core/stdarch/crates/core_arch/src/nvptx/mod.rs
1//! NVPTX intrinsics (experimental)
2//!
3//! These intrinsics form the foundation of the CUDA
4//! programming model.
5//!
6//! The reference is the [CUDA C Programming Guide][cuda_c]. Relevant is also
7//! the [LLVM NVPTX Backend documentation][llvm_docs].
8//!
9//! [cuda_c]:
10//! http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html
11//! [llvm_docs]:
12//! https://llvm.org/docs/NVPTXUsage.html
13
14use crate::ffi::c_void;
15
16mod packed;
17
18#[unstable(feature = "stdarch_nvptx", issue = "111199")]
19pub use packed::*;
20
21#[allow(improper_ctypes)]
22unsafe extern "C" {
23 #[link_name = "llvm.nvvm.barrier0"]
24 fn syncthreads() -> ();
25 #[link_name = "llvm.nvvm.read.ptx.sreg.ntid.x"]
26 fn block_dim_x() -> i32;
27 #[link_name = "llvm.nvvm.read.ptx.sreg.ntid.y"]
28 fn block_dim_y() -> i32;
29 #[link_name = "llvm.nvvm.read.ptx.sreg.ntid.z"]
30 fn block_dim_z() -> i32;
31 #[link_name = "llvm.nvvm.read.ptx.sreg.ctaid.x"]
32 fn block_idx_x() -> i32;
33 #[link_name = "llvm.nvvm.read.ptx.sreg.ctaid.y"]
34 fn block_idx_y() -> i32;
35 #[link_name = "llvm.nvvm.read.ptx.sreg.ctaid.z"]
36 fn block_idx_z() -> i32;
37 #[link_name = "llvm.nvvm.read.ptx.sreg.nctaid.x"]
38 fn grid_dim_x() -> i32;
39 #[link_name = "llvm.nvvm.read.ptx.sreg.nctaid.y"]
40 fn grid_dim_y() -> i32;
41 #[link_name = "llvm.nvvm.read.ptx.sreg.nctaid.z"]
42 fn grid_dim_z() -> i32;
43 #[link_name = "llvm.nvvm.read.ptx.sreg.tid.x"]
44 fn thread_idx_x() -> i32;
45 #[link_name = "llvm.nvvm.read.ptx.sreg.tid.y"]
46 fn thread_idx_y() -> i32;
47 #[link_name = "llvm.nvvm.read.ptx.sreg.tid.z"]
48 fn thread_idx_z() -> i32;
49}
50
51/// Synchronizes all threads in the block.
52#[inline]
53#[unstable(feature = "stdarch_nvptx", issue = "111199")]
54pub unsafe fn _syncthreads() -> () {
55 syncthreads()
56}
57
58/// x-th thread-block dimension.
59#[inline]
60#[unstable(feature = "stdarch_nvptx", issue = "111199")]
61pub unsafe fn _block_dim_x() -> i32 {
62 block_dim_x()
63}
64
65/// y-th thread-block dimension.
66#[inline]
67#[unstable(feature = "stdarch_nvptx", issue = "111199")]
68pub unsafe fn _block_dim_y() -> i32 {
69 block_dim_y()
70}
71
72/// z-th thread-block dimension.
73#[inline]
74#[unstable(feature = "stdarch_nvptx", issue = "111199")]
75pub unsafe fn _block_dim_z() -> i32 {
76 block_dim_z()
77}
78
79/// x-th thread-block index.
80#[inline]
81#[unstable(feature = "stdarch_nvptx", issue = "111199")]
82pub unsafe fn _block_idx_x() -> i32 {
83 block_idx_x()
84}
85
86/// y-th thread-block index.
87#[inline]
88#[unstable(feature = "stdarch_nvptx", issue = "111199")]
89pub unsafe fn _block_idx_y() -> i32 {
90 block_idx_y()
91}
92
93/// z-th thread-block index.
94#[inline]
95#[unstable(feature = "stdarch_nvptx", issue = "111199")]
96pub unsafe fn _block_idx_z() -> i32 {
97 block_idx_z()
98}
99
100/// x-th block-grid dimension.
101#[inline]
102#[unstable(feature = "stdarch_nvptx", issue = "111199")]
103pub unsafe fn _grid_dim_x() -> i32 {
104 grid_dim_x()
105}
106
107/// y-th block-grid dimension.
108#[inline]
109#[unstable(feature = "stdarch_nvptx", issue = "111199")]
110pub unsafe fn _grid_dim_y() -> i32 {
111 grid_dim_y()
112}
113
114/// z-th block-grid dimension.
115#[inline]
116#[unstable(feature = "stdarch_nvptx", issue = "111199")]
117pub unsafe fn _grid_dim_z() -> i32 {
118 grid_dim_z()
119}
120
121/// x-th thread index.
122#[inline]
123#[unstable(feature = "stdarch_nvptx", issue = "111199")]
124pub unsafe fn _thread_idx_x() -> i32 {
125 thread_idx_x()
126}
127
128/// y-th thread index.
