public void RunBasicScenario()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunBasicScenario));
UInt32[] values = new UInt32[ElementCount];
for (int i = 0; i < ElementCount; i++)
{
values[i] = TestLibrary.Generator.GetUInt32();
}
Vector256 <UInt32> value = Vector256.Create(values[0], values[1], values[2], values[3], values[4], values[5], values[6], values[7]);
Vector128 <UInt32> lowerResult = value.GetLower();
Vector128 <UInt32> upperResult = value.GetUpper();
ValidateGetResult(lowerResult, upperResult, values);
Vector256 <UInt32> result = value.WithLower(upperResult);
result = result.WithUpper(lowerResult);
ValidateWithResult(result, values);
}
public void RunBasicScenario()
{
TestLibrary.TestFramework.BeginScenario(nameof(RunBasicScenario));
SByte[] values = new SByte[ElementCount];
for (int i = 0; i < ElementCount; i++)
{
values[i] = TestLibrary.Generator.GetSByte();
}
Vector256 <SByte> value = Vector256.Create(values[0], values[1], values[2], values[3], values[4], values[5], values[6], values[7], values[8], values[9], values[10], values[11], values[12], values[13], values[14], values[15], values[16], values[17], values[18], values[19], values[20], values[21], values[22], values[23], values[24], values[25], values[26], values[27], values[28], values[29], values[30], values[31]);
Vector128 <SByte> lowerResult = value.GetLower();
Vector128 <SByte> upperResult = value.GetUpper();
ValidateGetResult(lowerResult, upperResult, values);
Vector256 <SByte> result = value.WithLower(upperResult);
result = result.WithUpper(lowerResult);
ValidateWithResult(result, values);
}
public static Vector256 <float> Divide(Vector256 <float> dividend, Vector256 <float> divisor)
{
if (Avx.IsSupported)
{
return(Avx.Divide(dividend, divisor));
}
return(FromLowHigh(Divide(dividend.GetLower(), divisor.GetLower()), Divide(dividend.GetUpper(), divisor.GetLower())));
}
public static void GetLowHigh(Vector256 <double> vector, out Vector128 <double> low, out Vector128 <double> high)
{
low = vector.GetLower();
high = vector.GetUpper();
}
private static float SumVector256(Vector256 <float> v)
{
v = Avx.HorizontalAdd(v, v); //0+1, 2+3, .., .., 4+5, 6+7, .., ..
v = Avx.HorizontalAdd(v, v); //0+1+2+3, .., .., .., 4+5+6+7, .., .., ..
return(v.GetUpper().ToScalar() + v.GetLower().ToScalar());
}
unsafe void IConvolver.ConvolveSourceLine(byte *istart, byte *tstart, int cb, byte *mapxstart, int smapx, int smapy)
{
float *tp = (float *)tstart, tpe = (float *)(tstart + cb);
float *pmapx = (float *)mapxstart;
int kstride = smapx * channels;
int tstride = smapy * 4;
int vcnt = smapx / Vector128 <float> .Count;
while (tp < tpe)
{
int ix = *(int *)pmapx++;
int lcnt = vcnt;
float *ip = (float *)istart + ix * channels;
float *mp = pmapx;
pmapx += kstride;
Vector128 <float> av0, av1, av2;
if (Avx.IsSupported && lcnt >= 2)
{
Vector256 <float> ax0 = Vector256 <float> .Zero, ax1 = ax0, ax2 = ax0;
for (; lcnt >= 2; lcnt -= 2)
{
var iv0 = Avx.LoadVector256(ip);
var iv1 = Avx.LoadVector256(ip + Vector256 <float> .Count);
var iv2 = Avx.LoadVector256(ip + Vector256 <float> .Count * 2);
ip += Vector256 <int> .Count * channels;
if (Fma.IsSupported)
{
ax0 = Fma.MultiplyAdd(Avx.LoadVector256(mp), iv0, ax0);
ax1 = Fma.MultiplyAdd(Avx.LoadVector256(mp + Vector256 <float> .Count), iv1, ax1);
ax2 = Fma.MultiplyAdd(Avx.LoadVector256(mp + Vector256 <float> .Count * 2), iv2, ax2);
}
else
{
ax0 = Avx.Add(ax0, Avx.Multiply(iv0, Avx.LoadVector256(mp)));
ax1 = Avx.Add(ax1, Avx.Multiply(iv1, Avx.LoadVector256(mp + Vector256 <float> .Count)));
ax2 = Avx.Add(ax2, Avx.Multiply(iv2, Avx.LoadVector256(mp + Vector256 <float> .Count * 2)));
}
mp += Vector256 <float> .Count * channels;
}
av0 = Sse.Add(ax0.GetLower(), ax1.GetUpper());
av1 = Sse.Add(ax0.GetUpper(), ax2.GetLower());
av2 = Sse.Add(ax1.GetLower(), ax2.GetUpper());
}
else
{
av0 = av1 = av2 = Vector128 <float> .Zero;
}
for (; lcnt != 0; lcnt--)
{
var iv0 = Sse.LoadVector128(ip);
var iv1 = Sse.LoadVector128(ip + Vector128 <float> .Count);
var iv2 = Sse.LoadVector128(ip + Vector128 <float> .Count * 2);
ip += Vector128 <float> .Count * channels;
if (Fma.IsSupported)
{
av0 = Fma.MultiplyAdd(Sse.LoadVector128(mp), iv0, av0);
av1 = Fma.MultiplyAdd(Sse.LoadVector128(mp + Vector128 <float> .Count), iv1, av1);
av2 = Fma.MultiplyAdd(Sse.LoadVector128(mp + Vector128 <float> .Count * 2), iv2, av2);
}
else
{
av0 = Sse.Add(av0, Sse.Multiply(iv0, Sse.LoadVector128(mp)));
av1 = Sse.Add(av1, Sse.Multiply(iv1, Sse.LoadVector128(mp + Vector128 <float> .Count)));
av2 = Sse.Add(av2, Sse.Multiply(iv2, Sse.LoadVector128(mp + Vector128 <float> .Count * 2)));
}
mp += Vector128 <float> .Count * channels;
}
var avs0 = Sse.Add(Sse.Add(
Sse.Shuffle(av0, av0, 0b_00_10_01_11),
Sse.Shuffle(av1, av1, 0b_00_01_11_10)),
Sse.Shuffle(av2, av2, 0b_00_11_10_01)
);
var avs1 = Sse3.IsSupported ?
Sse3.MoveHighAndDuplicate(avs0) :
Sse.Shuffle(avs0, avs0, 0b_11_11_01_01);
var avs2 = Sse.UnpackHigh(avs0, avs0);
tp[0] = Sse.AddScalar(av0, avs0).ToScalar();
tp[1] = Sse.AddScalar(av1, avs1).ToScalar();
tp[2] = Sse.AddScalar(av2, avs2).ToScalar();
tp += tstride;
}
}
public static void GetLowHigh <T>(Vector256 <T> vector, out Vector128 <T> low, out Vector128 <T> high) where T : struct
{
low = vector.GetLower();
high = vector.GetUpper();
}
