From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga06.intel.com (mga06.intel.com [134.134.136.31]) by dpdk.org (Postfix) with ESMTP id 3B12D4D27 for ; Tue, 9 Apr 2019 19:05:25 +0200 (CEST) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from fmsmga002.fm.intel.com ([10.253.24.26]) by orsmga104.jf.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 09 Apr 2019 10:05:24 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.60,330,1549958400"; d="scan'208";a="159635388" Received: from irsmsx106.ger.corp.intel.com ([163.33.3.31]) by fmsmga002.fm.intel.com with ESMTP; 09 Apr 2019 10:05:22 -0700 Received: from irsmsx103.ger.corp.intel.com ([169.254.3.30]) by IRSMSX106.ger.corp.intel.com ([169.254.8.235]) with mapi id 14.03.0415.000; Tue, 9 Apr 2019 18:05:21 +0100 From: "Richardson, Bruce" To: "Ananyev, Konstantin" , Aaron Conole CC: "dev@dpdk.org" , Jerin Jacob , Gavin Hu , Michael Santana Thread-Topic: [PATCH 3/3] acl: adjust the tests Thread-Index: AQHU7tRnFzUpsbeTZEqJJqSVVkAp/6Yz7P2AgAAh47A= Date: Tue, 9 Apr 2019 17:05:21 +0000 Message-ID: <59AF69C657FD0841A61C55336867B5B072787B4F@IRSMSX103.ger.corp.intel.com> References: <20190408182420.4398-1-aconole@redhat.com> <20190408182420.4398-4-aconole@redhat.com> <2601191342CEEE43887BDE71AB9772580148A94A44@irsmsx105.ger.corp.intel.com> <2601191342CEEE43887BDE71AB9772580148A94D87@irsmsx105.ger.corp.intel.com> In-Reply-To: <2601191342CEEE43887BDE71AB9772580148A94D87@irsmsx105.ger.corp.intel.com> Accept-Language: en-GB, en-US Content-Language: en-US X-MS-Has-Attach: X-MS-TNEF-Correlator: x-titus-metadata-40: eyJDYXRlZ29yeUxhYmVscyI6IiIsIk1ldGFkYXRhIjp7Im5zIjoiaHR0cDpcL1wvd3d3LnRpdHVzLmNvbVwvbnNcL0ludGVsMyIsImlkIjoiNTVlMDgyNjUtZmRlZC00NTcwLWIyYWQtZjQxYzc0NjI4OTczIiwicHJvcHMiOlt7Im4iOiJDVFBDbGFzc2lmaWNhdGlvbiIsInZhbHMiOlt7InZhbHVlIjoiQ1RQX05UIn1dfV19LCJTdWJqZWN0TGFiZWxzIjpbXSwiVE1DVmVyc2lvbiI6IjE3LjEwLjE4MDQuNDkiLCJUcnVzdGVkTGFiZWxIYXNoIjoiXC9uVXlYZldcL2EzK1dVS28rZjdhZ2ZIWW5FOFJGREN0clVGbk5vZ3BBNGtZY1UyUEpmS1ZWbitIdlNOam05azRcLyJ9 x-ctpclassification: CTP_NT dlp-product: dlpe-windows dlp-version: 11.0.600.7 dlp-reaction: no-action x-originating-ip: [163.33.239.181] Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: quoted-printable MIME-Version: 1.0 Subject: Re: [dpdk-dev] [PATCH 3/3] acl: adjust the tests X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Tue, 09 Apr 2019 17:05:25 -0000 > -----Original Message----- > From: Ananyev, Konstantin > Sent: Tuesday, April 9, 2019 5:03 PM > To: Aaron Conole > Cc: dev@dpdk.org; Jerin Jacob ; Gavin Hu > ; Richardson, Bruce ; > Michael Santana > Subject: RE: [PATCH 3/3] acl: adjust the tests >=20 >=20 > > > Hi Aaron, > > > > > >> > > >> This makes the tests pass, and also ensures that on platforms where > > >> the testing is supported, we can properly test the implementation > > >> specific code. One edge case is when we run on x86_64 systems that > > >> don't support AVX2, but where the compiler can generate such > > >> instructions. That could be an enhancement in the future, but for > > >> now at least the tests will pass. > > >> > > >> Signed-off-by: Aaron Conole > > >> --- > > >> app/test/test_acl.c | 62 +++++++++++++-----------------= - > -- > > >> lib/librte_acl/Makefile | 1 + > > >> lib/librte_acl/acl_run_notsup.c | 46 ++++++++++++++++++++++++ > > >> lib/librte_acl/meson.build | 4 +-- > > >> 4 files changed, 73 insertions(+), 40 deletions(-) create mode > > >> 100644 lib/librte_acl/acl_run_notsup.c > > >> > > >> diff --git a/app/test/test_acl.c b/app/test/test_acl.c index > > >> b1f75d1bc..c44faa251 100644 > > >> --- a/app/test/test_acl.c > > >> +++ b/app/test/test_acl.c > > >> @@ -408,6 +408,9 @@ test_classify(void) > > >> return -1; > > >> } > > >> > > >> + /* Always use the scalar testing for now. */ > > >> + rte_acl_set_ctx_classify(acx, RTE_ACL_CLASSIFY_SCALAR); > > >> + > > >> ret =3D 0; > > >> for (i =3D 0; i !