static unsafe int Main(string[] args)
{
int testResult = Pass;
if (Sse.IsSupported)
{
using (TestTable <float> floatTable = new TestTable <float>(new float[4] {
1, -5, 100, 0
}, new float[4] {
22, -1, -50, 0
}, new float[4]))
{
var vf1 = Unsafe.Read <Vector128 <float> >(floatTable.inArray1Ptr);
var vf2 = Unsafe.Read <Vector128 <float> >(floatTable.inArray2Ptr);
var vf3 = Sse.CompareGreaterThan(vf1, vf2);
Unsafe.Write(floatTable.outArrayPtr, vf3);
if (!floatTable.CheckResult((x, y, z) => BitConverter.SingleToInt32Bits(z) == ((x > y) ? -1 : 0)))
{
Console.WriteLine("SSE CompareGreaterThan failed on float:");
foreach (var item in floatTable.outArray)
{
Console.Write(item + ", ");
}
Console.WriteLine();
testResult = Fail;
}
}
}
return(testResult);
}
public void RunStructFldScenario(SimpleBinaryOpTest__CompareGreaterThanSingle testClass)
{
var result = Sse.CompareGreaterThan(_fld1, _fld2);
Unsafe.Write(testClass._dataTable.outArrayPtr, result);
testClass.ValidateResult(_fld1, _fld2, testClass._dataTable.outArrayPtr);
}
public void RunClassFldScenario()
{
var result = Sse.CompareGreaterThan(_fld1, _fld2);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(_fld1, _fld2, _dataTable.outArrayPtr);
}
public void RunStructLclFldScenario()
{
var test = TestStruct.Create();
var result = Sse.CompareGreaterThan(test._fld1, test._fld2);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(test._fld1, test._fld2, _dataTable.outArrayPtr);
}
public void RunClassLclFldScenario()
{
var test = new SimpleBinaryOpTest__CompareGreaterThanSingle();
var result = Sse.CompareGreaterThan(test._fld1, test._fld2);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(test._fld1, test._fld2, _dataTable.outArrayPtr);
}
public static Vector4F GreaterThanOrEqual(Vector4FParam1_3 left, Vector4FParam1_3 right)
{
if (Sse.IsSupported)
{
return(Sse.CompareGreaterThan(left, right));
}
return(GreaterThanOrEqual_Software(left, right));
}
public static Vector128 <float> CompareGreaterThan(Vector4FParam1_3 left, Vector4FParam1_3 right)
{
if (Sse.IsSupported)
{
return(Sse.CompareGreaterThan(left, right));
}
return(CompareGreaterThan_Software(left, right));
}
public void RunClassFldScenario()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunClassFldScenario));
var result = Sse.CompareGreaterThan(_fld1, _fld2);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(_fld1, _fld2, _dataTable.outArrayPtr);
}
public void RunLclVarScenario_LoadAligned()
{
var left = Sse.LoadAlignedVector128((Single *)(_dataTable.inArray1Ptr));
var right = Sse.LoadAlignedVector128((Single *)(_dataTable.inArray2Ptr));
var result = Sse.CompareGreaterThan(left, right);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(left, right, _dataTable.outArrayPtr);
}
public void RunLclVarScenario_UnsafeRead()
{
var left = Unsafe.Read <Vector128 <Single> >(_dataTable.inArray1Ptr);
var right = Unsafe.Read <Vector128 <Single> >(_dataTable.inArray2Ptr);
var result = Sse.CompareGreaterThan(left, right);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(left, right, _dataTable.outArrayPtr);
