Fix issues and warnings

crates/blastoff/src/level1.rs

@@ -46,7 +46,7 @@ impl CublasContext {
             Ok(T::amin(
                 ctx.raw,
                 n as i32,
-                x.as_device_ptr().as_raw(),
+                x.as_device_ptr().as_ptr(),
                 stride.unwrap_or(1) as i32,
                 result.as_device_ptr().as_mut_ptr(),
             )
@@ -108,7 +108,7 @@ impl CublasContext {
             Ok(T::amax(
                 ctx.raw,
                 n as i32,
-                x.as_device_ptr().as_raw(),
+                x.as_device_ptr().as_ptr(),
                 stride.unwrap_or(1) as i32,
                 result.as_device_ptr().as_mut_ptr(),
             )
@@ -172,10 +172,10 @@ impl CublasContext {
             Ok(T::axpy(
                 ctx.raw,
                 n as i32,
-                alpha.as_device_ptr().as_raw(),
-                x.as_device_ptr().as_raw(),
+                alpha.as_device_ptr().as_ptr(),
+                x.as_device_ptr().as_ptr(),
                 x_stride.unwrap_or(1) as i32,
-                y.as_device_ptr().as_raw_mut(),
+                y.as_device_ptr().as_mut_ptr(),
                 y_stride.unwrap_or(1) as i32,
             )
             .to_result()?)
@@ -245,9 +245,9 @@ impl CublasContext {
             Ok(T::copy(
                 ctx.raw,
                 n as i32,
-                x.as_device_ptr().as_raw(),
+                x.as_device_ptr().as_ptr(),
                 x_stride.unwrap_or(1) as i32,
-                y.as_device_ptr().as_raw_mut(),
+                y.as_device_ptr().as_mut_ptr(),
                 y_stride.unwrap_or(1) as i32,
             )
             .to_result()?)
@@ -314,11 +314,11 @@ impl CublasContext {
             Ok(T::dot(
                 ctx.raw,
                 n as i32,
-                x.as_device_ptr().as_raw(),
+                x.as_device_ptr().as_ptr(),
                 x_stride.unwrap_or(1) as i32,
-                y.as_device_ptr().as_raw(),
+                y.as_device_ptr().as_ptr(),
                 y_stride.unwrap_or(1) as i32,
-                result.as_device_ptr().as_raw_mut(),
+                result.as_device_ptr().as_mut_ptr(),
             )
             .to_result()?)
         })
@@ -390,11 +390,11 @@ impl CublasContext {
             Ok(T::dotu(
                 ctx.raw,
                 n as i32,
-                x.as_device_ptr().as_raw(),
+                x.as_device_ptr().as_ptr(),
                 x_stride.unwrap_or(1) as i32,
-                y.as_device_ptr().as_raw(),
+                y.as_device_ptr().as_ptr(),
                 y_stride.unwrap_or(1) as i32,
-                result.as_device_ptr().as_raw_mut(),
+                result.as_device_ptr().as_mut_ptr(),
             )
             .to_result()?)
         })
@@ -438,11 +438,11 @@ impl CublasContext {
             Ok(T::dotc(
                 ctx.raw,
                 n as i32,
-                x.as_device_ptr().as_raw(),
+                x.as_device_ptr().as_ptr(),
                 x_stride.unwrap_or(1) as i32,
-                y.as_device_ptr().as_raw(),
+                y.as_device_ptr().as_ptr(),
                 y_stride.unwrap_or(1) as i32,
-                result.as_device_ptr().as_raw_mut(),
+                result.as_device_ptr().as_mut_ptr(),
             )
             .to_result()?)
         })
@@ -483,9 +483,9 @@ impl CublasContext {
             Ok(T::nrm2(
                 ctx.raw,
                 n as i32,
-                x.as_device_ptr().as_raw(),
+                x.as_device_ptr().as_ptr(),
                 x_stride.unwrap_or(1) as i32,
-                result.as_device_ptr().as_raw_mut(),
+                result.as_device_ptr().as_mut_ptr(),
             )
             .to_result()?)
         })
@@ -559,12 +559,12 @@ impl CublasContext {
             Ok(T::rot(
                 ctx.raw,
                 n as i32,
-                x.as_device_ptr().as_raw_mut(),
+                x.as_device_ptr().as_mut_ptr(),
                 x_stride.unwrap_or(1) as i32,
-                y.as_device_ptr().as_raw_mut(),
+                y.as_device_ptr().as_mut_ptr(),
                 y_stride.unwrap_or(1) as i32,
-                c.as_device_ptr().as_raw(),
-                s.as_device_ptr().as_raw(),
+                c.as_device_ptr().as_ptr(),
+                s.as_device_ptr().as_ptr(),
             )
             .to_result()?)
         })

crates/cudnn/src/context.rs

@@ -397,9 +397,9 @@ impl CudnnContext {
         let x_data = x.data().as_device_ptr().as_raw();
 
         let y_desc = y.descriptor();
-        let y_data = y.data().as_device_ptr().as_raw_mut();
+        let y_data = y.data().as_device_ptr().as_ptr();
 
