feat(gpu): Implement fft128 in cuda backend

Author: bbarbakadze

_typos.toml

@@ -13,3 +13,8 @@ extend-ignore-identifiers-re = [
     # Example in trivium
     "C9217BA0D762ACA1"
 ]
+
+[files]
+extend-exclude = [
+    "backends/tfhe-cuda-backend/cuda/src/fft128/twiddles.cu"
+]

backends/tfhe-cuda-backend/build.rs

@@ -62,6 +62,7 @@ fn main() {
             "cuda/include/integer/integer.h",
             "cuda/include/keyswitch.h",
             "cuda/include/linear_algebra.h",
+            "cuda/include/fft/fft128.h",
             "cuda/include/pbs/programmable_bootstrap.h",
             "cuda/include/pbs/programmable_bootstrap_multibit.h",
         ];

backends/tfhe-cuda-backend/cuda/include/device.h

@@ -52,7 +52,7 @@ void *cuda_malloc_async(uint64_t size, cudaStream_t stream, uint32_t gpu_index);
 
 void cuda_check_valid_malloc(uint64_t size, uint32_t gpu_index);
 
-void cuda_memcpy_async_to_gpu(void *dest, void *src, uint64_t size,
+void cuda_memcpy_async_to_gpu(void *dest, const void *src, uint64_t size,
                               cudaStream_t stream, uint32_t gpu_index);
 
 void cuda_memcpy_async_gpu_to_gpu(void *dest, void const *src, uint64_t size,

backends/tfhe-cuda-backend/cuda/include/fft/fft128.h

@@ -0,0 +1,17 @@
+#include <stdint.h>
+extern "C" {
+void cuda_fourier_transform_forward_as_torus_f128_async(
+    void *stream, uint32_t gpu_index, void *re0, void *re1, void *im0,
+    void *im1, void const *standard, uint32_t const N,
+    const uint32_t number_of_samples);
+
+void cuda_fourier_transform_forward_as_integer_f128_async(
+    void *stream, uint32_t gpu_index, void *re0, void *re1, void *im0,
+    void *im1, void const *standard, uint32_t const N,
+    const uint32_t number_of_samples);
+
+void cuda_fourier_transform_backward_as_torus_f128_async(
+    void *stream, uint32_t gpu_index, void *standard, void const *re0,
+    void const *re1, void const *im0, void const *im1, uint32_t const N,
+    const uint32_t number_of_samples);
+}

backends/tfhe-cuda-backend/cuda/src/device.cu

@@ -135,7 +135,7 @@ bool cuda_check_support_thread_block_clusters() {
 }
 
 /// Copy memory to the GPU asynchronously
-void cuda_memcpy_async_to_gpu(void *dest, void *src, uint64_t size,
+void cuda_memcpy_async_to_gpu(void *dest, const void *src, uint64_t size,
                               cudaStream_t stream, uint32_t gpu_index) {
   if (size == 0)
     return;