On 11/10/20 10:59 AM, Stefan Kanthak via Gcc-patches wrote:
> The implementation of the __ashlDI3(), __ashrDI3() and __lshrDI3() functions
> is rather bad, it yields bad machine code at least on i386 and AMD64.
> Since GCC knows how to shift integers twice the register size these functions
> can be written as one-liners.
>
> The implementation of the __bswapsi2() function uses SIGNED instead of
> unsigned mask values; cf. __bswapdi2()
>
> Stefan Kanthak
>
> libgcc2.diff
>
> --- -/libgcc/libgcc2.h
> +++ +/libgcc/libgcc2.h
> @@ -391,7 +391,7 @@
> extern DWtype __negdi2 (DWtype);
> #endif
>
> -extern DWtype __lshrdi3 (DWtype, shift_count_type);
> +extern UDWtype __lshrdi3 (UDWtype, shift_count_type);
> extern DWtype __ashldi3 (DWtype, shift_count_type);
> extern DWtype __ashrdi3 (DWtype, shift_count_type);
>
> --- -/libgcc/libgcc2.c
> +++ +/libgcc/libgcc2.c
> @@ -398,30 +398,10 @@
> /* Unless shift functions are defined with full ANSI prototypes,
> parameter b will be promoted to int if shift_count_type is smaller than
> an int. */
> #ifdef L_lshrdi3
> -DWtype
> -__lshrdi3 (DWtype u, shift_count_type b)
> +UDWtype
> +__lshrdi3 (UDWtype u, shift_count_type b)
As has been pointed out, you can't implement these routines by doing the
operation in the same mode as the argument. It's potentially
self-recursive.
THe whole point of these routines is to provide double-word capabilities
on targets that don't have intrinsic double-word capabilities. That's
why they're written in a non-obvious way using word sized operations.
>
> @@ -486,10 +425,10 @@
> SItype
> __bswapsi2 (SItype u)
> {
> - return ((((u) & 0xff000000) >> 24)
> - | (((u) & 0x00ff0000) >> 8)
> - | (((u) & 0x0000ff00) << 8)
> - | (((u) & 0x000000ff) << 24));
> + return ((((u) & 0xff000000u) >> 24)
> + | (((u) & 0x00ff0000u) >> 8)
> + | (((u) & 0x0000ff00u) << 8)
> + | (((u) & 0x000000ffu) << 24));
What's the point of this change? I'm not sure how the signedness of the
constant really matters here.
jeff
