fanasina/y_project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

put all kernel opencl funcs in src

Browse Source
This commit is contained in:
fanasina
2024-01-24 19:02:16 +01:00
parent 4bf32c6501
commit b83a125bac
2 changed files with 57 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tensor_t/src/kernel_ProdContractnTensor.cl
+36
View File
@@ -0,0 +1,36 @@
__kernel void prodContractnTensorLin_TYPE_FLOAT(long unsigned int dSubRank, long unsigned int dMRank, __global const float *M0x , __global const float *M1x, __global float *Mx ){
//Get the index of the current element to be processed
size_t i = get_global_id(0);
size_t k, a0_id, a1_id, n0_id, n1_id;
a0_id = i / dSubRank;
a1_id = i % dSubRank;
Mx[i] = 0;
for (k = 0; k < dMRank; k++) {
n0_id = a0_id * dMRank + k;
n1_id = a1_id + dSubRank * k;
Mx[i] += M0x[n0_id] * M1x[n1_id];
}
}
__kernel void prodContractnTensorLin_TYPE_DOUBLE(long unsigned int dSubRank, long unsigned int dMRank, __global const double *M0x , __global const double *M1x, __global double *Mx ){
//Get the index of the current element to be processed
size_t i = get_global_id(0);
size_t k, a0_id, a1_id, n0_id, n1_id;
a0_id = i / dSubRank;
a1_id = i % dSubRank;
Mx[i] = 0;
for (k = 0; k < dMRank; k++) {
n0_id = a0_id * dMRank + k;
n1_id = a1_id + dSubRank * k;
Mx[i] += M0x[n0_id] * M1x[n1_id];
}
}