| Message ID | 20231103064113.823617-1-juzhe.zhong@rivai.ai |
|---|---|
| State | New |
| Headers | show |
| Series | [tree-optimization/111721,V2] VECT: Support SLP for MASK_LEN_GATHER_LOAD with dummy mask | expand |
On Fri, 3 Nov 2023, Juzhe-Zhong wrote: > This patch fixes following FAILs for RVV: > FAIL: gcc.dg/vect/vect-gather-1.c -flto -ffat-lto-objects scan-tree-dump vect "Loop contains only SLP stmts" > FAIL: gcc.dg/vect/vect-gather-1.c scan-tree-dump vect "Loop contains only SLP stmts" > > Bootstrap on X86 and regtest passed. > > Ok for trunk ? OK. We can walk back if problems with SVE appear. Thanks, Richard. > PR tree-optimization/111721 > > gcc/ChangeLog: > > * tree-vect-slp.cc (vect_get_and_check_slp_defs): Support SLP for dummy mask -1. > * tree-vect-stmts.cc (vectorizable_load): Ditto. > > --- > gcc/tree-vect-slp.cc | 5 ++--- > gcc/tree-vect-stmts.cc | 5 +++-- > 2 files changed, 5 insertions(+), 5 deletions(-) > > diff --git a/gcc/tree-vect-slp.cc b/gcc/tree-vect-slp.cc > index 43d742e3c92..6b8a7b628b6 100644 > --- a/gcc/tree-vect-slp.cc > +++ b/gcc/tree-vect-slp.cc > @@ -759,9 +759,8 @@ vect_get_and_check_slp_defs (vec_info *vinfo, unsigned char swap, > if ((dt == vect_constant_def > || dt == vect_external_def) > && !GET_MODE_SIZE (vinfo->vector_mode).is_constant () > - && (TREE_CODE (type) == BOOLEAN_TYPE > - || !can_duplicate_and_interleave_p (vinfo, stmts.length (), > - type))) > + && TREE_CODE (type) != BOOLEAN_TYPE > + && !can_duplicate_and_interleave_p (vinfo, stmts.length (), type)) > { > if (dump_enabled_p ()) > dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, > diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc > index 6ce4868d3e1..8c92bd5d931 100644 > --- a/gcc/tree-vect-stmts.cc > +++ b/gcc/tree-vect-stmts.cc > @@ -9825,6 +9825,7 @@ vectorizable_load (vec_info *vinfo, > > tree mask = NULL_TREE, mask_vectype = NULL_TREE; > int mask_index = -1; > + slp_tree slp_op = NULL; > if (gassign *assign = dyn_cast <gassign *> (stmt_info->stmt)) > { > scalar_dest = gimple_assign_lhs (assign); > @@ -9861,7 +9862,7 @@ vectorizable_load (vec_info *vinfo, > mask_index = vect_slp_child_index_for_operand (call, mask_index); > if (mask_index >= 0 > && !vect_check_scalar_mask (vinfo, stmt_info, slp_node, mask_index, > - &mask, NULL, &mask_dt, &mask_vectype)) > + &mask, &slp_op, &mask_dt, &mask_vectype)) > return false; > } > > @@ -10046,7 +10047,7 @@ vectorizable_load (vec_info *vinfo, > { > if (slp_node > && mask > - && !vect_maybe_update_slp_op_vectype (SLP_TREE_CHILDREN (slp_node)[0], > + && !vect_maybe_update_slp_op_vectype (slp_op, > mask_vectype)) > { > if (dump_enabled_p ()) >
Committed as passed the regression test of aarch64, thanks Richard. Pan -----Original Message----- From: Richard Biener <rguenther@suse.de> Sent: Friday, November 3, 2023 3:36 PM To: Juzhe-Zhong <juzhe.zhong@rivai.ai> Cc: gcc-patches@gcc.gnu.org; richard.sandiford@arm.com Subject: Re: [tree-optimization/111721 V2] VECT: Support SLP for MASK_LEN_GATHER_LOAD with dummy mask On Fri, 3 Nov 2023, Juzhe-Zhong wrote: > This patch fixes following FAILs for RVV: > FAIL: gcc.dg/vect/vect-gather-1.c -flto -ffat-lto-objects scan-tree-dump vect "Loop contains only SLP stmts" > FAIL: gcc.dg/vect/vect-gather-1.c scan-tree-dump vect "Loop contains only SLP stmts" > > Bootstrap on X86 and regtest passed. > > Ok for trunk ? OK. We can walk back if problems with SVE appear. Thanks, Richard. > PR tree-optimization/111721 > > gcc/ChangeLog: > > * tree-vect-slp.cc (vect_get_and_check_slp_defs): Support SLP for dummy mask -1. > * tree-vect-stmts.cc (vectorizable_load): Ditto. > > --- > gcc/tree-vect-slp.cc | 5 ++--- > gcc/tree-vect-stmts.cc | 5 +++-- > 2 files changed, 5 insertions(+), 5 deletions(-) > > diff --git a/gcc/tree-vect-slp.cc b/gcc/tree-vect-slp.cc > index 43d742e3c92..6b8a7b628b6 100644 > --- a/gcc/tree-vect-slp.cc > +++ b/gcc/tree-vect-slp.cc > @@ -759,9 +759,8 @@ vect_get_and_check_slp_defs (vec_info *vinfo, unsigned char swap, > if ((dt == vect_constant_def > || dt == vect_external_def) > && !GET_MODE_SIZE (vinfo->vector_mode).is_constant () > - && (TREE_CODE (type) == BOOLEAN_TYPE > - || !can_duplicate_and_interleave_p (vinfo, stmts.length (), > - type))) > + && TREE_CODE (type) != BOOLEAN_TYPE > + && !can_duplicate_and_interleave_p (vinfo, stmts.length (), type)) > { > if (dump_enabled_p ()) > dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, > diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc > index 6ce4868d3e1..8c92bd5d931 100644 > --- a/gcc/tree-vect-stmts.cc > +++ b/gcc/tree-vect-stmts.cc > @@ -9825,6 +9825,7 @@ vectorizable_load (vec_info *vinfo, > > tree mask = NULL_TREE, mask_vectype = NULL_TREE; > int mask_index = -1; > + slp_tree slp_op = NULL; > if (gassign *assign = dyn_cast <gassign *> (stmt_info->stmt)) > { > scalar_dest = gimple_assign_lhs (assign); > @@ -9861,7 +9862,7 @@ vectorizable_load (vec_info *vinfo, > mask_index = vect_slp_child_index_for_operand (call, mask_index); > if (mask_index >= 0 > && !vect_check_scalar_mask (vinfo, stmt_info, slp_node, mask_index, > - &mask, NULL, &mask_dt, &mask_vectype)) > + &mask, &slp_op, &mask_dt, &mask_vectype)) > return false; > } > > @@ -10046,7 +10047,7 @@ vectorizable_load (vec_info *vinfo, > { > if (slp_node > && mask > - && !vect_maybe_update_slp_op_vectype (SLP_TREE_CHILDREN (slp_node)[0], > + && !vect_maybe_update_slp_op_vectype (slp_op, > mask_vectype)) > { > if (dump_enabled_p ()) >
diff --git a/gcc/tree-vect-slp.cc b/gcc/tree-vect-slp.cc index 43d742e3c92..6b8a7b628b6 100644 --- a/gcc/tree-vect-slp.cc +++ b/gcc/tree-vect-slp.cc @@ -759,9 +759,8 @@ vect_get_and_check_slp_defs (vec_info *vinfo, unsigned char swap, if ((dt == vect_constant_def || dt == vect_external_def) && !GET_MODE_SIZE (vinfo->vector_mode).is_constant () - && (TREE_CODE (type) == BOOLEAN_TYPE - || !can_duplicate_and_interleave_p (vinfo, stmts.length (), - type))) + && TREE_CODE (type) != BOOLEAN_TYPE + && !can_duplicate_and_interleave_p (vinfo, stmts.length (), type)) { if (dump_enabled_p ()) dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc index 6ce4868d3e1..8c92bd5d931 100644 --- a/gcc/tree-vect-stmts.cc +++ b/gcc/tree-vect-stmts.cc @@ -9825,6 +9825,7 @@ vectorizable_load (vec_info *vinfo, tree mask = NULL_TREE, mask_vectype = NULL_TREE; int mask_index = -1; + slp_tree slp_op = NULL; if (gassign *assign = dyn_cast <gassign *> (stmt_info->stmt)) { scalar_dest = gimple_assign_lhs (assign); @@ -9861,7 +9862,7 @@ vectorizable_load (vec_info *vinfo, mask_index = vect_slp_child_index_for_operand (call, mask_index); if (mask_index >= 0 && !vect_check_scalar_mask (vinfo, stmt_info, slp_node, mask_index, - &mask, NULL, &mask_dt, &mask_vectype)) + &mask, &slp_op, &mask_dt, &mask_vectype)) return false; } @@ -10046,7 +10047,7 @@ vectorizable_load (vec_info *vinfo, { if (slp_node && mask - && !vect_maybe_update_slp_op_vectype (SLP_TREE_CHILDREN (slp_node)[0], + && !vect_maybe_update_slp_op_vectype (slp_op, mask_vectype)) { if (dump_enabled_p ())
This patch fixes following FAILs for RVV: FAIL: gcc.dg/vect/vect-gather-1.c -flto -ffat-lto-objects scan-tree-dump vect "Loop contains only SLP stmts" FAIL: gcc.dg/vect/vect-gather-1.c scan-tree-dump vect "Loop contains only SLP stmts" Bootstrap on X86 and regtest passed. Ok for trunk ? PR tree-optimization/111721 gcc/ChangeLog: * tree-vect-slp.cc (vect_get_and_check_slp_defs): Support SLP for dummy mask -1. * tree-vect-stmts.cc (vectorizable_load): Ditto. --- gcc/tree-vect-slp.cc | 5 ++--- gcc/tree-vect-stmts.cc | 5 +++-- 2 files changed, 5 insertions(+), 5 deletions(-)