In short, the reason is that eliminating bounds checks in JavaScript is harder than it seems. It also doesn't make a huge performance difference on most real-world code (i.e. aside from micro-benchmarks), so supporting it doesn't have very high priority. That said, we are hoping to implement a system one day that would at least handle the relatively straightforward case in your example.
The "most correct flags" are none at all. Everything that makes sense to enable is enabled by default anyway. On Sun, Apr 3, 2016 at 10:40 PM, Gem Dot Artigas <[email protected]> wrote: > Thanks a lot for your answer. > > But, There is any reason for this? In other words, why they have been > disabed it by defect? Usually, What do the current browsers do? I would > like to know which are the the most correct flags for achademic research. > > thanks in advance > > > El domingo, 3 de abril de 2016, 18:17:15 (UTC+2), Yang escribió: >> >> For Crankshaft, bounds check hoisting optimization has been disabled. You >> can try to turn the flag on and try your test case. >> Not sure if TurboFan already has similar optimization implemented or >> enabled. >> >> On Sunday, April 3, 2016 at 1:38:57 AM UTC-7, Gem Dot Artigas wrote: >>> >>> Hi, >>> >>> >>> I have an important question about check bounds. Why, in some very clear >>> cases, V8 put the instruction to obtain the length of the array out of the >>> loop, but the check bounds is still inside? (I refer the scenarios when the >>> loop control variable is clear to be incremented by 1 at each loop >>> iteration and it is not manipulated by no function) >>> >>> >>> >>> I put an example to clarify it: >>> >>> Given this JS function from Kraken suite: >>> >>> 1 findGraphNode = function(node) { >>> 2 for(var i=0;i<this.length;i++) { >>> 3 if(this[i].position == node.position) { >>> 4 return true; >>> 5 } >>> 6 } >>> 7 return false; >>> 8 } >>> >>> >>> The optimized generated x86 code is the following: >>> >>> I1 movq rax, this // load this >>> I2 test rax, 1 // check non-smi >>> I3 jz code_deoptimization // check non-smi >>> I4 movq r10, nodeList // check maps >>> I5 cmpq (rax-1), r10 // check maps >>> I6 jnz code_deoptimization // check maps >>> I7 movl rbx, (rax+27) // load elements length >>> I8 movq rdx,(rax+15) // load elements array >>> I9 movq rcx, (rbp+16) // load node >>> I10 testb rcx, 1 // check non-smi >>> I11 jz code_deoptimization // check non-smi >>> I12 movq r10, GraphNode // check maps >>> I13 cmpq (rcx-1), r10 // check maps >>> I14 jnz code_deoptimization // check maps >>> I15 movq rsi,(rcx+47) // load node.position >>> loop: >>> I16 cmpl rdi, rbx // compare i to this.length >>> I17 jge return_false // node not found: return false >>> I18 cmpq rsp, (stack_limit) // check stack >>> I19 jc external_exception // check stack >>> I20 cmpl rbx,rdi // check bounds >>> I21 jna code_deoptimization // check bounds >>> I22 movq r8, (rdx+rdi*8) // load this[i] >>> I23 testb r8, 1 // check non-smi >>> I24 jz code_deoptimization // check non-smi >>> I25 movq r10, GraphNode // check maps >>> I26 cmpq (r8-1), r10 // check maps >>> I27 jnz code_deoptimization // check maps >>> I28 movq r9,(r8+47) // load this[i].position >>> I29 cmpq r9,rsi // compare this[i].position to >>> node.position >>> I30 jz return_true // node found: return true >>> I31 addl rdi,0x1 // i++ >>> I32 jmp loop // loop back edge >>> >>> >>> So, Why check bounds can not be put out of the loop? You can put a check >>> bounds out of the loop and compare it with this.length variable directly... >>> >>> Thanks in Avdance >>> >> -- > -- > v8-dev mailing list > [email protected] > http://groups.google.com/group/v8-dev > --- > You received this message because you are subscribed to the Google Groups > "v8-dev" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/d/optout. > -- -- v8-dev mailing list [email protected] http://groups.google.com/group/v8-dev --- You received this message because you are subscribed to the Google Groups "v8-dev" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
