Changeset 525b5ef for benchmark/basic

Timestamp:

Jan 13, 2020, 12:49:14 PM (4 years ago)

Author:

Peter A. Buhr <pabuhr@…>

Branches:

ADT, arm-eh, ast-experimental, enum, forall-pointer-decay, jacob/cs343-translation, jenkins-sandbox, master, new-ast, new-ast-unique-expr, pthread-emulation, qualifiedEnum

Children:

768bd556

Parents:

a21dec4

Message:

add documentation

Location:

benchmark/basic

Files:

• fetch_add.c (modified) (1 diff)
• tls_fetch_add.c (modified) (1 diff)
• ttst_lock.c (modified) (1 diff)

benchmark/basic/fetch_add.c

@@ -2 +2 @@
 
 #include "bench.h"
+
+// Does a "lock add" on entry and a "lock sub" on exit => 2 atomic instructions making it the most expensive
+// atomic test.
 
 volatile int value;

benchmark/basic/tls_fetch_add.c

@@ -3 +3 @@
 
 #include "bench.h"
+
+// Does not do a fetch & add. It mimics the cfa protocol to disable interrupts locally, by writing true or false to a
+// thread_local Boolean. This means the entire protocol is just to "mov" instructions making it extremely cheap.
 
 #define thread_local _Thread_local

benchmark/basic/ttst_lock.c

@@ -3 +3 @@
 
 #include "bench.h"
+
+// Does a "lock xchg" on entry but a simple "mov" on exit => cheaper as 0 contention. While it has much more code, the
+// bulk is never run.
 
 #define CALIGN __attribute__(( aligned (CACHE_ALIGN) ))