File: cuda_extension_kernel2.cu

package info (click to toggle)
pytorch 1.13.1%2Bdfsg-4
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 139,252 kB
  • sloc: cpp: 1,100,274; python: 706,454; ansic: 83,052; asm: 7,618; java: 3,273; sh: 2,841; javascript: 612; makefile: 323; xml: 269; ruby: 185; yacc: 144; objc: 68; lex: 44
file content (25 lines) | stat: -rw-r--r-- 780 bytes parent folder | download | duplicates (3)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
#include <cuda.h>
#include <cuda_runtime.h>
#include <c10/cuda/CUDAException.h>

#include <ATen/ATen.h>

__global__ void tanh_add_kernel(
    const float* __restrict__ x,
    const float* __restrict__ y,
    float* __restrict__ output,
    const int size) {
  const int index = blockIdx.x * blockDim.x + threadIdx.x;
  if (index < size) {
    const float tanh_x = 2.0f / (1.0f + __expf(-2.0f * x[index])) - 1;
    const float tanh_y = 2.0f / (1.0f + __expf(-2.0f * y[index])) - 1;
    output[index] = tanh_x + tanh_y;
  }
}

void tanh_add_cuda(const float* x, const float* y, float* output, int size) {
  const int threads = 1024;
  const int blocks = (size + threads - 1) / threads;
  tanh_add_kernel<<<blocks, threads>>>(x, y, output, size);
  C10_CUDA_KERNEL_LAUNCH_CHECK();
}