new tests

intel · Nov 21, 2024 · a7db97c · a7db97c
1 parent 7283c7b
commit a7db97c
Show file tree

Hide file tree

Showing 5 changed files with 10 additions and 0 deletions.
diff --git a/sycl/test-e2e/Matrix/element_wise_all_ops_impl.hpp b/sycl/test-e2e/Matrix/element_wise_all_ops_impl.hpp
@@ -229,6 +229,7 @@ int main() {
         test_ewops_c<float, 16, 16>();
 // This combination is not currently supported for sub group size = 32 in IGC
 #if (!defined(SG_SZ) || SG_SZ != 32)
+        test_ewops_ab<bfloat16,  1, 32, use::a, layout::row_major, 1>();
         test_ewops_ab<bfloat16, 32, 16, use::a, layout::row_major, 1>();
         test_ewops_ab<bfloat16, 32, 32, use::a, layout::row_major, 1>();
         test_ewops_ab<bfloat16, 16, 64, use::b, layout::ext_intel_packed, 2>();

diff --git a/sycl/test-e2e/Matrix/element_wise_ops_impl.hpp b/sycl/test-e2e/Matrix/element_wise_ops_impl.hpp
@@ -133,6 +133,7 @@ int main() {
       // IGC
       passed &= test<bfloat16, float, 16, 16, 16, 2, class pvc_bf16_16x16x16>();
       passed &= test<bfloat16, float, 1, 64, 16, 2, class pvc_bf16_1x64x16>();
+      passed &= test<bfloat16, float, 1, 64, 32, 2, class pvc_bf16_1x64x32>();
       passed &= test<bfloat16, float, 32, 64, 16, 2, class pvc_bf16_32x64x16>();
       passed &= test<bfloat16, float, 32, 64, 32, 2, class pvc_bf16_32x64x32>();
 #endif

diff --git a/sycl/test-e2e/Matrix/joint_matrix_bf16_fill_k_cache_impl.hpp b/sycl/test-e2e/Matrix/joint_matrix_bf16_fill_k_cache_impl.hpp
@@ -483,6 +483,10 @@ size_t matrix_size = -1;
            MCache1, NCache1, KCache1, MCache2, NCache2, KCache2>(matrix_size);
       test<bfloat16, float, VnniFactor, /*TM*/ 32, /*TN*/ 64, /*TK*/ 16,
            MCache1, NCache1, KCache1, MCache2, NCache2, KCache2>(matrix_size);
+      test<bfloat16, float, VnniFactor, /*TM*/ 1, /*TN*/ 64, /*TK*/ 32,
+           MCache1, NCache1, KCache1, MCache2, NCache2, KCache2>(matrix_size);
+      test<bfloat16, float, VnniFactor, /*TM*/ 32, /*TN*/ 64, /*TK*/ 32,
+           MCache1, NCache1, KCache1, MCache2, NCache2, KCache2>(matrix_size);
 #endif
       break;
     }

diff --git a/sycl/test-e2e/Matrix/joint_matrix_rowmajorA_rowmajorB_impl.hpp b/sycl/test-e2e/Matrix/joint_matrix_rowmajorA_rowmajorB_impl.hpp
@@ -127,6 +127,8 @@ int main() {
                               bfloat16, float>();
         res += gemm_row_major<1, 64, 16, class bf16_1x64x16, bfloat16, bfloat16,
                               float>();
+        res += gemm_row_major<1, 64, 32, class bf16_1x64x32, bfloat16, bfloat16,
+                              float>();
         res += gemm_row_major<32, 64, 16, class bf16_32x64x16, bfloat16,
                               bfloat16, float>();
         res += gemm_row_major<32, 64, 32, class bf16_32x64x32, bfloat16,

diff --git a/sycl/test-e2e/Matrix/joint_matrix_transposeC_impl.hpp b/sycl/test-e2e/Matrix/joint_matrix_transposeC_impl.hpp
@@ -124,13 +124,15 @@ int main() {
     }
 
     if (combinations[i].nsize == 16) { // architecture::intel_gpu_pvc
+      run_matrix_test<float, /*TM*/32, /*TN*/ 64>();
       run_matrix_test<float, /*TM*/ 8, /*TN*/ 16>();
       run_matrix_test<float, /*TM*/ 7, /*TN*/ 16>();
       run_matrix_test<float, /*TM*/ 6, /*TN*/ 16>();
       run_matrix_test<float, /*TM*/ 5, /*TN*/ 16>();
       run_matrix_test<float, /*TM*/ 4, /*TN*/ 16>();
       run_matrix_test<float, /*TM*/ 3, /*TN*/ 16>();
       run_matrix_test<float, /*TM*/ 2, /*TN*/ 16>();
+      run_matrix_test<float, /*TM*/ 1, /*TN*/ 64>();
       run_matrix_test<float, /*TM*/ 1, /*TN*/ 16>();
       break;
     }