public static void fpbinary_gemm_cpu(IntPtr a, IntPtr b, IntPtr c, int m, int nn, int k, int transb, int beta, int alpha, IntPtr alphas)
{
if (THWrapper.THFloatTensor_nDimension(c) != 2 || THWrapper.THFloatTensor_size(c, 0) * THWrapper.THFloatTensor_size(c, 1) < m * k)
{
THWrapper.THFloatTensor_resize2d(c, m, k);
//THFloatTensor_resize2d(c, m, k);
}
/*
* uint32_t* A = (uint32_t*)THIntTensor_data(a);
* uint32_t* B = (uint32_t*)THIntTensor_data(b);
* float* C = THFloatTensor_data(c);
* float* D = THFloatTensor_data(alphas);
*/
var A = THWrapper.THIntTensor_data(a);
var B = THWrapper.THIntTensor_data(b);
var C = THWrapper.THFloatTensor_data(c);
var D = THWrapper.THFloatTensor_data(alphas);
var aa = InternalArray.FromTHIntTensor(a);
var bb = InternalArray.FromTHIntTensor(b);
var cc = InternalArray.FromTHFloatTensor(c);
var dd = InternalArray.FromTHFloatTensor(alphas);
int n = 1 + (nn - 1) / matmul.ENCODE_BIT;
int brow = transb != 0 ? 1 : k;
int bcol = transb != 0 ? n : 1;
//matmul.dgemm_nn(m, k, nn, A, n, 1, B, brow, bcol, C, k, 1, beta, alpha, D);
//matmul.dgemm_nn(m, k, nn, A, n, 1, B, brow, bcol, C, k, 1, beta, alpha, D);
matmul.fpdgemm_nn(m, k, nn, A, n, 1, B, brow, bcol, C, k, 1, beta);
if (alpha != 0)
{
for (int i = 0; i < m; i++)
{
for (int j = 0; j < k; j++)
{
//C[i * n + j] *= alphas[i];
var aa1 = matmul.GetFloat(C, i * k + j);
short aq1 = (short)(aa1 * 256);
var aa2 = matmul.GetFloat(D, i);
var aq2 = (short)(aa2 * 256);
var val4 = (short)((int)(aq1 * aq2) >> 8);
var quant_res = val4 / 256f;
var orig = aa1 * aa2;
//matmul.SetFloat(C, i * k + j, aa1 * aa2);
matmul.SetFloat(C, i * k + j, val4);
//C[i * n + j] = (float)(C[i * n + j] * alphas[i]);
}
}
}
}
internal static InternalArray bin_linear(InternalArray input, InternalArray weight, InternalArray bias, InternalArray alpha)
{
var m = input.Shape[0];
var n = input.Shape[1];
var k = weight.Shape[0];
/**
*
* m = input.data.shape[0]
* n = input.data.shape[1]
* k = weight.data.shape[0]
* out_tensor = torch.FloatTensor()
* bin_input = torch.IntTensor()
* use_cuda = input.is_cuda
* binop.encode_rows_cpu(input.data, bin_input)
* binop.binary_gemm_cpu(bin_input, weight.data, output.data, m, n, k, 1, 0, 0, alpha.data)
* output.data.mul_(alpha.data.t().expand(output.shape))
* if bias is not None:
* output.data.add_(bias.data.expand(output.shape))
* return output
*
*/
//InternalArray output = new InternalArray(new int[] { });
var _input = input.ToTHTensor();
var bin_input = THWrapper.THIntTensor_new();
encode_rows_cpu(_input, bin_input);
var temp = InternalArray.FromTHIntTensor(bin_input);
var _alpha = alpha.ToTHTensor();
//var _bin_input = bin_input.ToTHTensor();
var _weight = weight.ToTHTensor();
var _output = THWrapper.THFloatTensor_new();
binop.binary_gemm_cpu(bin_input, _weight, _output, m, n, k, 1, 0, 0, _alpha);
var temp2 = InternalArray.FromTHFloatTensor(_output);
THWrapper.THFloatTensor_free(_input);
THWrapper.THIntTensor_free(bin_input);
//var tt = alpha.ToTHTensor();
var ttt = alpha.Transpose2D();
//var newt=THWrapper.THFloatTensor_newTranspose(tt, 0, 1);
/*output.data.mul_(alpha.data.t().expand(output.shape))
*/
if (bias != null)
{/*
* if bias is not None:
* output.data.add_(bias.data.expand(output.shape))*/
}
var output = InternalArray.FromTHFloatTensor(_output);
for (int i = 0; i < ttt.Data.Length; i++)
{
output.Data[i] *= ttt.Data[i];
}
THWrapper.THFloatTensor_free(_output);
return(output);
}