Research window: 2026-11-11 04:30 ~ 2026-11-12 04:30 (Beijing time, past 24 hours; directly contiguous with the cutoff of the previous 11-11 report).

Sources: GitHub key repositories and organization push verification (RuyiAI-Stack, ruyisdk, openRuyi-Project, buddy-compiler; per-repository verification of commits, PRs, Releases, and tags within the window, with commit counts and times re-checked via API and branch-level atom as a fallback), Release notes for the ruyi package manager and ruyi-imager, packages-index, openRuyi packaging repository SPECS, ruyiai-daily daily intelligence archive, post-by-post verification of the RuyiSDK developer community forum, snapshots of three pages of the official RuyiAI homepage, the openRuyi distribution news page, targeted Chinese and English queries via Google News RSS (through a proxy), arXiv search, and verification of reports by China Science Daily.


This Issue’s Index

  • Today’s Focus: RuyiSDK biweekly progress issue 080 goes live on the community forum, covering three tools and six compiler upstream merges (11-11)
  • Today’s Focus: buddy-mlir fixes reduction lane loss on VLEN 1024, twenty-five tests flip from fail to pass (11-11)
  • I. RuyiAI Stack Progress (PyTorch / Triton / TileLang RISC-V adaptation)
    • 1.1 triton-riscv: all three merges fix the release chain; AMD backend docs add the ROCm-version torch prerequisite (11-11)
    • 1.2 triton-riscv nightly predates the three merges; main branch still has ten open PRs (11-11 status)
    • 1.3 pytorch, tilelang-riscv, llvm-project: zero commits in window (11-12 status)
  • II. RuyiSDK Toolchain and Distributions
    • 2.1 Ruyi Imager 1.1.0: four Release notes; K3 flashing support recorded via biweekly progress (11-10)
    • 2.2 openRuyi: thirty-six commits; ten OpenTelemetry packages and four Python auth packages all bumped (11-11)
    • 2.3 openRuyi: kernel bumped to 7.3.4 and 6.18.63; clang-wrap follows to 20261111 (11-11)
    • 2.4 openRuyi fills out device and peripheral coverage: three OpenCL packages, fwupd, qemu-rustsbi, and five Rust dependencies (11-11)
    • 2.5 ruyi-deck: openRuyi creates a new release-ops console repo with a single initialization commit (11-11)
  • III. Vector Extensions and Operator Optimization
    • 3.1 Lane headroom for strip-mined set_vl drops stepwise by half-width; YOLOv8 hotspots on K3 emerge from scalar dot products (11-11)
    • 3.2 The three vector repos remain silent; all vector-side increments are at the VIR layer (11-12 status)
  • IV. Buddy Compiler Updates
    • 4.1 Preserve scaling factor when cloning dynamic vector types (11-11)
    • 4.2 Nightly 0.1.4.dev20261111 ships, closing a six-day gap (11-11)
    • 4.3 VIR and triton-riscv point to the same hardware line: K3’s VLEN 1024 (11-11)
  • V. Community, Events, and Official Site
    • 5.1 Three forum posts on the same day: issue 080, open-source highlights roundup, and a T-Head repost (11-11)
    • 5.2 China Science Daily front page covers ten years of the Institute of Software; Ruyi community and the kernel same-origin plan appear in the article (11-10)
    • 5.3 Official site’s three pages unchanged; two blog posts, twenty-three doc chapters (11-12 status)
    • 5.4 ruyiai-daily archive still stuck at issue 11-08 (11-12 status)
    • 5.5 External news and arXiv: no additions in window (11-12 status)
  • VI. Trend Observations
    • 6.1 Release chain before features: what triton-riscv fixed this week was “can we ship a package”
    • 6.2 VLEN 1024 is becoming a shared pressure point for compilers and runtimes
    • 6.3 Biweekly progress draft day and publish day differ by six days; read its items accordingly

Today’s Focus: RuyiSDK Biweekly Progress Issue 080 Published on the Community Forum, Covering Three Tools and Six Compiler Upstream Merges (11-11)

Date: 2026-11-11 Source: Biweekly Progress Issue 080 (Community Forum), wechat-articles summary submission #405

Issue 080 was posted to the RuyiSDK community forum on 11-11 at 17:53 (Beijing time), six days after the in-repo draft of 11-05. This issue closes out the “three tool releases”: package manager 0.54.0, Ruyi Imager 1.1.0, and VS Code extension 0.1.7 are all named, of which 0.54.0 and Imager 1.1.0 are formal releases within this window, while extension 0.1.7 is in fact the 09-28 version, with 0.1.8-beta.1 (10-30) already following; it is carried over here per the draft’s wording, and readers should not treat it as a new version.

