[Bug rtl-optimization/26449] [4.2 Regression] ICE with -march=pentium4 -ftree-vectorize in matmul_i4.c in loop invariant motion

[rakdver at atrey dot karlin dot mff dot cuni dot cz]

------- Comment #5 from rakdver at atrey dot karlin dot mff dot cuni dot cz  2006-02-24 21:17 -------
Subject: Re:  [4.2 Regression] ICE with -march=pentium4 -ftree-vectorize in
matmul_i4.c in loop invariant motion

> I'm not sure how to fix this.  The debug trace here says it all: We try to
> force something into an operand using an optab with a mode for which we have no
> handler:

I have considered several possibilities; by far the simplest seems to be
to either just fail in this case in move_invariant_reg, or try calling
emit_move_insn on the original SET_SRC (set) if force_operand returns
NULL (http://gcc.gnu.org/ml/gcc-patches/2006-02/msg01009.html may be
useful in both cases).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26449