To me it just looks like it flutters a bit - the tests ought to be a little longer to get more significant digits. Note that the "total" column includes the load time in every case, so the "user" column is more interesting except for the first test.
/ Martin Stjernholm, Roxen IS
Previous text:
2003-01-16 04:16: Subject: Intel C++ / gcc 2.95.4 comparisions
Hmm. Actually, the test was between installed + dumped Pike 7.5.1 and non-installed, non-dumped with icc. Here's the results from icc when dumped:
test total user mem (runs) Pike start overhead........ 0.097s 0.001s 3560kb (25) Ackermann.................. 0.965s 0.863s 3788kb (6) Array & String Juggling.... 0.937s 0.828s 3956kb (6) Clone null-object.......... 0.347s 0.230s 3500kb (15) (19565218/s) Clone object............... 0.722s 0.624s 3480kb (7) (3363844/s) Compile.................... 1.508s 1.395s 5280kb (4) (69219 lines/s) Compile & Exec............. 1.536s 1.433s 3832kb (4) (1679302 lines/s) GC......................... 1.014s 0.898s 3688kb (5) Matrix multiplication...... 0.657s 0.560s 5568kb (8) Loops Nested (local)....... 0.647s 0.543s 3572kb (8) (247405936 iters/s) Loops Nested (global)...... 1.100s 0.902s 3560kb (5) (93000080 iters/s) Loops Recursed............. 0.695s 0.596s 3560kb (8) (14068944 iters/s)
I especially like the "pike start overhead" difference. HUGE gains over Pike compiled with gcc 2.95.4. Interestingly enough it's significantly in general over a non-dumped version, even when you think it wouldn't be.
/ David Hedbor