On Wed, 20 Dec 2023, Jakub Jelinek wrote:
> Hi!
>
> The following patch fixes 2 issues in handling of casts for mergeable
> stmts.
> The first hunk fixes the case when we have two nested casts (typically
> after optimization that is zero-extension of a sign-extension because
> everything else should have been folded into a single cast). If
> the lowering of the outer cast needs to make the code conditional
> (e.g.
> for (...)
> {
> if (idx <= 32)
> {
> if (idx < 32)
> { ... handle_operand (idx); ... }
> else
> { ... handle_operand (32); ... }
> }
> ...
> }
> ) and the lowering of the inner one as well, right now it creates invalid
> SSA form, because even for the inner cast we need a PHI on the loop
> and the PHI argument from the latch edge is a SSA_NAME initialized in
> the conditionally executed bb. The hunk fixes that by detecting such
> a case and adding further PHI nodes at the end of the ifs such that
> the right value propagates to the next loop iteration. We can use
> 0 arguments for the other edges because the inner operand handling
> is only done for the first set of iterations and then the other ifs take
> over.
>
> The rest fixes a case of again invalid SSA form, when for a sign extension
> we need to use the 0 or -1 value initialized by earlier iteration in
> a constant idx case, the code was using the value of the loop PHI argument
> from latch edge rather than result; that is correct for cases expanded
> in straight line code after the loop, but not inside of the loop for the
> cases of handle_cast conditionals, there we should use PHI result. This
> is done in the second hunk and supported by the remaining hunks, where
> it clears m_bb to tell the code we aren't in the loop anymore.
>
> Note, this patch doesn't deal with similar problems during multiplication,
> division, floating casts etc. where we just emit a library call. I'll
> need to make sure in that case we don't merge more than one cast per
> operand.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
OK.
Richard.
> 2023-12-20 Jakub Jelinek <ja...@redhat.com>
>
> PR tree-optimization/112941
> * gimple-lower-bitint.cc (bitint_large_huge::handle_cast): If
> save_cast_conditional, instead of adding assignment of t4 to
> m_data[save_data_cnt + 1] before m_gsi, add phi nodes such that
> t4 propagates to m_bb loop. For constant idx, use
> m_data[save_data_cnt] rather than m_data[save_data_cnt + 1] if inside
> of the m_bb loop.
> (bitint_large_huge::lower_mergeable_stmt): Clear m_bb when no longer
> expanding inside of that loop.
> (bitint_large_huge::lower_comparison_stmt): Likewise.
> (bitint_large_huge::lower_addsub_overflow): Likewise.
> (bitint_large_huge::lower_mul_overflow): Likewise.
> (bitint_large_huge::lower_bit_query): Likewise.
>
> * gcc.dg/bitint-55.c: New test.
>
> --- gcc/gimple-lower-bitint.cc.jj 2023-12-15 10:10:51.000000000 +0100
> +++ gcc/gimple-lower-bitint.cc 2023-12-19 18:32:52.937388747 +0100
> @@ -1401,8 +1401,29 @@ bitint_large_huge::handle_cast (tree lhs
