Evidence rules / Editorial record
Link notes
§01Start here
-
https://cacm.acm.org/research/roofline-an-insightful-visual-performance-model-for-multicore-architectures/: curl 403 with and without a browser user agent and WebFetch 403, while the repository link checker's GET with full browser headers is served; confirmed in a browser that the page is the full CACM article with a PDF download and no sign-in. The DOI 10.1145/1498765.1498785 resolves with a temporary redirect to dl.acm.org, which serves a Cloudflare challenge to non-browsers, so the magazine page is linked instead.
-
http://www.rdrop.com/users/paulmck/scalability/paper/whymb.2010.07.23a.pdf: 200 over plain HTTP; the host does not answer on the HTTPS port, so the http address is kept, as the memory section also does. The PDF was downloaded and its first page confirms the title, the author and the section plan.
-
https://www.intel.com/content/www/us/en/content-details/671488/intel-64-and-ia-32-architectures-optimization-reference-manual-volume-1.html: 200 to bare curl, 403 to a browser-like user agent, so the link checker will report it as blocked; WebFetch confirmed it is the document-library entry for Volume 1. Intel's download URL for the manual chains through three permanent redirects to the current PDF, document 248966-050US, April 2024, which is why the library page rather than a PDF is linked.
-
https://sites.google.com/site/analysismethods/yasin-pubs: 200; the author's own publications page, which links the ISPASS 2014 paper as a Google Drive file named TopDown-Yasin-ISPASS14.pdf (200, verified as the paper). The DOI 10.1109/ISPASS.2014.6844459 returns 202 with a bot challenge and is paywalled, so the author's page is linked as the brief prefers.
-
https://shop.elsevier.com/books/computer-architecture/hennessy/978-0-443-15406-5: 200; the publisher's page for the seventh edition, marked as the latest. The companion-site URL on www.elsevier.com redirects permanently to shop.elsevier.com, so the final URL is used in Rejected.
-
https://www.youtube.com/watch?v=bSkpMdDe4g4: 200; YouTube's oEmbed endpoint confirms the title and that the channel is CppCon.
-
https://github.com/dendibakh/perf-book: 200; the repository holds the book's source under CC0. The author's separate PDF distribution site did not answer curl from here and is not linked.
-
https://www.akkadia.org/drepper/cpumemory.pdf: 200 on the author's own domain; the LWN and FreeBSD copies in Rejected are also live but are mirrors.
-
https://www.agner.org/optimize/optimizing_cpp.pdf: 200; the copy read for this section is dated 2026-09-10, so the entry needs no version in its title.
-
https://support.arm.com/documentation/109898/latest/: redirected permanently from https://developer.arm.com/documentation/109898/latest/, so the final URL is used. Both hosts return 200 to bare curl and 403 to a browser-like user agent, so the link checker will report it as blocked. Confirmed in a browser that the page is the Neoverse V2 Core Software Optimization Guide, version 3.0, PDF download only, no sign-in. The N2 guide moved the same way. Both are listed under Rejected only.
-
https://chipsandcheese.com/p/arms-neoverse-v2-in-awss-graviton-4: 200. Listed under Rejected only.
-
https://www.barroso.org/publications/TheTailAtScale.pdf: 200 on the author's own domain. Listed under Rejected only.
-
https://mirrors.edge.kernel.org/pub/linux/kernel/people/paulmck/perfbook/perfbook.html: 200. Listed under Rejected only.
-
https://queue.acm.org/detail.cfm?id=3372264: curl 403 with and without a browser user agent; confirmed in a browser that the page is the article. Listed under Rejected only.
-
https://perfwiki.github.io/main/tutorial/: the old perf.wiki.kernel.org tutorial URL redirects permanently, via archive.kernel.org, to perfwiki.github.io, so the new home is used in Rejected.
-
https://www.intel.com/content/www/us/en/docs/vtune-profiler/cookbook/2025-0/top-down-microarchitecture-analysis-method.html: the "current" cookbook URL redirects permanently to this versioned one, so the final URL is used in Rejected; 200 to bare curl, 403 to a browser-like user agent like the other Intel page.
-
The Reproduce it line links the benchmark by the path relative to this file so that the format checker resolves it; on assembly it becomes benchmarks/04-cache-latency/README.md, relative to the README, as in the other sections' lines.
-
The repository link checker as first committed raised a TypeError on every URL (its opener was called with a context argument it does not accept); the working copy carries the two-line fix, and the verdicts above come from a run of that copy, which reported no dead and no moved links and only the blocked hosts noted here.
-
All other URLs in this file returned 200 at their stated address with no redirect.
§02One instruction, end to end
-
https://developer.arm.com/documentation/109898/latest/ and https://developer.arm.com/documentation/109914/latest/: both 301 to https://support.arm.com/documentation/109898/latest/ and https://support.arm.com/documentation/109914/latest/, so the final URLs are used. With a browser user agent the old host returns 403. The pages are JavaScript only; confirmed in a browser as "Arm Neoverse V2 Core Software Optimization Guide, Version 3.0 (Latest)" and "Arm Neoverse N2 Software Optimization Guide, Version 5.0 (Latest)", each offered as a PDF download. The V2 guide is listed once in this section, under Execute, with a reason that walks the core from fetch to issue before the tables, so that one Arm core is followed at every stage; the front-end copy and the N2 guide sit in Rejected.
-
https://ieeexplore.ieee.org/document/5392028, https://ieeexplore.ieee.org/document/4607, https://ieeexplore.ieee.org/document/945363 and https://ieeexplore.ieee.org/document/1310765: curl gets 202 (bot challenge) with and without a browser user agent, which the checker counts as ok; confirmed as the Tomasulo paper (IBM Journal of Research and Development), the Smith and Pleszkun paper (IEEE Transactions on Computers), the ISLPED 2001 micro-op cache paper (DOI 10.1109/LPE.2001.945363) and the ISCA 2004 memory-level parallelism paper (DOI 10.1109/ISCA.2004.1310765), the last two through the Crossref record for each DOI, which gives title, venue and authors. https://ieeexplore.ieee.org/document/476078 in Rejected behaves the same way and is the Smith and Sohi survey.
-
https://dl.acm.org/doi/10.1145/1028176.1006708: 403 to curl with and without a browser user agent, as with the other ACM pages; the Crossref record for the DOI confirms it as the SIGARCH Computer Architecture News printing of the same memory-level parallelism paper. The authors were at Sun, which hosts no copy, and the Semantic Scholar and ResearchGate copies are aggregators, so the IEEE proceedings page is listed and the ACM page sits in Rejected.
-
https://dl.acm.org/doi/10.1145/379240.379253: curl gets 403 with and without a browser user agent, and WebFetch gets 403; the Crossref record for the DOI confirms "Focusing processor policies via critical-path prediction", ISCA 2001. The author's publications page at https://homes.cs.washington.edu/~bodik/papers.html lists the paper with no PDF, and the copies at people.eecs.berkeley.edu/~kubitron, rsim.cs.illinois.edu and cs.cmu.edu are course mirrors, so the publisher page is used under rule 4.
-
https://cseweb.ucsd.edu/~calder/papers/MICRO-99-FDP.pdf: 200, application/pdf, on an author's own page; the PDF text confirms "Fetch Directed Instruction Prefetching", Reinman, Calder and Austin, MICRO-32, November 1999.
-
https://docs.amd.com/v/u/en-US/58455_1.00: curl gets 200 for an application shell; confirmed in a browser as "Software Optimization Guide for the AMD Zen5 Microarchitecture (PUB) (58455pub)", revision 1.00, released 2024-08-15, and the portal search API lists it as 58455_1.00.zip. WebFetch gets 401 on this host, so a browser was used.
-
https://docs.amd.com/v/u/en-US/57647_zen4_sog_1.01: curl gets 200 for the same shell, but the portal renders 401 and its search API lists no document 57647 (queries "57647" and "Zen4"). https://www.amd.com/en/support/tech-docs/software-optimization-guide-for-the-amd-zen4-microarchitecture times out for curl without a user agent and with one redirects to the documentation hub, which points back to the portal. https://www.amd.com/content/dam/amd/en/documents/processor-tech-docs/software-optimization-guides/57647.zip is 404 and https://docs.amd.com/v/u/en-US/57647 is Not Found. Left in Rejected until AMD restores the document.
-
https://shop.elsevier.com/books/computer-architecture/hennessy/978-0-12-811905-1 (6th edition) carries a "Newer edition is available" link to https://shop.elsevier.com/books/computer-architecture/hennessy/978-0-443-15406-5, which returns 200 and is confirmed by fetch as the 7th edition (Hennessy, Patterson and Kozyrakis, 2025) with chapter 3 titled "Instruction-Level Parallelism and Its Exploitation". Both pages now sit in Rejected, since the book is held to Start here. The companion site https://www.elsevier.com/books-and-journals/book-companion/9780128119051 301s to https://shop.elsevier.com/books/book-companion/9780128119051 and is not used.
-
https://people.csail.mit.edu/emer/media/papers/1998.06.isca.storesets.pdf (the path most search results give) is 404; the current path from the author's publications page is https://people.csail.mit.edu/emer/media/papers/1990s/1998/1998.06.isca.storesets.pdf, which returns application/pdf. Used because the ACM DL copy https://dl.acm.org/doi/10.1145/279358.279378 is paywalled and returns 403 to curl.
-
https://blog.stuffedcow.net/2013/05/measuring-rob-capacity/ and https://blog.stuffedcow.net/2014/01/x86-memory-disambiguation/: both http and https return 200 with no redirect; https is used.
-
https://fgiesen.wordpress.com/2022/09/05/entropy-decoding-in-oodle-data-x86-64-3-stream-huffman-decoders/ and https://fgiesen.wordpress.com/2023/10/29/entropy-decoding-in-oodle-data-x86-64-6-stream-huffman-decoders/ (Rejected): 200, no redirect; the page text confirms the Skylake critical-path figure and the measured figure quoted under Claims.
-
https://dougallj.wordpress.com/2022/08/20/faster-zlib-deflate-decompression-on-the-apple-m1-and-x86/: 200, no redirect; the author's own report of his implementation, with the code at https://github.com/dougallj/zlib-dougallj. The post is used rather than the repository because the reasoning about the refill chain is in the post, not the README.
-
https://research.google/pubs/asmdb-understanding-and-mitigating-front-end-stalls-in-warehouse-scale-computers/ (Rejected): 200, the authors' institutional page with the PDF linked; the ACM DOI 10.1145/3307650.3322234 returns 403 to curl.
-
https://www.intel.com/content/www/us/en/content-details/671488/intel-64-and-ia-32-architectures-optimization-reference-manual-volume-1.html and https://www.intel.com/content/www/us/en/content-details/841076/intel-mitigations-for-jump-conditional-code-erratum.html: 200 to curl's default agent, 403 to the browser-like agent that scripts/check_links.py sends (reported as blocked, not dead); confirmed by fetch as document 671488 version 050 and document 841076 dated 2019-11-01. The PDFs are served from cdrdv2-public.intel.com behind a download button, so the document library pages are linked. https://www.intel.com/content/www/us/en/developer/articles/technical/intel-sdm.html in Rejected behaves the same way.
-
https://support.arm.com/documentation/109898/latest/ and https://support.arm.com/documentation/109914/latest/: likewise 200 to curl's default agent and 403 to the checker's browser-like agent; see the first note for the browser confirmation.
-
scripts/check_links.py as first committed passed context= to OpenerDirector.open(), which raised TypeError and reported every link DEAD before any request was sent; the working-tree copy moves the SSL context into an HTTPSHandler and the results above come from that copy.
-
https://www.agner.org/optimize/microarchitecture.pdf, https://www.agner.org/optimize/instruction_tables.pdf, https://www.agner.org/optimize/optimizing_assembly.pdf: 200, application/pdf; these are the author's canonical paths and the index page states the current revision date of each. The PDF text of the microarchitecture manual carries a "Misprediction penalty" passage under each core's branch prediction section, a micro-op cache section per core that has one, and "Data delay between different execution domains" tables per core, which is what the reason line names.
-
https://en.wikichip.org/wiki/amd/microarchitectures/zen_4 was considered for Rejected as a rule 1 failure (secondary summary) but the host refused the connection during checking, so it is not linked.
§03Microarchitecture
-
https://ieeexplore.ieee.org/document/491460: curl returns 202 (bot challenge), not content; DOI 10.1109/40.491460 resolves to the page through doi.org and the Crossref registry confirms title, venue and year.
-
https://ieeexplore.ieee.org/document/5392028 and https://ieeexplore.ieee.org/document/945363 (both left to section 2): the same 202 bot challenge; DOIs 10.1147/rd.111.0025 and 10.1109/LPE.2001.945363 confirmed through Crossref. The micro-op cache paper's extended journal version is DOI 10.1109/TVLSI.2003.814327 (https://ieeexplore.ieee.org/document/1234399).
-
https://cseweb.ucsd.edu/~tullsen/isca95.pdf: curl 200, the ISCA 1995 paper from the first author's own page, twelve pages. The ACM record https://dl.acm.org/doi/10.1145/223982.224449 answers 403 to curl and the Crossref registry confirms title, venue and year; the project page https://dada.cs.washington.edu/smt/ (200) carries the same paper as PostScript only, which is why the PDF is linked.
-
https://www.intel.com/content/dam/www/public/us/en/documents/research/2002-vol06-iss-1-intel-technology-journal.pdf: curl 403 with and without a browser user agent; fetched and confirmed as Intel Technology Journal Volume 06 Issue 01, published 14 February 2002, the Hyper-Threading Technology issue, with the architecture and microarchitecture article at page 4 and its duplicated, partitioned and shared resource description in the text. The content/www path of the same file name answers 403 to curl and redirects a fetch into Intel's 404 handler, so the content/dam path, which is the file itself, is used.
-
https://users.elis.ugent.be/~leeckhou/papers/ispass06-eyerman.pdf: curl 200, the ISPASS 2006 paper from the third author's page, eleven pages. The IEEE record https://ieeexplore.ieee.org/document/1620789 answers 202 (bot challenge) and DOI 10.1109/ISPASS.2006.1620789 is confirmed through Crossref.
