Implement ABI stuff for -preview=in #3578
Merged
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
|
I am very worried about the semantic meaning of My worry is that with this PR, we combine the decision that we must pass by reference to not violate D language semantics, with the decision that we may pass by ref as an optimization (without changing language semantics). Edit: looking a bit closer at the PR, I see that my fear is not justified (all was behind preview=in before aswell). |
|
Yeah, this just implements the ABI-specific part for LDC (well, at least the 3 major ABIs, but the 32-bit x86 ABI is IMO okay too this way). The upstream ABI part for DMD needs work - I've promised to implement it, so the frontend changes in here will probably be upstreamed more or less as-is. |
This is hard to prove (could be aliased via global state as well), and explicit |
Ping me when you make that PR. |
kinke
added a commit
to kinke/dmd
that referenced
this pull request
Oct 7, 2020
This is the promised follow-up on dlang#11000 and makes DMD exploit the specifics of the few supported ABIs (Win64, SysV x86_64, 32-bit x86). It also almost perfectly matches the proposed LDC implementation in ldc-developers/ldc#3578 (just a minor divergence for Win64 and dynamic arrays, but in that point the LDC and DMD Win64 ABI diverges in general).
kinke
added a commit
to kinke/dmd
that referenced
this pull request
Oct 7, 2020
This is the promised follow-up on dlang#11000 and makes DMD exploit the specifics of the few supported ABIs (Win64, SysV x86_64, 32-bit x86). It also almost perfectly matches the proposed LDC implementation in ldc-developers/ldc#3578 (just a minor divergence for Win64 and dynamic arrays, but in that point the LDC and DMD Win64 ABI diverges in general).
kinke
added a commit
to kinke/dmd
that referenced
this pull request
Oct 31, 2020
This is the promised follow-up on dlang#11000 and makes DMD exploit the specifics of the few supported ABIs (Win64, SysV x86_64, 32-bit x86). It also almost perfectly matches the proposed LDC implementation in ldc-developers/ldc#3578 (just a minor divergence for Win64 and dynamic arrays, but in that point the LDC and DMD Win64 ABI diverges in general).
kinke
added a commit
to kinke/ldc
that referenced
this pull request
Nov 7, 2020
I expect this to be slightly more performant than the previous behavior, where a delegate was treated like a corresponding struct, passed via hidden pointer and returned via sret. The primary motivation is a smooth preparation for PR ldc-developers#3578 - in order to allow people to experiment with `-preview=in` without recompiling druntime and Phobos, `in` slices and delegates must not be passed by-ref with `-preview=in` (see dlang/dmd#11828). This would have required a special case for delegates on Win64, which is IMO better handled this way.
kinke
added a commit
to kinke/ldc
that referenced
this pull request
Nov 7, 2020
I expect this to be slightly more performant than the previous behavior, where a delegate was treated like a corresponding struct, passed via hidden pointer and returned via sret. The primary motivation is a smooth preparation for PR ldc-developers#3578 - in order to allow people to experiment with `-preview=in` without recompiling druntime and Phobos, `in` slices and delegates must not be passed by-ref with `-preview=in` (see dlang/dmd#11828). This would have required a special case for delegates on Win64, which is IMO better handled this way.
AsterMiha
pushed a commit
to AsterMiha/dmd
that referenced
this pull request
Nov 9, 2020
This is the promised follow-up on dlang#11000 and makes DMD exploit the specifics of the few supported ABIs (Win64, SysV x86_64, 32-bit x86). It also almost perfectly matches the proposed LDC implementation in ldc-developers/ldc#3578 (just a minor divergence for Win64 and dynamic arrays, but in that point the LDC and DMD Win64 ABI diverges in general).
kinke
added a commit
that referenced
this pull request
Nov 11, 2020
…ers (#3609) I expect this to be slightly more performant than the previous behavior, where a delegate was treated like a corresponding struct, passed via hidden pointer and returned via sret. The primary motivation is a smooth preparation for PR #3578 - in order to allow people to experiment with `-preview=in` without recompiling druntime and Phobos, `in` slices and delegates must not be passed by-ref with `-preview=in` (see dlang/dmd#11828). This would have required a special case for delegates on Win64, which is IMO better handled this way.
kinke
added a commit
to ldc-developers/dmd-testsuite
that referenced
this pull request
Nov 26, 2020
This is the promised follow-up on #11000 and makes DMD exploit the specifics of the few supported ABIs (Win64, SysV x86_64, 32-bit x86). It also almost perfectly matches the proposed LDC implementation in ldc-developers/ldc#3578 (just a minor divergence for Win64 and dynamic arrays, but in that point the LDC and DMD Win64 ABI diverges in general).
kinke
force-pushed
the
preview_in
branch
2 times, most recently
from
a83a362
to
153c5f6
Compare
I.e., the ABI-specific decision for which parameter types to prefer passing a ref over a value. The x86_64 and AArch64 ABIs are pretty clear - use a ref if the POD type cannot be passed in registers. For all others, use a simple default heuristic based on the type size - if larger than 2 machine words, use a ref. Note that this includes x87 `real` for the 32-bit x86 ABI.
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.
I.e., the ABI-specific decision for which parameter types to prefer passing a ref over a value.
The x86_64 and AArch64 ABIs are pretty clear - use a ref if the POD type cannot be passed in registers. For all others, use a simple default heuristic based on the type size - if larger than 2 machine words, use a ref. Note that this includes x87
realfor the 32-bit x86 ABI.