On Wed, Jan 18, 2017 at 10:45 PM, David Malcolm <dmalc...@redhat.com> wrote:
> The jit testcase test-nested-loops.c was crashing.
>
> Root cause is that deep inside loop optimization we're now exposing
> this call within fold-const.c which wasn't being hit before:
>
> 4082 /* Compute the mask to access the bitfield. */
> 4083 unsigned_type = lang_hooks.types.type_for_size (*pbitsize, 1);
>
> and the jit's implementation of LANG_HOOKS_TYPE_FOR_SIZE was a
> placeholder that asserted it wasn't called.
>
> This patch implements a proper LANG_HOOKS_TYPE_FOR_SIZE for jit,
> by taking LTO's implementation.
>
> Fixes test-nested-loops.c, along with the related failures in
> test-combination.c and test-threads.c due to reusing the test.
>
> This fixes all known failures in jit.sum, putting it at 8609 passes.
>
> Committed to trunk as r244600.
I suppose we could instead make the lto hook the default (thus move it
to langhooks.c as lhd_type_for_size). Note similar issues may arise
from type_for_mode? Ah, I see you have that one...
Richard.
> gcc/jit/ChangeLog:
> * dummy-frontend.c (jit_langhook_type_for_size): Implement, using
> lto's lto_type_for_size.
> ---
> gcc/jit/dummy-frontend.c | 49
> ++++++++++++++++++++++++++++++++++++++++++++----
> 1 file changed, 45 insertions(+), 4 deletions(-)
>
> diff --git a/gcc/jit/dummy-frontend.c b/gcc/jit/dummy-frontend.c
> index 8f28e7f..5955854 100644
> --- a/gcc/jit/dummy-frontend.c
> +++ b/gcc/jit/dummy-frontend.c
> @@ -207,12 +207,53 @@ jit_langhook_type_for_mode (enum machine_mode mode, int
> unsignedp)
> return NULL;
> }
>
> +/* Return an integer type with PRECISION bits of precision,
> + that is unsigned if UNSIGNEDP is nonzero, otherwise signed. */
> +
> static tree
> -jit_langhook_type_for_size (unsigned int bits ATTRIBUTE_UNUSED,
> - int unsignedp ATTRIBUTE_UNUSED)
> +jit_langhook_type_for_size (unsigned precision, int unsignedp)
> {
> - gcc_unreachable ();
> - return NULL;
> + int i;
> +
> + if (precision == TYPE_PRECISION (integer_type_node))
> + return unsignedp ? unsigned_type_node : integer_type_node;
> +
> + if (precision == TYPE_PRECISION (signed_char_type_node))
> + return unsignedp ? unsigned_char_type_node : signed_char_type_node;
> +
> + if (precision == TYPE_PRECISION (short_integer_type_node))
> + return unsignedp ? short_unsigned_type_node : short_integer_type_node;
> +
> + if (precision == TYPE_PRECISION (long_integer_type_node))
> + return unsignedp ? long_unsigned_type_node : long_integer_type_node;
> +
> + if (precision == TYPE_PRECISION (long_long_integer_type_node))
> + return unsignedp
> + ? long_long_unsigned_type_node
> + : long_long_integer_type_node;
> +
> + for (i = 0; i < NUM_INT_N_ENTS; i ++)
> + if (int_n_enabled_p[i]
> + && precision == int_n_data[i].bitsize)
> + return (unsignedp ? int_n_trees[i].unsigned_type
> + : int_n_trees[i].signed_type);
> +
> + if (precision <= TYPE_PRECISION (intQI_type_node))
> + return unsignedp ? unsigned_intQI_type_node : intQI_type_node;
> +
> + if (precision <= TYPE_PRECISION (intHI_type_node))
> + return unsignedp ? unsigned_intHI_type_node : intHI_type_node;
> +
> + if (precision <= TYPE_PRECISION (intSI_type_node))
> + return unsignedp ? unsigned_intSI_type_node : intSI_type_node;
> +
> + if (precision <= TYPE_PRECISION (intDI_type_node))
> + return unsignedp ? unsigned_intDI_type_node : intDI_type_node;
> +
> + if (precision <= TYPE_PRECISION (intTI_type_node))
> + return unsignedp ? unsigned_intTI_type_node : intTI_type_node;
> +
> + return NULL_TREE;
> }
>
> /* Record a builtin function. We just ignore builtin functions. */
> --
> 1.8.5.3
>