-
https://users.elis.ugent.be/~leeckhou/papers/tocs09.pdf: curl 200, the complete TOCS 2009 paper (thirty-seven pages) from the second author's page. The ACM record https://dl.acm.org/doi/10.1145/1534909.1534910 answers 403 and is confirmed through Crossref. The ISCA 2004 origin, DOI 10.1145/1028176.1006729 (https://dl.acm.org/doi/10.1145/1028176.1006729, 403, Crossref confirmed), has no author copy on either author's page and is paywalled, so the journal form is listed and the origin is recorded under Rejected.
-
https://jilp.org/jwac-2/program/cbp3_07_seznec.pdf: curl 200, four pages, first page confirmed as the ITTAGE paper; cbp3_03 at the same host is the ISL-TAGE paper, not this one.
-
https://dougallj.github.io/applecpu/firestorm.html: curl 200; the overview page states the counter method and links to the base and SIMD instruction tables, whose entries link to per-test measurement pages; the repository is https://github.com/dougallj/applecpu (200).
-
https://docs.kernel.org/admin-guide/pm/cpufreq.html: curl 200; the vendor-specific driver documents https://docs.kernel.org/admin-guide/pm/intel_pstate.html and https://docs.kernel.org/admin-guide/pm/amd-pstate.html (both 200) are reached from it and are not listed separately.
-
https://shop.elsevier.com/books/computer-architecture/hennessy/978-0-443-15406-5 (left to Start here and section 2): curl 200; the 7th edition page, which the publisher put in place of the 6th edition (ISBN 978-0-12-811905-1) in 2025.
-
https://davidwall.info/papers/WRL-TR-93.6.pdf: the HP Labs host for DEC WRL reports (hpl.hp.com/techreports) no longer resolves; the author's own site carries both the 1991 note and the 1993 report, and states the report replaces the note.
-
https://dl.acm.org/doi/10.1145/139669.139702: curl 403 (the ACM library refuses automated clients); the DOI is confirmed through the Crossref registry (ISCA 1992) and the paper is in the ACM open backfile. The authors' own HTML copy at https://suif.stanford.edu/papers/lam92/paper.html is the full paper, but the server sends an incomplete certificate chain that fails Python's verifier and so the repository link check, which is why the ACM page is linked instead.
-
https://ftp.cs.wisc.edu/sohi/papers/1997/isca.complexity.pdf: curl 200; a Wisconsin group archive rather than an author page, but the PDF is the ISCA 1997 paper. The canonical record is DOI 10.1145/264107.264201 at https://dl.acm.org/doi/10.1145/264107.264201, in the ACM open backfile, which answers 403 to curl.
-
https://www.intel.com/content/www/us/en/content-details/671488/intel-64-and-ia-32-architectures-optimization-reference-manual-volume-1.html (left to other sections): curl 403 with and without a browser user agent; confirmed by fetch as the document page (version 050, content ID 671488, download button).
-
https://www.intel.com/content/www/us/en/content-details/787036/intel-64-and-ia-32-architectures-optimization-reference-manual-documentation-volume-2-earlier-generations-of-intel-64-and-ia-32-processor-architectures-throughput-and-latency.html (rejected): curl 200 without a user agent and 403 with a browser one; confirmed by fetch as content ID 787036, version 050, dated 2023-08-25, and the downloaded PDF's chapter 7 tables list CPUID signatures up to Skylake.
-
https://www.intel.com/content/www/us/en/developer/articles/technical/software-security-guidance/technical-documentation/speculative-execution-side-channel-mitigations.html: curl 403; confirmed by fetch as revision 2.0 of the mitigations document.
-
https://support.arm.com/documentation/109898/latest/ (left to other sections): both this host and https://developer.arm.com/documentation/109898/latest/ answer 403 to curl (Akamai) with no redirect header; a browser load of the developer.arm.com path lands on this host and shows the Neoverse V2 Core Software Optimization Guide, version 3.0, PDF download only.
-
https://docs.amd.com/v/u/en-US/58455_1.00 and https://docs.amd.com/v/u/en-US/software-techniques-for-managing-speculation: the AMD portal is a JavaScript application that answers 200 to any path, so a 200 proves nothing; both documents were confirmed through the portal's search API (58455pub, revision 1.00, 2024-08-15; 70586 and 70628 for the white paper) and by a browser load. The white paper's old amd.com PDF URL under processor-tech-docs answers 301 to the portal path used here. The Zen 5 guide's Appendix A states that the spreadsheet is based on estimates.
-
https://newsroom.intel.com/editorials/intel-security-issue-update-initial-performance-data-results-client-systems/ (rejected): 403 from a corpredirect 404 redirector; the 2018 performance posts are gone from the vendor site.
-
https://developer.apple.com/documentation/apple-silicon/cpu-optimization-guide (rejected): curl 200, but a browser load shows a description page stating the guide needs a registered developer account and an additional agreement.
-
https://www.hpl.hp.com/techreports/Compaq-DEC/WRL-TN-36.pdf (rejected): curl 000, host does not resolve.
-
scripts/check_links.py reports ok 39, moved 0, blocked 8, dead 0 for this file; every blocked URL is one of the ACM, IEEE, Arm and Intel hosts confirmed by fetch or through Crossref above.
§04Memory hierarchy
-
https://www.intel.com/content/www/us/en/content-details/671190/...: curl 403 with and without a browser user agent; confirmed by fetch as SDM Volume 3A, the ID the intel-sdm.html library page links; the same note applies to 671488 (Optimization Reference Manual Volume 1) and 671098 (SDM Volume 4).
-
https://support.arm.com/documentation/102375/latest/: https://developer.arm.com/documentation/102375/latest/ answers 301 to this address; both hosts return 403 to curl; confirmed in a browser as the Neoverse V2 Core Technical Reference Manual, r0p2. The same 301 and 403 apply to ddi0487 (Arm ARM, version M.c) and to 109898 (Neoverse V2 Software Optimization Guide, issue 3.0, now under Rejected), so the final support.arm.com addresses are used.
-
https://www2.eecs.berkeley.edu/Pubs/TechRpts/1993/Archive/CSD-93-767.pdf: redirected 301 from https://www2.eecs.berkeley.edu/Pubs/TechRpts/1993/CSD-93-767.pdf; the final address is used.
-
https://www.kernel.org/doc/Documentation/ABI/testing/sysfs-devices-system-cpu: curl 200, plain text; the cache attribute block under cpu*/cache/index* lists coherency_line_size, number_of_sets, ways_of_associativity, size, level, type and shared_cpu_map. The stable ABI directory has no cache file, so the testing file is the canonical one.
-
https://dl.acm.org/doi/10.1145/1250662.1250709 and https://dl.acm.org/doi/10.1145/285930.285979: 403 to curl and to the fetch tool; both confirmed through Crossref as Adaptive insertion policies for high performance caching (ISCA 2007) and Lockup-free instruction fetch/prefetch cache organization (25 Years of ISCA, 1998). The original ISCA 1981 record, https://dl.acm.org/doi/10.5555/800052.801868, is not a registered handle and could not be confirmed by any client, so the reprint record of the same paper is used.
-
https://ieeexplore.ieee.org/document/134547: curl 202 with an empty body, which the link checker counts as live; confirmed by resolving the Crossref record 10.1109/ISCA.1990.134547 (Improving direct-mapped cache performance by the addition of a small fully-associative cache and prefetch buffers, ISCA 1990) to this document number. The same 202 and DOI confirmation apply to 814600, 1310765 and 9229804 under Rejected.
-
https://faculty.cc.gatech.edu/~hyesoon/lee_taco12.pdf: redirected 301 from https://www.cc.gatech.edu/~hyesoon/lee_taco12.pdf; the final address is used.
-
https://www.cl.cam.ac.uk/~pes20/armv8-mca/: curl 200; the fetch tool confirms the project page carries the paper PDF (armv8-mca-draft.pdf, the address listed before), an errata page, the Flat operational model and the herd form of the axiomatic model, so the page is listed rather than the bare PDF. https://dl.acm.org/doi/10.1145/3158107 is the PACMPL record and answers 403 to curl.
-
https://github.com/herd/herdtools7 and https://github.com/acmel/dwarves: curl 200; the herdtools7 README names litmus7 with Power, ARM, AArch64 and X86 targets and herd7 for cat models; the dwarves man page shows pahole printing holes and cache-line boundaries per struct.
-
https://gcc.gnu.org/onlinedocs/gcc/Other-Builtins.html#index-_005f_005fbuiltin_005fprefetch: curl 200 and the anchor is present in the page; the text gives the rw and locality arguments and the rule that an invalid address does not fault.
-
https://www.microsoft.com/en-us/research/publication/cache-conscious-structure-definition-2/: curl 403 with and without a browser user agent; the fetch tool confirms the publication record for Cache-Conscious Structure Definition, PLDI 1999, with the PDF at https://www.microsoft.com/en-us/research/wp-content/uploads/2016/12/definition_distr.pdf, which is also 403 to curl. The ACM record https://dl.acm.org/doi/10.1145/301618.301635 answers 403 as well and is confirmed through Crossref.
-
https://www.usenix.org/conference/osdi16/technical-sessions/presentation/kwon: curl 200 with a browser user agent; the fetch tool gets 403; the page body confirms the paper title and links the open PDF.
-
http://www.rdrop.com/users/paulmck/scalability/paper/whymb.2010.07.23a.pdf: served over plain HTTP only; the HTTPS port does not answer, so the http address is kept.
-
https://docs.amd.com/v/u/en-US/58455_1.00: curl 200; the page is the vendor portal record for publication 58455 rev 1.00 and serves a zip containing the PDF and the latency spreadsheet.
-
https://www.youtube.com/watch?v=rX0ItVEVjHc: curl 200; the oEmbed record names the uploading channel as CppCon, upload date 2014-09-29.
-
https://docs.kernel.org/admin-guide/mm/transhuge.html, hugetlbpage.html and https://docs.kernel.org/mm/page_tables.html: 200 with no redirect; www.kernel.org/doc/html/latest/ serves the same pages without redirecting, docs.kernel.org is used as the current documentation host.
-
https://dl.acm.org/doi/10.1145/1785414.1785443 and https://dl.acm.org/doi/10.1145/800015.808204: 403 to curl; the first is replaced by the authors' copy on cl.cam.ac.uk, the second is confirmed through Crossref as ISCA 1984 and sits under Rejected.
§05Measurement
-
https://perfwiki.github.io/main/tutorial/: the old https://perf.wiki.kernel.org/index.php/Tutorial returns 301 to archive.kernel.org/oldwiki and on to perfwiki.github.io; the perf wiki has moved to GitHub Pages and the new host is used. The page defines skid for interrupt-based samples in its own words, so the IBS paper owns the term in the section and the tutorial's reason no longer claims it.
-
https://man7.org/linux/man-pages/man2/perf_event_open.2.html and https://man7.org/linux/man-pages/man2/clock_gettime.2.html: curl 200, no redirect, with and without a browser user agent; the man-pages project's own renderings. The first states that events come in two flavours, counting and sampled, and defines PERF_SAMPLE_BRANCH_STACK; the second defines CLOCK_MONOTONIC, CLOCK_MONOTONIC_RAW, CLOCK_PROCESS_CPUTIME_ID and clock_getres.
-
https://support.arm.com/documentation/ddi0487/latest/: the developer.arm.com form returns 301 to support.arm.com, so the final host is used; plain curl gets 200, curl with a browser UA gets 403; the page is a JavaScript viewer that fetch reads as an empty shell, confirmed in a browser as "Arm Architecture Reference Manual for A-profile architecture" (DDI 0487, version M.c, latest). The three support.arm.com documents held under Rejected (109709, 109429 and 109528) answer plain curl and a browser UA the same way.
-
https://www.intel.com/content/www/us/en/developer/articles/technical/intel-sdm.html and https://www.intel.com/content/www/us/en/developer/tools/oneapi/vtune-profiler-documentation.html: curl with no UA gets 200, curl with a browser UA gets 403; both confirmed by fetch as the SDM download page and the VTune documentation index. The user guide alias .../vtune-profiler/user-guide/current/ is a 301 to a versioned path, so the index page is linked and its reason is written of the index. The SDM entry keeps the short title used in the single-thread and inference sections, one title per URL.
-
https://queue.acm.org/detail.cfm?id=2927301: 403 to curl with and without a UA and to fetch; confirmed in a browser, where the page title read "The Flame Graph ... Queue: Vol 14, No 2" before a bot check appeared. ACM Queue is the official open upload, so it is the canonical home despite the bot block.
-
https://www.amd.com/content/dam/amd/en/documents/archived-tech-docs/white-papers/AMD_IBS_paper_EN.pdf: HEAD returns 200 application/pdf, but the response time varies from under a second to over 25 seconds between runs, so a checker needs a 60 second timeout or it will report the link dead; a plain GET is held open until it times out; a GET with browser headers returns the PDF (PDF-1.4, the title and the 2007 AMD copyright on the first pages, and the word skid defined with a worked profile), so the document is confirmed.
-
https://docs.amd.com/v/u/en-US/57238 and https://docs.amd.com/r/en-US/57368-uProf-user-guide: curl 200; the PPR viewer confirmed by fetch as document 57238 for Family 1Ah Model 02h revision C1, the user guide as uProf User Guide 57368. The /v/u/ form is AMD's document-library listing with the ZIP download; the HTML reader form https://docs.amd.com/r/en-US/57238 answers 200 to HEAD but is a soft 404 on GET, so the listing is linked.
-
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/perf/Documentation and https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/perf/Documentation/perf-arm-spe.txt: curl 200 with and without a browser UA; the listing names perf-stat.txt, perf-record.txt, perf-mem.txt, perf-c2c.txt and perf-arm-spe.txt, and the plain view of perf-arm-spe.txt confirms the document as perf-arm-spe(1), with the five sampling stages, the filter options and decoding at perf report or perf script time; fetch gets 403 on the tree URLs.
-
https://web.eece.maine.edu/~vweaver/projects/deterministic/ispass2013_deterministic.pdf: curl 200 application/pdf, no redirect, on the first author's project page (https://web.eece.maine.edu/~vweaver/projects/deterministic/, also 200), which carries the ISPASS paper, the benchmark source and a later draft covering more parts; the PDF is linked because the entry title is the paper's.
-
https://llvm.org/docs/Benchmarking.html: curl 200, no redirect, with and without a browser UA; the page is "Benchmarking tips" in the LLVM documentation, whose Linux and Linux Intel sections set randomize_va_space, scaling_governor, a cpuset shield, SMT sibling offlining, tmpfs and intel_pstate no_turbo, and which cites the ASPLOS measurement-bias paper listed before it.
