[SWDEV-563266] [SWDEV-563264] Fix some conflicts related to inductor UTs (#2796)

Must have been some merge conflicts causing this

Author: jataylo

test/inductor/test_kernel_benchmark.py

@@ -172,8 +172,6 @@ def f(a, b):
         max_autotune=True, max_autotune_gemm_backends="TRITON", shape_padding=False
     )
     @fresh_cache()
-    @fresh_inductor_cache()
-    @skipIfRocm #This seems to be disabled upstream https://github.com/pytorch/pytorch/issues/118346
     def test_mm_triton_kernel_benchmark(self):
         M = 2048
         N = 2432

test/inductor/test_loop_ordering.py

@@ -415,10 +415,6 @@ def f(x):
         self.assertEqual(1, metrics.generated_kernel_count)
 
     @unittest.skipIf(not PLATFORM_SUPPORTS_FP8, "FP8 requires H100+ and MI300+")
-    @skipIfRocm
-    # Related PR: https://github.com/pytorch/pytorch/pull/149369
-    # This test can't function for ROCm because fp8 'mul_cuda' op is not supported
-    # in eager mode that is required here to check vs compiled results
     def test_fp8_cast_and_t(self):
         """
         This test repros the not able to fuses issue in
@@ -441,10 +437,6 @@ def f(x, scale):
         self.assertEqual(1, metrics.generated_kernel_count)
 
     @unittest.skipIf(not PLATFORM_SUPPORTS_FP8, "FP8 requires H100+ and MI300+")
-    @skipIfRocm
-    # Related PR: https://github.com/pytorch/pytorch/pull/149369
-    # This test can't function for ROCm because fp8 'mul_cuda' op is not supported
-    # in eager mode that is required here to check vs compiled results
     def test_fp8_pattern_2(self):
         """
         This test repros the fp8 fusion relation issue here: