underflow, context);
}
bool overflowToInf;
bool roundUp;
switch (context.Fpcr.GetRoundingMode())
{
default:
case FPRoundingMode.ToNearest:
roundUp = (error > 0.5d || (error == 0.5d && (intMant & 1u) == 1u));
overflowToInf = true;
break;
case FPRoundingMode.TowardsPlusInfinity:
roundUp = (error != 0d && !sign);
overflowToInf = !sign;
break;
case FPRoundingMode.TowardsMinusInfinity:
roundUp = (error != 0d && sign);
overflowToInf = sign;
break;
case FPRoundingMode.TowardsZero:
roundUp = false;
overflowToInf = false;
break;
}
if (roundUp)
{
intMant++;
if (intMant == 1u << f)
{
biasedExp = 1u;
}
if (intMant == 1u << (f + 1))
{
biasedExp++;
intMant >>= 1;
}
}
ushort resultBits;
if ((context.Fpcr & FPCR.Ahp) == 0)
{
if (biasedExp >= (1u << e) - 1u)
{
resultBits = overflowToInf ? FPInfinity(sign) : FPMaxNormal(sign);
FPProcessException(FPException.Overflow, context);
error = 1d;
}
else
{
resultBits = (ushort)((sign ? 1u : 0u) << 15 | (biasedExp & 0x1Fu) << 10 | (intMant & 0x03FFu));
}
}
else
{
if (biasedExp >= 1u << e)
{
resultBits = (ushort)((sign ? 1u : 0u) << 15 | 0x7FFFu);
FPProcessException(FPException.InvalidOp, context);
error = 0d;
}
else
{
resultBits = (ushort)((sign ? 1u : 0u) << 15 | (biasedExp & 0x1Fu) << 10 | (intMant & 0x03FFu));
}
}
if (error != 0d)
{
FPProcessException(FPException.Inexact, context);
}
return resultBits;
}
private static ushort FPConvertNaN(uint valueBits)
{
return (ushort)((valueBits & 0x80000000u) >> 16 | 0x7E00u | (valueBits & 0x003FE000u) >> 13);
}
private static void FPProcessException(FPException exc, ExecutionContext context)
{
int enable = (int)exc + 8;
if ((context.Fpcr & (FPCR)(1 << enable)) != 0)
{
throw new NotImplementedException("Floating-point trap handling.");
}
else
{
context.Fpsr |= (FPSR)(1 << (int)exc);
}
}
}
static class SoftFloat32
{
public static float FPAdd(float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out uint op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out uint op2, context);
float result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
if (inf1 && inf2 && sign1 == !sign2)
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
else if ((inf1 && !sign1) || (inf2 && !sign2))
{
result = FPInfinity(false);
}
else if ((inf1 && sign1) || (inf2 && sign2))
{
result = FPInfinity(true);
}
else if (zero1 && zero2 && sign1 == sign2)
{
result = FPZero(sign1);
}
else
{
result = value1 + value2;
if ((context.Fpcr & FPCR.Fz) != 0 && float.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0f);
}
}
}
return result;
}
public static int FPCompare(float value1, float value2, bool signalNaNs)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out _, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out _, context);
int result;
if (type1 == FPType.SNaN || type1 == FPType.QNaN || type2 == FPType.SNaN || type2 == FPType.QNaN)
{
result = 0b0011;
if (type1 == FPType.SNaN || type2 == FPType.SNaN || signalNaNs)
{
FPProcessException(FPException.InvalidOp, context);
}
}
else
{
if (value1 == value2)
{
result = 0b0110;
}
else if (value1 < value2)
{
result = 0b1000;
}
else
{
result = 0b0010;
}
}
return result;
}
public static float FPCompareEQ(float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out _, out _, context);
value2 = value2.FPUnpack(out FPType type2, out _, out _, context);
float result;
if (type1 == FPType.SNaN || type1 == FPType.QNaN || type2 == FPType.SNaN || type2 == FPType.QNaN)
{
result = ZerosOrOnes(false);
if (type1 == FPType.SNaN || type2 == FPType.SNaN)
{
FPProcessException(FPException.InvalidOp, context);
}
}
else
{
result = ZerosOrOnes(value1 == value2);
}
return result;
}
public static float FPCompareGE(float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out _, out _, context);
value2 = value2.FPUnpack(out FPType type2, out _, out _, context);
float result;
if (type1 == FPType.SNaN || type1 == FPType.QNaN || type2 == FPType.SNaN || type2 == FPType.QNaN)
{
result = ZerosOrOnes(false);
FPProcessException(FPException.InvalidOp, context);
}
else
{
result = ZerosOrOnes(value1 >= value2);
}
return result;
}
public static float FPCompareGT(float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out _, out _, context);
value2 = value2.FPUnpack(out FPType type2, out _, out _, context);
float result;
if (type1 == FPType.SNaN || type1 == FPType.QNaN || type2 == FPType.SNaN || type2 == FPType.QNaN)
{
result = ZerosOrOnes(false);
FPProcessException(FPException.InvalidOp, context);
}
else
{
result = ZerosOrOnes(value1 > value2);
}
return result;
}
public static float FPCompareLE(float value1, float value2)
{
return FPCompareGE(value2, value1);
}
public static float FPCompareLT(float value1, float value2)
{
return FPCompareGT(value2, value1);
}
public static float FPDiv(float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out uint op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out uint op2, context);
float result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
if ((inf1 && inf2) || (zero1 && zero2))
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
else if (inf1 || zero2)
{
result = FPInfinity(sign1 ^ sign2);
if (!inf1)
{
FPProcessException(FPException.DivideByZero, context);
}
}
else if (zero1 || inf2)
{
result = FPZero(sign1 ^ sign2);
}
else
{
result = value1 / value2;
if ((context.Fpcr & FPCR.Fz) != 0 && float.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0f);
}
}
}
return result;
}
