digitalmars.D - Weird DMD codegen issue
- Stanislav Blinov (15/15) Nov 04 2021 So I have a function:
- Temtaime (3/18) Nov 04 2021 Just don't use dmd, go with ldc
- user1234 (11/26) Nov 04 2021 As you cannot reduce I suggest to try with every .0 DMD version,
- max haughton (5/20) Nov 04 2021 Dustmite?
- Stanislav Blinov (3/5) Nov 05 2021 Narrowed it down somewhat. Thanks guys!
So I have a function:
```d
auto foo(C, K)(Strukt1* s, Strukt2 a, uint b, scope C c, scope K
k, out Strukt3 r)
{ /* ... */ }
```
Where return type is `Strukt4`. dmd borks the first argument on
call (at least on Linux). Workarounds so far I found:
1) adding `extern(C)`
2) reducing number of parameters - five is fine, as soon as I add
the sixth or more - it breaks (but not with `extern(C)`).
Problem is, I tried reproducing it outside of codebase, with no
success :( Does that sound familiar to anyone? Maybe there's
already a report about this on bugzilla? Any pointers on tracking
this down?
Nov 04 2021
Temtaime <temtaime gmail.com>
On Thursday, 4 November 2021 at 16:32:54 UTC, Stanislav Blinov
wrote:
So I have a function:
```d
auto foo(C, K)(Strukt1* s, Strukt2 a, uint b, scope C c, scope
K k, out Strukt3 r)
{ /* ... */ }
```
Where return type is `Strukt4`. dmd borks the first argument on
call (at least on Linux). Workarounds so far I found:
1) adding `extern(C)`
2) reducing number of parameters - five is fine, as soon as I
add the sixth or more - it breaks (but not with `extern(C)`).
Problem is, I tried reproducing it outside of codebase, with no
success :( Does that sound familiar to anyone? Maybe there's
already a report about this on bugzilla? Any pointers on
tracking this down?
Just don't use dmd, go with ldc
Nov 04 2021
On Thursday, 4 November 2021 at 16:32:54 UTC, Stanislav Blinov
wrote:
So I have a function:
```d
auto foo(C, K)(Strukt1* s, Strukt2 a, uint b, scope C c, scope
K k, out Strukt3 r)
{ /* ... */ }
```
Where return type is `Strukt4`. dmd borks the first argument on
call (at least on Linux). Workarounds so far I found:
1) adding `extern(C)`
2) reducing number of parameters - five is fine, as soon as I
add the sixth or more - it breaks (but not with `extern(C)`).
Problem is, I tried reproducing it outside of codebase, with no
success :( Does that sound familiar to anyone? Maybe there's
already a report about this on bugzilla? Any pointers on
tracking this down?
As you cannot reduce I suggest to try with every .0 DMD version,
degressively, to see if it's a regression.
Is it only with -O and are the structure very small? that could
be a SROA problem then.
Otherwise I have a specific PR in mind, that is [the big
backend](https://github.com/dlang/dmd/pull/10200) one that fixed
system V ABI. If it's the culprit, manually checking with
different DMD version will be useful (unless it did work
previously either)
Nov 04 2021
On Thursday, 4 November 2021 at 16:32:54 UTC, Stanislav Blinov
wrote:
So I have a function:
```d
auto foo(C, K)(Strukt1* s, Strukt2 a, uint b, scope C c, scope
K k, out Strukt3 r)
{ /* ... */ }
```
Where return type is `Strukt4`. dmd borks the first argument on
call (at least on Linux). Workarounds so far I found:
1) adding `extern(C)`
2) reducing number of parameters - five is fine, as soon as I
add the sixth or more - it breaks (but not with `extern(C)`).
Problem is, I tried reproducing it outside of codebase, with no
success :( Does that sound familiar to anyone? Maybe there's
already a report about this on bugzilla? Any pointers on
tracking this down?
Dustmite?
At very least can you give the definitions of the types used at
the call site where it breaks.
Nov 04 2021
On Thursday, 4 November 2021 at 17:46:02 UTC, max haughton wrote:At very least can you give the definitions of the types used at the call site where it breaks.Narrowed it down somewhat. Thanks guys! https://issues.dlang.org/show_bug.cgi?id=22484
Nov 05 2021