Re: RISC-V: Folding memory for FP + constant case

[Vineet Gupta <vineetg@rivosinc.com>]

[-- Attachment #1: Type: text/plain, Size: 2508 bytes --]



On 7/25/23 20:31, Jeff Law via Gcc-patches wrote:
>
>
> On 7/25/23 05:24, Jivan Hakobyan wrote:
>> Hi.
>>
>> I re-run the benchmarks and hopefully got the same profit.
>> I also compared the leela's code and figured out the reason.
>>
>> Actually, my and Manolis's patches do the same thing. The difference 
>> is only execution order.
> But shouldn't your patch also allow for for at the last the potential 
> to pull the fp+offset computation out of a loop?  I'm pretty sure 
> Manolis's patch can't do that.
>
>> Because of f-m-o held after the register allocation it cannot 
>> eliminate redundant move 'sp' to another register.
> Actually that's supposed to be handled by a different patch that 
> should already be upstream.  Specifically;
>
>> commit 6a2e8dcbbd4bab374b27abea375bf7a921047800
>> Author: Manolis Tsamis <manolis.tsamis@vrull.eu>
>> Date:   Thu May 25 13:44:41 2023 +0200
>>
>>     cprop_hardreg: Enable propagation of the stack pointer if possible
>>         Propagation of the stack pointer in cprop_hardreg is currenty
>>     forbidden in all cases, due to maybe_mode_change returning NULL.
>>     Relax this restriction and allow propagation when no mode change is
>>     requested.
>>         gcc/ChangeLog:
>>                 * regcprop.cc (maybe_mode_change): Enable stack pointer
>>             propagation.
> I think there were a couple-follow-ups.  But that's the key change 
> that should allow propagation of copies from the stack pointer and 
> thus eliminate the mov gpr,sp instructions.  If that's not happening, 
> then it's worth investigating why.
>
>>
>> Besides that, I have checked the build failure on x264_r. It is 
>> already fixed on the third version.
> Yea, this was a problem with re-recognition.  I think it was fixed by:
>
>> commit ecfa870ff29d979bd2c3d411643b551f2b6915b0
>> Author: Vineet Gupta <vineetg@rivosinc.com>
>> Date:   Thu Jul 20 11:15:37 2023 -0700
>>
>>     RISC-V: optim const DF +0.0 store to mem [PR/110748]
>>         Fixes: ef85d150b5963 ("RISC-V: Enable TARGET_SUPPORTS_WIDE_INT")
>>         DF +0.0 is bitwise all zeros so int x0 store to mem can be 
>> used to optimize it.
> [ ... ]
>
>
> So I think the big question WRT your patch is does it still help the 
> case where we weren't pulling the fp+offset computation out of a loop.

I have some numbers for f-m-o v3 vs this. Attached here (vs. inline to 
avoid the Thunderbird mangling the test formatting)

[-- Attachment #2: fmo-vs-fold-fp --]
[-- Type: text/plain, Size: 2698 bytes --]

benchmark	workload #       upstream        upstream +              upstream +
                                 g54e54f77c1        f-m-o               fold-fp-off

500.perlbench_r	0		1217932817476	1217884553366	0.004%	1217928953834	0.000%
		1		743757241201	743655528133	0.014%	743695820426	0.008%
		2		703455646090	703423559298	0.005%	703455296251	0.000%
502.gcc_r	0		195004369104	194973478945	0.016%	194984188400	0.010%
		1		232719938778	232688491113	0.014%	232692379085	0.012%
		2		223443280459	223413616368	0.013%	223424151848	0.009%
		3		186233704624	186206516421	0.015%	186231137616	0.001%
		4		287406394232	287378870279	0.010%	287403707466	0.001%
503.bwaves_r	0		316194043679	316194043679	0.000%	316194043662	0.000%
		1		499293490380	499293490380	0.000%	499293490363	0.000%
		2		389365401615	389365401615	0.000%	389365401598	0.000%
		3		473514310679	473514310679	0.000%	473514310662	0.000%
505.mcf_r	0		689258694902	689254740344	0.001%	689258694887	0.000%
507.cactuBSSN_r	0		3966612364613	3966498234698	0.003%	3966612365068	0.000%
508.namd_r	0		1903766272166	1903766271701	0.000%	1903765987301	0.000%
510.parest_r	0		3512678127316	3512676752062	0.000%	3512677505662	0.008%
511.povray_r	0		3036725558618	3036722265149	0.000%	3036725556997	0.000%
519.lbm_r	0		1134454304533	1134454304533	0.000%	1134454304518	0.000%
520.omnetpp_r	0		1001937885126	1001937884542	0.000%	1001937883931	0.000%
521.wrf_r	0		3959642601629	3959541912013	0.003%	3959642615086	0.000%
523.xalancbmk_r	0		1065004269065	1064981413043	0.002%	1065004132070	0.000%
525.x264_r	0		496492857533	496459367582	0.007%	496477988435	0.003%
		1		1891248078083	1891222197535	0.001%	1890990911614	0.014%
		2		1815609267498	1815561397105	0.003%	1815341248007	0.015%
526.blender_r	0		1672203767444	1671549923427	0.039%	1672224626743  -0.001%
527.cam4_r	0		2326424925038	2320567166886	0.252%	2326333566227	0.004% <-
531.deepsjeng_r	0		1668993359340	1662816376544	0.370%	1668993353038	0.000% <-
538.imagick_r	0		3260965672876	3260965672712	0.000%	3260965672777	0.000%
541.leela_r	0		2034139863891	2034101807341	0.002%	2026647843672	0.368%    <--
544.nab_r	0		1566465507272	1565420628706	0.067%	1566465379674	0.000%
548.exchange2_r	0		2228112071994	2228109962469	0.000%	2228114278251	0.000%
549.fotonik3d_r	0		2255238867247	2255238867246	0.000%	2255238865924	0.000%
554.roms_r	0		2653150555486	2651884870455	0.048%	2653150554877	0.000%
557.xz_r	0		367892301169	367892301167	0.000%	367892301154	0.000%
		1		979549393200	979549393198	0.000%	979549393185	0.000%
		2		525066235331	525066235329	0.000%	525066235316	0.000%
997.specrand_fr	0		453112389	453112389	0.000%	453112374	0.000%
999.specrand_ir	0		453112389	453112389	0.000%	453112374	0.000%