June 2012 - squirrelfish-dev - lists.macosforge.org

Re: [squirrelfish] Porting DFG JIT
by Zoltan Herczeg 30 Jun '12

30 Jun '12

Hi, thanks for the reply. > label1: > ... some code > branch label3 > label2: > <watchpoint> > ... some short op (smaller than maxJumpReplacementSize) > label3: Is there anything which ensures that a watchpoint is always immediately followed by a label, which inserts the nops? So the code layout look always like this: watchpoint - multiple times allowed label - terminate the watchpoints, inserts the nops (btw on ARM we insert one nop like arithmetic instruction and one constant pool entry) wouldn't it be better, if the watchpoint would insert the nops, detects if called again (and no more nops are inserted) and the label would work as it worked before (we could get rid of one label type in this case). The operations performed by inlined labels seems too much at the moment. Regards, Zoltan

2 2

Porting DFG JIT
by Zoltan Herczeg 29 Jun '12

29 Jun '12

Hi, I think this is the best mailing list to discuss the technical questions about the porting of DFG JIT since everybody is here, still we do not pollute webkit-dev. My questions: - What is the difference between labelIgnoringWatchpoints, labelForWatchpoint, and label. - What is the difference between *_FUNCTION_WRAPPER_WITH_RETURN_ADDRESS_* macros? Are they similar to the DEFINE_STUB_FUNCTION macros? Thanks, Zoltan

2 2

while (regex.exec(string)) is surprisingly slow
by Zoltan Herczeg 15 Jun '12

15 Jun '12

Hi, I compared the following two in JSC and in v8: A: result = str.match(regex); B: while (regex.exec(str)) ++count; str was about 20M input, and regex was /./gm JSC: A: 618 ms B: 3469 ms V8: A: 2640 ms B: 1833 ms I was quite surprised that although the result is a really huge array, str.match was much faster in JSC. Is there a known reason to that? Can B be improved in JSC? Regards, Zoltan

2 1

sin is slow on 64 bit
by Zoltan Herczeg 08 Jun '12

08 Jun '12

Hi guys, I played with result differences on 32/64 bit systems today, and here is something interesting revealed by one sputnik test: sputnik/Conformance/15_Native_Objects/15.8_Math/15.8.2/15.8.2.16_sin/S15.8.2.16_A7.html Try this snippet on a 32 and a 64 bit system: #include <math.h> #include <stdio.h> int main() { int i; double d = 6.283185307179586, k; for (i = 0; i < 10000000; ++i) k = sin(d); return 0; } The 32 bit is 15x faster. Why? Because it simply calls fsin, while the 64 bit has an own SSE2 based implementation. I know x87 is obsolote and everything, but it is 15x faster... Shall we optimize this? Regards, Zoltan

3 6