Reduce the amount of llvm IR generated #1903

Marwes · 2021-01-29T11:06:22Z

Rusoto generates a lot of code for each of its services which is noticable in its long compile time. These few commits follows the same ideas as in #1773 . Commonly generated code are extracted into helper functions that can be reused and wherever possible this are written in a manner to not have any type parameters so the code is only instantiated once.

On the s3 crate this reduces the amount of generated LLVM IR by ~21% (400099 / 503440 = 0.79473025584)

Before

  Lines          Copies        Function name
  -----          ------        -------------
  503440 (100%)  10403 (100%)  (TOTAL)
   53787 (10.7%)  1359 (13.1%) core::ptr::drop_in_place
   46645 (9.3%)     95 (0.9%)  rusoto_s3::generated::S3Client::sign_and_dispatch::{{closure}}
   31374 (6.2%)    124 (1.2%)  rusoto_core::proto::xml::util::deserialize_elements
   25451 (5.1%)     42 (0.4%)  rusoto_core::proto::xml::util::parse_response::{{closure}}
   11934 (2.4%)    234 (2.2%)  <core::future::from_generator::GenFuture<T> as core::future::future::Future>::poll
   10296 (2.0%)    234 (2.2%)  core::pin::Pin<&mut T>::map_unchecked_mut
    6267 (1.2%)     98 (0.9%)  core::result::Result<T,E>::map_err
    6040 (1.2%)    117 (1.1%)  <core::option::Option<T> as core::fmt::Debug>::fmt
    5642 (1.1%)     31 (0.3%)  alloc::raw_vec::RawVec<T,A>::grow_amortized
    4631 (0.9%)    362 (3.5%)  <core::result::Result<T,E> as core::ops::try::Try>::from_error
    3787 (0.8%)    472 (4.5%)  core::pin::Pin<P>::new_unchecked
    3423 (0.7%)    161 (1.5%)  core::ops::function::FnOnce::call_once
    3325 (0.7%)     95 (0.9%)  <rusoto_core::error::RusotoError<E> as core::convert::From<rusoto_core::client::SignAndDispatchError>>::from
    3176 (0.6%)      1 (0.0%)  <rusoto_s3::generated::S3Client as rusoto_s3::generated::S3>::put_object::__put_object::{{closure}}
    3072 (0.6%)     32 (0.3%)  core::iter::traits::exact_size::ExactSizeIterator::len
    2890 (0.6%)    289 (2.8%)  <&T as core::fmt::Debug>::fmt
    2822 (0.6%)     34 (0.3%)  core::fmt::builders::DebugList::entries
    2560 (0.5%)     32 (0.3%)  <alloc::vec::Vec<T,A> as core::iter::traits::collect::IntoIterator>::into_iter
    2322 (0.5%)     27 (0.3%)  core::option::Option<T>::get_or_insert_with
    2320 (0.5%)    232 (2.2%)  core::future::from_generator
    2310 (0.5%)      1 (0.0%)  <rusoto_s3::generated::S3Client as rusoto_s3::generated::S3>::get_object::__get_object::{{closure}}
    2237 (0.4%)      1 (0.0%)  <rusoto_s3::generated::S3Client as rusoto_s3::generated::S3>::copy_object::__copy_object::{{closure}}
    2232 (0.4%)     31 (0.3%)  core::alloc::layout::Layout::array
    2178 (0.4%)     33 (0.3%)  alloc::raw_vec::RawVec<T,A>::current_memory
    2036 (0.4%)      1 (0.0%)  <rusoto_s3::generated::S3Client as rusoto_s3::generated::S3>::head_object::__head_object::{{closure}}
    2033 (0.4%)    406 (3.9%)  <core::result::Result<T,E> as core::ops::try::Try>::into_result
    1995 (0.4%)     95 (0.9%)  rusoto_s3::generated::S3Client::sign_and_dispatch

