From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from aserp2130.oracle.com (aserp2130.oracle.com [141.146.126.79]) by sourceware.org (Postfix) with ESMTPS id 2CD2B39730FB for ; Fri, 15 Jan 2021 16:16:47 +0000 (GMT) DMARC-Filter: OpenDMARC Filter v1.3.2 sourceware.org 2CD2B39730FB Received: from pps.filterd (aserp2130.oracle.com [127.0.0.1]) by aserp2130.oracle.com (8.16.0.42/8.16.0.42) with SMTP id 10FGAK6L145343; Fri, 15 Jan 2021 16:16:44 GMT Received: from aserp3020.oracle.com (aserp3020.oracle.com [141.146.126.70]) by aserp2130.oracle.com with ESMTP id 360kg25ns3-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=OK); Fri, 15 Jan 2021 16:16:43 +0000 Received: from pps.filterd (aserp3020.oracle.com [127.0.0.1]) by aserp3020.oracle.com (8.16.0.42/8.16.0.42) with SMTP id 10FG9oKe093261; Fri, 15 Jan 2021 16:16:43 GMT Received: from aserv0121.oracle.com (aserv0121.oracle.com [141.146.126.235]) by aserp3020.oracle.com with ESMTP id 360kebbcdu-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-GCM-SHA384 bits=256 verify=OK); Fri, 15 Jan 2021 16:16:42 +0000 Received: from abhmp0005.oracle.com (abhmp0005.oracle.com [141.146.116.11]) by aserv0121.oracle.com (8.14.4/8.13.8) with ESMTP id 10FGGf0a018616; Fri, 15 Jan 2021 16:16:42 GMT Received: from dhcp-10-154-190-241.vpn.oracle.com (/10.154.190.241) by default (Oracle Beehive Gateway v4.0) with ESMTP ; Fri, 15 Jan 2021 08:16:41 -0800 From: Qing Zhao Message-Id: Mime-Version: 1.0 (Mac OS X Mail 13.4 \(3608.120.23.2.4\)) Subject: Re: The performance data for two different implementation of new security feature -ftrivial-auto-var-init Date: Fri, 15 Jan 2021 10:16:40 -0600 In-Reply-To: Cc: Richard Sandiford , Richard Biener via Gcc-patches To: Richard Biener References: <33955130-9D2D-43D5-818D-1DCC13FC1988@ORACLE.COM> <89D58812-0F3E-47AE-95A5-0A07B66EED8C@ORACLE.COM> <9585CBB2-0082-4B9A-AC75-250F54F0797C@ORACLE.COM> <51911859-45D5-4566-B588-F828B9D7313B@ORACLE.COM> <9127AAB9-92C8-4A1B-BAD5-2F5F8762DCF9@ORACLE.COM> <5A0F7219-DAFA-4EAA-B845-0E236A108738@ORACLE.COM> <7E70D6B0-CA52-4957-BF84-401AA6E094D7@ORACLE.COM> X-Mailer: Apple Mail (2.3608.120.23.2.4) X-Proofpoint-Virus-Version: vendor=nai engine=6000 definitions=9864 signatures=668683 X-Proofpoint-Spam-Details: rule=notspam policy=default score=0 adultscore=0 suspectscore=0 spamscore=0 mlxlogscore=999 malwarescore=0 bulkscore=0 mlxscore=0 phishscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.12.0-2009150000 definitions=main-2101150100 X-Proofpoint-Virus-Version: vendor=nai engine=6000 definitions=9864 signatures=668683 X-Proofpoint-Spam-Details: rule=notspam policy=default score=0 malwarescore=0 suspectscore=0 clxscore=1015 impostorscore=0 spamscore=0 priorityscore=1501 mlxscore=0 phishscore=0 mlxlogscore=999 bulkscore=0 adultscore=0 lowpriorityscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.12.0-2009150000 definitions=main-2101150100 X-Spam-Status: No, score=-4.1 required=5.0 tests=BAYES_00, DKIMWL_WL_HIGH, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, HTML_MESSAGE, RCVD_IN_MSPIKE_H2, SPF_HELO_PASS, SPF_PASS, TXREP, UNPARSEABLE_RELAY autolearn=ham autolearn_force=no version=3.4.2 X-Spam-Checker-Version: SpamAssassin 3.4.2 (2018-09-13) on server2.sourceware.org Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: quoted-printable X-Content-Filtered-By: Mailman/MimeDel 2.1.29 X-BeenThere: gcc-patches@gcc.gnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Gcc-patches mailing list List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 15 Jan 2021 16:16:51 -0000 > On Jan 15, 2021, at 2:11 AM, Richard Biener wrote: >=20 >=20 >=20 > On Thu, 14 Jan 2021, Qing Zhao wrote: >=20 >> Hi,=20 >> More data on code size and compilation time with CPU2017: >> ********Compilation time data: the numbers are the slowdown against = the >> default =E2=80=9Cno=E2=80=9D: >> benchmarks A/no D/no >> =20 >> 500.perlbench_r 5.19% 1.95% >> 502.gcc_r 0.46% -0.23% >> 505.mcf_r 0.00% 0.00% >> 520.omnetpp_r 0.85% 0.00% >> 523.xalancbmk_r 0.79% -0.40% >> 525.x264_r -4.48% 0.00% >> 531.deepsjeng_r 16.67% 16.67% >> 541.leela_r 0.00% 0.00% >> 557.xz_r 0.00% 0.00% >> =20 >> 507.cactuBSSN_r 1.16% 0.58% >> 508.namd_r 9.62% 8.65% >> 510.parest_r 0.48% 1.19% >> 511.povray_r 3.70% 3.70% >> 519.lbm_r 0.00% 0.00% >> 521.wrf_r 0.05% 0.02% >> 526.blender_r 0.33% 1.32% >> 527.cam4_r -0.93% -0.93% >> 538.imagick_r 1.32% 3.95% >> 544.nab_r 0.00% 0.00% >> =46rom the above data, looks like that the compilation time impact >> from implementation A and D are almost the same. >> *******code size data: the numbers are the code size increase against = the >> default =E2=80=9Cno=E2=80=9D: >> benchmarks A/no D/no >> =20 >> 500.perlbench_r 2.84% 0.34% >> 502.gcc_r 2.59% 0.35% >> 505.mcf_r 3.55% 0.39% >> 520.omnetpp_r 0.54% 0.03% >> 523.xalancbmk_r 0.36% 0.39% >> 525.x264_r 1.39% 0.13% >> 531.deepsjeng_r 2.15% -1.12% >> 541.leela_r 0.50% -0.20% >> 557.xz_r 0.31% 0.13% >> =20 >> 507.cactuBSSN_r 5.00% -0.01% >> 508.namd_r 3.64% -0.07% >> 510.parest_r 1.12% 0.33% >> 511.povray_r 4.18% 1.16% >> 519.lbm_r 8.83% 6.44% >> 521.wrf_r 0.08% 0.02% >> 526.blender_r 1.63% 0.45% >> 527.cam4_r 0.16% 0.06% >> 538.imagick_r 3.18% -0.80% >> 544.nab_r 5.76% -1.11% >> Avg 2.52% 0.36% >> =46rom the above data, the implementation D is always better than A, = it=E2=80=99s a >> surprising to me, not sure what=E2=80=99s the reason for this. >=20 > D probably inhibits most interesting loop transforms (check SPEC FP > performance). The call to .DEFERRED_INIT is marked as ECF_CONST: /* A function to represent an artifical initialization to an = uninitialized automatic variable. The first argument is the variable itself, the second argument is the initialization type. */ DEF_INTERNAL_FN (DEFERRED_INIT, ECF_CONST | ECF_LEAF | ECF_NOTHROW, = NULL) So, I assume that such const call should minimize the impact to loop = optimizations. But yes, it will still inhibit some of the loop = transformations. > It will also most definitely disallow SRA which, when > an aggregate is not completely elided, tends to grow code. Make sense to me.=20 The run-time performance data for D and A are actually very similar as I = posted in the previous email (I listed it here for convenience) Run-time performance overhead with A and D: benchmarks A / no D /no 500.perlbench_r 1.25% 1.25% 502.gcc_r 0.68% 1.80% 505.mcf_r 0.68% 0.14% 520.omnetpp_r 4.83% 4.68% 523.xalancbmk_r 0.18% 1.96% 525.x264_r 1.55% 2.07% 531.deepsjeng_ 11.57% 11.85% 541.leela_r 0.64% 0.80% 557.xz_ -0.41% -0.41% 507.cactuBSSN_r 0.44% 0.44% 508.namd_r 0.34% 0.34% 510.parest_r 0.17% 0.25% 511.povray_r 56.57% 57.27% 519.lbm_r 0.00% 0.00% 521.wrf_r -0.28% -0.37% 526.blender_r 16.96% 17.71% 527.cam4_r 0.70% 0.53% 538.imagick_r 2.40% 2.40% 544.nab_r 0.00% -0.65% avg 5.17% 5.37% Especially for the SPEC FP benchmarks, I didn=E2=80=99t see too much = performance difference between A and D.=20 I guess that the RTL optimizations might be enough to get rid of most of = the overhead introduced by the additional initialization.