internal narrowing conversion error in ppu/include/vectormath/cpp/quat_aos.h

[ParzivalWolfram]

Potential issue with PSL1GHT:

Reproduction code: make -f Makefile.ps3 from https://github.com/emukidid/wii64-ps3 current master (commit 85b23eeaf0c3ef224cdde6074bcda99339b6274f for future reference)

Error:

/usr/local/ps3dev/ppu/include/vectormath/cpp/quat_aos.h:502:100: error: narrowing conversion of '2147483648' from 'unsigned int' to 'int' inside { } [-Wnarrowing]
     return Quat( vec_xor( quat.get128(), ((vec_float4)(vec_int4){0x80000000,0x80000000,0x80000000,0}) ) );```

[miigotu]

You can add -Wno-narrowing to the cpp flags in that makefile.

It should work fine, but we should also explicitly static_cast that line or update the include.

[zeldin]

Is the value supposed to be 2147483648? In that case I guess the code should simply say
(vec_uint4){0x80000000U,0x80000000U,0x80000000U,0}
instead?

[ParzivalWolfram]

I'm unsure, but from the surrounding code, it seems 0x7FFFFFFF would be sufficient. I'm not the greatest at C, however, so take someone else's word for it, not mine.

…

On Thu, May 7, 2020 at 3:11 PM Marcus Comstedt ***@***.***> wrote: Is the value supposed to be 2147483648? In that case I guess the code should simply say (vec_uint4){0x80000000U,0x80000000U,0x80000000U,0} instead? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#70 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHE6CQNZNZZ5MVRPXS34X5DRQMIWRANCNFSM4M3ELFNQ> .

[zeldin]

Since the value is an operand to XOR, using 0x7FFFFFFF would give a very different result I think. The intention seems to be to invert the sign bit, not every bit except the sign bit. 😸

[zeldin]

Come to think of it, maybe the whole thing can be rewritten to
return Quat( vec_xor( quat.get128(), ((vec_float4){-0.0, -0.0, -0.0, 0.0})));
without involving any integers at all?

[ParzivalWolfram]

I didn't catch that that was the intent, like I said, i'm not great at C... hence the disclaimer.

…

On Thursday, May 7, 2020, Marcus Comstedt ***@***.***> wrote: Come to think of it, maybe the whole thing can be rewritten to return Quat( vec_xor( quat.get128(), ((vec_float4){-0.0, -0.0, -0.0, 0.0}))); without involving any integers at all? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#70 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHE6CQJUWX4T2DMSQINLKI3RQMKRPANCNFSM4M3ELFNQ> .

[miigotu]

Pretty sure just changing the cast to vec_uint4 is enough:

return Quat( vec_xor( quat.get128(), ((vec_float4)(vec_uint4){0x80000000,0x80000000,0x80000000,0}) ) );

I don't have the toolchain/libraries set up to test that out though

[bucanero]

I've got a similar problem trying to build the rsxtest sample from PSL1GHT ( https://github.com/ps3dev/PSL1GHT/tree/master/samples/graphics/rsxtest )

by running make, I get the same error about the narrowing conversion.

Following @miigotu comment, I tried casting to vec_uint4 (modified line 502 in /usr/local/ps3dev/ppu/include/vectormath/cpp/quat_aos.h )

Tried to compile again, and to my surprise, I've got a very weird error (at least for my limited knowledge):

% make
main.cpp
/Users/dparrino/github/psl1ght-libs/samples/psl1ght/graphics/rsxtest/source/main.cpp: In function 'int main(int, const char**)':
/Users/dparrino/github/psl1ght-libs/samples/psl1ght/graphics/rsxtest/source/main.cpp:556:6: warning: variable 'ret' set but not used [-Wunused-but-set-variable]
  s32 ret,i;
      ^~~
/Users/dparrino/github/psl1ght-libs/samples/psl1ght/graphics/rsxtest/source/main.cpp: In function 'void program_exit_callback()':
/Users/dparrino/github/psl1ght-libs/samples/psl1ght/graphics/rsxtest/source/main.cpp:410:1: internal compiler error: in rs6000_savres_routine_name, at config/rs6000/rs6000.c:28661
 }
 ^
libbacktrace could not find executable to open
Please submit a full bug report,
with preprocessed source if appropriate.
See <https://gcc.gnu.org/bugs/> for instructions.
make[1]: *** [main.o] Error 1
make: *** [build] Error 2

did I truly broke gcc, or I did something wrong on my side?

thanks for the help

[miigotu]

@bucanero the first error is easily worked around, by passing -Wno-unused-but-set-variable as a cflag.

[bucanero]

@bucanero the first error is easily worked around, by passing -Wno-unused-but-set-variable as a cflag.

oh yes, the unused variable is not a big deal, what I have no clue about is the internal compiler error 😕

/Users/dparrino/github/psl1ght-libs/samples/psl1ght/graphics/rsxtest/source/main.cpp:410:1: internal compiler error: in rs6000_savres_routine_name, at config/rs6000/rs6000.c:28661
 }
 ^
libbacktrace could not find executable to open
Please submit a full bug report,
with preprocessed source if appropriate.
See <https://gcc.gnu.org/bugs/> for instructions.

it's the first time I see gcc telling me to submit a bug report . I just wanted to check if it was a problem on my side , or someone else also got this issue when trying to build rsxtest

[miigotu]

Ideally someone should just update vectormath in PSL1GHT
In the updated library, there is a commit that fixes this same error:
erwincoumans/sce_vectormath@5cd9fd0

Maybe we should replace vectormath from here: https://github.com/erwincoumans/sce_vectormath/tree/master/include/vectormath

[miigotu]

@bucanero can you try checking out the bug/issue-70 branch and compile that?

[bucanero]

@miigotu sure, I'll check out and try to compile. 👍

I'll report back with my results.

[bucanero]

Sorry for the delay, using the fix from bug/issue-70 removed the narrowing conversion error, but the example still fails to compile.

I'll try #77 next.

Edit: at least from my tests, after applying patch from bug/issue-70 and removing the -Os compiler flag from the Makefile, everything worked. I could build the rsxtest example, and execute it correctly on my PS3.