After

cargo llvm-lines --lib | head -30
  Lines          Copies       Function name
  -----          ------       -------------
  400099 (100%)  8535 (100%)  (TOTAL)
   38832 (9.7%)  1131 (13.3%) core::ptr::drop_in_place
   31374 (7.8%)   124 (1.5%)  rusoto_core::proto::xml::util::deserialize_elements
    6511 (1.6%)   102 (1.2%)  core::result::Result<T,E>::map_err
    6040 (1.5%)   117 (1.4%)  <core::option::Option<T> as core::fmt::Debug>::fmt
    5642 (1.4%)    31 (0.4%)  alloc::raw_vec::RawVec<T,A>::grow_amortized
    5049 (1.3%)    99 (1.2%)  <core::future::from_generator::GenFuture<T> as core::future::future::Future>::poll
    4356 (1.1%)    99 (1.2%)  core::pin::Pin<&mut T>::map_unchecked_mut
    3501 (0.9%)   164 (1.9%)  core::ops::function::FnOnce::call_once
    3420 (0.9%)   269 (3.2%)  <core::result::Result<T,E> as core::ops::try::Try>::from_error
    3072 (0.8%)    32 (0.4%)  core::iter::traits::exact_size::ExactSizeIterator::len
    2907 (0.7%)     1 (0.0%)  <rusoto_s3::generated::S3Client as rusoto_s3::generated::S3>::put_object::__put_object::{{closure}}
    2890 (0.7%)   289 (3.4%)  <&T as core::fmt::Debug>::fmt
    2822 (0.7%)    34 (0.4%)  core::fmt::builders::DebugList::entries
    2560 (0.6%)    32 (0.4%)  <alloc::vec::Vec<T,A> as core::iter::traits::collect::IntoIterator>::into_iter
    2322 (0.6%)    27 (0.3%)  core::option::Option<T>::get_or_insert_with
    2232 (0.6%)    31 (0.4%)  core::alloc::layout::Layout::array
    2221 (0.6%)     1 (0.0%)  <rusoto_s3::generated::S3Client as rusoto_s3::generated::S3>::get_object::__get_object::{{closure}}
    2178 (0.5%)    33 (0.4%)  alloc::raw_vec::RawVec<T,A>::current_memory
    1947 (0.5%)     1 (0.0%)  <rusoto_s3::generated::S3Client as rusoto_s3::generated::S3>::head_object::__head_object::{{closure}}
    1938 (0.5%)     1 (0.0%)  <rusoto_s3::generated::S3Client as rusoto_s3::generated::S3>::copy_object::__copy_object::{{closure}}
    1855 (0.5%)    35 (0.4%)  core::slice::iter::Iter<T>::new
    1766 (0.4%)     1 (0.0%)  <rusoto_s3::generated::S3Client as rusoto_s3::generated::S3>::create_multipart_upload::__create_multipart_upload::{{closure}}
    1733 (0.4%)    31 (0.4%)  alloc::vec::Vec<T,A>::push
    1715 (0.4%)    35 (0.4%)  core::slice::iter::Iter<T>::post_inc_start
    1680 (0.4%)    42 (0.5%)  rusoto_core::proto::xml::util::<impl core::convert::From<rusoto_core::proto::xml::util::XmlResponseError> for rusoto_core::error::RusotoError<E>>::from
    1627 (0.4%)   202 (2.4%)  core::pin::Pin<P>::new_unchecked
    1610 (0.4%)    35 (0.4%)  <core::slice::iter::Iter<T> as core::iter::traits::iterator::Iterator>::next

Reduces the llvm-lines for s3 to 400k from 460k (-13%)

Removes another ~6% of llvm-lines from s3

dacut

I'm not a maintainer, but in case a maintainer needs some time saving here: I've reviewed these, and they're all what they say they do -- getting rid of extra .to_string() calls littered throughout the code.

(This applies to commits aeffede and 92f680c).

dacut

Not a maintainer -- just hoping this helps the maintainers.
Just had one question about the change from the HttpDispatch error variant to Unknown in a couple places.

dacut · 2021-04-02T22:24:48Z

service_crategen/src/commands/generate/codegen/query.rs

                    let mut response = self.client.sign_and_dispatch(request).await?;
                    if !response.status.is_success() {{
-                        let response = response.buffer().await.map_err(RusotoError::HttpDispatch)?;


What is the rationale for changing the returned variant from HttpDispatch to Unknown here? I'm imagine this might break a few clients, and I would think HttpDispatch is more appropriate -- but feel free to set me straight here.

I think it should still be HttpDispatch though? It just gets wrapped by the From impl instead?

dacut · 2021-04-02T22:26:36Z

service_crategen/src/commands/generate/codegen/json.rs

                    let mut response = self.client.sign_and_dispatch(request).await?;
                    if !response.status.is_success() {{
-                        let response = response.buffer().await.map_err(RusotoError::HttpDispatch)?;


Same comment as for codegen/query.rs -- not sure about the rationale for changing from the HttpDispatch variant to Unknown.

Markus Westerlind added 7 commits January 11, 2021 16:33

perf: Avoid doing to_string on string elements

aeffede

perf: Avoid calling to_string on String fields

92f680c

perf: Make the sign_and_dispatch helper non-generic

38a8728

Reduces the llvm-lines for s3 to 400k from 460k (-13%)

Shrink parse_response a bit

4daebb5

perf: Make parse_response non-generic

4a86edc

Removes another ~6% of llvm-lines from s3

Avoid calling as_ref() repeatedly

a224a33

No need for refine in rest_json

b3991f4

Marwes mentioned this pull request Jan 29, 2021

Reduce the amount of llvm IR generated #1890

Merged

dacut reviewed Apr 2, 2021

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Reduce the amount of llvm IR generated #1903

Reduce the amount of llvm IR generated #1903

Marwes commented Jan 29, 2021

dacut left a comment •

edited

dacut left a comment

dacut Apr 2, 2021

Marwes Apr 4, 2021

dacut Apr 2, 2021

Reduce the amount of llvm IR generated #1903

Are you sure you want to change the base?

Reduce the amount of llvm IR generated #1903

Conversation

Marwes commented Jan 29, 2021

Before

After

dacut left a comment • edited

Choose a reason for hiding this comment

dacut left a comment

Choose a reason for hiding this comment

dacut Apr 2, 2021

Choose a reason for hiding this comment

Marwes Apr 4, 2021

Choose a reason for hiding this comment

dacut Apr 2, 2021

Choose a reason for hiding this comment

dacut left a comment •

edited