Merge branch 'master' of github.com:cwpearson/triad-gpu-bandwidth

cwpearson · Jun 13, 2019 · ce44219 · ce44219
2 parents 739a7cc + 42f84b2
commit ce44219
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/README.md b/README.md
@@ -74,13 +74,13 @@ Depending on your installed CUDA, you'll need a different version of clang.
 | CUDA | Clang | Installer |
 |-|-|-|
 | 9.2  | 5.0.0 | https://gist.github.com/cwpearson/c5521dfc50175b1d977643b2fc5a2bb1 |
-| 10.1 | 8.0.0 | https://gist.github.com/cwpearson/fc91b92c3d49d75a1b3a559aacb1d38e |
+| 10.1 | 5.0.0 | https://gist.github.com/cwpearson/c13ac7c25bde8c8644300e211faf4e78 |
 
 Add the clang to your path, and have CMake use clang in the build.
 
 ```
 mkdir build && cd build
 cmake .. -DCMAKE_TOOLCHAIN_FILE=`readlink -f ../toolchains/clang.toolchain`
 ```
-
+The CUDA documentation claims that clang 8.0.0 is supported for CUDA 10.1, but if you actually try it says it requires clang>=3.2 and clang<8. Clang 7.1.0 fails on CUDA 10.0.0 with some errors about `__fp16`.