47_Sum_reduction_over_a_dimension • fully_unrolled_warp_sum_reduction_baseimport torch
import torch.nn as nn
import torch.nn.functional as F
def module_fn(x: torch.Tensor, dim: int) -> torch.Tensor:
"""
Applies sum reduction over the specified dimension.
Args:
x (torch.Tensor): Input tensor of shape (..., dim, ...).
dim (int): Dimension to reduce over.
Returns:
torch.Tensor: Output tensor after sum reduction, shape (..., 1, ...).
"""
return torch.sum(x, dim=dim, keepdim=True)
class Model(nn.Module):
"""
Simple model that performs sum reduction over a specified dimension.
"""
def __init__(self, dim: int):
"""
Initializes the model with the dimension to reduce over.
Args:
dim (int): Dimension to reduce over.
"""
super(Model, self).__init__()
self.dim = dim
def forward(self, x: torch.Tensor, fn=module_fn) -> torch.Tensor:
"""
Applies sum reduction over the specified dimension.
Args:
x (torch.Tensor): Input tensor of shape (..., dim, ...).
Returns:
torch.Tensor: Output tensor after sum reduction, shape (..., 1, ...).
"""
return fn(x, self.dim)
batch_size = 16
dim1 = 256
dim2 = 256
reduce_dim = 1
def get_inputs():
x = torch.randn(batch_size, dim1, dim2)
return [x]
def get_init_inputs():
return [reduce_dim]
import torch
import torch.nn as nn
class Model(nn.Module):
"""
Simple model that performs sum reduction over a specified dimension.
"""
def __init__(self, dim: int):
"""
Initializes the model with the dimension to reduce over.
Args:
dim (int): Dimension to reduce over.
"""
super(Model, self).__init__()
self.dim = dim
def forward(self, x: torch.Tensor) -> torch.Tensor:
"""
Applies sum reduction over the specified dimension.
Args:
x (torch.Tensor): Input tensor of shape (..., dim, ...).
Returns:
torch.Tensor: Output tensor after sum reduction, shape (..., 1, ...).
"""
return torch.sum(x, dim=self.dim, keepdim=True)
batch_size = 16
dim1 = 256
dim2 = 256
reduce_dim = 1
def get_inputs():
x = torch.randn(batch_size, dim1, dim2)
return [x]
def get_init_inputs():
return [reduce_dim]#include <torch/extension.h>
#include <cuda.h>
#include <cuda_runtime.h>
// This kernel performs sum reduction over a given dimension using warp-level primitives with full unrolling.
// Each warp is responsible for computing one output element by summing over a segment of the reduction dimension.
// The reduction within each warp is fully unrolled using __shfl_down_sync, eliminating the use of shared memory and reducing overhead.
template <typename scalar_t>
__global__ void fully_unrolled_warp_reduce_sum_kernel(
const scalar_t* __restrict__ input,
scalar_t* __restrict__ output,
int64_t reduce_size,
int64_t inner_size,
int64_t total_outputs) {
const int warpSize = 32;
int global_thread_id = blockIdx.x * blockDim.x + threadIdx.x;
int warp_id = global_thread_id / warpSize; // Which warp this thread belongs to
int lane = global_thread_id % warpSize; // Lane index within the warp
int total_warps = (gridDim.x * blockDim.x) / warpSize;
// Each warp computes one output element, with a grid-stride loop over outputs
for (int out_idx = warp_id; out_idx < total_outputs; out_idx += total_warps) {
// Map the 1D output index to (outer, inner) indices
int outer_idx = out_idx / inner_size;
int inner_idx = out_idx % inner_size;
// Base index for the given outer and inner indices
int64_t base = outer_idx * reduce_size * inner_size + inner_idx;
scalar_t sum = 0;
// Each thread accumulates a partial sum from its portion of the reduction dimension
for (int i = lane; i < reduce_size; i += warpSize) {
sum += input[base + i * inner_size];
}
// Fully unroll warp-level reduction using shuffle down
unsigned int mask = 0xffffffff;
sum += __shfl_down_sync(mask, sum, 16);
sum += __shfl_down_sync(mask, sum, 8);
sum += __shfl_down_sync(mask, sum, 4);
sum += __shfl_down_sync(mask, sum, 2);
sum += __shfl_down_sync(mask, sum, 1);
// The first lane writes the final result
if (lane == 0) {
output[out_idx] = sum;
}
}
}
// CUDA wrapper function
torch::Tensor sum_reduce_cuda(torch::Tensor input, int64_t dim) {
// Adjust negative dimensions
if (dim < 0) dim += input.dim();
auto sizes = input.sizes().vec();
int64_t reduce_size = sizes[dim];
// Compute outer and inner sizes
int64_t outer_size = 1;
for (int i = 0; i < dim; i++) {
outer_size *= sizes[i];
}
int64_t inner_size = 1;
for (int i = dim + 1; i < sizes.size(); i++) {
inner_size *= sizes[i];
}
// Set the reduced dimension to 1
sizes[dim] = 1;
auto output = torch::empty(sizes, input.options());
// Total number of outputs is outer_size * inner_size
int64_t total_outputs = outer_size * inner_size;