}
public void RunBasicScenario_UnsafeRead()
{
var result = Sse.CompareGreaterThan(
Unsafe.Read <Vector128 <Single> >(_dataTable.inArray1Ptr),
Unsafe.Read <Vector128 <Single> >(_dataTable.inArray2Ptr)
);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(_dataTable.inArray1Ptr, _dataTable.inArray2Ptr, _dataTable.outArrayPtr);
}
public void RunStructLclFldScenario()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunStructLclFldScenario));
var test = TestStruct.Create();
var result = Sse.CompareGreaterThan(test._fld1, test._fld2);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(test._fld1, test._fld2, _dataTable.outArrayPtr);
}
public void RunClsVarScenario()
{
var result = Sse.CompareGreaterThan(
_clsVar1,
_clsVar2
);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(_clsVar1, _clsVar2, _dataTable.outArrayPtr);
}
public void RunBasicScenario_LoadAligned()
{
var result = Sse.CompareGreaterThan(
Sse.LoadAlignedVector128((Single *)(_dataTable.inArray1Ptr)),
Sse.LoadAlignedVector128((Single *)(_dataTable.inArray2Ptr))
);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(_dataTable.inArray1Ptr, _dataTable.inArray2Ptr, _dataTable.outArrayPtr);
}
public void RunLclVarScenario_UnsafeRead()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunLclVarScenario_UnsafeRead));
var op1 = Unsafe.Read <Vector128 <Single> >(_dataTable.inArray1Ptr);
var op2 = Unsafe.Read <Vector128 <Single> >(_dataTable.inArray2Ptr);
var result = Sse.CompareGreaterThan(op1, op2);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(op1, op2, _dataTable.outArrayPtr);
}
public void RunLclVarScenario_LoadAligned()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunLclVarScenario_LoadAligned));
var op1 = Sse.LoadAlignedVector128((Single *)(_dataTable.inArray1Ptr));
var op2 = Sse.LoadAlignedVector128((Single *)(_dataTable.inArray2Ptr));
var result = Sse.CompareGreaterThan(op1, op2);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(op1, op2, _dataTable.outArrayPtr);
}
public void RunLclVarScenario_Load()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunLclVarScenario_Load));
var left = Sse.LoadVector128((Single *)(_dataTable.inArray1Ptr));
var right = Sse.LoadVector128((Single *)(_dataTable.inArray2Ptr));
var result = Sse.CompareGreaterThan(left, right);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(left, right, _dataTable.outArrayPtr);
}
internal static bool IsIntersectingSse(Vector128 <float> a, Vector128 <float> b)
{
var aMin = Sse.MoveLowToHigh(a, a);
var aMax = Sse.MoveHighToLow(a, a);
var bMin = Sse.MoveLowToHigh(b, b);
var bMax = Sse.MoveHighToLow(b, b);
var lt = Sse.CompareGreaterThan(aMin, bMax);
var gt = Sse.CompareLessThan(aMax, bMin);
var oob = Sse.Or(gt, lt);
return(Sse.MoveMask(oob) == 0);
}
public void RunStructLclFldScenario_Load()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunStructLclFldScenario_Load));
var test = TestStruct.Create();
var result = Sse.CompareGreaterThan(
Sse.LoadVector128((Single *)(&test._fld1)),
Sse.LoadVector128((Single *)(&test._fld2))
