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

debug tensor prod contract 2d cl and pthread

Browse Source
This commit is contained in:
fanasina
2024-02-12 12:08:16 +01:00
parent 60e4b16308
commit c07ed1b160
5 changed files with 38 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tensor_t/src/kernel_2d_ProdContractnTensor.cl
+4 -4
View File
@@ -49,8 +49,8 @@ __kernel void prodContractnTensor2dLiniNotEndian_TYPE_FLOAT(long unsigned int dS
size_t ind = i + dSubRank * j;
Mx[ind] = 0;
for (k = 0; k < dMRank; k++) {
n0_id = i + dMRank * k;
n1_id = j * dSubRank + k;
n0_id = i + dSubRank * k;
n1_id = j * dMRank + k;
Mx[ind] += M0x[n0_id] * M1x[n1_id];
}
}
@@ -68,8 +68,8 @@ __kernel void prodContractnTensor2dLinNotEndian_TYPE_DOUBLE(long unsigned int dS
size_t ind = i + dSubRank * j;
Mx[ind] = 0;
for (k = 0; k < dMRank; k++) {
n0_id = i + dMRank * k;
n1_id = j * dSubRank + k;
n0_id = i + dSubRank * k;
n1_id = j * dMRank + k;
Mx[ind] += M0x[n0_id] * M1x[n1_id];
}
}