# Examples

To build the examples:
``` bash
# Make sure `rust-ptx-linker` is in the $PATH
$ which `rust-ptx-linker`
$ cd example_dir
$ cargo build --target nvptx64-nvidia-cuda --release
```

## Deep Dependencies Example: `depenencies`
This example shows that the linker is actually "linking" :)

The dependencies hierarchy is:
```
example (is the CUDA kernel dylib crate)
└─ dummy_math (rlib crate)
   └─ dummy_utils (rlib crate)
```

Both `dummy_math` and `dummy_utils` exports functions and also a kernel.

A PTX output at `target/nvptx64-nvidia-cuda/release/example.ptx` should contain no `.extern` function declaration and have 3 kernels.

## Undefined Reference Example: `undefined-ref`
This example shows that the linker can find unresoved external references and reject linking because the output won't be a valid PTX.

When you try to run the example the linker should fail with error message:
```
[ERROR ptx_linker] Unable to link modules
[ERROR ptx_linker]   caused by: Undefined references: ["bar"]
```

## Intrinsics Example: `intrinsics`
This example shows a way to use NVPTX intrinsics with a helper lib `japaric/nvptx-builtins`.

A PTX output at `target/nvptx64-nvidia-cuda/release/example.ptx` should use special PTX registers:
```
mov.u32 %r_, %ntid.x;
mov.u32 %r_, %ctaid.x;
mov.u32 %r_, %tid.x;

mov.u32 %r_, %ntid.y;
mov.u32 %r_, %ctaid.y;
mov.u32 %r_, %tid.y;
```