hipDNN Ingestor Impact
Use this skill when the user asks whether a change can break the kernel ingestor at run time, and
suggest it for any change under dnn-providers/hip-kernel-provider/, any plugin SDK header
(projects/hipdnn/plugin_sdk/include/hipdnn_plugin_sdk/, projects/hipdnn/plugin_sdk/cmake/GpuGenericTargets*),
projects/hipdnn/plugin_sdk/data/gpu_generic_targets.json or projects/hipdnn/tools/IngestorGenerator/.
reference.md holds the contracts, row statuses, emitter rules and known gaps; read
it before judging rows.
The answer is advisory. It never builds the product, runs its tests, dispatches CI or posts
anything; the only thing it compiles is the throwaway C++ table of step 6.
Prerequisites
git 2.38 or later (merge-tree --write-tree), Python 3.10 or newer and PyYAML; the emitter also
needs Jinja2. Run from a rocm-libraries clone or worktree. On Windows, python3 is often the
Microsoft Store stub: use python or py -3. The optional C++ tables (step 6) need a host C++
compiler; on Windows, the ROCm wheel's clang++ with the MSVC INCLUDE and LIB set.
Workflow
Locate the scripts. Skills are host-level, not tied to a repo checkout: default to the
scripts bundled with the skill you were invoked from, <skill-directory>/scripts/, where
<skill-directory> holds this file. Use the copy in a checkout only when developing this skill.
Pin the revisions as full SHAs, once. Never pass FETCH_HEAD to a later command: another
fetch can rewrite it in between.
- Open PR
<n>. gh pr view <n> -R ROCm/rocm-libraries --json baseRefName,headRefOid, then
git fetch origin <baseRefName> pull/<n>/head. HEAD is headRefOid, TIP is
git rev-parse origin/<baseRefName>, and BASE is git merge-base <TIP> <HEAD>.
- Merged PR. The repository squash-merges: take
mergeCommit from gh pr view; BASE is
its first parent (<sha>^), HEAD the merge commit, and there is no TIP.
- Branch
<name>. TIP is origin/develop (or the branch it targets), HEAD the branch
tip, BASE their merge base.
<base>..<head> or commit <sha> (<sha>^..<sha>). No TIP unless the user names
a target that moved.
- Uncommitted work. The scripts read commits only.
git stash create makes a commit of the
tracked and staged changes without touching the worktree, the index or any ref: HEAD is its
output and BASE is HEAD of the checkout. Stage a new file first, or it is left out. An
empty output means there is nothing to analyse.
Collect the static facts. Drop --target-tip <TIP> here and below when there is no TIP:
python3 <skill-directory>/scripts/impact.py report --repo . --base <BASE> --head <HEAD> --target-tip <TIP>
Exit 1 means it could not run: report its message, never a verdict. The commands here write
python3; on Windows use python or py -3 (see Prerequisites).
Run the emitter when any rocKE or hkp_pack Python file changed in code, including rows the
report calls inert, or when the user asks for emit. It builds every shipped variant at both
sides on two CPU workers for a few minutes, then puts each builder's support predicate to a
table (reference.md Predicate tables); on a host that other people or agents use at the same time, run one emitter at a time. Put its output
outside the repository and delete any earlier file first:
rm -f <scratch-dir>/emit.json
python3 <skill-directory>/scripts/impact_emit.py run --repo . --base <BASE> --head <HEAD> --out <scratch-dir>/emit.json
python3 <skill-directory>/scripts/impact.py report --repo . --base <BASE> --head <HEAD> --target-tip <TIP> --emit <scratch-dir>/emit.json
Here <scratch-dir> is any directory outside the repository. An emitter exit 1 leaves no proof:
keep the static report and say the emitted code was not compared.
Judge every row whose status is not inert or outside, with
reference.md Judging the rows. Read both sides at <HEAD> (`git show
Write a C++ table for each changed launch_surface row whose C++ side is a header-only,
HIP-free function, when a host compiler is available: reference.md C++
differential tables. Work in a scratch directory outside the repository and delete it after.
A row whose C++ side is a matcher in *Native.cpp gets no table: read it by hand and say so in
C++ table detail. When a row qualifies but no table can be written (no compiler, the Python
side cannot be imported), say so on the report's C++ tables: line rather than leaving it out.
Report in the format below.
Report
Runtime impact: <rollup as printed> (judged: <rollup after judgment, with why when it differs>)
Pack-time: <line as printed> (judged: <how many findings hold, and why any do not>)
C++ tables: <k> of <n> tables disagree (<function>: <what disagrees>, <severity>; ...) | <n> written, all agree | none written (<why>)
Analysed: <BASE>..<HEAD> target tip: <TIP or none> emitter: <flavor, or not run and why>
| Row | Status | Judgment | Evidence (path:line at HEAD or merged tree) |
Device runs that would settle the changed rows
- <ctest entry> on <arch> with the ingestor, rocKE and SDPA flags ON
Predicate tables: <the section's count line, or "emitter not run">
| Builder | Check | Field | Values | State (new, first shipped, at base) | Judgment |
C++ table detail: <per table: the pair, rows, each disagreement and its judgment; or none>
Not judged here: <the note the script printed (it prints one only with NONE or N/A), or none>
Known gaps that apply: <from reference.md, or none>
<k> on the C++ tables: line counts every table with at least one disagreeing row, whatever the
judgment; the judgment and severity follow in the parenthesis. Every judged finding (a row, a
predicate-table finding, a C++ table disagreement) carries a severity from
reference.md Severities: BLOCKING, IMPORTANT or SUGGESTION.