The package manager item is the same 0.54.0 as in the previous daily report; what Issue 080 adds is the testing angle: 0.54.0-beta.20261102 and extension v0.1.7 were judged to meet the exit criteria, see the test report for details; on the toolchain side, CI is not yet run on Python 3.15 pending pygit2 support for Python 3.15, which is the only explicit deferral in this issue. The next version is set for late November.

The compiler section is the most substantive content in this issue. Three RISC-V GCC backend optimizations entered the GCC development mainline on 10-27 (r17-5638, r17-5639, r17-5640): Zba adds combined recognition of shift-add with integer division and remainder; Zbs exploits bclr’s own shift-index semantics to eliminate redundant masks, improving code generation for RV32/RV64 bitmap operations; Zbkb fixes the cost model and match priority of packh—previously, when Zbb and Zbkb were both enabled, the better instruction could not be selected. All three come with regression tests. On the RISE-CI side, container services and GitHub Actions were connected, a multilib build scenario check was added, and integration of LLM-based evaluation of CI errors is planned; performance benchmarking is moving toward SPEC CPU 2026, starting with environment and resource assessment, which has already identified a memory shortfall for running the full SPECspeed 2026 on K3.

On the LLVM side, six PRs for the RISC-V P extension and GlobalISel have been merged upstream (#235306, #235245, #234450, #237815, #238214, #234484), focused on accumulation merging for packed multiply-accumulate, branch elimination for 64-bit funnel shifts on RV32, and widening of narrow-type leading-zero and trailing-zero counts under XTHeadBb. RuyiSDK’s own LLVM branch has two more: preliminary support for the RV32/RV64 matrix extension per the Ztt specification and intrinsics documentation v0.6 (#169, covering assembly/disassembly, Clang types and intrinsics, IR and code generation, M/A/CC register allocation and spill handling), and GitHub Actions CI for the AME branch (#170).

One item each on the runtime and language sides is worth noting: OpenJDK’s JEP 544 (AOT static compilation) RISC-V port passed functional tests, met expected performance gains, and has been merged into the mainline (#32311), with ten additional RISC-V-related PRs entering the mainline and three backported to 17u/21u/25u; the V8 review list includes “Allow vlen=1024” (CL8585179), pointing to the same hardware fact as the two K3 fixes in this issue’s buddy-mlir. On the Go side, five CLs were submitted, of which the RVV implementation of GreenGC’s filterNil has been merged.

Community and content: packages-index added GNU toolchain support for SpacemiT K1/K3 and supplemented ABI compatibility information for some packages; the development board support matrix added Chinese and English reports for Milk-V Jupiter2 with Bianbu 4.0.1 and SpacemiT K3 Pico-ITX with Bianbu 4.0.8 LXQt, and updated the Ubuntu 24.04.5 report for VisionFive 2 Lite; the examples repo added Dhrystone and SiFive Kernel Library Chinese examples for HiFive Premier P550, a llama-server deployment example for Jupiter2, and extended the K3 ROS 2 course entry to chapters 11 through 15 (scan matching, SLAM, and localization).


Today’s Focus: buddy-mlir Fixes Lane Loss in Reductions on VLEN 1024, Twenty-Five Tests Go from Wrong to Right (11-11)

Date: 2026-11-11 Source: PR #1051

PR #1051 was merged on 11-11 at 23:51 (Beijing time), fixing a bug that only surfaces on long-vector hardware: performing vir.reduce on a scalable vector of width 1 leaves only lane 0 in the result. The cause is not VIR’s own canonicalization but downstream—vector.reduction’s ElideSingleElementReduction treats vector<[1]xT> as a “single-element” vector, rewriting it to vector.extract of lane 0 combined with the accumulator, and the remaining vscale - 1 lanes vanish.

