import torch
import torch.nn as nn
import torch.nn.functional as F
def module_fn(predictions: torch.Tensor, targets: torch.Tensor) -> torch.Tensor:
"""
Computes the Mean Squared Error loss for regression tasks.
Args:
predictions (torch.Tensor): Predicted values.
targets (torch.Tensor): Target values.
Returns:
torch.Tensor: Mean Squared Error loss.
"""
return F.mse_loss(predictions, targets, reduction="mean")
class Model(nn.Module):
"""
A model that computes the Mean Squared Error loss for regression tasks.
Parameters:
None
"""
def __init__(self):
super(Model, self).__init__()
def forward(self, predictions, targets, fn=module_fn):
return fn(predictions, targets)
batch_size = 128
input_shape = (4096,)
dim = 1
def get_inputs():
return [
torch.randn(batch_size, *input_shape),
torch.randn(batch_size, *input_shape),
]
def get_init_inputs():
return []
import torch
import torch.nn as nn
class Model(nn.Module):
"""
A model that computes the Mean Squared Error loss for regression tasks.
Parameters:
None
"""
def __init__(self):
super(Model, self).__init__()
def forward(self, predictions, targets):
return torch.mean((predictions - targets) ** 2)
batch_size = 128
input_shape = (4096, )
dim = 1
def get_inputs():
return [torch.randn(batch_size, *input_shape), torch.randn(batch_size, *input_shape)]
def get_init_inputs():
return []
#include <pybind11/pybind11.h>
#include <torch/extension.h>
#include <cuda.h>
#include <cuda_runtime.h>
#include <algorithm>
static const int BLOCK_SIZE = 256;
template <typename scalar_t>
__global__ void optimized_mse_kernel(
const scalar_t* __restrict__ preds,
const scalar_t* __restrict__ tgts,
double* __restrict__ sum_out,
const int64_t num_elements
) {
__shared__ double smem[BLOCK_SIZE];
int tid = threadIdx.x;
double thread_sum = 0.0;
const int grid_stride = blockDim.x * gridDim.x;
int idx = blockIdx.x * blockDim.x + tid;
// Optimized grid-stride loop
while (idx < num_elements) {
double diff = static_cast<double>(preds[idx]) - static_cast<double>(tgts[idx]);
thread_sum += diff * diff;
idx += grid_stride;
}
smem[tid] = thread_sum;
__syncthreads();
// Unrolled reduction with warp-level optimization
for (int s = blockDim.x/2; s > 32; s >>= 1) {
if (tid < s) {
smem[tid] += smem[tid + s];
}
__syncthreads();
}
// Warp-level reduction
if (tid < 32) {
volatile double* vsmem = smem;
vsmem[tid] += vsmem[tid + 32];
vsmem[tid] += vsmem[tid + 16];
vsmem[tid] += vsmem[tid + 8];
vsmem[tid] += vsmem[tid + 4];
vsmem[tid] += vsmem[tid + 2];
vsmem[tid] += vsmem[tid + 1];
}
if (tid == 0) {
atomicAdd(sum_out, smem[0]);
}
}
torch::Tensor forward(torch::Tensor predictions, torch::Tensor targets) {
TORCH_CHECK(predictions.is_cuda(), "predictions must be a CUDA tensor");
TORCH_CHECK(targets.is_cuda(), "targets must be a CUDA tensor");
TORCH_CHECK(predictions.numel() == targets.numel(),
"predictions and targets must have the same number of elements");
const int64_t num_elements = predictions.numel();
auto accumulator = torch::zeros({1}, predictions.options().dtype(at::kDouble));
int grid_size = (num_elements + BLOCK_SIZE - 1) / BLOCK_SIZE;
grid_size = std::min(grid_size, 1024);