// Each output element is computed by one warp (32 threads)
const int warpSize = 32;
int total_threads = total_outputs * warpSize;
int threads = 256; // Must be a multiple of 32
int blocks = (total_threads + threads - 1) / threads;
AT_DISPATCH_FLOATING_TYPES(input.scalar_type(), "sum_reduce_cuda", ([&] {
fully_unrolled_warp_reduce_sum_kernel<scalar_t><<<blocks, threads>>>(
input.data_ptr<scalar_t>(),
output.data_ptr<scalar_t>(),
reduce_size,
inner_size,
total_outputs
);
}));
return output;
}
PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
m.def("forward", &sum_reduce_cuda, "Sum reduction forward (CUDA) using fully unrolled warp-level primitives");
}
| Metric | Value | Unit | Variance | Samples |
|---|---|---|---|---|
| Executed Ipc Active | 0.626 | inst/cycle | 0.000 | 5 |
| Executed Ipc Elapsed | 0.462 | inst/cycle | 0.000 | 5 |
| Issue Slots Busy | 16.022 | % | 0.009 | 5 |
| Issued Ipc Active | 0.644 | inst/cycle | 0.000 | 5 |
| SM Busy | 16.022 | % | 0.009 | 5 |
| Memory Throughput | 440144928057.034 | byte/second | 19189658725823696896.000 | 5 |
| Mem Busy | 54.014 | % | 0.268 | 5 |
| Max Bandwidth | 13.176 | % | 0.014 | 5 |
| L1/TEX Hit Rate | 87.482 | % | 0.000 | 5 |
| L2 Hit Rate | 47.544 | % | 0.002 | 5 |
| Mem Pipes Busy | 3.786 | % | 0.001 | 5 |
| Warp Cycles Per Issued Instruction | 41.524 | cycle | 0.246 | 5 |
| Warp Cycles Per Executed Instruction | 42.450 | cycle | 0.259 | 5 |
| Avg. Active Threads Per Warp | 32.000 | 0.000 | 5 | |
| Avg. Not Predicated Off Threads Per Warp | 30.160 | 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 | 6.000 | block | 0.000 | 5 |
| Block Limit Shared Mem | 32.000 | block | 0.000 | 5 |
| Block Limit Warps | 8.000 | block | 0.000 | 5 |
| Theoretical Active Warps per SM | 48.000 | warp | 0.000 | 5 |
| Theoretical Occupancy | 75.000 | % | 0.000 | 5 |
| Achieved Occupancy | 41.290 | % | 0.019 | 5 |
| Achieved Active Warps Per SM | 26.424 | warp | 0.008 | 5 |
| Rule | Description |
|---|---|
| WRN HighPipeUtilization | All compute pipelines are under-utilized. Either this kernel is very small or it doesn't issue enough warps per scheduler. Check the Launch Statistics and Scheduler Statistics sections for further details. |
| 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 (75.0%) is limited by the number of required registers. The difference between calculated theoretical (75.0%) and measured achieved occupancy (41.5%) 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 | 463973.28 | μs |
| Device Time | 344.89 | μs |
| Self CPU Time | 37.67 | 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 | 463935.61 | μs |
| Device Time | 344.89 | μs |
| Self CPU Time | 90.34 | 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::empty_strided | ||
| CPU Time | 463281.89 | μs |
| Device Time | 0.00 | μs |
| Self CPU Time | 70.53 | 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 |
| cudaDeviceGetStreamPriorityRange | ||
| CPU Time | 463001.66 | μs |
| Device Time | 0.00 | μs |
| Self CPU Time | 463001.66 | 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 |
| cudaLaunchKernel | ||
| CPU Time | 563014.21 | μs |
| Device Time | 22219.64 | μs |
| Self CPU Time | 563014.21 | μs |
| Self Device Time | 22219.64 | μs |
| CPU Memory Usage | 0 | B |
| Device Memory Usage | 0 | B |
| Self CPU Memory Usage | 0 | B |
| Self Device Memory Usage | 0 | B |
| void fully_unrolled_warp_reduce_sum_kernel<float>(float const*, float*, long, long, long) | ||
| CPU Time | 0.00 | μs |
| Device Time | 67675.48 | μs |
| Self CPU Time | 0.00 | μs |
| Self Device Time | 67675.48 | μs |
| CPU Memory Usage | 0 | B |
| Device Memory Usage | 0 | B |
| Self CPU Memory Usage | 0 | B |
| Self Device Memory Usage | 0 | B |
| cudaEventRecord | ||
| CPU Time | 18553.05 | μs |
| Device Time | 44019.60 | μs |
| Self CPU Time | 18553.05 | μs |
| Self Device Time | 44019.60 | 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 | 67525.17 | μs |
| Device Time | 657990.82 | μs |
| Self CPU Time | 14991.00 | 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 | 52535.05 | μs |
| Device Time | 657990.82 | μs |
| Self CPU Time | 15991.54 | μs |
| Self Device Time | 657990.82 | 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 | 658069.47 | μs |
| Self CPU Time | 0.00 | μs |
| Self Device Time | 658069.47 | μs |
| CPU Memory Usage | 0 | B |
| Device Memory Usage | 0 | B |
| Self CPU Memory Usage | 0 | B |
| Self Device Memory Usage | 0 | B |
45285 warnings generated when compiling for host. Suppressed 45322 warnings (45275 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.