The trigger condition is VLEN 1024: a region of 16 f32 values is vector<[1]xf32> (vscale 16) in a single vector, whereas at VLEN 128 and 256 it is vector<[4]xf32> and wider, never reaching this path. The author measured on the A100 core of SpacemiT K3, where all 25 reduction tests produced wrong values—for example, amax over 512 values yielded 1.853 instead of 4.101; the same tests pass on the CI riscv64 runner (VLEN 256). This is precisely why such defects are hard to catch: CI’s vector width coverage does not reach the width of shipping hardware.

The fix bypasses canonicalization and constructs the reduction directly with the LLVM intrinsics that vector.reduction ultimately lowers to—llvm.vector.reduce.fadd for floating point with the accumulator as the starting value, and maxnum plus the accumulator for fmax—changing both lowering paths of vir.reduce. The author notes this is the same workaround as splat’s broadcastTo. Accompanying tests go into the VIR cases under tests/Conversion, and check-buddy passes on both x86 (using main’s LLVM) and SG2044 natively.

What this means for the stack: Triton-RISCV’s reduction operators on K3 (amax, max, mean, sum, cumsum, rms_norm, mse_loss, vector_norm, euclidean_dist, attention) were previously silently wrong on this board, not crashing. Only after the fix does discussing performance become meaningful.


I. RuyiAI Stack Progress (PyTorch / Triton / TileLang RISC-V Porting)

1.1 triton-riscv: All three merges fix the release chain; AMD backend docs add the ROCm torch prerequisite (11-11)

Date: 2026-11-11 Source: PR #111 fix release script, PR #113 shared git auth, PR #112 AMD backend torch prerequisite

Three merges in the window, none of which touch compiler functionality — all are on the release chain. #113 extracts the “one-off global git config” that CI and release.sh each duplicated into a shared scripts/_git-auth.sh: anonymous git traffic on the self-hosted riscv64 runner gets no authentication, yet curl to the same host works fine, so when upstream FetchContent pulls googletest it returns 401 and hangs at the username prompt with no tty. #111 fixes the release script itself. #112 is documentation: after #110 merged, the CPU path no longer needs torch at all, so that install line is removed from the README; the AMD section gains the missing half-sentence — torch is required, and it must be a build that can use the AMD GPU.

All three were authored by the same person (Mingzhu Yan, Institute of Software, CAS), between 17:59 and 21:11. The week’s priority was clearly making the nightly packages reproducible rather than expanding functionality.

1.2 triton-riscv nightly predates the three merges; main branch still has ten open PRs (11-11 status)

Date: 2026-11-11 (status) Source: nightly 2026-11-11-e02c9e1, open PR list

Nightly 2026-11-11-e02c9e1 was produced at 17:33, with release notes stating “authenticate git clones across the whole build, not just the triton one” — two steps of the same effort that #113 later consolidated. The package was cut before the day’s three merges (17:59, 21:10, 21:11), so as usual it excludes subsequent changes from that day.

Ten PRs remain open; the most recently updated are #103 (tutorial additions) and #104 (BOSCAME FPGA SPM example and AOT contract frontend), both touched on the morning of 11-11; #109’s plugin rename (unifying triton_shared to riscv, along with the resulting change to TRITON_DEFAULT_BACKEND values) is still under review — detailed in yesterday’s report, unchanged this issue.

1.3 pytorch, tilelang-riscv, llvm-project: zero commits in the window (11-12 status)

Date: 2026-11-12 (status) Source: pytorch riscv branch, tilelang-riscv, llvm-project

The three repos’ head commits sit at 11-08 18:35, 11-03 06:13, and 11-02 14:48 (UTC) respectively. This issue’s RuyiAI stack delta comes entirely from the three triton-riscv merges. The pytorch riscv branch has not synced for four days; the last batch was the merges around 11-04.