-
https://docs.kernel.org/admin-guide/pm/cpufreq.html: curl 200, no redirect, with and without a browser UA; the page is "CPU Performance Scaling" and links intel_pstate.html and amd-pstate.html on the same host, both 200, held under Rejected.
-
https://sape.inf.usi.ch/publications/asplos09: curl 200; the page's PDF link goes through the ACM author-authorised download, which is free.
-
https://research.google/pubs/google-wide-profiling-a-continuous-profiling-infrastructure-for-data-centers/: curl 200; the publication page is linked rather than the static PDF, matching the abs-over-pdf rule.
-
https://emeryberger.com/research/coz/ and https://emeryberger.com/research/stabilizer/ (not used): both now 302 to github.com/emeryberger, so the author's university copy and arXiv are used for the papers.
-
All remaining URLs returned 200 with no redirect to plain curl.
§06Models
-
https://cacm.acm.org/research/roofline-an-insightful-visual-performance-model-for-multicore-architectures/: curl 403 with and without a browser user agent, fetch 403; confirmed in a browser as the full article with a PDF download and no sign-in, and its reference list cites the listed queueing book and the listed proof of L = λW. The DOI 10.1145/1498765.1498785 resolves to dl.acm.org, which serves a bot challenge, so the magazine page is linked as in the Start here section.
-
https://amcr.lbl.gov/departments/computer-science-department/ppan/roofline-performance-model/ (Rejected): the old crd.lbl.gov roofline URL returns 404 (via a hop to amcr.lbl.gov), so the current AMCR page is recorded; curl 200 with and without a user agent, fetch 403, content read through curl and confirmed as the roofline hub, which names LIKWID as the Linux roofline tool alongside the toolkit, the ERT and Advisor.
-
https://bitbucket.org/berkeleylab/cs-roofline-toolkit/: curl 200; the page is a JavaScript app, so the README was read from the raw master branch and names the Empirical Roofline Tool and the Roofline Visualizer. The last commit is from November 2020, so the toolkit measures the roofs but is not where new counter groups appear.
-
https://spiral.ece.cmu.edu/pub-spiral/pubfile/ispass-2013_177.pdf: curl 200 with and without a browser user agent, downloaded and read. The Spiral project page (abstract.jsp?id=177), which the ETH group's publication list links, serves this PDF, and its title page names the five ETH Zurich authors, so it is the authors' own copy. The IEEE copy is in Rejected as the paywalled duplicate.
-
https://github.com/RRZE-HPC/likwid: curl 200 with and without a user agent; the README was read from the raw master branch and lists likwid-perfctr and likwid-bench. The groups directory carries MEM_DP groups with an operational intensity metric for Sapphire Rapids, Granite Rapids and Grace, FLOPS_DP and MEM groups for Zen 4, Zen 4c and Zen 5, and TMA groups for Sapphire Rapids, Granite Rapids and Zen 5. Last push September 2026.
-
https://ieeexplore.ieee.org/document/6506838: curl returns 202 with a bot challenge, with and without a user agent; the DOI 10.1109/L-CA.2013.6 is a temporary redirect to the same document, so the document page is linked. Confirmed in a browser as the paper with abstract, section list and a sign-in wall for the body; the Crossref record for the DOI matches. No author or institutional copy exists (the INESC-ID PDF is 404 and the ULisboa portal carries only the abstract).
-
https://www.intel.com/content/www/us/en/docs/advisor/user-guide/2025-1/analyze-cpu-roofline.html: the "current" alias of the page returns 301 to this versioned path, so the final URL is used; curl 200, confirmed by fetch as the Analyze CPU Roofline page of the Advisor user guide.
-
https://sites.google.com/site/analysismethods/yasin-pubs: curl 200; the author's page links the ISPASS 2014 PDF on Google Drive, which was downloaded and confirmed as the paper (ten pages, v1.02).
-
https://github.com/intel/perfmon/blob/main/TMA_Metrics-full.xlsx: curl 200; the README of the repository names the TMA spreadsheets as the official TMA releases. The README's reference to a
TMA_Metrics.xlsxat the root is stale, that path is 404, and the full spreadsheet is the file present. -
https://perfwiki.github.io/main/top-down-analysis/ (Rejected): curl 200; confirmed by fetch as the perf maintainers' top-down page for Linux 6.1 and later, Intel-only, with a note that multiplexing lowers accuracy.
-
https://github.com/andikleen/pmu-tools: curl 200; the toplev manual, read from the repository wiki, states that multiplexing can cause measurement errors because several nodes correlate data from different groups, that it works best on workloads that repeat the same thing and badly on many short events, and that --no-multiplex reruns a repeatable workload instead.
-
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/perf/pmu-events/arch/x86/amdzen4/pipeline.json: curl 200 with and without a user agent; the file was read through the plain view and defines the PipelineL1 and PipelineL2 metric groups with each metric's event formula and description, including smt_contention. The amdzen5 directory carries the same file with longer metric names.
-
https://docs.amd.com/r/en-US/57368-uProf-user-guide/Pipeline-Utilization (Rejected): curl 200 but the portal is a JavaScript app that answers 200 to any path, so the page was confirmed by fetch as the Pipeline Utilization page of the uProf User Guide, revision 5.1, which defines the level-one and level-two categories in words and names the AMDuProfPcm option that reports them for Zen 4 and Zen 5.
-
https://support.arm.com/documentation/109542/latest/: https://developer.arm.com/documentation/109542/latest/ returns 301 to this host, so the final URL is used. Curl 200, fetch returns only the viewer shell; confirmed in a browser as "Arm CPU Telemetry Solution Topdown Methodology Specification", version 2.0, issue 109542_02_02_en. The same host form is used for the V2 telemetry specification in Rejected.
-
https://gitlab.arm.com/telemetry-solution/telemetry-solution: curl 200 bare and 202 with a browser user agent, both of which the link checker counts as ok; the README was read from the raw main branch. It links the Neoverse V1 methodology white paper as the example of the methodology with case studies. The github.com/ARM-software/telemetry-solution path is 404, and GitHub copies are third-party mirrors.
-
https://support.arm.com/documentation/109199/latest/: https://developer.arm.com/documentation/109199/latest/ returns 301 to this host, so the final URL is used. Curl 200 bare and 403 to a browser user agent, and fetch returns only the viewer shell. The document library lists the number as "Arm Neoverse V1 Performance Analysis Methodology", and the same paper is served from Arm's own CDN at https://armkeil.blob.core.windows.net/developer/Files/pdf/white-paper/neoverse-v1-core-performance-analysis.pdf, the address the telemetry repository README links; it was downloaded from there and confirmed as "Arm Neoverse V1 Core: Performance Analysis Methodology", whose chapter 4 is the case study on the Arrow CSV parser. The document library page is linked as the canonical home, in the same host form as the specification.
-
https://arxiv.org/abs/1208.2908: curl 200; the abs page, whose abstract states that ECM predicts the scaling and saturation of bandwidth-limited loop kernels on a multicore chip, recorded in Rejected so the choice of the stencil paper is documented.
-
https://dl.acm.org/doi/10.1145/1465482.1465560: curl 403 with and without a browser user agent, fetch 403, and a browser shows a Cloudflare human-verification step that was not completed. Identity confirmed through the Crossref record for the DOI (AFIPS '67 Spring, ACM Press). No author copy exists; course mirrors are in Rejected. The link checker reports this host as blocked, not dead.
-
http://www.johngustafson.net/pubs/pub13/amdahl.htm: the author's own HTML of the CACM article; curl 200 over http, and https fails at the TLS handshake, so the http form is the only working address. Confirmed by reading the page, which carries the full text and references.
-
https://research.cs.wisc.edu/multifacet/papers/ieeecomputer08_amdahl_multicore.pdf: curl 200; the authors' group copy of the IEEE Computer article, confirmed by reading the first page.
-
https://www.perfdynamics.com/Papers/njgCMG93.pdf: curl 200; a ten-page scan with no text layer, page one rendered and confirmed as the CMG 1993 paper.
-
https://www.perfdynamics.com/iBook/gcap.html: curl 200; the author's book page with contents and corrigenda, used in place of the Springer page, whose plain URL returns 303 through a cookie check.
-
https://github.com/smoeding/usl: curl 200; the brief's "usl4r" was not found under that name, and the R package usl by this author is the package that name refers to, with CRAN as its distribution.
-
https://cacm.acm.org/research/measuring-parallel-processor-performance/ (Rejected): curl 403 with and without a browser user agent, fetch 403. Identity confirmed through the Crossref record for the DOI 10.1145/78607.78614 (Communications of the ACM, 1990), whose dl.acm.org copy is paywalled; the magazine page carries the abstract only and no author copy was found.
-
https://pubsonline.informs.org/doi/10.1287/opre.9.3.383: curl 403 with and without a browser user agent, fetch 403, and a browser shows a Cloudflare human-verification step that was not completed. Identity confirmed through the Crossref record for the DOI (Operations Research, 1961). Paywalled, no author copy found. The link checker reports this host as blocked, not dead.
-
https://projecteuclid.org/journals/annals-of-mathematical-statistics/volume-24/issue-3/Stochastic-Processes-Occurring-in-the-Theory-of-Queues-and-their-Analysis-by-the/10.1214/aoms/1177728975.full: curl 200; confirmed in a browser as the paper with a plain PDF download and no sign-in, while the download URL itself serves a bot check to curl.
-
https://www.cambridge.org/core/journals/mathematical-proceedings-of-the-cambridge-philosophical-society/article/abs/single-server-queue-in-heavy-traffic/81C55BC00A68FE6D5385638AA0B0AF37: the DOI 10.1017/S0305004100036094 is a temporary redirect to this page, so the publisher's page is linked; curl 200, confirmed by fetch as the note, paywalled beyond the first page, no author copy found.
-
https://homes.cs.washington.edu/~lazowska/qsp/ and https://www.cs.cmu.edu/~harchol/PerformanceModeling/book.html: curl 200; the authors' own pages, the first with the full text free by returned copyright, the second with a first chapter, errata and tutorial slides.
-
https://researchportal.ulisboa.pt/en/publications/cache-aware-roofline-model-upgrading-the-loft (not used): 301 to a trailing-slash form; carries only the abstract and DOI.
-
https://www.intel.com/content/www/us/en/docs/advisor/user-guide/2025-1/analyze-cpu-roofline.html and all three support.arm.com pages: 200 to bare curl and 403 to a browser-like user agent, so the link checker reports them as blocked; each was confirmed by fetch, by download or in a browser as noted above.
-
scripts/check_links.py run on this file after the edits: 48 unique links, 39 ok, 0 moved, 0 dead, 9 blocked (dl.acm.org x2, pubsonline.informs.org x2, support.arm.com x3, intel.com x2), each confirmed above. Both cacm.acm.org pages answered the checker's client with 200 on this run while plain curl still receives a Cloudflare 403, so the host can appear as either ok or blocked between runs.
-
All remaining URLs returned 200 with no redirect to plain curl.
§07Single-thread optimisation
-
https://developer.arm.com/documentation/ddi0487/latest/ and https://developer.arm.com/documentation/102476/latest/: 301 to https://support.arm.com/documentation/ddi0487/latest/ and https://support.arm.com/documentation/102476/latest/; a real browser lands on the support host too, so the final URLs are used. Plain curl gets 200 there; curl or the repo checker with a browser user agent gets 403 (Akamai). Confirmed in a browser: "Arm Architecture Reference Manual for A-profile architecture", DDI 0487 M.c, and "Introduction to SVE" version 1.1.
-
https://docs.riscv.org/reference/isa/unpriv/v-st-ext.html (Rejected block): the ratified text in the RISC-V Ratified Specifications Library, 200 to curl, the same URL the watchlist carries; https://github.com/riscv/riscv-v-spec 301s to the archived https://github.com/riscvarchive/riscv-v-spec, whose v1.0 release still says "frozen for public review".
-
https://www.intel.com/content/www/us/en/developer/articles/technical/intel-sdm.html and https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html: 200 to plain curl, 403 to the repo checker's user agent; confirmed by fetch as the manual index (document 671200 and the volume set) and the Intrinsics Guide (version 3.6.9).
-
https://dl.acm.org/doi/10.1145/113445.113449 and https://dl.acm.org/doi/10.1145/106972.106981: curl 403 with and without a browser user agent, fetch 403, and the browser shows a Cloudflare bot check that was not attempted. Confirmed instead through doi.org (302 to dl.acm.org, via portal.acm.org for the second) and Crossref metadata (Wolf and Lam, PLDI 1991; Lam, Rothberg and Wolf, ASPLOS-IV 1991). Both are in the ACM backfile that was made open access in 2022. The SUIF group copy at https://suif.stanford.edu/papers/lam-asplos91.pdf serves 200 to curl but the host's TLS chain is incomplete, which the repo checker reports as blocked, so the ACM URL is used.
-
https://dl.acm.org/doi/10.1145/53990.54022, https://dl.acm.org/doi/10.1145/197320.197366 and https://dl.acm.org/doi/10.1145/29873.29875 (Rejected block): curl 403; Crossref confirms Lam, PLDI 1988; Carr and Kennedy, TOPLAS November 1994; Allen and Kennedy, TOPLAS October 1987.
-
https://gmplib.org/~tege/divcnst-pldi94.pdf (Rejected block): 200, the author's copy of the PLDI 1994 paper; https://dl.acm.org/doi/10.1145/178243.178249 is 403 and Crossref confirms Granlund and Montgomery.
-
https://siboehm.com/articles/22/Fast-MMM-on-CPU (Rejected block): 200 to curl and to fetch; confirmed as the walkthrough with its results table, hardware paragraph and code listings.
-
https://clang.llvm.org/docs/UsersManual.html#controlling-floating-point-behavior (Rejected block): 200, anchor present, and the section defines -ffp-contract and -fassociative-math. The scoped pragma clang fp reassociate that benchmarks/07 uses is defined on https://clang.llvm.org/docs/LanguageExtensions.html#extensions-to-specify-floating-point-flags (200), not on this page. https://gcc.gnu.org/wiki/FloatingPointMath (Rejected block): 200.
