← Back to Leaderboard

The AI CUDA Engineer 👷

29_Softplus • optimized_softplus_base

Level 1 • Task 29
import torch
import torch.nn as nn
import torch.nn.functional as F


def module_fn(x: torch.Tensor) -> torch.Tensor:
    """
    Applies Softplus activation to the input tensor.

    Args:
        x (torch.Tensor): Input tensor of any shape.

    Returns:
        torch.Tensor: Output tensor with Softplus applied, same shape as input.
    """
    return F.softplus(x)


class Model(nn.Module):
    """
    Simple model that performs a Softplus activation.
    """

    def __init__(self):
        super(Model, self).__init__()

    def forward(self, x: torch.Tensor, fn=module_fn) -> torch.Tensor:
        return fn(x)


batch_size = 16
dim = 16384


def get_inputs():
    x = torch.randn(batch_size, dim)
    return [x]


def get_init_inputs():
    return []  # No special initialization inputs needed

import torch
import torch.nn as nn

class Model(nn.Module):
    """
    Simple model that performs a Softplus activation.
    """
    def __init__(self):
        super(Model, self).__init__()
    
    def forward(self, x: torch.Tensor) -> torch.Tensor:
        """
        Applies Softplus activation to the input tensor.

        Args:
            x (torch.Tensor): Input tensor of any shape.

        Returns:
            torch.Tensor: Output tensor with Softplus applied, same shape as input.
        """
        return torch.nn.functional.softplus(x)

batch_size = 16
dim = 16384

def get_inputs():
    x = torch.randn(batch_size, dim)
    return [x]

def get_init_inputs():
    return []  # No special initialization inputs needed
Download Evaluation Download PyTorch Download CUDA Download Profiles

Kernel Information

Operation Name 29_Softplus
Level ID 1
Task ID 29
Kernel Name optimized_softplus_base
CUDA Speedup (Native) 0.993x
CUDA Speedup (Compile) 4.180x
CUDA Runtime 0.007 ms
PyTorch Runtime (Native) 0.007 ms
PyTorch Runtime (Compile) 0.029 ms
Correct True
Max Diff (vs. Reference) 0.000000
Model azure-gpt-4o-2024-08-06
Temperature 1.00
View Experiment Progress Details

Related Kernels (Level 1, Task 29 • 29_Softplus)

Rank Kernel Name Runtime (ms) Speedup Native Speedup Compile
🥇 softplus_modular_base_base 0.01 1.16 4.88
🥇 warp_and_alignment_optimized_softplus_edit_1 0.01 1.16 4.88
🥇 branchless_softplus_edit_1 0.01 1.16 4.88
🥇 warp_optimized_softplus_base 0.01 1.16 4.88
5 softplus_unrolled_base_base 0.01 0.99 4.18
5 softplus_coalesced_base 0.01 0.99 4.18
5 softplus_2d_block_thread_base 0.01 0.99 4.18
5 optimized_softplus_cuda_base 0.01 0.99 4.18
5 softplus_coalesced_memory_access_base 0.01 0.99 4.18
5 softplus_tuned_indexing_base_base 0.01 0.99 4.18
5 softplus_blockstride_base 0.01 0.99 4.18
5 softplus_loop_unroll_base_base 0.01 0.99 4.18
5 softplus_branchless_base 0.01 0.99 4.18
5 softplus_blocksize_experiment_base 0.01 0.99 4.18
5 softplus_unrolled_base 0.01 0.99 4.18
5 softplus_constant_memory_base_base 0.01 0.99 4.18
5 optimized_softplus_cuda_base 0.01 0.99 4.18
5 29_Softplus 0.01 0.99 4.18
5 softplus_constant_memory_base 0.01 0.99 4.18
5 optimized_softplus_base 0.01 0.99 4.18 
#include <torch/extension.h>
#include <cuda.h>
#include <cuda_runtime.h>

// Defined optimal block size for H100 GPU
const int OPTIMAL_THREADS = 512;

template <typename scalar_t>
__global__ void softplus_kernel(
    const scalar_t* __restrict__ input,
    scalar_t* __restrict__ output,
    const int size) {
    
    const int idx = blockIdx.x * blockDim.x + threadIdx.x;
    
    if (idx < size) {
        const scalar_t x = input[idx];
        if (x > 20.0) {
            output[idx] = x;
        } else if (x < -20.0) {
            output[idx] = exp(x);
        } else {
            output[idx] = log1p(exp(x));
        }
    }
}

torch::Tensor softplus_cuda_forward(torch::Tensor input) {
    auto output = torch::empty_like(input);
    const int size = input.numel();
    const int blocks = (size + OPTIMAL_THREADS - 1) / OPTIMAL_THREADS;