);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(test._fld1, test._fld2, _dataTable.outArrayPtr);
}
public void RunStructFldScenario_Load(SimpleBinaryOpTest__CompareGreaterThanSingle testClass)
{
fixed(Vector128 <Single> *pFld1 = &_fld1)
fixed(Vector128 <Single> *pFld2 = &_fld2)
{
var result = Sse.CompareGreaterThan(
Sse.LoadVector128((Single *)(pFld1)),
Sse.LoadVector128((Single *)(pFld2))
);
Unsafe.Write(testClass._dataTable.outArrayPtr, result);
testClass.ValidateResult(_fld1, _fld2, testClass._dataTable.outArrayPtr);
}
}
public void RunClassFldScenario_Load()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunClassFldScenario_Load));
fixed(Vector128 <Single> *pFld1 = &_fld1)
fixed(Vector128 <Single> *pFld2 = &_fld2)
{
var result = Sse.CompareGreaterThan(
Sse.LoadVector128((Single *)(pFld1)),
Sse.LoadVector128((Single *)(pFld2))
);
Unsafe.Write(_dataTable.outArrayPtr, result);
ValidateResult(_fld1, _fld2, _dataTable.outArrayPtr);
}
}
static bool TestSseCompareGreaterThan()
{
if (Sse.IsSupported)
{
const int expectedResult = 0b0100;
Vector128 <float> value1 = Vector128.Create(float.NaN, 1.0f, 2.0f, 3.0f);
Vector128 <float> value2 = Vector128.Create(0.0f, 2.0f, 1.0f, 3.0f);
Vector128 <float> result = Sse.CompareGreaterThan(value1, value2);
int actualResult = Sse.MoveMask(result);
if (actualResult != expectedResult)
{
Console.WriteLine($"{nameof(Sse)}.{nameof(Sse.CompareGreaterThan)}({value1}, {value2}) returned {Convert.ToString(actualResult, 2)}; expected {Convert.ToString(expectedResult, 2)}");
return(false);
}
}
return(true);
}
public Intro()
{
var middleVector = Vector128.Create(1.0f); // middleVector = <1,1,1,1>
middleVector = Vector128.CreateScalar(-1.0f); // middleVector = <-1,0,0,0>
var floatBytes = Vector64.AsByte(Vector64.Create(1.0f, -1.0f)); // floatBytes = <0, 0, 128, 63, 0, 0, 128, 191>
if (Avx.IsSupported)
{
var left = Vector256.Create(-2.5f); // <-2.5, -2.5, -2.5, -2.5, -2.5, -2.5, -2.5, -2.5>
var right = Vector256.Create(5.0f); // <5, 5, 5, 5, 5, 5, 5, 5>
Vector256 <float> result = Avx.AddSubtract(left, right); // result = <-7.5, 2.5, -7.5, 2.5, -7.5, 2.5, -7.5, 2.5>xit
left = Vector256.Create(-1.0f, -2.0f, -3.0f, -4.0f, -50.0f, -60.0f, -70.0f, -80.0f);
right = Vector256.Create(0.0f, 2.0f, 3.0f, 4.0f, 50.0f, 60.0f, 70.0f, 80.0f);
result = Avx.UnpackHigh(left, right); // result = <-3, 3, -4, 4, -70, 70, -80, 80>
result = Avx.UnpackLow(left, right); // result = <-1, 1, -2, 2, -50, 50, -60, 60>
result = Avx.DotProduct(left, right, 0b1111_0001); // result = <-30, 0, 0, 0, -17400, 0, 0, 0>
bool testResult = Avx.TestC(left, right); // testResult = true
testResult = Avx.TestC(right, left); // testResult = false
Vector256 <float> result1 = Avx.Divide(left, right);
var plusOne = Vector256.Create(1.0f);
result = Avx.Compare(right, result1, FloatComparisonMode.OrderedGreaterThanNonSignaling);
result = Avx.Compare(right, result1, FloatComparisonMode.UnorderedNotLessThanNonSignaling);
left = Vector256.Create(0.0f, 3.0f, -3.0f, 4.0f, -50.0f, 60.0f, -70.0f, 80.0f);
