diff options
| author | Matthew Maurer <mmaurer@google.com> | 2025-09-02 18:43:48 +0000 |
|---|---|---|
| committer | Matthew Maurer <mmaurer@google.com> | 2025-09-02 18:47:59 +0000 |
| commit | 5d9f8fcd3ea44bdd633ecddac2aff471ffcb797f (patch) | |
| tree | b81c9920be537e7957f9eb7ae394e11aee262175 | |
| parent | a2c8b0b92c14b02f0b3f96a0d5296f1090dc286b (diff) | |
| download | rust-5d9f8fcd3ea44bdd633ecddac2aff471ffcb797f.tar.gz rust-5d9f8fcd3ea44bdd633ecddac2aff471ffcb797f.zip | |
llvm: nvptx: Layout update to match LLVM
LLVM upstream switched layouts to support 256-bit vector load/store.
| -rw-r--r-- | compiler/rustc_codegen_llvm/src/context.rs | 4 | ||||
| -rw-r--r-- | compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs | 2 |
2 files changed, 5 insertions, 1 deletions
diff --git a/compiler/rustc_codegen_llvm/src/context.rs b/compiler/rustc_codegen_llvm/src/context.rs index 4fd6110ac4a..257c7b95666 100644 --- a/compiler/rustc_codegen_llvm/src/context.rs +++ b/compiler/rustc_codegen_llvm/src/context.rs @@ -217,6 +217,10 @@ pub(crate) unsafe fn create_module<'ll>( // LLVM 22.0 updated the default layout on avr: https://github.com/llvm/llvm-project/pull/153010 target_data_layout = target_data_layout.replace("n8:16", "n8") } + if sess.target.arch == "nvptx64" { + // LLVM 22 updated the NVPTX layout to indicate 256-bit vector load/store: https://github.com/llvm/llvm-project/pull/155198 + target_data_layout = target_data_layout.replace("-i256:256", ""); + } } // Ensure the data-layout values hardcoded remain the defaults. diff --git a/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs b/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs index 598f0f19f0d..cada0dd640a 100644 --- a/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs +++ b/compiler/rustc_target/src/spec/targets/nvptx64_nvidia_cuda.rs @@ -6,7 +6,7 @@ use crate::spec::{ pub(crate) fn target() -> Target { Target { arch: "nvptx64".into(), - data_layout: "e-p6:32:32-i64:64-i128:128-v16:16-v32:32-n16:32:64".into(), + data_layout: "e-p6:32:32-i64:64-i128:128-i256:256-v16:16-v32:32-n16:32:64".into(), llvm_target: "nvptx64-nvidia-cuda".into(), metadata: TargetMetadata { description: Some("--emit=asm generates PTX code that runs on NVIDIA GPUs".into()), |