> add_phi_arg (phi, m_data[save_data_cnt], edge_false,
> UNKNOWN_LOCATION);
> add_phi_arg (phi, ext, edge_true_true, UNKNOWN_LOCATION);
> - g = gimple_build_assign (m_data[save_data_cnt + 1], t4);
> - insert_before (g);
> + if (!save_cast_conditional)
> + {
> + g = gimple_build_assign (m_data[save_data_cnt + 1], t4);
> + insert_before (g);
> + }
> + else
> + for (basic_block bb = gsi_bb (m_gsi);;)
> + {
> + edge e1 = single_succ_edge (bb);
> + edge e2 = find_edge (e1->dest, m_bb), e3;
> + tree t5 = (e2 ? m_data[save_data_cnt + 1]
> + : make_ssa_name (m_limb_type));
> + phi = create_phi_node (t5, e1->dest);
> + edge_iterator ei;
> + FOR_EACH_EDGE (e3, ei, e1->dest->preds)
> + add_phi_arg (phi, (e3 == e1 ? t4
> + : build_zero_cst (m_limb_type)),
> + e3, UNKNOWN_LOCATION);
> + if (e2)
> + break;
> + t4 = t5;
> + bb = e1->dest;
> + }
> }
> if (m_bitfld_load)
> {
> @@ -1470,6 +1491,8 @@ bitint_large_huge::handle_cast (tree lhs
> m_data_cnt = tree_to_uhwi (m_data[save_data_cnt + 2]);
> if (TYPE_UNSIGNED (rhs_type))
> t = build_zero_cst (m_limb_type);
> + else if (m_bb)
> + t = m_data[save_data_cnt];
> else
> t = m_data[save_data_cnt + 1];
> }
> @@ -2586,6 +2609,7 @@ bitint_large_huge::lower_mergeable_stmt
> m_gsi = gsi_after_labels (edge_bb);
> else
> m_gsi = gsi_for_stmt (stmt);
> + m_bb = NULL;
> }
> }
> }
> @@ -2712,6 +2736,7 @@ bitint_large_huge::lower_mergeable_stmt
> NULL_TREE, NULL_TREE);
> insert_before (g);
> m_gsi = gsi_for_stmt (stmt);
> + m_bb = NULL;
> }
> }
> }
> @@ -2890,6 +2915,7 @@ bitint_large_huge::lower_comparison_stmt
> extract_true_false_edges_from_block (gsi_bb (m_gsi),
> &true_edge, &false_edge);
> m_gsi = gsi_after_labels (false_edge->dest);
> + m_bb = NULL;
> }
> }
>
> @@ -4208,6 +4234,7 @@ bitint_large_huge::lower_addsub_overflow
> NULL_TREE, NULL_TREE);
> insert_before (g);
> m_gsi = gsi_for_stmt (final_stmt);
> + m_bb = NULL;
> }
> }
> }
> @@ -4405,6 +4432,7 @@ bitint_large_huge::lower_mul_overflow (t
> &true_edge,
> &false_edge);
> m_gsi = gsi_after_labels (false_edge->dest);
> + m_bb = NULL;
> }
> }
>
> @@ -4744,6 +4772,7 @@ bitint_large_huge::lower_bit_query (gimp
> m_gsi = gsi_after_labels (edge_bb);
> else
> m_gsi = gsi_for_stmt (stmt);
> + m_bb = NULL;
> }
> }
> }
> @@ -4905,6 +4934,7 @@ bitint_large_huge::lower_bit_query (gimp
> extract_true_false_edges_from_block (gsi_bb (m_gsi),
> &true_edge, &false_edge);
> m_gsi = gsi_after_labels (false_edge->dest);
> + m_bb = NULL;
> }
> }
> }
> --- gcc/testsuite/gcc.dg/bitint-55.c.jj 2023-12-19 23:37:41.161537400
> +0100
> +++ gcc/testsuite/gcc.dg/bitint-55.c 2023-12-19 23:37:08.886986817 +0100
> @@ -0,0 +1,129 @@
> +/* PR tree-optimization/112941 */
> +/* { dg-do compile { target bitint } } */
> +/* { dg-options "-std=c23 -O2" } */
> +
> +#if __BITINT_MAXWIDTH__ >= 4096
> +void
> +f1 (_BitInt(4096) *p, int r, _BitInt(115) s, _BitInt(128) t, _BitInt(231) u)
> +{
> + p[0] += (unsigned _BitInt(2048)) r;
> + p[1] += (unsigned _BitInt(2048)) s;
> + p[2] += (unsigned _BitInt(2048)) t;