-
https://gcc.gnu.org/onlinedocs/gcc/Common-Attributes.html#index-target_005fclones (Rejected block): 200, anchor present; https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html now meta-refreshes to Common-Attributes.html, so the latter is canonical.
-
https://lemire.me/blog/2022/06/23/filtering-numbers-quickly-with-sve-on-amazon-graviton-3-processors/ (Rejected block) and https://lemire.me/blog/2021/07/14/faster-sorted-array-unions-by-reducing-branches/: 200 to plain curl and to fetch (confirmed as the posts, each with its results table and code link), 403 to the repo checker's user agent.
-
https://arxiv.org/abs/2205.05982 (Rejected block) and https://arxiv.org/abs/1509.05053: 200; the abs pages are used, and the second is the authors' copy of the ACM JEA paper.
-
https://www.informit.com/store/hackers-delight-9780321842688: 200, title confirmed as "Hacker's Delight, 2nd Edition". The author's former domain hackersdelight.org now serves an unrelated French site, so the publisher page is the canonical home, as shop.elsevier.com is for other books in the list.
-
https://inria.hal.science/hal-01100647: redirected from hal.inria.fr; plain curl gets the record (200), a browser user agent gets the Anubis proof-of-work page. Confirmed through the HAL API record and the PDF, "Branch Prediction and the Performance of Interpreters - Don't Trust Folklore", CGO 2015.
-
http://0x80.pl/notesen/2016-11-28-simd-strfind.html (Rejected block): plain HTTP by design; the HTTPS certificate on that host belongs to another site, so the URL stays http. 200 to curl.
-
https://www.usenix.org/conference/nsdi19/presentation/wang-xiang: 200 to curl, 403 to fetch; confirmed by curl body as the NSDI '19 open-access page for the paper.
-
https://www.intel.com/content/www/us/en/content-details/671488/intel-64-and-ia-32-architectures-optimization-reference-manual-volume-1.html: 200 to plain curl, 403 to the repo checker's user agent; confirmed by downloading document 671488 (248966-050US, April 2024, with the Data Layout Optimizations and AoS to SoA sections). https://www.intel.com/content/www/us/en/content-details/828965/intel-advanced-vector-extensions-10-2-intel-avx10-2-architecture-specification.html (Rejected block): 200 to plain curl, 403 to the checker, confirmed by fetch as document 828965; the separate AVX10.1 specification, document 784267, now 301s to Intel's 404 redirector.
-
https://pharr.org/matt/assets/ispc.pdf: author's copy of the InPar 2012 paper; the publisher copy is paywalled. 200.
-
https://www.agner.org/optimize/optimizing_cpp.pdf: 200; the PDF is dated 2026-09-10 in its front matter.
-
https://github.com/riscv-non-isa/rvv-intrinsic-doc (Rejected block): 301 to https://github.com/riscv-non-isa/riscv-rvv-intrinsic-doc; the final URL is used.
-
https://developer.arm.com/documentation/102340/latest/ (Rejected block): 301 to https://support.arm.com/documentation/102340/latest/, which is used; 200 to plain curl, 403 to a browser user agent and to the repo checker, the same Akamai behaviour as the other Arm documents. Confirmed in a browser: "Learn the architecture - Introducing SVE2 guide", version 1.0.
-
scripts/check_links.py run on this file after the trim reports ok 35, moved 0, blocked 18, dead 0; every blocked URL is one of the ACM, Arm, Intel and lemire.me pages confirmed above.
§08Compilers and codegen
-
https://research.google/pubs/autofdo-automatic-feedback-directed-optimization-for-warehouse-scale-applications/: redirected from https://research.google/pubs/pub45290/ (the old numeric slug); the final URL is used. The page has no PDF link, and the ACM DOI 10.1145/2854038.2854044 returns 403 to curl and to WebFetch and presents a bot check in a browser, so the authors' page is the listed source. Google still serves the full text at https://storage.googleapis.com/gweb-research2023-media/pubtools/pdf/45290.pdf (curl 200, application/pdf, confirmed to be the CGO 2016 paper), which was read to fill the Claims fields.
-
https://dl.acm.org/doi/10.1145/93542.93550: curl 403 with and without a browser user agent, WebFetch 403; no author page or preprint found, so the paper is listed under Rejected rather than as an entry.
-
https://lists.llvm.org/pipermail/llvm-dev/2019-September/135393.html: curl 200. Chosen over the ASPLOS paper's DOI 10.1145/3575693.3575727, which returns 403 to curl and to WebFetch and presents a bot check in a browser.
-
https://www.youtube.com/watch?v=bSkpMdDe4g4, https://www.youtube.com/watch?v=fHNmRkzxHWs, https://www.youtube.com/watch?v=nXaxk27zwlk, https://www.youtube.com/watch?v=2EWejmkKlxs, https://www.youtube.com/watch?v=rHIkrotSwcc: curl 200; WebFetch returns only the page footer, so the title and uploader were read from the page metadata, which gives ownerChannelName "CppCon" for all five. The bSkpMdDe4g4 entry now carries the same title as its Start here listing so one URL has one title in the README.
-
https://arxiv.org/abs/1807.06735: abs page used over the PDF per rule 4; the PDF was read to fill the Claims fields.
-
https://github.com/llvm/llvm-project/tree/main/bolt: replaces the README blob https://github.com/llvm/llvm-project/blob/main/bolt/README.md listed before review; the directory renders the same README and is the project's home in the monorepo. Both curl 200.
-
https://clang.llvm.org/docs/UsersManual.html#controlling-floating-point-behavior, https://clang.llvm.org/docs/UsersManual.html#options-to-emit-optimization-reports and https://clang.llvm.org/docs/UsersManual.html#profile-guided-optimization: all three fragments confirmed present as element ids in the current page; the -O level descriptions that older manuals carried are no longer in this document, which is why the -O level entry points at the clang Command Guide. The three fragments are deliberate separate entries for three distinct sections of one document, in two subsections. The floating-point section was read to confirm that -ffast-math implies -fassociative-math and -ffp-contract=fast, and the PGO section to confirm it names llvm-profgen and AutoFDO's create_llvm_prof as the sampled-profile converters.
-
https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html and https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html#index-fprofile-use: the fragment confirmed as an element id on the page; the bare page and the fragment were separate entries for two distinct sections of one document until the fragment entry was trimmed for length, so the bare page now stands for both. The page also carries index-ffast-math and index-fsemantic-interposition, the GCC counterparts of the Clang floating-point entry and the interposition cost in How To Write Shared Libraries.
-
https://clang.llvm.org/docs/CommandGuide/clang.html#cmdoption-O0: curl 200; fragment confirmed as an element id, and the text under it carries the -O level definitions and the -Ofast deprecation notice.
-
https://gcc.gnu.org/onlinedocs/gcc/Developer-Options.html#index-fopt-info: curl 200; fragment confirmed as an element id, and the text under it defines the missed, optimized and note message kinds and the vec pass group.
-
https://gcc.gnu.org/onlinedocs/gcc/x86-Options.html and https://gcc.gnu.org/onlinedocs/gcc/AArch64-Options.html: curl 200, no redirect; the x86 page read to confirm the -march against -mtune definition, the psABI micro-architecture levels and -mprefer-vector-width, and the AArch64 page to confirm -msve-vector-bits and its vector-length agnostic and vector-length specific modes.
-
https://gcc.gnu.org/onlinedocs/gcc/Function-Multiversioning.html: curl 200, no redirect; the page read to confirm target_clones, target_version and the resolver the dynamic linker runs through the ELF indirect function extension. The old Common-Function-Attributes.html URL for the attribute text is now a redirect stub, so the attribute page was not listed.
-
https://itanium-cxx-abi.github.io/cxx-abi/abi.html: curl 200, no redirect; section 3.1.2.3 Non-Trivial Parameters read to confirm the caller-allocated temporary passed by reference.
-
https://www.akkadia.org/drepper/dsohowto.pdf: curl 200, application/pdf; title, author and the interposition, GOT and PLT, and -fvisibility sections confirmed from the PDF text.
-
https://gcc.gnu.org/onlinedocs/gccint/LTO-Overview.html: curl 200, no redirect; the LTO and WHOPR modes and the LGEN, WPA and LTRANS stages confirmed on the page.
-
https://blog.llvm.org/2011/05/what-every-c-programmer-should-know.html: curl 200, no redirect; the signed-overflow and type-based alias analysis passages confirmed by fetch, with the zero_array loop as the worked example.
-
https://github.com/google/autofdo: curl 200; create_llvm_prof.cc, create_gcov.cc, the perf.data reader, the LBR aggregators and the Arm SPE sample reader confirmed in the repository tree, and the README notes that the Propeller code has moved to llvm-propeller.
-
https://llvm.org/docs/CommandGuide/llvm-exegesis.html: curl 200, no redirect; moved from the vectorisation subsection to sit beside llvm-mca, whose scheduling model it checks.
-
https://uops.info/: curl 200 with and without a browser user agent. Removed from this section on review as the third listing in the README.
-
All other URLs: curl 200 with no redirect.
§09Concurrency
-
https://www.labs.hpe.com/techreports/2004/HPL-2004-209.pdf: curl 200 with and without a browser user agent, PDF, nine pages, read to confirm it is the report version of the paper. The former host www.hpl.hp.com does not answer, the successor's HTML landing page redirects (301) to a generic Hewlett Packard Labs page, and its http form does the same, so the https PDF is the only address. The server sends only its leaf certificate, so OpenSSL-based clients fail verification and scripts/check_links.py reports it blocked with a TLS chain note, while macOS curl and browsers fetch the issuer and succeed. The author's own site links the dead host and holds only the talk slides, the ACM copy is behind a bot check, and no other author-controlled copy was found, so the URL stays as the canonical live copy with this note. Now under Rejected, trimmed for length.
-
https://rsim.cs.illinois.edu/Pubs/08PLDI.pdf: curl 200 over https and http, PDF, eleven pages, read to confirm it is the paper. The co-author's group page, used because the originating lab's copy fails certificate verification as above.
-
https://eel.is/c++draft/atomics: curl 200, the [atomics] clause of the current working draft.
-
https://www.open-std.org/jtc1/sc22/wg14/www/docs/n1570.pdf: curl 200. The C11 committee draft, whose clause 7.17 is the atomics text. Now under Rejected.
-
https://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html: curl 200. The page covers x86, POWER, Armv7, Armv8 in both AArch32 and AArch64, and Itanium.
-
https://www.kernel.org/doc/Documentation/memory-barriers.txt: curl 200, plain text served straight from the kernel tree. Section 4 rejected this URL as belonging here.
-
https://github.com/herd/herdtools7: curl 200 with and without a browser user agent. The README names herd7, litmus7, diy7 and klitmus7 and points at http://diy.inria.fr/ as the documentation home, which also answers 200 over https.
-
https://www.cs.rochester.edu/u/scott/papers/1991_TOCS_synch.pdf and https://www.cs.rochester.edu/u/scott/papers/1996_PODC_queues.pdf: curl 200, the authors' own copies. The fetch tool cannot parse the scanned first paper; it was read with pdftotext.
-
https://docs.kernel.org/locking/mutex-design.html, https://docs.kernel.org/locking/seqlock.html and https://docs.kernel.org/RCU/whatisRCU.html: curl 200. The kernel.org/doc/html/latest form of the same pages also answers 200 and is not used. The mutex design and seqlock pages are now under Rejected, trimmed for length.
-
https://www.akkadia.org/drepper/futex.pdf: curl 200, version 1.6 of the paper.
-
https://pdos.csail.mit.edu/papers/linux:lock.pdf: curl 200 with and without a browser user agent, PDF on the group's own paper host, read with pdftotext to confirm the title, the machine and the four workloads. Now under Rejected, trimmed for length.
-
https://google.github.io/tcmalloc/design.html: curl 200. The same document lives at docs/design.md in the repository, which also answers 200; the rendered form is listed because the project publishes it as its documentation site.
-
https://shop.elsevier.com/books/the-art-of-multiprocessor-programming/herlihy/978-0-12-415950-1: curl 200, the publisher's page for the second edition.
-
https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2023/p2530r3.pdf: curl 200, P2530R3, read to confirm the inventor is first author.
-
https://lwn.net/Articles/262464/: curl 200, free to read. Now under Rejected, trimmed for length.
-
https://www.efficios.com/pub/rcu/urcu-main.pdf: curl 200 with and without a browser user agent, PDF on the first author's company host, read with pdftotext to confirm it is the TPDS paper and to take the machine and method fields above. The host answers non-browser fetch tools with a block page, so curl and pdftotext are the way to read it. The project home https://liburcu.org/ answers 200, and git.liburcu.org redirects to git.lttng.org.
-
https://mirrors.edge.kernel.org/pub/linux/kernel/people/paulmck/perfbook/perfbook.html: curl 200. The page offers the current release in three PDF layouts under a Creative Commons licence.
-
https://publications.csail.mit.edu/lcs/pubs/pdf/MIT-LCS-TM-548.pdf: curl 200, the MIT LCS technical memo dated December 1995, read to confirm the title and authors. The publications page at supertech.mit.edu renders no entries to a fetcher and its bibliography pages carry citations only.
-
https://www.fftw.org/~athena/papers/cilk5.ps.gz: curl 200, served as application/octet-stream. It is gzipped PostScript produced by dvips in March 1998, twelve pages, on the first author's own paper directory, the only author-hosted copy found, and no PDF exists on any author or MIT host. The reason line now says so.
-
https://www.chargueraud.org/research/2018/heartbeat/heartbeat.pdf: curl 200 with and without a browser user agent, PDF on the second author's own host, read with pdftotext to confirm it is the PLDI paper and to take the machine and method fields above. The directory above it answers 403 and the ACM DOI is behind a bot check, so the PDF is the only author address. Now under Rejected, trimmed for length.
-
https://inria.hal.science/hal-00802885: curl 200 without a user agent and the record is returned. With a browser user agent the host answers 200 with an Anubis JavaScript challenge page, which a browser passes. The PDF is at the same address with /document appended. The first author's copy at di.ens.fr redirects (302) to fzn.fr and is not used because the redirect is temporary.
-
https://github.com/uxlfoundation/oneTBB and https://www.openmp.org/specifications/: curl 200. Section 13 lists both with the same URLs for inference threading, with reasons that name the inference-side mechanisms, so both stay here with the general statements.
