[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 Xi Ruoyao changed: What|Removed |Added Status|NEW |RESOLVED Resolution|--- |FIXED --- Comment #21 from Xi Ruoyao --- The de-optimization is fixed as well. Closing.
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 --- Comment #20 from CVS Commits --- The master branch has been updated by Xi Ruoyao : https://gcc.gnu.org/g:9e9279fadbd1c673c875b9d20261d2de0473f63f commit r14-5542-g9e9279fadbd1c673c875b9d20261d2de0473f63f Author: Andrew Pinski Date: Mon Nov 13 20:18:34 2023 + Only allow (copysign x, NEG_CONST) -> (fneg (fabs x)) simplification for constant folding [PR112483] On targets with native copysign instructions, (copysign x, -1) is usually more efficient than (fneg (fabs x)). Since r14-5284, in the middle end we always optimize (fneg (fabs x)) to (copysign x, -1), not vice versa. If the target does not support native fcopysign, expand_COPYSIGN will expand it as (fneg (fabs x)) anyway. gcc/ChangeLog: PR rtl-optimization/112483 * simplify-rtx.cc (simplify_binary_operation_1) : Call simplify_unary_operation for NEG instead of simplify_gen_unary.
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 --- Comment #19 from Xi Ruoyao --- (In reply to Andrew Pinski from comment #18) > (In reply to Xi Ruoyao from comment #17) > > Works for me. Will you send the simplify_gen_unary -> > > simplify_unary_operation change to gcc-patches? > > I don't have a platform where it makes a difference to test on. Can you post > it? https://gcc.gnu.org/pipermail/gcc-patches/2023-November/636445.html
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 --- Comment #18 from Andrew Pinski --- (In reply to Xi Ruoyao from comment #17) > Works for me. Will you send the simplify_gen_unary -> > simplify_unary_operation change to gcc-patches? I don't have a platform where it makes a difference to test on. Can you post it?
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 --- Comment #17 from Xi Ruoyao --- (In reply to Xi Ruoyao from comment #16) > (In reply to Andrew Pinski from comment #12) > > What about this patch: > > ``` > > diff --git a/gcc/simplify-rtx.cc b/gcc/simplify-rtx.cc > > index 69d87579d9c..f3745d86aea 100644 > > --- a/gcc/simplify-rtx.cc > > +++ b/gcc/simplify-rtx.cc > > @@ -4392,7 +4392,7 @@ simplify_ashift: > > real_convert (, mode, CONST_DOUBLE_REAL_VALUE (trueop1)); > > rtx tmp = simplify_gen_unary (ABS, mode, op0, mode); > > if (REAL_VALUE_NEGATIVE (f1)) > > - tmp = simplify_gen_unary (NEG, mode, op0, mode); > > + tmp = simplify_unary_operation (NEG, mode, tmp, mode); > > return tmp; > > } > >if (GET_CODE (op0) == NEG || GET_CODE (op0) == ABS) > > > > ``` > > > > We only generate NEG if it simplifies from the ABS. Most likely due to > > constants. > > I'm trying this out on LoongArch. Works for me. Will you send the simplify_gen_unary -> simplify_unary_operation change to gcc-patches?
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 Richard Biener changed: What|Removed |Added Priority|P3 |P4
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 --- Comment #16 from Xi Ruoyao --- (In reply to Andrew Pinski from comment #12) > What about this patch: > ``` > diff --git a/gcc/simplify-rtx.cc b/gcc/simplify-rtx.cc > index 69d87579d9c..f3745d86aea 100644 > --- a/gcc/simplify-rtx.cc > +++ b/gcc/simplify-rtx.cc > @@ -4392,7 +4392,7 @@ simplify_ashift: > real_convert (, mode, CONST_DOUBLE_REAL_VALUE (trueop1)); > rtx tmp = simplify_gen_unary (ABS, mode, op0, mode); > if (REAL_VALUE_NEGATIVE (f1)) > - tmp = simplify_gen_unary (NEG, mode, op0, mode); > + tmp = simplify_unary_operation (NEG, mode, tmp, mode); > return tmp; > } >if (GET_CODE (op0) == NEG || GET_CODE (op0) == ABS) > > ``` > > We only generate NEG if it simplifies from the ABS. Most likely due to > constants. I'm trying this out on LoongArch.
