1074509 - SIMD: Implement NaN-correct greaterThan and greaterThanOrEqual for x86

Status: RESOLVED FIXED

(Core :: JavaScript Engine: JIT, defect)

Description

[Dan Gohman [:sunfish]]

Attachment: simd-float32x4-gt-ge.patch

The greaterThan and greaterThanOrEqual functions should return false when either operand is a NaN. To implement this on x86 without using copies, we must reverse the operands to form lessThan or lessThanOrEqual before register allocation, since these instructions are two-address.

Comment 1

[Benjamin Bouvier [:bbouvier] (inactive)]

Comment on attachment 8497124 [details] [diff] [review]
simd-float32x4-gt-ge.patch

Review of attachment 8497124 [details] [diff] [review]:
-----------------------------------------------------------------

Thanks, looks good to me. I think this patch is based on the one in bug 1072368, so you might want to rebase it on trunk or to ask the other one for r?.

::: js/src/jit/MIR.h
@@ +1608,5 @@
> +        switch (operation()) {
> +          case greaterThan:        operation_ = lessThan; break;
> +          case greaterThanOrEqual: operation_ = lessThanOrEqual; break;
> +          case lessThan:           operation_ = greaterThan; break;
> +          case lessThanOrEqual:    operation_ = greaterThanOrEqual; break;  

nit: trailing whitespace

::: js/src/jit/arm/Lowering-arm.h
@@ +59,5 @@
>      bool lowerForFPU(LInstructionHelper<1, 2, 0> *ins, MDefinition *mir,
>                       MDefinition *lhs, MDefinition *rhs);
> +
> +    bool lowerForCompIx4(LSimdBinaryCompIx4 *ins, MSimdBinaryComp *mir,
> +                         MDefinition *lhs, MDefinition *rhs) {

nit: { goes to the next line, here and below.

::: js/src/jit/mips/Lowering-mips.h
@@ +59,5 @@
>      bool lowerForFPU(LInstructionHelper<1, 2, 0> *ins, MDefinition *mir,
>                       MDefinition *lhs, MDefinition *rhs);
> +
> +    bool lowerForCompIx4(LSimdBinaryCompIx4 *ins, MSimdBinaryComp *mir,
> +                         MDefinition *lhs, MDefinition *rhs) {

nit: { goes to the next line, here and below.

::: js/src/jit/shared/CodeGenerator-x86-shared.cpp
@@ +2396,5 @@
>  
>          // scr := scr > lhs (i.e. lhs < rhs)
>          // Improve by doing custom lowering (rhs is tied to the output register)
>          masm.packedGreaterThanInt32x4(ToOperand(ins->lhs()), ScratchSimdReg);
> +        masm.moveAlignedInt32x4(ScratchSimdReg, lhs);

This code seems to be based on bug 1072368's patch...

Comment 2

[Dan Gohman [:sunfish]]

Rebased, and nits addressed:
https://hg.mozilla.org/integration/mozilla-inbound/rev/696baf50aabd

Comment 3

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/696baf50aabd