public static float FPMax(float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out uint op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out uint op2, context);
float result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
if (value1 > value2)
{
if (type1 == FPType.Infinity)
{
result = FPInfinity(sign1);
}
else if (type1 == FPType.Zero)
{
result = FPZero(sign1 && sign2);
}
else
{
result = value1;
}
}
else
{
if (type2 == FPType.Infinity)
{
result = FPInfinity(sign2);
}
else if (type2 == FPType.Zero)
{
result = FPZero(sign1 && sign2);
}
else
{
result = value2;
if ((context.Fpcr & FPCR.Fz) != 0 && float.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0f);
}
}
}
}
return result;
}
public static float FPMaxNum(float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1.FPUnpack(out FPType type1, out _, out _, context);
value2.FPUnpack(out FPType type2, out _, out _, context);
if (type1 == FPType.QNaN && type2 != FPType.QNaN)
{
value1 = FPInfinity(true);
}
else if (type1 != FPType.QNaN && type2 == FPType.QNaN)
{
value2 = FPInfinity(true);
}
return FPMax(value1, value2);
}
public static float FPMin(float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out uint op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out uint op2, context);
float result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
if (value1 < value2)
{
if (type1 == FPType.Infinity)
{
result = FPInfinity(sign1);
}
else if (type1 == FPType.Zero)
{
result = FPZero(sign1 || sign2);
}
else
{
result = value1;
}
}
else
{
if (type2 == FPType.Infinity)
{
result = FPInfinity(sign2);
}
else if (type2 == FPType.Zero)
{
result = FPZero(sign1 || sign2);
}
else
{
result = value2;
if ((context.Fpcr & FPCR.Fz) != 0 && float.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0f);
}
}
}
}
return result;
}
public static float FPMinNum(float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1.FPUnpack(out FPType type1, out _, out _, context);
value2.FPUnpack(out FPType type2, out _, out _, context);
if (type1 == FPType.QNaN && type2 != FPType.QNaN)
{
value1 = FPInfinity(false);
}
else if (type1 != FPType.QNaN && type2 == FPType.QNaN)
{
value2 = FPInfinity(false);
}
return FPMin(value1, value2);
}
public static float FPMul(float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out uint op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out uint op2, context);
float result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
if ((inf1 && zero2) || (zero1 && inf2))
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
else if (inf1 || inf2)
{
result = FPInfinity(sign1 ^ sign2);
}
else if (zero1 || zero2)
{
result = FPZero(sign1 ^ sign2);
}
else
{
result = value1 * value2;
if ((context.Fpcr & FPCR.Fz) != 0 && float.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0f);
}
}
}
return result;
}
public static float FPMulAdd(float valueA, float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
valueA = valueA.FPUnpack(out FPType typeA, out bool signA, out uint addend, context);
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out uint op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out uint op2, context);
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
float result = FPProcessNaNs3(typeA, type1, type2, addend, op1, op2, out bool done, context);
if (typeA == FPType.QNaN && ((inf1 && zero2) || (zero1 && inf2)))
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
if (!done)
{
bool infA = typeA == FPType.Infinity; bool zeroA = typeA == FPType.Zero;
bool signP = sign1 ^ sign2;
bool infP = inf1 || inf2;
bool zeroP = zero1 || zero2;
if ((inf1 && zero2) || (zero1 && inf2) || (infA && infP && signA != signP))
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
else if ((infA && !signA) || (infP && !signP))
{
result = FPInfinity(false);
}
else if ((infA && signA) || (infP && signP))
{
result = FPInfinity(true);
}
else if (zeroA && zeroP && signA == signP)
{
result = FPZero(signA);
}
else
{
result = MathF.FusedMultiplyAdd(value1, value2, valueA);
Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 14:56:22 -04:00
if ((context.Fpcr & FPCR.Fz) != 0 && float.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0f);
}
}
}
return result;
}
public static float FPMulSub(float valueA, float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPNeg();
return FPMulAdd(valueA, value1, value2);
}
public static float FPMulX(float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out uint op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out uint op2, context);
float result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
if ((inf1 && zero2) || (zero1 && inf2))
{
result = FPTwo(sign1 ^ sign2);
}
else if (inf1 || inf2)
{
result = FPInfinity(sign1 ^ sign2);
}
else if (zero1 || zero2)
{
result = FPZero(sign1 ^ sign2);
}
else
{
result = value1 * value2;
if ((context.Fpcr & FPCR.Fz) != 0 && float.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0f);
}
}
}
return result;
}
public static float FPRecipEstimate(float value)
{
ExecutionContext context = NativeInterface.GetContext();
value.FPUnpack(out FPType type, out bool sign, out uint op, context);
float result;
if (type == FPType.SNaN || type == FPType.QNaN)
{
result = FPProcessNaN(type, op, context);
}
else if (type == FPType.Infinity)
{
result = FPZero(sign);
}
else if (type == FPType.Zero)
{
result = FPInfinity(sign);
FPProcessException(FPException.DivideByZero, context);
}
else if (MathF.Abs(value) < MathF.Pow(2f, -128))
{
bool overflowToInf;
switch (context.Fpcr.GetRoundingMode())
{
default:
