Merge branch 'dev' into develop/dima/msm_improvements

wrappers/rust/icicle-core/src/msm/tests.rs

@@ -31,28 +31,21 @@ where
     C::ScalarField: ArkConvertible<ArkEquivalent = <C::ArkSWConfig as ArkCurveConfig>::ScalarField>,
     C::BaseField: ArkConvertible<ArkEquivalent = <C::ArkSWConfig as ArkCurveConfig>::BaseField>,
 {
-    let test_sizes = [4, 8, 16, 32, 64, 128, 256, 1000, 1 << 18];
-    let mut msm_results = HostOrDeviceSlice::cuda_malloc(1).unwrap();
-    for test_size in test_sizes {
-        let points = generate_random_affine_points_with_zeroes(test_size, 2);
-        let scalars = <C::ScalarField as FieldImpl>::Config::generate_random(test_size);
-        let points_ark: Vec<_> = points
-            .iter()
-            .map(|x| x.to_ark())
-            .collect();
-        let scalars_ark: Vec<_> = scalars
-            .iter()
-            .map(|x| x.to_ark())
-            .collect();
-        // if we simply transmute arkworks types, we'll get scalars or points in Montgomery format
-        // (just beware the possible extra flag in affine point types, can't transmute ark Affine because of that)
-        let scalars_mont = unsafe { &*(&scalars_ark[..] as *const _ as *const [C::ScalarField]) };
+    let device_count = get_device_count().unwrap();
+    (0..device_count) // TODO: this is proto-loadbalancer
+        .into_par_iter()
+        .for_each(move |device_id| {
+            //TODO: currently supported multi-GPU workflow:
+            //      1) User starts child host thread from parent host thread
+            //      2) Calls set_device once with selected device_id (0 is default device .. < device_count)
+            //      3) Perform all operations (without changing device on the thread)
+            //      4) If necessary - export results to parent host thread
 
             set_device(device_id).unwrap();
             let test_sizes = [4, 8, 16, 32, 64, 128, 256, 1000, 1 << 18];
             let mut msm_results = HostOrDeviceSlice::cuda_malloc(1).unwrap();
             for test_size in test_sizes {
-                let points = C::generate_random_affine_points(test_size);
+                let points = generate_random_affine_points_with_zeroes(test_size, 2);
                 let scalars = <C::ScalarField as FieldImpl>::Config::generate_random(test_size);
                 let points_ark: Vec<_> = points
                     .iter()