Repository navigation
GPU CommonIR:适配冻结 FLIR main 并保持显式 TileIR 转换 - #88
Draft
macqueen09 wants to merge 4 commits into
Draft
macqueen09 wants to merge 4 commits into
macqueen09 wants to merge 4 commits into
Conversation
LeBlanc9
reviewed
Sep 7, 2026
|
|
||
| #include "mlir/Dialect/Arith/IR/Arith.h" | ||
| #include "mlir/IR/BuiltinOps.h" | ||
| #include "tle/dialect/include/IR/Dialect.h" |
Collaborator
There was a problem hiding this comment.
commonIR 为什么需要 depend on tle ir ?
LeBlanc9
reviewed
Sep 7, 2026
| Type ptrType = getLocalPointerType(valueType, memDescType.getElementType()); | ||
| SmallVector<Value> indices = | ||
| createFullIndices(builder, loc, memDescType.getShape()); | ||
| return builder.create<tle::LocalPointersOp>(loc, ptrType, memDesc, indices); |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
本次更新
本 PR 在
triton_v3.6.x@3f863acd7基础上更新 GPU CommonIR POC,并将 FLIR 依赖固定到 GitHubkateyijian/flir的main@a884fd3c608db2203f08ec5a281327f5dc0153b0。FLIR 仓库本身不再修改,也不需要配套 FLIR 分支。CommonIR目录、Dialect 与 CMake target 名称。tle.gpu.* -> tile.alloc/copy/subview/to_tensor/store_tensor -> CommonIRToTTGIR -> ttg.local_alloc/local_load/local_store/memdesc_index。tle.gpu.local_ptr复用现有tle.local_pointers,以builtin.unrealized_conversion_cast作为临时!tile.buf -> !ttg.memdesc桥接;该桥接在CommonIRToTTGIR中解析并清除,不残留到 TTGIR。ttg.local_alloc,也没有修改 FLIR/community 已有tile.copy等接口。依赖
triton_v3.6.x@3f863acd700bda89514cf3dce6c57020d593d8f3main@a884fd3c608db2203f08ec5a281327f5dc0153b0(已在 FlagTree 下载配置中固定)59d3f12ef验证结果
gpu-tileir-to-ttgir.mlirlit/FileCheck 通过。tle_gpu_primitives.py的 TileIR/TTGIR dump 通过。native_matmul_gpu.py、native_flash_gpu.py均精度 PASS。ttg.local_alloc;TTGIR 阶段不存在残留tile.alloc/copy/subview/to_tensor/store_tensor或桥接 cast。-0.60%、+2.63%;归一化路径后的最终 PTX 分别 76/76、4/4 完全一致。pre-commit --all-files本地复跑全部通过。Wheel
已按最终提交重新生成并在独立
ftree环境强制安装验证:flagtree-0.6.0+git59d3f12e-cp311-cp311-linux_x86_64.whld0a42394e468775cc53b4304b9232cd2cc70d509087d51dc2bc32a3ea05bd324is_common_ir_enabled() == True,matmul/flash 精度及 TileIR/TTGIR 流程检查通过。