support sm_100 and llvm v19

Author: brandonros

Cargo.toml

@@ -22,5 +22,8 @@ exclude = [
     "crates/optix/examples/common",
 ]
 
-[profile.dev.package.rustc_codegen_nvvm]
+[profile.dev.package.rustc_codegen_nvvm_v7]
+opt-level = 3
+
+[profile.dev.package.rustc_codegen_nvvm_v19]
 opt-level = 3

container/ubuntu22-cuda12/Dockerfile

@@ -1,4 +1,4 @@
-FROM nvidia/cuda:12.8.1-cudnn-devel-ubuntu22.04
+FROM nvidia/cuda:12.9.0-cudnn-devel-ubuntu22.04
 
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get -qq -y install \
     build-essential \

container/ubuntu24-cuda12/Dockerfile

@@ -1,4 +1,4 @@
-FROM nvidia/cuda:12.8.1-cudnn-devel-ubuntu24.04
+FROM nvidia/cuda:12.9.0-cudnn-devel-ubuntu24.04
 
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get -qq -y install \
     build-essential \

crates/cuda_builder/Cargo.toml

@@ -9,7 +9,8 @@ repository = "https://github.com/Rust-GPU/Rust-CUDA"
 readme = "../../README.md"
 
 [dependencies]
-rustc_codegen_nvvm = { version = "0.3", path = "../rustc_codegen_nvvm" }
+rustc_codegen_nvvm_v7 = { version = "0.3", path = "../rustc_codegen_nvvm_v7" }
+rustc_codegen_nvvm_v19 = { version = "0.3", path = "../rustc_codegen_nvvm_v19" }
 nvvm = { path = "../nvvm", version = "0.1" }
 serde = { version = "1.0.217", features = ["derive"] }
 serde_json = "1.0.138"

crates/cuda_std/src/cfg.rs

@@ -16,6 +16,11 @@ pub enum ComputeCapability {
     Compute72,
     Compute75,
     Compute80,
+    Compute86,
+    Compute87,
+    Compute89,
+    Compute90,
+    Compute100
 }
 
 impl ComputeCapability {
@@ -42,6 +47,11 @@ impl ComputeCapability {
             "720" => ComputeCapability::Compute72,
             "750" => ComputeCapability::Compute75,
             "800" => ComputeCapability::Compute80,
+            "860" => ComputeCapability::Compute86,  // Ampere (RTX 30 series, A100)
+            "870" => ComputeCapability::Compute87,  // Ampere (Jetson AGX Orin)
+            "890" => ComputeCapability::Compute89,  // Ada Lovelace (RTX 40 series)
+            "900" => ComputeCapability::Compute90,  // Hopper (H100)
+            "1000" => ComputeCapability::Compute100, // Blackwell (RTX 50 series, H200, B100)
             _ => panic!("CUDA_ARCH had an invalid value"),
         }
     }

crates/cust/src/module.rs

@@ -56,6 +56,10 @@ pub enum JitTarget {
     Compute75 = 75,
     Compute80 = 80,
     Compute86 = 86,
+    Compute87 = 87,
+    Compute89 = 89,
+    Compute90 = 90,
+    Compute100 = 100,
 }
 
 /// How to handle cases where a loaded module's data does not contain an exact match for the

crates/nvvm/src/lib.rs

@@ -4,7 +4,6 @@ use std::{
     ffi::{CStr, CString},
     fmt::Display,
     mem::MaybeUninit,
-    ptr::null_mut,
     str::FromStr,
 };
 