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 Tamar Christina changed: What|Removed |Added CC|tamar.christina at arm dot com | --- Comment #15 from Tamar Christina --- removing duplicate mail
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 Xi Ruoyao changed: What|Removed |Added Status|ASSIGNED|NEW Assignee|xry111 at gcc dot gnu.org |unassigned at gcc dot gnu.org --- Comment #14 from Xi Ruoyao --- I've pushed the obvious fix for now (as the wrong-code issue is causing wide-spreading test failures). Leaving the ticket open for the missed-optimization.
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 --- Comment #13 from CVS Commits --- The master branch has been updated by Xi Ruoyao : https://gcc.gnu.org/g:7ba7529ee3974667a8e68d089b606ac2fb159415 commit r14-5377-g7ba7529ee3974667a8e68d089b606ac2fb159415 Author: Xi Ruoyao Date: Sun Nov 12 14:16:20 2023 + Fix (fcopysign x, NEGATIVE_CONST) -> (fneg (fabs x)) simplification [PR112483] (fcopysign x, NEGATIVE_CONST) can be simplified to (fneg (fabs x)), but a logic error in the code caused it mistakenly simplified to (fneg x) instead. gcc/ChangeLog: PR rtl-optimization/112483 * simplify-rtx.cc (simplify_binary_operation_1) : Fix the simplification of (fcopysign x, NEGATIVE_CONST).
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 --- Comment #12 from Andrew Pinski --- What about this patch: ``` diff --git a/gcc/simplify-rtx.cc b/gcc/simplify-rtx.cc index 69d87579d9c..f3745d86aea 100644 --- a/gcc/simplify-rtx.cc +++ b/gcc/simplify-rtx.cc @@ -4392,7 +4392,7 @@ simplify_ashift: real_convert (, mode, CONST_DOUBLE_REAL_VALUE (trueop1)); rtx tmp = simplify_gen_unary (ABS, mode, op0, mode); if (REAL_VALUE_NEGATIVE (f1)) - tmp = simplify_gen_unary (NEG, mode, op0, mode); + tmp = simplify_unary_operation (NEG, mode, tmp, mode); return tmp; } if (GET_CODE (op0) == NEG || GET_CODE (op0) == ABS) ``` We only generate NEG if it simplifies from the ABS. Most likely due to constants.
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 Xi Ruoyao changed: What|Removed |Added Keywords||missed-optimization --- Comment #11 from Xi Ruoyao --- As we've discussed via gcc-patches, the (fcopysign x, NEG_CONST) -> (fneg (fabs x)) conversion seems a de-optimization.
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 Andrew Pinski changed: What|Removed |Added Target Milestone|--- |14.0 Keywords||wrong-code
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 Xi Ruoyao changed: What|Removed |Added Assignee|unassigned at gcc dot gnu.org |xry111 at gcc dot gnu.org Status|NEW |ASSIGNED Keywords||patch URL||https://gcc.gnu.org/piperma ||il/gcc-patches/2023-Novembe ||r/636200.html --- Comment #10 from Xi Ruoyao --- (In reply to Xi Ruoyao from comment #8) > (In reply to Tamar Christina from comment #7) > > Yeah, that fold-rtx code is bogus. It's a latent bug. > > > > Optimizing copysign(x, -y) to neg(x) is just wrong. > > > > Will you be sending a patch Xi or do you want me to? > > Yes, I'm running a regression test now and I'll send the patch after it > finishes. https://gcc.gnu.org/pipermail/gcc-patches/2023-November/636200.html
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 --- Comment #9 from Xi Ruoyao --- *** Bug 112484 has been marked as a duplicate of this bug. ***
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 --- Comment #8 from Xi Ruoyao --- (In reply to Tamar Christina from comment #7) > Yeah, that fold-rtx code is bogus. It's a latent bug. > > Optimizing copysign(x, -y) to neg(x) is just wrong. > > Will you be sending a patch Xi or do you want me to? Yes, I'm running a regression test now and I'll send the patch after it finishes.
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483 Tamar Christina changed: What|Removed |Added CC||tnfchris at gcc dot gnu.org --- Comment #7 from Tamar Christina --- Yeah, that fold-rtx code is bogus. It's a latent bug. Optimizing copysign(x, -y) to neg(x) is just wrong. Will you be sending a patch Xi or do you want me to?
