[Need Help] A simple C2C 1D FFT produces different result with NumPy FFT

[AtomicVar]

I'm new to FFT and I need to use FFT in my Metal project so I found VkFFT a suitable choice. But after I tried a simple 1D C2C FFT on [0, 1, 2, 3, 4, 5, 6, 7], and compared the results with NumPy FFT, they were different.

Here is my simple code:

VkFFTResult res = VKFFT_SUCCESS;
  
// Prepare input data [0, 1, 2, 3, 4, 5, 6, 7]
float* input_buffer = new float[2 * N];
for (int i = 0; i < N; i++) {
  input_buffer[i * 2] = (float)i;
  input_buffer[i * 2 + 1] = 0.0f;
}

// Configuration + FFT application
VkFFTConfiguration configuration = {};
VkFFTApplication app = {};

configuration.FFTdim = 1; //FFT dimension, 1D, 2D or 3D (default 1).
configuration.size[0] = sizeof(float) * N;
configuration.numberBatches = 1;
configuration.device = MTL::CreateSystemDefaultDevice();
configuration.queue = configuration.device->newCommandQueue();

// Prepare Metal Buffer
uint64_t bufferSize = (uint64_t)sizeof(float) * 2 * N;
MTL::Buffer* buffer = configuration.device->newBuffer(input_buffer, bufferSize, MTL::ResourceStorageModeShared);
configuration.buffer = &buffer;
configuration.bufferSize = &bufferSize;

res = initializeVkFFT(&app, configuration);
if (res != VKFFT_SUCCESS) {
  printf("[ERROR] initializeVkFFT returned %d\n", res);
}

// Prepare launch params
VkFFTLaunchParams launchParams = {};
MTL::CommandBuffer* commandBuffer = configuration.queue->commandBuffer();
if (commandBuffer == 0) {
  printf("[ERROR] failed to create command buffer.\n");
}
launchParams.commandBuffer = commandBuffer;
MTL::ComputeCommandEncoder* commandEncoder = commandBuffer->computeCommandEncoder();
if (commandEncoder == 0) {
  printf("[ERROR] failed to create command encoder.\n");
}
launchParams.commandEncoder = commandEncoder;

// encode FFT
res = VkFFTAppend(&app, -1, &launchParams);
if (res != VKFFT_SUCCESS) {
  printf("[ERROR] VkFFTAppend returned %d\n", res);
}
commandEncoder->endEncoding();

// commit and wait
commandBuffer->commit();
commandBuffer->waitUntilCompleted();

// Get outputs
float* output = (float*)buffer->contents();
for (int i = 0; i < N; i++) {
  printf("%f + %fj\n", output[i * 2], output[i * 2 + 1]);
}

// Post-processings
commandEncoder->release();
commandBuffer->release();
buffer->release();
deleteVkFFT(&app);
delete[] input_buffer;

printf("Done.\n");

The output is:

28.000000 + 0.000000j
14.590963 + -22.021717j
-9.137072 + -20.109358j
-16.153057 + -1.033177j
-4.000000 + 9.656855j
6.358437 + 2.874962j
2.380085 + -5.986423j
-5.495201 + -3.378813j

But NumPy FFT gives different output:

import numpy as np
x = np.arange(start=0.0, stop=8.0, dtype=np.float32)
y = np.fft.fft(x)

output:

array([28.+0.j        , -4.+9.65685425j, -4.+4.j        , -4.+1.65685425j,
       -4.+0.j        , -4.-1.65685425j, -4.-4.j        , -4.-9.65685425j])

I don't know what went wrong. Can someone help me? Thanks.

[AtomicVar]

Sorry, I misunderstood the size member of VkFFTConfiguration. After I fixed it, the output is the same. I'm closing this discussion.

// configuration.size[0] = sizeof(float) * N;
configuration.size[0] = N;