← Back to Leaderboard

The AI CUDA Engineer 👷

2_ShallowWideMLP • warp_uniform_control_flow_base

Level 3 • Task 2
import torch
import torch.nn as nn
import torch.nn.functional as F


def module_fn(
    x: torch.Tensor, weights: nn.ParameterList, biases: nn.ParameterList
) -> torch.Tensor:
    """
    Implements a shallow wide multi-layer perceptron with ReLU activation.

    Args:
        x (torch.Tensor): The input tensor, shape (batch_size, input_size)
        weights (nn.ParameterList): A list of weight tensors for each linear layer
        biases (nn.ParameterList): A list of bias tensors for each linear layer

    Returns:
        torch.Tensor: The output tensor, shape (batch_size, output_size)
    """
    for weight, bias in zip(weights[:-1], biases[:-1]):
        x = F.linear(x, weight, bias)
        x = F.relu(x)
    x = F.linear(x, weights[-1], biases[-1])
    return x


class Model(nn.Module):
    def __init__(self, input_size, hidden_layer_sizes, output_size):
        """
        :param input_size: The number of input features
        :param hidden_layer_sizes: A list of ints containing the sizes of each hidden layer
        :param output_size: The number of output features
        """
        super(Model, self).__init__()

        self.weights = nn.ParameterList()
        self.biases = nn.ParameterList()

        current_input_size = input_size
        for hidden_size in hidden_layer_sizes:
            linear = nn.Linear(current_input_size, hidden_size)
            self.weights.append(nn.Parameter(linear.weight.data.clone()))
            self.biases.append(nn.Parameter(linear.bias.data.clone()))
            current_input_size = hidden_size

        linear = nn.Linear(current_input_size, output_size)
        self.weights.append(nn.Parameter(linear.weight.data.clone()))
        self.biases.append(nn.Parameter(linear.bias.data.clone()))

    def forward(self, x, fn=module_fn):
        return fn(x, self.weights, self.biases)


# Test code
batch_size = 1
input_size = 1000
hidden_layer_sizes = [2000, 2000]  # Example of deep and narrow layers
output_size = 10


def get_inputs():
    return [torch.randn(batch_size, input_size)]


def get_init_inputs():
    return [input_size, hidden_layer_sizes, output_size]

import torch
import torch.nn as nn
import torch.nn.functional as F

class Model(nn.Module):
    def __init__(self, input_size, hidden_layer_sizes, output_size):
        """
        :param input_size: The number of input features
        :param hidden_layer_sizes: A list of ints containing the sizes of each hidden layer
        :param output_size: The number of output features
        """
        super(Model, self).__init__()
        
        layers = []
        current_input_size = input_size
        
        for hidden_size in hidden_layer_sizes:
            layers.append(nn.Linear(current_input_size, hidden_size))
            layers.append(nn.ReLU())
            current_input_size = hidden_size
        
        layers.append(nn.Linear(current_input_size, output_size))
        
        self.network = nn.Sequential(*layers)
    
    def forward(self, x):
        """
        :param x: The input tensor, shape (batch_size, input_size)
        :return: The output tensor, shape (batch_size, output_size)
        """
        return self.network(x)

# Test code
batch_size = 1
input_size = 1000
hidden_layer_sizes = [2000, 2000]  # Example of deep and narrow layers
output_size = 10

def get_inputs():
    return [torch.randn(batch_size, input_size)]

def get_init_inputs():
    return [input_size, hidden_layer_sizes, output_size]
Download Evaluation Download PyTorch Download CUDA Download Profiles

Kernel Information

Operation Name 2_ShallowWideMLP
Level ID 3
Task ID 2
Kernel Name warp_uniform_control_flow_base
CUDA Speedup (Native) 2.134x
CUDA Speedup (Compile) 4.052x
CUDA Runtime 0.043 ms
PyTorch Runtime (Native) 0.092 ms
PyTorch Runtime (Compile) 0.174 ms
Correct True
Max Diff (vs. Reference) 0.000000
Model azure-gpt-4o-2024-08-06
Temperature 0.50
View Experiment Progress Details

