Improve LLVM IR ECC doubling (Nvidia & AMD GPU backends)

[mratsim]

Port #510 to

constantine/constantine/math_compiler/impl_curves_ops_jacobian.nim

Lines 482 to 524 in 5d66b52

	# "dbl-2009-l" doubling formula - https://www.hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-0.html#doubling-dbl-2009-l
	#
	# Cost: 2M + 5S + 6add + 3*2 + 1*3 + 1*8.
	# Source: 2009.04.01 Lange.
	# Explicit formulas:
	#
	# A = X₁²
	# B = Y₁²
	# C = B²
	# D = 2*((X₁+B)²-A-C)
	# E = 3*A
	# F = E²
	# X₃ = F-2*D
	# Y₃ = E*(D-X₃)-8*C
	# Z₃ = 2*Y₁*Z₁
	#
	A.square(P.x)
	B.square(P.y)
	C.square(B)
	B += P.x
	# aliasing: we don't use P.x anymore
	B.square()
	B -= A
	B -= C
	B.double() # D = 2*((X₁+B)²-A-C)
	A *= 3 # E = 3*A
	rA.x.square(A) # F = E²
	rA.x -= B
	rA.x -= B # X₃ = F-2*D
	B -= rA.x # (D-X₃)
	A *= B # E*(D-X₃)
	C *= 8
	rA.z.prod(P.z, P.y)
	rA.z.double() # Z₃ = 2*Y₁*Z₁
	# aliasing: we don't use P.y, P.z anymore
	rA.y.diff(A, C) # Y₃ = E*(D-X₃)-8*C
	asy.br.retVoid()

[shreyas-londhe]

Hi @mratsim I’m interested in working on this.