-        let reserve_space_ptr = reserve_space.as_device_ptr().as_raw_mut();
+        let reserve_space_ptr = reserve_space.as_device_ptr().as_ptr();
 
         unsafe {
             sys::cudnnDropoutForward(
@@ -454,9 +454,9 @@ impl CudnnContext {
         let dy_data = dy.data().as_device_ptr().as_raw();
 
         let dx_desc = dx.descriptor();
-        let dx_data = dx.data().as_device_ptr().as_raw_mut();
+        let dx_data = dx.data().as_device_ptr().as_ptr();
 
-        let reserve_space_ptr = reserve_space.as_device_ptr().as_raw_mut();
+        let reserve_space_ptr = reserve_space.as_device_ptr().as_ptr();
 
         unsafe {
             sys::cudnnDropoutBackward(
@@ -528,7 +528,7 @@ impl CudnnContext {
                 raw,
                 self.raw,
                 dropout,
-                states.as_device_ptr().as_raw_mut() as *mut std::ffi::c_void,
+                states.as_device_ptr().as_ptr() as *mut std::ffi::c_void,
                 states.len(),
                 seed,
             )
@@ -1185,14 +1185,14 @@ impl CudnnContext {
         let w_data = w.data().as_device_ptr().as_raw();
         let w_desc = w.descriptor();
 
-        let y_data = y.data().as_device_ptr().as_raw_mut();
+        let y_data = y.data().as_device_ptr().as_ptr();
         let y_desc = y.descriptor();
 
         // If the _ size is 0 then the algorithm can work in-place and cuDNN expects a null
         // pointer.
         let (work_space_ptr, work_space_size): (*mut u8, usize) = {
             work_space.map_or((std::ptr::null_mut(), 0), |work_space| {
-                (work_space.as_device_ptr().as_raw_mut(), work_space.len())
+                (work_space.as_device_ptr().as_mut_ptr(), work_space.len())
             })
         };
 
@@ -1287,12 +1287,12 @@ impl CudnnContext {
         let dy_data = dy.data().as_device_ptr().as_raw();
         let dy_desc = dy.descriptor();
 
-        let dx_data = dx.data().as_device_ptr().as_raw_mut();
+        let dx_data = dx.data().as_device_ptr().as_ptr();
         let dx_desc = dx.descriptor();
 
         let (work_space_ptr, work_space_size): (*mut u8, usize) = {
             work_space.map_or((std::ptr::null_mut(), 0), |work_space| {
-                (work_space.as_device_ptr().as_raw_mut(), work_space.len())
+                (work_space.as_device_ptr().as_mut_ptr(), work_space.len())
             })
         };
 
@@ -1388,12 +1388,12 @@ impl CudnnContext {
         let dy_data = dy.data().as_device_ptr().as_raw();
         let dy_desc = dy.descriptor();
 
-        let dw_data = dw.data().as_device_ptr().as_raw_mut();
+        let dw_data = dw.data().as_device_ptr().as_ptr();
         let dw_desc = dw.descriptor();
 
         let (work_space_ptr, work_space_size): (*mut u8, usize) = {
             work_space.map_or((std::ptr::null_mut(), 0), |work_space| {
-                (work_space.as_device_ptr().as_raw_mut(), work_space.len())
+                (work_space.as_device_ptr().as_mut_ptr(), work_space.len())
             })
         };
 
@@ -1615,28 +1615,28 @@ impl CudnnContext {
         L: RnnDataLayout,
         NCHW: SupportedType<T1>,
     {
-        let device_sequence_lengths_ptr = device_seq_lengths.as_device_ptr().as_raw();
+        let device_sequence_lengths_ptr = device_seq_lengths.as_device_ptr().as_ptr();
 
         let x_ptr = x.as_device_ptr().as_raw();
-        let y_ptr = y.as_device_ptr().as_raw_mut();
+        let y_ptr = y.as_device_ptr().as_ptr();
 
