On Tue, Feb 16, 2016 at 10:38 AM, <[email protected]> wrote: > 1) 0x36c28f53eecf 47 e8cc92fdff call 0x36c28f5181a0 ;; code: > contextual, LOAD_IC, GENERIC > Who does generate this stub? Full-codegen? >
No, the IC system. You can start reading in LoadIC::Load in src/ic/ic.cc and trace your way to the various HandlerCompilers, but be warned: there's a lot of complexity there. > 2) As I undertood the effictive code is placed in IC stub? > Depends on what you mean by "effective code". Originally you said you wanted to understand full codegen. Stubs are not part of full codegen. > If this is true then where can I see the stub's code? > --print-code-stubs. However, for this specific example, you won't see a stub there, because loading properties from numbers is handled specially. Of course you can also use a debugger of your choice, interrupt the program when it's in the state you want to inspect, and start disassembling whatever you're interested in. > On Tuesday, February 16, 2016 at 3:18:07 PM UTC+6, Jakob Kummerow wrote: > >> On Tue, Feb 16, 2016 at 10:08 AM, <[email protected]> wrote: >> >>> Thank Jakob. >>> >>> Consider much more easier function: >>> >>> >>> function foo(o) { >>> return o.x; >>> } >>> >>> >>> print(foo(37)); >>> >>> >>> listing <http://pastebin.com/u0ZM8C3w> >>> >>> I've got two question: >>> 1) 0x36c28f53eecf 47 e8cc92fdff call 0x36c28f5181a0 ;; code: >>> contextual, LOAD_IC, GENERIC >>> Does that code create new IC stub? Where is the check for IC placed? >>> >> >> That's the call to the IC stub. The stub is created elsewhere >> (dynamically, as needed). All the type checks happen inside the stub, not >> in the enclosing function. >> >> >>> >>> I believe that check should looks like: >>> if (actual_type1 == int) -> generate IC stub for int, if it exists then >>> we call it >>> if (actual_type1 == double) ... >>> >> >> More like: >> if (hidden class of o == previously seen hidden class) { call previously >> generated handler } >> else miss /* call into C++ to create a new handler for the new hidden >> class */ >> >> At least that's the standard "monomorphic" case. The IC system is quite >> flexible, each IC can be in one of many different states. >> >> >>> 2) How to remove lines that looks like debug *. For example : debug: >>> statement 19 >>> >> >> They're just comments, part of the --code-comments output. Their purpose >> is to make it easier for you to trace machine instructions back to the >> JavaScript source. >> >> >>> On Tuesday, February 16, 2016 at 2:31:58 PM UTC+6, Jakob Kummerow wrote: >>>> >>>> On Tue, Feb 16, 2016 at 8:10 AM, <[email protected]> wrote: >>>> >>>>> Thank Ben, but I need to some example: >>>>> >>>>> function add(a, b) { >>>>> return a + b; >>>>> } >>>>> >>>>> >>>>> print(add(37, 73)); >>>>> I execute this code via d8. >>>>> >>>>> What is context in that case? >>>>> >>>> >>>> Your function "add" is so simple that it doesn't need a context, and >>>> indeed no context is allocated. >>>> >>>> >>>>> I've got the asm listing via --print-code: listing >>>>> <http://pastebin.com/NhWyhRDK> >>>>> Actually I can't understand why the listing consists so many codes? >>>>> The adder function is very simple. >>>>> >>>>> 0xda955b3eff6 150 49ba0000000025000000 REX.W movq r10,0x2500000000 >>>>> 0xda955b3f000 160 4152 push r10 >>>>> 0xda955b3f002 162 49ba0000000049000000 REX.W movq r10,0x4900000000 >>>>> 0xda955b3f00c 172 4152 push r10 >>>>> 0xda955b3f00e 174 48ba0000000007000000 REX.W movq rdx,0x700000000 >>>>> 0xda955b3f018 184 488b7c2418 REX.W movq rdi,[rsp+0x18] >>>>> 0xda955b3f01d 189 e85efcffff call 0xda955b3ec80 ;; code: >>>>> CALL_IC, GENERIC >>>>> 0xda955b3f022 194 488b75f8 REX.W movq rsi,[rbp-0x8] >>>>> 0xda955b3f026 198 48890424 REX.W movq [rsp],rax >>>>> >>>>> >>>> ^ That's the compiled top-level code. The compiled code for function >>>> "add" is in lines 109 - 170 in your pastebin. You can tell either by >>>> looking at the "--- Raw source ---" section, or at the "name = add" line. >>>> >>>> If you run with --code-comments, you'll get comments embedded into the >>>> code that'll explain what each section was generated for. >>>> >>>> >>>>> On Monday, February 15, 2016 at 11:14:58 PM UTC+6, Ben Noordhuis wrote: >>>>> >>>>>> On Mon, Feb 15, 2016 at 1:57 PM, <[email protected]> wrote: >>>>>> > Hi all! >>>>>> > >>>>>> > I'm trying to understand the full-codegen compiler in v8. >>>>>> > >>>>>> > Ok, I've taken the full-codegen code for x64 architecture. >>>>>> > >>>>>> > As I can see in code, it does following things: >>>>>> > Build x64 frame >>>>>> > Allocate locals >>>>>> > And then allocate context <- this is magic for me >>>>>> > >>>>>> > The code's: >>>>>> > >>>>>> > // Possibly allocate a local context. >>>>>> > >>>>>> > if (info->scope()->num_heap_slots() > 0) { >>>>>> > >>>>>> > Comment cmnt(masm_, "[ Allocate context"); >>>>>> > >>>>>> > bool need_write_barrier = true; >>>>>> > >>>>>> > int slots = info->scope()->num_heap_slots() - >>>>>> > Context::MIN_CONTEXT_SLOTS; >>>>>> > >>>>>> > // Argument to NewContext is the function, which is still in >>>>>> rdi. >>>>>> > >>>>>> > ... >>>>>> > >>>>>> > >>>>>> > What does the full-codegen does here? >>>>>> >>>>>> The context it allocates is the storage for a function's free >>>>>> variables. V8's garbage collector is generational; the write barrier >>>>>> is for tracking object references from the old space to the new >>>>>> space. >>>>>> >>>>> -- >>>>> >>>> -- > -- > 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.