right = Vector256.Create(0.0f, 2.0f, 3.0f, 2.0f, 50.0f, -60.0f, 70.0f, -80.0f);
Vector256 <float> nanInFirstPosition = Avx.Divide(left, right);
left = Vector256.Create(1.1f, 3.3333333f, -3.0f, 4.22f, -50.0f, 60.0f, -70.0f, 80.0f);
Vector256 <float> InfInFirstPosition = Avx.Divide(left, right);
left = Vector256.Create(-1.1f, 3.0f, 1.0f / 3.0f, MathF.PI, -50.0f, 60.0f, -70.0f, 80.0f);
right = Vector256.Create(0.0f, 2.0f, 3.1f, 2.0f, 50.0f, -60.0f, 70.0f, -80.0f);
Vector256 <float> compareResult = Avx.Compare(left, right, FloatComparisonMode.OrderedGreaterThanNonSignaling); // compareResult = <0, NaN, 0, NaN, 0, NaN, 0, NaN>
Vector256 <float> mixed = Avx.BlendVariable(left, right, compareResult); // mixed = <-1, 2, -3, 2, -50, -60, -70, -80>
//left = Vector256.Create(-1.0f, 1.0f, -1.0f, 1.0f, -1.0f, 1.0f, -1.0f, 1.0f);
//right = Vector256.Create(1.0f, 1.0f, -1.0f, 1.0f, 1.0f, 1.0f, -1.0f, 1.0f);
Vector256 <float> other = right = Vector256.Create(0.0f, 2.0f, 3.0f, 2.0f, 50.0f, -60.0f, 70.0f, -80.0f);
bool bRes = Avx.TestZ(plusOne, compareResult);
bool bRes2 = Avx.TestC(plusOne, compareResult);
bool allTrue = !Avx.TestZ(compareResult, compareResult);
compareResult = Avx.Compare(nanInFirstPosition, right, FloatComparisonMode.OrderedEqualNonSignaling); // compareResult = <0, NaN, 0, NaN, 0, NaN, 0, NaN>
compareResult = Avx.Compare(nanInFirstPosition, right, FloatComparisonMode.UnorderedEqualNonSignaling);
compareResult = Avx.Compare(InfInFirstPosition, right, FloatComparisonMode.UnorderedNotLessThanOrEqualNonSignaling);
compareResult = Avx.Compare(InfInFirstPosition, right, FloatComparisonMode.OrderedGreaterThanNonSignaling);
var left128 = Vector128.Create(1.0f, 2.0f, 3.0f, 4.0f);
var right128 = Vector128.Create(2.0f, 3.0f, 4.0f, 5.0f);
Vector128 <float> compResult128 = Sse.CompareGreaterThan(left128, right128); // compResult128 = <0, 0, 0, 0>
int res = Avx.MoveMask(compareResult);
if (Fma.IsSupported)
{
Vector256 <float> resultFma = Fma.MultiplyAdd(left, right, other); // = left * right + other for each element
resultFma = Fma.MultiplyAddNegated(left, right, other); // = -(left * right + other) for each element
resultFma = Fma.MultiplySubtract(left, right, other); // = left * right - other for each element
Fma.MultiplyAddSubtract(left, right, other); // even elements (0, 2, ...) like MultiplyAdd, odd elements like MultiplySubtract
}
result = Avx.DotProduct(left, right, 0b1010_0001); // result = <-20, 0, 0, 0, -10000, 0, 0, 0>
result = Avx.Floor(left); // result = <-3, -3, -3, -3, -3, -3, -3, -3>
result = Avx.Add(left, right); // result = <2.5, 2.5, 2.5, 2.5, 2.5, 2.5, 2.5, 2.5>
result = Avx.Ceiling(left); // result = <-2, -2, -2, -2, -2, -2, -2, -2>
result = Avx.Multiply(left, right); // result = <-12.5, -12.5, -12.5, -12.5, -12.5, -12.5, -12.5, -12.5>
result = Avx.HorizontalAdd(left, right); // result = <-5, -5, 10, 10, -5, -5, 10, 10>
result = Avx.HorizontalSubtract(left, right); // result = <0, 0, 0, 0, 0, 0, 0, 0>
double[] someDoubles = new double[] { 1.0, 3.0, -2.5, 7.5, 10.8, 0.33333 };
double[] someOtherDoubles = new double[] { 1.0, 1.0, 1.0, 1.0, 1.0, 1.0 };
