[Libre-soc-bugs] [Bug 558] gcc SV intrinsics concept

bugzilla-daemon at libre-soc.org bugzilla-daemon at libre-soc.org
Mon Jan 11 22:49:32 GMT 2021


https://bugs.libre-soc.org/show_bug.cgi?id=558

--- Comment #42 from Luke Kenneth Casson Leighton <lkcl at lkcl.net> ---
(In reply to Jacob Lifshay from comment #40)
> I think that CRs should be handled entirely by the gcc backend, they will
> have mask values allocated to them automatically.

in the "final" version (the final target) i could not agree more. results (all
results) can have a CR associated with them: when Vectorised this becomes a
Vector of associated CRs and i would anticipate there not being much in the way
of changes to gcc-ppc64 scalar code to transform it to support "scalar is now
vector yes even CRs"

we are however at a weird state where for the minimum amount of work we need an
interim step that at least gives developers such as Lauri, plus the
cryptoprimitives Grant proposal, a level just above assembler that does not
require 2 years of gcc development before they can even begin work.

to expect gcc to automatically support vectors-of-CRs unfortunately does
exactly that: places significant gcc work smack on the critical path.  

> In the gcc frontend, masks
> are just bitvectors (basically a uint64_t with a different type).

ah that might work.  ahh i know.  each CR Vector is considered to be a Vector
of eq bits, a Vector of lt bits etc.

the uint64_t with different type you mention is:

   bit 0 is CR0.eq
   bit 1 is CR1.eq
   etc

then in the crweirder intrinsic it takes these parameters:

   * struct of 4 uint64_t named eqltgtso
   * same as dest   
   * mode
   * mask

(or, just 10 arguments, 4x uint64_t src, 4x dest)

this sounds very much like what Tim Forsyth was talking about.  design the ISA
so that the compiler can understand it.



> masks are generated by using compare intrinsics and by using bit-wise logic
> ops on masks. masks can be cast to/from integers.
> 
> the cr logic ops and the cr from int and cr to int ops are generated
> automatically by the instruction selector and by the register allocator.

it woule be spectacularly weird to be striping the CR eq/so/lt/gt bits across

-- 
You are receiving this mail because:
You are on the CC list for the bug.


More information about the libre-soc-bugs mailing list