```python
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.cpp_extension import load_inline


_cuda_source = r"""
#include <torch/extension.h>
#include <ATen/cuda/CUDAContext.h>
#include <c10/cuda/CUDAGuard.h>
#include <c10/cuda/CUDAException.h>
#include <cuda_runtime.h>
#include <algorithm>

__device__ __forceinline__ float transform(
    float x, float subtract_value, float multiply_value
) {
    float y = __fmul_rn(__fsub_rn(x, subtract_value), multiply_value);
    return y < 0.0f ? 0.0f : y;
}

__global__ void fused_sub_mul_relu_kernel(
    const float* __restrict__ input,
    float* __restrict__ output,
    int64_t size,
    float subtract_value,
    float multiply_value
) {
    const int64_t index =
        static_cast<int64_t>(blockIdx.x) * blockDim.x + threadIdx.x;
    const int64_t stride =
        static_cast<int64_t>(gridDim.x) * blockDim.x;
    const int64_t vector_count = size / 4;

    const float4* input4 = reinterpret_cast<const float4*>(input);
    float4* output4 = reinterpret_cast<float4*>(output);

    for (int64_t i = index; i < vector_count; i += stride) {
        float4 value = input4[i];
        value.x = transform(value.x, subtract_value, multiply_value);
        value.y = transform(value.y, subtract_value, multiply_value);
        value.z = transform(value.z, subtract_value, multiply_value);
        value.w = transform(value.w, subtract_value, multiply_value);
        output4[i] = value;
    }

    const int64_t tail_index = vector_count * 4 + index;
    if (tail_index < size) {
        output[tail_index] = transform(
            input[tail_index], subtract_value, multiply_value
        );
    }
}

torch::Tensor fused_sub_mul_relu_cuda(
    torch::Tensor input,
    double subtract_value,
    double multiply_value
) {
    TORCH_CHECK(input.is_cuda(), "Expected a CUDA tensor");
    TORCH_CHECK(input.scalar_type() == at::kFloat, "Expected float32");
    TORCH_CHECK(input.is_contiguous(), "Expected contiguous input");

    const c10::cuda::CUDAGuard device_guard(input.device());
    auto output = torch::empty_like(input);
    const int64_t size = input.numel();

    if (size == 0) {
        return output;
    }

    constexpr int threads = 256;
    const int64_t work_items = (size + 3) / 4;
    const int blocks = static_cast<int>(
        std::min<int64_t>((work_items + threads - 1) / threads, 4096)
    );

    fused_sub_mul_relu_kernel<<<
        blocks, threads, 0, at::cuda::getCurrentCUDAStream()
    >>>(
        input.data_ptr<float>(),
        output.data_ptr<float>(),
        size,
        static_cast<float>(subtract_value),
        static_cast<float>(multiply_value)
    );

    C10_CUDA_KERNEL_LAUNCH_CHECK();
    return output;
}
"""

_fused_ops = load_inline(
    name="linear_sub_mul_relu_fused_cuda",
    cpp_sources=r"""
torch::Tensor fused_sub_mul_relu_cuda(
    torch::Tensor input,
    double subtract_value,
    double multiply_value
);
""",
    cuda_sources=_cuda_source,
    functions=["fused_sub_mul_relu_cuda"],
    extra_cflags=["-O3"],
    extra_cuda_cflags=["-O3"],
    verbose=False,
)


class ModelNew(nn.Module):
    def __init__(
        self, in_features, out_features, subtract_value, multiply_value
    ):
        super().__init__()
        self.linear = nn.Linear(in_features, out_features)
        self.subtract_value = subtract_value
        self.multiply_value = multiply_value

    def forward(self, x):
        y = F.linear(x, self.linear.weight, self.linear.bias)
        if y.is_cuda and y.dtype == torch.float32:
            return _FusedActivation.apply(
                y, self.subtract_value, self.multiply_value
            )
        return torch.relu(
            (y - self.subtract_value) * self.multiply_value
        )


class _FusedActivation(torch.autograd.Function):
    @staticmethod
    def forward(ctx, x, subtract_value, multiply_value):
        output = _fused_ops.fused_sub_mul_relu_cuda(
            x.contiguous(), float(subtract_value), float(multiply_value)
        )
        if ctx.needs_input_grad[0]:
            ctx.save_for_backward(output)
            ctx.multiply_value = float(multiply_value)
        return output

    @staticmethod
    def backward(ctx, grad_output):
        output, = ctx.saved_tensors
        grad_input = grad_output.masked_fill(output == 0, 0)
        grad_input = grad_input * ctx.multiply_value
        return grad_input, None, None
```