forked from ccf-ai-infra/GPUCodeForces
finish CrossEntropyDiceLoss #59
This commit is contained in:
parent
10eed82956
commit
d44eb0f537
|
|
@ -0,0 +1,166 @@
|
|||
import torch
|
||||
import torch.nn as nn
|
||||
from torch.utils.cpp_extension import load_inline
|
||||
|
||||
|
||||
class ModelNew(nn.Module):
|
||||
def __init__(self, alpha=0.5, smooth=1.0):
|
||||
super().__init__()
|
||||
self.alpha = alpha
|
||||
self.smooth = smooth
|
||||
self._compile_cuda_kernel()
|
||||
|
||||
def _compile_cuda_kernel(self):
|
||||
cpp_source = """
|
||||
torch::Tensor ce_dice_loss_cuda(torch::Tensor logits, torch::Tensor targets, double smooth, int batch_size);
|
||||
"""
|
||||
|
||||
cuda_source = """
|
||||
#include <torch/extension.h>
|
||||
#include <cuda_runtime.h>
|
||||
#include <math.h>
|
||||
|
||||
__device__ __forceinline__ double sigmoid_d(double x) {
|
||||
if (x >= 0.0) {
|
||||
double z = exp(-x);
|
||||
return 1.0 / (1.0 + z);
|
||||
} else {
|
||||
double z = exp(x);
|
||||
return z / (1.0 + z);
|
||||
}
|
||||
}
|
||||
|
||||
__device__ __forceinline__ double log_sigmoid_d(double x) {
|
||||
if (x >= 0.0) {
|
||||
return -log(1.0 + exp(-x));
|
||||
} else {
|
||||
return x - log(1.0 + exp(x));
|
||||
}
|
||||
}
|
||||
|
||||
__global__ void ce_dice_loss_kernel(
|
||||
const double* __restrict__ logits,
|
||||
const double* __restrict__ targets,
|
||||
double* __restrict__ intersection_out,
|
||||
double* __restrict__ sum_probs_out,
|
||||
double* __restrict__ sum_targets_out,
|
||||
double* __restrict__ ce_out,
|
||||
const int batch_size,
|
||||
const int feature_dim)
|
||||
{
|
||||
const int batch_idx = blockIdx.x;
|
||||
const int tid = threadIdx.x;
|
||||
const int stride = blockDim.x;
|
||||
|
||||
if (batch_idx >= batch_size) return;
|
||||
|
||||
__shared__ double s_inter[256];
|
||||
__shared__ double s_probs[256];
|
||||
__shared__ double s_targets[256];
|
||||
__shared__ double s_ce[256];
|
||||
|
||||
double local_inter = 0.0;
|
||||
double local_probs = 0.0;
|
||||
double local_targets = 0.0;
|
||||
double local_ce = 0.0;
|
||||
|
||||
const int offset = batch_idx * feature_dim;
|
||||
|
||||
for (int i = tid; i < feature_dim; i += stride) {
|
||||
double z = logits[offset + i];
|
||||
double y = targets[offset + i];
|
||||
|
||||
double p = sigmoid_d(z);
|
||||
|
||||
local_inter += p * y;
|
||||
local_probs += p;
|
||||
local_targets += y;
|
||||
|
||||
double log_p = log_sigmoid_d(z);
|
||||
double log_1mp = log_sigmoid_d(-z);
|
||||
double ce = -(y * log_p + (1.0 - y) * log_1mp);
|
||||
|
||||
local_ce += ce;
|
||||
}
|
||||
|
||||
s_inter[tid] = local_inter;
|
||||
s_probs[tid] = local_probs;
|
||||
s_targets[tid] = local_targets;
|
||||
s_ce[tid] = local_ce;
|
||||
__syncthreads();
|
||||
|
||||
for (int s = blockDim.x / 2; s > 0; s >>= 1) {
|
||||
if (tid < s) {
|
||||
s_inter[tid] += s_inter[tid + s];
|
||||
s_probs[tid] += s_probs[tid + s];
|
||||
