https://gcc.gnu.org/g:9954ab48cfdfd4accb54885ea6bfdf8709a97579
commit r16-9833-g9954ab48cfdfd4accb54885ea6bfdf8709a97579 Author: Igor Shevlyakov <[email protected]> Date: Thu Oct 22 20:16:36 2026 -0700 match.pd: Keep the ORDERED & NE to LTGT fold quiet [PR127785] ORDERED and NE do not raise an invalid operation exception for a quiet NaN, but LTGT does. The (ordered & ne) -> ltgt rule added with "i386: Optimize ordered and nonequal" has no trapping-math check, so !__builtin_isunordered (a, b) & (a != b) raises FE_INVALID for a NaN operand even at -O0, since the rule also fires in GENERIC. On x86 the result is a signaling comiss. Fold only with -fno-trapping-math or when NaNs are not honored, as the (ltgt @0 @0) rule next to it does. GCC has no quiet comparison code for "ordered and not equal", so under the default flags the two comparisons stay separate, and x86 goes back to the quiet ucomiss/setp/setnp sequence that the rule replaced. The tests added with the rule check the fold itself, so they now pass -fno-trapping-math. With that option GENERIC can already fold the && forms through combine_comparisons, so opt-ordered-and-nonequal-1.c scans the optimized dump instead of forwprop1. The new opt-ordered-and-nonequal-2.c checks that the default flags keep <> out, and cmp-of-cmp-fenv-1.c checks at run time that FE_INVALID is not raised. Assisted-by: Claude Opus 5.5 PR middle-end/127785 gcc/ChangeLog: * match.pd ((ordered @0 @1) & (ne @0 @1)): Only fold to LTGT without trapping math or when NaNs are not honored. gcc/testsuite/ChangeLog: * gcc.dg/opt-ordered-and-nonequal-1.c: Add -fno-trapping-math and scan the optimized dump. * gcc.dg/opt-ordered-and-nonequal-2.c: New test. * gcc.dg/torture/cmp-of-cmp-fenv-1.c: New test. * gcc.target/i386/opt-comi-1.c: Add -fno-trapping-math. * gcc.target/i386/avx10_2-compare-1b.c: Likewise. Signed-off-by: Igor Shevlyakov <[email protected]> Diff: --- gcc/match.pd | 4 ++- gcc/testsuite/gcc.dg/opt-ordered-and-nonequal-1.c | 4 +-- gcc/testsuite/gcc.dg/opt-ordered-and-nonequal-2.c | 27 +++++++++++++++++++ gcc/testsuite/gcc.dg/torture/cmp-of-cmp-fenv-1.c | 31 ++++++++++++++++++++++ gcc/testsuite/gcc.target/i386/avx10_2-compare-1b.c | 2 +- gcc/testsuite/gcc.target/i386/opt-comi-1.c | 2 +- 6 files changed, 65 insertions(+), 5 deletions(-) diff --git a/gcc/match.pd b/gcc/match.pd index 54bbf8b8856b..135e17bbac37 100644 --- a/gcc/match.pd +++ b/gcc/match.pd @@ -7598,9 +7598,11 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT) (ltgt @0 @0) (if (!flag_trapping_math || !tree_expr_maybe_nan_p (@0)) { constant_boolean_node (false, type); })) +/* LTGT raises an exception for a quiet NaN, ORDERED and NE do not. */ (simplify (bit_and:c (ordered @0 @1) (ne @0 @1)) - (ltgt @0 @1)) + (if (!flag_trapping_math || !HONOR_NANS (@0)) + (ltgt @0 @1))) /* x == ~x -> false */ /* x != ~x -> true */ diff --git a/gcc/testsuite/gcc.dg/opt-ordered-and-nonequal-1.c b/gcc/testsuite/gcc.dg/opt-ordered-and-nonequal-1.c index d61c3322214b..de7d4324816d 100644 --- a/gcc/testsuite/gcc.dg/opt-ordered-and-nonequal-1.c +++ b/gcc/testsuite/gcc.dg/opt-ordered-and-nonequal-1.c @@ -1,5 +1,5 @@ /* { dg-do compile } */ -/* { dg-options "-O2 -fdump-tree-forwprop1-details" } */ +/* { dg-options "-O2 -fno-trapping-math -fdump-tree-optimized" } */ /* Make this work for default 'LOGICAL_OP_NON_SHORT_CIRCUIT == false' configurations: { dg-additional-options "--param logical-op-non-short-circuit=1" } */ @@ -48,4 +48,4 @@ int is_ordered_or_nonequal_sh_9 (float a, float b) return !