On 28 March 2017 at 12:42, Dibyendu Majumdar <mobile@xxxxxxxxxxxxxxx> wrote:
> I have this test:
>
> extern int printf(const char *, ...);
>
> int main(void)
> {
> int x = 6;
> const char txt[] = "local string %d %f %f\n";
> float g = 4.2;
> double e = 42.34;
>
> printf(txt, x, g, e);
>
> return 0;
> }
>
> When I generate the linearized output with simplifications disabled, I get:
>
> C:\d\github\dmr_c\tests\bugs>..\..\build\Debug\linearize.exe init1.c
> main:
> .L0:
> <entry-point>
> store.32 $6 -> 0[x]
> load.184* %r1 <- 0["local string %d %f %f\n"]
> store.184* %r1 -> 0[txt]
> set.f32 %r2 <- 4.200000
> store.f32 %r2 -> 0[g]
> set.f64 %r3 <- 42.340000
> store.f64 %r3 -> 0[e]
> symaddr.64* %r4 <- txt
> set.f64 %r5 <- 4.200000
> set.f64 %r6 <- 42.340000
> push.64* %r4
> push.32 $6
> push.f64 %r5
> push.f64 %r6
> call.32 %r7 <- printf
> phisrc.32 %phi1(return) <- $0
> br .L1
>
> .L1:
> phi.32 %r8 <- %phi1(return)
> ret.32 $0
>
> What is interesting is that the load and store ops have a size that is
> 184 bits. This scenario is not handled by sparse-llvm correctly.
>
> Presumably for the store here we need to copy the data? What about the load?
>
So it seems that when dealing with store we need to check if the
target type is an aggregate type - array, or struct - because stores
to these types are not yet implemented in sparse-llvm. I am not sure
whether load needs a check too.
Regards
Dibyendu
--
To unsubscribe from this list: send the line "unsubscribe linux-sparse" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
