Re: [Qemu-devel] [PATCH 2/8] target/arm: optimize smul_dual() and neon_trn_u8() using extract op

[Philippe Mathieu-Daudé]

Hi Richard,

On 05/10/2017 05:32 PM, Philippe Mathieu-Daudé wrote:
> On 05/10/2017 05:20 PM, Richard Henderson wrote:
> > On 05/10/2017 01:05 PM, Philippe Mathieu-Daudé wrote:
> > > -    tcg_gen_shri_i32(t1, t1, 8);
> > > -    tcg_gen_andi_i32(t1, t1, 0x00ff00ff);
> > > +    tcg_gen_extract_i32(t1, t1, 8, 0x00ff00ff);
> >
> > This is very wrong.  See my previous comment.

Indeed, after correcting the script:

$ docker run -it -v `pwd`:`pwd` -w `pwd` petersenna/coccinelle --sp-file 
scripts/coccinelle/tcg_gen_extract.cocci --macro-file 
scripts/cocci-macro-file.h target/arm/translate.c --in-place
init_defs_builtins: /usr/lib64/coccinelle/standard.h
init_defs: scripts/cocci-macro-file.h
HANDLING: target/arm/translate.c
candidate at target/arm/translate.c:4703
  op_size: i32/i32 (same)
  low_bits: 8 (value: 0xff)
  len: 0xff00ff
  len_bits != low_bits
  candidate is NOT optimizable

candidate at target/arm/translate.c:342
  op_size: i32/i32 (same)
  low_bits: 8 (value: 0xff)
  len: 0xff00ff
  len_bits != low_bits
  candidate is NOT optimizable

> Arghhh I see, I checked manually and though I had it...
>
> I'll first check with the cocci script if it can handles this better
> then review the serie manually before bother you again.
>
> Thinking about it, this should be quite easy unit-testable somehow ...
> Not sure if I want to start this path although.
>
> Sorry for the noise and thank a lot for the review!
>
> Phil.