II. RuyiSDK Toolchain and Distributions

2.1 Ruyi Imager 1.1.0: Four Release Notes Items, K3 Flashing Support Added via Biweekly Progress (11-10)

Date: 2026-11-10 Source: Ruyi Imager 1.1.0, commit list

1.1.0 was released on 11-10 at 15:46 (Beijing time), falling within the previous window but only the dependency side was reported at the time; the version contents are covered here. The release notes list four items, all from the same new contributor (qingwan12138, their first contribution): adding a download action to the image selection step (i.e., the standalone image pre-download button mentioned in the biweekly progress); making CLI output follow the console character set while JSON remains UTF-8; supporting double-click confirmation in the selection dialog; and requiring an explicit selection before the dialog can proceed.

The biweekly progress additionally records something not in the release notes: new SpacemiT K3 flashing support. The two are not contradictory—K3 support may be provided on the image manifest side rather than the code side, and the release notes do not list it separately; when taking the actual release contents as authoritative, both the Release and the biweekly progress should be consulted.

2.2 openRuyi: Thirty-Six Commits, Ten OpenTelemetry Packages and Four Python Auth Packages Upgraded Together (11-11)

Date: 2026-11-11 Source: openRuyi commit list

The packaging repo saw thirty-six commits in the window, the densest day in the past week. The largest cluster is OpenTelemetry: eight packages upgraded to 1.45.2 (exporter-otlp and its three sub-packages grpc/http/proto-common, proto, api, sdk), semantic-conventions upgraded to 0.66b2, plus two new packages exporter-otlp-common and exporter-http-transport. One of these, python-opentelemetry-api, was tagged security-fixes. The observability stack previously had only a few scattered packages in the distribution; this round fills it out to cover the common OTLP combinations.

The second cluster is four Python auth packages, all tagged security-fixes: pyjwt 2.15.1, oauthlib 4.0.1, joserfc 1.8.0, authlib 1.9.3. All four were committed at the same time, a batch action tracking upstream security advisories. For a RISC-V distribution that needs to run containers and cloud-native workloads, the speed of security follow-up on such packages says more about maintenance status than new features do.

2.3 openRuyi: Kernel Upgraded to 7.3.4 and 6.18.63, clang-wrap Follows to 20261111 (11-11)

Date: 2026-11-11 Source: linux commit, openjdk CA patch commit

The mainline kernel and the LTS kernel each got a version bump on the same day: linux to 7.3.4, linux-lts to 6.18.63, both with patchset and config refreshes. clang-wrap follows to 20261111, consistent with the llvm-snapshot cadence; the fact that downstream gets a compiler that changes daily was mentioned in the previous issue, and this is the third consecutive day.

There is also a commit refreshing the CA (certificate authority) enhancement patches for openjdk-17/21/25 simultaneously, three JDK lines moving together, a packaging consistency maintenance; cmake enabling the ccmake interface, htop upgraded to 3.5.3, and kactivitymanagerd fixing a dependency version number are three scattered desktop and toolchain fixes.

2.4 openRuyi Fills Out Device and Peripheral Coverage: Three OpenCL Packages, fwupd, qemu-rustsbi, and Five Rust Dependencies (11-11)

Date: 2026-11-11 Source: three OpenCL packages, qemu-rustsbi and Rust dependencies

Three OpenCL components added: opencl-headers, opencl-icd-loader, clinfo. For RISC-V desktops and AI acceleration this is a long-missing piece—triton-riscv drives AMD cards on RISC-V hosts via ROCm, and ROCm shares the ICD loading mechanism with OpenCL, so third-party implementations can only plug in once icd-loader is in place. fwupd was also added in the same batch, giving device firmware updates a standard entry point.

On the Rust side, five dependencies were added at once along with qemu-rustsbi: rust-memchr 2.7.6 and rust-libc 0.2.185 are new, rust-hashbrown upgraded to 0.17.1, plus rustc-literal-escaper 0.0.7 and clap-verbosity-flag 3. The appearance of qemu-rustsbi indicates that the RustSBI RISC-V firmware route has entered the QEMU rehearsal stage in the packaging repo.

