Add mixed integer precision test for op_mul

[swolchok]

This tests a possibly-surprising result: int8(100) * int8(100) with output type of long is 16 in ATen, even though the output type can hold 10000.

[swolchok]

Stack from ghstack (oldest at bottom):

• Implement unary_ufunc functions using elementwise_util #9386
• -> Add mixed integer precision test for op_mul #11206
• Add SupportedTensorDtypes::BOOL #9584
• relax tolerances for all unary float ops #9585
• Add vectorization in elementwise_util #9432
• Add optimized_portable_kernels test #11205
• Add vectorized_math.h #11204

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/11206

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 1 New Failure, 1 Unrelated Failure

As of commit 9573d9f with merge base 71025df ():

NEW FAILURE - The following job has failed:

• Lint / lintrunner / linux-job (gh)
  >>> Lint for kernels/portable/cpu/util/dtype_util.h:

FLAKY - The following job failed but was likely due to flakiness present on trunk:

• pull / test-models-linux (resnet18, xnnpack-quantization-delegation, linux.2xlarge) / linux-job (gh) (detected as infra flaky with no log or failing log classifier)

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[kimishpatel]

wait what? I dont get why is it 16?

[manuelcandales]

wait what? I dont get why is it 16?

because 100 * 100 mod 256 = 16

this is how ATen works, because of functionalization
mul.out(x, y, out=out) must be equivalent to z = mul(x, y); out = z.to(out.dtype)
if x & y are int8, then z is also int8, and so multiplication turns into multiplication modulo 256
this is the downside of functionalization, it is sort of a compromise