Related Kernels (Level 3, Task 2 • 2_ShallowWideMLP)

Rank Kernel Name Runtime (ms) Speedup Native Speedup Compile
🥇 min_sync_warp_base 0.04 2.29 4.36
🥈 optimized_reduction_warp_v2_base 0.04 2.24 4.25
🥈 efficient_mlp_forward_kernel_base 0.04 2.24 4.25
4 warp_index_optimized_base 0.04 2.18 4.15
4 warp_shfl_optimized_base 0.04 2.18 4.15
4 stride_loop_optimization_base 0.04 2.18 4.15
4 warp_dot_base 0.04 2.18 4.15
4 coalesced_vectorized_base 0.04 2.18 4.15
4 tiled_warp_forward_base 0.04 2.18 4.15
4 uniform_control_flow_base_base 0.04 2.18 4.15
4 warp_dot_optimized_block_base 0.04 2.18 4.15
4 tuned_block_size_reduction_v2_base 0.04 2.18 4.15
4 tuned_block_size_reduction_v2_edit_1 0.04 2.18 4.15
14 warp_uniform_control_flow_base 0.04 2.13 4.05
15 block_reduce_mlp_2d_base 0.04 2.09 3.96
16 warp_coalesced_optimized_block_base 0.05 1.99 3.79
17 reduced_shared_warp_edit_1 0.05 1.91 3.63
18 reduced_shared_warp_base 0.05 1.87 3.56
19 hybrid_mlp_forward_base 0.06 1.61 3.06
19 hybrid_mlp_forward_base 0.06 1.61 3.06 
#include <torch/extension.h>
#include <cuda.h>
#include <cuda_runtime.h>

#define NUM_THREADS 128

// Kernel with improved warp uniform control flow
// Each warp computes one output element for a given row

template <typename scalar_t>
__global__ void mlp_forward_kernel_uniform_control(
    const scalar_t* __restrict__ input,
    const scalar_t* __restrict__ weight,
    const scalar_t* __restrict__ bias,
    scalar_t* __restrict__ output,
    const int batch_size,
    const int in_features,
    const int out_features) {

    int threadId = threadIdx.x; 
    int warpId = threadId / 32; 
    int lane = threadId % 32; 

    int batch_row = blockIdx.x; // Process each row in the batch
    int neuron_col = blockIdx.y * blockDim.x / 32 + warpId; // Calculate neuron index

    if (batch_row < batch_size && neuron_col < out_features) {
        scalar_t sum = 0;
        for (int i = lane; i < in_features; i += 32) {
            sum += input[batch_row * in_features + i] * weight[neuron_col * in_features + i];
        }

        for (int offset = 16; offset > 0; offset /= 2) {
            sum += __shfl_down_sync(0xffffffff, sum, offset);
        }

        if (lane == 0) {
            output[batch_row * out_features + neuron_col] = sum + bias[neuron_col];
        }
    }
}

// ReLU kernel with stride loop optimization

template <typename scalar_t>
__global__ void relu_kernel_index(
    scalar_t* __restrict__ data,
    const int size) {
    int idx = blockIdx.x * blockDim.x + threadIdx.x;
    int stride = blockDim.x * gridDim.x;
    for (int i = idx; i < size; i += stride) {
        data[i] = data[i] > 0 ? data[i] : 0;
    }
}

// Main forward function applying the MLP using the improved kernel

torch::Tensor mlp_cuda_forward(
    torch::Tensor input,
    std::vector<torch::Tensor> weights,
    std::vector<torch::Tensor> biases) {

    auto device = input.device();
    auto num_layers = weights.size();
    auto current = input;

    for (size_t i = 0; i < num_layers; i++) {
        const auto batch_size = current.size(0);
        const auto in_features = current.size(1);
        const auto out_features = weights[i].size(0);

        auto output = torch::empty({batch_size, out_features}, 
                                 torch::dtype(current.dtype()).device(device));

        dim3 block(NUM_THREADS);
        dim3 grid(batch_size, (out_features + NUM_THREADS / 32 - 1) / (NUM_THREADS / 32));