=3D TEST_CLASSIFY_ITER; i++) { > > >> > > >> @@ -547,6 +550,7 @@ test_build_ports_range(void) > > >> for (i =3D 0; i !=3D RTE_DIM(test_data); i++) > > >> data[i] =3D (uint8_t *)&test_data[i]; > > >> > > >> + rte_acl_set_ctx_classify(acx, RTE_ACL_CLASSIFY_SCALAR); > > >> for (i =3D 0; i !=3D RTE_DIM(test_rules); i++) { > > >> rte_acl_reset(acx); > > >> ret =3D test_classify_buid(acx, test_rules, i + 1); @@ - > 911,6 > > >> +915,8 @@ test_convert_rules(const char *desc, > > >> return -1; > > >> } > > >> > > >> + rte_acl_set_ctx_classify(acx, RTE_ACL_CLASSIFY_SCALAR); > > >> + > > > > > > As I understand here and above, on x86 you replaced default algo > > > (SSE, AVX2) with scalar one, right? > > > That looks like reduction of test coverage for x86. > > > > In one way, you're right. However, the tests weren't testing what > > they purported anyway. >=20 > Could you explain a bit more here? > What I am seeing: tests were running bot sse(or avx2) and scalar > classify() method. > Now they always running scalar only. > To me it definitely looks like reduction in coverage. >=20 > > Actually, it's just a shift I think (previously, it would have tested > > the AVX2 but I don't see AVX2 having a fallback into the SSE code - > > unlike the SSE code falling back into scalar). >=20 > Not sure I understand you here. > What fallback for AVX2 you expect that you think is missing? >=20 > > > > The tests were failing for a number of reasons when built with meson, >=20 > Ok, but with legacy build system (make) on x86 all tests passes, right? > So the problem is in new build system, not in the test itself. > Why we should compromise our test coverage to make it work with new tools= ? > That just hides the problem without fixing it. > Instead I think the build system needs to be fixed. > Looking at it a bit closely, for .so meson+ninja generates code with > correct version of the function: >=20 > nm x86_64-native-linuxapp-gcc-meson/lib/librte_acl.so.2 | grep > acl_classify_sse > 000000000000fa50 t rte_acl_classify_sse >=20 > So for 'meson -Ddefault_library=3Dshared' > acl_autotest passes without the problem. >=20 > Though for static lib we have both: > nm x86_64-native-linuxapp-gcc-meson/lib/librte_acl.a | grep > acl_classify_sse > 0000000000000000 W rte_acl_classify_sse > 0000000000004880 T rte_acl_classify_sse >=20 > And then linker pickups the wrong one: > nm x86_64-native-linuxapp-gcc-meson/app/test/dpdk-test | grep > acl_classify_sse > 00000000005f6100 W rte_acl_classify_sse >=20 > While for make: > $ nm x86_64-native-linuxapp-gcc-aesmb/lib/librte_acl.a | grep > acl_classify_sse > 0000000000000000 W rte_acl_classify_sse > 0000000000004880 T rte_acl_classify_sse > $ nm x86_64-native-linuxapp-gcc-aesmb/app/test | grep acl_classify_sse > 0000000000240440 T rte_acl_classify_sse >=20 > Linker pickups the right one. >=20 I assume the same issues occurs for AVX2, but for SSE specifically why do w= e even compile up two copies of the function for x86 platforms, since SSE w= ill always be supported? /Bruce From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from dpdk.org (dpdk.org [92.243.14.124]) by dpdk.space (Postfix) with ESMTP id 288E1A0096 for ; Tue, 9 Apr 2019 19:05:27 +0200 (CEST) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id CC3784D3A; Tue, 9 Apr 2019 19:05:26 +0200 (CEST) Received: from mga06.intel.com (mga06.intel.com [134.134.136.31]) by dpdk.org (Postfix) with ESMTP id 3B12D4D27 for ; Tue, 9 Apr 2019 19:05:25 +0200 (CEST) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from fmsmga002.fm.intel.com ([10.253.24.26]) by orsmga104.jf.