Definition
-Control the precision of the Sugama cmat constants.
+Control the precision of the cmat constants.
Choices
COLLISION_PRECISION_MODE = 0: Full 64-bit precision
COLLISION_PRECISION_MODE = 1: Use lower-precision 32-bit precision for off-diagonal element.
COLLISION_PRECISION_MODE = 0: Full 64-bit precision.
COLLISION_PRECISION_MODE = 1: Heuristics-based lower-precision setup, slightly better than 32-bit precision.
COLLISION_PRECISION_MODE = 32: Reduced 32-bit precision.
COLLISION_PRECISION_MODE = 64: Full 64-bit precision. (Same as 0)
Comments
DEFAULT = 0
Only active for COLLISION_MODEL =4
Selecting COLLISION_PRECISION_MODE = 1 results in significant memory saving and negligible precision loss in most use-cases.
Definition
-Number of full 64-bit stripes to use close to the diagonal in low-precision mode.
-Comments
-DEFAULT = 8
Only active for COLLISION_MODEL =4 and COLLISION_PRECISION_MODE =1
Selecting COLLISION_PRECISION_MODE 1 and 32 results in significant memory saving and negligible precision loss in most use-cases.
Together with KY, this controls the toroidal resolution.
Together with TOROIDALS_PER_PROC, this controls the number of MPI processes needed.
Definition
+Number of toroidal harmonics to process in a single MPI process.
+Comments
+DEFAULT: 1
Must be a divisor of N_TOROIDAL.
A higher number typically results in faster simulation on low node count.
Definition
diff --git a/docs/cgyro/cgyro_table.html b/docs/cgyro/cgyro_table.html index 96d0694..774f079 100644 --- a/docs/cgyro/cgyro_table.html +++ b/docs/cgyro/cgyro_table.html @@ -110,6 +110,7 @@| - | What internal velocity order to use |
-1 |
-
-
| + | ||
Use reduced precision communication |
0 |
- | Enable GPU offload |
-1 |
-
input.cgyro parameter |
+Short description |
+Default |
+
|---|---|---|
| + | How many toroidal harmonics per MPI process |
+1 |
+
| + | Relative ordering of MPI ranks |
+2 |
+
| + | What internal velocity order to use |
+1 |
+
| + | Enable GPU offload when possible |
+1 |
+