rtx-cfd/rtx-fsi PERF-2 P3-iii K-stream mode: each lane thread's device operator lives on its own CUDA stream (no rendezvous; set_plane_streams, RTX_FSI2O_PLANE_MODE=streams default | batch); stream-mode unit test digit-identical to the solo apply

Co-Authored-By: Claude Fable 5.1 <[email protected]>

crates/specialized/rtx-fsi/tests/fsi2_overset_plane.rs

@@ -10,6 +10,9 @@
 //!   (the `p7_points.txt` format; `#` comments and blank lines skipped).
 //!   Without it the test returns.
 //! - `RTX_FSI2O_PLANE_DIR`: each lane's CSV goes to `<dir>/<tag>.csv`.
+//! - `RTX_FSI2O_PLANE_MODE`: `streams` (default; one CUDA stream per lane
+//!   thread, no rendezvous — the K-stream design) or `batch` (the
+//!   rendezvous: every lane inside a CG gathered into one batched launch).
 //! - `RTX_FSI2O_PLANE_STACK_MB`: the lane threads' stack (default 64).
 //! Requires `RTX_FSI2O_MG_RB=1 RTX_FSI2O_MG_DEVICE=1` on a
 //! `--features rtx-cfd/cuda` build; the test fails loudly otherwise (a
@@ -21,7 +24,7 @@ mod fsi2_harness;
 
 use fsi2_harness::overset_march::{OversetMarchConfig, run_march_overset};
 use fsi2_harness::{BenchmarkCase, FSI2, FSI3};
-use rtx_cfd::solvers::incompressible::{plane_counters, set_plane_lane};
+use rtx_cfd::solvers::incompressible::{plane_counters, set_plane_lane, set_plane_streams};
 use std::time::Instant;
 
 struct Point {
@@ -80,6 +83,13 @@ fn fsi2_overset_plane() {
         device_on,
         "the plane needs RTX_FSI2O_MG_RB=1 RTX_FSI2O_MG_DEVICE=1 on a cuda build"
     );
+    let mode = std::env::var("RTX_FSI2O_PLANE_MODE").unwrap_or_else(|_| "streams".into());
+    let streams = match mode.as_str() {
+        "streams" => true,
+        "batch" => false,
+        other => panic!("RTX_FSI2O_PLANE_MODE must be streams or batch, got {other:?}"),
+    };
+    set_plane_streams(streams);
     let points = read_points(&points_path);
     let k = points.len();
     assert!(k >= 1, "no plane points in {points_path}");
@@ -113,7 +123,7 @@ fn fsi2_overset_plane() {
         FSI2
     };
     println!(
-        "  MARCH PLANE: {k} lanes in one process ({} base, ny {}, t_end {}, coupler {}, s = {}); points from {points_path}, CSVs under {dir}",
+        "  MARCH PLANE: {k} lanes in one process, mode {mode} ({} base, ny {}, t_end {}, coupler {}, s = {}); points from {points_path}, CSVs under {dir}",
         base.name, config.ny, config.t_end, config.coupler, config.subcycle
     );
     for (i, p) in points.iter().enumerate() {
@@ -176,7 +186,7 @@ fn fsi2_overset_plane() {
     let plane_wall = t_plane.elapsed().as_secs_f64();
     let (batches, served) = plane_counters();
     println!(
-        "  MARCH PLANE done: {k} lanes in {plane_wall:.0} s wall; {batches} batched V-cycle launches served {served} lane calls ({:.2} lanes per batch)",
+        "  MARCH PLANE done: {k} lanes in {plane_wall:.0} s wall, mode {mode}; {batches} batched V-cycle launches served {served} lane calls ({:.2} lanes per batch)",
         if batches > 0 {
             served as f64 / batches as f64
         } else {