[SYCL Joint Matrix][E2E] Add 1d vector load/store tests for big combinations (#13814)

This change is partially addressing requirement to enable 1d block load
testing (IGC_JointMatrixLoadStoreOpt=2) for only big combinations
32x64x16, 1x64x16, 16x16x16.
This environment variable was added to as minimum tests as possible,
while keeping the coverage of big combinations to maximum.
This environment variable was not added to Matrix/element_wise_ops.cpp
yet, because, IGC compiler needs to be fixed to pass this test with
IGC_JointMatrixLoadStoreOpt=2

Author: YuriPlyakhin

sycl/test-e2e/Matrix/joint_matrix_bfloat16_packedB.cpp

@@ -10,6 +10,7 @@
 
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
+// RUN: %if gpu %{ env IGC_JointMatrixLoadStoreOpt=2 %{run} %t.out %}
 
 #include "common.hpp"
 #include "joint_matrix_bfloat16_packedB_impl.hpp"

sycl/test-e2e/Matrix/joint_matrix_rowmajorA_rowmajorB.cpp

@@ -11,6 +11,7 @@
 
 // RUN: %{build} -o %t.out
 // RUN: %{run} %t.out
+// RUN: %if gpu %{ env IGC_JointMatrixLoadStoreOpt=2 %{run} %t.out %}
 
 // This tests support of row major layout for matrix B which does automatic VNNI
 // transform. This is currently only available on AMX and XMX of PVC

sycl/test-e2e/Matrix/joint_matrix_rowmajorA_rowmajorB_impl.hpp

@@ -123,6 +123,8 @@ int main() {
                             int32_t>();
 
       if (combination.nsize == 16) { // architecture::intel_gpu_pvc
+        res += gemm_row_major<16, 16, 16, class bf16_16x16x16, bfloat16,
+                              bfloat16, float>();
         res += gemm_row_major<1, 64, 16, class bf16_1x64x16, bfloat16, bfloat16,
                               float>();
         res += gemm_row_major<32, 64, 16, class bf16_32x64x16, bfloat16,