chore(gpu): add vectorized bitand bench (64 bits)

Author: agnesLeroy

tfhe-benchmark/Cargo.toml

@@ -90,6 +90,12 @@ path = "benches/high_level_api/noise_squash.rs"
 harness = false
 required-features = ["integer", "internal-keycache"]
 
+[[bench]]
+name = "hlapi-arrays"
+path = "benches/high_level_api/arrays.rs"
+harness = false
+required-features = ["integer", "internal-keycache"]
+
 [[bench]]
 name = "glwe_packing_compression-integer-bench"
 path = "benches/integer/glwe_packing_compression.rs"

tfhe-benchmark/benches/high_level_api/arrays.rs

@@ -0,0 +1,59 @@
+use benchmark::utilities::{write_to_json, OperatorType};
+use criterion::{Criterion};
+use rand::prelude::*;
+use benchmark::params_aliases::BENCH_PARAM_GPU_MULTI_BIT_GROUP_4_MESSAGE_2_CARRY_2_KS_PBS_TUNIFORM_2M128;
+use tfhe::keycache::NamedParam;
+use tfhe::prelude::*;
+use tfhe::{
+    ClientKey, CompressedServerKey,
+};
+use tfhe::array::GpuFheUint64Array;
+
+#[cfg(feature = "gpu")]
+fn main() {
+    let cks = {
+        use tfhe::{set_server_key, ConfigBuilder};
+        let config = ConfigBuilder::with_custom_parameters(
+            BENCH_PARAM_GPU_MULTI_BIT_GROUP_4_MESSAGE_2_CARRY_2_KS_PBS_TUNIFORM_2M128,
+        )
+            .build();
+        let cks = ClientKey::generate(config);
+        let compressed_sks = CompressedServerKey::new(&cks);
+
+        set_server_key(compressed_sks.decompress_to_gpu());
+        cks
+    };
+
+    let array_dim = 32;
+    let num_elems = array_dim * array_dim;
+    let mut rng = thread_rng();
+    let clear_xs = (0..num_elems as u64).map(| _ | rng.gen::<u64>()).collect::<Vec<_>>();
+    let clear_ys = (0..num_elems as u64).map(| _ | rng.gen::<u64>()).collect::<Vec<_>>();
+
+    let xs = GpuFheUint64Array::try_encrypt((clear_xs.as_slice(), vec![array_dim, array_dim]), &cks).unwrap();
+    let ys = GpuFheUint64Array::try_encrypt((clear_ys.as_slice(), vec![array_dim, array_dim]), &cks).unwrap();
+
+    let mut c = Criterion::default().configure_from_args();
+    let bench_id = format!("bench::hlapi::array::cuda::bitand::");
+    c.bench_function(&bench_id, |b| {
+
+        b.iter(|| {
+            let _ = &xs & &ys;
+        })
+    });
+
+    let params = cks.computation_parameters();
+
+    write_to_json::<u64, _>(
+        &bench_id,
+        params,
+        params.name(),
+        "erc20-transfer",
+        &OperatorType::Atomic,
+        64,
+        vec![],
+    );
+
+
+    c.final_summary();
+}

tfhe/src/integer/gpu/mod.rs

@@ -7732,9 +7732,7 @@ pub unsafe fn unchecked_bitop_vec_radix_kb_assign<T: UnsignedInteger, B: Numeric
         num_radix_ciphertexts,
     );
     scratch_cuda_integer_radix_bitop_kb_64(
-        streams.ptr.as_ptr(),
-        streams.gpu_indexes_ptr(),
-        streams.len() as u32,
+        streams.ffi(),
         std::ptr::addr_of_mut!(mem_ptr),
         glwe_dimension.0 as u32,
         polynomial_size.0 as u32,
@@ -7754,9 +7752,7 @@ pub unsafe fn unchecked_bitop_vec_radix_kb_assign<T: UnsignedInteger, B: Numeric
         noise_reduction_type as u32,
     );
     cuda_bitop_integer_radix_ciphertext_kb_64(
-        streams.ptr.as_ptr(),
-        streams.gpu_indexes_ptr(),
-        streams.len() as u32,
+        streams.ffi(),
         &raw mut cuda_ffi_radix_lwe_left,
         &raw const cuda_ffi_radix_lwe_left,
         &raw const cuda_ffi_radix_lwe_right,
@@ -7766,9 +7762,7 @@ pub unsafe fn unchecked_bitop_vec_radix_kb_assign<T: UnsignedInteger, B: Numeric
         &raw const ms_noise_reduction_key_ffi,
     );
     cleanup_cuda_integer_bitop(
-        streams.ptr.as_ptr(),
-        streams.gpu_indexes_ptr(),
-        streams.len() as u32,
+        streams.ffi(),
         std::ptr::addr_of_mut!(mem_ptr),
     );
     update_noise_degree(radix_lwe_left, &cuda_ffi_radix_lwe_left);