Incorrectly warns about mixing code and declarations. about clspv HOT 9 OPEN

Could you provide an example showing this message please?

from clspv.

Here is an example kernel:

#define uint32_t        uint

__kernel
void foo
(
        __global uint32_t* __restrict__ a,
        __global uint32_t* __restrict__ b,
        __global uint32_t* __restrict__ c
)
{
        const uint32_t pindex = get_global_id(0);
        uint32_t aa = a[pindex] * a[pindex];
        if (aa < 1)
                aa = 1;
        uint32_t bb = b[pindex] * b[pindex];
        if (bb < 1)
                bb = 1;
        c[pindex] = aa + bb;
}

Which will give the compiler warning:

$ ~/src/clspv/build/bin/clspv foo.cl 
foo.cl:15:11: warning: mixing declarations and code is incompatible with standards before C99
   15 |         uint32_t bb = b[pindex] * b[pindex];
      |    

... which makes no sense, as OpenCL follows C99.

from clspv.

clspv is build to enable all warnings, which is equivalent to compiling with -Weverything.

Compiling a simple C code can produce the message even with C11:

$ clang -std=c11 test.c  -Weverything
test.c:7:9: warning: mixing declarations and code is incompatible with standards before C99 [-Wdeclaration-after-statement]
    int toto = 0;

For more information: llvm/llvm-project#53438

@alan-baker should we add no-declaration-after-statement in the warning list of clspv?

from clspv.

Note that the behavior is the same with gcc, expect that -Weverything does not exist, thus one need to use -Wdeclaration-after-statement to trigger the warning message (even with C11 or newer).

from clspv.

I think clspv needs to add a flag internally, because it will not accept it on its own command line:

~/src/clspv/build/bin/clspv -Wno-declaration-after-statement -o foo.spirv foo.cl 
clspv: Unknown command line argument '-Wno-declaration-after-statement'.  Try: '/home/bram/src/clspv/build/bin/clspv --help'
clspv: Did you mean '--enable-ext-tsp-block-placement'?

from clspv.

I don't think we want to make it configurable. Either we want to remove this warning from clspv, or we want to keep it like it is today.
Is this blocking something on your side for some reason?

from clspv.

No, it is not blocking. It is just unnecessary noise.

You can close it if you don't think it needs addressing.

from clspv.

I think this is worth discussing. Let's wait for @alan-baker opinion before closing it.
Thank you for reporting it.

from clspv.

I generally prefer to have as few deviations from clang as possible. If anything I'd rather expose an option that can be forwarded to clang to let the user control which warnings they want.

from clspv.