[libc-commits] [libc] 5083885 - [libc][docs] Update NVPTX using documentation now that linking works

[Joseph Huber via libc-commits]

Author: Joseph Huber
Date: 2024-10-04T18:09:28-05:00
New Revision: 50838851e82281635deacd017273947d9259f8fe

URL: https://github.com/llvm/llvm-project/commit/50838851e82281635deacd017273947d9259f8fe
DIFF: https://github.com/llvm/llvm-project/commit/50838851e82281635deacd017273947d9259f8fe.diff

LOG: [libc][docs] Update NVPTX using documentation now that linking works

Summary:
I added a wrapper linker awhile back but this still says it doesn't
work.

Added: 
    

Modified: 
    libc/docs/gpu/using.rst

Removed: 
    


################################################################################
diff  --git a/libc/docs/gpu/using.rst b/libc/docs/gpu/using.rst
index 4034c04867c99a..e56b6f634bb31e 100644
--- a/libc/docs/gpu/using.rst
+++ b/libc/docs/gpu/using.rst
@@ -34,10 +34,10 @@ described in the `clang documentation
 by the OpenMP toolchain, but is currently opt-in for the CUDA and HIP toolchains
 through the ``--offload-new-driver``` and ``-fgpu-rdc`` flags.
 
-In order or link the GPU runtime, we simply pass this library to the embedded 
-device linker job. This can be done using the ``-Xoffload-linker`` option, which 
-forwards an argument to a ``clang`` job used to create the final GPU executable. 
-The toolchain should pick up the C libraries automatically in most cases, so 
+In order or link the GPU runtime, we simply pass this library to the embedded
+device linker job. This can be done using the ``-Xoffload-linker`` option, which
+forwards an argument to a ``clang`` job used to create the final GPU executable.
+The toolchain should pick up the C libraries automatically in most cases, so
 this shouldn't be necessary.
 
 .. code-block:: sh
@@ -189,7 +189,7 @@ final executable.
 
   #include <stdio.h>
 
-  int main() { fputs("Hello from AMDGPU!\n", stdout); }
+  int main() { printf("Hello from AMDGPU!\n"); }
 
 This program can then be compiled using the ``clang`` compiler. Note that
 ``-flto`` and ``-mcpu=`` should be defined. This is because the GPU
@@ -227,28 +227,26 @@ Building for NVPTX targets
 ^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 The infrastructure is the same as the AMDGPU example. However, the NVPTX binary
-utilities are very limited and must be targeted directly. There is no linker
-support for static libraries so we need to link in the ``libc.bc`` bitcode and
-inform the compiler driver of the file's contents.
+utilities are very limited and must be targeted directly. A utility called
+``clang-nvlink-wrapper`` instead wraps around the standard link job to give the
+illusion that ``nvlink`` is a functional linker.
 
 .. code-block:: c++
 
   #include <stdio.h>
 
   int main(int argc, char **argv, char **envp) {
-    fputs("Hello from NVPTX!\n", stdout);
+    printf("Hello from NVPTX!\n");
   }
 
 Additionally, the NVPTX ABI requires that every function signature matches. This
 requires us to pass the full prototype from ``main``. The installation will
 contain the ``nvptx-loader`` utility if the CUDA driver was found during
-compilation.
+compilation. Using link time optimization will help hide this.
 
 .. code-block:: sh
 
-  $> clang hello.c --target=nvptx64-nvidia-cuda -march=native \
-       -x ir <install>/lib/nvptx64-nvidia-cuda/libc.bc \
-       -x ir <install>/lib/nvptx64-nvidia-cuda/crt1.o
+  $> clang hello.c --target=nvptx64-nvidia-cuda -mcpu=native -flto -lc <install>/lib/nvptx64-nvidia-cuda/crt1.o
   $> nvptx-loader --threads 2 --blocks 2 a.out
   Hello from NVPTX!
   Hello from NVPTX!