CK: Extract shared boilerplate from 47 gemm_quant test files#6323
Merged
AviralGoelAMD merged 49 commits intoApr 11, 2026
Conversation
The disabled loop applying b_element_op was superseded by the active code path using ReferenceGemm with PassThrough ops.
…instances Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
…m_b_scale/device_gemm_b_scale_xdl_f16_i4_f16 Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
…m_universal/device_gemm_xdl_universal_bf16_i4_bf16 Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
…m_universal/device_gemm_xdl_universal_f16_i4_f16 Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete #if 0 disabled code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
…pu/gemm_streamk Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
Remove obsolete commented-out code blocks identified by automated dead code scan with expert triage.
- test_gemm_pipeline_compiler.cpp: split into 13 smaller test files - test_grouped_gemm_quant.cpp: split into 5 smaller test files - 2 unsplit f8_f8_f16 instance files: superseded by _part1/_part2 splits Each file verified to have clear replacements already in the build.
Create test_gemm_quant_common.hpp with shared includes, layout aliases, data type aliases, quant type aliases, and group size aliases that were copy-pasted across all 47 test_gemm_quant_*.cpp files. Each test file now includes the common header instead of duplicating ~20 lines of identical boilerplate. Separate .cpp files are preserved for parallel compilation. No functional changes. Net reduction: ~733 lines of duplicated code.
This was referenced Apr 9, 2026
shumway
approved these changes
Apr 10, 2026
Contributor
shumway
left a comment
shumway
left a comment
There was a problem hiding this comment.
Thanks, looks good.
The real problem here is that we're testing wrong. This kind of systematic testing should be data-driven, thought the dispatcher, but we don't have all the pieces available yet. I'll keep this in mind and get back to you once we have that capability. (I've tested bquant with the rocm-ck prototype, and I'm planning on adding that to the dispatcher once all that new work has a design review and is in production.)
Base automatically changed from
users/avirgoel/ck/remove-unused-files
to
develop
April 10, 2026 15:22
AviralGoelAMD
added a commit
that referenced
this pull request
Apr 10, 2026
…N_INSTANCE macro (#6325) Depends on #6324 ## Summary Refactor all 144 contraction library instance `.cpp` files (bilinear + scale, 2D/6D, f32/f64/bf16/cf32/cf64) to use a shared `CK_CONTRACTION_INSTANCE` macro defined in `contraction_instance_common.hpp`. Each 58-line file is reduced to 12 lines — zero unique logic, pure parameterization. | Metric | Value | |--------|-------| | Files changed | 145 (144 instances + 1 new shared header) | | Insertions | +1,373 | | Deletions | −7,890 | | **Net lines removed** | **−6,517** | ### What changed | Before | After | |--------|-------| | 144 instance files, each ~58 lines of identical boilerplate (includes, namespace, type alias, registration function) differing only in 12 template parameters | 1 shared macro header + 144 files at ~12 lines each (copyright + include + macro invocation) | ### Macro parameters (12) ``` CK_CONTRACTION_INSTANCE(INST_TPL, OP_NAME, CDE_OP, NDIM_VAL, NAME_SUFFIX, ADATA, BDATA, ACC, CSHUFFLE, DS_TUPLE, EDATA, COMPUTE) ``` | Parameter | Example | Purpose | |-----------|---------|---------| | `INST_TPL` | `device_contraction_kk_instance` | Device template to instantiate | | `OP_NAME` | `bilinear` | Lowercase, used in `##` token pasting for function/type names | | `CDE_OP` | `Bilinear` | C++ type name (capitalized) for template args | | `NDIM_VAL` | `2` or `6` | Number of dimensions | | `NAME_SUFFIX` | `f32_f32_f32_f32_kknn` | Data type + layout suffix | | `ADATA..COMPUTE` | `F32, F32, ...` | Template type arguments | ### Readability assessment A code realist review confirmed this change **improves readability**: the original 58-line files contained zero unique logic — just mechanical boilerplate wrappers that varied only in 12 template parameters. After the macro, each file's intent is immediately clear from a single macro call, and the 12 parameters serve as a concise specification of what the instance does (data types, layout, operation). Adding a new contraction instance requires writing 1 line instead of copying and modifying 58 lines. The realist also noted that the area has very low activity (1 functional commit in 18 months), so merge conflict risk is negligible. ### Cumulative cleanup series stats | PR | Description | Net lines | |----|-------------|-----------| | #6300 | Remove 61 dead `#if 0` blocks | −2,648 | | #6302 | Remove 41 commented-out dead code blocks | −2,861 | | #6303 | Remove 4 orphaned files | −3,886 | | #6323 | Extract gemm_quant test boilerplate | −693 | | #6324 | Extract contraction example boilerplate | −1,016 | | This PR | Refactor 144 contraction instance files | −6,517 | | **Total** | | **−17,621** |
