Repository navigation
Conversation
Add gfx1171 to GPU_ARCHS and trim its Tensile logic to the same four HHS/HSS GridBased stems as gfx1170, so the SDK ships bin\hipblaslt\library\gfx1171. Without it hipBLASLt finds no Tensile library on gfx1171 and every matmul fails. Co-Authored-By: Copilot <copilot@github.com>
|
Thanks for opening a PR! This project follows LLVM's incremental-development and AI-tool-use Before requesting review, please check that:
Reviewers are assigned through |
zz002
added a commit
to ROCm/hip-ep
that referenced
this pull request
Oct 27, 2026
Point therock_windows at rocm-dist-windows-gfx115x-gfx117x-7.16.0-hipblaslt-gfx117x-trimmed (v0.5.2 asset, built by zz002#7). It adds bin\hipblaslt\library\gfx1171, trimmed to the same HHS/HSS stems as gfx1170; without it hipBLASLt finds no Tensile library on gfx1171. Co-Authored-By: Copilot <copilot@github.com>
2 of 6 tasks
gemma-4's vision pooler multiplies by a transposed operand, which reaches hipBLASLt as transA=N, transB=T, fp16 in/out, f32 compute. The gfx1170 and gfx1171 keep-set has only the Bljk (transB=N) stems, so the heuristic returns no algorithm, the matmul fails and the image features are garbage. Add Cijk_Ailk_Bjlk_HHS to the keep-set: one more .co per gfx117x arch (about 1 MB), well inside the trimmed-set limit. Fail the step when a keep-set stem is missing, instead of skipping it, so the trimmed library cannot silently lose a stem the workloads need. Co-authored-by: Cursor <cursoragent@cursor.com>
zz002
added a commit
to ROCm/hip-ep
that referenced
this pull request
Oct 28, 2026
Point therock_windows at rocm-dist-windows-gfx115x-gfx117x-7.16.0-hipblaslt-gfx117x-trimmed (v0.5.2 asset, built by zz002#7). It adds bin\hipblaslt\library\gfx1171, trimmed to the same HHS/HSS stems as gfx1170; without it hipBLASLt finds no Tensile library on gfx1171. Co-Authored-By: Copilot <copilot@github.com>
zz002
added a commit
to ROCm/hip-ep
that referenced
this pull request
Oct 30, 2026
… there (#1329) ## Summary Build the gfx117x custom kernels for `gfx11-7-generic`, so one DLL covers gfx1170 / gfx1171 / gfx1172. Switch to an SDK that ships gfx1171 hipBLASLt data, and fix the fp16 WMMA conv, which compiled to a trap on gfx117x. ## Related issue or design None ## Why - The pinned SDK (ROCm/llvm-project `a2529029`) defines `gfx11-7-generic`, so one generic target covers the whole gfx117x family. - The previous SDK has no `hipblaslt\library\gfx1171`, so hipBLASLt cannot find its Tensile library on gfx1171. - gfx117x is in the gfx11 family but implements only the gfx12-style WMMA. `conv_kernel.hip` called the gfx11 builtin directly, which falls back to `__builtin_trap()` on gfx117x, so every fp16 WMMA conv faulted on the GPU. This is what broke the gemma3 vision encoder on gfx1171. ## What - `HIP_ARCHITECTURES`: `gfx11-generic;gfx1170` -> `gfx11-generic;gfx11-7-generic`. The four generic-target tables (`genericTargetFor()`, `kGenericTargets`, `_lut_arch_members()`, `_GENERIC_MEMBERS`) map gfx1170 / gfx1171 / gfx1172 to `gfx11-7-generic`. - `cmake/deps.txt`: SDK -> v0.5.2 `rocm-dist-windows-gfx115x-gfx117x-7.16.0-hipblaslt-gfx117x-trimmed`, which adds gfx1171 hipBLASLt data; gfx1170 / gfx1171 keep the same five fp16 Tensile stems (built by zz002#7). - `conv_kernel.hip`: `conv_wmma_kernel` and `conv_window_kernel` use the WMMA helpers from `hip_arch_compat.h` (fragment width, K offset, accumulator row, builtin), as `gemm_wmma_kernel.hip` does. The gfx11 encoding and memory access are unchanged. ## Test plan - [ ] build-windows green; `custom_kernels_gfx11-7-generic.dll` and `kernel-tests/gfx11-7-generic/` are staged - [ ] gpu-test-package contains `bin\hipblaslt\library\gfx1171` ## Notes for reviewers The DLL size gate reports `custom_kernels_gfx11-7-generic.dll` +5.58 MB FAIL because it compares by file name and main has no file by that name, so the baseline counts as 0. The DLL replaces main's `custom_kernels_gfx1170.dll`; `hipgpu.dll` and `custom_kernels_gfx11-generic.dll` are unchanged (Δ 0.00). ## Checklist - [x] The change is focused, or links a design/series explaining its scope. - [ ] Relevant tests were added or updated and the results are documented. - [ ] User-facing or design documentation was updated when needed. - [x] Substantial AI assistance is disclosed, and I reviewed and understand the result. --------- Co-authored-by: Copilot <copilot@github.com> Co-authored-by: Cursor <cursoragent@cursor.com>
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
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.
Summary
Builds gfx1171 in the SDK selfbuild, with its Tensile logic trimmed to the same four
HHS/HSSGridBased stems as gfx1170. Stacked onci/sdk-rc7-11d(#5).Why
The current SDK ships no
bin\hipblaslt\library\gfx1171, so on gfx1171 hipBLASLt finds no Tensile library, every hipBLASLt matmul fails and VLM output is garbage. ROCm#1329 moves the gfx1171 custom kernels togfx11-7-generic; hipBLASLt needs matching data.What
GPU_ARCHS: addgfx1171.bin\hipblaslt\library\gfx1171is a required path.rocm-dist-windows-gfx115x-gfx117x-<ver>-hipblaslt-gfx117x-trimmed.Test plan
Windows SDK SelfbuildcompleteshipBLASLt logic: gfx1171reports more than 0 and at most 20.co; gfx1150-gfx1153 counts unchanged