-
https://dl.acm.org/doi/10.1145/324133.324234, https://dl.acm.org/doi/10.1145/1073970.1073974 and https://dl.acm.org/doi/10.1145/1065010.1065042: curl 403 with and without a browser user agent, fetch 403, and a browser is shown a Cloudflare verification page, which was not passed.
-
https://dominoweb.draco.res.ibm.com/reports/rj5118.pdf: curl 000 with and without a browser user agent, connection never completes.
-
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/memory-model/Documentation/explanation.txt: curl 200 without a user agent and the rendered file is returned, headed Explanation of the Linux-Kernel Memory Consistency Model. With a browser user agent the host answers 200 with an Anubis JavaScript challenge page, which a browser passes and which the repository link checker counts as ok, and the fetch tool gets 403. The kernel tree is the only home of the file, and memory-barriers.txt points to it. Now under Rejected, trimmed for length.
-
https://www.microsoft.com/en-us/research/publication/mimalloc-free-list-sharding-in-action/: curl 403 without a user agent and with the short Mozilla/5.0 one, 200 with a full browser user agent and an Accept header, which is what scripts/check_links.py sends, so the checker reports it ok. Confirmed by fetch as the Microsoft Research page for Mimalloc: Free List Sharding in Action, MSR-TR-2019-18, June 2019. The mimalloc repository README links this page as the design document, so it is the vendor's own address. Now under Rejected.
-
All other URLs: curl 200 with no redirect.
-
scripts/check_links.py on this file: ok 55, moved 0, blocked 5, dead 0. The blocked five are the three ACM DOIs and the Arm manual, all in Rejected, and the HPE certificate case above, which the checker classifies as blocked rather than dead.
§10NUMA and multi-socket
-
https://ieeexplore.ieee.org/document/121510: curl returns 202 (bot challenge) with and without a browser user agent, not content, the same behaviour as the IEEE Xplore pages in sections 2 and 3; DOI 10.1109/2.121510 resolves to this page through doi.org and the Crossref registry confirms the title (The Stanford Dash multiprocessor), venue (Computer, volume 25, issue 3) and year (1992). Was the origin paper entry; now under Rejected for length. The ISCA 1990 protocol paper on the ACM Digital Library sits behind a bot-verification challenge and was not attempted.
-
https://support.arm.com/documentation/102308/latest/: https://developer.arm.com/documentation/102308/latest/ answers 301 to this address, so the final host is used, as for every other Arm document in the list; 200 to bare curl, 403 to a browser-like user agent and to the link checker's agent (blocked, not dead). The page is a JavaScript viewer that fetch reads as an empty shell, and no browser was available in the session that added it, so the document was confirmed from the PDF Arm's documentation service issues for it (102308_0302_07_en, revision r3p2, non-confidential, July 2023), whose feature list names the distributed system level cache and snoop filter across the HN-F home nodes and the CCG devices that carry CML_SMP or CXL 2.0 links; a search of the same host shows the current viewer path at revision r3p7.
-
https://chipsandcheese.com/p/arms-neoverse-v2-in-awss-graviton-4 and https://chipsandcheese.com/p/a-look-into-intel-xeon-6s-memory: curl 200 with and without a browser user agent, no redirect; the /p/ addresses are used. Both articles now have section 14 as their one home and appear here under Rejected only.
-
https://queue.acm.org/doi/10.1145/2508834.2513149: curl 403 with and without a browser user agent (Cloudflare), fetch tool 403; confirmed in a browser session that the page is the article (Queue volume 11 issue 7, August 2013). The older https://queue.acm.org/detail.cfm?id=2513149 form redirects in the browser to this DOI form, so the DOI form is used. Kept as an entry because benchmark 10 supports it.
-
http://halobates.de/numaapi3.pdf: curl 200 with no redirect, and the site does not redirect http to https. The https form is served with only the leaf certificate and no intermediate, so it verifies with macOS curl and browsers (which chase the issuer) but fails in OpenSSL-based clients including the fetch tool and the repository link checker; the http form is used so the checker sees the file. Read from the downloaded PDF: twelve pages, dated August 2004, and the title on the PDF is "An NUMA API for Linux", which is why the line keeps that article. Now under Rejected for length.
-
https://www.intel.com/content/www/us/en/developer/articles/technical/xeon-processor-scalable-family-technical-overview.html: curl 200 with the default agent, 403 when
-A 'Mozilla/5.0'is passed and 403 to the repository link checker's agent (reported as blocked, not dead); confirmed by fetch as the article, updated December 2022. The article's own terms are Caching and Home Agent (CHA) and directory mode, which the reason now uses. -
https://www.intel.com/content/www/us/en/content-details/858491/intel-xeon-6-with-p-cores-configuration-and-tuning-guide-for-hpc-applications.html: curl 200 with the default agent, 403 with
-A 'Mozilla/5.0'and to the link checker's agent (blocked, not dead); confirmed by fetch as document 858491 (revision 1.4, April 2026). The PDF behind it was read from https://cdrdv2-public.intel.com/858491/858491-Xeon6_HPC_Tuning_Guide_v1.4.pdf; the library page is used because it tracks the latest revision. -
https://www.intel.com/content/www/us/en/developer/articles/tool/intelr-memory-latency-checker.html: curl 200 with the default agent, 403 with
-A 'Mozilla/5.0'and to the link checker's agent (blocked, not dead); confirmed by fetch as the tool page (version 3.13 at the time of reading). The tool is x86 only, which the reason now says; on Arm the lmbench entries in section 15, pinned with numactl, are the fallback for the node matrix. -
https://docs.amd.com/v/u/en-US/58011-epyc-9004-tg-bios-and-workload: redirected (301) from https://www.amd.com/content/dam/amd/en/documents/epyc-technical-docs/tuning-guides/58011-epyc-9004-tg-bios-and-workload.pdf. The portal returns 200 and the same single-page shell for any slug, so the page was confirmed in a browser session as document 58011; the PDF was read via the portal's khub API (revision 1.4, May 2025).
-
https://docs.amd.com/v/u/en-US/58002_amd-epyc-9004-tg-hpc: redirected (301) from https://www.amd.com/content/dam/amd/en/documents/epyc-technical-docs/tuning-guides/58002_amd-epyc-9004-tg-hpc.pdf. Note the underscore: the hyphenated slug 58002-epyc-9004-tg-hpc is a 404 inside the portal app although curl reports 200. Confirmed in a browser session as document 58002; the PDF was read via the khub API (revision 1.7, July 2024).
-
https://docs.amd.com/v/u/en-US/NUMA-Topology-for-AMD-EPYC-Naples-Family-Processors: curl 200 (the portal shell, as above) and the slug loads document 56308 in a browser session, so the earlier Rejected line's claim that the URL was dead was wrong; only the old www.amd.com PDF path is gone. Kept under Rejected for size.
-
https://lameter.com/gelato2005.pdf: curl 200 application/pdf with no redirect on the bare https host (23 pages, Gelato conference, May 2005, the author's own site); only the http://www.lameter.com form serves a registrar lander page, which is what the earlier Rejected line mistook for a parked domain. A second copy is live at https://www.kernel.org/pub/linux/kernel/people/christoph/gelato/gelato2005-paper.pdf. Kept under Rejected for scope.
-
https://www.usenix.org/conference/atc15/technical-session/presentation/lepers: curl 200 with no redirect; the official USENIX page with the abstract and the paper PDF at https://www.usenix.org/system/files/conference/atc15/atc15-paper-lepers.pdf (also 200). Kept under Rejected for size.
-
https://docs.kernel.org/admin-guide/sysctl/kernel.html#numa-balancing: curl 200; the anchor
numa-balancingis present in the rendered HTML as a section id. -
https://people.ece.ubc.ca/sasha/papers/asplos284-dashti.pdf: curl 200 with no redirect; the author's own copy, used in preference to the ACM Digital Library page, which returns 403.
-
https://www.kernel.org/doc/Documentation/ABI/stable/sysfs-devices-node: curl 200 with no redirect; plain text, and it also documents the accessY and memory_side_cache attributes that the numaperf entry explains. Now under Rejected for length.
-
https://man7.org/linux/man-pages/man7/numa.7.html: curl 200 with no redirect; proc_pid_numa_maps(5) on the same host only defers to it, so it is the only document of the numa_maps format.
-
All remaining URLs (docs.kernel.org, man7.org, github.com) returned 200 with no redirect.
-
scripts/check_links.py as committed (c9b3a4e) does not run:
opener.open(req, timeout=TIMEOUT, context=ctx)raises TypeError because OpenerDirector.open has nocontextargument, so every link is reported dead before a request is made. The working tree now carries the one-line fix (the SSL context passed to an HTTPSHandler in build_opener) as an uncommitted change made outside this section, and the results above come from running it: 0 dead, 0 moved, and the blocked set is exactly the ACM, UEFI, Intel and Arm pages listed here.
§11OS and I/O
-
https://www.kernel.dk/io_uring.pdf: the document is served only from the www host. https://kernel.dk/io_uring.pdf fails TLS (certificate for another domain) and returns 404 over HTTP, so the www URL is used.
-
https://www.usenix.org/conference/osdi10/flexsc-flexible-system-call-scheduling-exception-less-system-calls: curl 200; WebFetch is refused with 403, confirmed by curl that the page is the paper record, whose PDF link https://www.usenix.org/events/osdi10/tech/full_papers/Soares.pdf answers 302 to https://www.usenix.org/legacy/events/osdi10/tech/full_papers/Soares.pdf.
-
https://atlarge-research.com/pdfs/2022-systor-apis.pdf: curl 200, application/pdf, the author group's copy; confirmed from the PDF text as the SYSTOR 2022 paper, DOI 10.1145/3534056.3534945, whose ACM copy answers 403.
-
https://docs.kernel.org/admin-guide/pm/cpufreq.html and https://docs.kernel.org/admin-guide/pm/cpuidle.html: curl 200; confirmed by fetch as the kernel admin guide pages that define the scaling governors and the boost file, and the idle states, their governors and the per-state disable attribute. The CPU Isolation guide links the cpuidle page from its checklist, and the cpuidle page is under Rejected since the length trim.
-
https://cs.uwaterloo.ca/~mkarsten/papers/sigmetrics2024.html: curl 200; the author's paper page, which links the preprint https://cs.uwaterloo.ca/~mkarsten/papers/sigmetrics2024-preprint.pdf (curl 200, PDF confirmed) and the supplementary page https://cs.uwaterloo.ca/~mkarsten/netstack/ (curl 200) with the F-Stack Memcached port, the run and plot scripts and the experiment data, and whose addendum records that the interrupt mitigation scheme was submitted to the kernel. The DOI https://doi.org/10.1145/3626780 resolves to the ACM copy, 403 to curl.
-
https://docs.kernel.org/arch/arm64/mpam.html: curl 200; confirmed by fetch as the kernel's arm64 MPAM document, which states the cache portion and bandwidth controls and the monitors exposed through resctrl and the platform requirements. Arm's own MPAM supplement DDI0598 is retired by Arm's page for it, and the memory-system specification IHI0099 answers 403 and a single-page shell, so it is under Rejected as unconfirmed.
-
https://www.intel.com/content/www/us/en/content-details/789566/intel-resource-director-technology-intel-rdt-architecture-specification.html: curl 403 with and without a browser user agent; confirmed by fetch as the Intel RDT Architecture Specification record. The direct download https://cdrdv2.intel.com/v1/dl/getContent/789566 returns 301 to the latest PDF, currently revision 1.4 (document 356688-005US).
-
https://docs.amd.com/v/u/en-US/24593_3.45_APM_Vol2_PUB: the AMD portal is a single-page application that returns 200 for any path, so a curl 200 proves nothing; confirmed by fetch as publication 24593 revision 3.45, dated 2026-07-29. Chapter 19 is the Platform Quality of Service extension that replaced the withdrawn 56375 document. The same manual is expected in the ISA section; the same URL may appear in two sections. Under Rejected since the length trim. The URL pins the revision because no unpinned path on the portal could be confirmed; a newer revision will not trip the link checker, so the pin is to be re-checked by hand.
-
https://docs.amd.com/v/u/en-US/69127_1.00_PQOSWP: same single-page portal; confirmed by fetch and by the downloaded PDF as publication 69127 revision 1.00. The PDF is at https://docs.amd.com/api/khub/documents/VuNrmUG_yfhPVgGYcFlkZg/content. Listed under Rejected.
-
https://docs.kernel.org/filesystems/resctrl.html: the older path https://docs.kernel.org/arch/x86/resctrl.html is 404; the document moved to the filesystems tree. The AMD white paper still cites the old x86 path.
-
https://docs.kernel.org/networking/ethtool-netlink.html#coalesce-set: the page documents every ethtool netlink operation; the fragment lands on the COALESCE_SET heading, which exists in the rendered page.
-
https://github.com/tohojo/xdp-paper: the original repository; https://github.com/xdp-project/xdp-paper is a fork of it. The paper PDF is xdp-the-express-data-path.pdf at the repository root.
-
http://halobates.de/lk09-scalability.pdf (Rejected only): the HTTPS endpoint sends only its leaf certificate and no intermediate, so strict verifiers fail on it (Python ssl, and with it scripts/check_links.py, reports CERTIFICATE_VERIFY_FAILED; WebFetch fails too) while browsers and curl succeed with 200; the same host serves the file over plain HTTP with 200 and no redirect, so the HTTP URL is used for every halobates.de link here until the host serves its chain.
-
https://ieeexplore.ieee.org/document/7446102 (Rejected only): curl 202, the IEEE bot gate, and fetch returns nothing; confirmed by search as the record for the HPCA 2016 paper, with no author copy found.
-
https://pdos.csail.mit.edu/papers/linux:osdi10.pdf (Rejected only): curl 200, application/pdf, the author copy of the OSDI 2010 paper whose USENIX record is the Rejected line.
-
https://people.ece.ubc.ca/sasha/papers/eurosys16-final29.pdf: 200; the same file is also served under www.ece.ubc.ca. The people host is the author's page and is used.
-
https://csl.stanford.edu/~christos/publications/2015.heracles.isca.pdf: 200 over HTTPS; search engines list the HTTP form.
-
https://research.google/pubs/pub40737/ and https://research.google.com/pubs/pub40737.html: both 301 to https://research.google/pubs/cpi2-cpu-performance-isolation-for-shared-compute-clusters/, a landing page; the author copy is used instead.
