1130089 - Clean-up: factor definitions of '' JitStackAlignment / sizeof(Value); ''

Status: RESOLVED FIXED

(Core :: JavaScript Engine: JIT, defect)

Description

[Nicolas B. Pierron [:nbp]]

The alignment patches are adding the same computation as a neat trick to remove paths at compile time.  We should move this definition to a header, and use MOZ_CONSTEXPR_VAR to define it.

Comment 1

[Nicolas B. Pierron [:nbp]]

Attachment: Use constexpr for JitStackValueAlignment.

Comment 2

[Benjamin Bouvier [:bbouvier] (inactive)]

Comment on attachment 8580906 [details] [diff] [review]
Use constexpr for JitStackValueAlignment.

Review of attachment 8580906 [details] [diff] [review]:
-----------------------------------------------------------------

Nice

::: js/src/jit/LIR.h
@@ +1731,5 @@
>      // the number of slots per Value.
>      uint32_t paddedLocalSlotCount() const {
>          // Round to ABIStackAlignment, but also round to at least sizeof(Value)
>          // in case that's greater, because StackOffsetOfPassedArg rounds
>          // argument slots to 8-byte boundaries.

comment is now stall, please update to say that JitStackAlignment is a multiple of sizeof(Value)

::: js/src/jit/Lowering.cpp
@@ +393,5 @@
>      // Align the arguments of a call such that the callee would keep the same
>      // alignment as the caller.
>      uint32_t baseSlot = 0;
> +    if (JitStackValueAlignment > 1)
> +        baseSlot = AlignBytes(argc, JitStackValueAlignment);

feel free to make this a ternary, if it fits in 100 chars

::: js/src/jit/arm/Assembler-arm.h
@@ +151,5 @@
> +static MOZ_CONSTEXPR_VAR uint32_t JitStackAlignment = 8;
> +
> +static MOZ_CONSTEXPR_VAR uint32_t JitStackValueAlignment = JitStackAlignment / sizeof(Value);
> +static_assert(JitStackAlignment % sizeof(Value) == 0 && JitStackValueAlignment >= 1,
> +  "Stack alignment should be larger than a Value");

same as in assembler x64

::: js/src/jit/mips/Assembler-mips.h
@@ +166,5 @@
> +static MOZ_CONSTEXPR_VAR uint32_t JitStackAlignment = 8;
> +
> +static MOZ_CONSTEXPR_VAR uint32_t JitStackValueAlignment = JitStackAlignment / sizeof(Value);
> +static_assert(JitStackAlignment % sizeof(Value) == 0 && JitStackValueAlignment >= 1,
> +  "Stack alignment should be larger than a Value");

same as in assembler x64

::: js/src/jit/x64/Assembler-x64.h
@@ +183,5 @@
> +static MOZ_CONSTEXPR_VAR uint32_t JitStackAlignment = 16;
> +
> +static MOZ_CONSTEXPR_VAR uint32_t JitStackValueAlignment = JitStackAlignment / sizeof(Value);
> +static_assert(JitStackAlignment % sizeof(Value) == 0 && JitStackValueAlignment >= 1,
> +  "Stack alignment should be larger than a Value");

not only larger, but also a multiple

::: js/src/jit/x64/Trampoline-x64.cpp
@@ +404,5 @@
>        "No need to consider the JitFrameLayout for aligning the stack");
>      static_assert(JitStackAlignment % sizeof(Value) == 0,
>        "Ensure that we can pad the stack by pushing extra UndefinedValue");
>  
> +    MOZ_ASSERT(IsPowerOfTwo(JitStackValueAlignment));

Can it be made an static assert, ooc? I suspect no, because IsPowerOfTwo might not be applicable to known constants...

::: js/src/jit/x86/Assembler-x86.h
@@ +113,5 @@
> +static MOZ_CONSTEXPR_VAR uint32_t JitStackAlignment = 16;
> +
> +static MOZ_CONSTEXPR_VAR uint32_t JitStackValueAlignment = JitStackAlignment / sizeof(Value);
> +static_assert(JitStackAlignment % sizeof(Value) == 0 && JitStackValueAlignment >= 1,
> +  "Stack alignment should be larger than a Value");

same as in assembler x64

Comment 3

[Nicolas B. Pierron [:nbp]]

(In reply to Benjamin Bouvier [:bbouvier] from comment #2)
> ::: js/src/jit/x64/Trampoline-x64.cpp
> @@ +404,5 @@
> >        "No need to consider the JitFrameLayout for aligning the stack");
> >      static_assert(JitStackAlignment % sizeof(Value) == 0,
> >        "Ensure that we can pad the stack by pushing extra UndefinedValue");
> >  
> > +    MOZ_ASSERT(IsPowerOfTwo(JitStackValueAlignment));
> 
> Can it be made an static assert, ooc? I suspect no, because IsPowerOfTwo
> might not be applicable to known constants...

No because IsPowerOfTwo is not a constexpr, and recently I noticed that gcc 4.7 has issues with constexpr templates.  So, not yet :(

Comment 4

[Nicolas B. Pierron [:nbp]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/c2e1dd959091

https://treeherder.mozilla.org/#/jobs?repo=try&revision=054b26094100

Comment 5

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/c2e1dd959091