s_targets[tid] += s_targets[tid + s];
|
||||
s_ce[tid] += s_ce[tid + s];
|
||||
}
|
||||
__syncthreads();
|
||||
}
|
||||
|
||||
if (tid == 0) {
|
||||
intersection_out[batch_idx] = s_inter[0];
|
||||
sum_probs_out[batch_idx] = s_probs[0];
|
||||
sum_targets_out[batch_idx] = s_targets[0];
|
||||
ce_out[batch_idx] = s_ce[0];
|
||||
}
|
||||
}
|
||||
|
||||
torch::Tensor ce_dice_loss_cuda(torch::Tensor logits, torch::Tensor targets, double smooth, int batch_size) {
|
||||
auto Z_c = logits.contiguous();
|
||||
auto Y_c = targets.contiguous();
|
||||
|
||||
const int feature_dim = Z_c.size(1);
|
||||
|
||||
auto intersection_out = torch::zeros({batch_size}, Z_c.options());
|
||||
auto sum_probs_out = torch::zeros({batch_size}, Z_c.options());
|
||||
auto sum_targets_out = torch::zeros({batch_size}, Z_c.options());
|
||||
auto ce_out = torch::zeros({batch_size}, Z_c.options());
|
||||
|
||||
const int threads = 256;
|
||||
const int blocks = batch_size;
|
||||
|
||||
ce_dice_loss_kernel<<<blocks, threads>>>(
|
||||
Z_c.data_ptr<double>(),
|
||||
Y_c.data_ptr<double>(),
|
||||
intersection_out.data_ptr<double>(),
|
||||
sum_probs_out.data_ptr<double>(),
|
||||
sum_targets_out.data_ptr<double>(),
|
||||
ce_out.data_ptr<double>(),
|
||||
batch_size,
|
||||
feature_dim
|
||||
);
|
||||
|
||||
return torch::cat({intersection_out, sum_probs_out, sum_targets_out, ce_out}, 0);
|
||||
}
|
||||
"""
|
||||
|
||||
self.op = load_inline(
|
||||
name="ce_dice_loss_v1",
|
||||
cpp_sources=cpp_source,
|
||||
cuda_sources=cuda_source,
|
||||
functions=["ce_dice_loss_cuda"],
|
||||
extra_cuda_cflags=["-O3"],
|
||||
verbose=False
|
||||
)
|
||||
|
||||
def forward(self, logits, targets):
|
||||
targets_f = targets.to(logits.dtype)
|
||||
|
||||
batch_size = logits.size(0)
|
||||
feature_dim = logits.size(1)
|
||||
|
||||
comp_flat = self.op.ce_dice_loss_cuda(logits, targets_f, self.smooth, batch_size)
|
||||
|
||||
intersection = comp_flat[:batch_size]
|
||||
sum_probs = comp_flat[batch_size:2 * batch_size]
|
||||
sum_targets = comp_flat[2 * batch_size:3 * batch_size]
|
||||
ce_sum = comp_flat[3 * batch_size:]
|
||||
|
||||
dice = (2.0 * intersection + self.smooth) / (sum_probs + sum_targets + self.smooth)
|
||||
dice_loss = 1.0 - dice.mean()
|
||||
|
||||
ce_loss = ce_sum.sum() / (batch_size * feature_dim)
|
||||
|
||||
return self.alpha * dice_loss + (1.0 - self.alpha) * ce_loss
|
||||
|
|
@ -0,0 +1,39 @@
|
|||
import torch
|
||||
import torch.nn as nn
|
||||
import torch.nn.functional as F
|
||||
|
||||
|
||||
class Model(nn.Module):
|
||||
def __init__(self, alpha=0.5, smooth=1.0):
|
||||
super().__init__()
|
||||
self.alpha = alpha
|
||||
self.smooth = smooth
|
||||
|
||||
def forward(self, logits: torch.Tensor, targets: torch.Tensor) -> torch.Tensor:
|
||||
targets_f = targets.float()
|
||||
|
||||
probs = logits.sigmoid()
|
||||
probs = probs.flatten(1)
|
||||
targets_f_flat = targets_f.flatten(1)
|
||||
|
||||
