[SYCL][Matrix] Add matrix combinations 16x16x16 and 32x64x16 bfloat16 for PVC (intel#12026)

Author: dkhaldi

sycl/doc/extensions/experimental/sycl_ext_matrix/sycl_ext_oneapi_matrix.asciidoc

@@ -915,9 +915,11 @@ architecture::intel_gpu_dg2_g11, architecture::intel_gpu_dg2_g12`
 `matrix_type::fp32`   .2+|  +<=+ 8 |  16   .2+|  16 |
 `architecture::intel_gpu_pvc`|8| `architecture::intel_gpu_dg2_g10,
 architecture::intel_gpu_dg2_g11, architecture::intel_gpu_dg2_g12`
-.2+|  `matrix_type::bf16`       .2+|  `matrix_type::bf16`   .2+|
-`matrix_type::fp32`   .2+|  +<=+ 8 |  16   .2+|  16 |
-`architecture::intel_gpu_pvc` |8| `architecture::intel_gpu_dg2_g10,
+.4+|  `matrix_type::bf16`       .4+|  `matrix_type::bf16`   .4+|
+`matrix_type::fp32`  | 16 | 16 | 16 .3+|`architecture::intel_gpu_pvc` |
+32 | 64 | 16
+.2+|  +<=+ 8 |  16   .2+|  16
+|8 | `architecture::intel_gpu_dg2_g10,
 architecture::intel_gpu_dg2_g11, architecture::intel_gpu_dg2_g12`
 |  `matrix_type::tf32`       |  `matrix_type::tf32`   |
 `matrix_type::fp32`   |  +<=+ 8 |  16   |  8 |

sycl/source/detail/device_info.hpp

@@ -763,6 +763,10 @@ struct get_device_info_impl<
            matrix_type::fp32, matrix_type::fp32},
           {8, 0, 0, 0, 16, 16, matrix_type::bf16, matrix_type::bf16,
            matrix_type::fp32, matrix_type::fp32},
+          {0, 0, 0, 16, 16, 16, matrix_type::bf16, matrix_type::bf16,
+           matrix_type::fp32, matrix_type::fp32},
+          {0, 0, 0, 32, 64, 16, matrix_type::bf16, matrix_type::bf16,
+           matrix_type::fp32, matrix_type::fp32},
           {8, 0, 0, 0, 16, 8, matrix_type::tf32, matrix_type::tf32,
            matrix_type::fp32, matrix_type::fp32},
       };

sycl/test-e2e/Matrix/runtime_query_pvc.cpp

@@ -28,6 +28,10 @@ int main() {
        matrix_type::fp32, matrix_type::fp32},
       {8, 0, 0, 0, 16, 16, matrix_type::bf16, matrix_type::bf16,
        matrix_type::fp32, matrix_type::fp32},
+      {0, 0, 0, 16, 16, 16, matrix_type::bf16, matrix_type::bf16,
+       matrix_type::fp32, matrix_type::fp32},
+      {0, 0, 0, 32, 64, 16, matrix_type::bf16, matrix_type::bf16,
+       matrix_type::fp32, matrix_type::fp32},
       {8, 0, 0, 0, 16, 8, matrix_type::tf32, matrix_type::tf32,
        matrix_type::fp32, matrix_type::fp32},
   };