> + p[3] += (unsigned _BitInt(2048)) u;
> +}
> +
> +void
> +f2 (_BitInt(4094) *p, int r, _BitInt(115) s, _BitInt(128) t, _BitInt(231) u)
> +{
> + p[0] -= (unsigned _BitInt(2048)) r;
> + p[1] -= (unsigned _BitInt(2048)) s;
> + p[2] -= (unsigned _BitInt(2048)) t;
> + p[3] -= (unsigned _BitInt(2048)) u;
> +}
> +
> +void
> +f3 (_BitInt(4096) *p, int r, _BitInt(115) s, _BitInt(128) t, _BitInt(231) u)
> +{
> + p[0] += (unsigned _BitInt(2110)) r;
> + p[1] += (unsigned _BitInt(2110)) s;
> + p[2] += (unsigned _BitInt(2110)) t;
> + p[3] += (unsigned _BitInt(2110)) u;
> +}
> +
> +void
> +f4 (_BitInt(4094) *p, int r, _BitInt(115) s, _BitInt(128) t, _BitInt(231) u)
> +{
> + p[0] -= (unsigned _BitInt(2110)) r;
> + p[1] -= (unsigned _BitInt(2110)) s;
> + p[2] -= (unsigned _BitInt(2110)) t;
> + p[3] -= (unsigned _BitInt(2110)) u;
> +}
> +
> +void
> +f5 (unsigned _BitInt(4096) *p, int r, _BitInt(115) s, _BitInt(128) t,
> _BitInt(231) u)
> +{
> + p[0] += (unsigned _BitInt(2048)) r;
> + p[1] += (unsigned _BitInt(2048)) s;
> + p[2] += (unsigned _BitInt(2048)) t;
> + p[3] += (unsigned _BitInt(2048)) u;
> +}
> +
> +void
> +f6 (unsigned _BitInt(4094) *p, int r, _BitInt(115) s, _BitInt(128) t,
> _BitInt(231) u)
> +{
> + p[0] -= (unsigned _BitInt(2048)) r;
> + p[1] -= (unsigned _BitInt(2048)) s;
> + p[2] -= (unsigned _BitInt(2048)) t;
> + p[3] -= (unsigned _BitInt(2048)) u;
> +}
> +
> +void
> +f7 (unsigned _BitInt(4096) *p, int r, _BitInt(115) s, _BitInt(128) t,
> _BitInt(231) u)
> +{
> + p[0] += (unsigned _BitInt(2110)) r;
> + p[1] += (unsigned _BitInt(2110)) s;
> + p[2] += (unsigned _BitInt(2110)) t;
> + p[3] += (unsigned _BitInt(2110)) u;
> +}
> +
> +void
> +f8 (unsigned _BitInt(4094) *p, int r, _BitInt(115) s, _BitInt(128) t,
> _BitInt(231) u)
> +{
> + p[0] -= (unsigned _BitInt(2110)) r;
> + p[1] -= (unsigned _BitInt(2110)) s;
> + p[2] -= (unsigned _BitInt(2110)) t;
> + p[3] -= (unsigned _BitInt(2110)) u;
> +}
> +
> +#if __SIZEOF_INT128__
> +void
> +f9 (_BitInt(4096) *p, __int128 r)
> +{
> + p[0] += (unsigned _BitInt(2048)) r;
> +}
> +
> +void
> +f10 (_BitInt(4094) *p, __int128 r)
> +{
> + p[0] -= (unsigned _BitInt(2048)) r;
> +}
> +
> +void
> +f11 (_BitInt(4096) *p, __int128 r)
> +{
> + p[0] += (unsigned _BitInt(2110)) r;
> +}
> +
> +void
> +f12 (_BitInt(4094) *p, __int128 r)
> +{
> + p[0] -= (unsigned _BitInt(2110)) r;
> +}
> +
> +void
> +f13 (unsigned _BitInt(4096) *p, __int128 r)
> +{
> + p[0] += (unsigned _BitInt(2048)) r;
> +}
> +
> +void
> +f14 (unsigned _BitInt(4094) *p, __int128 r)
> +{
> + p[0] -= (unsigned _BitInt(2048)) r;
> +}
> +
> +void
> +f15 (unsigned _BitInt(4096) *p, __int128 r)
> +{
> + p[0] += (unsigned _BitInt(2110)) r;
> +}
> +
> +void
> +f16 (unsigned _BitInt(4094) *p, __int128 r)
> +{
> + p[0] -= (unsigned _BitInt(2110)) r;
> +}
> +#endif
> +#else
> +int i;
> +#endif
>
> Jakub
>
>
--
Richard Biener <rguent...@suse.de>
SUSE Software Solutions Germany GmbH,
Frankenstrasse 146, 90461 Nuernberg, Germany;
GF: Ivo Totev, Andrew McDonald, Werner Knoblich; (HRB 36809, AG Nuernberg)