…ro (#6324) Depends on #6323 ## Summary Extract repeated `DeviceOpInstance` type alias boilerplate from 20 contraction example files into a single macro `CK_CONTRACTION_DEVICE_OP_INSTANCES(BASE, SUFFIX)` in `common_instances.hpp`. Each example file's 4 device-op-instance blocks (~56 lines) are replaced by 4 one-line macro calls. | Metric | Value | |--------|-------| | Files changed | 22 | | Insertions | +124 | | Deletions | −1,140 | | **Net lines removed** | **−1,016** | ### What changed | Before | After | |--------|-------| | 20 example files, each with 4 identical `DeviceOpInstance` type alias blocks (~14 lines each) | 1 macro definition in `common_instances.hpp` + 20 files with 4 one-line macro calls each | ### Readability assessment A code realist review confirmed this change **improves readability**: the 14-line `DeviceOpInstance` blocks were pure noise — identical across all 20 files and obscuring the actual example logic (data types, element operations). After the macro, each file's intent is immediately clear from the 4 one-liner macro calls, and a developer adding a new contraction example only needs to specify the varying parameters instead of copying 56 lines of boilerplate. ### GPU verification All 20 contraction examples verified on MI300X: **20/20 passed**. ### Cumulative cleanup series stats | PR | Description | Net lines | |----|-------------|-----------| | #6300 | Remove 61 dead `#if 0` blocks | −2,648 | | #6302 | Remove 41 commented-out dead code blocks | −2,861 | | #6303 | Remove 4 orphaned files | −3,886 | | #6323 | Extract gemm_quant test boilerplate | −693 | | This PR | Extract contraction example boilerplate | −1,016 | | **Total** | | **−11,104** |
AviralGoelAMD
deleted the
users/avirgoel/ck/dedup-gemm-quant-boilerplate
branch
assistant-librarian Bot
pushed a commit
to ROCm/composable_kernel
that referenced
this pull request
Apr 11, 2026
CK: Extract shared boilerplate from 47 gemm_quant test files (#6323) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Depends on #6303 ## Summary Extract shared test boilerplate (includes, type aliases, test fixture macros) from 47 `test_gemm_quant_*` files into a single `test_gemm_quant_common.hpp` header. Each test file is reduced from ~50 lines of boilerplate to ~5 lines. | Metric | Value | |--------|-------| | Files changed | 48 | | Insertions | +413 | | Deletions | −1,106 | | **Net lines removed** | **−693** | ### What changed | Before | After | |--------|-------| | 47 test files, each with ~50 lines of identical includes, type aliases, and fixture macros | 1 shared header (`test_gemm_quant_common.hpp`) + 47 thin files (~5 lines each: include + params) | ### Readability assessment A code realist review confirmed this change **improves readability**: the 47 test files had identical boilerplate obscuring the only meaningful content — the `GemmConfig` type alias and test dimensions. After the refactoring, each file's unique configuration is immediately visible, and adding a new test variant requires specifying only the varying parameters instead of copying 50 lines. ### Cumulative cleanup series stats | PR | Description | Net lines | |----|-------------|-----------| | #6300 | Remove 61 dead `#if 0` blocks | −2,648 | | #6302 | Remove 41 commented-out dead code blocks | −2,861 | | #6303 | Remove 4 orphaned files | −3,886 | | This PR | Extract gemm_quant test boilerplate | −693 | | **Total** | | **−10,088** |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Depends on #6303
Summary
Extract shared test boilerplate (includes, type aliases, test fixture macros) from 47
test_gemm_quant_*files into a singletest_gemm_quant_common.hppheader. Each test file is reduced from ~50 lines of boilerplate to ~5 lines.What changed
test_gemm_quant_common.hpp) + 47 thin files (~5 lines each: include + params)Readability assessment
A code realist review confirmed this change improves readability: the 47 test files had identical boilerplate obscuring the only meaningful content — the
GemmConfigtype alias and test dimensions. After the refactoring, each file's unique configuration is immediately visible, and adding a new test variant requires specifying only the varying parameters instead of copying 50 lines.Cumulative cleanup series stats
#if 0blocks