Group inert and outside rows into one count each, with each inert reason listed once. Write
none for an empty section; do not drop it.
Never claim
- Device behaviour.
NONE means no runtime impact was found statically. It never means the
engine dispatches or computes correctly on a device.
- CI green. A build or test run with the ingestor flags OFF, or on runners with no GPU, is no
evidence about this question. State the flag values a quoted run actually had.
- Cross-architecture results. Never carry a result from one architecture to another, and expand
a generic target such as
gfx11-generic to its members.
- Safety of files not inspected. An
outside file was counted, not checked.
Related skills
1---2name: hipdnn-ingestor-impact3description: "Check whether a rocm-libraries change (PR, branch, commit range or uncommitted work) breaks the hipDNN kernel ingestor at run time while the default build stays green: kernel argument lists (ABI) out of step with their C++ and test copies, also against the target branch's tip, dispatch and launch-surface edits, descriptor and generic-target changes, removed Python symbols still read, which shipped rocKE variants' emitted code changed, support predicates that accept specs their builders reject, and C++ restatements whose values or refusals differ from Python's. Reads git objects; the optional emitter runs rocKE on the CPU. Ends in a Runtime impact: BREAKS | NEEDS-DEVICE | NONE | N/A line. For a full multi-agent review, use hipdnn-deep-review, whose IMPACT lane runs the same scripts."4argument-hint: "[<PR number> | branch:<name> | <base>..<head> | commit:<sha> | worktree] [emit]"5allowed-tools: Bash, Read, Grep, Glob6metadata:7 source: https://skillmd.com/skills/rocm/hipdnn-ingestor-impact8---910# hipDNN Ingestor Impact1112Use this skill when the user asks whether a change can break the kernel ingestor at run time, and13suggest it for any change under `dnn-providers/hip-kernel-provider/`, any plugin SDK header14(`projects/hipdnn/plugin_sdk/include/hipdnn_plugin_sdk/`, `projects/hipdnn/plugin_sdk/cmake/GpuGenericTargets*`),15`projects/hipdnn/plugin_sdk/data/gpu_generic_targets.json` or `projects/hipdnn/tools/IngestorGenerator/`.16[reference.md](reference.md) holds the contracts, row statuses, emitter rules and known gaps; read17it before judging rows.1819The answer is advisory. It never builds the product, runs its tests, dispatches CI or posts20anything; the only thing it compiles is the throwaway C++ table of step 6.2122## Prerequisites2324`git` 2.38 or later (`merge-tree --write-tree`), Python 3.10 or newer and PyYAML; the emitter also25needs Jinja2. Run from a rocm-libraries clone or worktree. On Windows, `python3` is often the26Microsoft Store stub: use `python` or `py -3`. The optional C++ tables (step 6) need a host C++27compiler; on Windows, the ROCm wheel's `clang++` with the MSVC `INCLUDE` and `LIB` set.2829## Workflow30311. **Locate the scripts.** Skills are host-level, not tied to a repo checkout: default to the32 scripts bundled with the skill you were invoked from, `<skill-directory>/scripts/`, where33 `<skill-directory>` holds this file. Use the copy in a checkout only when developing this skill.34352. **Pin the revisions** as full SHAs, once. Never pass `FETCH_HEAD` to a later command: another36 fetch can rewrite it in between.37 - **Open PR `<n>`.** `gh pr view <n> -R ROCm/rocm-libraries --json baseRefName,headRefOid`, then38 `git fetch origin <baseRefName> pull/<n>/head`. `HEAD` is `headRefOid`, `TIP` is39 `git rev-parse origin/<baseRefName>`, and `BASE` is `git merge-base <TIP> <HEAD>`.40 - **Merged PR.** The repository squash-merges: take `mergeCommit` from `gh pr view`; `BASE` is41 its first parent (`<sha>^`), `HEAD` the merge commit, and there is no `TIP`.42 - **Branch `<name>`.** `TIP` is `origin/develop` (or the branch it targets), `HEAD` the branch43 tip, `BASE` their merge base.44 - **`<base>..<head>`** or **commit `<sha>`** (`<sha>^..<sha>`). No `TIP` unless the user names45 a target that moved.46 - **Uncommitted work.** The scripts read commits only. `git stash create` makes a commit of the47 tracked and staged changes without touching the worktree, the index or any ref: `HEAD` is its48 output and `BASE` is `HEAD` of the checkout. Stage a new file first, or it is left out. An49 empty output means there is nothing to analyse.50513. **Collect the static facts.