[SYCL][Joint Matrix][Tests] Add tests for 16x16x16 and 32x64x32 joint_matrix shape combinations (#11649)

Note that I did not touch the file in XMX8 as optimal combinations on
DG2 are different. So it is better to address DG2 case as a separate
task

Author: dkhaldi

sycl/test-e2e/Matrix/SG32/joint_matrix_bfloat16_32x64.cpp renamed to sycl/test-e2e/Matrix/SG32/joint_matrix_bfloat16_16x16x16.cpp

@@ -1,4 +1,4 @@
-//==----- joint_matrix_bfloat16_32x64.cpp  - DPC++ joint_matrix-------------==//
+//==----- joint_matrix_bfloat16_16x16x16.cpp  - DPC++ joint_matrix----------==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
@@ -17,6 +17,9 @@
 using namespace sycl;
 using namespace sycl::ext::oneapi::experimental::matrix;
 
-constexpr size_t SG_SZ = 32;
+#define SG_SZ 32
+constexpr size_t TM = 16;
+constexpr size_t TN = 16;
+constexpr size_t TK = 16;
 
-#include "../joint_matrix_bfloat16_32x64_impl.hpp"
+#include "../joint_matrix_bfloat16_packedB_impl.hpp"

sycl/test-e2e/Matrix/SG32/joint_matrix_bfloat16_32x64x16.cpp

@@ -0,0 +1,25 @@
+//==----- joint_matrix_bfloat16_32x64x16.cpp  - DPC++ joint_matrix----------==//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+// REQUIRES: matrix
+
+// RUN: %{build} -o %t.out -DSYCL_EXT_ONEAPI_MATRIX_VERSION=4
+// RUN: %{run} %t.out
+
+// XFAIL: *
+
+#include "../common.hpp"
+
+using namespace sycl;
+using namespace sycl::ext::oneapi::experimental::matrix;
+
+#define SG_SZ 32
+constexpr size_t TM = 32;
+constexpr size_t TN = 64;
+constexpr size_t TK = 16;
+
+#include "../joint_matrix_bfloat16_packedB_impl.hpp"

sycl/test-e2e/Matrix/SG32/joint_matrix_bfloat16_32x64x32.cpp

@@ -0,0 +1,25 @@
+//==----- joint_matrix_bfloat16_32x64x32.cpp  - DPC++ joint_matrix----------==//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+// REQUIRES: matrix
+
+// RUN: %{build} -o %t.out -DSYCL_EXT_ONEAPI_MATRIX_VERSION=4
+// RUN: %{run} %t.out
+
+// XFAIL: *
+
+#include "../common.hpp"
+
+using namespace sycl;
+using namespace sycl::ext::oneapi::experimental::matrix;
+
+#define SG_SZ 32
+constexpr size_t TM = 32;
+constexpr size_t TN = 64;
+constexpr size_t TK = 32;
+
+#include "../joint_matrix_bfloat16_packedB_impl.hpp"

sycl/test-e2e/Matrix/joint_matrix_bfloat16_32x64.cpp renamed to sycl/test-e2e/Matrix/joint_matrix_bfloat16_16x16x16.cpp

@@ -1,4 +1,4 @@
-//==----- joint_matrix_bfloat16_32x64.cpp  - DPC++ joint_matrix-------------==//
+//==----- joint_matrix_bfloat16_16x16x16.cpp  - DPC++ joint_matrix----------==//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
@@ -18,5 +18,8 @@ using namespace sycl;
 using namespace sycl::ext::oneapi::experimental::matrix;
 
 #define SG_SZ 16
+constexpr size_t TM = 16;
+constexpr size_t TN = 16;
+constexpr size_t TK = 16;
 
-#include "joint_matrix_bfloat16_32x64_impl.hpp"
+#include "joint_matrix_bfloat16_packedB_impl.hpp"

sycl/test-e2e/Matrix/joint_matrix_bfloat16_32x64x16.cpp

@@ -0,0 +1,25 @@
+//==----- joint_matrix_bfloat16_32x64x16.cpp  - DPC++ joint_matrix----------==//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+// REQUIRES: matrix
+
+// RUN: %{build} -o %t.out -DSYCL_EXT_ONEAPI_MATRIX_VERSION=4
+// RUN: %{run} %t.out
+
+// XFAIL: *
+
+#include "common.hpp"
+
+using namespace sycl;
+using namespace sycl::ext::oneapi::experimental::matrix;
+
+#define SG_SZ 16
+constexpr size_t TM = 32;
+constexpr size_t TN = 64;
+constexpr size_t TK = 16;
+
+#include "joint_matrix_bfloat16_packedB_impl.hpp"

sycl/test-e2e/Matrix/joint_matrix_bfloat16_32x64x32.cpp

@@ -0,0 +1,25 @@
+//==----- joint_matrix_bfloat16_32x64x32.cpp  - DPC++ joint_matrix----------==//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+// REQUIRES: matrix
+
+// RUN: %{build} -o %t.out -DSYCL_EXT_ONEAPI_MATRIX_VERSION=4
+// RUN: %{run} %t.out
+
+// XFAIL: *
+
+#include "common.hpp"
+
+using namespace sycl;
+using namespace sycl::ext::oneapi::experimental::matrix;
+
+#define SG_SZ 16
+constexpr size_t TM = 32;
+constexpr size_t TN = 64;
+constexpr size_t TK = 32;
+
+#include "joint_matrix_bfloat16_packedB_impl.hpp"

sycl/test-e2e/Matrix/joint_matrix_bfloat16_32x64_impl.hpp renamed to sycl/test-e2e/Matrix/joint_matrix_bfloat16_packedB_impl.hpp

@@ -1,7 +1,3 @@
-#define TM 32
-#define TN 64
-#define TK 16
-
 template <typename T1, typename T2, size_t M, size_t N, size_t K>
 void matrix_multiply(big_matrix<T1, M, N> &C, big_matrix<T2, M, K> &A,
                      big_matrix<T2, K / 2, N * 2> &B) {