Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add split_tensor_along method #917

Draft
wants to merge 8 commits into
base: main
Choose a base branch
from
Draft

Add split_tensor_along method #917

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
5c532ec
remove deprecated ftz intrinsics
rainiwu Jan 26, 2024
fb91f13
suppress spurious cargo clippy warning
rainiwu Jan 26, 2024
24a8593
Merge pull request #1 from rainiwu/remove-ftz
swfsql Feb 7, 2024
4e3f7c7
avoid conv1d bound for cudnn
swfsql Feb 6, 2024
a8bc54c
bump gemm
swfsql Feb 9, 2024
557687c
clippy fix
swfsql Feb 9, 2024
1175903
Merge pull request #2 from swfsql/avoid-ci-errors
swfsql Mar 1, 2024
aa44eb5
Add split_tensor_along method
swfsql Feb 1, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Next Next commit
remove deprecated ftz intrinsics
  • Loading branch information
@rainiwu
rainiwu committed Jan 26, 2024
commit 5c532ec5dc51cd17cd4bb9ae940ecf2c9baf89f6
38 changes: 0 additions & 38 deletions dfdx-core/src/lib.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -128,44 +128,6 @@ pub mod prelude {
pub use crate::tensor_ops::*;
}

/// Sets a CPU `sse` flag to flush denormal floating point numbers to zero. The opposite of this is [keep_denormals()].
///
/// Some resources:
/// 1. [Effects of Flush-To-Zero mode](https://developer.arm.com/documentation/dui0473/c/neon-and-vfp-programming/the-effects-of-using-flush-to-zero-mode?lang=en)
/// 2. [When to use Flush-To-Zero mode](https://developer.arm.com/documentation/dui0473/c/neon-and-vfp-programming/when-to-use-flush-to-zero-mode?lang=en)
pub fn flush_denormals_to_zero() {
#[cfg(all(target_arch = "x86", target_feature = "sse"))]
{
use std::arch::x86::{_MM_FLUSH_ZERO_ON, _MM_SET_FLUSH_ZERO_MODE};
unsafe { _MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_ON) }
}

#[cfg(all(target_arch = "x86_64", target_feature = "sse"))]
{
use std::arch::x86_64::{_MM_FLUSH_ZERO_ON, _MM_SET_FLUSH_ZERO_MODE};
unsafe { _MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_ON) }
}
}

/// Sets a CPU flag to keep denormal floating point numbers. The opposite of this is [flush_denormals_to_zero()].
///
/// Some resources:
/// 1. [Effects of Flush-To-Zero mode](https://developer.arm.com/documentation/dui0473/c/neon-and-vfp-programming/the-effects-of-using-flush-to-zero-mode?lang=en)
/// 2. [When to use Flush-To-Zero mode](https://developer.arm.com/documentation/dui0473/c/neon-and-vfp-programming/when-to-use-flush-to-zero-mode?lang=en)
pub fn keep_denormals() {
#[cfg(all(target_arch = "x86", target_feature = "sse"))]
{
use std::arch::x86::{_MM_FLUSH_ZERO_OFF, _MM_SET_FLUSH_ZERO_MODE};
unsafe { _MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_OFF) }
}

#[cfg(all(target_arch = "x86_64", target_feature = "sse"))]
{
use std::arch::x86_64::{_MM_FLUSH_ZERO_OFF, _MM_SET_FLUSH_ZERO_MODE};
unsafe { _MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_OFF) }
}
}

#[cfg(test)]
pub(crate) mod tests {
pub use num_traits::{Float, NumCast, Zero};
Expand Down
3 changes: 0 additions & 3 deletions dfdx/examples/12-mnist.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,9 +62,6 @@ type Mlp = (
const BATCH_SIZE: usize = 32;

fn main() {
// ftz substantially improves performance
dfdx::flush_denormals_to_zero();

let mnist_path = std::env::args()
.nth(1)
.unwrap_or_else(|| "./datasets/MNIST/raw".to_string());
Expand Down
Loading