double[] someResult = new double[someDoubles.Length];
float[] someFloats = new float[] { 1, 2, 3, 4, 10, 20, 30, 40, 0 };
float[] someOtherFloats = new float[] { 1, 1, 1, 1, 1, 1, 1, 1 };
unsafe
{
fixed(double *ptr = &someDoubles[1])
{
fixed(double *ptr2 = &someResult[0])
{
Vector256 <double> res2 = Avx.LoadVector256(ptr); // res2 = <3, -2.5, 7.5, 10.8>
Avx.Store(ptr2, res2);
}
}
fixed(float *ptr = &someFloats[0])
{
fixed(float *ptr2 = &someOtherFloats[0])
{
Vector256 <float> res2 = Avx.DotProduct(Avx.LoadVector256(ptr), Avx.LoadVector256(ptr2), 0b0001_0001);
//Avx.Store(ptr2, res2);
}
}
}
}
}
public void ResizeBicubic(FastBitmap rtnImage)
{
float scaleX = (float)this.width / rtnImage.width;
float scaleY = (float)this.height / rtnImage.height;
if (scaleX > 1 || scaleY > 1)
{
throw new Exception("拡大のみ対応");
}
float[] tmpa = new float[rtnImage.width * 4 * this.height];
fixed(float *tmpp = tmpa)
{
float *tmp = tmpp;
var _00mask = Vector128.Create(0, 255, 255, 255, 1, 255, 255, 255, 2, 255, 255, 255, 3, 255, 255, 255);
var _01mask = Vector128.Create(4, 255, 255, 255, 5, 255, 255, 255, 6, 255, 255, 255, 7, 255, 255, 255);
var _10mask = Vector128.Create(8, 255, 255, 255, 9, 255, 255, 255, 10, 255, 255, 255, 11, 255, 255, 255);
var _11mask = Vector128.Create(12, 255, 255, 255, 13, 255, 255, 255, 14, 255, 255, 255, 15, 255, 255, 255);
var _vmask = Vector128.Create(0, 4, 8, 12, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255);
var _1012 = Vector128.Create(-1, 0, 1, 2);
var _0123i = Vector128.Create(0, 1, 2, 3);
var _0000 = Vector128.Create(0, 0, 0, 0);
var _0000f = Vector128.Create(0f, 0, 0, 0);
var _255f = Vector128.Create(255f, 255, 255, 255);
var _1111 = Vector128.Create(1, 1, 1, 1);
var _1111f = Vector128.Create(1f, 1, 1, 1);
var _4444f = Vector128.Create(4f, 4, 4, 4);
var _4444 = Vector128.Create(4, 4, 4, 4);
var _5555f = Vector128.Create(5f, 5, 5, 5);
var _2222f = Vector128.Create(2f, 2, 2, 2);
var _8888f = Vector128.Create(8f, 8, 8, 8);
var _7f = Vector128.Create(0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff).AsSingle();
var _ff = Vector128.Create(-1, -1, -1, -1);
var _stride = Vector128.Create(rtnImage.width * 4, rtnImage.width * 4, rtnImage.width * 4, rtnImage.width * 4);
Parallel.For(0, this.height, (y) =>
{
float py = (y * scaleY);
float *tmpPos = tmp + y * rtnImage.width * 4;
for (int x = 0; x < rtnImage.width; x++)
{
float px = (x * scaleX);
int sx = (int)px;
var _px = Vector128.CreateScalar(px);
_px = Sse.Shuffle(_px, _px, 0);
var _sx = Vector128.CreateScalar(sx);
_sx = Sse2.Shuffle(_sx, 0);
var _width = Vector128.CreateScalar(this.width);
_width = Sse2.Shuffle(_width, 0);
var _x2 = Sse2.Add(_sx, _1012);
var _d = Sse.And(Sse.Subtract(_px, Sse2.ConvertToVector128Single(_x2)), _7f);
var _d2 = Sse.Multiply(_d, _d);
var _d3 = Sse.Multiply(_d2, _d);
var w1 = Sse.Add(_1111f, Sse.Subtract(_d3, Sse.Multiply(_2222f, _d2)));
var w2 = Sse.Subtract(Sse.Subtract(Sse.Add(_4444f, Sse.Multiply(_5555f, _d2)), Sse.Multiply(_d, _8888f)), _d3);