case FPRoundingMode.ToNearest: overflowToInf = true; break;
case FPRoundingMode.TowardsPlusInfinity: overflowToInf = !sign; break;
case FPRoundingMode.TowardsMinusInfinity: overflowToInf = sign; break;
case FPRoundingMode.TowardsZero: overflowToInf = false; break;
}
result = overflowToInf ? FPInfinity(sign) : FPMaxNormal(sign);
FPProcessException(FPException.Overflow, context);
FPProcessException(FPException.Inexact, context);
}
else if ((context.Fpcr & FPCR.Fz) != 0 && (MathF.Abs(value) >= MathF.Pow(2f, 126)))
{
result = FPZero(sign);
context.Fpsr |= FPSR.Ufc;
}
else
{
ulong fraction = (ulong)(op & 0x007FFFFFu) << 29;
uint exp = (op & 0x7F800000u) >> 23;
if (exp == 0u)
{
if ((fraction & 0x0008000000000000ul) == 0ul)
{
fraction = (fraction & 0x0003FFFFFFFFFFFFul) << 2;
exp -= 1u;
}
else
{
fraction = (fraction & 0x0007FFFFFFFFFFFFul) << 1;
}
}
uint scaled = (uint)(((fraction & 0x000FF00000000000ul) | 0x0010000000000000ul) >> 44);
uint resultExp = 253u - exp;
uint estimate = (uint)SoftFloat.RecipEstimateTable[scaled - 256u] + 256u;
fraction = (ulong)(estimate & 0xFFu) << 44;
if (resultExp == 0u)
{
fraction = ((fraction & 0x000FFFFFFFFFFFFEul) | 0x0010000000000000ul) >> 1;
}
else if (resultExp + 1u == 0u)
{
fraction = ((fraction & 0x000FFFFFFFFFFFFCul) | 0x0010000000000000ul) >> 2;
resultExp = 0u;
}
result = BitConverter.Int32BitsToSingle(
(int)((sign ? 1u : 0u) << 31 | (resultExp & 0xFFu) << 23 | (uint)(fraction >> 29) & 0x007FFFFFu));
}
return result;
}
public static float FPRecipStepFused(float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPNeg();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out uint op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out uint op2, context);
float result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
if ((inf1 && zero2) || (zero1 && inf2))
{
result = FPTwo(false);
}
else if (inf1 || inf2)
{
result = FPInfinity(sign1 ^ sign2);
}
else
{
result = MathF.FusedMultiplyAdd(value1, value2, 2f);
Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 14:56:22 -04:00
if ((context.Fpcr & FPCR.Fz) != 0 && float.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0f);
}
}
}
return result;
}
public static float FPRecpX(float value)
{
ExecutionContext context = NativeInterface.GetContext();
value.FPUnpack(out FPType type, out bool sign, out uint op, context);
float result;
if (type == FPType.SNaN || type == FPType.QNaN)
{
result = FPProcessNaN(type, op, context);
}
else
{
uint notExp = (~op >> 23) & 0xFFu;
uint maxExp = 0xFEu;
result = BitConverter.Int32BitsToSingle(
(int)((sign ? 1u : 0u) << 31 | (notExp == 0xFFu ? maxExp : notExp) << 23));
}
return result;
}
public static float FPRSqrtEstimate(float value)
{
ExecutionContext context = NativeInterface.GetContext();
value.FPUnpack(out FPType type, out bool sign, out uint op, context);
float result;
if (type == FPType.SNaN || type == FPType.QNaN)
{
result = FPProcessNaN(type, op, context);
}
else if (type == FPType.Zero)
{
result = FPInfinity(sign);
FPProcessException(FPException.DivideByZero, context);
}
else if (sign)
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
else if (type == FPType.Infinity)
{
result = FPZero(false);
}
else
{
ulong fraction = (ulong)(op & 0x007FFFFFu) << 29;
uint exp = (op & 0x7F800000u) >> 23;
if (exp == 0u)
{
while ((fraction & 0x0008000000000000ul) == 0ul)
{
fraction = (fraction & 0x0007FFFFFFFFFFFFul) << 1;
exp -= 1u;
}
fraction = (fraction & 0x0007FFFFFFFFFFFFul) << 1;
}
uint scaled;
if ((exp & 1u) == 0u)
{
scaled = (uint)(((fraction & 0x000FF00000000000ul) | 0x0010000000000000ul) >> 44);
}
else
{
scaled = (uint)(((fraction & 0x000FE00000000000ul) | 0x0010000000000000ul) >> 45);
}
uint resultExp = (380u - exp) >> 1;
uint estimate = (uint)SoftFloat.RecipSqrtEstimateTable[scaled - 128u] + 256u;
result = BitConverter.Int32BitsToSingle((int)((resultExp & 0xFFu) << 23 | (estimate & 0xFFu) << 15));
}
return result;
}
public static float FPRSqrtStepFused(float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPNeg();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out uint op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out uint op2, context);
float result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
if ((inf1 && zero2) || (zero1 && inf2))
{
result = FPOnePointFive(false);
}
else if (inf1 || inf2)
{
result = FPInfinity(sign1 ^ sign2);
}
else
{
result = MathF.FusedMultiplyAdd(value1, value2, 3f) / 2f;
Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 14:56:22 -04:00
if ((context.Fpcr & FPCR.Fz) != 0 && float.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0f);
}
}
}
return result;
}
public static float FPSqrt(float value)
{
ExecutionContext context = NativeInterface.GetContext();
value = value.FPUnpack(out FPType type, out bool sign, out uint op, context);
float result;
if (type == FPType.SNaN || type == FPType.QNaN)
{
result = FPProcessNaN(type, op, context);
}
else if (type == FPType.Zero)
{
result = FPZero(sign);
}
else if (type == FPType.Infinity && !sign)
{
result = FPInfinity(sign);
}
else if (sign)
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
else
{
result = MathF.Sqrt(value);
if ((context.Fpcr & FPCR.Fz) != 0 && float.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0f);
}
}
return result;
}
public static float FPSub(float value1, float value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out uint op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out uint op2, context);
