> -----Original Message-----
> From: Jeffrey Law <[email protected]>
> Sent: 04 August 2026 17:26
> To: [email protected]; [email protected]
> Cc: Tamar Christina <[email protected]>;
> [email protected]
> Subject: Re: [PATCH 1/2] genopinit: Distribute generated code across multiple
> files
> 
> 
> 
> On 7/23/2026 7:20 AM, [email protected] wrote:
> > From: Kyrylo Tkachov <[email protected]>
> >
> > insn-opinit.cc contains large generated initialization functions with
> > thousands of target-feature expressions.  The single file is among the
> > slowest objects in every AArch64 bootstrap stage.
> >
> > Teach genopinit to accept repeatable -O output options, following genemit
> and
> > genrecog.  Reuse the existing --with-insnemit-partitions count, while
> > preserving the default init-opinit.c output and -cFILE interface when -O is 
> > not
> > used.  Collect and validate all output names before opening them, rejecting
> > duplicates, resolvable aliases, header collisions, and incompatible -c and 
> > -O
> > options.
> >
> > Keep shared lookup code and all helper declarations in the first output, 
> > then
> > place helper definitions in the shortest output.  One-file output retains 
> > the
> > historical static helper names.  Split output uses generator-private 
> > external
> > names for cross-file references.
> >
> > With the default ten AArch64 partitions, the largest part compiles about 90%
> > faster than the original single object.
> >
> > Bootstrapped and tested on aarch64-none-linux-gnu and x86_64-linux.
> > Ok for trunk?
> > Thanks,
> > Kyrill
> >
> > gcc/ChangeLog:
> >
> >     * Makefile.in (INSNOPINIT_SPLITS_SEQ, INSNOPINIT_SEQ_SRC)
> >     (INSNOPINIT_SEQ_TMP, INSNOPINIT_SEQ_O): New variables.
> >     (OBJS): Replace insn-opinit.o with $(INSNOPINIT_SEQ_O).
> >     (MOSTLYCLEANFILES): Add $(INSNOPINIT_SEQ_SRC), retaining the
> legacy
> >     unnumbered source for cleanup.
> >     (s-opinit): Generate and move numbered insn-opinit-N.cc files.  Stop if
> >     moving an output fails.
> >     * configure.ac (--with-insnemit-partitions): Mention genopinit
> outputs.
> >     * configure: Regenerate.
> >     * genopinit.cc (source_file_option, output_file_names, output_files):
> >     New variables.
> >     (handle_arg): Record and validate repeatable -O output names.
> >     (main): Validate all output modes before opening files.  Preserve static
> >     helper linkage for one output and distribute private split helpers.
> >
> > Signed-off-by: Kyrylo Tkachov <[email protected]>
> It'll be interesting to see if this improves the RISC-V build times
> too.  Robin fixed the worst of the offenders a while back, but we've got
> daily build data and if there's a meaningful improvement we should be
> able to see it.
> 
> The only concern is the INSNOPINIT_SPLITS_SEQ is slightly different than
> the other _SEQ variables.    The others use:
> 
> MATCH_SPLITS_SEQ = $(wordlist 1,$(NUM_MATCH_SPLITS),$(one_to_9999))
> INSNEMIT_SPLITS_SEQ = $(wordlist
> 1,$(NUM_INSNEMIT_SPLITS),$(one_to_9999))
> INSNRECOG_SPLITS_SEQ = $(wordlist
> 1,$(NUM_INSNEMIT_SPLITS),$(one_to_9999))
> 
> Is there a reason you didn't use that form?

So I've been ignoring this since it's mostly Richard's thing and we seem to 
disagree
on design a lot.

But in my opinion we don't want to keep adding split code to every gen* file 
individually
but concentrate that code in gensupport so ever gen* can use it.

At the same time I also think we should have one configure option. Back in the 
day my
unsubmitted patch series used `:` to denote the different split values. 
So`5:6:8` meant
Split XX in 5 pieces, XY in 6 and XZ in 8.  This because the size of the 
generated files aren't
the same for all gen* code.

Lastly I also don't think splitting on an iterative way is going to give you 
the best compile
time increase.  Because that tends to bias the output.

In genmatch I use ftell to tell how big the files is so far an pick the 
smallest file. This allows
me to balance out the compile time over all files. I believe Robin eventually 
respun his
gen<something> match to do this a swell as that allows you to amortize the 
compile time
over cores.

So I really would like to see the infrastructure for splitting unified... but 
this is a step into
Improvement.  FWIW, I abandoned my original changes here wrt to splitting to do 
it in a
fundamentally different way.   But I'm doing that in my own free time :)

anyway feel free to ignore this, but it would be nice if we could refactor the 
splitting code.

Thanks,
Tamar

> 
> jeff

Reply via email to