-
https://man7.org/linux/man-pages/man7/io_uring.7.html: 200; the page footer states it is part of the liburing project, which is why the repository rather than the rendered page held the entry until the length trim.
-
https://sourceforge.net/projects/lmbench/ (Rejected only): 403 with a browser user agent and 200 without one, the pattern the benchmarks section records; http://www.bitmover.com/lmbench/ gives no response and is not linked.
§12Tail latency and production systems
-
https://www.barroso.org/publications/TheTailAtScale.pdf: curl 200. The publisher's page https://cacm.acm.org/research/the-tail-at-scale/ answers 403 (Cloudflare) to curl with a browser user agent and to fetch, so the author's own copy is used; the Google Research page https://research.google/pubs/the-tail-at-scale/ is live but only links to the blocked CACM page.
-
https://www.barroso.org/publications/AttackoftheKillerMicroseconds.pdf: curl 200. https://cacm.acm.org/research/attack-of-the-killer-microseconds/ answers 403 (Cloudflare) to curl and fetch; https://research.google/pubs/attack-of-the-killer-microseconds/ is live but has no download link (https://research.google/pubs/pub51993/ redirects there).
-
https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/44271.pdf: curl 200. The publication page https://research.google/pubs/profiling-a-warehouse-scale-computer/ is live (https://research.google/pubs/pub44271/ redirects there) but carries no link to the paper, so the Google-hosted PDF is the form cited under Rejected and in section 15.
-
https://www.youtube.com/watch?v=lJ8ydIuPFeU: curl 200; the page title and channel (Strange Loop Conference, published September 2015) were read from the page source because fetch returns only the footer.
-
https://groups.google.com/g/mechanical-sympathy/c/icNZJejUHfE: curl 200; the message permalink https://groups.google.com/g/mechanical-sympathy/c/icNZJejUHfE/m/BfDekfBEs_sJ also answers 200; the thread URL is used as the canonical form.
-
https://hdrhistogram.org/: curl HEAD answers 405, GET answers 200 with a frameset that embeds http://hdrhistogram.github.io/HdrHistogram/, which is the actual site; listed under Rejected. https://hdrhistogram.github.io/HdrHistogram/plotFiles.html answers 200.
-
https://git.kernel.org/pub/scm/utils/rt-tests/rt-tests.git/ and https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/perf/Documentation/perf-sched.txt: curl HEAD and GET answer 200, but the GET body is an Anubis proof-of-work page for non-browser clients; the README and the man page were confirmed from the kernel.googlesource.com mirrors of the same repositories.
-
https://docs.kernel.org/trace/timerlat-tracer.html and https://docs.kernel.org/virt/kvm/halt-polling.html: curl 200, no redirect; the same pages under www.kernel.org/doc/html/latest/ serve without redirecting, and docs.kernel.org is the form the kernel's own index uses.
-
https://docs.redhat.com/en/documentation/red_hat_enterprise_linux_for_real_time/9/html/optimizing_rhel_9_for_real_time_for_low_latency_operation/index: curl 403 with and without a browser user agent; confirmed by fetch as the document with its chapter list; listed under Rejected.
-
https://wiki.linuxfoundation.org/realtime/documentation/howto/tools/cyclictest/start: curl 403 and fetch 403, both a Cloudflare challenge; could not be confirmed, so listed under Rejected.
-
https://drkp.net/papers/latency-socc14.pdf: curl 200, author's copy; the Google Research page for the paper links only to the ACM DL, which answers 403 to curl.
-
https://www.usenix.org/conference/nsdi-06/open-versus-closed-cautionary-tale: curl 200, the publisher's page for the paper with title, authors and a PDF link that lands on https://www.usenix.org/legacy/events/nsdi06/tech/full_papers/schroeder/schroeder.pdf (curl 200); the earlier entry URL https://www.usenix.org/legacy/event/nsdi06/tech/full_papers/schroeder/schroeder.pdf also answers 200, and the text was read from the downloaded PDF because fetch answers 403 on it. Author copies at https://www.cs.cmu.edu/~harchol/Papers/nsdi06.pdf and https://users.cms.caltech.edu/~adamw/papers/openvsclosed.pdf are live; USENIX is the publisher of record.
-
https://github.com/leverich/mutilate: curl 200; the README credits the Facebook workload paper for its fb_key, fb_value and fb_ia distributions.
-
https://tailbench.csail.mit.edu/: curl 200; the project page is the home of the suite and links the paper at http://people.csail.mit.edu/sanchez/papers/2016.tailbench.iiswc.pdf (the https form answers 200 and was read for the Claims line) and the v0.9 source and input tarballs, so the project page is used over the paper.
-
https://jiangs.utasites.cloud/pubs/papers/atikoglu12-memcached.pdf: curl 200, a co-author's own copy linked from his university home page (the older path https://ranger.uta.edu/~sjiang/pubs/papers/atikoglu12-memcached.pdf serves the same file and its home page refreshes to the new host). The employer's publication page https://research.facebook.com/publications/workload-analysis-of-a-large-scale-key-value-store/ answers 200 to a bare user agent but 400 to a browser-like one, which the link checker counts as dead, and its PDF link redirects to a signed CDN URL that expires; the ACM DL copy https://dl.acm.org/doi/10.1145/2254756.2254766 answers 403.
-
https://csl.stanford.edu/~christos/publications/2014.mutilate.eurosys.pdf and https://csl.stanford.edu/~christos/publications/2015.heracles.isca.pdf: curl 200, copies on the authors' own publication page; https://research.google/pubs/heracles-improving-resource-efficiency-at-scale/ is live but carries no PDF link.
-
https://rigtorp.se/ringbuffer/ and https://rigtorp.se/virtual-memory/: curl 200, no redirect.
-
https://github.com/aeron-io/aeron and https://github.com/aeron-io/simple-binary-encoding: curl 200; the older https://github.com/real-logic/aeron redirects (301) to the aeron-io organisation, so the new organisation is used.
-
https://lmax-exchange.github.io/disruptor/disruptor.html: curl 200; this is the project's maintained copy of the paper, which adds a newer results table to the 2011 text.
-
https://dl.acm.org/doi/10.1145/2408776.2408794, https://dl.acm.org/doi/10.1145/3015146, https://dl.acm.org/doi/10.1145/2749469.2750392, https://dl.acm.org/doi/10.1145/2592798.2592821, https://dl.acm.org/doi/10.1145/2749469.2749475: curl 403; cited only under Rejected.
§13Inference on CPU
-
http://justine.lol/matmul/: the https certificate expired on 2026-06-09 (notAfter Jun 9 2026, curl exit 60 when checked on 2026-09-15), so the http URL is used, which returns 200 with the page. Switch back to https once the certificate is renewed.
-
https://github.com/OpenMathLib/OpenBLAS: redirected (301) from https://github.com/xianyi/OpenBLAS.
-
https://github.com/ggml-org/llama.cpp and https://github.com/ggml-org/ggml: redirected (301) from the ggerganov account, pull request, discussion and blob URLs follow the same redirect.
-
https://github.com/uxlfoundation/oneDNN and https://uxlfoundation.github.io/oneDNN/: the project moved from oneapi-src to the UXL Foundation, the old GitHub URL redirects (301) and the old docs site still serves copies. The README states the AArch64 CPU engine can be built with Arm Compute Library integration, minimum version 53.1.0, which is the basis of the oneDNN reason and of the Arm Compute Library line under Rejected.
-
https://www.intel.com/content/www/us/en/developer/articles/technical/intel-sdm.html, https://www.intel.com/content/www/us/en/content-details/671488/intel-64-and-ia-32-architectures-optimization-reference-manual-volume-1.html and https://www.intel.com/content/www/us/en/content-details/671279/bfloat16-hardware-numerics-definition.html: 200 to plain curl, 403 to a browser user agent and to the repository checker, which reports them as blocked; the curl copies confirm the manual index (every volume at version 092, with the volume 2 instruction pages that define VPDPBUSD), the document-library record for Optimization Reference Manual Volume 1 (version 050), and the document-library record for bfloat16 Hardware Numerics Definition, document 671279, dated 2018-11-14. The revision-numbered PDFs the first two replace, 253665-092 and 248966-050, are the copies read for the AMX, int8 and VNNI chapters and are held under Rejected because their addresses change at every release. The Optimization Reference Manual chapters and the bfloat16 definition are now under Rejected, trimmed for length, and only the manual index stays as an entry.
-
https://support.arm.com/documentation/ddi0487/latest/ and https://support.arm.com/documentation/109246/latest/: the developer.arm.com forms answer 301 to these addresses to plain curl and 403 to a browser user agent; on the support host plain curl gets 200, curl with a browser user agent gets 403, and fetch sees a script-only shell. Confirmed in a browser on 2026-09-15 as DDI 0487 version M.c, and SME Programmer's Guide version 1.1 with the SME2 matmul_int8, gemv and lut_gemv examples. Section 16 lists the guide as a watchlist line with a promotion condition; it is core here because the corpus names SME and SME2 under this section, and its reason now carries the same warning the watchlist line does, that the unit is so far only in client parts, so the editor should drop the watchlist line.
-
https://cacm.acm.org/research/roofline-an-insightful-visual-performance-model-for-multicore-architectures/: curl 403 with and without a browser user agent, WebFetch 403, and the repository checker reports it as blocked; confirmed in a browser on 2026-09-15 as the full article with a PDF download and no sign-in, the same copy sections 1 and 6 link. Now under Rejected, trimmed for length, with sections 1 and 6 as its homes and the preamble carrying the batch-one decode case. The measured roof a decode rate is compared with is the STREAM entry in section 15.
-
https://ieeexplore.ieee.org/document/6506838: 202 with a WAF challenge to curl, the copy section 6 already verified; now under Rejected, trimmed for length, with section 6 as its home.
-
https://gitlab.arm.com/kleidi/kleidiai: returns 202 with an AWS WAF challenge to curl. Confirmed in a browser as the KleidiAI project, the GitHub copy states it is a read-only mirror. Now under Rejected, trimmed for length.
-
https://www.cs.utexas.edu/~flame/pubs/GotoTOMS_revision.pdf and https://www.cs.utexas.edu/~flame/pubs/blis1_toms_rev3.pdf: 200 and confirmed by their first pages, the directory index above them returns 403.
-
https://docs.openvino.ai/2026/...: the version-pinned path is used because https://docs.openvino.ai/latest/... redirects to the archive index rather than to the page.
-
https://github.com/ggml-org/llama.cpp/pull/7707: closed unmerged, it holds the design report and measurements, the code merged as https://github.com/ggml-org/llama.cpp/pull/8998 whose body only points back to it.
-
https://github.com/google/ruy: curl 200; the README states it is the default matrix multiplication library of TensorFlow Lite on Arm and that it targets the small and rectangular shapes of inference rather than large sizes, which is why it is under Rejected.
-
https://github.com/ARM-software/ComputeLibrary: curl 200; the README names Cortex-A and Neoverse as the CPU targets and lists BFLOAT16 and INT8 among the data types. Now under Rejected, trimmed for length.
-
https://github.com/amd/ZenDNN and https://github.com/ggml-org/llama.cpp/blob/master/docs/backend/ZenDNN.md: curl 200, both read for the Rejected line.
-
https://arxiv.org/abs/2309.05516 and https://github.com/intel/auto-round: curl 200, both read for the Rejected line.
-
https://docs.kernel.org/arch/x86/xstate.html: curl 200; the page is the kernel's own document, with the ARCH_REQ_XCOMP_PERM request, the trap on first use and the AMX TILE_DATA example.
-
https://docs.pytorch.org/tutorials/recipes/recipes/tuning_guide.html: curl 200; the CPU specific optimizations section carries the numactl, OMP_NUM_THREADS, GOMP_CPU_AFFINITY and KMP_AFFINITY recipe. Now under Rejected, trimmed for length, beside the Intel extension guide.
-
https://uxlfoundation.github.io/oneDNN/dev_guide_graph_sdpa.html: curl 200, oneDNN v3.14 documentation; the page states f32 intermediates under bf16 or f16 inputs, that the reference implementation stores the query-key product in O(S²) memory, and the 4D shapes the optimised CPU implementation for inference accepts.
-
https://github.com/ggml-org/ggml/blob/master/examples/gpt-j/README.md: curl 200; the raw file was read for the MPS passage and the per-token figure.
-
https://johannesgaessler.github.io/llamacpp_performance: curl 200; the page is marked WIP and states the two machines, the test setup and the memory-frequency sweep.
-
https://arxiv.org/abs/2502.12444 and https://github.com/IntelLabs/Hardware-Aware-Automated-Machine-Learning/tree/main/SparAMX: curl 200; the HTML version of the paper was read for the Xeon Gold 6430L and the VTune table, and the abstract links the code path.
-
https://github.com/mlcommons/inference_results_v6.0: curl 200; closed/Intel/systems/1-node-2S-GNR_128C.json and the Offline summary logs under closed/Intel and closed/NVIDIA were read through raw.githubusercontent.com for the Claims line.
-
https://mlcommons.org/benchmarks/inference-datacenter/: 403 to curl, fetch confirmed the page, the results repository is linked instead.
-
https://github.com/mlcommons/inference_policies/blob/master/inference_rules.adoc: listed in section 15 for the scenario and latency-bound clauses; now under Rejected as a section 15 duplicate, trimmed for length, and the FAQ clause that image data may not be preloaded closer to the accelerator is read from the same document there. The calibration-only quantization clause the earlier reason named is covered by the section 15 entry.
-
https://www.openmp.org/specifications/: curl 200, fetch 403; now under Rejected as a section 9 duplicate.
-
https://github.com/uxlfoundation/oneTBB: curl 200; now under Rejected as a section 9 duplicate, trimmed for length.
-
https://www.cs.virginia.edu/stream/ref.html: curl 200, already verified by section 15; held under Rejected as a cross-section duplicate with the pointer in the Roofline link note above.
§14Hardware generations
-
https://www.intel.com/content/www/us/en/content-details/671488/intel-64-and-ia-32-architectures-optimization-reference-manual-volume-1.html and https://www.intel.com/content/www/us/en/content-details/858491/intel-xeon-6-with-p-cores-configuration-and-tuning-guide-for-hpc-applications.html (both Rejected only): curl 200 with its default agent and 403 with a browser-like agent, so the link checker reports both as blocked; the download endpoint behind each answers 200 and resolves to the current PDF on the public bucket (the manual at revision 050, the guide at revision 1.4 dated April 2026), and both PDFs were read for the chapters the reasons cite, the guide's Latency Optimized Mode section for the mode, the BIOS menu, the pcm script and the sysfs check.