    AT_DISPATCH_FLOATING_TYPES(input.type(), "softplus_forward_cuda", ([&] {
        softplus_kernel<scalar_t><<<blocks, OPTIMAL_THREADS>>>(
            input.data_ptr<scalar_t>(),
            output.data_ptr<scalar_t>(),
            size);
    }));

    return output;
}

PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
    m.def("forward", &softplus_cuda_forward, "Softplus forward (CUDA)");
}
Performance Metrics
Metric Value Unit Variance Samples
Executed Ipc Active 1.236 inst/cycle 0.001 5
Executed Ipc Elapsed 0.512 inst/cycle 0.000 5
Issue Slots Busy 32.696 % 0.741 5
Issued Ipc Active 1.308 inst/cycle 0.001 5
SM Busy 32.696 % 0.741 5
Memory Throughput 283404369679.568 byte/second 7042058468524385280.000 5
Mem Busy 13.498 % 0.017 5
Max Bandwidth 12.496 % 0.018 5
L1/TEX Hit Rate 0.000 % 0.000 5
L2 Hit Rate 66.904 % 0.040 5
Mem Pipes Busy 10.476 % 0.012 5
Warp Cycles Per Issued Instruction 39.532 cycle 0.173 5
Warp Cycles Per Executed Instruction 41.852 cycle 0.194 5
Avg. Active Threads Per Warp 32.000 0.000 5
Avg. Not Predicated Off Threads Per Warp 30.000 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 8.000 block 0.000 5
Block Limit Shared Mem 16.000 block 0.000 5
Block Limit Warps 4.000 block 0.000 5
Theoretical Active Warps per SM 64.000 warp 0.000 5
Theoretical Occupancy 100.000 % 0.000 5
Achieved Occupancy 83.402 % 0.083 5
Achieved Active Warps Per SM 53.376 warp 0.034 5
Analysis Rules
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 is not impacted by any block limit. The difference between calculated theoretical (100.0%) and measured achieved occupancy (83.3%) 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 234393.85 μs
Device Time 39.97 μs
Self CPU Time 35.88 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 234357.97 μs
Device Time 39.97 μs
Self CPU Time 86.67 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::empty_strided
CPU Time 253904.02 μs
Device Time 0.00 μs
Self CPU Time 20000.53 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 233713.60 μs
Device Time 0.00 μs
Self CPU Time 233713.60 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 516984.95 μs
Device Time 23391.38 μs
Self CPU Time 516984.95 μs
Self Device Time 23391.38 μs
CPU Memory Usage 0 B
Device Memory Usage 0 B
Self CPU Memory Usage 0 B
Self Device Memory Usage 0 B
void softplus_kernel<float>(float const*, float*, int)
CPU Time 0.00 μs
Device Time 35040.05 μs
Self CPU Time 0.00 μs
Self Device Time 35040.05 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 23545.11 μs
Device Time 45053.82 μs
Self CPU Time 23545.11 μs
Self Device Time 45053.82 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 68382.53 μs
Device Time 665737.59 μs
Self CPU Time 13983.75 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 54402.10 μs
Device Time 665737.59 μs
Self CPU Time 17688.70 μs
Self Device Time 665737.59 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 665737.59 μs
Self CPU Time 0.00 μs
Self Device Time 665737.59 μs
CPU Memory Usage 0 B
Device Memory Usage 0 B
Self CPU Memory Usage 0 B
Self Device Memory Usage 0 B
Status: Completed 
45281 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.
/home/robert_sakana_ai/llm_cuda/experiments/20250202_optimize_b10_s4_e0_sweep/level_1/task_29/b1_s1_optimized_softplus/base/base.cu:14:21 bugprone-narrowing-conversions
14 | const int idx = blockIdx.x * blockDim.x + threadIdx.x;
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250202_optimize_b10_s4_e0_sweep/level_1/task_29/b1_s1_optimized_softplus/base/base.cu:30:22: warning: narrowing conversion from 'int64_t' (aka 'long') to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
30 | const int size = input.numel();
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250202_optimize_b10_s4_e0_sweep/level_1/task_29/b1_s1_optimized_softplus/base/base.cu:33:5: warning: inside a lambda, '__func__' expands to the name of the function call operator; consider capturing the name of the enclosing function explicitly [bugprone-lambda-function-name]