@@ -255,6 +254,11 @@ impl FromStr for NvvmOption {
                     "72" => NvvmArch::Compute72,
                     "75" => NvvmArch::Compute75,
                     "80" => NvvmArch::Compute80,
+                    "86" => NvvmArch::Compute86,
+                    "87" => NvvmArch::Compute87,
+                    "89" => NvvmArch::Compute89,
+                    "90" => NvvmArch::Compute90,
+                    "100" => NvvmArch::Compute100,
                     _ => return Err("unknown arch"),
                 };
                 Self::Arch(arch)
@@ -279,6 +283,11 @@ pub enum NvvmArch {
     Compute72,
     Compute75,
     Compute80,
+    Compute86,
+    Compute87,
+    Compute89,
+    Compute90,
+    Compute100,
 }
 
 impl Display for NvvmArch {
@@ -403,8 +412,21 @@ impl NvvmProgram {
 
     /// Verify the program without actually compiling it. In the case of invalid IR, you can find
     /// more detailed error info by calling [`compiler_log`](Self::compiler_log).
-    pub fn verify(&self) -> Result<(), NvvmError> {
-        unsafe { nvvm_sys::nvvmVerifyProgram(self.raw, 0, null_mut()).to_result() }
+    pub fn verify(&self, options: &[NvvmOption]) -> Result<(), NvvmError> {
+        let option_strings: Vec<_> = options.iter().map(|opt| opt.to_string()).collect();
+        let option_cstrings: Vec<_> = option_strings.iter()
+            .map(|s| std::ffi::CString::new(s.as_str()).unwrap())
+            .collect();
+        let mut option_ptrs: Vec<_> = option_cstrings.iter()
+            .map(|cs| cs.as_ptr())
+            .collect();
+        unsafe { 
+            nvvm_sys::nvvmVerifyProgram(
+                self.raw, 
+                option_ptrs.len() as i32, 
+                option_ptrs.as_mut_ptr()
+            ).to_result() 
+        }
     }
 }
 
@@ -433,6 +455,11 @@ mod tests {
             "-arch=compute_72",
             "-arch=compute_75",
             "-arch=compute_80",
+            "-arch=compute_86",
+            "-arch=compute_87",
+            "-arch=compute_89",
+            "-arch=compute_90",
+            "-arch=compute_100",
             "-ftz=1",
             "-prec-sqrt=0",
             "-prec-div=0",
@@ -454,6 +481,11 @@ mod tests {
             Arch(Compute72),
             Arch(Compute75),
             Arch(Compute80),
+            Arch(Compute86),
+            Arch(Compute87),
+            Arch(Compute89),
+            Arch(Compute90),
+            Arch(Compute100),
             Ftz,
             FastSqrt,
             FastDiv,

crates/rustc_codegen_nvvm/CHANGELOG.md renamed to crates/rustc_codegen_nvvm_v19/CHANGELOG.md

@@ -0,0 +1,42 @@
+[package]
+name = "rustc_codegen_nvvm_v19"
+version = "0.3.0"
+authors = [
+    "Riccardo D'Ambrosio <[email protected]>",
+    "The Rust Project Developers",
+]
+edition = "2024"
+license = "MIT OR Apache-2.0"
+description = "A codegen backend for Rustc which targets the libnvvm CUDA library"
+repository = "https://github.com/Rust-GPU/Rust-CUDA"
+readme = "../../README.md"
+
+[lib]
+crate-type = ["dylib"]
+
+[dependencies]
+nvvm = { version = "0.1", path = "../nvvm" }
+rustc-demangle = "0.1.24"
+libc = "0.2.169"
+libloading = "0.8.0"
+tar = "0.4.43"
+object = "0.36.7"
+bitflags = "2.8.0"
+# To avoid duplicate dependencies, this should match the version of gimli used
+# by `rustc_codegen_ssa` via its `thorin-dwp` dependency.
+gimli = "0.30"
+tracing = { version = "0.1.41", features = ["release_max_level_debug"] }
+tracing-subscriber = { version = "0.3.19", features = ["env-filter"] }
+rustc_codegen_nvvm_macros = { version = "0.1", path = "../rustc_codegen_nvvm_macros" }
+smallvec = { version = "1.14.0", features = ["union", "may_dangle"] }
+itertools = "0.14.0"
+
+[build-dependencies]
+build-helper = "0.1.1"
+cc = { version = "1.0", features = ["parallel"] }
+xz = "0.1.0"
+tar = "0.4.37"
+curl = "0.4.40"
+
+[package.metadata.rust-analyzer]
+rustc_private = true