From cda4fcb0fd3e9766a161cf3e5aa7c3283e7f7c9e Mon Sep 17 00:00:00 2001 From: Tim Hall Date: Thu, 19 May 2022 12:36:58 +0100 Subject: MLBEDSW-6563: networks failing with memory area exceeded in vela - For allocations that have a hard memory limit the Hill Climb allocator should be given more attempts to find a solution that would fit - The fix is to use a memory limit when there is a hard constraint, and a minimum iteration count, reset on every improvement, when there is a soft constraint - Added maximum number iterations CLI option Signed-off-by: Tim Hall Change-Id: I19ff53a0b68412de280263626778a3102cbe52fa --- OPTIONS.md | 14 ++++++++++++++ 1 file changed, 14 insertions(+) (limited to 'OPTIONS.md') diff --git a/OPTIONS.md b/OPTIONS.md index b4132d3a..ddda6971 100644 --- a/OPTIONS.md +++ b/OPTIONS.md @@ -255,6 +255,20 @@ systems, as there is a hard limit on thread stack size. vela network.tflite --recursion-limit 2000 ``` +### HillClimb Max Iterations + +Sets the maximum number of iterations the Hill Climb tensor allocator will run. +This is a hard limit on the total number of iterations of the algorithm. +Reducing this value is unlikely to reduce the compilation time of a working +solution, and it may cause the algorithm to terminate before finding a workable +solution. +**Type: Integer** +**Default: 99999** + +```bash +vela network.tflite --hillclimb-max-iterations 1000 +``` + ## Verbose Print Options All of the options below are disabled by default and enabling them will add -- cgit v1.2.1