AT_DISPATCH_FLOATING_TYPES(predictions.scalar_type(), "optimized_mse_cuda", ([&] {
optimized_mse_kernel<scalar_t><<<grid_size, BLOCK_SIZE>>>(
predictions.data_ptr<scalar_t>(),
targets.data_ptr<scalar_t>(),
accumulator.data_ptr<double>(),
num_elements
);
}));
auto result = accumulator.div_(static_cast<double>(num_elements));
return result.to(predictions.dtype());
}
PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
m.def("forward", &forward, "Optimized MSE forward (CUDA)");
}| Metric | Value | Unit | Variance | Samples |
|---|---|---|---|---|
| Executed Ipc Active | 1.352 | inst/cycle | 0.000 | 5 |
| Executed Ipc Elapsed | 0.686 | inst/cycle | 0.000 | 5 |
| Issue Slots Busy | 36.046 | % | 0.041 | 5 |
| Issued Ipc Active | 1.442 | inst/cycle | 0.000 | 5 |
| SM Busy | 36.046 | % | 0.041 | 5 |
| Memory Throughput | 724896736366.736 | byte/second | 44080681346011742208.000 | 5 |
| Mem Busy | 14.084 | % | 0.015 | 5 |
| Max Bandwidth | 21.732 | % | 0.048 | 5 |
| L1/TEX Hit Rate | 0.000 | % | 0.000 | 5 |
| L2 Hit Rate | 18.686 | % | 0.008 | 5 |
| Mem Pipes Busy | 9.850 | % | 0.007 | 5 |
| Warp Cycles Per Issued Instruction | 35.452 | cycle | 0.028 | 5 |
| Warp Cycles Per Executed Instruction | 37.836 | cycle | 0.031 | 5 |
| Avg. Active Threads Per Warp | 31.730 | 0.000 | 5 | |
| Avg. Not Predicated Off Threads Per Warp | 28.440 | 0.000 | 5 | |
| Max Active Clusters | 0.000 | cluster | 0.000 | 5 |
| Max Cluster Size | 8.000 | block | 0.000 | 5 |
| Overall GPU Occupancy | 0.000 | % | 0.000 | 5 |
| Cluster Occupancy | 0.000 | % | 0.000 | 5 |
| Block Limit SM | 32.000 | block | 0.000 | 5 |
| Block Limit Registers | 10.000 | block | 0.000 | 5 |
| Block Limit Shared Mem | 21.000 | block | 0.000 | 5 |
| Block Limit Warps | 8.000 | block | 0.000 | 5 |
| Theoretical Active Warps per SM | 64.000 | warp | 0.000 | 5 |
| Theoretical Occupancy | 100.000 | % | 0.000 | 5 |
| Achieved Occupancy | 79.834 | % | 0.009 | 5 |
| Achieved Active Warps Per SM | 51.094 | warp | 0.004 | 5 |
| Rule | Description |
|---|---|
| INF HighPipeUtilization | ALU is the highest-utilized pipeline (20.2%) based on active cycles, taking into account the rates of its different instructions. It executes integer and logic operations. It is well-utilized, but should not be a bottleneck. |
| INF CPIStall | Check the Warp Stall Sampling (All Cycles) table for the top stall locations in your source based on sampling data. The Kernel Profiling Guide (https://docs.nvidia.com/nsight-compute/ProfilingGuide/index.html#metrics-reference) provides more details on each stall reason. |
| WRN Occupancy | This kernel's theoretical occupancy is not impacted by any block limit. The difference between calculated theoretical (100.0%) and measured achieved occupancy (79.7%) can be the result of warp scheduling overheads or workload imbalances during the kernel execution. Load imbalances can occur between warps within a block as well as across blocks of the same kernel. See the CUDA Best Practices Guide (https://docs.nvidia.com/cuda/cuda-c-best-practices-guide/index.html#occupancy) for more details on optimizing occupancy. |