=20 >=20 >> ********stack usage data, I added -fstack-usage to the compilation = line when >> compiling CPU2017 benchmarks. And all the *.su files were generated = for each >> of the modules. >> Since there a lot of such files, and the stack size information are = embedded >> in each of the files. I just picked up one benchmark 511.povray to >> check. Which is the one that=20 >> has the most runtime overhead when adding initialization (both A and = D).=20 >> I identified all the *.su files that are different between A and D = and do a >> diff on those *.su files, and looks like that the stack size is much = higher >> with D than that with A, for example: >> $ diff build_base_auto_init.D.0000/bbox.su >> build_base_auto_init.A.0000/bbox.su5c5 >> < bbox.cpp:1782:12:int pov::sort_and_split(pov::BBOX_TREE**, >> pov::BBOX_TREE**&, long int*, long int, long int) 160 static >> --- >> > bbox.cpp:1782:12:int pov::sort_and_split(pov::BBOX_TREE**, >> pov::BBOX_TREE**&, long int*, long int, long int) 96 static >> $ diff build_base_auto_init.D.0000/image.su >> build_base_auto_init.A.0000/image.su >> 9c9 >> < image.cpp:240:6:void pov::bump_map(double*, pov::TNORMAL*, double*) = 624 >> static >> --- >> > image.cpp:240:6:void pov::bump_map(double*, pov::TNORMAL*, double*) = 272 >> static >> =E2=80=A6. >> Looks like that implementation D has more stack size impact than A.=20= >> Do you have any insight on what the reason for this? >=20 > D will keep all initialized aggregates as aggregates and live which > means stack will be allocated for it. With A the usual optimizations > to reduce stack usage can be applied. I checked the routine =E2=80=9Cpoverties::bump_map=E2=80=9D in = 511.povray_r since it has a lot stack increase=20 due to implementation D, by examine the IR immediate before RTL = expansion phase. =20 (image.cpp.244t.optimized), I found that we have the following = additional statements for the array elements: void pov::bump_map (double * EPoint, struct TNORMAL * Tnormal, double * = normal) { =E2=80=A6 double p3[3]; double p2[3]; double p1[3]; float colour3[5]; float colour2[5]; float colour1[5]; =E2=80=A6 # DEBUG BEGIN_STMT colour1 =3D .DEFERRED_INIT (colour1, 2); colour2 =3D .DEFERRED_INIT (colour2, 2); colour3 =3D .DEFERRED_INIT (colour3, 2); # DEBUG BEGIN_STMT MEM [(double[3] *)&p1] =3D p1$0_144(D); MEM [(double[3] *)&p1 + 8B] =3D p1$1_135(D); MEM [(double[3] *)&p1 + 16B] =3D p1$2_138(D); p1 =3D .DEFERRED_INIT (p1, 2); # DEBUG D#12 =3D> MEM [(double[3] *)&p1] # DEBUG p1$0 =3D> D#12 # DEBUG D#11 =3D> MEM [(double[3] *)&p1 + 8B] # DEBUG p1$1 =3D> D#11 # DEBUG D#10 =3D> MEM [(double[3] *)&p1 + 16B] # DEBUG p1$2 =3D> D#10 MEM [(double[3] *)&p2] =3D p2$0_109(D); MEM [(double[3] *)&p2 + 8B] =3D p2$1_111(D); MEM [(double[3] *)&p2 + 16B] =3D p2$2_254(D); p2 =3D .DEFERRED_INIT (p2, 2); # DEBUG D#9 =3D> MEM [(double[3] *)&p2] # DEBUG p2$0 =3D> D#9 # DEBUG D#8 =3D> MEM [(double[3] *)&p2 + 8B] # DEBUG p2$1 =3D> D#8 # DEBUG D#7 =3D> MEM [(double[3] *)&p2 + 16B] # DEBUG p2$2 =3D> D#7 MEM [(double[3] *)&p3] =3D p3$0_256(D); MEM [(double[3] *)&p3 + 8B] =3D p3$1_258(D); MEM [(double[3] *)&p3 + 16B] =3D p3$2_260(D); p3 =3D .DEFERRED_INIT (p3, 2); =E2=80=A6. } I guess that the above =E2=80=9CMEM =E2=80=A6.. =3D =E2=80=A6=E2=80= =9D are the ones that make the differences. Which phase introduced them? >=20 >> Let me know if you have any comments and suggestions. >=20 > First of all I would check whether the prototype implementations > work as expected. I have done such check with small testing cases already, checking the IR = generated with the implementation A or D, mainly Focus on *.c.006t.gimple. and *.c.