From f3ebc8c7ebad7b649fef16ae5737b105ca06010e Mon Sep 17 00:00:00 2001 From: Patrick Pelissier Date: Sat, 1 Jun 2024 10:16:47 +0200 Subject: [PATCH] Update comments --- m-dict.h | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/m-dict.h b/m-dict.h index 428ca9c6..7389abbd 100644 --- a/m-dict.h +++ b/m-dict.h @@ -189,7 +189,8 @@ ARRAY_DEF(m_array_index, m_indexhash_t, M_POD_OPLIST) /* Upper Bound of the hash table The container is designed for an upper bound in the range [0.5-0.75] Using higher upper bound (even by a slow amount) will deteriorate a lot the performance: - an upper-bound of 0.875 is ~3 times slower than an upper bound of 0.7 for lockup of key not present. + an upper-bound of 0.875 is ~3 times slower than an upper bound of 0.7 for lockup of key not present + (without considering prefetch of cache-lines which improve performance). This is due to the esperance of the number of reads before reading an empty value which is: (1-q)*sum(q^k*(k+1),k=0 .. infinity) = 1/(1-q) which is the estimated number of iteration before stopping the search loop.