-
https://www.intel.com/content/www/us/en/developer/articles/technical/fourth-generation-xeon-scalable-family-overview.html: curl 200 with no redirect and 403 with a browser-like agent, the same agent filter as the rest of the host; the fetch tool confirms the title and that the article carries no performance percentage.
-
https://builders.intel.com/solutionslibrary/intel-xeon-6-processors-performance-and-power-profiles (Rejected only): curl 200; the page offers the PDF at a timestamped path under builders.intel.com/docs/networkbuilders/; the PDF (document 826934, revision 1.06, April 2025) was read, and its own text says the results may be estimated or simulated.
-
https://real.mtak.hu/215594/: curl 200 with and without a browser agent, no redirect; the authors' deposit in the Hungarian Academy's repository, which serves the PDF (Intel_CPU_energy_tiling_IEEE.pdf, 237 kB) with no login and was read for Appendix A, the run method and the results text; the Crossref record for DOI 10.1109/SCW63240.2024.00182 gives the same title in the SC24 workshops proceedings, and the publisher's page answers a bot challenge to curl like the other IEEE pages, so the repository copy is linked as the authors' own page.
-
https://ieeexplore.ieee.org/document/10454434 and https://ieeexplore.ieee.org/document/10466769: curl answers 202 (bot challenge) with and without a browser agent, as sibling sections found for the host; DOIs 10.1109/ISSCC49657.2024.10454434 and 10.1109/MM.2024.3375070 resolve to these pages through the Crossref registry, which confirms title, venue and year, and the abstracts were read through Semantic Scholar. No author copy of either paper was found. The four IEEE pages under Rejected behave the same way.
-
https://docs.amd.com/v/u/en-US/58002_amd-epyc-9004-tg-hpc, https://docs.amd.com/v/u/en-US/58455_1.00 and https://docs.amd.com/v/u/en-US/58462_amd-epyc-9005-tg-architecture-overview: curl 200 for an application shell; confirmed in a browser as the HPC tuning guide (revision 1.7, July 2024), the Zen 5 guide (revision 1.00, released 2024-08-15) and the 9005 overview (revision 1.0, released 2025-04-23), and the portal's search API lists all three. The amd.com content/dam PDF paths answer 301 to these viewer addresses with a browser agent and fail with an HTTP/2 stream error without one, so the viewer addresses are used. The same shell answers 200 for 57647 and 58015, which the browser then renders as 401 and 404, so a 200 from curl on this host proves nothing.
-
https://support.arm.com/documentation/102375/latest/, https://support.arm.com/documentation/109898/latest/, https://support.arm.com/documentation/109914/latest/ and https://support.arm.com/documentation/101427/latest/ (102375 and 101427 Rejected only): the developer.arm.com forms answer 301 to these addresses, so the final host is used; plain curl gets 200, curl with a browser agent gets 403, and the page is a script shell; confirmed in a browser as the Neoverse V2 Core Technical Reference Manual, r0p2, the Neoverse V2 Core Software Optimization Guide, version 3.0, the Neoverse N2 Software Optimization Guide, version 5.0, and the Neoverse V1 Core Technical Reference Manual, r1p2, the guides a PDF download with no sign-in and the manuals rendered as pages; Arm's documentation service at documentation-service.arm.com/documentation/ID/latest returns each id's title, version and non-confidential status as JSON, which is the same confirmation without a browser. The Graviton repository's generation table links 102375 as the Graviton 4 core document and 101427 as the Graviton 3 core document.
-
https://support.arm.com/documentation/109678/latest/ and https://support.arm.com/documentation/109637/latest/ (Rejected only): the same 301, 200 and 403 behaviour; the documentation service names them the Neoverse V3 Core Software Optimization Guide, version 3.0, and the Neoverse N3 Software Optimization Guide, version 3.0, and both are linked from section 16.
-
https://support.arm.com/documentation/109897/latest/ (Rejected only, not linked): plain curl 200 for the same script shell, so the link checker would pass it, but the documentation service answers 404 with the body "Document '109897' does not exist", a browser is redirected to the site's not-found page while 109898 and 101427 render, and the older PJDOC address the Graviton repository links for the V1 guide gets the same 404 from the service; the search index's title for the id is the only trace of the guide.
-
https://github.com/aws/aws-graviton-getting-started: curl 200; the README's generation table was read from the raw file on the main branch and carries a Graviton5 column on Neoverse-V3 with CMN-S3 beside the Graviton2, Graviton3 and Graviton4 columns.
-
https://amperecomputing.com/assets/Altra_Rev_A1_DS_v1_50_20240130_3375c3dec5_1c5d4604fa.pdf: curl 200 with and without a browser agent, a PDF by its header bytes, document issue 1.50 dated 2024-01-30, read for the core, cache, mesh and memory statements; the vendor's document library holds a newer revision behind a login.
-
https://github.com/llvm/llvm-project/blob/main/llvm/lib/Target/AArch64/AArch64SchedAmpere1B.td (Rejected only): curl 200; the file header names no author, and the commit history on the main branch shows it was written and adjusted from a compiler contractor's address with one later edit from Arm, so no vendor provenance is claimed for it.
-
https://learn.microsoft.com/en-us/azure/virtual-machines/sizes/cobalt-overview (Rejected only): curl 200; the locale-free form answers 301 to this address; page dated 2024-11-07 and updated 2025-07-03, and it covers Cobalt 100 only.
-
https://docs.cloud.google.com/compute/docs/general-purpose-machines (Rejected only): the cloud.google.com form answers 301 to this host, so the final address is used; curl 200; the C4A and N4A sentences naming Neoverse V2 and Neoverse N3 were confirmed by fetch.
-
https://docs.nvidia.com/dccpu/grace-perf-tuning-guide/index.html: curl 200 with and without a browser agent, no redirect; the index page was read for the coherency fabric, LPDDR5X and MPAM statements and for the performance-per-watt multiple recorded under Claims, and the fetch tool confirms the title; the NUMA and SMT facts in the system chapter appear only inside a sample lscpu listing, so they are not claimed as design statements.
-
l14.16 chipsandcheese.comchipsandcheese.comchipsandcheese.comchipsandcheese.comchipsandcheese.comchipsandcheese.comchipsandcheese.comchipsandcheese.comchipsandcheese.com
https://chipsandcheese.com/p/a-peek-at-sapphire-rapids, https://chipsandcheese.com/p/a-look-into-intel-xeon-6s-memory, https://chipsandcheese.com/p/testing-amds-bergamo-zen-4c-spam, https://chipsandcheese.com/p/amds-turin-5th-gen-epyc-launched, https://chipsandcheese.com/p/arms-neoverse-v2-in-awss-graviton-4, https://chipsandcheese.com/p/arms-neoverse-n2-cortex-a710-for-servers, https://chipsandcheese.com/p/ampereone-at-hot-chips-2024-maximizing-density and, Rejected only, https://chipsandcheese.com/p/meteor-lakes-e-cores-crestmont-makes-incremental-progress and https://chipsandcheese.com/p/zen-5-variants-and-more-clock-for-clock: curl 200 with and without a browser agent; the dated legacy paths of the form chipsandcheese.com/2023/03/12/... answer 404 on the main host and survive only on old.chipsandcheese.com, so the /p/ addresses are used. The Bergamo article's Zen 4 comparison part is a Ryzen 9 7950X3D capped to Bergamo's clock, the N2 article's N1 comparison is Ampere Altra and Graviton 2, and the AmpereOne article names Altra once in prose and compares against no Altra measurement, all confirmed by fetch.
-
https://www.agner.org/optimize/microarchitecture.pdf (Rejected only): curl 200, application/pdf, front matter dated 2026-05-23, with chapters for Alder Lake (Golden Cove and Gracemont), Zen 4 and Zen 5 and none for a server-only core.
-
https://uops.info/ (Rejected only): curl 200 with and without a browser agent; the table page lists Emerald Rapids, Meteor Lake, Zen 4 and Zen 5 among the measured microarchitectures.
-
https://arxiv.org/abs/2409.08108 and https://arxiv.org/abs/2506.09505: curl 200; abs pages used rather than pdf; both PDFs were read for the testbed sections cited under Claims.
-
l14.20 intel.comdownload.intel.comtechcommunity.microsoft.comamperecomputing.comresources.nvidia.com
Rejected only: https://www.intel.com/content/www/us/en/content-details/787386/... and .../787431/... answer 200, their download links under cdrdv2.intel.com/v1/dl/getContent answer 404 through Intel's not-found redirector, and the cdrdv2-public bucket paths answer 403 with an AccessDenied body, so neither deck is linked; https://download.intel.com/newsroom/2023/data-center-hpc/Hot_Chips_23_Granite_Rapids_Sierra_Forest_Xeon_Press_Briefing.pdf answers 200 and was read; https://techcommunity.microsoft.com/blog/azureinfrastructureblog/announcing-cobalt-200-azure%E2%80%99s-next-cloud-native-cpu/4469807 answers 403 to plain curl and 400 with a browser agent, and is not linked; https://amperecomputing.com/customer-connect/products/AmpereOne-device-documentation and the Altra equivalent answer 200 but every file behind them is a login download; https://resources.nvidia.com/en-us-grace-cpu/nvidia-grace-cpu-superchip answers 200 after appending a tracking query.
§15Benchmarks
-
https://www.spec.org/cpu2017/Docs/runrules.html (Rejected) and https://www.spec.org/cpu2026/Docs/runrules.html: curl 200 with no redirect for both; the 2026 page is the run and reporting rules for the 2026 suite, and its text names the 2018 Ampere eMag 8180 as the reference machine (rule 4.10), defines benchmark specials (1.4.2), the threading models a speed run may select (2.3.5.1) and cloud bare-metal systems (4.4.5).
-
https://arxiv.org/abs/2605.01575 and https://arxiv.org/abs/2605.03713: curl 200; abs pages used rather than pdf. The first is the SPEC CPU committee's own paper, thirty-four authors, presented at ISCA 2026, whose abstract defines Rolling-Round-Robin Rate; the second is the characterisation of SPEC CPU 2026 on nine named Intel, AMD, Ampere and Nvidia systems, and its HTML rendering was read to confirm the platform table and the gcc -O3 build.
-
https://research.spec.org/icpe_proceedings/2018/companion/p41.pdf (Rejected): curl 200 with no redirect; the ACM DOI 10.1145/3185768.3185771 answers 403 to curl; the SPEC Research Group archive is the co-publisher's copy of the ICPE companion.
-
https://lca.ece.utexas.edu/pubs/HPCA_SPEC17_ShuangSong.pdf (Rejected): curl 200, the camera-ready HPCA 2018 paper; a second file on the same site is a confidential submission draft and is not linked; the IEEE record for the paper answers 202 with an interstitial.
-
https://github.com/mlcommons/inference_policies/blob/master/inference_rules.adoc: curl 200; the master branch is the live rules document and the repository has no release tag for it.
-
https://arxiv.org/abs/1911.02549 (Rejected): curl 200; cut under rule 5 because the rules document carries the same ground for a CPU reader.
-
https://aisystemcodesign.github.io/papers/DCPerf-ISCA25.pdf: curl 200; the PDF was downloaded and opened, title, authors and the ISCA 2025 venue confirmed; the page is the Meta AI and Systems Co-Design team's publication list. The ACM DOI 10.1145/3695053.3731411 answers 403 to curl and is confirmed through the Crossref record.
-
https://github.com/facebookresearch/DCPerf (Rejected): curl 200; the README states x86_64 and aarch64 as supported architectures and names TaoBench, FeedSim, DjangoBench, Mediawiki, SparkBench and WDLBench.
-
https://github.com/google/fleetbench (Rejected): curl 200; the README states that the input distributions are derived from data collected in production and that the multi-core mode arrived with release v2.0.0. The design paper it cites, DOI 10.1109/ISPASS61541.2024.00046, resolves to https://ieeexplore.ieee.org/document/10590038/, which answers 202 to curl with and without a browser user agent; the Crossref record confirms the title and venue, and no author copy was found on research.google, so the record sits in Rejected.
-
https://github.com/RRZE-HPC/likwid (Rejected): curl 200; the README lists Intel through Emerald Rapids, AMD through Zen 4, ARMv8 including Neoverse N1 on Graviton 2, Neoverse V1, A64FX and Apple M1 under Linux, and the wiki page https://github.com/RRZE-HPC/likwid/wiki/Likwid-Bench (curl 200) documents the stream kernels and the placement option that puts data in a chosen NUMA domain. Section 6 lists the same repository under the same title for the roofs and the counter groups, and is now its one home; the Rejected line names the pinned kernels and per-node placement that were the reason for a second listing.
-
https://www.geekbench.com/doc/geekbench6-benchmark-internals.pdf (Rejected): curl 200 with and without a browser user agent; the fetch tool gets 403.
-
https://github.com/eembc/coremark (Rejected): curl 200; the README on the main branch carries the run rules, the minimum run time and the reporting line.
-
https://www.cs.virginia.edu/stream/ref.html: curl 200; the home page https://www.cs.virginia.edu/stream/ is also 200 and is the results portal.
-
https://www.cs.virginia.edu/~mccalpin/papers/balance/: curl 200; the FAQ cites the TCCA newsletter paper and says it has no link to the published version; the historical PostScript at tab.computer.org answers 404, so the author's own copy is used.
-
https://www.usenix.org/legacy/publications/library/proceedings/sd96/mcvoy.html: curl 200 with and without a browser user agent; the modern conference URL for the 1996 programme answers 404; the legacy abstract page links the PDF, ASCII and PostScript, and the PDF was opened to confirm the paper.
-
https://github.com/travisdowns/uarch-bench: curl 200; sections 2 and 3 record it under Rejected and defer to this section, so the README has one home for it, and the reason here names the memory-level parallelism and load-latency tests and the DVFS hold-off.
-
https://github.com/andreas-abel/nanoBench (Rejected): curl 200; sections 3 and 5 record it under Rejected and defer to this section, where its paper https://arxiv.org/abs/1911.03282 (curl 200, abs page) is listed and names the repository, so the README has one home for the harness through the paper.