*t.expand, all worked as expected.=20 For the CPU2017, for example as the above, I also checked the IR for = both A and D, looks like all worked as expected. Thanks.=20 Qing >=20 > Richard. >=20 >=20 >> thanks. >> Qing >> On Jan 13, 2021, at 1:39 AM, Richard Biener >> wrote: >>=20 >> On Tue, 12 Jan 2021, Qing Zhao wrote: >>=20 >> Hi,=20 >>=20 >> Just check in to see whether you have any comments >> and suggestions on this: >>=20 >> FYI, I have been continue with Approach D >> implementation since last week: >>=20 >> D. Adding calls to .DEFFERED_INIT during >> gimplification, expand the .DEFFERED_INIT during >> expand to >> real initialization. Adjusting uninitialized pass >> with the new refs with =E2=80=9C.DEFFERED_INIT=E2=80=9D. >>=20 >> For the remaining work of Approach D: >>=20 >> ** complete the implementation of >> -ftrivial-auto-var-init=3Dpattern; >> ** complete the implementation of uninitialized >> warnings maintenance work for D.=20 >>=20 >> I have completed the uninitialized warnings >> maintenance work for D. >> And finished partial of the >> -ftrivial-auto-var-init=3Dpattern implementation.=20 >>=20 >> The following are remaining work of Approach D: >>=20 >> ** -ftrivial-auto-var-init=3Dpattern for VLA; >> **add a new attribute for variable: >> __attribute((uninitialized) >> the marked variable is uninitialized intentionaly >> for performance purpose. >> ** adding complete testing cases; >>=20 >> Please let me know if you have any objection on my >> current decision on implementing approach D.=20 >>=20 >> Did you do any analysis on how stack usage and code size are >> changed=20 >> with approach D? How does compile-time behave (we could gobble >> up >> lots of .DEFERRED_INIT calls I guess)? >>=20 >> Richard. >>=20 >> Thanks a lot for your help. >>=20 >> Qing >>=20 >> On Jan 5, 2021, at 1:05 PM, Qing Zhao >> via Gcc-patches >> wrote: >>=20 >> Hi, >>=20 >> This is an update for our previous >> discussion.=20 >>=20 >> 1. I implemented the following two >> different implementations in the latest >> upstream gcc: >>=20 >> A. Adding real initialization during >> gimplification, not maintain the >> uninitialized warnings. >>=20 >> D. Adding calls to .DEFFERED_INIT >> during gimplification, expand the >> .DEFFERED_INIT during expand to >> real initialization. Adjusting >> uninitialized pass with the new refs >> with =E2=80=9C.DEFFERED_INIT=E2=80=9D. >>=20 >> Note, in this initial implementation, >> ** I ONLY implement >> -ftrivial-auto-var-init=3Dzero, the >> implementation of >> -ftrivial-auto-var-init=3Dpattern=20 >> is not done yet. Therefore, the >> performance data is only about >> -ftrivial-auto-var-init=3Dzero.=20 >>=20 >> ** I added an temporary option >> -fauto-var-init-approach=3DA|B|C|D to >> choose implementation A or D for=20 >> runtime performance study. >> ** I didn=E2=80=99t finish the uninitialized >> warnings maintenance work for D. (That >> might take more time than I expected).=20 >>=20 >> 2. I collected runtime data for CPU2017 >> on a x86 machine with this new gcc for >> the following 3 cases: >>=20 >> no: default. (-g -O2 -march=3Dnative ) >> A: default + >> -ftrivial-auto-var-init=3Dzero >> -fauto-var-init-approach=3DA=20 >> D: default + >> -ftrivial-auto-var-init=3Dzero >> -fauto-var-init-approach=3DD=20 >>=20 >> And then compute the slowdown data for >> both A and D as following: >>=20 >> benchmarks A / no D /no >>=20 >> 500.perlbench_r 1.25% 1.25% >> 502.gcc_r 0.68% 1.80% >> 505.mcf_r 0.68% 0.14% >> 520.omnetpp_r 4.83% 4.68% >> 523.xalancbmk_r 