-        let hx_ptr = hx.map_or(std::ptr::null(), |buff| buff.as_device_ptr().as_raw());
+        let hx_ptr = hx.map_or(std::ptr::null(), |buff| buff.as_device_ptr().as_ptr());
         let hy_ptr = hy.map_or(std::ptr::null_mut(), |buff| {
-            buff.as_device_ptr().as_raw_mut()
+            buff.as_device_ptr().as_mut_ptr()
         });
 
         let c_desc = c_desc.map_or(std::ptr::null_mut(), |desc| desc.raw);
 
-        let cx_ptr = cx.map_or(std::ptr::null(), |buff| buff.as_device_ptr().as_raw());
+        let cx_ptr = cx.map_or(std::ptr::null(), |buff| buff.as_device_ptr().as_ptr());
         let cy_ptr = cy.map_or(std::ptr::null_mut(), |buff| {
-            buff.as_device_ptr().as_raw_mut()
+            buff.as_device_ptr().as_mut_ptr()
         });
 
-        let weight_space_ptr = weight_space.as_device_ptr().as_raw_mut();
-        let work_space_ptr = work_space.as_device_ptr().as_raw_mut();
+        let weight_space_ptr = weight_space.as_device_ptr().as_ptr();
+        let work_space_ptr = work_space.as_device_ptr().as_ptr();
         let (reserve_space_ptr, reserve_space_size) = reserve_space
             .map_or((std::ptr::null_mut(), 0), |buff| {
-                (buff.as_device_ptr().as_raw_mut(), buff.len())
+                (buff.as_device_ptr().as_mut_ptr(), buff.len())
             });
 
         unsafe {
@@ -1814,32 +1814,32 @@ impl CudnnContext {
         L: RnnDataLayout,
         NCHW: SupportedType<T1>,
     {
-        let device_sequence_lengths_ptr = device_seq_lengths.as_device_ptr().as_raw();
+        let device_sequence_lengths_ptr = device_seq_lengths.as_device_ptr().as_ptr();
 
         let y_ptr = y.as_device_ptr().as_raw();
         let dy_ptr = dy.as_device_ptr().as_raw();
 
-        let dx_ptr = dx.as_device_ptr().as_raw_mut();
+        let dx_ptr = dx.as_device_ptr().as_ptr();
 
         let h_desc = h_desc.map_or(std::ptr::null_mut(), |desc| desc.raw);
 
-        let hx_ptr = hx.map_or(std::ptr::null(), |buff| buff.as_device_ptr().as_raw());
-        let dhy_ptr = dhy.map_or(std::ptr::null(), |buff| buff.as_device_ptr().as_raw());
+        let hx_ptr = hx.map_or(std::ptr::null(), |buff| buff.as_device_ptr().as_ptr());
+        let dhy_ptr = dhy.map_or(std::ptr::null(), |buff| buff.as_device_ptr().as_ptr());
         let dhx_ptr = dhx.map_or(std::ptr::null_mut(), |buff| {
-            buff.as_device_ptr().as_raw_mut()
+            buff.as_device_ptr().as_mut_ptr()
         });
 
         let c_desc = c_desc.map_or(std::ptr::null_mut(), |desc| desc.raw);
 
-        let cx_ptr = cx.map_or(std::ptr::null(), |buff| buff.as_device_ptr().as_raw());
-        let dcy_ptr = dcy.map_or(std::ptr::null(), |buff| buff.as_device_ptr().as_raw());
+        let cx_ptr = cx.map_or(std::ptr::null(), |buff| buff.as_device_ptr().as_ptr());
+        let dcy_ptr = dcy.map_or(std::ptr::null(), |buff| buff.as_device_ptr().as_mut_ptr());
         let dcx_ptr = dcx.map_or(std::ptr::null_mut(), |buff| {
-            buff.as_device_ptr().as_raw_mut()
+            buff.as_device_ptr().as_mut_ptr()
         });
 
-        let weight_space_ptr = weight_space.as_device_ptr().as_raw_mut();
-        let work_space_ptr = work_space.as_device_ptr().as_raw_mut();
-        let reserve_space_ptr = reserve_space.as_device_ptr().as_raw_mut();
+        let weight_space_ptr = weight_space.as_device_ptr().as_ptr();
+        let work_space_ptr = work_space.as_device_ptr().as_ptr();
+        let reserve_space_ptr = reserve_space.as_device_ptr().as_ptr();
 
         unsafe {
             sys::cudnnRNNBackwardData_v8(
@@ -1947,15 +1947,15 @@ impl CudnnContext {
         L: RnnDataLayout,
         NCHW: SupportedType<T1>,
     {
-        let device_sequence_lengths_ptr = device_seq_lengths.as_device_ptr().as_raw();
+        let device_sequence_lengths_ptr = device_seq_lengths.as_device_ptr().as_mut_ptr();
 
         let x_ptr = x.as_device_ptr().as_raw();
         let hx_ptr = x.as_device_ptr().as_raw();
         let y_ptr = y.as_device_ptr().as_raw();
 