intersection = (probs * targets_f_flat).sum(dim=1)
|
||||
dice = (2.0 * intersection + self.smooth) / (probs.sum(dim=1) + targets_f_flat.sum(dim=1) + self.smooth)
|
||||
dice_loss = 1.0 - dice.mean()
|
||||
|
||||
ce_loss = F.binary_cross_entropy_with_logits(logits, targets_f, reduction='mean')
|
||||
|
||||
return self.alpha * dice_loss + (1.0 - self.alpha) * ce_loss
|
||||
|
||||
|
||||
batch_size = 128
|
||||
feature_dim = 100
|
||||
|
||||
|
||||
def get_inputs():
|
||||
logits = torch.randn(batch_size, feature_dim, dtype=torch.float64)
|
||||
targets = torch.randint(0, 2, (batch_size, feature_dim), dtype=torch.float64)
|
||||
return [logits, targets]
|
||||
|
||||
|
||||
def get_init_inputs():
|
||||
return [0.5, 1.0]
|
||||
|
|
@ -0,0 +1,69 @@
|
|||
You have complete freedom to choose the set of operators you want to replace. You may make the decision to replace some operators with custom CUDA kernels and leave others unchanged. You may replace multiple operators with custom implementations, consider operator fusion opportunities (combining multiple operators into a single kernel, for example, combining chunk+gelu+elementwise_mul), or algorithmic changes (such as optimized memory access patterns). You are only limited by your imagination.
|
||||
CUDA kernel for CrossEntropy-Dice Loss with shared memory parallel reduction.
|
||||
|
||||
Optimizations:
|
||||
|
||||
Numerically stable sigmoid/log-sigmoid using exp(-|x|).
|
||||
|
||||
Parallel tree reduction in shared memory (4 concurrent reductions).
|
||||
|
||||
Batch-level parallelism (one block per sample).
|
||||
|
||||
Double precision for accuracy.
|
||||
|
||||
Kernel computes per batch:
|
||||
|
||||
intersection = Σ(p*y) for Dice
|
||||
|
||||
sum_probs = Σ(p)
|
||||
|
||||
sum_targets = Σ(y)
|
||||
|
||||
ce_sum = Σ(BCE loss)
|
||||
|
||||
Final loss:
|
||||
L = α·(1-mean(Dice)) + (1-α)·mean(CE)
|
||||
|
||||
|
||||
|
||||
|
||||
Here's an example to show you the syntax of inline embedding custom CUDA operators in torch: The example given architecture is:
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
import torch.nn.functional as F
|
||||
|
||||
|
||||
class Model(nn.Module):
|
||||
def __init__(self, alpha=0.5, smooth=1.0):
|
||||
super().__init__()
|
||||
self.alpha = alpha
|
||||
self.smooth = smooth
|
||||
|
||||
def forward(self, logits: torch.Tensor, targets: torch.Tensor) -> torch.Tensor:
|
||||
targets_f = targets.float()
|
||||
|
||||
probs = logits.sigmoid()
|
||||
probs = probs.flatten(1)
|
||||
targets_f_flat = targets_f.flatten(1)
|
||||
|
||||
intersection = (probs * targets_f_flat).sum(dim=1)
|
||||
dice = (2.0 * intersection + self.smooth) / (probs.sum(dim=1) + targets_f_flat.sum(dim=1) + self.smooth)
|
||||
dice_loss = 1.0 - dice.mean()
|
||||
|
||||
ce_loss = F.binary_cross_entropy_with_logits(logits, targets_f, reduction='mean')
|
||||
|
||||
return self.alpha * dice_loss + (1.0 - self.alpha) * ce_loss
