public static unsafe void MultiplyBlockUnroll8xhParallelWithPadding(float *Ap, int AN, int AM,
float *Bp, int BN, int BM,
float *Cp, int CN, int CM, int bs,
bool transposeA = false, bool transposeB = false)
{
if (transposeA)
{
var tmp = AN; AN = AM; AM = tmp;
}
if (transposeB)
{
var tmp = BN; BN = BM; BM = tmp;
}
int N = AN;
int M = AM;
int K = BM;
{
Assert.IsTrue(bs >= 8, "Matrix Mul block size should be >= 8");
Parallel.For(0, (BM / bs) + (BM % bs > 0 ? 1 : 0), colB =>
{
float[] blockA = new float[bs * bs];
float[] blockB = new float[bs * bs];
float[] blockC = new float[bs * bs];
for (int rowA = 0; rowA < N; rowA += bs)
{
//for (int colB = 0; colB < BM; colB += bs)
{
for (int l = 0; l < AM; l += bs)
{
CopyBlockWithPadding(Ap, rowA, AN, l, AM, blockA, bs, transposeA);
CopyBlockWithPadding(Bp, l, BN, colB * bs, BM, blockB, bs, transposeB);
CopyBlockWithPadding(Cp, rowA, CN, colB * bs, CM, blockC, bs);
fixed(float *blockAp = blockA, blockBp = blockB, blockCp = blockC)
{
MatrixUtils.MultiplyBlockUnroll8xhPadded(blockAp, blockBp, blockCp, bs);
}
CopyBlockWithPadding(blockC, Cp, rowA, CN, colB * bs, CM, bs);
}
}
}
});
}
}
private unsafe void ExecutOverRowA(float *blockA, float *blockB, float *blockC, int rowA)
{
//for (int rowA = 0; rowA < AN; rowA += bs)
{
for (int colB = 0; colB < BM; colB += bs)
{
for (int l = 0; l < AM; l += bs)
{
MatrixUtils.CopyBlockWithPadding(A, rowA, AN, l, AM, blockA, bs, transposeA);
MatrixUtils.CopyBlockWithPadding(B, l, BN, colB, BM, blockB, bs, transposeB);
MatrixUtils.CopyBlockWithPadding(C, rowA, CN, colB, CM, blockC, bs);
MatrixUtils.MultiplyBlockUnroll8xhPadded(blockA, blockB, blockC, bs);
MatrixUtils.CopyBlockWithPadding(blockC, C, rowA, CN, colB, CM, bs);
}
}
}
}