Hello! I have a piece of auto-generated code and I noticed that if a function has more than certain number of expressions it starts to run much slower. I'm guessing this has something to do with what JIT considers to be still worth of compiling but I just would like to confirm this: is there any limit on how big (in term of operations) a function needs to be in order to get compiled by JIT? Is it possible to increase the limit from user code? Or is there any other reason for the slowdown? Please find below an arficial example. Here it turn in ~0.13 seconds but when I uncomment the first line it drops to 5.3 seconds. Thanks, Tuom Larsen def f(x): a0 = x*0.1; a1 = x*0.1; a2 = x*0.2; a3 = x*0.3; a4 = x*0.4 a5 = x*0.5; a6 = x*0.6; a7 = x*0.7; a8 = x*0.8; a9 = x*0.9 b0 = x*0.1; b1 = x*0.1; b2 = x*0.2; b3 = x*0.3; b4 = x*0.4 b5 = x*0.5; b6 = x*0.6; b7 = x*0.7; b8 = x*0.8; b9 = x*0.9 return sum([ # (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9)]) for i in range(100000): f(0.1)
Hi Tuom. It's the limit of number of operations while compiling. It can be (globally) edited using: pypy --jit trace_limit=some-number default is 6000. --jit help will show you some information. We should improve the heuristic to always trace a function that's the only one maybe though On Tue, Oct 6, 2015 at 7:32 PM, Tuom Larsen <tuom.larsen@gmail.com> wrote:
Hello!
I have a piece of auto-generated code and I noticed that if a function has more than certain number of expressions it starts to run much slower.
I'm guessing this has something to do with what JIT considers to be still worth of compiling but I just would like to confirm this: is there any limit on how big (in term of operations) a function needs to be in order to get compiled by JIT? Is it possible to increase the limit from user code? Or is there any other reason for the slowdown?
Please find below an arficial example. Here it turn in ~0.13 seconds but when I uncomment the first line it drops to 5.3 seconds.
Thanks,
Tuom Larsen
def f(x): a0 = x*0.1; a1 = x*0.1; a2 = x*0.2; a3 = x*0.3; a4 = x*0.4 a5 = x*0.5; a6 = x*0.6; a7 = x*0.7; a8 = x*0.8; a9 = x*0.9 b0 = x*0.1; b1 = x*0.1; b2 = x*0.2; b3 = x*0.3; b4 = x*0.4 b5 = x*0.5; b6 = x*0.6; b7 = x*0.7; b8 = x*0.8; b9 = x*0.9 return sum([ # (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9)])
for i in range(100000): f(0.1) _______________________________________________ pypy-dev mailing list pypy-dev@python.org https://mail.python.org/mailman/listinfo/pypy-dev
Thanks, even trace_limit=9000 is sufficient in my case! That heuristic would be most awesome so that other people don't have to manually override the limit. I'm guessing it is more complicated than that but simplify lifting the limit to, say, 12000 wouldn't suffice? On Wed, Oct 7, 2015 at 11:26 AM, Maciej Fijalkowski <fijall@gmail.com> wrote:
Hi Tuom. It's the limit of number of operations while compiling. It can be (globally) edited using:
pypy --jit trace_limit=some-number
default is 6000. --jit help will show you some information. We should improve the heuristic to always trace a function that's the only one maybe though
On Tue, Oct 6, 2015 at 7:32 PM, Tuom Larsen <tuom.larsen@gmail.com> wrote:
Hello!
I have a piece of auto-generated code and I noticed that if a function has more than certain number of expressions it starts to run much slower.
I'm guessing this has something to do with what JIT considers to be still worth of compiling but I just would like to confirm this: is there any limit on how big (in term of operations) a function needs to be in order to get compiled by JIT? Is it possible to increase the limit from user code? Or is there any other reason for the slowdown?
Please find below an arficial example. Here it turn in ~0.13 seconds but when I uncomment the first line it drops to 5.3 seconds.
Thanks,
Tuom Larsen
def f(x): a0 = x*0.1; a1 = x*0.1; a2 = x*0.2; a3 = x*0.3; a4 = x*0.4 a5 = x*0.5; a6 = x*0.6; a7 = x*0.7; a8 = x*0.8; a9 = x*0.9 b0 = x*0.1; b1 = x*0.1; b2 = x*0.2; b3 = x*0.3; b4 = x*0.4 b5 = x*0.5; b6 = x*0.6; b7 = x*0.7; b8 = x*0.8; b9 = x*0.9 return sum([ # (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9)])
for i in range(100000): f(0.1) _______________________________________________ pypy-dev mailing list pypy-dev@python.org https://mail.python.org/mailman/listinfo/pypy-dev
we found quite a few problems with the current heuristic so we'll put some effort into it some time soon. On Wed, Oct 7, 2015 at 12:17 PM, Tuom Larsen <tuom.larsen@gmail.com> wrote:
Thanks, even trace_limit=9000 is sufficient in my case!
That heuristic would be most awesome so that other people don't have to manually override the limit. I'm guessing it is more complicated than that but simplify lifting the limit to, say, 12000 wouldn't suffice?
On Wed, Oct 7, 2015 at 11:26 AM, Maciej Fijalkowski <fijall@gmail.com> wrote:
Hi Tuom. It's the limit of number of operations while compiling. It can be (globally) edited using:
pypy --jit trace_limit=some-number
default is 6000. --jit help will show you some information. We should improve the heuristic to always trace a function that's the only one maybe though
On Tue, Oct 6, 2015 at 7:32 PM, Tuom Larsen <tuom.larsen@gmail.com> wrote:
Hello!
I have a piece of auto-generated code and I noticed that if a function has more than certain number of expressions it starts to run much slower.
I'm guessing this has something to do with what JIT considers to be still worth of compiling but I just would like to confirm this: is there any limit on how big (in term of operations) a function needs to be in order to get compiled by JIT? Is it possible to increase the limit from user code? Or is there any other reason for the slowdown?
Please find below an arficial example. Here it turn in ~0.13 seconds but when I uncomment the first line it drops to 5.3 seconds.
Thanks,
Tuom Larsen
def f(x): a0 = x*0.1; a1 = x*0.1; a2 = x*0.2; a3 = x*0.3; a4 = x*0.4 a5 = x*0.5; a6 = x*0.6; a7 = x*0.7; a8 = x*0.8; a9 = x*0.9 b0 = x*0.1; b1 = x*0.1; b2 = x*0.2; b3 = x*0.3; b4 = x*0.4 b5 = x*0.5; b6 = x*0.6; b7 = x*0.7; b8 = x*0.8; b9 = x*0.9 return sum([ # (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9), (a0+a1+a2+a3+a4+a5+a6+a7+a8+a9)*(b0+b1+b2+b3+b4+b5+b6+b7+b8+b9)])
for i in range(100000): f(0.1) _______________________________________________ pypy-dev mailing list pypy-dev@python.org https://mail.python.org/mailman/listinfo/pypy-dev
participants (2)
-
Maciej Fijalkowski -
Tuom Larsen