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483
Thomas Schwinge changed:
What|Removed |Added
Target|loongarch64-linux-gnu |loongarch64-linux-gnu
||powerpc64le-linux-gnu
CC||tschwinge at gcc dot gnu.org
--- Comment #6 from Thomas Schwinge ---
(In reply to Xi Ruoyao from comment #5)
> In simplify_binary_operation_1, simplify-rtx.cc:
>
> case COPYSIGN:
> if (rtx_equal_p (trueop0, trueop1) && ! side_effects_p (op0))
> return op0;
> if (CONST_DOUBLE_AS_FLOAT_P (trueop1))
> {
> REAL_VALUE_TYPE f1;
> real_convert (, mode, CONST_DOUBLE_REAL_VALUE (trueop1));
> rtx tmp = simplify_gen_unary (ABS, mode, op0, mode);
> if (REAL_VALUE_NEGATIVE (f1))
> tmp = simplify_gen_unary (NEG, mode, op0, mode);
> ^^^
> return tmp;
> }
>
> shouldn't the "op0" with caret be "tmp" instead??
I have no knowledge at all about that code, but your suggested change appears
legit, and I do confirm that it does fix the following powerpc64le-linux-gnu
regressions:
@@ -169421,7 +169467,7 @@ PASS: gcc.target/powerpc/p8vector-fp.c
scan-assembler xsdivdp
PASS: gcc.target/powerpc/p8vector-fp.c scan-assembler xsdivsp
PASS: gcc.target/powerpc/p8vector-fp.c scan-assembler xsmuldp
PASS: gcc.target/powerpc/p8vector-fp.c scan-assembler xsmulsp
[-PASS:-]{+FAIL:+} gcc.target/powerpc/p8vector-fp.c scan-assembler xsnabsdp
PASS: gcc.target/powerpc/p8vector-fp.c scan-assembler xsnegdp
PASS: gcc.target/powerpc/p8vector-fp.c scan-assembler xssqrtdp
PASS: gcc.target/powerpc/p8vector-fp.c scan-assembler xssqrtsp
PASS: gfortran.dg/ieee/ieee_2.f90 -O0 (test for excess errors)
PASS: gfortran.dg/ieee/ieee_2.f90 -O0 execution test
PASS: gfortran.dg/ieee/ieee_2.f90 -O1 (test for excess errors)
[-PASS:-]{+FAIL:+} gfortran.dg/ieee/ieee_2.f90 -O1 execution test
PASS: gfortran.dg/ieee/ieee_2.f90 -O2 (test for excess errors)
PASS: gfortran.dg/ieee/ieee_2.f90 -O2 execution test
PASS: gfortran.dg/ieee/ieee_2.f90 -O3 -fomit-frame-pointer -funroll-loops
-fpeel-loops -ftracer -finline-functions (test for excess errors)
@@ -63715,7 +63751,7 @@ PASS: gfortran.dg/ieee/ieee_2.f90 -O3
-fomit-frame-pointer -funroll-loops -fpe
PASS: gfortran.dg/ieee/ieee_2.f90 -O3 -g (test for excess errors)
PASS: gfortran.dg/ieee/ieee_2.f90 -O3 -g execution test
PASS: gfortran.dg/ieee/ieee_2.f90 -Os (test for excess errors)
[-PASS:-]{+FAIL:+} gfortran.dg/ieee/ieee_2.f90 -Os execution test
[Bug rtl-optimization/112483] [14 Regression] gfortran.dg/ieee/ieee_2.f90 fails on loongarch64-linux-gnu at -O1 or above
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112483
Xi Ruoyao changed:
What|Removed |Added
Component|target |rtl-optimization
--- Comment #5 from Xi Ruoyao ---
In simplify_binary_operation_1, simplify-rtx.cc:
case COPYSIGN:
if (rtx_equal_p (trueop0, trueop1) && ! side_effects_p (op0))
return op0;
if (CONST_DOUBLE_AS_FLOAT_P (trueop1))
{
REAL_VALUE_TYPE f1;
real_convert (, mode, CONST_DOUBLE_REAL_VALUE (trueop1));
rtx tmp = simplify_gen_unary (ABS, mode, op0, mode);
if (REAL_VALUE_NEGATIVE (f1))
tmp = simplify_gen_unary (NEG, mode, op0, mode);
^^^
return tmp;
}
shouldn't the "op0" with caret be "tmp" instead??