var wb = Sse2.CompareGreaterThan(_d, _1111f);
var _w = Sse41.BlendVariable(w1, w2, wb);
var _xpb = Sse2.Or(Sse2.CompareLessThan(_x2, _0000), Sse41.MultiplyLow(Sse2.AndNot(Sse2.CompareLessThan(_x2, _width), _1111).AsInt32(), _ff));
var _xpp = Sse2.And(_sx, _xpb);
var _xp = Sse41.BlendVariable(_x2, _xpp, _xpb);
var p = Avx2.GatherVector128((uint *)(this._ptr + this._stride * y), _xp, 4).AsByte();
var _p0 = Sse2.ConvertToVector128Single(Ssse3.Shuffle(p, _00mask).AsInt32());
var _p1 = Sse2.ConvertToVector128Single(Ssse3.Shuffle(p, _01mask).AsInt32());
var _p2 = Sse2.ConvertToVector128Single(Ssse3.Shuffle(p, _10mask).AsInt32());
var _p3 = Sse2.ConvertToVector128Single(Ssse3.Shuffle(p, _11mask).AsInt32());
var _w0 = Sse.Shuffle(_w, _w, 0);
var _w1 = Sse.Shuffle(_w, _w, 0b01010101);
var _w2 = Sse.Shuffle(_w, _w, 0b10101010);
var _w3 = Sse.Shuffle(_w, _w, 0b11111111);
var rgbaf = Sse.Add(Sse.Add(Sse.Multiply(_p0, _w0), Sse.Multiply(_p1, _w1)), Sse.Add(Sse.Multiply(_p2, _w2), Sse.Multiply(_p3, _w3)));
Sse2.Store(tmpPos + x * 4, rgbaf);
}
});
Parallel.For(0, rtnImage.height, (y) =>
{
float py = (y * scaleY);
int sy = (int)py;
uint *store = stackalloc uint[4];
var _py = Vector128.CreateScalar(py);
_py = Sse.Shuffle(_py, _py, 0);
var _sy = Vector128.CreateScalar(sy);
_sy = Sse2.Shuffle(_sy, 0);
var _height = Vector128.CreateScalar(this.height);
_height = Sse2.Shuffle(_height, 0);
var _y2 = Sse2.Add(_sy, _1012);
var _d = Sse.And(Sse.Subtract(_py, Sse2.ConvertToVector128Single(_y2)), _7f);
var _d2 = Sse.Multiply(_d, _d);
var _d3 = Sse.Multiply(_d2, _d);
var w1 = Sse.Add(_1111f, Sse.Subtract(_d3, Sse.Multiply(_2222f, _d2)));
var w2 = Sse.Subtract(Sse.Subtract(Sse.Add(_4444f, Sse.Multiply(_5555f, _d2)), Sse.Multiply(_d, _8888f)), _d3);
var wb = Sse2.CompareGreaterThan(_d, _1111f);
var _w = Sse41.BlendVariable(w1, w2, wb);
var _ypb = Sse2.Or(Sse2.CompareLessThan(_y2, _0000), Sse41.MultiplyLow(Sse2.AndNot(Sse2.CompareLessThan(_y2, _height), _1111).AsInt32(), _ff));
var _ypp = Sse2.And(_sy, _ypb);
var _yp = Sse41.BlendVariable(_y2, _ypp, _ypb);
var _yps = Sse41.MultiplyLow(_yp, _stride);
var _yp0 = Sse2.Add(Sse2.Shuffle(_yps, 0), _0123i);
var _yp1 = Sse2.Add(Sse2.Shuffle(_yps, 0b01010101), _0123i);
var _yp2 = Sse2.Add(Sse2.Shuffle(_yps, 0b10101010), _0123i);
var _yp3 = Sse2.Add(Sse2.Shuffle(_yps, 0b11111111), _0123i);
uint *rtn = (uint *)(rtnImage._ptr + rtnImage._stride * y);
for (int x = 0; x < rtnImage.width; x++)
{
var _p0 = Avx2.GatherVector128((float *)(tmp), _yp0, 4);
var _p1 = Avx2.GatherVector128((float *)(tmp), _yp1, 4);
var _p2 = Avx2.GatherVector128((float *)(tmp), _yp2, 4);
var _p3 = Avx2.GatherVector128((float *)(tmp), _yp3, 4);
var _w0 = Sse.Shuffle(_w, _w, 0);
var _w1 = Sse.Shuffle(_w, _w, 0b01010101);
var _w2 = Sse.Shuffle(_w, _w, 0b10101010);
var _w3 = Sse.Shuffle(_w, _w, 0b11111111);
var rgbaf = Sse.Add(Sse.Add(Sse.Multiply(_p0, _w0), Sse.Multiply(_p1, _w1)), Sse.Add(Sse.Multiply(_p2, _w2), Sse.Multiply(_p3, _w3)));
var _b0 = Sse.CompareLessThan(rgbaf, _0000f);
rgbaf = Sse41.BlendVariable(rgbaf, _0000f, _b0);
