struct field reordering and optimization#37429

Merged

bors merged 28 commits intorust-lang:masterfrom

ahicks92:univariant_layout_optimization

Dec 18, 2016

Commits on Dec 14, 2016

Optimize anything using a layout::Struct by introducing a mapping from source code field order to in-memory field order and sorting by alignment.
ahicks92
committed
Fix bugs to optimizing enums:
ahicks92
committed
Make constant field access account for field reordering.
ahicks92
committed
Fix extern-pass-empty test, which needed repr(C)
ahicks92
committed
Fix tuple and closure literals.
ahicks92
committed
Add yet more missing #[repr(C)] to tests
ahicks92
committed
Incorporate a bunch of review comments.
ahicks92
committed
Struct::new takes a vec, avoiding double allocation in some cases
ahicks92
committed
Modify debuginfo to deal with the difference between source and memory order
ahicks92
committed
Make tidy
ahicks92
committed
Use an enum to differentiate between kinds of structs.
ahicks92
committed
First attempt at detecting if structs can ever be unsized
ahicks92
committed
Don't optimize pairs
ahicks92
committed
Fix checking to see if the last field of a struct can be unsized.
ahicks92
committed
Fix type-sizes test
ahicks92
committed
Make tidy
ahicks92
committed
Change how type-sizes works slightly: we want to ensure that [i16; 0] introduces padding
ahicks92
committed
Fix having multiple reprs on the same type.
ahicks92
committed
Fix closure arguments which are immediate because of field reordering.
ahicks92
committed
Make tidy
ahicks92
committed
Some small fixes to how structs/enums are optimized
ahicks92
committed
Incorporate review comments
ahicks92
committed
Fix -Z print-type-sizes and tests.
ahicks92
committed
Fix error introduced during last rebase
ahicks92
committed
Fix computation of enum names based off the discrfield in the case of the null pointer optimization. This functionality is needed by pretty printers for gdb and lldb.
ahicks92
committed
Add a case to type-sizes to explicitly verify that field reordering triggers.
ahicks92
committed

Commits on Dec 16, 2016

Incorporate review comments.
ahicks92
committed
flock needs repr(C)
ahicks92
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

struct field reordering and optimization#37429

Commits on Dec 14, 2016

Optimize anything using a layout::Struct by introducing a mapping from source code field order to in-memory field order and sorting by alignment.

Fix bugs to optimizing enums:

Make constant field access account for field reordering.

Fix extern-pass-empty test, which needed repr(C)

Fix tuple and closure literals.

Add yet more missing #[repr(C)] to tests

Incorporate a bunch of review comments.

Struct::new takes a vec, avoiding double allocation in some cases

Modify debuginfo to deal with the difference between source and memory order

Make tidy

Use an enum to differentiate between kinds of structs.

First attempt at detecting if structs can ever be unsized

Don't optimize pairs

Fix checking to see if the last field of a struct can be unsized.

Fix type-sizes test

Make tidy

Change how type-sizes works slightly: we want to ensure that [i16; 0] introduces padding

Fix having multiple reprs on the same type.

Fix closure arguments which are immediate because of field reordering.

Make tidy

Some small fixes to how structs/enums are optimized

Incorporate review comments

Fix -Z print-type-sizes and tests.

Fix error introduced during last rebase

Fix computation of enum names based off the discrfield in the case of the null pointer optimization. This functionality is needed by pretty printers for gdb and lldb.

Add a case to type-sizes to explicitly verify that field reordering triggers.

Commits on Dec 16, 2016

Incorporate review comments.

flock needs repr(C)