add nvptx architecture

library/stdarch/src/lib.rs

@@ -153,6 +153,8 @@ pub mod vendor {
 
     #[cfg(target_arch = "aarch64")]
     pub use aarch64::*;
+
+    pub use nvptx::*;
 }
 
 #[macro_use]
@@ -205,3 +207,5 @@ mod x86;
 mod arm;
 #[cfg(target_arch = "aarch64")]
 mod aarch64;
+
+mod nvptx;

library/stdarch/src/nvptx/mod.rs

@@ -0,0 +1,109 @@
+//! nvptx intrinsics
+
+#[allow(improper_ctypes)]
+extern "C" {
+    #[link_name = "llvm.cuda.syncthreads"]
+    fn syncthreads() -> ();
+    #[link_name = "llvm.nvvm.read.ptx.sreg.ntid.x"]
+    fn block_dim_x() -> i32;
+    #[link_name = "llvm.nvvm.read.ptx.sreg.ntid.y"]
+    fn block_dim_y() -> i32;
+    #[link_name = "llvm.nvvm.read.ptx.sreg.ntid.z"]
+    fn block_dim_z() -> i32;
+    #[link_name = "llvm.nvvm.read.ptx.sreg.ctaid.x"]
+    fn block_idx_x() -> i32;
+    #[link_name = "llvm.nvvm.read.ptx.sreg.ctaid.y"]
+    fn block_idx_y() -> i32;
+    #[link_name = "llvm.nvvm.read.ptx.sreg.ctaid.z"]
+    fn block_idx_z() -> i32;
+    #[link_name = "llvm.nvvm.read.ptx.sreg.nctaid.x"]
+    fn grid_dim_x() -> i32;
+    #[link_name = "llvm.nvvm.read.ptx.sreg.nctaid.y"]
+    fn grid_dim_y() -> i32;
+    #[link_name = "llvm.nvvm.read.ptx.sreg.nctaid.z"]
+    fn grid_dim_z() -> i32;
+    #[link_name = "llvm.nvvm.read.ptx.sreg.tid.x"]
+    fn thread_idx_x() -> i32;
+    #[link_name = "llvm.nvvm.read.ptx.sreg.tid.y"]
+    fn thread_idx_y() -> i32;
+    #[link_name = "llvm.nvvm.read.ptx.sreg.tid.z"]
+    fn thread_idx_z() -> i32;
+}
+
+/// Synchronizes all threads in the block.
+#[inline(always)]
+pub unsafe fn _syncthreads() -> () {
+    syncthreads()
+}
+
+/// x-th thread-block dimension.
+#[inline(always)]
+pub unsafe fn _block_dim_x() -> i32 {
+    block_dim_x()
+}
+
+/// y-th thread-block dimension.
+#[inline(always)]
+pub unsafe fn _block_dim_y() -> i32 {
+    block_dim_y()
+}
+
+/// z-th thread-block dimension.
+#[inline(always)]
+pub unsafe fn _block_dim_z() -> i32 {
+    block_dim_z()
+}
+
+/// x-th thread-block index.
+#[inline(always)]
+pub unsafe fn _block_idx_x() -> i32 {
+    block_idx_x()
+}
+
+/// y-th thread-block index.
+#[inline(always)]
+pub unsafe fn _block_idx_y() -> i32 {
+    block_idx_y()
+}
+
+/// z-th thread-block index.
+#[inline(always)]
+pub unsafe fn _block_idx_z() -> i32 {
+    block_idx_z()
+}
+
+/// x-th block-grid dimension.
+#[inline(always)]
+pub unsafe fn _grid_dim_x() -> i32 {
+    grid_dim_x()
+}
+
+/// y-th block-grid dimension.
+#[inline(always)]
+pub unsafe fn _grid_dim_y() -> i32 {
+    grid_dim_y()
+}
+
+/// z-th block-grid dimension.
+#[inline(always)]
+pub unsafe fn _grid_dim_z() -> i32 {
+    grid_dim_z()
+}
+
+/// x-th thread index.
+#[inline(always)]
+pub unsafe fn _thread_idx_x() -> i32 {
+    thread_idx_x()
+}
+
+/// y-th thread index.
+#[inline(always)]
+pub unsafe fn _thread_idx_y() -> i32 {
+    thread_idx_y()
+}
+
+/// z-th thread index.
+#[inline(always)]
+pub unsafe fn _thread_idx_z() -> i32 {
+    thread_idx_z()
+}