Repository navigation

Implement CUDA support and GPU operations for tensor processing - #8

Merged

Alwaysproblem merged 34 commits into

mainfrom

cuda-tile

Mar 8, 2026

Merged

Implement CUDA support and GPU operations for tensor processing#8
Alwaysproblem merged 34 commits into
mainfrom
cuda-tile

Commits on Dec 29, 2025
- Inital commits
  Alwaysproblem
  committed
  3679d7a
- Make ch6 works
  Alwaysproblem
  committed
  e3204ff
- Added Matmul Toy Op
  Alwaysproblem
  committed
  2b310d3
Commits on Dec 30, 2025
- Added the validation for cuda tile
  Alwaysproblem
  committed
  ca1bd69
Commits on Dec 31, 2025
- Try gpu and failed
  Alwaysproblem
  committed
  afe7997
- update the build dep
  Alwaysproblem
  committed
  88ce121
- Change float type to FP32
  Alwaysproblem
  committed
  8a0ba30
Commits on Jan 1, 2026
- Added the GPU C API
  Alwaysproblem
  committed
  8be2741
- Added the GPU related operation
  Alwaysproblem
  committed
  7e1f45c
Commits on Jan 2, 2026
- Create the gpu outline pass
  Alwaysproblem
  committed
  37800da
- Added Affine pass code
  Alwaysproblem
  committed
  fd70514
Commits on Jan 3, 2026
- Added the EntryOp for cuda tile IR
  Alwaysproblem
  committed
  60b8083
- Sync the command line history
  Alwaysproblem
  committed
  d12717f
- Added devcontainer
  Alwaysproblem
  committed
  bdb9a66
Commits on Jan 5, 2026
- Added the make tensor view
  Alwaysproblem
  committed
  07c722f
Commits on Jan 11, 2026
- Sync
  Alwaysproblem
  committed
  43b2739
Commits on Jan 29, 2026
- Added the return, add, mul lowering
  Alwaysproblem
  committed
  0a3ccf4
Commits on Feb 4, 2026
- Added the pass that can compile cuda tile IR
  Alwaysproblem
  committed
  f2d06f6
Commits on Feb 12, 2026
- Verified the cuda shim API and POC is ready for the cuda shim
  Alwaysproblem
  committed
  99596b0
Commits on Feb 27, 2026
- sync the progress not finish
  Alwaysproblem
  committed
  80e2354
- upload file
  Alwaysproblem
  committed
  034cc87
- sync
  Alwaysproblem
  committed
  5007d48
Commits on Mar 1, 2026
- Sync: added the getglobal memref
  Alwaysproblem
  committed
  b9cfa7e
- Added the input and allocation for the cuda shim
  Alwaysproblem
  committed
  dbf81a7
Commits on Mar 2, 2026
- Tested with cuda 12.x cubin worked
  Alwaysproblem
  committed
  96a122c
- Move the helper function into cuda shim builder
  Alwaysproblem
  committed
  b2e34d0
Commits on Mar 5, 2026
- Tested on the GPU RTX4090 with cuda 12.x
  Alwaysproblem
  committed
  45a5e01
Commits on Mar 7, 2026
- Tested on the RTX4090
  Alwaysproblem
  committed
  8ad69b7
- Update ReadMe
  Alwaysproblem
  committed
  5cdabd4
Commits on Mar 8, 2026
- Tested on RTX5090
  Alwaysproblem
  committed
  1d98f48
- Test on RTX4090 after grid change
  Alwaysproblem
  committed
  d41cea8
- Update mlir/cuda-tile/Toy/mlir/EmitCudaTile.cpp
  
  Alwaysproblem
  and
  Copilot
  authored
  87330ea
- Update mlir/cuda-tile/Toy/mlir/EmitCudaTile.cpp
  
  Alwaysproblem
  and
  Copilot
  authored
  c7b1673
- Fix the comments from github copilot
  Alwaysproblem
  committed
  bcb8038

Commits

Commits on Dec 29, 2025

Commits on Dec 30, 2025

Commits on Dec 31, 2025

Commits on Jan 1, 2026

Commits on Jan 2, 2026

Commits on Jan 3, 2026

Commits on Jan 5, 2026

Commits on Jan 11, 2026

Commits on Jan 29, 2026

Commits on Feb 4, 2026

Commits on Feb 12, 2026

Commits on Feb 27, 2026

Commits on Mar 1, 2026

Commits on Mar 2, 2026

Commits on Mar 5, 2026

Commits on Mar 7, 2026

Commits on Mar 8, 2026