2.5 ruyi-deck: openRuyi Creates a New Release Operations Console Repo, with Only an Initialization Commit (11-11)

Date: 2026-11-11 Source: ruyi-deck repo

ruyi-deck was newly created under openRuyi-Project, described as “The Operations Deck for openRuyi Repositories and Releases”, licensed MulanPSL-2.0, with only one “Init repo” commit in the window and a README containing just a one-line title. The naming and positioning indicate it is meant to be an operations console for repositories and releases—openRuyi currently ships a monthly release across multiple repos (packaging, images, firmware), and what is missing is exactly this kind of unified entry point. There is no code in the repo yet; this issue only records its existence.


III. Vector Extensions and Operator Optimization

3.1 Strip-Mined set_vl Lane Remainder Steps Down by Half-Width, YOLOv8 Hotspot on K3 Emerges from Scalar Dot Product (11-11)

Date: 2026-11-11 Source: PR #1050

#1050 was merged on 11-11 at 20:12. Previously a strip-mined vir.set_vl ran the main loop only at the selected width, dumping all remaining lanes into a scalar tail loop; regions with fewer lanes than one vector were entirely scalar. Running f32 YOLOv8 on K3’s A100 core (VLEN 1024), the pose head’s 51 output channels yield vectors of 64 lanes, and the result was that the hottest loop (5% of total model time) was a scalar flw plus fmadd.s dot product, with scalar floating-point instructions accounting for 39% of the 150,000 perf samples.

The fix steps the lanes after the main loop down successively to half-width and quarter-width, each level used at most once and in lane order, before finally handing off to the scalar tail loop, which retains only the portion below the minimum width (fixed 2 lanes, scalable [1]). There are no masks: scalable vectors check the remainder at runtime with scf.if, fixed-length vectors resolve it at compile time when vl is constant, so the affine tail loop’s bounds remain affine; when vl is dynamic, the scalar tail is retained.

Numerics are unchanged—each lane computes the same operation, and the lane order seen by ordered reductions is also unchanged; the author verified bit-identity using an FNV hash of the forward output on K3. This fixes performance, not correctness, but what it buys is that the hottest code on long-vector hardware finally goes to the vector unit.

3.2 Three Vector Repos Still Silent, Vector-Side Progress All at the VIR Layer (11-12 status)

Date: 2026-11-12 (status) Source: VecIntrinBench, IntrinTrans, FBGEMM

The head commits of the three vector-related repos remain at 02-13, 06-25, and 07-13, with no activity for months. This does not mean no one is working on the vector side: both buddy-mlir changes in this window (reduction lane handling and strip-mined remainder width stepping) directly determine what vector instructions are generated; the silence of the measurement tool and transpiler repos shows that investment remains concentrated in the compiler rather than benchmarks.


IV. Buddy Compiler News

4.1 Preserve Scaling Factor When Cloning Dynamic Vector Types (11-11)

Date: 2026-11-11 Source: PR #1048

#1048 was merged on 11-11 at 08:32, with the title being the change itself: carry the scaling factor along when cloning dynamic vector types. This is the wrap-up of the previous issue’s 4.3 (printing, parsing, and verification of dynamic vector types)—that batch made the scaling factor visible and verifiable at the text level, and this commit ensures it is not lost on the type cloning path. The change is small, but if the scaling factor is lost, the vector’s actual width is computed incorrectly, and all subsequent size inference goes wrong with it.

4.2 Nightly 0.1.4.dev20261111 Released, Filling the Six-Day Gap (11-11)

Date: 2026-11-11 Source: nightly 0.1.4.dev20261111

The nightly was released on 11-11 at 09:32, six days after the previous one 0.1.4.dev20261105, filling in the “six commits merged without a release” noted in the previous issue. The notes reference #1047—the reassoc reduction that first blocks by lane, fixing bit inconsistency on VLEN 128 hosts, which was the content of the previous issue’s 4.4. The official stable release remains BUDDY MLIR 0.1.3 from 10-31.

