public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r12-5490] middle-end: Fix failures with bitclear patterns on signed values
@ 2021-11-24 6:41 Tamar Christina
0 siblings, 0 replies; only message in thread
From: Tamar Christina @ 2021-11-24 6:41 UTC (permalink / raw)
To: gcc-cvs
https://gcc.gnu.org/g:755c2e7d71cbab89b2bd1d787db46428a604efb2
commit r12-5490-g755c2e7d71cbab89b2bd1d787db46428a604efb2
Author: Tamar Christina <tamar.christina@arm.com>
Date: Wed Nov 24 06:39:05 2021 +0000
middle-end: Fix failures with bitclear patterns on signed values
During testing after rebasing to commit I noticed a failing testcase with the
bitmask compare patch.
Consider the following C++ testcase:
#include <compare>
#define A __attribute__((noipa))
A bool f5 (double i, double j) { auto c = i <=> j; return c >= 0; }
This turns into a comparison against chars, on systems where chars are signed
the pattern inserts an unsigned convert such that it's able to do the
transformation.
i.e.:
# RANGE [-1, 2]
# c$_M_value_22 = PHI <-1(3), 0(2), 2(5), 1(4)>
# RANGE ~[3, 254]
_11 = (unsigned char) c$_M_value_22;
_19 = _11 <= 1;
# .MEM_24 = VDEF <.MEM_6(D)>
D.10434 ={v} {CLOBBER};
# .MEM_14 = VDEF <.MEM_24>
D.10407 ={v} {CLOBBER};
# VUSE <.MEM_14>
return _19;
instead of:
# RANGE [-1, 2]
# c$_M_value_5 = PHI <-1(3), 0(2), 2(5), 1(4)>
# RANGE [-2, 2]
_3 = c$_M_value_5 & -2;
_19 = _3 == 0;
# .MEM_24 = VDEF <.MEM_6(D)>
D.10440 ={v} {CLOBBER};
# .MEM_14 = VDEF <.MEM_24>
D.10413 ={v} {CLOBBER};
# VUSE <.MEM_14>
return _19;
This causes much worse codegen under -ffast-math due to phiops no longer
recognizing the pattern. It turns out that phiopts spaceship_replacement is
looking for the exact form that was just changed.
The comments seems to suggest this code only checks for (res & ~1) == 0 but the
implementation seems to suggest it's broader.
As such I added a case to check to see if the value comparison we found is a
type cast. and strips away the type cast and continues.
In match.pd the typecasts are only added for signed comparisons to == 0 and != 0
which are then rewritten into comparisons with 1.
As such I only check for 1 and LE and GT, which is what match.pd would have
rewritten it to.
This fixes the regression but this is not code I 100% understand, since I don't
really know the semantics of the spaceship operator so would appreciate an extra
look.
gcc/ChangeLog:
* tree-ssa-phiopt.c (spaceship_replacement): Handle new canonical
codegen.
Diff:
---
gcc/tree-ssa-phiopt.c | 94 ++++++++++++++++++++++++++++++++++++++++++++++-----
1 file changed, 86 insertions(+), 8 deletions(-)
diff --git a/gcc/tree-ssa-phiopt.c b/gcc/tree-ssa-phiopt.c
index 2fa7069f824..3eac9b1ce46 100644
--- a/gcc/tree-ssa-phiopt.c
+++ b/gcc/tree-ssa-phiopt.c
@@ -2055,11 +2055,36 @@ spaceship_replacement (basic_block cond_bb, basic_block middle_bb,
gimple *orig_use_stmt = use_stmt;
tree orig_use_lhs = NULL_TREE;
int prec = TYPE_PRECISION (TREE_TYPE (phires));
- if (is_gimple_assign (use_stmt)
- && gimple_assign_rhs_code (use_stmt) == BIT_AND_EXPR
- && TREE_CODE (gimple_assign_rhs2 (use_stmt)) == INTEGER_CST
- && (wi::to_wide (gimple_assign_rhs2 (use_stmt))
- == wi::shifted_mask (1, prec - 1, false, prec)))
+ bool is_cast = false;
+
+ /* Deal with the case when match.pd has rewritten the (res & ~1) == 0
+ into res <= 1 and has left a type-cast for signed types. */
+ if (gimple_assign_cast_p (use_stmt))
+ {
+ orig_use_lhs = gimple_assign_lhs (use_stmt);
+ /* match.pd would have only done this for a signed type,
+ so the conversion must be to an unsigned one. */
+ tree ty1 = TREE_TYPE (gimple_assign_rhs1 (use_stmt));
