On Fri, May 26, 2023 at 4:12 AM Jiang, Haochen <haochen.ji...@intel.com> wrote:
>
> > gcc/ChangeLog:
> >
> > * config/i386/i386-expand.cc (ix86_expand_vecop_qihi2):
> > Rewrite to expand to 2x-wider (e.g. V16QI -> V16HImode)
> > instructions when available. Emulate truncation via
> > ix86_expand_vec_perm_const_1 when native truncate insn
> > is not available.
> > (ix86_expand_vecop_qihi_partial) <case MULT>: Use pmovzx
> > when available. Trivially rename some variables.
> > (ix86_expand_vecop_qihi): Unconditionally call ix86_expand_vecop_qihi2.
>
> Hi Uros,
>
> I suppose you pushed wrong patch to trunk.
Ouch...
I was preparing before/after asm sequences for the ChangeLog entry and
took a shortcut by disabling the call in my main devel worktree. I
forgot to remove it before committing...
Anyway, the regression test was done with the correct version (there
are several tests already present in the testsuite that would fail
otherwise). I fixed it with an obvious patch.
gcc/ChangeLog:
* config/i386/i386-expand.cc (ix86_expand_vecop_qihi):
Do not disable call to ix86_expand_vecop_qihi2.
Thanks for spotting this oversight!
Uros.
> On trunk, we see this:
>
> @@ -23409,9 +23457,7 @@ ix86_expand_vecop_qihi (enum rtx_code code, rtx dest,
> rtx op1, rtx op2)
> && ix86_expand_vec_shift_qihi_constant (code, dest, op1, op2))
> return;
>
> - if (TARGET_AVX512BW
> - && VECTOR_MODE_P (GET_MODE (op2))
> - && ix86_expand_vecop_qihi2 (code, dest, op1, op2))
> + if (0 && ix86_expand_vecop_qihi2 (code, dest, op1, op2))
> return;
>
> switch (qimode)
>
> It should not be if (0 && ix86_expand_vecop_qihi2 (code, dest, op1, op2))
>
> The patch in this thread is correct, where is:
>
> @@ -23409,9 +23457,7 @@ ix86_expand_vecop_qihi (enum rtx_code code, rtx dest,
> rtx op1, rtx op2)
> && ix86_expand_vec_shift_qihi_constant (code, dest, op1, op2))
> return;
>
> - if (TARGET_AVX512BW
> - && VECTOR_MODE_P (GET_MODE (op2))
> - && ix86_expand_vecop_qihi2 (code, dest, op1, op2))
> + if (ix86_expand_vecop_qihi2 (code, dest, op1, op2))
> return;
>
> switch (qimode)
>
> Thx,
> Haochen
>
> > * config/i386/i386.cc (ix86_multiplication_cost): Rewrite cost
> > calculation of V*QImode emulations to account for generation of
> > 2x-wider mode instructions.
> > (ix86_shift_rotate_cost): Update cost calculation of V*QImode
> > emulations to account for generation of 2x-wider mode instructions.
> >
> > gcc/testsuite/ChangeLog:
> >
> > * gcc.target/i386/avx512vl-pr95488-1.c: Revert 2023-05-18 change.
> >
> > Bootstrapped and regression tested on x86_64-linux-gnu {,-m32}.
> >
> > Uros.