| Operation / Metric | Value | Unit |
|---|---|---|
| aten::to | ||
| CPU Time | 1274560.64 | μs |
| Device Time | 225454.44 | μs |
| Self CPU Time | 76165.23 | μs |
| Self Device Time | 0.00 | μs |
| CPU Memory Usage | 0 | B |
| Device Memory Usage | 0 | B |
| Self CPU Memory Usage | 0 | B |
| Self Device Memory Usage | 0 | B |
| aten::_to_copy | ||
| CPU Time | 1198395.41 | μs |
| Device Time | 225454.44 | μs |
| Self CPU Time | 175706.42 | μs |
| Self Device Time | 0.00 | μs |
| CPU Memory Usage | 0 | B |
| Device Memory Usage | 0 | B |
| Self CPU Memory Usage | 0 | B |
| Self Device Memory Usage | 0 | B |
| aten::zero_ | ||
| CPU Time | 4326439.45 | μs |
| Device Time | 6695453.63 | μs |
| Self CPU Time | 301330.07 | μs |
| Self Device Time | 0.00 | μs |
| CPU Memory Usage | 0 | B |
| Device Memory Usage | 0 | B |
| Self CPU Memory Usage | 0 | B |
| Self Device Memory Usage | 0 | B |
| aten::fill_ | ||
| CPU Time | 4025111.13 | μs |
| Device Time | 6695453.63 | μs |
| Self CPU Time | 351206.85 | μs |
| Self Device Time | 6695453.63 | μs |
| CPU Memory Usage | 0 | B |
| Device Memory Usage | 0 | B |
| Self CPU Memory Usage | 0 | B |
| Self Device Memory Usage | 0 | B |
| cudaLaunchKernel | ||
| CPU Time | 4651041.03 | μs |
| Device Time | 1319.86 | μs |
| Self CPU Time | 4651041.03 | μs |
| Self Device Time | 1319.86 | μs |
| CPU Memory Usage | 0 | B |
| Device Memory Usage | 0 | B |
| Self CPU Memory Usage | 0 | B |
| Self Device Memory Usage | 0 | B |
| void optimized_mse_kernel<float>(float const*, float const*, double*, long) | ||
| CPU Time | 0.00 | μs |
| Device Time | 440105.23 | μs |
| Self CPU Time | 0.00 | μs |
| Self Device Time | 440105.23 | μs |
| CPU Memory Usage | 0 | B |
| Device Memory Usage | 0 | B |
| Self CPU Memory Usage | 0 | B |
| Self Device Memory Usage | 0 | B |
| void at::native::vectorized_elementwise_kernel<4, at::native::BUnaryFunctor<double, double, double, at::native::binary_internal::MulFunctor<double> >, at::detail::Array<char*, 2> >(int, at::native::BUnaryFunctor<double, double, double, at::native::binary_internal::MulFunctor<double> >, at::detail::Array<char*, 2>) | ||
| CPU Time | 0.00 | μs |
| Device Time | 228436.47 | μs |
| Self CPU Time | 0.00 | μs |
| Self Device Time | 228436.47 | μs |
| CPU Memory Usage | 0 | B |
| Device Memory Usage | 0 | B |
| Self CPU Memory Usage | 0 | B |
| Self Device Memory Usage | 0 | B |
| void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor<int>, at::detail::Array<char*, 1> >(int, at::native::FillFunctor<int>, at::detail::Array<char*, 1>) | ||
| CPU Time | 0.00 | μs |
| Device Time | 6495396.64 | μs |
| Self CPU Time | 0.00 | μs |
| Self Device Time | 6495396.64 | μs |
| CPU Memory Usage | 0 | B |
| Device Memory Usage | 0 | B |
| Self CPU Memory Usage | 0 | B |
| Self Device Memory Usage | 0 | B |
45282 warnings generated when compiling for host. Suppressed 45321 warnings (45274 in non-user code, 47 NOLINT). Use -header-filter=.* to display errors from all non-system headers. Use -system-headers to display errors from system headers as well.