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 09 Apr 2019 10:05:24 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.60,330,1549958400"; d="scan'208";a="159635388" Received: from irsmsx106.ger.corp.intel.com ([163.33.3.31]) by fmsmga002.fm.intel.com with ESMTP; 09 Apr 2019 10:05:22 -0700 Received: from irsmsx103.ger.corp.intel.com ([169.254.3.30]) by IRSMSX106.ger.corp.intel.com ([169.254.8.235]) with mapi id 14.03.0415.000; Tue, 9 Apr 2019 18:05:21 +0100 From: "Richardson, Bruce" To: "Ananyev, Konstantin" , Aaron Conole CC: "dev@dpdk.org" , Jerin Jacob , Gavin Hu , Michael Santana Thread-Topic: [PATCH 3/3] acl: adjust the tests Thread-Index: AQHU7tRnFzUpsbeTZEqJJqSVVkAp/6Yz7P2AgAAh47A= Date: Tue, 9 Apr 2019 17:05:21 +0000 Message-ID: <59AF69C657FD0841A61C55336867B5B072787B4F@IRSMSX103.ger.corp.intel.com> References: <20190408182420.4398-1-aconole@redhat.com> <20190408182420.4398-4-aconole@redhat.com> <2601191342CEEE43887BDE71AB9772580148A94A44@irsmsx105.ger.corp.intel.com> <2601191342CEEE43887BDE71AB9772580148A94D87@irsmsx105.ger.corp.intel.com> In-Reply-To: <2601191342CEEE43887BDE71AB9772580148A94D87@irsmsx105.ger.corp.intel.com> Accept-Language: en-GB, en-US Content-Language: en-US X-MS-Has-Attach: X-MS-TNEF-Correlator: x-titus-metadata-40: eyJDYXRlZ29yeUxhYmVscyI6IiIsIk1ldGFkYXRhIjp7Im5zIjoiaHR0cDpcL1wvd3d3LnRpdHVzLmNvbVwvbnNcL0ludGVsMyIsImlkIjoiNTVlMDgyNjUtZmRlZC00NTcwLWIyYWQtZjQxYzc0NjI4OTczIiwicHJvcHMiOlt7Im4iOiJDVFBDbGFzc2lmaWNhdGlvbiIsInZhbHMiOlt7InZhbHVlIjoiQ1RQX05UIn1dfV19LCJTdWJqZWN0TGFiZWxzIjpbXSwiVE1DVmVyc2lvbiI6IjE3LjEwLjE4MDQuNDkiLCJUcnVzdGVkTGFiZWxIYXNoIjoiXC9uVXlYZldcL2EzK1dVS28rZjdhZ2ZIWW5FOFJGREN0clVGbk5vZ3BBNGtZY1UyUEpmS1ZWbitIdlNOam05azRcLyJ9 x-ctpclassification: CTP_NT dlp-product: dlpe-windows dlp-version: 11.0.600.7 dlp-reaction: no-action x-originating-ip: [163.33.239.181] Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable MIME-Version: 1.0 Subject: Re: [dpdk-dev] [PATCH 3/3] acl: adjust the tests X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" Message-ID: <20190409170521.nIEtksmKYApu167AegB6wHdaxMcPOf_sRr6L75Yfz7k@z> > -----Original Message----- > From: Ananyev, Konstantin > Sent: Tuesday, April 9, 2019 5:03 PM > To: Aaron Conole > Cc: dev@dpdk.org; Jerin Jacob ; Gavin Hu > ; Richardson, Bruce ; > Michael Santana > Subject: RE: [PATCH 3/3] acl: adjust the tests >=20 >=20 > > > Hi Aaron, > > > > > >> > > >> This makes the tests pass, and also ensures that on platforms where > > >> the testing is supported, we can properly test the implementation > > >> specific code. One edge case is when we run on x86_64 systems that > > >> don't support AVX2, but where the compiler can generate such > > >> instructions. That could be an enhancement in the future, but for > > >> now at least the tests will pass. > > >> > > >> Signed-off-by: Aaron Conole > > >> --- > > >> app/test/test_acl.c | 62 +++++++++++++-----------------= - > -- > > >> lib/librte_acl/Makefile | 1 + > > >> lib/librte_acl/acl_run_notsup.c | 46 ++++++++++++++++++++++++ > > >> lib/librte_acl/meson.build | 4 +-- > > >> 4 files changed, 73 insertions(+), 40 deletions(-) create mode > > >> 100644 lib/librte_acl/acl_run_notsup.c > > >> > > >> diff --git a/app/test/test_acl.c b/app/test/test_acl.c index > > >> b1f75d1bc..c44faa251 100644 > > >> --- a/app/test/test_acl.c > > >> +++ b/app/test/test_acl.c > > >> @@ -408,6 +408,9 @@ test_classify(void) > > >> return -1; > > >> } > > >> > > >> + /* Always use the scalar testing for now. */ > > >> + rte_acl_set_ctx_classify(acx, RTE_ACL_CLASSIFY_SCALAR); > > >> + > > >> ret =3D 0; > > >> for (i =3D 0; i !=3D TEST_CLASSIFY_ITER; i++) { > > >> > > >> @@ -547,6 +550,7 @@ test_build_ports_range(void) > > >> for (i =3D 0; i !=3D RTE_DIM(test_data); i++) > > >> data[i] =3D (uint8_t *)&test_data[i]; > > >> > > >> + rte_acl_set_ctx_classify(acx, RTE_ACL_CLASSIFY_SCALAR); > > >> for (i =3D 0; i !=3D RTE_DIM(test_rules); i++) { > > >> rte_acl_reset(acx); > > >> ret =3D test_classify_buid(acx, test_rules, i + 1); @@ - > 911,6 > > >> +915,8 @@ test_convert_rules(const char *desc, > > >> return -1; > > >> } > > >> > > >> + rte_acl_set_ctx_classify(acx, RTE_ACL_CLASSIFY_SCALAR); > > >> + > > > > > > As I understand here and above, on x86 you replaced default algo > > > (SSE, AVX2) with scalar one, right? > > > That looks like reduction of test coverage for x86. > > > > In one way, you're right. However, the tests weren't testing what > > they purported anyway. >=20 > Could you explain a bit more here? > What I am seeing: tests were running bot sse(or avx2) and scalar > classify() method. > Now they always running scalar only. > To me it definitely looks like reduction in coverage. >=20 > > Actually, it's just a shift I think (previously, it would have tested > > the AVX2 but I don't see AVX2 having a fallback into the SSE code - > > unlike the SSE code falling back into scalar). >=20 > Not sure I understand you here. > What fallback for AVX2 you expect that you think is missing? >=20 > > > > The tests were failing for a number of reasons when built with meson, >=20 > Ok, but with legacy build system (make) on x86 all tests passes, right? > So the problem is in new build system, not in the test itself. > Why we should compromise our test coverage to make it work with new tools= ? > That just hides the problem without fixing it. > Instead I think the build system needs to be fixed. > Looking at it a bit closely, for .so meson+ninja generates code with > correct version of the function: >=20 > nm x86_64-native-linuxapp-gcc-meson/lib/librte_acl.so.2 | grep > acl_classify_sse > 000000000000fa50 t rte_acl_classify_sse >=20 > So for 'meson -Ddefault_library=3Dshared' > acl_autotest passes without the problem. >=20 > Though for static lib we have both: > nm x86_64-native-linuxapp-gcc-meson/lib/librte_acl.a | grep > acl_classify_sse > 0000000000000000 W rte_acl_classify_sse > 0000000000004880 T rte_acl_classify_sse >=20 > And then linker pickups the wrong one: > nm x86_64-native-linuxapp-gcc-meson/app/test/dpdk-test | grep > acl_classify_sse > 00000000005f6100 W rte_acl_classify_sse >=20 > While for make: > $ nm x86_64-native-linuxapp-gcc-aesmb/lib/librte_acl.a | grep > acl_classify_sse > 0000000000000000 W rte_acl_classify_sse > 0000000000004880 T rte_acl_classify_sse > $ nm x86_64-native-linuxapp-gcc-aesmb/app/test | grep acl_classify_sse > 0000000000240440 T rte_acl_classify_sse >=20 > Linker pickups the right one. >=20 I assume the same issues occurs for AVX2, but for SSE specifically why do w= e even compile up two copies of the function for x86 platforms, since SSE w= ill always be supported? /Bruce