|
||||
|
||||
|
||||
batch_size = 128
|
||||
feature_dim = 100
|
||||
|
||||
|
||||
def get_inputs():
|
||||
logits = torch.randn(batch_size, feature_dim, dtype=torch.float64)
|
||||
targets = torch.randint(0, 2, (batch_size, feature_dim), dtype=torch.float64)
|
||||
return [logits, targets]
|
||||
|
||||
|
||||
def get_init_inputs():
|
||||
return [0.5, 1.0]
|
||||
|
|
@ -0,0 +1,77 @@
|
|||
###########################################################
|
||||
# 性能和精度验证程序
|
||||
###########################################################
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
import time
|
||||
from CrossEntropyDiceLoss_torch import Model, get_inputs, get_init_inputs
|
||||
from CrossEntropyDiceLoss_cuda import ModelNew
|
||||
|
||||
|
||||
def run_benchmark():
|
||||
# 检查 CUDA 是否可用
|
||||
if not torch.cuda.is_available():
|
||||
print("CUDA 不可用，请确保您有可用的 NVIDIA GPU 并已正确安装 PyTorch CUDA 版本。")
|
||||
return
|
||||
else:
|
||||
device = torch.device("cuda")
|
||||
|
||||
# 初始化模型
|
||||
init_inputs = get_init_inputs()
|
||||
init_inputs = [
|
||||
x.cuda(device=device) if isinstance(x, torch.Tensor) else x for x in init_inputs
|
||||
]
|
||||
inputs = get_inputs()
|
||||
inputs = [
|
||||
x.cuda(device=device) if isinstance(x, torch.Tensor) else x for x in inputs
|
||||
]
|
||||
|
||||
torch_model = Model(*init_inputs).cuda()
|
||||
cuda_model = ModelNew(*init_inputs).cuda()
|
||||
|
||||
torch_model.eval()
|
||||
cuda_model.eval()
|
||||
|
||||
print("-------------------- 精度对齐验证 --------------------")
|
||||
with torch.no_grad():
|
||||
output_torch = torch_model(*inputs)
|
||||
output_cuda = cuda_model(*inputs)
|
||||
|
||||
precision_flag = torch.allclose(output_torch, output_cuda, rtol=1e-03)
|
||||
if precision_flag:
|
||||
print("✅ 精度对齐：两个模型的输出结果非常接近。")
|
||||
else:
|
||||
print("❌ 精度不一致！")
|
||||
|
||||
print("\n-------------------- 性能加速比测试 --------------------")
|
||||
num_iterations = 100
|
||||
|
||||
# PyTorch 模型计时
|
||||
torch.cuda.synchronize()
|
||||
start_time = time.time()
|
||||
for _ in range(num_iterations):
|
||||
_ = torch_model(*inputs)
|
||||
torch.cuda.synchronize()
|
||||
torch_time = (time.time() - start_time) / num_iterations
|
||||
|
||||
# 自定义 CUDA 内核计时
|
||||
torch.cuda.synchronize()
|
||||
start_time = time.time()
|
||||
for _ in range(num_iterations):
|
||||
_ = cuda_model(*inputs)
|
||||
torch.cuda.synchronize()
|
||||
cuda_time = (time.time() - start_time) / num_iterations
|
||||
|
||||
print(f"PyTorch torch.relu 平均执行时间: {torch_time:.6f} 秒")
|
||||
print(f"自定义 CUDA 内核 平均执行时间: {cuda_time:.6f} 秒")
|
||||
speedup = 0
|
||||
if cuda_time > 0:
|
||||
speedup = torch_time / cuda_time
|
||||
print(f"加速比 (Speedup): {speedup:.2f}x")
|
||||
else:
|
||||
print("CUDA 内核执行时间为0，无法计算加速比。")
|
||||
return precision_flag, speedup
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
precision_flag, speedup = run_benchmark()
|
||||
Loading…
Reference in New Issue