0.18% 1.96% >> 525.x264_r 1.55% 2.07% >> 531.deepsjeng_ 11.57% 11.85% >> 541.leela_r 0.64% 0.80% >> 557.xz_ -0.41% -0.41% >>=20 >> 507.cactuBSSN_r 0.44% 0.44% >> 508.namd_r 0.34% 0.34% >> 510.parest_r 0.17% 0.25% >> 511.povray_r 56.57% 57.27% >> 519.lbm_r 0.00% 0.00% >> 521.wrf_r -0.28% -0.37% >> 526.blender_r 16.96% 17.71% >> 527.cam4_r 0.70% 0.53% >> 538.imagick_r 2.40% 2.40% >> 544.nab_r 0.00% -0.65% >>=20 >> avg 5.17% 5.37% >>=20 >> =46rom the above data, we can see that in >> general, the runtime performance >> slowdown for=20 >> implementation A and D are similar for >> individual benchmarks. >>=20 >> There are several benchmarks that have >> significant slowdown with the new added >> initialization for both >> A and D, for example, 511.povray_r, >> 526.blender_, and 531.deepsjeng_r, I >> will try to study a little bit >> more on what kind of new initializations >> introduced such slowdown.=20 >>=20 >> =46rom the current study so far, I think >> that approach D should be good enough >> for our final implementation.=20 >> So, I will try to finish approach D with >> the following remaining work >>=20 >> ** complete the implementation of >> -ftrivial-auto-var-init=3Dpattern; >> ** complete the implementation of >> uninitialized warnings maintenance work >> for D.=20 >>=20 >> Let me know if you have any comments and >> suggestions on my current and future >> work. >>=20 >> Thanks a lot for your help. >>=20 >> Qing >>=20 >> On Dec 9, 2020, at 10:18 AM, >> Qing Zhao via Gcc-patches >> >> wrote: >>=20 >> The following are the >> approaches I will implement >> and compare: >>=20 >> Our final goal is to keep >> the uninitialized warning >> and minimize the run-time >> performance cost. >>=20 >> A. Adding real >> initialization during >> gimplification, not maintain >> the uninitialized warnings. >> B. Adding real >> initialization during >> gimplification, marking them >> with =E2=80=9Cartificial_init=E2=80=9D.=20 >> Adjusting uninitialized >> pass, maintaining the >> annotation, making sure the >> real init not >> Deleted from the fake >> init.=20 >> C. Marking the DECL for an >> uninitialized auto variable >> as =E2=80=9Cno_explicit_init=E2=80=9D during >> gimplification, >> maintain this >> =E2=80=9Cno_explicit_init=E2=80=9D bit till >> after >> pass_late_warn_uninitialized, >> or till pass_expand,=20 >> add real initialization >> for all DECLs that are >> marked with >> =E2=80=9Cno_explicit_init=E2=80=9D. >> D. Adding .DEFFERED_INIT >> during gimplification, >> expand the .DEFFERED_INIT >> during expand to >> real initialization. >> Adjusting uninitialized pass >> with the new refs with >> =E2=80=9C.DEFFERED_INIT=E2=80=9D. >>=20 >> In the above, approach A >> will be the one that have >> the minimum run-time cost, >> will be the base for the >> performance >> comparison.=20 >>=20 >> I will implement approach D >> then, this one is expected >> to have the most run-time >> overhead among the above >> list, but >> Implementation should be the >> cleanest among B, C, D. >> Let=E2=80=99s see how much more >> performance overhead this >> approach >> will be. If the data is >> good, maybe we can avoid the >> effort to implement B, and >> C.=20 >>=20 >> If the performance of D is >> not good, I will implement B >> or C at that time. >>=20 >> Let me know if you have any >> comment or suggestions. >>=20 >> Thanks. >>=20 >> Qing >>=20 >> --=20 >> Richard Biener >> SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 >> Nuernberg, >> Germany; GF: Felix Imend=C3=B6rffer; HRB 36809 (AG Nuernberg)