129#[inline]
130#[unstable(feature = "stdarch_nvptx", issue = "111199")]
131pub unsafe fn _thread_idx_y() -> i32 {
132 thread_idx_y()
133}
134
135/// z-th thread index.
136#[inline]
137#[unstable(feature = "stdarch_nvptx", issue = "111199")]
138pub unsafe fn _thread_idx_z() -> i32 {
139 thread_idx_z()
140}
141
142/// Generates the trap instruction `TRAP`
143#[inline]
144#[unstable(feature = "stdarch_nvptx", issue = "111199")]
145pub unsafe fn trap() -> ! {
146 crate::intrinsics::abort()
147}
148
149// Basic CUDA syscall declarations.
150unsafe extern "C" {
151 /// Print formatted output from a kernel to a host-side output stream.
152 ///
153 /// Syscall arguments:
154 /// * `status`: The status value that is returned by `vprintf`.
155 /// * `format`: A pointer to the format specifier input (uses common `printf` format).
156 /// * `valist`: A pointer to the valist input.
157 ///
158 /// ```
159 /// #[repr(C)]
160 /// struct PrintArgs(f32, f32, f32, i32);
161 ///
162 /// vprintf(
163 /// "int(%f + %f) = int(%f) = %d\n".as_ptr(),
164 /// transmute(&PrintArgs(a, b, a + b, (a + b) as i32)),
165 /// );
166 /// ```
167 ///
168 /// Sources:
169 /// [Programming Guide](https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#formatted-output),
170 /// [PTX Interoperability](https://docs.nvidia.com/cuda/ptx-writers-guide-to-interoperability/index.html#system-calls).
171 #[unstable(feature = "stdarch_nvptx", issue = "111199")]
172 pub fn vprintf(format: *const u8, valist: *const c_void) -> i32;
173
174 /// Allocate memory dynamically from a fixed-size heap in global memory.
175 ///
176 /// The CUDA in-kernel `malloc()` function allocates at least `size` bytes
177 /// from the device heap and returns a pointer to the allocated memory
178 /// or `NULL` if insufficient memory exists to fulfill the request.
179 ///
180 /// The returned pointer is guaranteed to be aligned to a 16-byte boundary.
181 ///
182 /// The memory allocated by a given CUDA thread via `malloc()` remains allocated
183 /// for the lifetime of the CUDA context, or until it is explicitly released
184 /// by a call to `free()`. It can be used by any other CUDA threads
185 /// even from subsequent kernel launches.
186 ///
187 /// Sources:
188 /// [Programming Guide](https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#dynamic-global-memory-allocation-and-operations),
189 /// [PTX Interoperability](https://docs.nvidia.com/cuda/ptx-writers-guide-to-interoperability/index.html#system-calls).
190 // FIXME(denzp): assign `malloc` and `nothrow` attributes.
191 #[unstable(feature = "stdarch_nvptx", issue = "111199")]
192 pub fn malloc(size: usize) -> *mut c_void;
193
194 /// Free previously dynamically allocated memory.
195 ///
196 /// The CUDA in-kernel `free()` function deallocates the memory pointed to by `ptr`,
197 /// which must have been returned by a previous call to `malloc()`. If `ptr` is NULL,
198 /// the call to `free()` is ignored.
199 ///
200 /// Any CUDA thread may free memory allocated by another thread, but care should be taken
201 /// to ensure that the same pointer is not freed more than once. Repeated calls to `free()`
202 /// with the same `ptr` has undefined behavior.
203 ///
204 /// Sources:
205 /// [Programming Guide](https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#dynamic-global-memory-allocation-and-operations),
206 /// [PTX Interoperability](https://docs.nvidia.com/cuda/ptx-writers-guide-to-interoperability/index.html#system-calls).
207 // FIXME(denzp): assign `nothrow` attribute.
208 #[unstable(feature = "stdarch_nvptx", issue = "111199")]
209 pub fn free(ptr: *mut c_void);
210
211 // Internal declaration of the syscall. Exported variant has
212 // the `char_size` parameter set to `1` (single char size in bytes).
213 fn __assertfail(
214 message: *const u8,
215 file: *const u8,
216 line: u32,
217 function: *const u8,
218 char_size: usize,
219 );
220}
221
222/// Syscall to be used whenever the *assert expression produces a `false` value*.
223///
224/// Syscall arguments:
225/// * `message`: The pointer to the string that should be output.
226/// * `file`: The pointer to the file name string associated with the assert.
227/// * `line`: The line number associated with the assert.
228/// * `function`: The pointer to the function name string associated with the assert.
229///
230/// Source:
231/// [PTX Interoperability](https://docs.nvidia.com/cuda/ptx-writers-guide-to-interoperability/index.html#system-calls).
232#[inline]
233#[unstable(feature = "stdarch_nvptx", issue = "111199")]
234pub unsafe fn __assert_fail(message: *const u8, file: *const u8, line: u32, function: *const u8) {
235 __assertfail(message, file, line, function, 1)
236}