vectorizer: Remove dead scalar .COND_* calls from vectorized loops [PR99767]

Hi!

The following testcase ICEs because disabling of DCE means there are dead
stmts in the loop (though, in theory they could become dead only shortly
before if-conv through some optimization), ifcvt which goes through all
stmts in the loop if-converts them into .COND_DIV etc. internal fn calls
in the copy of the loop meant for vectorization only, the loop is
successfully vectorized but the particular .COND_* call is not because
it isn't a live statement and the scalar .COND_* remains in the IL until
expansion where it ICEs because these ifns only support vectors and not
scalars.

These ifns are similar to .MASK_{LOAD,STORE} in this behavior.

One possible fix could be to expand scalar versions of them during
expansion, basically undoing what if-conv did to create them, i.e.
expand them as the lhs = else; if (mask) { lhs = statement; } or so.

For .MASK_LOAD we have code to replace them in vect_transform_loop already
though (not needed for .MASK_STORE, as stores should be always live
and thus always vectorized), so this patch instead replaces .COND_*
similarly to .MASK_LOAD in that loop, with the small difference
that lhs = .MASK_LOAD (...); is replaced by lhs = 0; while
lhs = .COND_* (..., else_arg); is replaced by lhs = else_arg.
The statement must be dead, otherwise it would be vectorized, so I think
it is not a big deal we don't turn it back into multiple basic blocks etc.
(and it might be not possible to do that at that point).

Bootstrapped/regtested on {x86_64,i686,aarch64}-linux, ok for trunk?

2021-04-16  Jakub Jelinek  <jakub@redhat.com>

	PR target/99767
	* tree-vect-loop.c (vect_transform_loop): Don't remove just
	dead scalar .MASK_LOAD calls, but also dead .COND_* calls - replace
	them by their last argument.

	* gcc.target/aarch64/pr99767.c: New test.


	Jakub

On Fri, 16 Apr 2021, Jakub Jelinek wrote:

> Hi!
> 
> The following testcase ICEs because disabling of DCE means there are dead
> stmts in the loop (though, in theory they could become dead only shortly
> before if-conv through some optimization), ifcvt which goes through all
> stmts in the loop if-converts them into .COND_DIV etc. internal fn calls
> in the copy of the loop meant for vectorization only, the loop is
> successfully vectorized but the particular .COND_* call is not because
> it isn't a live statement and the scalar .COND_* remains in the IL until
> expansion where it ICEs because these ifns only support vectors and not
> scalars.
> 
> These ifns are similar to .MASK_{LOAD,STORE} in this behavior.
> 
> One possible fix could be to expand scalar versions of them during
> expansion, basically undoing what if-conv did to create them, i.e.
> expand them as the lhs = else; if (mask) { lhs = statement; } or so.
> 
> For .MASK_LOAD we have code to replace them in vect_transform_loop already
> though (not needed for .MASK_STORE, as stores should be always live
> and thus always vectorized), so this patch instead replaces .COND_*
> similarly to .MASK_LOAD in that loop, with the small difference
> that lhs = .MASK_LOAD (...); is replaced by lhs = 0; while
> lhs = .COND_* (..., else_arg); is replaced by lhs = else_arg.
> The statement must be dead, otherwise it would be vectorized, so I think
> it is not a big deal we don't turn it back into multiple basic blocks etc.
> (and it might be not possible to do that at that point).
> 
> Bootstrapped/regtested on {x86_64,i686,aarch64}-linux, ok for trunk?

OK.

Thanks,
Richard.

> 2021-04-16  Jakub Jelinek  <jakub@redhat.com>
> 
> 	PR target/99767
> 	* tree-vect-loop.c (vect_transform_loop): Don't remove just
> 	dead scalar .MASK_LOAD calls, but also dead .COND_* calls - replace
> 	them by their last argument.
> 
> 	* gcc.target/aarch64/pr99767.c: New test.
> 
> --- gcc/tree-vect-loop.c.jj	2021-04-07 12:35:13.000000000 +0200
> +++ gcc/tree-vect-loop.c	2021-04-15 17:21:58.750101894 +0200
> @@ -9676,7 +9676,10 @@ vect_transform_loop (loop_vec_info loop_
>  	   !gsi_end_p (gsi); gsi_next (&gsi))
>  	{
>  	  gcall *call = dyn_cast <gcall *> (gsi_stmt (gsi));
> -	  if (call && gimple_call_internal_p (call, IFN_MASK_LOAD))
> +	  if (!call || !gimple_call_internal_p (call))
> +	    continue;
> +	  internal_fn ifn = gimple_call_internal_fn (call);
> +	  if (ifn == IFN_MASK_LOAD)
>  	    {
>  	      tree lhs = gimple_get_lhs (call);
>  	      if (!VECTOR_TYPE_P (TREE_TYPE (lhs)))
> @@ -9686,6 +9689,17 @@ vect_transform_loop (loop_vec_info loop_
>  		  gsi_replace (&gsi, new_stmt, true);
>  		}
>  	    }
> +	  else if (conditional_internal_fn_code (ifn) != ERROR_MARK)
> +	    {
> +	      tree lhs = gimple_get_lhs (call);
> +	      if (!VECTOR_TYPE_P (TREE_TYPE (lhs)))
> +		{
> +		  tree else_arg
> +		    = gimple_call_arg (call, gimple_call_num_args (call) - 1);
> +		  gimple *new_stmt = gimple_build_assign (lhs, else_arg);
> +		  gsi_replace (&gsi, new_stmt, true);
> +		}
> +	    }
>  	}
>      }				/* BBs in loop */
>  
> --- gcc/testsuite/gcc.target/aarch64/pr99767.c.jj	2021-04-15 17:28:10.466928380 +0200
> +++ gcc/testsuite/gcc.target/aarch64/pr99767.c	2021-04-15 17:28:45.653533566 +0200
> @@ -0,0 +1,16 @@
> +/* PR target/99767 */
> +/* { dg-do compile } */
> +/* { dg-options " -O1 -fopenmp-simd -fno-tree-dce -march=armv8-a+sve" } */
> +
> +int a[1024], b[1024];
> +
> +void
> +foo (void)
> +{
> +  #pragma omp simd
> +  for (int i = 0; i < 1024; i++)
> +    if (b[i] > 23) {
> +      a[i] = b[i] + 1;
> +      int v = 1 / 0;	/* { dg-warning "division by zero" } */
> +    }
> +}
> 
> 	Jakub
> 
>