Further reduce large gemm tests

I suspect that we're running into more oom issues because of tharnsB
and transC and the like. So, move large gemm tests into their own file
where there'll only be one or two tests that run, thus ensuring we
don't overload the machines.

mlir/test/e2e/CMakeLists.txt

@@ -47,6 +47,7 @@ if (ROCK_E2E_TEST_ENABLED)
   if (ROCK_E2E_TEST_SUITES STREQUAL "" OR ROCK_E2E_TEST_SUITES STREQUAL "part3")
     list(APPEND CONFIGS
     Resnext101Config
+    LargeGemm
     PaddedGemmConfig
     )
   endif()

mlir/test/e2e/LargeGemm.toml

@@ -0,0 +1,20 @@
+directory = "LargeGemm"
+prefix = "rocmlir-gen"
+suffix = "--operation gemm --arch %arch %pv %random_data %rocmlir_gen_flags | rocmlir-driver -c | mlir-cpu-runner -O2 --shared-libs=%linalg_test_lib_dir/libmlir_rocm_runtime%shlibext,%conv_validation_wrapper_library_dir/libconv-validation-wrappers%shlibext,%linalg_test_lib_dir/libmlir_runner_utils%shlibext,%linalg_test_lib_dir/libmlir_float16_utils%shlibext --entry-point-result=void | FileCheck %s --check-prefix="
+
+# Note: Just f32 for ensuring lower machine load
+[[axis]]
+name = "data type"
+values = ["f32"]
+prefix = "-t "
+
+## Gemm variants
+[[suite]]
+name = "large_gemm"
+
+# Large gemm, input padding
+[[suite.test]]
+config = "-g 1 -m 1 -k 32768 -n 32769"
+# Large gemm, no padding.
+[[suite.test]]
+config = "-g 1 -m 64 -k 32768 -n 32768"