-
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/tools/perf/Documentation/perf-bench.txt (Rejected): curl 200 without a browser user agent; the host puts a challenge page in front of browser agents, blocked rather than dead, as with the perf documentation tree section 5 links.
-
https://www.davidhbailey.com/dhbpapers/twelve-ways.pdf (Rejected): curl 200; the author's own papers directory, the PDF states the original publication and date.
-
https://dl.acm.org/doi/10.1145/5666.5673: curl 403 with and without a browser user agent and 403 to the fetch tool, as with every other ACM page in the README; identity confirmed through the Crossref record for the DOI (Communications of the ACM, volume 29, number 3, March 1986, Fleming and Wallace). No author copy was found and the university course copies are third-party mirrors, so the publisher record is used as sections 2 and 6 use it for the critical-path and Amdahl papers. The link checker reports the host as blocked, not dead.
-
https://gernot-heiser.org/benchmarking-crimes.html: curl 200 with no redirect.
-
https://htor.inf.ethz.ch/publications/img/hoefler-scientific-benchmarking.pdf: curl 200; the ACM DOI 10.1145/2807591.2807644 answers 403 to curl; the PDF was opened to confirm the twelve rules.
-
https://kar.kent.ac.uk/33611/: curl 200; the record links the PDF and notes that it corrects the ISMM 2013 printed version; the ACM DOI 10.1145/2464157.2464160 answers 403 to curl.
-
https://arxiv.org/abs/1801.02381 (Rejected): curl 200; abs page.
-
https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/44271.pdf: curl 200; the publication page https://research.google/pubs/profiling-a-warehouse-scale-computer/ is live (pub44271 redirects there with 301) but carries no PDF link; section 12 now records the same address under its Rejected list and defers to this section, so the README has one home for the paper; https://storage.googleapis.com/gweb-research2023-media/pubtools/pdf/44271.pdf also serves it.
-
https://mlcommons.org/benchmarks/inference-datacenter/: curl 403 with a browser user agent and 200 without one, so the host filters on the agent string rather than the client; not listed, it is a leaderboard.
-
https://infoscience.epfl.ch/record/173764: 301 to https://infoscience.epfl.ch/entities/publication/674caf70-abd5-40a6-b096-713741ee689c, which answers 200 to curl and 429 to the fetch tool; the page title confirms the paper; the final address is used in Rejected.
-
https://sourceforge.net/projects/lmbench/ (Rejected): curl 200 with a full page body when no browser user agent is sent and 403 with a two-byte body when one is, confirmed by fetch as the lmbench project page, maintainer staelin, latest release lmbench-3.0-a9, a download page rather than a browsable repository. http://www.bitmover.com/lmbench/ gives no response (curl exit without an HTTP code) and is not linked.
§16Watchlist
-
https://www.intel.com/content/www/us/en/content-details/784266/intel-advanced-performance-extensions-intel-apx-architecture-specification.html, https://www.intel.com/content/www/us/en/content-details/671368/intel-architecture-instruction-set-extensions-programming-reference.html, https://www.intel.com/content/www/us/en/content-details/828965/intel-advanced-vector-extensions-10-2-intel-avx10-2-architecture-specification.html and https://www.intel.com/content/www/us/en/content-details/919018/intel-xeon-6-processors-with-mrdimm-solution-brief.html (Rejected only): curl 200 with its default agent, 403 with a browser-like agent, so the repository checker reports them as blocked; confirmed by fetch as the Intel document library landing pages (the APX page shows "Intel Advanced Performance Extensions (Intel APX) Architecture Specification", version 9, and names no processor), and the PDFs behind them were downloaded through Intel's own cdrdv2 endpoint and read: APX specification revision 9.0 (August 2026), ISE programming reference 319433-062 (June 2026, whose Table 1-2 lists APX, AVX10.2, AMX-FP8, AMX-TF32, AMX-AVX512 and AMX-MOVRS against Diamond Rapids, lists APX and AVX10.2 against Nova Lake as well, and whose CPUID table still calls both future processors), AVX10.2 specification revision 7.0 (April 2026), solution brief dated May 2026.
-
https://www.intel.com/content/www/us/en/products/details/processors/xeon/6-plus-series.html: 200 to plain curl and 403 to a browser-like agent, the same pattern as the other Intel pages; confirmed by fetch as "Intel Xeon 6+ processors", listing the 6960E+, 6970E+, 6980E+ and 6990E+ (144 to 288 E-cores on Intel 18A, launch marked Q2'26) with five multiples footnoted to the vendor claims site.
-
https://www.intel.com/content/www/us/en/products/details/processors/xeon/6-e-core-series.html and https://www.intel.com/content/www/us/en/products/docs/xeon-6-product-brief.html (both Rejected only): the same 200 to plain curl and 403 to a browser-like agent; confirmed by fetch as "Intel Xeon 6 with E-Core Processors" (seven parts, up to the 6780E) and the product brief.
-
https://www.intel.com/content/www/us/en/newsroom/news/client-computing/intel-outlines-architectures-for-agentic-ai-at-hot-chips-2026.html and https://www.intel.com/content/www/us/en/newsroom/news/new-ultrafast-memory-boosts-intel-xeon-chips.html: the newsroom.intel.com forms answer 301 through Intel's not-found redirector and land on these www.intel.com addresses, which answer 200 to plain curl and 403 to a browser-like agent, so the final URLs are used; confirmed by fetch (Hot Chips post dated August 2026, saying "enhanced" AMX and "new" APX with no FP8 or ship date; memory post dated November 2024).
-
https://newsroom.amd.com/news/aai-2026-6th-gen-epyc/, https://newsroom.amd.com/news/amd-announces-production-ramp-of-next-generation-a/ and https://newsroom.amd.com/news/aai-2026-full-stack-compute-agentic-ai/: curl 200 with and without a browser agent; the www.amd.com press-release forms answer 301 to the newsroom host (plain curl fails against www.amd.com with an HTTP/2 stream error, a browser agent gets the redirect), so the final URLs are used; confirmed by fetch and in a browser (launch release dated 23 July 2026, titled "AAI 2026: 6th Gen AMD EPYC Server CPUs Power the Agentic Data Center" and trimmed in the entry to the family name, naming the 9006 SP7, SP8, 9006X SP7 and 9006 LP series with no ship dates and footnotes that call the agent figures estimates; production ramp dated May 2026; AAI post dated July 2026).
-
https://www.amd.com/en/products/processors/server/epyc/9006-series.html and https://www.amd.com/en/legal/claims/epyc.html: plain curl fails with an HTTP/2 stream error, curl with a browser agent gets 200, fetch times out; the product page was confirmed in a browser (rack multiples under a "Performance Projections Methodology" link, model tables marked subject to change) and the claims page renders only through script (Rejected only).
-
https://www.nvidia.com/en-us/data-center/vera-cpu/: curl 200 with and without a browser agent; confirmed by fetch as "Next Gen Data Center CPU | NVIDIA Vera CPU", naming 88 Olympus cores, 176 threads through partitioned core resources, LPDDR5X on SOCAMM at up to 1.2 TB/s, first systems delivered, three multiples marked "subject to change" and no link to an architecture or ISA document.
-
https://aws.amazon.com/about-aws/whats-new/2026/06/ec2-m9g-m9gd-instances-graviton5-processors-available/ (Rejected only): curl 200 with and without a browser agent; confirmed by fetch as "Amazon EC2 M9g and M9gd general purpose instances are now available", posted 10 June 2026, general availability in four regions, four percentages against M8g and no fields. https://aws.amazon.com/ec2/graviton/ (Rejected only): curl 200.
-
https://docs.cloud.google.com/compute/docs/general-purpose-machines#n4a_series (Rejected only): the cloud.google.com form answers 301 to docs.cloud.google.com, so the final host is used; curl 200; the page carries an "N4A machine series" heading with that anchor and the sentence naming Google's Axion processor "built on the Arm Neoverse N3 compute core", with no preview marker and no performance number. https://cloud.google.com/products/axion was not linked, since the documentation page is the one that names the core.
-
https://support.arm.com/documentation/109246/latest/, https://support.arm.com/documentation/109678/latest/ and https://support.arm.com/documentation/109637/latest/ (the first Rejected only, the guides in the body since section 14 ceded them): the developer.arm.com forms answer 301 to these addresses, so the final URLs are used; plain curl gets 200, curl with a browser agent gets 403, and fetch sees only a script shell; confirmed in a browser as "SME Programmer's Guide", version 1.1 (issue 0101-01, October 2025), "Arm Neoverse V3 Core Software Optimization Guide", version 3.0, and "Arm Neoverse N3 Core Software Optimization Guide", version 3.0, each a PDF download with no sign-in. https://support.arm.com/documentation/ddi0616/latest/ (SME supplement) was confirmed in a browser to carry the notice "This document is now RETIRED".
-
https://docs.riscv.org/reference/isa/unpriv/v-st-ext.html: curl 200; the page forwards in the browser to the dated build https://docs.riscv.org/reference/isa/v20260120/unpriv/v-st-ext.html, confirmed by fetch as the chapter "V Standard Extension for Vector Operations, Version 1.0" in the RISC-V ratified specifications library; the unversioned address is linked, under the title section 7 uses for it, since that section ceded the URL to the watchlist. There is no /latest/ path (404).
-
https://computeexpresslink.org/cxl-specification/: curl 200; the specification download (CXL 4.0, February 2026) sits behind a name-and-email form, so the page rather than a PDF is linked.
-
https://arxiv.org/abs/2303.15375: curl 200; confirmed by fetch as "Demystifying CXL Memory with Genuine CXL-Ready Systems and Devices", MICRO 2023, and the PDF was downloaded and read for Table 1 (two Xeon 6430 at 2.1 GHz, Hyper-Threading disabled, Ubuntu 22.04.2 with kernel 6.2, three CXL devices, ten thousand repeats with the median taken), which states no compiler or flags.
-
https://www.jedec.org/standards-documents/docs/jesd82-552 and https://www.jedec.org/news/pressreleases/jedec%C2%AE-advances-ddr5-mrdimm-ecosystem-new-memory-interface-logic-and-expanded: curl 403 with and without a browser agent (also with the checker's own agent and headers) and fetch 403, while the repository checker's urllib requests get 200 and report both live; confirmed in a browser as "DDR5MDB02 Multiplexed Rank Data Buffer", JESD82-552, published February 2026, free download after registration, with no performance number on the page, and as the press release dated 30 April 2026 (Rejected only) stating JESD82-552 published, JESD82-542 expected soon and the MRDIMM Gen2 module standard nearing completion.
-
https://docs.kernel.org/scheduler/sched-ext.html and https://docs.kernel.org/networking/iou-zcrx.html: curl 200; both render from the 7.3.0-rc3 tree at verification time and contain no performance numbers.
-
https://lore.kernel.org/io-uring/20250215000947.789731-1-dw@davidwei.uk/: curl HEAD 200 with and without a browser agent, so the repository checker reports it live; a GET from a non-browser agent, including fetch, is answered by lore's Anubis challenge page, so the content was confirmed from the raw message of the same id on the gnuweeb lore mirror (https://lore.gnuweeb.org/io-uring/20250215000947.789731-1-dw@davidwei.uk/raw), which carries the subject "[PATCH v14 00/11] io_uring zero copy rx", the test setup and the two throughput tables quoted under Claims; the title in the entry drops the bracketed tag to the end so the link renders.
-
https://www.nature.com/articles/s41586-023-06004-9: curl answers 303 to an idp.nature.com cookie check and back to the article with a cookies_not_supported query, so the clean article address is kept and the checker reports it live; confirmed by fetch as the open-access article "Faster sorting algorithms discovered using deep reinforcement learning", which states the sort3, sort4 and sort5 routines were merged into LLVM libc++, names ARMv8, Intel Skylake and AMD Zen 2 as the measured architectures and no compiler, flags, core count or frequency. https://reviews.llvm.org/D118029 (Rejected only): curl 200.
-
https://github.com/microsoft/BitNet, https://github.com/microsoft/T-MAC, https://github.com/sched-ext/scx, https://github.com/parallelcodefoundry/ParEval and https://github.com/riscvarchive/riscv-v-spec: curl 200; https://github.com/riscv/riscv-v-spec answers 301 to the archive organisation. The BitNet README attributes its speedup range to arXiv 2410.16144 and also links 2502.11880.
-
https://security.apple.com/blog/private-cloud-compute/ (Rejected only): the brief's HEAD form of curl gets 403 with and without a browser agent, a GET gets 200, so the repository checker reports it live; confirmed by fetch as "Private Cloud Compute: A new frontier for AI privacy in the cloud", dated June 2024, with the sentence describing custom-built server hardware on Apple silicon and no part name or measurement.
-
https://www.arm.com/technologies/sme2, https://www.arm.com/products/silicon-ip-cpu/neoverse/neoverse-v3, https://www.arm.com/products/cloud-datacenter/arm-agi-cpu, https://www.arm.com/products/cloud-datacenter/neoverse-compute-subsystems/css-n4 and https://newsroom.arm.com/blog/microsoft-azure-cobalt-200-arm-neoverse-css-v3: curl 200 (Rejected only). The CSS N4 page was re-read by fetch on the verification date: it names the Neoverse N4 core, Armv9.3 and eight to 128 cores per die, and links product support but no technical reference manual or software optimisation guide.
-
https://tenstorrent.com/en/newsroom/tenstorrent-announces-availability-of-tt-ascalon: curl 200 (Rejected only). https://www.sifive.com/cores/performance-p870: 307 to https://www.sifive.com/cores/performance-p800 (Rejected only). https://edc.intel.com/content/www/us/en/products/performance/benchmarks/intel-xeon-6/: curl 200 (Rejected only).
-
https://riscv.org/blog/ventana-introduces-veyron-v2-worlds-highest-performance-data-center-class-risc-v-processor-and-platform/ (Rejected only): curl 200 with and without a browser agent; confirmed by fetch as RISC-V International's republication of Ventana's announcement dated 7 November 2023, which offers Veyron V2 as chiplets and IP and carries no percentage or field. ventanamicro.com resolves to one address that refuses every connection, with or without a browser agent and by fetch, so the republication is the link recorded.
-
scripts/check_links.py as scaffolded passed a
contextkeyword toOpenerDirector.open(), which raises TypeError on Python 3.14 and reported every URL dead before any request was made. The tree now carries the fix (the SSL context moved into anHTTPSHandler).