forked from ccf-ai-infra/GPUCodeForces
71 lines
1.7 KiB
Python
71 lines
1.7 KiB
Python
import torch
|
|
import torch.nn as nn
|
|
from torch.utils.cpp_extension import load_inline
|
|
|
|
cuda_source = """
|
|
#include <torch/extension.h>
|
|
#include <cuda_runtime.h>
|
|
#include <math.h>
|
|
|
|
__global__ void complex_abs_angle_polar_kernel(
|
|
const float* __restrict__ input,
|
|
float* __restrict__ output,
|
|
int N
|
|
) {
|
|
int idx = blockIdx.x * blockDim.x + threadIdx.x;
|
|
if (idx < N) {
|
|
int input_offset = idx * 2;
|
|
int output_offset = idx * 2;
|
|
|
|
float X = input[input_offset];
|
|
float Y = input[input_offset + 1];
|
|
|
|
// 1. Magnitude (Abs): R = hypot(X, Y)
|
|
float R = hypotf(X, Y);
|
|
|
|
// 2. Angle (Argument): Theta = atan2(Y, X)
|
|
float Theta = atan2f(Y, X);
|
|
|
|
// 3. Polar Conversion (Output [R, Theta])
|
|
output[output_offset] = R;
|
|
output[output_offset + 1] = Theta;
|
|
}
|
|
}
|
|
|
|
torch::Tensor complex_abs_angle_polar_cuda(torch::Tensor input) {
|
|
auto output = torch::empty_like(input);
|
|
int N = input.size(0);
|
|
|
|
const int block_size = 256;
|
|
int num_blocks = (N + block_size - 1) / block_size;
|
|
|
|
complex_abs_angle_polar_kernel<<<num_blocks, block_size>>>(
|
|
input.data_ptr<float>(),
|
|
output.data_ptr<float>(),
|
|
N
|
|
);
|
|
|
|
return output;
|
|
}
|
|
"""
|
|
|
|
cpp_source = """
|
|
torch::Tensor complex_abs_angle_polar_cuda(torch::Tensor input);
|
|
"""
|
|
|
|
module = load_inline(
|
|
name="complex_abs_angle_polar",
|
|
cpp_sources=cpp_source,
|
|
cuda_sources=cuda_source,
|
|
functions=["complex_abs_angle_polar_cuda"],
|
|
verbose=True
|
|
)
|
|
|
|
|
|
class ModelNew(nn.Module):
|
|
def __init__(self):
|
|
super(ModelNew, self).__init__()
|
|
self.module = module
|
|
|
|
def forward(self, x):
|
|
return self.module.complex_abs_angle_polar_cuda(x) |