-        let dweight_space_ptr = dweight_space.as_device_ptr().as_raw_mut();
-        let work_space_ptr = work_space.as_device_ptr().as_raw_mut();
-        let reserve_space_ptr = reserve_space.as_device_ptr().as_raw_mut();
+        let dweight_space_ptr = dweight_space.as_device_ptr().as_mut_ptr();
+        let work_space_ptr = work_space.as_device_ptr().as_mut_ptr();
+        let reserve_space_ptr = reserve_space.as_device_ptr().as_mut_ptr();
 
         unsafe {
             sys::cudnnRNNBackwardWeights_v8(

crates/cust/CHANGELOG.md

@@ -30,6 +30,24 @@ any breaking changes, the API is the same.
 - `Linker::complete` now only returns the built cubin, and not the cubin and a duration.
 - `Stream`, `Module`, `Linker`, `Function`, `Event`, `UnifiedBox`, `ArrayObject`, `LockedBuffer`, `LockedBox`, `DeviceSlice`, `DeviceBuffer`, and `DeviceBox` all now impl `Send` and `Sync`, this makes
 it much easier to write multigpu code. The CUDA API is fully thread-safe except for graph objects.
+- Features such as `vek` for implementing DeviceCopy are now `impl_cratename`, e.g. `impl_vek`, `impl_half`, etc.
+- `DevicePointer::as_raw` now returns a `CUdeviceptr` instead of a `*const T`.
+- Added `DevicePointer::as_ptr` and `DevicePointer::as_mut_ptr` for returning `*const T` or `*mut T`.
+- Added mint integration behind `impl_mint`.
+- Added half integration behind `impl_half`.
+- Added glam integration behind `impl_glam`.
+- Added experimental linux external memory import APIs through `cust::external::ExternalMemory`.
+- `vek` is no longer re-exported.
+- `DeviceBox` now requires `T: DeviceCopy` (previously it didn't but almost all its methods did).
+- `DeviceBox::from_raw` now takes a `CUdeviceptr` instead of a `*mut T`.
+- `DeviceBox::as_device_ptr` now requires `&self` instead of `&mut self`.
+- `DeviceBuffer` now requires `T: DeviceCopy`.
+- `DeviceSlice` now requires `T: DeviceCopy`.
+- `DeviceSlice` is now represented by a ptr and a len instead of `[T]` which was likely unsound.
+- `DeviceSlice::as_ptr` and `DeviceSlice::as_ptr_mut` now both return a `DevicePointer<T>`.
+- Added `DeviceSlice::as_device_ptr`.
+- `DeviceSlice::chunks` and `DeviceSlice::chunks_mut` have been deleted.
+- `DeviceSlice::from_slice` and `DeviceSlice::from_slice_mut` have been deleted.
 
 ## 0.2.2 - 12/5/21
 

crates/cust/Cargo.toml

@@ -17,7 +17,7 @@ cust_core = { path = "../cust_core", version = "0.1.0"}
 cust_raw = { path = "../cust_raw", version = "0.11.2"}
 bitflags = "1.2"
 cust_derive = { path = "../cust_derive", version = "0.1" }
-glam = { git = "https://github.com/anderslanglands/glam-rs", branch = "cuda", features=["cuda"], optional=true }
+glam = { version = "0.20", features=["cuda"], optional = true }
 mint = { version = "^0.5", optional = true }
 num-complex = { version = "0.4", optional = true }
 vek = { version = "0.15.1", optional = true, default-features = false }

crates/cust/src/external.rs

@@ -10,6 +10,7 @@ pub struct ExternalMemory(sys::CUexternalMemory);
 
 impl ExternalMemory {
     // Import an external memory referenced by `fd` with `size`
+    #[allow(clippy::missing_safety_doc)]
     pub unsafe fn import(fd: i32, size: usize) -> CudaResult<ExternalMemory> {
         let desc = sys::CUDA_EXTERNAL_MEMORY_HANDLE_DESC {
             type_: sys::CUexternalMemoryHandleType_enum::CU_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD,
@@ -26,6 +27,7 @@ impl ExternalMemory {
             .map(|_| ExternalMemory(memory))
     }
 
+    #[allow(clippy::missing_safety_doc)]
     pub unsafe fn reimport(&mut self, fd: i32, size: usize) -> CudaResult<()> {
         // import new memory - this will call drop to destroy the old one
         *self = ExternalMemory::import(fd, size)?;