• bearophile (5/5) Oct 31 2014 Do you want to make LDC2 recognize core.checkedint functions and
• David Nadlinger via digitalmars-d-ldc (5/8) Oct 31 2014 Sure, it maps very directly to the corresponding LLVM intrinsics
• Kai Nacke (5/14) Nov 01 2014 See https://github.com/ldc-developers/ldc/pull/772
• safety0ff (23/23) Nov 01 2014 I'm not terribly interested in preparing a PR for this, but I
• bearophile (5/9) Nov 01 2014 Good.
• safety0ff (3/12) Nov 01 2014 They're intrinsics:
• bearophile (4/6) Nov 01 2014 Sorry, I meant the functions like mulu, etc.

"bearophile" <bearophileHUGS lycos.com> writes:

Do you want to make LDC2 recognize core.checkedint functions and 
implement them efficiently reading the overflow/carry flags of 
the CPU?

Bye,
bearophile

David Nadlinger via digitalmars-d-ldc <digitalmars-d-ldc puremagic.com> writes:

On 1 Nov 2014, at 1:58, bearophile via digitalmars-d-ldc wrote:
 Do you want to make LDC2 recognize core.checkedint functions and 
 implement them efficiently reading the overflow/carry flags of the 
 CPU?

Sure, it maps very directly to the corresponding LLVM intrinsics 
(conditionally setting the by-ref parameter has to be done in a small 
wrapper, though). Pull request are very welcome. ;)

David

"Kai Nacke" <kai redstar.de> writes:

On Saturday, 1 November 2014 at 04:04:48 UTC, David Nadlinger via 
digitalmars-d-ldc wrote:
 On 1 Nov 2014, at 1:58, bearophile via digitalmars-d-ldc wrote:
 Do you want to make LDC2 recognize core.checkedint functions 
 and implement them efficiently reading the overflow/carry 
 flags of the CPU?

 Sure, it maps very directly to the corresponding LLVM 
 intrinsics (conditionally setting the by-ref parameter has to 
 be done in a small wrapper, though). Pull request are very 
 welcome. ;)

 David

See https://github.com/ldc-developers/ldc/pull/772

Regards,
Kai

"safety0ff" <safety0ff.dev gmail.com> writes:

I'm not terribly interested in preparing a PR for this, but I 
believe the code should be something like this (untested):

ldc/src/core/checkedint.d:

uint mulu(uint x, uint y, ref bool overflow)
{
+ version (LDC)
+ {
+     auto res = llvm_umul_with_overflow(x, y);
+     overflow |= res.overflow;
+     return res.result;
+ }
+ else
+ {
       ulong r = ulong(x) * ulong(y);
       if (r > uint.max)
           overflow = true;
       return cast(uint)r;
+ }
}

Rinse and repeat with:
llvm_uadd_with_overflow, llvm_sadd_with_overflow, 
llvm_usub_with_overflow, llvm_ssub_with_overflow, 
llvm_smul_with_overflow.

"bearophile" <bearophileHUGS lycos.com> writes:

safety0ff:

 Rinse and repeat with:
 llvm_uadd_with_overflow, llvm_sadd_with_overflow, 
 llvm_usub_with_overflow, llvm_ssub_with_overflow, 
 llvm_smul_with_overflow.

Good.
Is it 100% guaranteed ldc2 will inline this function?

Bye,
bearophile

"safety0ff" <safety0ff.dev gmail.com> writes:

On Saturday, 1 November 2014 at 18:06:41 UTC, bearophile wrote:
 safety0ff:

 Rinse and repeat with:
 llvm_uadd_with_overflow, llvm_sadd_with_overflow, 
 llvm_usub_with_overflow, llvm_ssub_with_overflow, 
 llvm_smul_with_overflow.

 Good.
 Is it 100% guaranteed ldc2 will inline this function?

 Bye,
 bearophile

They're intrinsics:
https://github.com/ldc-developers/druntime/blob/ldc/src/ldc/intrinsics.di#L462

"bearophile" <bearophileHUGS lycos.com> writes:

safety0ff:

 They're intrinsics:
 https://github.com/ldc-developers/druntime/blob/ldc/src/ldc/intrinsics.di#L462

Sorry, I meant the functions like mulu, etc.

Bye,
bearophile