Keep necsim-rust-analysis Rust-code-only up-to-date with necsim-rust

Cargo.toml.patch

@@ -1,17 +1,17 @@
 diff --git a/necsim-rust/Cargo.toml b/necsim-rust/Cargo.toml
 --- a/necsim-rust/Cargo.toml
 +++ b/necsim-rust/Cargo.toml
-@@ -36,6 +36,13 @@ members = [
-     "rust-cuda",
-     "rust-cuda/rust-cuda-derive",
-
+@@ -31,6 +31,13 @@ members = [
+     "rustcoalescence/algorithms/cuda",
+     "rustcoalescence/algorithms/cuda/gpu-kernel",
+     "rustcoalescence/algorithms/cuda/cpu-kernel",
++
 +    "analysis/rng/randomness",
 +    "analysis/rng/hash",
 +    "analysis/rng/correlation",
 +    "analysis/performance/exponential",
 +    "analysis/performance/exponential/kernel",
 +    "analysis/performance/reporting",
-+
-     "third-party/array2d-no-std",
-     "third-party/contracts",
-     "third-party/float-next-after-no-std",
+ ]
+
+ default-members = [

analysis/performance/exponential/Cargo.toml

@@ -8,14 +8,13 @@ edition = "2018"
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
+analysis-performance-exponential-kernel = { path = "kernel" }
+
 necsim-core = { path = "../../../necsim/core", features = ["cuda"] }
-necsim-core-bond = { path = "../../../necsim/core/bond", features = ["cuda"] }
+necsim-core-bond = { path = "../../../necsim/core/bond" }
+necsim-core-maths = { path = "../../../necsim/core/maths" }
 necsim-impls-no-std = { path = "../../../necsim/impls/no-std" }
 
-rust-cuda = { path = "../../../rust-cuda", features = ["host"] }
-
-contracts = { path = "../../../third-party/contracts" }
-structopt = "0.3.21"
+rust-cuda = { git = "https://github.com/MomoLangenstein/rust-cuda", branch = "main", features = ["host"] }
 
-[build-dependencies]
-ptx-builder = { path = "../../../third-party/rust-ptx-builder" }
+structopt = "0.3"

analysis/performance/exponential/kernel/.cargo/config.toml

@@ -2,7 +2,7 @@
 pipelining = false
 
 [target.nvptx64-nvidia-cuda]
-rustflags = ["-Clink-args=--arch sm_35", "-Cpanic=abort", "-Clto=no", "-Clink-arg=-Olto"]
+rustflags = ["-Clink-args=--arch=sm_35", "-Cpanic=abort", "-Clinker-plugin-lto", "-Ccodegen-units=1", "-Clink-arg=-Olto"]
 
 [unstable]
 build-std = ["core", "alloc"]

analysis/performance/exponential/kernel/Cargo.toml

@@ -5,12 +5,15 @@ authors = ["Momo Langenstein <ml5717@ic.ac.uk>"]
 license = "MIT OR Apache-2.0"
 edition = "2018"
 
+[lib]
+crate-type = ["cdylib", "rlib"]
+
 # See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
 [dependencies]
 necsim-core = { path = "../../../../necsim/core", features = ["cuda"] }
-necsim-core-bond = { path = "../../../../necsim/core/bond", features = ["cuda"] }
-rust-cuda = { path = "../../../../rust-cuda", features = [] }
+necsim-core-bond = { path = "../../../../necsim/core/bond" }
+rust-cuda = { git = "https://github.com/MomoLangenstein/rust-cuda", branch = "main", features = [] }
 necsim-impls-no-std = { path = "../../../../necsim/impls/no-std", features = ["cuda"] }
 necsim-impls-cuda = { path = "../../../../necsim/impls/cuda" }
-contracts = { path = "../../../../third-party/contracts" }
+contracts = "0.6.3"

analysis/performance/exponential/kernel/src/benchmark.rs

@@ -0,0 +1,55 @@
+use core::{
+    num::NonZeroU32,
+    sync::atomic::{AtomicU64, Ordering},
+};
+
+use necsim_core::{
+    cogs::SeedableRng,
+    landscape::{IndexedLocation, Location},
+};
+use necsim_core_bond::{OffByOneU32, PositiveF64};
+
+use necsim_impls_cuda::cogs::maths::NvptxMathsCore;
+use necsim_impls_no_std::cogs::{
+    active_lineage_sampler::independent::event_time_sampler::EventTimeSampler,
+    habitat::non_spatial::NonSpatialHabitat, rng::wyhash::WyHash,
+};
+
+use crate::{sample, UniformTurnoverRate};
+
+#[inline]
+#[allow(dead_code)]
+pub fn inter_event_times<
+    E: EventTimeSampler<
+        NvptxMathsCore,
+        NonSpatialHabitat<NvptxMathsCore>,
+        WyHash<NvptxMathsCore>,
+        UniformTurnoverRate,
+    >,
+>(
+    event_time_sampler: E,
+    seed: u64,
+    lambda: PositiveF64,
+    limit: u128,
+    total_cycles_sum: &AtomicU64,
+    total_time_sum: &AtomicU64,
+) {
+    let habitat = NonSpatialHabitat::new((OffByOneU32::one(), OffByOneU32::one()), unsafe {
+        NonZeroU32::new_unchecked(1)
+    });
+    let rng = WyHash::seed_from_u64(seed + (rust_cuda::device::utils::index() as u64));
+    let turnover_rate = UniformTurnoverRate::new(lambda);
+    let indexed_location = IndexedLocation::new(Location::new(0, 0), 0);
+
+    let (cycles, time) = sample::exponential_inter_event_times(
+        habitat,
+        rng,
+        turnover_rate,
+        event_time_sampler,
+        indexed_location,
+        limit,
+    );
+
+    total_cycles_sum.fetch_add(cycles, Ordering::Relaxed);
+    total_time_sum.fetch_add(time, Ordering::Relaxed);
+}

analysis/performance/exponential/kernel/src/clock.rs

@@ -0,0 +1,17 @@
+/// A predefined, read-only 64-bit unsigned cycle counter.
+#[inline]
+#[must_use]
+pub fn counter() -> u64 {
+    let counter: u64;
+    unsafe { core::arch::asm!("mov.u64  {}, %clock64;", out(reg64) counter, options(nostack)) };
+    counter
+}
+
+/// A predefined, 64-bit global nanosecond timer.
+#[inline]
+#[must_use]
+pub fn timer_ns() -> u64 {
+    let timer: u64;
+    unsafe { core::arch::asm!("mov.u64  {}, %globaltimer;", out(reg64) timer, options(nostack)) };
+    timer
+}