[SYCL][Matrix] Fix the bug in element_wise_all_ops_tf32_impl.hpp (#10469)

Author: arnamoy10

sycl/test-e2e/Matrix/element_wise_all_ops_tf32_impl.hpp

@@ -52,7 +52,7 @@ void matrix_verify_add(queue q, big_matrix<Ts, M, N> &A, nd_range<2> &r,
            ext::intel::experimental::matrix::joint_matrix_store(
                sg, sub_a,
                accA.template get_multi_ptr<access::decorated::no>() +
-                   (sg_startx * TM) * N + sg_starty / SG_SZ * TN,
+                   (sg_startx * TM) * N + sg_starty / SG_SZ * TK,
                N);
          }); // parallel for
    }).wait();
@@ -87,7 +87,7 @@ void matrix_verify_sub(queue q, big_matrix<Ts, M, N> &A, nd_range<2> &r,
            ext::intel::experimental::matrix::joint_matrix_store(
                sg, sub_a,
                accA.template get_multi_ptr<access::decorated::no>() +
-                   (sg_startx * TM) * N + sg_starty / SG_SZ * TN,
+                   (sg_startx * TM) * N + sg_starty / SG_SZ * TK,
                N);
          }); // parallel for
    }).wait();
@@ -121,7 +121,7 @@ void matrix_verify_mul(queue q, big_matrix<Ts, M, N> &A, nd_range<2> &r,
            ext::intel::experimental::matrix::joint_matrix_store(
                sg, sub_a,
                accA.template get_multi_ptr<access::decorated::no>() +
-                   (sg_startx * TM) * N + sg_starty / SG_SZ * TN,
+                   (sg_startx * TM) * N + sg_starty / SG_SZ * TK,
                N);
          }); // parallel for
    }).wait();
@@ -156,7 +156,7 @@ void matrix_verify_div(queue q, big_matrix<Ts, M, N> &A, nd_range<2> &r,
            ext::intel::experimental::matrix::joint_matrix_store(
                sg, sub_a,
                accA.template get_multi_ptr<access::decorated::no>() +
-                   (sg_startx * TM) * N + sg_starty / SG_SZ * TN,
+                   (sg_startx * TM) * N + sg_starty / SG_SZ * TK,
                N);
          }); // parallel for
    }).wait();
@@ -206,7 +206,7 @@ void matrix_verify_logic(queue q, big_matrix<Ts, M, N> &A, nd_range<2> &r,
            ext::intel::experimental::matrix::joint_matrix_store(
                sg, sub_a,
                accA.template get_multi_ptr<access::decorated::no>() +
-                   (sg_startx * TM) * N + sg_starty / SG_SZ * TN,
+                   (sg_startx * TM) * N + sg_starty / SG_SZ * TK,
                N);
          }); // parallel for
    }).wait();
@@ -224,7 +224,7 @@ int main() {
   big_matrix<float, MATRIX_M, MATRIX_N> MA((float *)&A);
 
   size_t NDRangeM = MATRIX_M / TM;
-  size_t NDRangeN = MATRIX_N / TN;
+  size_t NDRangeN = MATRIX_N / TK;
   queue q;
   nd_range<2> r({NDRangeM, NDRangeN * SG_SZ}, {1, 1 * SG_SZ});