float result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
if (inf1 && inf2 && sign1 == sign2)
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
else if ((inf1 && !sign1) || (inf2 && sign2))
{
result = FPInfinity(false);
}
else if ((inf1 && sign1) || (inf2 && !sign2))
{
result = FPInfinity(true);
}
else if (zero1 && zero2 && sign1 == !sign2)
{
result = FPZero(sign1);
}
else
{
result = value1 - value2;
if ((context.Fpcr & FPCR.Fz) != 0 && float.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0f);
}
}
}
return result;
}
private static float FPDefaultNaN()
{
return -float.NaN;
}
private static float FPInfinity(bool sign)
{
return sign ? float.NegativeInfinity : float.PositiveInfinity;
}
private static float FPZero(bool sign)
{
return sign ? -0f : +0f;
}
private static float FPMaxNormal(bool sign)
{
return sign ? float.MinValue : float.MaxValue;
}
private static float FPTwo(bool sign)
{
return sign ? -2f : +2f;
}
private static float FPOnePointFive(bool sign)
{
return sign ? -1.5f : +1.5f;
}
private static float FPNeg(this float value)
{
return -value;
}
private static float ZerosOrOnes(bool ones)
{
return BitConverter.Int32BitsToSingle(ones ? -1 : 0);
}
private static float FPUnpack(
this float value,
out FPType type,
out bool sign,
out uint valueBits,
ExecutionContext context)
{
valueBits = (uint)BitConverter.SingleToInt32Bits(value);
sign = (~valueBits & 0x80000000u) == 0u;
if ((valueBits & 0x7F800000u) == 0u)
{
if ((valueBits & 0x007FFFFFu) == 0u || (context.Fpcr & FPCR.Fz) != 0)
{
type = FPType.Zero;
value = FPZero(sign);
if ((valueBits & 0x007FFFFFu) != 0u)
{
FPProcessException(FPException.InputDenorm, context);
}
}
else
{
type = FPType.Nonzero;
}
}
else if ((~valueBits & 0x7F800000u) == 0u)
{
if ((valueBits & 0x007FFFFFu) == 0u)
{
type = FPType.Infinity;
}
else
{
type = (~valueBits & 0x00400000u) == 0u ? FPType.QNaN : FPType.SNaN;
value = FPZero(sign);
}
}
else
{
type = FPType.Nonzero;
}
return value;
}
private static float FPProcessNaNs(
FPType type1,
FPType type2,
uint op1,
uint op2,
out bool done,
ExecutionContext context)
{
done = true;
if (type1 == FPType.SNaN)
{
return FPProcessNaN(type1, op1, context);
}
else if (type2 == FPType.SNaN)
{
return FPProcessNaN(type2, op2, context);
}
else if (type1 == FPType.QNaN)
{
return FPProcessNaN(type1, op1, context);
}
else if (type2 == FPType.QNaN)
{
return FPProcessNaN(type2, op2, context);
}
done = false;
return FPZero(false);
}
private static float FPProcessNaNs3(
FPType type1,
FPType type2,
FPType type3,
uint op1,
uint op2,
uint op3,
out bool done,
ExecutionContext context)
{
done = true;
if (type1 == FPType.SNaN)
{
return FPProcessNaN(type1, op1, context);
}
else if (type2 == FPType.SNaN)
{
return FPProcessNaN(type2, op2, context);
}
else if (type3 == FPType.SNaN)
{
return FPProcessNaN(type3, op3, context);
}
else if (type1 == FPType.QNaN)
{
return FPProcessNaN(type1, op1, context);
}
else if (type2 == FPType.QNaN)
{
return FPProcessNaN(type2, op2, context);
}
else if (type3 == FPType.QNaN)
{
return FPProcessNaN(type3, op3, context);
}
done = false;
return FPZero(false);
}
private static float FPProcessNaN(FPType type, uint op, ExecutionContext context)
{
if (type == FPType.SNaN)
{
op |= 1u << 22;
FPProcessException(FPException.InvalidOp, context);
}
if ((context.Fpcr & FPCR.Dn) != 0)
{
return FPDefaultNaN();
}
return BitConverter.Int32BitsToSingle((int)op);
}
private static void FPProcessException(FPException exc, ExecutionContext context)
{
int enable = (int)exc + 8;
if ((context.Fpcr & (FPCR)(1 << enable)) != 0)
{
throw new NotImplementedException("Floating-point trap handling.");
}
else
{
context.Fpsr |= (FPSR)(1 << (int)exc);
}
}
}
static class SoftFloat64
{
public static double FPAdd(double value1, double value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out ulong op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out ulong op2, context);
double result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
if (inf1 && inf2 && sign1 == !sign2)
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
else if ((inf1 && !sign1) || (inf2 && !sign2))
{
result = FPInfinity(false);
}
else if ((inf1 && sign1) || (inf2 && sign2))
{
result = FPInfinity(true);
}
else if (zero1 && zero2 && sign1 == sign2)
{
result = FPZero(sign1);
}
else
{
result = value1 + value2;
if ((context.Fpcr & FPCR.Fz) != 0 && double.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0d);
}
}
}
return result;
}
public static int FPCompare(double value1, double value2, bool signalNaNs)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out _, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out _, context);
int result;
if (type1 == FPType.SNaN || type1 == FPType.QNaN || type2 == FPType.SNaN || type2 == FPType.QNaN)
{
result = 0b0011;
if (type1 == FPType.SNaN || type2 == FPType.SNaN || signalNaNs)
{
FPProcessException(FPException.InvalidOp, context);
}
}
else
{
if (value1 == value2)
{
result = 0b0110;
}
else if (value1 < value2)
{
result = 0b1000;
}
else
{
result = 0b0010;
}
}
return result;
}
public static double FPCompareEQ(double value1, double value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out _, out _, context);
value2 = value2.FPUnpack(out FPType type2, out _, out _, context);
double result;
if (type1 == FPType.SNaN || type1 == FPType.QNaN || type2 == FPType.SNaN || type2 == FPType.QNaN)
{
result = ZerosOrOnes(false);
if (type1 == FPType.SNaN || type2 == FPType.SNaN)
{
FPProcessException(FPException.InvalidOp, context);
