Skip to content

[AMP] Disallow fp16 conversion for arange op #8644

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
masahi merged 3 commits into from
Aug 4, 2021
Merged

[AMP] Disallow fp16 conversion for arange op #8644

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
b1f6467
[AMP] Do not allow fp16 cast on arange inputs
masahi Aug 4, 2021
082b1ff
add test
masahi Aug 4, 2021
e21470b
Add comment explaining the issue with fp16 "end"
masahi Aug 4, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions python/tvm/relay/transform/mixed_precision.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -128,6 +128,9 @@
# Error function doesn't seem to be able to be lowered into fp16 version in llvm.
# Move to follow list when it does.
"erf",
# Do not allow arange arguments (begin/end) to be fp16. "end" can be a big fp32 number
# not representable in fp16.
"arange",
]


Expand Down
11 changes: 11 additions & 0 deletions tests/python/relay/test_to_mixed_precision.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -230,6 +230,17 @@ def test_do_not_convert_softmax():
assert tvm.ir.structural_equal(mod, output_mod)


def test_do_not_convert_arange():
"""Arange is a red listed operation and therefore should never be fp16."""
dtype = "float32"
arange = relay.arange(relay.const(1, dtype), relay.const(128, dtype))
mod = tvm.IRModule.from_expr(arange)
mod = tvm.relay.transform.InferType()(mod)

output_mod = verify_mixed_precision_output_close(mod, {}, atol=0.0, rtol=0)
assert tvm.ir.structural_equal(mod, output_mod)


def test_green_gray_propagates_simple():
"""Conv is a green listed operation, while addition is gray.

Expand Down