embedded3 PERF-3 P2: the Poisson initial guess by projection onto the last K corrections of the same corrector index (GuessBasis, DeviceCg::project_guess: Gram matrix rebuilt per solve, Cholesky with pivot dropping; e3_cg_axpy) behind RTX_E3_POISSON_PROJECT=K (default 0, digit-identical); static DFG 2D ny 122: CG 9.5 -> 2.5 per step, loads identical to 4 digits; the moving flag: -4 % only (consecutive corrections nearly uncorrelated); RTX_E3_POISSON_PROJECT_DEBUG prints the residual before / after

Co-Authored-By: Claude Fable 5.1 <[email protected]>
2026-09-19 18:57:50 -05:00
parent 845e0ae01a
commit d4cd7d9545
5 changed files with 309 additions and 1 deletions
@@ -138,6 +138,17 @@ extern "C" __global__ void e3_cg_axpy2(
    r[g] -= alpha * q[g];
 }

+/* y += alpha x over the cells (PERF-3 P2: the projected initial guess). */
+extern "C" __global__ void e3_cg_axpy(
+    int n_cells, const unsigned int* __restrict__ cells, double alpha,
+    const double* __restrict__ x, double* __restrict__ y)
+{
+    int t = blockIdx.x * blockDim.x + threadIdx.x;
+    if (t >= n_cells) return;
+    int g = cells[t];
+    y[g] += alpha * x[g];
+}
+
 /* d = z + beta d. */
 extern "C" __global__ void e3_cg_xpay(
    int n_cells, const unsigned int* __restrict__ cells, double beta,