Hi Roger,
On 03/05/2026 17:36, Jeffrey Law wrote:
>
>
> On 3/28/2026 7:09 AM, Roger Sayle wrote:
> > This patch resolves PR middle-end/122871 by improving RTL expansion of
> > doubleword multiplications. The main change is to synth_mult adding
> > support for the case where the constant being multiplied has BITS_PER_WORD
> > or more trailing zeros. The shift_cost tables in expmed are only
> > parameterized for shifts less than BITS_PER_WORD, so doubleword shifts
> > by more than this can't use the usual code path. This patch teaches
> > synth_mult that for scalar doubleword multiplications, a doubleword shift
> > by more than BITS_PER_WORD typically requires two instructions; one to
> > set the result lowpart to zero, and the other a wordmode shift to
> > calculate the result highpart.
It looks like this patch breaks bootstrap on aarch64-linux-gnu. I see builds
failing in stage2 with:
In file included from /home/alecop01/toolchain/src/gcc/gcc/expmed.cc:35:
In function ‘int* shift_cost_ptr(bool, machine_mode, int)’,
inlined from ‘int shift_cost(bool, machine_mode, int)’ at
/home/alecop01/toolchain/src/gcc/gcc/expmed.h:415:26,
inlined from ‘void synth_mult(algorithm*, long unsigned int, const
mult_cost*, machine_mode)’ at
/home/alecop01/toolchain/src/gcc/gcc/expmed.cc:2969:30:
/home/alecop01/toolchain/src/gcc/gcc/expmed.h:398:61: error: array subscript -1
is below array bounds of ‘int [64]’ [-Werror=array-bounds=]
398 | return &this_target_expmed->x_shift_cost[speed][midx][bits];
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
/home/alecop01/toolchain/src/gcc/gcc/expmed.h: In function ‘void
synth_mult(algorithm*, long unsigned int, const mult_cost*, machine_mode)’:
/home/alecop01/toolchain/src/gcc/gcc/expmed.h:168:7: note: while referencing
‘target_expmed::x_shift_cost’
168 | int x_shift_cost[2][NUM_MODE_IPV_INT][MAX_BITS_PER_WORD];
| ^~~~~~~~~~~~
cc1plus: all warnings being treated as errors
make[3]: *** [Makefile:1217: expmed.o] Error 1
make[3]: Leaving directory '/work/builds/bstrap/gcc'
make[2]: *** [Makefile:5182: all-stage2-gcc] Error 2
make[2]: Leaving directory '/work/builds/bstrap'
make[1]: *** [Makefile:27065: stage2-bubble] Error 2
make[1]: Leaving directory '/work/builds/bstrap'
make: *** [Makefile:1140: all] Error 2
Might you be able to take a look?
Thanks,
Alex
> >
> > For the testcase given in the PR:
> >
> > long long ashll_fn (long long a)
> > {
> > long long c;
> >
> > c = a << 33;
> > c += a;
> > return c;
> > }
> >
> > GCC for arm-linux-gnueabihf currently generates with -O2:
> >
> > ashll_fn:
> > lsl r2, r1, #11
> > lsl ip, r0, #11
> > subs ip, ip, r0
> > orr r2, r2, r0, lsr #21
> > sbc r2, r2, r1
> > lsl r3, ip, #11
> > lsl r2, r2, #11
> > adds r3, r3, r0
> > orr r2, r2, ip, lsr #21
> > adc r1, r1, r2
> > lsl r2, r1, #11
> > lsl r0, r3, #11
> > adds r0, r3, r0
> > orr r2, r2, r3, lsr #21
> > adc r1, r1, r2
> > bx lr
> >
> > with this patch, we instead generate:
> >
> > ashll_fn:
> > add r1, r1, r0, lsl #1
> > bx lr
> >
> >
> > Additionally, this patch includes a clean-up (identified by Andrew
> > Pinski) to prevent RTL expansion of doubleword multiplications from
> > initially emitting multiply instructions by immediate constants 0, 1
> > or 2. These dubious multiplications eventually get tidied up by later
> > RTL optimization passes, but being sensible during RTL expansion
> > both speeds up the compiler and reduces unnecessary memory usage.
> You might want to consider handling -1 and perhaps -2 as well if it's
> trivial to do so. That could be a follow-up IMHO.
>
> >
> >
> > This patch has been tested on x86_64-pc-linux-gnu with make bootstrap
> > and make -k check, both with and without --target_board=unix{-m32} with
> > no new failures. Ok for mainline?
> >
> >
> > 2026-03-28 Roger Sayle <[email protected]>
> >
> > gcc/ChangeLog
> > PR middle-end/122871
> > * expmed.cc (synth_mult): Handle doubleword left shifts by
> > BITS_PER_MODE bits or more, for scalar modes.
> > * optabs.cc (expand_doubleword_mult): Avoid generating multiply
> > instructions by immediate constants 0, 1 or 2.
> Consider adding the example as a testcase.
>
> OK with or without the testcase and handling of -1, -2.
>
> jeff
