Implement finite field `ccopy`, `neg`, `cneg`, `nsqr`, ... for CUDA target#466

Merged

mratsim merged 85 commits intomasterfrom moreFFnvidia

Oct 27, 2024

+3,782-47

Commits on Oct 27, 2024

minor docstring fixes / improvements

Vindaar
authored and
mratsim
committed
borrow isNil for BasicBlockRef

Vindaar
authored and
mratsim
committed
fix ABI for LLVM neg procedures

Vindaar
authored and
mratsim
committed
add phi and branch related LLVM ABI functions

Vindaar
authored and
mratsim
committed
inject LLVM fn in llvmFnDef for caller

Vindaar
authored and
mratsim
committed
add logic to generate ccopy, neg, cneg, nsqr

Vindaar
authored and
mratsim
committed
[tests] add test cases for ccopy, neg, cneg, nsqr, nsqrRT

Vindaar
authored and
mratsim
committed
change ccopy to not have additional extra argument

Vindaar
authored and
mratsim
committed
split cneg into internal/public part, fix logic for ccopy change

Vindaar
authored and
mratsim
committed
[tests] fix ccopy test case for API change

Vindaar
authored and
mratsim
committed
[tests] remove extra pointers in nsqr test

Vindaar
authored and
mratsim
committed
add setZero internal / public finite field proc

Vindaar
authored and
mratsim
committed
split finite field sub into internal/public procs

Vindaar
authored and
mratsim
committed
split finite field mul into internal/public procs

Vindaar
authored and
mratsim
committed
split finite field nsqr into internal / public procs

Vindaar
authored and
mratsim
committed
add finite field isZero proc

Vindaar
authored and
mratsim
committed
add basic CurveDescriptor type

Vindaar
authored and
mratsim
committed
add asArray overload taking BuilderRef

Vindaar
authored and
mratsim
committed
add more generic getElementPtr

Vindaar
authored and
mratsim
committed
add wrapper Field, EcPoint that are distinct arrays

Vindaar
authored and
mratsim
committed
port required code precompute -> impl_field_globas for p+1 div 2

Vindaar
authored and
mratsim
committed
add proc to get pointer to the prime plus 1 div 2 value

Vindaar
authored and
mratsim
committed
add internal add, fix up docstrings for internal procs

Vindaar
authored and
mratsim
committed
add cadd, csub, shiftRight, div2, double, isOdd for Fp

Vindaar
authored and
mratsim
committed
[tests] add test case for EC point component retrieval

Vindaar
authored and
mratsim
committed
[tests] add test case for mul

Vindaar
authored and
mratsim
committed
[tests] add "test" case implementing EC addition

Vindaar
authored and
mratsim
committed
rename t_ec_sum_impl to t_ec_sum_port, add doc comment

Vindaar
authored and
mratsim
committed
[nvidia] add macro execCuda to execute CUDA kernel with var args

Vindaar
authored and
mratsim
committed
[nvidia] add NvidiaAssembler helper to simplify init & compile

Vindaar
authored and
mratsim
committed
add elliptic curve implementations for LLVM / Nvidia target

Vindaar
authored and
mratsim
committed
[tests] use NvidiaAssembler in Nvidia tests

Vindaar
authored and
mratsim
committed
[tests] add EC sum on Nvidia test using pub_curves

Vindaar
authored and
mratsim
committed
run all nvidia tests for test_nvidia nimble task

Vindaar
authored and
mratsim
committed
add overload of execCuda without inputs argument

Vindaar
authored and
mratsim
committed
add neg and cneg for EC points

Vindaar
authored and
mratsim
committed
port scalar multiplication with CT integer for finite fields to LLVM

Vindaar
authored and
mratsim
committed
add double for EC points

Vindaar
authored and
mratsim
committed
prepare for different EcPoint types, move to pub_curves_jacobian

Vindaar
authored and
mratsim
committed
port precompute code for Montgomery 'One' for LLVM

Vindaar
authored and
mratsim
committed
remove Assembler_LLVM argument from store for field / EC