}
}
else
{
result = ZerosOrOnes(value1 == value2);
}
return result;
}
public static double FPCompareGE(double value1, double value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out _, out _, context);
value2 = value2.FPUnpack(out FPType type2, out _, out _, context);
double result;
if (type1 == FPType.SNaN || type1 == FPType.QNaN || type2 == FPType.SNaN || type2 == FPType.QNaN)
{
result = ZerosOrOnes(false);
FPProcessException(FPException.InvalidOp, context);
}
else
{
result = ZerosOrOnes(value1 >= value2);
}
return result;
}
public static double FPCompareGT(double value1, double value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out _, out _, context);
value2 = value2.FPUnpack(out FPType type2, out _, out _, context);
double result;
if (type1 == FPType.SNaN || type1 == FPType.QNaN || type2 == FPType.SNaN || type2 == FPType.QNaN)
{
result = ZerosOrOnes(false);
FPProcessException(FPException.InvalidOp, context);
}
else
{
result = ZerosOrOnes(value1 > value2);
}
return result;
}
public static double FPCompareLE(double value1, double value2)
{
return FPCompareGE(value2, value1);
}
public static double FPCompareLT(double value1, double value2)
{
return FPCompareGT(value2, value1);
}
public static double FPDiv(double value1, double value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out ulong op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out ulong op2, context);
double result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
if ((inf1 && inf2) || (zero1 && zero2))
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
else if (inf1 || zero2)
{
result = FPInfinity(sign1 ^ sign2);
if (!inf1)
{
FPProcessException(FPException.DivideByZero, context);
}
}
else if (zero1 || inf2)
{
result = FPZero(sign1 ^ sign2);
}
else
{
result = value1 / value2;
if ((context.Fpcr & FPCR.Fz) != 0 && double.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0d);
}
}
}
return result;
}
public static double FPMax(double value1, double value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out ulong op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out ulong op2, context);
double result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
if (value1 > value2)
{
if (type1 == FPType.Infinity)
{
result = FPInfinity(sign1);
}
else if (type1 == FPType.Zero)
{
result = FPZero(sign1 && sign2);
}
else
{
result = value1;
}
}
else
{
if (type2 == FPType.Infinity)
{
result = FPInfinity(sign2);
}
else if (type2 == FPType.Zero)
{
result = FPZero(sign1 && sign2);
}
else
{
result = value2;
if ((context.Fpcr & FPCR.Fz) != 0 && double.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0d);
}
}
}
}
return result;
}
public static double FPMaxNum(double value1, double value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1.FPUnpack(out FPType type1, out _, out _, context);
value2.FPUnpack(out FPType type2, out _, out _, context);
if (type1 == FPType.QNaN && type2 != FPType.QNaN)
{
value1 = FPInfinity(true);
}
else if (type1 != FPType.QNaN && type2 == FPType.QNaN)
{
value2 = FPInfinity(true);
}
return FPMax(value1, value2);
}
public static double FPMin(double value1, double value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out ulong op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out ulong op2, context);
double result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
if (value1 < value2)
{
if (type1 == FPType.Infinity)
{
result = FPInfinity(sign1);
}
else if (type1 == FPType.Zero)
{
result = FPZero(sign1 || sign2);
}
else
{
result = value1;
}
}
else
{
if (type2 == FPType.Infinity)
{
result = FPInfinity(sign2);
}
else if (type2 == FPType.Zero)
{
result = FPZero(sign1 || sign2);
}
else
{
result = value2;
if ((context.Fpcr & FPCR.Fz) != 0 && double.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0d);
}
}
}
}
return result;
}
public static double FPMinNum(double value1, double value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1.FPUnpack(out FPType type1, out _, out _, context);
value2.FPUnpack(out FPType type2, out _, out _, context);
if (type1 == FPType.QNaN && type2 != FPType.QNaN)
{
value1 = FPInfinity(false);
}
else if (type1 != FPType.QNaN && type2 == FPType.QNaN)
{
value2 = FPInfinity(false);
}
return FPMin(value1, value2);
}
public static double FPMul(double value1, double value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out ulong op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out ulong op2, context);
double result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
if ((inf1 && zero2) || (zero1 && inf2))
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
else if (inf1 || inf2)
{
result = FPInfinity(sign1 ^ sign2);
}
else if (zero1 || zero2)
{
result = FPZero(sign1 ^ sign2);
}
else
{
result = value1 * value2;
if ((context.Fpcr & FPCR.Fz) != 0 && double.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0d);
}
}
}
return result;
}
public static double FPMulAdd(double valueA, double value1, double value2)
{
ExecutionContext context = NativeInterface.GetContext();
valueA = valueA.FPUnpack(out FPType typeA, out bool signA, out ulong addend, context);
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out ulong op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out ulong op2, context);
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
double result = FPProcessNaNs3(typeA, type1, type2, addend, op1, op2, out bool done, context);
if (typeA == FPType.QNaN && ((inf1 && zero2) || (zero1 && inf2)))