+ tree ty2 = TREE_TYPE (orig_use_lhs);
+
+ if (!TYPE_UNSIGNED (ty2) || !INTEGRAL_TYPE_P (ty2))
+ return false;
+ if (TYPE_PRECISION (ty1) != TYPE_PRECISION (ty2))
+ return false;
+ if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (orig_use_lhs))
+ return false;
+ if (EDGE_COUNT (phi_bb->preds) != 4)
+ return false;
+ if (!single_imm_use (orig_use_lhs, &use_p, &use_stmt))
+ return false;
+
+ is_cast = true;
+ }
+ else if (is_gimple_assign (use_stmt)
+ && gimple_assign_rhs_code (use_stmt) == BIT_AND_EXPR
+ && TREE_CODE (gimple_assign_rhs2 (use_stmt)) == INTEGER_CST
+ && (wi::to_wide (gimple_assign_rhs2 (use_stmt))
+ == wi::shifted_mask (1, prec - 1, false, prec)))
{
/* For partial_ordering result operator>= with unspec as second
argument is (res & 1) == res, folded by match.pd into
@@ -2116,7 +2141,43 @@ spaceship_replacement (basic_block cond_bb, basic_block middle_bb,
|| !tree_fits_shwi_p (rhs)
|| !IN_RANGE (tree_to_shwi (rhs), -1, 1))
return false;
- if (orig_use_lhs)
+
+ if (is_cast)
+ {
+ if (TREE_CODE (rhs) != INTEGER_CST)
+ return false;
+ /* As for -ffast-math we assume the 2 return to be
+ impossible, canonicalize (unsigned) res <= 1U or
+ (unsigned) res < 2U into res >= 0 and (unsigned) res > 1U
+ or (unsigned) res >= 2U as res < 0. */
+ switch (cmp)
+ {
+ case LE_EXPR:
+ if (!integer_onep (rhs))
+ return false;
+ cmp = GE_EXPR;
+ break;
+ case LT_EXPR:
+ if (wi::ne_p (wi::to_widest (rhs), 2))
+ return false;
+ cmp = GE_EXPR;
+ break;
+ case GT_EXPR:
+ if (!integer_onep (rhs))
+ return false;
+ cmp = LT_EXPR;
+ break;
+ case GE_EXPR:
+ if (wi::ne_p (wi::to_widest (rhs), 2))
+ return false;
+ cmp = LT_EXPR;
+ break;
+ default:
+ return false;
+ }
+ rhs = build_zero_cst (TREE_TYPE (phires));
+ }
+ else if (orig_use_lhs)
{
if ((cmp != EQ_EXPR && cmp != NE_EXPR) || !integer_zerop (rhs))
return false;
@@ -2411,6 +2472,7 @@ spaceship_replacement (basic_block cond_bb, basic_block middle_bb,
use_operand_p use_p;
imm_use_iterator iter;
bool has_debug_uses = false;
+ bool has_cast_debug_uses = false;
FOR_EACH_IMM_USE_FAST (use_p, iter, phires)
{
gimple *use_stmt = USE_STMT (use_p);
@@ -2422,12 +2484,14 @@ spaceship_replacement (basic_block cond_bb, basic_block middle_bb,
}
if (orig_use_lhs)
{
- if (!has_debug_uses)
+ if (!has_debug_uses || is_cast)
FOR_EACH_IMM_USE_FAST (use_p, iter, orig_use_lhs)
{
gimple *use_stmt = USE_STMT (use_p);
gcc_assert (is_gimple_debug (use_stmt));
has_debug_uses = true;
+ if (is_cast)
+ has_cast_debug_uses = true;
}
gimple_stmt_iterator gsi = gsi_for_stmt (orig_use_stmt);
tree zero = build_zero_cst (TREE_TYPE (orig_use_lhs));
@@ -2459,7 +2523,21 @@ spaceship_replacement (basic_block cond_bb, basic_block middle_bb,
gsi_insert_before (&gsi, g, GSI_SAME_STMT);
replace_uses_by (phires, temp2);
if (orig_use_lhs)
- replace_uses_by (orig_use_lhs, temp2);
+ {
+ if (has_cast_debug_uses)
+ {
+ tree temp3 = make_node (DEBUG_EXPR_DECL);
+ DECL_ARTIFICIAL (temp3) = 1;
+ TREE_TYPE (temp3) = TREE_TYPE (orig_use_lhs);
+ SET_DECL_MODE (temp3, TYPE_MODE (type));
+ t = fold_convert (TREE_TYPE (temp3), temp2);
+ g = gimple_build_debug_bind (temp3, t, phi);
+ gsi_insert_before (&gsi, g, GSI_SAME_STMT);
+ replace_uses_by (orig_use_lhs, temp3);
+ }
+ else
+ replace_uses_by (orig_use_lhs, temp2);
+ }
}
}
^ permalink raw reply [flat|nested] only message in thread
only message in thread, other threads:[~2021-11-24 6:41 UTC | newest]
Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-11-24 6:41 [gcc r12-5490] middle-end: Fix failures with bitclear patterns on signed values Tamar Christina
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).