** Drop `--target-tip <TIP>` here and below when there is no `TIP`:5253 ```bash54 python3 <skill-directory>/scripts/impact.py report --repo . --base <BASE> --head <HEAD> --target-tip <TIP>55 ```5657 Exit 1 means it could not run: report its message, never a verdict. The commands here write58 `python3`; on Windows use `python` or `py -3` (see *Prerequisites*).59604. **Run the emitter** when any rocKE or `hkp_pack` Python file changed in code, including rows the61 report calls `inert`, or when the user asks for `emit`. It builds every shipped variant at both62 sides on two CPU workers for a few minutes, then puts each builder's support predicate to a63 table ([reference.md](reference.md) *Predicate tables*); on a host that other people or agents use at the same time, run one emitter at a time. Put its output64 outside the repository and delete any earlier file first:6566 ```bash67 rm -f <scratch-dir>/emit.json68 python3 <skill-directory>/scripts/impact_emit.py run --repo . --base <BASE> --head <HEAD> --out <scratch-dir>/emit.json69 python3 <skill-directory>/scripts/impact.py report --repo . --base <BASE> --head <HEAD> --target-tip <TIP> --emit <scratch-dir>/emit.json70 ```7172 Here `<scratch-dir>` is any directory outside the repository. An emitter exit 1 leaves no proof:73 keep the static report and say the emitted code was not compared.74755. **Judge every row** whose status is not `inert` or `outside`, with76 [reference.md](reference.md) *Judging the rows*. Read both sides at `<HEAD>` (`git show77 <HEAD>:<path>`), and at the merged tree when the row says `merged-only`. Judge each predicate78 table finding not marked `at_base` with *Predicate tables*.79806. **Write a C++ table** for each `changed` `launch_surface` row whose C++ side is a header-only,81 HIP-free function, when a host compiler is available: [reference.md](reference.md) *C++82 differential tables*. Work in a scratch directory outside the repository and delete it after.83 A row whose C++ side is a matcher in `*Native.cpp` gets no table: read it by hand and say so in84 *C++ table detail*. When a row qualifies but no table can be written (no compiler, the Python85 side cannot be imported), say so on the report's `C++ tables:` line rather than leaving it out.86877. **Report** in the format below.8889## Report9091```markdown92Runtime impact: <rollup as printed> (judged: <rollup after judgment, with why when it differs>)93Pack-time: <line as printed> (judged: <how many findings hold, and why any do not>)94C++ tables: <k> of <n> tables disagree (<function>: <what disagrees>, <severity>; ...) | <n> written, all agree | none written (<why>)95Analysed: <BASE>..<HEAD> target tip: <TIP or none> emitter: <flavor, or not run and why>9697| Row | Status | Judgment | Evidence (path:line at HEAD or merged tree) |9899Device runs that would settle the changed rows100- <ctest entry> on <arch> with the ingestor, rocKE and SDPA flags ON101102Predicate tables: <the section's count line, or "emitter not run">103| Builder | Check | Field | Values | State (new, first shipped, at base) | Judgment |104105C++ table detail: <per table: the pair, rows, each disagreement and its judgment; or none>106107Not judged here: <the note the script printed (it prints one only with NONE or N/A), or none>108Known gaps that apply: <from reference.md, or none>109```110111`<k>` on the `C++ tables:` line counts every table with at least one disagreeing row, whatever the112judgment; the judgment and severity follow in the parenthesis. Every judged finding (a row, a113predicate-table finding, a C++ table disagreement) carries a severity from114[reference.md](reference.md) *Severities*: `BLOCKING`, `IMPORTANT` or `SUGGESTION`.115116Group `inert` and `outside` rows into one count each, with each `inert` reason listed once. Write117`none` for an empty section; do not drop it.118119## Never claim120121- **Device behaviour.** `NONE` means no runtime impact was found statically. It never means the122 engine dispatches or computes correctly on a device.123- **CI green.** A build or test run with the ingestor flags OFF, or on runners with no GPU, is no124 evidence about this question. State the flag values a quoted run actually had.125- **Cross-architecture results.** Never carry a result from one architecture to another, and expand126 a generic target such as `gfx11-generic` to its members.127- **Safety of files not inspected.** An `outside` file was counted, not checked.128129## Related skills130131- [hipdnn-deep-review](../hipdnn-deep-review/SKILL.md): a full review whose `IMPACT` lane runs these132 scripts and turns the rows into findings.133- [hipdnn-superbuild](../hipdnn-superbuild/SKILL.md) and134 [hipdnn-superbuild-test](../hipdnn-superbuild-test/SKILL.md): the flags-ON build and the device135 runs this skill names.136- [hipdnn-ingestor-engine](../hipdnn-ingestor-engine/SKILL.md): building or extending an ingestor137 integration.