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
if (!done)
{
bool infA = typeA == FPType.Infinity; bool zeroA = typeA == FPType.Zero;
bool signP = sign1 ^ sign2;
bool infP = inf1 || inf2;
bool zeroP = zero1 || zero2;
if ((inf1 && zero2) || (zero1 && inf2) || (infA && infP && signA != signP))
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
else if ((infA && !signA) || (infP && !signP))
{
result = FPInfinity(false);
}
else if ((infA && signA) || (infP && signP))
{
result = FPInfinity(true);
}
else if (zeroA && zeroP && signA == signP)
{
result = FPZero(signA);
}
else
{
result = Math.FusedMultiplyAdd(value1, value2, valueA);
Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 14:56:22 -04:00
if ((context.Fpcr & FPCR.Fz) != 0 && double.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0d);
}
}
}
return result;
}
public static double FPMulSub(double valueA, double value1, double value2)
{
value1 = value1.FPNeg();
return FPMulAdd(valueA, value1, value2);
}
public static double FPMulX(double value1, double value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out ulong op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out ulong op2, context);
double result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
if ((inf1 && zero2) || (zero1 && inf2))
{
result = FPTwo(sign1 ^ sign2);
}
else if (inf1 || inf2)
{
result = FPInfinity(sign1 ^ sign2);
}
else if (zero1 || zero2)
{
result = FPZero(sign1 ^ sign2);
}
else
{
result = value1 * value2;
if ((context.Fpcr & FPCR.Fz) != 0 && double.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0d);
}
}
}
return result;
}
public static double FPRecipEstimate(double value)
{
ExecutionContext context = NativeInterface.GetContext();
value.FPUnpack(out FPType type, out bool sign, out ulong op, context);
double result;
if (type == FPType.SNaN || type == FPType.QNaN)
{
result = FPProcessNaN(type, op, context);
}
else if (type == FPType.Infinity)
{
result = FPZero(sign);
}
else if (type == FPType.Zero)
{
result = FPInfinity(sign);
FPProcessException(FPException.DivideByZero, context);
}
else if (Math.Abs(value) < Math.Pow(2d, -1024))
{
bool overflowToInf;
switch (context.Fpcr.GetRoundingMode())
{
default:
case FPRoundingMode.ToNearest: overflowToInf = true; break;
case FPRoundingMode.TowardsPlusInfinity: overflowToInf = !sign; break;
case FPRoundingMode.TowardsMinusInfinity: overflowToInf = sign; break;
case FPRoundingMode.TowardsZero: overflowToInf = false; break;
}
result = overflowToInf ? FPInfinity(sign) : FPMaxNormal(sign);
FPProcessException(FPException.Overflow, context);
FPProcessException(FPException.Inexact, context);
}
else if ((context.Fpcr & FPCR.Fz) != 0 && (Math.Abs(value) >= Math.Pow(2d, 1022)))
{
result = FPZero(sign);
context.Fpsr |= FPSR.Ufc;
}
else
{
ulong fraction = op & 0x000FFFFFFFFFFFFFul;
uint exp = (uint)((op & 0x7FF0000000000000ul) >> 52);
if (exp == 0u)
{
if ((fraction & 0x0008000000000000ul) == 0ul)
{
fraction = (fraction & 0x0003FFFFFFFFFFFFul) << 2;
exp -= 1u;
}
else
{
fraction = (fraction & 0x0007FFFFFFFFFFFFul) << 1;
}
}
uint scaled = (uint)(((fraction & 0x000FF00000000000ul) | 0x0010000000000000ul) >> 44);
uint resultExp = 2045u - exp;
uint estimate = (uint)SoftFloat.RecipEstimateTable[scaled - 256u] + 256u;
fraction = (ulong)(estimate & 0xFFu) << 44;
if (resultExp == 0u)
{
fraction = ((fraction & 0x000FFFFFFFFFFFFEul) | 0x0010000000000000ul) >> 1;
}
else if (resultExp + 1u == 0u)
{
fraction = ((fraction & 0x000FFFFFFFFFFFFCul) | 0x0010000000000000ul) >> 2;
resultExp = 0u;
}
result = BitConverter.Int64BitsToDouble(
(long)((sign ? 1ul : 0ul) << 63 | (resultExp & 0x7FFul) << 52 | (fraction & 0x000FFFFFFFFFFFFFul)));
}
return result;
}
public static double FPRecipStepFused(double value1, double value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPNeg();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out ulong op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out ulong op2, context);
double result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
if ((inf1 && zero2) || (zero1 && inf2))
{
result = FPTwo(false);
}
else if (inf1 || inf2)
{
result = FPInfinity(sign1 ^ sign2);
}
else
{
result = Math.FusedMultiplyAdd(value1, value2, 2d);
Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 14:56:22 -04:00
if ((context.Fpcr & FPCR.Fz) != 0 && double.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0d);
}
}
}
return result;
}
public static double FPRecpX(double value)
{
ExecutionContext context = NativeInterface.GetContext();
value.FPUnpack(out FPType type, out bool sign, out ulong op, context);
double result;
if (type == FPType.SNaN || type == FPType.QNaN)
{
result = FPProcessNaN(type, op, context);
}
else
{
ulong notExp = (~op >> 52) & 0x7FFul;
ulong maxExp = 0x7FEul;
result = BitConverter.Int64BitsToDouble(
(long)((sign ? 1ul : 0ul) << 63 | (notExp == 0x7FFul ? maxExp : notExp) << 52));
}
return result;
}
public static double FPRSqrtEstimate(double value)
{
ExecutionContext context = NativeInterface.GetContext();
value.FPUnpack(out FPType type, out bool sign, out ulong op, context);
double result;
if (type == FPType.SNaN || type == FPType.QNaN)
{
result = FPProcessNaN(type, op, context);
}
else if (type == FPType.Zero)
{
result = FPInfinity(sign);
FPProcessException(FPException.DivideByZero, context);
}
else if (sign)
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
else if (type == FPType.Infinity)
{
result = FPZero(false);
}
else
{