        AT_DISPATCH_FLOATING_TYPES(current.type(), "mlp_forward_kernel_uniform_control", ([&] {
            mlp_forward_kernel_uniform_control<scalar_t><<<grid, block>>>(
                current.data_ptr<scalar_t>(),
                weights[i].data_ptr<scalar_t>(),
                biases[i].data_ptr<scalar_t>(),
                output.data_ptr<scalar_t>(),
                batch_size,
                in_features,
                out_features
            );
        }));

        if (i < num_layers - 1) {
            const int size = batch_size * out_features;
            const int thread_per_block = 256;
            const int num_blocks = (size + thread_per_block - 1) / thread_per_block;

            AT_DISPATCH_FLOATING_TYPES(output.type(), "relu_kernel_index", ([&] {
                relu_kernel_index<scalar_t><<<num_blocks, thread_per_block>>>(
                    output.data_ptr<scalar_t>(),
                    size
                );
            }));
        }
        
        current = output;
    }

    return current;
}

PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
    m.def("forward", &mlp_cuda_forward, "MLP forward (CUDA warp uniform control)");
}
Performance Metrics
Metric Value Unit Variance Samples
Executed Ipc Active 0.218 inst/cycle 0.001 5
Executed Ipc Elapsed 0.000 inst/cycle 0.000 5
Issue Slots Busy 6.102 % 0.410 5
Issued Ipc Active 0.244 inst/cycle 0.001 5
SM Busy 6.102 % 0.410 5
Memory Throughput 3466119884.664 byte/second 47339296222217672.000 5
Mem Busy 9.714 % 0.252 5
Max Bandwidth 5.000 % 0.056 5
L1/TEX Hit Rate 50.000 % 0.000 5
L2 Hit Rate 100.808 % 0.038 5
Mem Pipes Busy 0.132 % 0.000 5
Warp Cycles Per Issued Instruction 33.914 cycle 5.824 5
Warp Cycles Per Executed Instruction 38.032 cycle 7.318 5
Avg. Active Threads Per Warp 31.800 0.000 5
Avg. Not Predicated Off Threads Per Warp 27.370 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 32.000 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 12.112 % 0.010 5
Achieved Active Warps Per SM 7.754 warp 0.004 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 (12.2%) 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 388403.69 μs
Device Time 2255.86 μs
Self CPU Time 72.15 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 388331.54 μs
Device Time 2255.86 μs
Self CPU Time 140.31 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 385471.19 μs
Device Time 0.00 μs
Self CPU Time 216.91 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 371188.60 μs
Device Time 0.00 μs
Self CPU Time 371188.60 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 650948.81 μs
Device Time 70220.06 μs
Self CPU Time 650948.81 μs
Self Device Time 70220.06 μs
CPU Memory Usage 0 B
Device Memory Usage 0 B
Self CPU Memory Usage 0 B
Self Device Memory Usage 0 B
void mlp_forward_kernel_uniform_control<float>(float const*, float const*, float const*, float*, int, int, int)
CPU Time 0.00 μs
Device Time 246601.46 μs
Self CPU Time 0.00 μs
Self Device Time 246601.46 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 65474.43 μs
Device Time 603447.43 μs
Self CPU Time 13852.98 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 51623.38 μs
Device Time 603447.43 μs
Self CPU Time 16169.32 μs
Self Device Time 603447.43 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 603447.43 μs
Self CPU Time 0.00 μs
Self Device Time 603447.43 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 
45295 warnings generated when compiling for host.
Suppressed 45323 warnings (45276 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/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:13:5 bugprone-easily-swappable-parameters
13 | const scalar_t* __restrict__ weight,
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
14 | const scalar_t* __restrict__ bias,
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/robert_sakana_ai/llm_cuda/experiments/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:13:34: note: the first parameter in the range is 'weight'
13 | const scalar_t* __restrict__ weight,
| ^~~~~~
/home/robert_sakana_ai/llm_cuda/experiments/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:14:34: note: the last parameter in the range is 'bias'
14 | const scalar_t* __restrict__ bias,
| ^~~~