4.3 VIR and triton-riscv Point to the Same Hardware Line: K3’s VLEN 1024 (11-11)

Date: 2026-11-11 Source: PR #1051, PR #1050

Of the three buddy-mlir commits in this window, two were tested on SpacemiT K3’s A100 core (VLEN 1024), and #1051 explicitly states the issue was exposed by Triton-RISCV on that core. Around the same time, the RuyiSDK biweekly progress records V8 merging “Allow vlen=1024”, and triton-riscv’s ROCm validation on K3 was also covered in the previous issue. Three independent lines converge on the same parameter, indicating that 1024-bit vectors are already the default premise for this stack’s shipping form, and that the 128- and 256-bit configurations common in CI are no longer a sufficient test surface.


V. Community, Events, and Official Homepage

5.1 Forum publishes three posts on the same day: Issue 080, open-source highlights roundup, and a T-Head repost (11-11)

Date: 2026-11-11 Source: Latest topics in the RuyiSDK community, Open-source technology highlights post

Three new topics appeared in the forum window: the biweekly progress Issue 080 at 17:53 (today’s highlights already covered), a repost at 12:29, and a “2026-11-11 Open-Source Technology News Highlights” roundup at 11:50. The latter two posts have no direct relation to the RuyiAI stack: the repost’s title claims Alibaba plans to spin off T-Head for an independent listing in Hong Kong, but it is a reposted article and unverified; the highlights post is a routine daily aggregation of industry news. The forum’s substantive content remains concentrated on the biweekly progress thread.

5.2 China Science Daily front page covers the Institute of Software’s decade, with Ruyi community and the kernel homology plan mentioned (11-10)

Date: 2026-11-10 Source: China Science Daily report

The 11-10 front page of China Science Daily published a ten-year retrospective of the Institute of Software’s RISC-V basic software team. Several verifiable facts are worth noting: the Ruyi community was founded in November 2024; as of the end of 2025, code contributed by industry partners such as Alibaba and ZTE under the “RISC-V kernel homology” plan exceeded half of the total; and the RVA23 releases published by openEuler and openKylin all adopt the results of this homologous kernel work. The article also mentions that the software migration tool follows a “neuro-symbolic fusion” path (program analysis and formal verification as the foundation, large models for efficiency gains, and engineers’ experience distilled into rules and knowledge bases to delineate boundaries), already deployed in communities such as openEuler; and that a master’s student added a RISC-V backend and vector instruction optimizations to OpenCV, which were merged into the official mainline.

The “open-source AI compiler project for RISC-V” mentioned in the article is buddy-mlir, whose VIR commits mostly come from this team. This report provides the organizational background of this stack: the Ruyi community, kernel homology, and the compiler and migration tools all come from the same team; the changes scattered across different repositories in the daily briefings are, at the institutional level, one and the same line.

5.3 Official homepage’s three pages unchanged; two blog posts, twenty-three documentation chapters (11-12 status)

Date: 2026-11-12 (status) Source: RuyiAI official homepage, Blog page

The hashes of the index, blog, and docs pages match the previous inspection; the blog still has two posts (Triton’s RISC-V adaptation, PyTorch RISC-V CI practices), and the documentation index has twenty-three chapters. The recent changes to triton-riscv’s GPU backend and release chain still have no corresponding article on the site.

5.4 ruyiai-daily archive still stops at the 11-08 issue (11-12 status)

Date: 2026-11-12 (status) Source: ruyiai-daily commit list

The last commit in the daily intelligence repository is still 11-08; the four issues from 11-09 to 11-12 have not been generated. Previously the repository produced one issue per day; the gap has now lasted four days.

5.5 External news and arXiv: no new items in the window (11-12 status)

Date: 2026-11-12 (status) Source: Google News, arXiv RISC-V search

Targeted queries in Chinese and English for RuyiAI, RuyiSDK, Ruyi community, and openRuyi returned zero hits within the window (only one China Science Daily report in three days, already listed separately in 5.2); the latest paper in the arXiv RISC-V vector search is still 11-03, with no new papers in the window. General RISC-V crawls mostly returned teaching-oriented pipeline implementations and personal CPU repositories, unrelated to the Ruyi community.


