feat(gpu): add support for GPU-accelerated expand on the HL Api

Author: pdroalves

Makefile

@@ -839,7 +839,7 @@ test_high_level_api: install_rs_build_toolchain
 
 test_high_level_api_gpu: install_rs_build_toolchain install_cargo_nextest
 	RUSTFLAGS="$(RUSTFLAGS)" cargo $(CARGO_RS_BUILD_TOOLCHAIN) nextest run --cargo-profile $(CARGO_PROFILE) \
-		--features=integer,internal-keycache,gpu -p $(TFHE_SPEC) \
+		--features=integer,internal-keycache,gpu,zk-pok -p $(TFHE_SPEC) \
 		-E "test(/high_level_api::.*gpu.*/)"
 
 .PHONY: test_strings # Run the tests for strings ci

tfhe/src/core_crypto/gpu/entities/lwe_ciphertext_list.rs

@@ -7,7 +7,7 @@ use crate::core_crypto::prelude::{
 use tfhe_cuda_backend::cuda_bind::cuda_memcpy_async_gpu_to_gpu;
 
 /// A structure representing a vector of LWE ciphertexts with 64 bits of precision on the GPU.
-#[derive(Debug)]
+#[derive(Clone, Debug)]
 pub struct CudaLweCiphertextList<T: UnsignedInteger>(pub(crate) CudaLweList<T>);
 
 #[allow(dead_code)]

tfhe/src/core_crypto/gpu/entities/lwe_compact_ciphertext_list.rs

@@ -10,6 +10,10 @@ use crate::core_crypto::prelude::{
 pub struct CudaLweCompactCiphertextList<T: UnsignedInteger>(pub CudaLweList<T>);
 
 impl<T: UnsignedInteger> CudaLweCompactCiphertextList<T> {
+    pub fn duplicate(&self, streams: &CudaStreams) -> Self {
+        Self(self.0.duplicate(streams))
+    }
+
     pub fn from_lwe_compact_ciphertext_list<C: Container<Element = T>>(
         h_ct: &LweCompactCiphertextList<C>,
         streams: &CudaStreams,

tfhe/src/core_crypto/gpu/entities/lwe_keyswitch_key.rs

@@ -11,6 +11,7 @@ use crate::core_crypto::prelude::{
 };
 
 #[allow(dead_code)]
+#[derive(Clone)]
 pub struct CudaLweKeyswitchKey<T: UnsignedInteger> {
     pub(crate) d_vec: CudaVec<T>,
     input_lwe_size: LweSize,

tfhe/src/core_crypto/gpu/mod.rs

@@ -914,7 +914,7 @@ pub unsafe fn fourier_transform_backward_as_torus_f128_async<T: UnsignedInteger>
     );
 }
 
-#[derive(Debug)]
+#[derive(Clone, Debug)]
 pub struct CudaLweList<T: UnsignedInteger> {
     // Pointer to GPU data
     pub d_vec: CudaVec<T>,
@@ -926,6 +926,17 @@ pub struct CudaLweList<T: UnsignedInteger> {
     pub ciphertext_modulus: CiphertextModulus<T>,
 }
 
+impl<T: UnsignedInteger> CudaLweList<T> {
+    pub fn duplicate(&self, streams: &CudaStreams) -> Self {
+        Self {
+            d_vec: self.d_vec.duplicate(streams),
+            lwe_ciphertext_count: self.lwe_ciphertext_count,
+            lwe_dimension: self.lwe_dimension,
+            ciphertext_modulus: self.ciphertext_modulus,
+        }
+    }
+}
+
 #[derive(Debug, Clone)]
 pub struct CudaGlweList<T: UnsignedInteger> {
     // Pointer to GPU data

tfhe/src/high_level_api/backward_compatibility/compact_list.rs

@@ -3,6 +3,9 @@ use tfhe_versionable::{Upgrade, Version, VersionsDispatch};
 
 use crate::{CompactCiphertextList, Tag};
 
+#[cfg(feature = "zk-pok")]
+use crate::ProvenCompactCiphertextList;
+
 #[derive(Version)]
 pub struct CompactCiphertextListV0(crate::integer::ciphertext::CompactCiphertextList);
 
@@ -17,9 +20,6 @@ impl Upgrade<CompactCiphertextList> for CompactCiphertextListV0 {
     }
 }
 
-#[cfg(feature = "zk-pok")]
-use crate::ProvenCompactCiphertextList;
-
 #[derive(VersionsDispatch)]
 pub enum CompactCiphertextListVersions {
     V0(CompactCiphertextListV0),