[Topi][Unittests] Parametrized tests in test_topi_dense.py, split out gpu-independent implementations
#8336
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Lunderberg
force-pushed
the
topi_dense_parametrize
branch
2 times, most recently
from
0dde14d to
98a145b
Compare
jcf94
reviewed
Jun 28, 2021
| return (a_np, b_np, c_np, d_np) | ||
|
|
||
|
|
||
| def test_dense( |
There was a problem hiding this comment.
I'm thinking that wheather this can test different targets as you expected.
Shouldn't this function be decorated with "@tvm.testing.parametrize_targets"?
There was a problem hiding this comment.
As of #8010 , the decorator for @tvm.testing.parametrize_targets is only needed if the test should override the default targets. Otherwise, having the target and/or dev fixture arguments is sufficient to have the test run on all enabled targets.
There was a problem hiding this comment.
Oh, my bad. That really a helpful feature.
Now, tests run for multiple data types, can be extended with additional datatypes.
Lunderberg
force-pushed
the
topi_dense_parametrize
branch
from
98a145b to
4786a21
Compare
|
Added a few fixes in the codegen needed for the vulkan tests to run correctly on newer nvidia GPUs. (@masahi ) |
…ense As a follow-up to the renaming of "gpu" to "cuda", separating implementations that require CUDA (e.g. dense_cublas.cuda) from implementations that require any GPU, but not necessarily a CUDA GPU (e.g. dense_small_batch.gpu). My intent is to pair this migration with the extension of unit tests to cover additional GPU runtimes, migrating only implementations that run correctly on non-CUDA GPU devices.
…lts on some GPUs - In ThreadAllreduceBuilder, separate out load/store so that they can have a memory barrier in-between. - In Vulkan codegen, added Workgroup memory sync for subgroup thread sync, since the different subgroup threads can still access workgroup memory. Longer-term, may need tir enhancements to separate out sync of control/memory.
Lunderberg
force-pushed
the
topi_dense_parametrize
branch
from
4786a21 to
adb2431
Compare
ylc
pushed a commit
to ylc/tvm
that referenced
this pull request
Sep 29, 2021
…ut gpu-independent implementations (apache#8336) * [Topi][UnitTests] Parametrized tests in test_topi_dense.py Now, tests run for multiple data types, can be extended with additional datatypes. * [Topi] Separated generic-gpu nn.dense implementations into topi.gpu.dense As a follow-up to the renaming of "gpu" to "cuda", separating implementations that require CUDA (e.g. dense_cublas.cuda) from implementations that require any GPU, but not necessarily a CUDA GPU (e.g. dense_small_batch.gpu). My intent is to pair this migration with the extension of unit tests to cover additional GPU runtimes, migrating only implementations that run correctly on non-CUDA GPU devices. * [Vulkan][Codegen] Updated storage sync to avoid incorrect matmul results on some GPUs - In ThreadAllreduceBuilder, separate out load/store so that they can have a memory barrier in-between. - In Vulkan codegen, added Workgroup memory sync for subgroup thread sync, since the different subgroup threads can still access workgroup memory. Longer-term, may need tir enhancements to separate out sync of control/memory. Co-authored-by: Eric Lunderberg <[email protected]>
zxy844288792
pushed a commit
to zxy844288792/tvm
that referenced
this pull request
Mar 4, 2022
…ut gpu-independent implementations (apache#8336) * [Topi][UnitTests] Parametrized tests in test_topi_dense.py Now, tests run for multiple data types, can be extended with additional datatypes. * [Topi] Separated generic-gpu nn.dense implementations into topi.gpu.dense As a follow-up to the renaming of "gpu" to "cuda", separating implementations that require CUDA (e.g. dense_cublas.cuda) from implementations that require any GPU, but not necessarily a CUDA GPU (e.g. dense_small_batch.gpu). My intent is to pair this migration with the extension of unit tests to cover additional GPU runtimes, migrating only implementations that run correctly on non-CUDA GPU devices. * [Vulkan][Codegen] Updated storage sync to avoid incorrect matmul results on some GPUs - In ThreadAllreduceBuilder, separate out load/store so that they can have a memory barrier in-between. - In Vulkan codegen, added Workgroup memory sync for subgroup thread sync, since the different subgroup threads can still access workgroup memory. Longer-term, may need tir enhancements to separate out sync of control/memory. Co-authored-by: Eric Lunderberg <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
[Topi][UnitTests] Parametrized tests in test_topi_dense.py
Now, tests run for multiple data types, can be extended with additional datatypes.
[Topi] Separated generic-gpu nn.dense implementations into topi.gpu.dense
As a follow-up to the renaming of "gpu" to "cuda", separating implementations that require CUDA (e.g. dense_cublas.cuda) from implementations that require any GPU, but not necessarily a CUDA GPU (e.g. dense_small_batch.gpu).
My intent is to pair this migration with the extension of unit tests to cover additional GPU runtimes, migrating only implementations that run correctly on non-CUDA GPU devices.