VI. Trend Observations

6.1 Release chain before features: what triton-riscv fixed this week was “can we ship a package at all”

Of the three merges in triton-riscv during the window, two were CI and release scripts and one was a README note on installation prerequisites — zero functional changes. The nightly build’s own release notes are also about git authentication on the runner. This is not stagnation; it is a project that just graduated from “runs on real hardware” to “GPU backend enabled by default” filling in reproducibility: anonymous git traffic on the self-hosted riscv64 runner cannot obtain authentication, which looks like a minor infrastructure issue but in practice determines whether a nightly package comes out every day.

The contrast with RuyiSDK makes this clear: the package manager can turn ABI compatibility into a pre-install warning because its release chain is already stable enough for that kind of fine-grained work. triton-riscv is on the same path, just a few steps behind.

6.2 VLEN 1024 is becoming a shared pressure point for compilers and runtimes

Within a single day, three independent pieces of evidence pointed to the same parameter: buddy-mlir’s reduction lane loss only shows up at VLEN 1024 (VLEN 128 and 256 both pass), the motivation for narrowing the strip-mined remainder comes from 64-lane vectors on the K3 A100 core, and “Allow vlen=1024” appears in V8’s reviewer list. Previously this kind of discussion stayed at the architectural parameter level; now it has landed on “which test fails at which width.”

The direct requirement for testing is to cover shipping widths: the original CI riscv64 runner is VLEN 256, which happens to pass on the #1051 issue. That lesson is worth more than the few lines of fixed code — any reasoning involving vscale must be verified at 1024.

6.3 The biweekly progress draft and publish dates are six days apart; read its items separately

Issue 080 took shape in the repo on 11-05 but was only posted to the forum on 11-11. So in the same text, 0.54.0 and Imager 1.1.0 are “released” (things that happened on 11-10), and plugin 0.1.7 is also “released” (a 09-28 event, followed by 0.1.8-beta.1 on 10-30). Reading it straight through by publish date makes it easy to mistake three tools for a triple launch on the same day.

A workable reading is two-layered: for tool versions, look at Release dates; for compiler and runtime items, look at upstream merge dates (the three GCC items are dated 10-27). The daily briefing has already been split along these lines, and readers comparing against the biweekly progress should do the same.


Source Verification Table

Source Verification result
RuyiAI-Stack key repos triton-riscv: three merges (#111, #112, #113) and one nightly build; pytorch, tilelang-riscv, llvm-project, FBGEMM, cpuinfo, VecIntrinBench, IntrinTrans, harness, torq-tile all had zero commits
ruyisdk org ruyi: zero commits, zero releases; ruyi-imager 1.1.0 released 11-10 15:46 (supplementary content for this issue); packages-index: zero commits; wechat-articles: one summary commit (#405)
openRuyi-Project Packaging repo: 36 commits (API review: OpenTelemetry 10, Python auth 4, OpenCL 3, kernel 2, Rust dependencies 6, other 11); new ruyi-deck repo created (one initialization commit); infrastructure repos such as linux and kernel-team-tools had pushes; tags and news page still at 2026.09
buddy-compiler buddy-mlir: three merges (#1048, #1050, #1051) and nightly 0.1.4.dev20261111; stable release still 0.1.3
RuyiSDK forum Three new topics in the window: biweekly progress issue 080, open-source hotspot roundup, T-Head repost; only issue 080 has substantive content
RuyiAI official site index, blog, and docs pages all unchanged; 2 blog posts, 23 documentation chapters
ruyiai-daily Archive stops at the 11-08 issue; the four issues from 11-09 to 11-12 were not generated
openRuyi distro news page The latest two posts are still the two from 2026-10 (RISC-V software ecosystem public infrastructure, Ruyi Firmware); monthly releases not updated past 2026.10
Google News Zero hits across four sets of targeted Chinese and English queries; within three days only one article from China Science Daily (listed separately)
arXiv No new papers in the window; latest stops at 11-03
Media coverage One front-page article in China Science Daily on 11-10, on ten years of the Institute of Software, Chinese Academy of Sciences RISC-V foundational software team

Complete Source List