var _b1 = Sse.CompareGreaterThan(rgbaf, _255f);
rgbaf = Sse41.BlendVariable(rgbaf, _255f, _b1);
var rgbab = Sse2.ConvertToVector128Int32(rgbaf).AsByte();
var rgba = Ssse3.Shuffle(rgbab, _vmask).AsUInt32();
Sse2.Store(store, rgba);
_yp0 = Sse2.Add(_yp0, _4444);
_yp1 = Sse2.Add(_yp1, _4444);
_yp2 = Sse2.Add(_yp2, _4444);
_yp3 = Sse2.Add(_yp3, _4444);
*rtn = *store;
rtn++;
}
});
public static m32 GreaterThan(f32 lhs, f32 rhs) => Sse.CompareGreaterThan(lhs, rhs).AsInt32();
unsafe void IConvolver.SharpenLine(byte *cstart, byte *ystart, byte *bstart, byte *ostart, int ox, int ow, float amt, float thresh, bool gamma)
{
float *ip = (float *)cstart + (uint)ox * channels, yp = (float *)ystart + (uint)ox, bp = (float *)bstart, op = (float *)ostart;
float *ipe = ip + (uint)ow * channels;
bool threshold = thresh > 0f;
if (Avx.IsSupported && ip <= ipe - VectorAvx.Count)
{
var vthresh = Vector256.Create(threshold ? thresh : -1f);
var vmsk = Vector256.Create(0x7fffffff).AsSingle();
var vamt = Vector256.Create(amt);
var vmin = VectorAvx.Zero;
ipe -= VectorAvx.Count;
do
{
var vd = Avx.Subtract(Avx.LoadVector256(yp), Avx.LoadVector256(bp));
yp += VectorAvx.Count;
bp += VectorAvx.Count;
if (threshold)
{
var sm = HWIntrinsics.AvxCompareGreaterThan(Avx.And(vd, vmsk), vthresh);
vd = Avx.And(vd, sm);
}
vd = Avx.Multiply(vd, vamt);
var v0 = Avx.LoadVector256(ip);
ip += VectorAvx.Count;
if (gamma)
{
v0 = Avx.Max(v0, vmin);
v0 = Avx.Multiply(v0, Avx.ReciprocalSqrt(v0));
v0 = Avx.Add(v0, vd);
v0 = Avx.Max(v0, vmin);
v0 = Avx.Multiply(v0, v0);
}
else
{
v0 = Avx.Add(v0, vd);
}
Avx.Store(op, v0);
op += VectorAvx.Count;
} while (ip <= ipe);
ipe += VectorAvx.Count;
}
else if (ip <= ipe - VectorSse.Count)
{
var vthresh = Vector128.Create(threshold ? thresh : -1f);
var vmsk = Vector128.Create(0x7fffffff).AsSingle();
var vamt = Vector128.Create(amt);
var vmin = VectorSse.Zero;
ipe -= VectorSse.Count;
do
{
var vd = Sse.Subtract(Sse.LoadVector128(yp), Sse.LoadVector128(bp));
yp += VectorSse.Count;
bp += VectorSse.Count;
if (threshold)
{
var sm = Sse.CompareGreaterThan(Sse.And(vd, vmsk), vthresh);
vd = Sse.And(vd, sm);
}
vd = Sse.Multiply(vd, vamt);
var v0 = Sse.LoadVector128(ip);
ip += VectorSse.Count;
if (gamma)
{
v0 = Sse.Max(v0, vmin);
v0 = Sse.Multiply(v0, Sse.ReciprocalSqrt(v0));
v0 = Sse.Add(v0, vd);
v0 = Sse.Max(v0, vmin);
v0 = Sse.Multiply(v0, v0);
}
else
{
v0 = Sse.Add(v0, vd);
}
Sse.Store(op, v0);
op += VectorSse.Count;
} while (ip <= ipe);
ipe += VectorSse.Count;
}
float fmin = VectorSse.Zero.ToScalar();
while (ip < ipe)
{
float dif = *yp++ - *bp++;
float c0 = *ip++;
if (!threshold || Math.Abs(dif) > thresh)
{
dif *= amt;
if (gamma)
{
c0 = MathUtil.MaxF(c0, fmin).Sqrt();
c0 = MathUtil.MaxF(c0 + dif, fmin);
c0 *= c0;
}
else
{
c0 += dif;
}
}
*op++ = c0;
}
}
public static Vector128 <float> _mm_cmpgt_ps(Vector128 <float> left, Vector128 <float> right)
{
return(Sse.CompareGreaterThan(left, right));
}
public static Vector128 <float> op_GreaterThan(Vector128 <float> left, Vector128 <float> right) => Sse.CompareGreaterThan(left, right);