Enable autotuning level for triton test and disable fp8 dot tests

[hsharsha]

No description provided.

[i-chaochen]

d6d8535 I have disabled this due to triton backport is not done yet. wondering is it still need to disable?

[hsharsha]

• @zoranjovanovic-ns

[zoranjovanovic-ns]

At the moment, I do not see any reason why F32 should not be supported now.

[i-chaochen]

@hsharsha could you have a try based on Zoran's backport, please?

[hsharsha]

Yes the test fails with @zoranjovanovic-ns autoner enable commit
[ FAILED ] DotTestTestSuite/DotTest.IsTritonSupportedExecutesCorrectlyForDot/f32_dot, where GetParam() = (11, dot)

[hsharsha]

[ RUN      ] DotTestTestSuite/DotTest.IsTritonSupportedExecutesCorrectlyForDot/f32_dot
2025-02-07 14:58:39.560103: W ./xla/service/compiler.h:213] Ignoring the buffer assignment proto provided.
xla/service/gpu/fusions/triton/triton_support_legacy_test.cc:119: Failure
Value of: RunAndCompareNoHloPasses( std::move(ti.Module()), ErrorSpec{ primitive_util::IsF8Type(lhs_type) ? 1.0 : 2e-4, 2e-4})
  Actual: false (INTERNAL: Failed to launch ROCm kernel: triton_op with block dimensions: 256x1x1: HIP_ERROR_InvalidValue)Expected: true
[  FAILED  ] DotTestTestSuite/DotTest.IsTritonSupportedExecutesCorrectlyForDot/f32_dot, where GetParam() = (11, dot) (1045 ms)

[zoranjovanovic-ns]

I saw same error in some other tests too. Will investigate.