Vindaar
authored and
mratsim
committed
add setOne and csetZero for finite fields

Vindaar
authored and
mratsim
committed
add file for LLVM EC points in affine coords w/ isNeutral

Vindaar
authored and
mratsim
committed
add asEcPointJac overload taking CurveDescritpor

Vindaar
authored and
mratsim
committed
add fromAffine for LLVM

Vindaar
authored and
mratsim
committed
fix type passed in genEcIsNeutral for affine coords

Vindaar
authored and
mratsim
committed
add template fieldOps, ellipticOps to simplify operations

Vindaar
authored and
mratsim
committed
allocate limbs for BigNum in LLVM 'precompute'

Vindaar
authored and
mratsim
committed
remove leftover ptrBool line

Vindaar
authored and
mratsim
committed
fix setOne template pointing to setZero

Vindaar
authored and
mratsim
committed
fix setOne for finite field

Vindaar
authored and
mratsim
committed
improve derefBool logic to only maybe deref & raise if not bool

Vindaar
authored and
mratsim
committed
use derefBool in other conditionals

Vindaar
authored and
mratsim
committed
add csetOne for fields

Vindaar
authored and
mratsim
committed
add csetOne for fields

Vindaar
authored and
mratsim
committed
define destructor for Assembler_LLVM if ARC/ORC

Vindaar
authored and
mratsim
committed
fix global retrieval possibly yielding nil if already called

Vindaar
authored and
mratsim
committed
call destroy for Assembler_LLVM field for NvidiaAssembler

Vindaar
authored and
mratsim
committed
rename isNeutral for affine coords to avoid name clash

Vindaar
authored and
mratsim
committed
add ellipticAffOps for affine coordinates

Vindaar
authored and
mratsim
committed
adjust EcPointJac templates for added affine template

Vindaar
authored and
mratsim
committed
add mixedSum between Jacobian and Affine coordinates

Vindaar
authored and
mratsim
committed
[tests] add test case for jacobian + affine coords

Vindaar
authored and
mratsim
committed
[fields] use final reduce only in last iteration, expose finalReduce

Vindaar
authored and
mratsim
committed
[misc] remove int type TODO & forgotten ptrBool

Vindaar
authored and
mratsim
committed
make scalarMul take Nim RT value for LLVM

Vindaar
authored and
mratsim
committed
assign curve coefficients a, b from configureCurve

Vindaar
authored and
mratsim
committed
[fields] support to skip final subtraction in mul related ops

Vindaar
authored and
mratsim
committed
partially use skipFinalSub in EC sum

Vindaar
authored and
mratsim
committed
support other branches of coef_a in EC (mixed) sum

Vindaar
authored and
mratsim
committed
move internal LLVM procs to impl files

Vindaar
authored and
mratsim
committed
remove _internal suffix from procs, repl by "_impl" as string name

Vindaar
authored and
mratsim
committed
add requiresCopy for execCuda macro logic back in

Vindaar
authored and
mratsim
committed
simplify neg by avoiding slct, use select

Vindaar
authored and
mratsim
committed
[codegen] allow non ident / sym arguments to execCuda

Vindaar
authored and
mratsim
committed
[codegen] expand requires copy to allow copy for res parameters

Vindaar
authored and
mratsim
committed
[codegen] allow literals, consts, gen locals for them, allow ptr T

Vindaar
authored and
mratsim
committed
[llvm] wrap (subset of) float types of LLVM

Vindaar
authored and
mratsim
committed
[tests] add test case for LLVM execCuda with different types of args

Vindaar
authored and
mratsim
committed
[nimble] add execCuda test case to nimble file

Vindaar
authored and
mratsim
committed
[docs] remove outdated TODO and update execCuda docstring

Vindaar
authored and
mratsim
committed
implement setNeutral for EC points on LLVM

Vindaar
authored and
mratsim
committed
[tests] add WIP test case for misc EC functions

Vindaar
authored and
mratsim
committed
ec-nvidia: rename internal field/templates
mratsim
committed
ec-nvidia: switch to 32-bit due to upstream fused-multiply-add-with-carry 64-bit bug
mratsim
committed