huggingface/candle
1
0
Fork 0
mirror of https://github.com/huggingface/candle.git synced 2025-06-18 19:47:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add more binary kernels.

Browse Source
This commit is contained in:
laurent
2023-06-22 14:07:02 +01:00
parent 97fe1fac85
commit b8f514d9c6
9 changed files with 54 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
kernels/src/binary.cu Normal file
View File

@ -0,0 +1,29 @@
#include "binary_op_macros.cuh"
#if __CUDA_ARCH__ >= 530
BINARY_OP(__half, badd_f16, x + y)
#endif
BINARY_OP(float, badd_f32, x + y)
BINARY_OP(double, badd_fwd_f64, x + y);
#if __CUDA_ARCH__ >= 530
BINARY_OP(__half, bdiv_f16, x / y)
#endif
BINARY_OP(float, bdiv_f32, x / y)
BINARY_OP(double, bdiv_f64, x / y);
#if __CUDA_ARCH__ >= 530
BINARY_OP(__half, bmul_f16, x * y)
#endif
BINARY_OP(float, bmul_f32, x * y)
BINARY_OP(double, bmul_f64, x * y);
#if __CUDA_ARCH__ >= 530
BINARY_OP(__half, bsub_f16, x - y)
#endif
BINARY_OP(float, bsub_f32, x - y)
BINARY_OP(double, bsub_f64, x - y);

8
kernels/src/binary_add.cu
View File

@ -1,8 +0,0 @@
#include "binary_op_macros.cuh"
#if __CUDA_ARCH__ >= 530
BINARY_OP(__half, badd_f16, x + y)
#endif
BINARY_OP(float, badd_f32, x + y)
BINARY_OP(double, badd_fwd_f64, x + y);

8
kernels/src/binary_div.cu
View File

@ -1,8 +0,0 @@
#include "binary_op_macros.cuh"
#if __CUDA_ARCH__ >= 530
BINARY_OP(__half, bdiv_f16, x / y)
#endif
BINARY_OP(float, bdiv_f32, x / y)
BINARY_OP(double, bdiv_f64, x / y);

8
kernels/src/binary_mul.cu
View File

@ -1,8 +0,0 @@
#include "binary_op_macros.cuh"
#if __CUDA_ARCH__ >= 530
BINARY_OP(__half, bmul_f16, x * y)
#endif
BINARY_OP(float, bmul_f32, x * y)
BINARY_OP(double, bmul_f64, x * y);

8
kernels/src/binary_sub.cu
View File

@ -1,8 +0,0 @@
#include "binary_op_macros.cuh"
#if __CUDA_ARCH__ >= 530
BINARY_OP(__half, bsub_f16, x - y)
#endif
BINARY_OP(float, bsub_f32, x - y)
BINARY_OP(double, bsub_f64, x - y);

5
kernels/src/lib.rs
View File

@ -1,6 +1,3 @@
pub const AFFINE: &str = include_str!(concat!(env!("OUT_DIR"), "/affine.ptx"));
pub const BINARY_ADD: &str = include_str!(concat!(env!("OUT_DIR"), "/binary_add.ptx"));
pub const BINARY_DIV: &str = include_str!(concat!(env!("OUT_DIR"), "/binary_div.ptx"));
pub const BINARY_MUL: &str = include_str!(concat!(env!("OUT_DIR"), "/binary_mul.ptx"));
pub const BINARY_SUB: &str = include_str!(concat!(env!("OUT_DIR"), "/binary_sub.ptx"));
pub const BINARY: &str = include_str!(concat!(env!("OUT_DIR"), "/binary.ptx"));
pub const FILL: &str = include_str!(concat!(env!("OUT_DIR"), "/fill.ptx"));