((a == b) || __builtin_isunordered (b, a)); } -/* { dg-final { scan-tree-dump-times "gimple_simplified to\[^\n\r]*<>" 9 "forwprop1" } } */ +/* { dg-final { scan-tree-dump-times "<>" 9 "optimized" } } */ diff --git a/gcc/testsuite/gcc.dg/opt-ordered-and-nonequal-2.c b/gcc/testsuite/gcc.dg/opt-ordered-and-nonequal-2.c new file mode 100644 index 000000000000..d1fbae9d5e79 --- /dev/null +++ b/gcc/testsuite/gcc.dg/opt-ordered-and-nonequal-2.c @@ -0,0 +1,27 @@ +/* ORDERED and NE do not raise FE_INVALID for a quiet NaN but LTGT does, + so without -fno-trapping-math they must not be combined. */ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-optimized" } */ +/* { dg-add-options ieee } */ + +int is_ordered_and_nonequal_1 (double a, double b) +{ + return !__builtin_isunordered (a, b) & (a != b); +} + +int is_ordered_and_nonequal_2 (double a, double b) +{ + return (b != a) & !__builtin_isunordered (a, b); +} + +int is_ordered_and_nonequal_3 (double a, double b) +{ + return !__builtin_isunordered (a, b) && (a != b); +} + +int is_ordered_or_nonequal_4 (double a, double b) +{ + return !(__builtin_isunordered (a, b) || (a == b)); +} + +/* { dg-final { scan-tree-dump-not "<>" "optimized" } } */ diff --git a/gcc/testsuite/gcc.dg/torture/cmp-of-cmp-fenv-1.c b/gcc/testsuite/gcc.dg/torture/cmp-of-cmp-fenv-1.c new file mode 100644 index 000000000000..043f51fcb54e --- /dev/null +++ b/gcc/testsuite/gcc.dg/torture/cmp-of-cmp-fenv-1.c @@ -0,0 +1,31 @@ +/* PR middle-end/127785 */ +/* Combining ORDERED and NE must not raise FE_INVALID for a quiet NaN. */ +/* { dg-do run } */ +/* { dg-add-options ieee } */ +/* { dg-require-effective-target fenv_exceptions_double } */ + +#include <fenv.h> + +#define NOIPA __attribute__ ((noipa)) + +/* Quiet & quiet: LTGT would signal. */ +NOIPA int ord_and_ne (double a, double b) +{ return !__builtin_isunordered (a, b) & (a != b); } + +int +main (void) +{ + static const double as[4] = { __builtin_nan (""), 0.0, 1.0, 2.0 }; + /* Results for (NaN, 1), (0, 1), (1, 1) and (2, 1). */ + static const int res[4] = { 0, 1, 0, 1 }; + + for (unsigned j = 0; j < 4; j++) + { + feclearexcept (FE_INVALID); + if (ord_and_ne (as[j], 1.0) != res[j]) + __builtin_abort (); + if (fetestexcept (FE_INVALID)) + __builtin_abort (); + } + return 0; +} diff --git a/gcc/testsuite/gcc.target/i386/avx10_2-compare-1b.c b/gcc/testsuite/gcc.target/i386/avx10_2-compare-1b.c index cc7f8205f5f1..73457c9096a6 100644 --- a/gcc/testsuite/gcc.target/i386/avx10_2-compare-1b.c +++ b/gcc/testsuite/gcc.target/i386/avx10_2-compare-1b.c @@ -1,5 +1,5 @@ /* { dg-do compile } */ -/* { dg-options "-O2 -march=x86-64-v3 -mavx10.2 -mfpmath=sse" } */ +/* { dg-options "-O2 -fno-trapping-math -march=x86-64-v3 -mavx10.2 -mfpmath=sse" } */ /* { dg-final { scan-assembler-times "comi" 6 } } */ /* { dg-final { scan-assembler-times "comx" 12 } } */ diff --git a/gcc/testsuite/gcc.target/i386/opt-comi-1.c b/gcc/testsuite/gcc.target/i386/opt-comi-1.c index a2c5219d98df..715847faf33e 100644 --- a/gcc/testsuite/gcc.target/i386/opt-comi-1.c +++ b/gcc/testsuite/gcc.target/i386/opt-comi-1.c @@ -1,5 +1,5 @@ /* { dg-do compile } */ -/* { dg-options "-O2 -mfpmath=sse -msse2" } */ +/* { dg-options "-O2 -fno-trapping-math -mfpmath=sse -msse2" } */ /* { dg-additional-options "-fno-unwind-tables -fno-asynchronous-unwind-tables" { target *-*-darwin* } } */ /* { dg-final { scan-assembler-times "comiss" 9 } } */ /* { dg-final { scan-assembler-times "set" 9 } } */