33 | AT_DISPATCH_FLOATING_TYPES(input.type(), "softplus_forward_cuda", ([&] {
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/ATen/Dispatch.h:237:34: note: expanded from macro 'AT_DISPATCH_FLOATING_TYPES'
237 | AT_DISPATCH_SWITCH(TYPE, NAME, AT_DISPATCH_CASE_FLOATING_TYPES(__VA_ARGS__))
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/ATen/Dispatch.h:233:3: note: expanded from macro 'AT_DISPATCH_CASE_FLOATING_TYPES'
233 | AT_DISPATCH_CASE(at::ScalarType::Double, __VA_ARGS__) \
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/ATen/Dispatch.h:74:3: note: expanded from macro 'AT_DISPATCH_CASE'
74 | AT_PRIVATE_CASE_TYPE_USING_HINT(enum_type, scalar_t, __VA_ARGS__)
| ^
note: (skipping 1 expansions in backtrace; use -fmacro-backtrace-limit=0 to see all)
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/ATen/Dispatch.h:58:7: note: expanded from macro 'AT_PRIVATE_CHECK_SELECTIVE_BUILD'
58 | AT_ERROR( \
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/c10/util/Exception.h:711:32: note: expanded from macro 'AT_ERROR'
711 | C10_EXPAND_MSVC_WORKAROUND(TORCH_CHECK(false, ::c10::str(__VA_ARGS__))); \
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/c10/util/Exception.h:536:9: note: expanded from macro 'TORCH_CHECK'
536 | __func__, \
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250202_optimize_b10_s4_e0_sweep/level_1/task_29/b1_s1_optimized_softplus/base/base.cu:33:5: warning: 'scalar_type' is deprecated: passing at::DeprecatedTypeProperties to an AT_DISPATCH macro is deprecated, pass an at::ScalarType instead [clang-diagnostic-deprecated-declarations]
33 | AT_DISPATCH_FLOATING_TYPES(input.type(), "softplus_forward_cuda", ([&] {
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/ATen/Dispatch.h:237:3: note: expanded from macro 'AT_DISPATCH_FLOATING_TYPES'
237 | AT_DISPATCH_SWITCH(TYPE, NAME, AT_DISPATCH_CASE_FLOATING_TYPES(__VA_ARGS__))
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/ATen/Dispatch.h:218:36: note: expanded from macro 'AT_DISPATCH_SWITCH'
218 | at::ScalarType _st = ::detail::scalar_type(the_type); \
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/ATen/Dispatch.h:106:1: note: 'scalar_type' has been explicitly marked deprecated here
106 | C10_DEPRECATED_MESSAGE(
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/c10/util/Deprecated.h:24:43: note: expanded from macro 'C10_DEPRECATED_MESSAGE'
24 | #define C10_DEPRECATED_MESSAGE(message) [[deprecated(message)]]
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250202_optimize_b10_s4_e0_sweep/level_1/task_29/b1_s1_optimized_softplus/base/base.cu:33:38: warning: 'type' is deprecated: Tensor.type() is deprecated. Instead use Tensor.options(), which in many cases (e.g. in a constructor) is a drop-in replacement. If you were using data from type(), that is now available from Tensor itself, so instead of tensor.type().scalar_type(), use tensor.scalar_type() instead and instead of tensor.type().backend() use tensor.device(). [clang-diagnostic-deprecated-declarations]
33 | AT_DISPATCH_FLOATING_TYPES(input.type(), "softplus_forward_cuda", ([&] {
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/ATen/core/TensorBody.h:224:3: note: 'type' has been explicitly marked deprecated here
224 | C10_DEPRECATED_MESSAGE("Tensor.type() is deprecated. Instead use Tensor.options(), which in many cases (e.g. in a constructor) is a drop-in replacement. If you were using data from type(), that is now available from Tensor itself, so instead of tensor.type().scalar_type(), use tensor.scalar_type() instead and instead of tensor.type().backend() use tensor.device().")
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/c10/util/Deprecated.h:24:43: note: expanded from macro 'C10_DEPRECATED_MESSAGE'
24 | #define C10_DEPRECATED_MESSAGE(message) [[deprecated(message)]]
| ^