mirror of
https://github.com/ml-explore/mlx.git
synced 2025-12-16 01:49:05 +08:00
13 lines
387 B
C
13 lines
387 B
C
// Copyright © 2025 Apple Inc.
|
|
|
|
// This file is used by both CUDA kernel code and host-only C++ code.
|
|
|
|
#pragma once
|
|
|
|
// The maximum dimensions of shape/strides passed as kernel parameters.
|
|
#define MAX_NDIM 8
|
|
|
|
// All existing NVIDIA hardware has a fixed 32 warp size. Though a built-in
|
|
// warpSize variable exists, using it would prevent compile-time optimizations.
|
|
#define WARP_SIZE 32
|