There's a tool in the perl src repository which uses cachegrind behind the scenes to accurately measure how many CPU instructions, data reads etc a small snippet of code uses. With the following initial setup (so the hash already exists and has some keys):
my %h = qw(a 1 b 2 c 3 d 4); my $key = "foo";
Running the following benchmark (using a non-constant key so the key's hash gets recalculated each time):
$h{$key} = 1; delete $h{$key}
Shows the following results on various perls:
Key:
Ir Instruction read
Dr Data read
Dw Data write
COND conditional branches
IND indirect branches
_m branch predict miss
_m1 level 1 cache miss
_mm last cache (e.g. L3) miss
- indeterminate percentage (e.g. 1/0)
The numbers represent raw counts per loop iteration.
perl589o perl5101o perl5125o perl5144o perl5163o perl5184o perl
+5203o perl5222o perl5240o perl5258o
-------- --------- --------- --------- --------- --------- ----
+----- --------- --------- ---------
Ir 1348.0 1340.4 1378.0 1383.0 1423.0 1453.0 1
+466.0 1368.0 1356.0 1300.0
Dr 414.0 403.0 411.0 404.0 408.0 403.0
+411.0 379.0 373.0 362.0
Dw 226.0 214.0 222.0 227.0 228.0 231.0
+231.0 208.0 206.0 196.0
COND 202.0 210.1 210.0 204.0 213.0 204.0
+210.0 199.0 197.0 188.0
IND 16.0 16.0 17.0 18.0 18.0 18.0
+ 17.0 14.0 12.0 14.0
COND_m 2.0 1.0 4.0 2.0 3.0 3.0
+ 1.0 2.0 2.0 3.0
IND_m 9.0 9.0 11.0 9.0 9.0 11.0
+ 9.0 5.0 5.0 5.0
Ir_m1 0.0 0.0 0.0 0.0 0.0 0.0
+ 0.0 -0.1 0.0 0.0
Dr_m1 0.0 0.0 0.0 0.0 0.0 0.0
+ 0.0 0.0 0.0 0.0
Dw_m1 0.0 0.0 0.0 0.0 0.0 0.0
+ 0.0 0.0 0.0 0.0
Ir_mm 0.0 0.0 0.0 0.0 0.0 0.0
+ 0.0 0.0 0.0 0.0
Dr_mm 0.0 0.0 0.0 0.0 0.0 0.0
+ 0.0 0.0 0.0 0.0
Dw_mm 0.0 0.0 0.0 0.0 0.0 0.0
+ 0.0 0.0 0.0 0.0
Which shows everything being much the same before 5.22 (and in particular no significant slowdown in 5.16), and things getting better since.
Dave.
