[Bug tree-optimization/103007] [12 Regression] ice in vect_normalize_conj_loc, at tree-vect-slp-patterns.c:722 since r12-4785-ged3de62ac949c92ad41ef6de7cc926fbb2a510ce
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103007 Tamar Christina changed: What|Removed |Added Resolution|--- |FIXED Status|ASSIGNED|RESOLVED --- Comment #12 from Tamar Christina --- Fixed, thanks for the report.
[Bug tree-optimization/103007] [12 Regression] ice in vect_normalize_conj_loc, at tree-vect-slp-patterns.c:722 since r12-4785-ged3de62ac949c92ad41ef6de7cc926fbb2a510ce
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103007 --- Comment #11 from CVS Commits --- The master branch has been updated by Tamar Christina : https://gcc.gnu.org/g:6cc8aa65fdeaefe9774d5e0d4e72c91f52313be1 commit r12-4846-g6cc8aa65fdeaefe9774d5e0d4e72c91f52313be1 Author: Tamar Christina Date: Tue Nov 2 13:26:36 2021 + middle-end: Fix PR103007, add missing check on complex fms detection. The complex FMS detection is missing a check on if the nodes of the VEC_PERM has the amount of children we expect before it recurses. This check is there on MUL and FMA but was missing for FMS, due to this the compiler goes on further than it should and hits an assert. gcc/ChangeLog: PR tree-optimization/103007 * tree-vect-slp-patterns.c (complex_fms_pattern::matches): Add elem check. gcc/testsuite/ChangeLog: PR tree-optimization/103007 * g++.dg/pr103007.C: New test.
[Bug tree-optimization/103007] [12 Regression] ice in vect_normalize_conj_loc, at tree-vect-slp-patterns.c:722 since r12-4785-ged3de62ac949c92ad41ef6de7cc926fbb2a510ce
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103007 Richard Biener changed: What|Removed |Added Priority|P3 |P1
[Bug tree-optimization/103007] [12 Regression] ice in vect_normalize_conj_loc, at tree-vect-slp-patterns.c:722 since r12-4785-ged3de62ac949c92ad41ef6de7cc926fbb2a510ce
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103007 --- Comment #10 from Mario Baumann --- Added smaller Fortran example foo.F. Same behaviour, fails with ICE for -O2 and works with -O1 and -O0
[Bug tree-optimization/103007] [12 Regression] ice in vect_normalize_conj_loc, at tree-vect-slp-patterns.c:722 since r12-4785-ged3de62ac949c92ad41ef6de7cc926fbb2a510ce
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103007 Mario Baumann changed: What|Removed |Added CC||mario-baumann at web dot de --- Comment #9 from Mario Baumann --- Created attachment 51718 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51718=edit smaller Fortran example
[Bug tree-optimization/103007] [12 Regression] ice in vect_normalize_conj_loc, at tree-vect-slp-patterns.c:722 since r12-4785-ged3de62ac949c92ad41ef6de7cc926fbb2a510ce
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103007 --- Comment #8 from Jürgen Reuter --- (In reply to Jürgen Reuter from comment #6) > Created attachment 51716 [details] > gfortran appearance of the ICE Just for completeness, this example needs to be compiled with -O2, while -O0 and -O1 work fine.
[Bug tree-optimization/103007] [12 Regression] ice in vect_normalize_conj_loc, at tree-vect-slp-patterns.c:722 since r12-4785-ged3de62ac949c92ad41ef6de7cc926fbb2a510ce
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103007 --- Comment #7 from Tamar Christina --- Yes, I've posted a patch for this already https://gcc.gnu.org/pipermail/gcc-patches/2021-November/583008.html waiting review.