ulong fraction = op & 0x000FFFFFFFFFFFFFul;
uint exp = (uint)((op & 0x7FF0000000000000ul) >> 52);
if (exp == 0u)
{
while ((fraction & 0x0008000000000000ul) == 0ul)
{
fraction = (fraction & 0x0007FFFFFFFFFFFFul) << 1;
exp -= 1u;
}
fraction = (fraction & 0x0007FFFFFFFFFFFFul) << 1;
}
uint scaled;
if ((exp & 1u) == 0u)
{
scaled = (uint)(((fraction & 0x000FF00000000000ul) | 0x0010000000000000ul) >> 44);
}
else
{
scaled = (uint)(((fraction & 0x000FE00000000000ul) | 0x0010000000000000ul) >> 45);
}
uint resultExp = (3068u - exp) >> 1;
uint estimate = (uint)SoftFloat.RecipSqrtEstimateTable[scaled - 128u] + 256u;
result = BitConverter.Int64BitsToDouble((long)((resultExp & 0x7FFul) << 52 | (estimate & 0xFFul) << 44));
}
return result;
}
public static double FPRSqrtStepFused(double value1, double value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPNeg();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out ulong op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out ulong op2, context);
double result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
if ((inf1 && zero2) || (zero1 && inf2))
{
result = FPOnePointFive(false);
}
else if (inf1 || inf2)
{
result = FPInfinity(sign1 ^ sign2);
}
else
{
result = Math.FusedMultiplyAdd(value1, value2, 3d) / 2d;
Add a new JIT compiler for CPU code (#693) * Start of the ARMeilleure project * Refactoring around the old IRAdapter, now renamed to PreAllocator * Optimize the LowestBitSet method * Add CLZ support and fix CLS implementation * Add missing Equals and GetHashCode overrides on some structs, misc small tweaks * Implement the ByteSwap IR instruction, and some refactoring on the assembler * Implement the DivideUI IR instruction and fix 64-bits IDIV * Correct constant operand type on CSINC * Move division instructions implementation to InstEmitDiv * Fix destination type for the ConditionalSelect IR instruction * Implement UMULH and SMULH, with new IR instructions * Fix some issues with shift instructions * Fix constant types for BFM instructions * Fix up new tests using the new V128 struct * Update tests * Move DIV tests to a separate file * Add support for calls, and some instructions that depends on them * Start adding support for SIMD & FP types, along with some of the related ARM instructions * Fix some typos and the divide instruction with FP operands * Fix wrong method call on Clz_V * Implement ARM FP & SIMD move instructions, Saddlv_V, and misc. fixes * Implement SIMD logical instructions and more misc. fixes * Fix PSRAD x86 instruction encoding, TRN, UABD and UABDL implementations * Implement float conversion instruction, merge in LDj3SNuD fixes, and some other misc. fixes * Implement SIMD shift instruction and fix Dup_V * Add SCVTF and UCVTF (vector, fixed-point) variants to the opcode table * Fix check with tolerance on tester * Implement FP & SIMD comparison instructions, and some fixes * Update FCVT (Scalar) encoding on the table to support the Half-float variants * Support passing V128 structs, some cleanup on the register allocator, merge LDj3SNuD fixes * Use old memory access methods, made a start on SIMD memory insts support, some fixes * Fix float constant passed to functions, save and restore non-volatile XMM registers, other fixes * Fix arguments count with struct return values, other fixes * More instructions * Misc. fixes and integrate LDj3SNuD fixes * Update tests * Add a faster linear scan allocator, unwinding support on windows, and other changes * Update Ryujinx.HLE * Update Ryujinx.Graphics * Fix V128 return pointer passing, RCX is clobbered * Update Ryujinx.Tests * Update ITimeZoneService * Stop using GetFunctionPointer as that can't be called from native code, misc. fixes and tweaks * Use generic GetFunctionPointerForDelegate method and other tweaks * Some refactoring on the code generator, assert on invalid operations and use a separate enum for intrinsics * Remove some unused code on the assembler * Fix REX.W prefix regression on float conversion instructions, add some sort of profiler * Add hardware capability detection * Fix regression on Sha1h and revert Fcm** changes * Add SSE2-only paths on vector extract and insert, some refactoring on the pre-allocator * Fix silly mistake introduced on last commit on CpuId * Generate inline stack probes when the stack allocation is too large * Initial support for the System-V ABI * Support multiple destination operands * Fix SSE2 VectorInsert8 path, and other fixes * Change placement of XMM callee save and restore code to match other compilers * Rename Dest to Destination and Inst to Instruction * Fix a regression related to calls and the V128 type * Add an extra space on comments to match code style * Some refactoring * Fix vector insert FP32 SSE2 path * Port over the ARM32 instructions * Avoid memory protection races on JIT Cache * Another fix on VectorInsert FP32 (thanks to LDj3SNuD * Float operands don't need to use the same register when VEX is supported * Add a new register allocator, higher quality code for hot code (tier up), and other tweaks * Some nits, small improvements on the pre allocator * CpuThreadState is gone * Allow changing CPU emulators with a config entry * Add runtime identifiers on the ARMeilleure project * Allow switching between CPUs through a config entry (pt. 2) * Change win10-x64 to win-x64 on projects * Update the Ryujinx project to use ARMeilleure * Ensure that the selected register is valid on the hybrid allocator * Allow exiting on returns to 0 (should fix test regression) * Remove register assignments for most used variables on the hybrid allocator * Do not use fixed registers as spill temp * Add missing namespace and remove unneeded using * Address PR feedback * Fix types, etc * Enable AssumeStrictAbiCompliance by default * Ensure that Spill and Fill don't load or store any more than necessary