/home/robert_sakana_ai/llm_cuda/experiments/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:16:5: warning: 2 adjacent parameters of 'mlp_forward_kernel_uniform_control' of similar type ('const int') are easily swapped by mistake [bugprone-easily-swappable-parameters]
16 | const int batch_size,
| ^~~~~~~~~~~~~~~~~~~~~
17 | const int in_features,
| ~~~~~~~~~~~~~~~~~~~~~
/home/robert_sakana_ai/llm_cuda/experiments/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:16:15: note: the first parameter in the range is 'batch_size'
16 | const int batch_size,
| ^~~~~~~~~~
/home/robert_sakana_ai/llm_cuda/experiments/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:17:15: note: the last parameter in the range is 'in_features'
17 | const int in_features,
| ^~~~~~~~~~~
/home/robert_sakana_ai/llm_cuda/experiments/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:20:20: warning: narrowing conversion from 'unsigned int' to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
20 | int threadId = threadIdx.x;
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:24:21: warning: narrowing conversion from 'unsigned int' to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
24 | int batch_row = blockIdx.x; // Process each row in the batch
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:25:22: warning: narrowing conversion from 'unsigned int' to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
25 | int neuron_col = blockIdx.y * blockDim.x / 32 + warpId; // Calculate neuron index
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:49:15: warning: narrowing conversion from 'unsigned int' to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
49 | int idx = blockIdx.x * blockDim.x + threadIdx.x;
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:50:18: warning: narrowing conversion from 'unsigned int' to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
50 | int stride = blockDim.x * gridDim.x;
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:59:19: warning: the parameter 'input' is copied for each invocation but only used as a const reference; consider making it a const reference [performance-unnecessary-value-param]
59 | torch::Tensor input,
| ^
| const &
/home/robert_sakana_ai/llm_cuda/experiments/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:78:9: 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]
78 | AT_DISPATCH_FLOATING_TYPES(current.type(), "mlp_forward_kernel_uniform_control", ([&] {
| ^
/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/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:78:9: warning: 'scalar_type' is deprecated: passing at::DeprecatedTypeProperties to an AT_DISPATCH macro is deprecated, pass an at::ScalarType instead [clang-diagnostic-deprecated-declarations]
78 | AT_DISPATCH_FLOATING_TYPES(current.type(), "mlp_forward_kernel_uniform_control", ([&] {
| ^
/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/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:78:44: 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]
78 | AT_DISPATCH_FLOATING_TYPES(current.type(), "mlp_forward_kernel_uniform_control", ([&] {
| ^
/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)]]
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:91:30: warning: narrowing conversion from 'int64_t' (aka 'long') to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
91 | const int size = batch_size * out_features;
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:95:13: 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]
95 | AT_DISPATCH_FLOATING_TYPES(output.type(), "relu_kernel_index", ([&] {
| ^
/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/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:95:13: warning: 'scalar_type' is deprecated: passing at::DeprecatedTypeProperties to an AT_DISPATCH macro is deprecated, pass an at::ScalarType instead [clang-diagnostic-deprecated-declarations]
95 | AT_DISPATCH_FLOATING_TYPES(output.type(), "relu_kernel_index", ([&] {
| ^
/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/20250203_optimize_b10_s4_e0_sweep/level_3/task_2/b6_s0_warp_uniform_control_flow/base/base.cu:95:47: 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]
95 | AT_DISPATCH_FLOATING_TYPES(output.type(), "relu_kernel_index", ([&] {
| ^
/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)]]
| ^