[Bug tree-optimization/103007] [12 Regression] ice in vect_normalize_conj_loc, at tree-vect-slp-patterns.c:722 since r12-4785-ged3de62ac949c92ad41ef6de7cc926fbb2a510ce
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103007 --- Comment #6 from Jürgen Reuter --- Created attachment 51716 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51716=edit gfortran appearance of the ICE
[Bug tree-optimization/103007] [12 Regression] ice in vect_normalize_conj_loc, at tree-vect-slp-patterns.c:722 since r12-4785-ged3de62ac949c92ad41ef6de7cc926fbb2a510ce
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103007 Jürgen Reuter changed: What|Removed |Added CC||juergen.reuter at desy dot de --- Comment #5 from Jürgen Reuter --- I can confirm this ICE also with (legacy) Fortran code, cf. below. The full code is attached. C*** C...PYQQBH C...Calculates the matrix element for the processes C...g + g or q + qbar -> Q + Qbar + H (normally with Q = t). C...REDUCE output and part of the rest courtesy Z. Kunszt, see C...Z. Kunszt, Nucl. Phys. B247 (1984) 339. SUBROUTINE PYQQBH(WTQQBH) C...Double precision and integer declarations. IMPLICIT DOUBLE PRECISION(A-H, O-Z) IMPLICIT INTEGER(I-N) INTEGER PYK,PYCHGE,PYCOMP C...Commonblocks. COMMON/PYDAT1/MSTU(200),PARU(200),MSTJ(200),PARJ(200) COMMON/PYDAT2/KCHG(500,4),PMAS(500,4),PARF(2000),VCKM(4,4) COMMON/PYPARS/MSTP(200),PARP(200),MSTI(200),PARI(200) COMMON/PYINT1/MINT(400),VINT(400) COMMON/PYINT2/ISET(500),KFPR(500,2),COEF(500,20),ICOL(40,4,2) SAVE /PYDAT1/,/PYDAT2/,/PYPARS/,/PYINT1/,/PYINT2/ C...Local arrays and function. DIMENSION PP(15,4),CLR(8,8),FM(10,10),RM(8,8),DX(8) DOT(I,J)=PP(I,4)*PP(J,4)-PP(I,1)*PP(J,1)-PP(I,2)*PP(J,2)- (I,3)*PP(J,3) C...Mass parameters. WTQQBH=0D0 ISUB=MINT(1) SHPR=SQRT(VINT(26))*VINT(1) PQ=PMAS(PYCOMP(KFPR(ISUB,2)),1) PH=SQRT(VINT(21))*VINT(1) SPQ=PQ**2 SPH=PH**2 C...Set up outgoing kinematics: 1=t, 2=tbar, 3=H. DO 100 I=1,2 PT=SQRT(MAX(0D0,VINT(197+5*I))) PP(I,1)=PT*COS(VINT(198+5*I)) PP(I,2)=PT*SIN(VINT(198+5*I)) 100 CONTINUE PP(3,1)=-PP(1,1)-PP(2,1) PP(3,2)=-PP(1,2)-PP(2,2) PMS1=SPQ+PP(1,1)**2+PP(1,2)**2 PMS2=SPQ+PP(2,1)**2+PP(2,2)**2 PMS3=SPH+PP(3,1)**2+PP(3,2)**2 PMT3=SQRT(PMS3) PP(3,3)=PMT3*SINH(VINT(211)) PP(3,4)=PMT3*COSH(VINT(211)) PMS12=(SHPR-PP(3,4))**2-PP(3,3)**2 PP(1,3)=(-PP(3,3)*(PMS12+PMS1-PMS2)+ (213)*(SHPR-PP(3,4))*VINT(220))/(2D0*PMS12) PP(2,3)=-PP(1,3)-PP(3,3) PP(1,4)=SQRT(PMS1+PP(1,3)**2) PP(2,4)=SQRT(PMS2+PP(2,3)**2) C...Set up incoming kinematics and derived momentum combinations. DO 110 I=4,5 PP(I,1)=0D0 PP(I,2)=0D0 PP(I,3)=-0.5D0*SHPR*(-1)**I PP(I,4)=-0.5D0*SHPR 110 CONTINUE DO 120 J=1,4 PP(6,J)=PP(1,J)+PP(2,J) PP(7,J)=PP(1,J)+PP(3,J) PP(8,J)=PP(1,J)+PP(4,J) PP(9,J)=PP(1,J)+PP(5,J) PP(10,J)=-PP(2,J)-PP(3,J) PP(11,J)=-PP(2,J)-PP(4,J) PP(12,J)=-PP(2,J)-PP(5,J) PP(13,J)=-PP(4,J)-PP(5,J) 120 CONTINUE C...Derived kinematics invariants. X1=DOT(1,2) X2=DOT(1,3) X3=DOT(1,4) X4=DOT(1,5) X5=DOT(2,3) X6=DOT(2,4) X7=DOT(2,5) X8=DOT(3,4) X9=DOT(3,5) X10=DOT(4,5) C...Propagators. SS1=DOT(7,7)-SPQ SS2=DOT(8,8)-SPQ SS3=DOT(9,9)-SPQ SS4=DOT(10,10)-SPQ SS5=DOT(11,11)-SPQ SS6=DOT(12,12)-SPQ SS7=DOT(13,13) DX(1)=SS1*SS6 DX(2)=SS2*SS6 DX(3)=SS2*SS4 DX(4)=SS1*SS5 DX(5)=SS3*SS5 DX(6)=SS3*SS4 DX(7)=SS7*SS1 DX(8)=SS7*SS4 C...Define colour coefficients for g + g -> Q + Qbar + H. IF(ISUB.EQ.121.OR.ISUB.EQ.181.OR.ISUB.EQ.186) THEN DO 140 I=1,3 DO 130 J=1,3 CLR(I,J)=16D0/3D0 CLR(I+3,J+3)=16D0/3D0 CLR(I,J+3)=-2D0/3D0 CLR(I+3,J)=-2D0/3D0 130 CONTINUE 140 CONTINUE DO 160 L=1,2 DO 150 I=1,3 CLR(I,6+L)=-6D0 CLR(I+3,6+L)=6D0 CLR(6+L,I)=-6D0 CLR(6+L,I+3)=6D0 150 CONTINUE 160 CONTINUE DO 180 K1=1,2 DO 170 K2=1,2 CLR(6+K1,6+K2)=12D0 170 CONTINUE 180 CONTINUE C...Evaluate matrix elements for g + g -> Q + Qbar + H. FM(1,1)=64*PQ**6+16*PQ**4*PH**2+32*PQ**4*(X1+2*X2+X4+X9+2* & X7+X5)+8*PQ**2*PH**2*(-X1-X4+2*X7)+16*PQ**2*(X2*X9+4*X2* & X7+X2*X5-2*X4*X7-2*X9*X7)+8*PH**2*X4*X7-16*X2*X9*X7 FM(1,2)=16*PQ**6+8*PQ**4*(-2*X1+X2-2*X3-2*X4-4*X10+X9-X8+2 & *X7-4*X6+X5)+8*PQ**2*(-2*X1*X2-2*X2*X4-2*X2*X10+X2*X7-2* & X2*X6-2*X3*X7+2*X4*X7+4*X10*X7-X9*X7-X8*X7)+16*X2*X7*(X4+ & X10) FM(1,3)=16*PQ**6-4*PQ**4*PH**2+8*PQ**4*(-2*X1+2*X2-2*X3-4* & X4-8*X10+X9+X8-2*X7-4*X6+2*X5)-(4*PQ**2*PH**2)*(X1+X4+X10 & +X6)+8*PQ**2*(-2*X1*X2-2*X1*X10+X1*X9+X1*X8-2*X1*X5+X2**2 & -4*X2*X4-5*X2*X10+X2*X8-X2*X7-3*X2*X6+X2*X5+X3*X9+2*X3*X7 & -X3*X5+X4*X8+2*X4*X6-3*X4*X5-5*X10*X5+X9*X8+X9*X6+X9*X5+ & X8*X7-4*X6*X5+X5**2)-(16*X2*X5)*(X1+X4+X10+X6) FM(1,4)=16*PQ**6+4*PQ**4*PH**2+16*PQ**4*(-X1+X2-X3-X4+X10- &
[Bug tree-optimization/103007] [12 Regression] ice in vect_normalize_conj_loc, at tree-vect-slp-patterns.c:722 since r12-4785-ged3de62ac949c92ad41ef6de7cc926fbb2a510ce
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103007 Tamar Christina changed: What|Removed |Added Status|NEW |ASSIGNED Assignee|unassigned at gcc dot gnu.org |tnfchris at gcc dot gnu.org
[Bug tree-optimization/103007] [12 Regression] ice in vect_normalize_conj_loc, at tree-vect-slp-patterns.c:722 since r12-4785-ged3de62ac949c92ad41ef6de7cc926fbb2a510ce
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103007 --- Comment #4 from Tamar Christina --- Yes a lane check size is missing from fms, will be fixed in a bit.
[Bug tree-optimization/103007] [12 Regression] ice in vect_normalize_conj_loc, at tree-vect-slp-patterns.c:722 since r12-4785-ged3de62ac949c92ad41ef6de7cc926fbb2a510ce
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103007 Martin Liška changed: What|Removed |Added CC||marxin at gcc dot gnu.org Summary|[12 Regression] ice in |[12 Regression] ice in |vect_normalize_conj_loc, at |vect_normalize_conj_loc, at |tree-vect-slp-patterns.c:72 |tree-vect-slp-patterns.c:72 |2 |2 since ||r12-4785-ged3de62ac949c92ad ||41ef6de7cc926fbb2a510ce Blocks||26163 --- Comment #3 from Martin Liška --- Btw. it affects quite some SPEC benchmarks. Referenced Bugs: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=26163 [Bug 26163] [meta-bug] missed optimization in SPEC (2k17, 2k and 2k6 and 95)