2019-08-08 14:56:22 -04:00
if ((context.Fpcr & FPCR.Fz) != 0 && double.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0d);
}
}
}
return result;
}
public static double FPSqrt(double value)
{
ExecutionContext context = NativeInterface.GetContext();
value = value.FPUnpack(out FPType type, out bool sign, out ulong op, context);
double result;
if (type == FPType.SNaN || type == FPType.QNaN)
{
result = FPProcessNaN(type, op, context);
}
else if (type == FPType.Zero)
{
result = FPZero(sign);
}
else if (type == FPType.Infinity && !sign)
{
result = FPInfinity(sign);
}
else if (sign)
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
else
{
result = Math.Sqrt(value);
if ((context.Fpcr & FPCR.Fz) != 0 && double.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0d);
}
}
return result;
}
public static double FPSub(double value1, double value2)
{
ExecutionContext context = NativeInterface.GetContext();
value1 = value1.FPUnpack(out FPType type1, out bool sign1, out ulong op1, context);
value2 = value2.FPUnpack(out FPType type2, out bool sign2, out ulong op2, context);
double result = FPProcessNaNs(type1, type2, op1, op2, out bool done, context);
if (!done)
{
bool inf1 = type1 == FPType.Infinity; bool zero1 = type1 == FPType.Zero;
bool inf2 = type2 == FPType.Infinity; bool zero2 = type2 == FPType.Zero;
if (inf1 && inf2 && sign1 == sign2)
{
result = FPDefaultNaN();
FPProcessException(FPException.InvalidOp, context);
}
else if ((inf1 && !sign1) || (inf2 && sign2))
{
result = FPInfinity(false);
}
else if ((inf1 && sign1) || (inf2 && !sign2))
{
result = FPInfinity(true);
}
else if (zero1 && zero2 && sign1 == !sign2)
{
result = FPZero(sign1);
}
else
{
result = value1 - value2;
if ((context.Fpcr & FPCR.Fz) != 0 && double.IsSubnormal(result))
{
context.Fpsr |= FPSR.Ufc;
result = FPZero(result < 0d);
}
}
}
return result;
}
private static double FPDefaultNaN()
{
return -double.NaN;
}
private static double FPInfinity(bool sign)
{
return sign ? double.NegativeInfinity : double.PositiveInfinity;
}
private static double FPZero(bool sign)
{
return sign ? -0d : +0d;
}
private static double FPMaxNormal(bool sign)
{
return sign ? double.MinValue : double.MaxValue;
}
private static double FPTwo(bool sign)
{
return sign ? -2d : +2d;
}
private static double FPOnePointFive(bool sign)
{
return sign ? -1.5d : +1.5d;
}
private static double FPNeg(this double value)
{
return -value;
}
private static double ZerosOrOnes(bool ones)
{
return BitConverter.Int64BitsToDouble(ones ? -1L : 0L);
}
private static double FPUnpack(
this double value,
out FPType type,
out bool sign,
out ulong valueBits,
ExecutionContext context)
{
valueBits = (ulong)BitConverter.DoubleToInt64Bits(value);
sign = (~valueBits & 0x8000000000000000ul) == 0ul;
if ((valueBits & 0x7FF0000000000000ul) == 0ul)
{
if ((valueBits & 0x000FFFFFFFFFFFFFul) == 0ul || (context.Fpcr & FPCR.Fz) != 0)
{
type = FPType.Zero;
value = FPZero(sign);
if ((valueBits & 0x000FFFFFFFFFFFFFul) != 0ul)
{
FPProcessException(FPException.InputDenorm, context);
}
}
else
{
type = FPType.Nonzero;
}
}
else if ((~valueBits & 0x7FF0000000000000ul) == 0ul)
{
if ((valueBits & 0x000FFFFFFFFFFFFFul) == 0ul)
{
type = FPType.Infinity;
}
else
{
type = (~valueBits & 0x0008000000000000ul) == 0ul ? FPType.QNaN : FPType.SNaN;
value = FPZero(sign);
}
}
else
{
type = FPType.Nonzero;
}
return value;
}
private static double FPProcessNaNs(
FPType type1,
FPType type2,
ulong op1,
ulong op2,
out bool done,
ExecutionContext context)
{
done = true;
if (type1 == FPType.SNaN)
{
return FPProcessNaN(type1, op1, context);
}
else if (type2 == FPType.SNaN)
{
return FPProcessNaN(type2, op2, context);
}
else if (type1 == FPType.QNaN)
{
return FPProcessNaN(type1, op1, context);
}
else if (type2 == FPType.QNaN)
{
return FPProcessNaN(type2, op2, context);
}
done = false;
return FPZero(false);
}
private static double FPProcessNaNs3(
FPType type1,
FPType type2,
FPType type3,
ulong op1,
ulong op2,
ulong op3,
out bool done,
ExecutionContext context)
{
done = true;
if (type1 == FPType.SNaN)
{
return FPProcessNaN(type1, op1, context);
}
else if (type2 == FPType.SNaN)
{
return FPProcessNaN(type2, op2, context);
}
else if (type3 == FPType.SNaN)
{
return FPProcessNaN(type3, op3, context);
}
else if (type1 == FPType.QNaN)
{
return FPProcessNaN(type1, op1, context);
}
else if (type2 == FPType.QNaN)
{
return FPProcessNaN(type2, op2, context);
}
else if (type3 == FPType.QNaN)
{
return FPProcessNaN(type3, op3, context);
}
done = false;
return FPZero(false);
}
private static double FPProcessNaN(FPType type, ulong op, ExecutionContext context)
{
if (type == FPType.SNaN)
{
op |= 1ul << 51;
FPProcessException(FPException.InvalidOp, context);
}
if ((context.Fpcr & FPCR.Dn) != 0)
{
return FPDefaultNaN();
}
return BitConverter.Int64BitsToDouble((long)op);
}
private static void FPProcessException(FPException exc, ExecutionContext context)
{
int enable = (int)exc + 8;
if ((context.Fpcr & (FPCR)(1 << enable)) != 0)
{
throw new NotImplementedException("Floating-point trap handling.");
}
else
{
context.Fpsr |= (FPSR)(1 << (int)exc);
}
}
}
}