Wafer logo
Back to all blogs

Cloud Compiler Analyzer (PTX/SASS) Inside Your IDE

Cloud CUDA compilation with PTX/SASS output, PyTorch headers, and VS Code integration. No local CUDA install required.

February 3, 2026·Wafer Team
CUDA Compiler Explorer in VS Code

TLDR

  • Launching our Cloud Compiler: Analyze CUDA remotely to show PTX/SASS directly from VS Code or Cursor
  • No local CUDA toolkit or GPU required
  • PyTorch and CUTLASS headers included
  • 12+ GPU architectures supported (sm80 to sm120a)
  • Completely free of use

The Problem

Discord message from gau.nernst asking about cloud CUDA compilation

As @gau.nernst mentioned, if you're doing kernel optimization, you need to see what the compiler actually produced. PTX shows you the intermediate representation. SASS shows you the machine instructions hitting the GPU.

But getting there is annoying. Godbolt is great for quick experiments, but it doesn't have PyTorch or CUTLASS headers. Setting up a local CUDA toolkit just to inspect assembly is overkill. And copying code back and forth between your editor and a browser breaks your flow.

You want to change a line, see the assembly, iterate. Fast.

What We Built

Wafer now includes cloud CUDA compilation. You can write CUDA in your editor, do Command + Shift + P, and type: "Compile CUDA" and get PTX and SASS back. No local CUDA install required. No GPU required either. Completely free.

The compiler button appears in your editor title bar when you're editing a .cu file. Click it, pick your target architecture, and see the output. That's it.

We support 12+ GPU architectures: sm80 (Ampere) through sm120a (Blackwell). Whatever you're targeting, we've got it.

Built-in Headers

We include PyTorch headers and CUTLASS by default.

You can #include or pull in CUTLASS 4.3.5 and compile against them directly.

Custom nvcc flags work too. Want -O3? --maxrregcount=32? -lineinfo for debugging? Pass them through.

Try Out Today

Install the Wafer extension and open any .cu file. To use it, either do Command + Shift + P and then type "Compile CUDA", with any CUDA file open, or click the three dots "..." at the top of the editor window and click the first option of Compile CUDA there.

Give Us Feedback

If there's something else that would make your kernel development faster, let us know.

Reach out at emilio@wafer.ai or find us on Twitter/X.