diff --git a/core/crypto/_fiat/README.md b/core/crypto/_fiat/README.md
new file mode 100644
index 000000000..cd510d442
--- /dev/null
+++ b/core/crypto/_fiat/README.md
@@ -0,0 +1,35 @@
+# fiat
+
+This package contains low level arithmetic required to implement certain
+cryptographic primitives, ported from the [fiat-crypto project][1]
+along with some higher-level helpers.
+
+## Notes
+
+fiat-crypto gives the choice of 3 licenses for derived works.  The 1-Clause
+BSD license is chosen as it is compatible with Odin's existing licensing.
+
+The routines are intended to be timing-safe, as long as the underlying
+integer arithmetic is constant time.  This is true on most systems commonly
+used today, with the notable exception of WASM.
+
+While fiat-crypto provides both output targeting both 32-bit and 64-bit
+architectures, only the 64-bit versions were used, as 32-bit architectures
+are becoming increasingly uncommon and irrelevant.
+
+With the current Odin syntax, the Go output is trivially ported in most
+cases and was used as the basis of the port.
+
+In the future, it would be better to auto-generate Odin either directly
+by adding an appropriate code-gen backend written in Coq, or perhaps by
+parsing the JSON output.
+
+As this is a port rather than autogenerated output, none of fiat-crypto's
+formal verification guarantees apply, unless it is possible to prove binary
+equivalence.
+
+For the most part, alterations to the base fiat-crypto generated code was
+kept to a minimum, to aid auditability.  This results in a somewhat
+ideosyncratic style, and in some cases minor performance penalties.
+
+[1]: https://github.com/mit-plv/fiat-crypto
diff --git a/core/crypto/_fiat/fiat.odin b/core/crypto/_fiat/fiat.odin
new file mode 100644
index 000000000..ae9727149
--- /dev/null
+++ b/core/crypto/_fiat/fiat.odin
@@ -0,0 +1,24 @@
+package fiat
+
+// This package provides various helpers and types common to all of the
+// fiat-crypto derived backends.
+
+// This code only works on a two's complement system.
+#assert((-1 & 3) == 3)
+
+u1 :: distinct u8
+i1 :: distinct i8
+
+cmovznz_u64 :: #force_inline proc "contextless" (arg1: u1, arg2, arg3: u64) -> (out1: u64) {
+	x1 := (u64(arg1) * 0xffffffffffffffff)
+	x2 := ((x1 & arg3) | ((~x1) & arg2))
+	out1 = x2
+	return
+}
+
+cmovznz_u32 :: #force_inline proc "contextless" (arg1: u1, arg2, arg3: u32) -> (out1: u32) {
+	x1 := (u32(arg1) * 0xffffffff)
+	x2 := ((x1 & arg3) | ((~x1) & arg2))
+	out1 = x2
+	return
+}
diff --git a/core/crypto/_fiat/field_curve25519/field.odin b/core/crypto/_fiat/field_curve25519/field.odin
new file mode 100644
index 000000000..faf8ae3f7
--- /dev/null
+++ b/core/crypto/_fiat/field_curve25519/field.odin
@@ -0,0 +1,138 @@
+package field_curve25519
+
+import "core:crypto"
+import "core:mem"
+
+fe_relax_cast :: #force_inline proc "contextless" (arg1: ^Tight_Field_Element) -> ^Loose_Field_Element {
+	return transmute(^Loose_Field_Element)(arg1)
+}
+
+fe_tighten_cast :: #force_inline proc "contextless" (arg1: ^Loose_Field_Element) -> ^Tight_Field_Element {
+	return transmute(^Tight_Field_Element)(arg1)
+}
+
+fe_from_bytes :: proc "contextless" (out1: ^Tight_Field_Element, arg1: ^[32]byte) {
+	// Ignore the unused bit by copying the input and masking the bit off
+	// prior to deserialization.
+	tmp1: [32]byte = ---
+	copy_slice(tmp1[:], arg1[:])
+	tmp1[31] &= 127
+
+	_fe_from_bytes(out1, &tmp1)
+
+	mem.zero_explicit(&tmp1, size_of(tmp1))
+}
+
+fe_equal :: proc "contextless" (arg1, arg2: ^Tight_Field_Element) -> int {
+	tmp2: [32]byte = ---
+
+	fe_to_bytes(&tmp2, arg2)
+	ret := fe_equal_bytes(arg1, &tmp2)
+
+	mem.zero_explicit(&tmp2, size_of(tmp2))
+
+	return ret
+}
+
+fe_equal_bytes :: proc "contextless" (arg1: ^Tight_Field_Element, arg2: ^[32]byte) -> int {
+	tmp1: [32]byte = ---
+
+	fe_to_bytes(&tmp1, arg1)
+
+	ret := crypto.compare_constant_time(tmp1[:], arg2[:])
+
+	mem.zero_explicit(&tmp1, size_of(tmp1))
+
+	return ret
+}
+
+fe_carry_pow2k :: proc (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element, arg2: uint) {
+	// Special case: `arg1^(2 * 0) = 1`, though this should never happen.
+	if arg2 == 0 {
+		fe_one(out1)
+		return
+	}
+
+	fe_carry_square(out1, arg1)
+	for _ in 1..<arg2 {
+		fe_carry_square(out1, fe_relax_cast(out1))
+	}
+}
+
+fe_carry_opp :: #force_inline proc "contextless" (out1, arg1: ^Tight_Field_Element) {
+	fe_opp(fe_relax_cast(out1), arg1)
+	fe_carry(out1, fe_relax_cast(out1))
+}
+
+fe_carry_invsqrt :: proc (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element) -> int {
+	// Inverse square root taken from Monocypher.
+
+	tmp1, tmp2, tmp3: Tight_Field_Element = ---, ---, ---
+
+	// t0 = x^((p-5)/8)
+	// Can be achieved with a simple double & add ladder,
+	// but it would be slower.
+	fe_carry_pow2k(&tmp1, arg1, 1)
+	fe_carry_pow2k(&tmp2, fe_relax_cast(&tmp1), 2)
+	fe_carry_mul(&tmp2, arg1, fe_relax_cast(&tmp2))
+	fe_carry_mul(&tmp1, fe_relax_cast(&tmp1), fe_relax_cast(&tmp2))
+	fe_carry_pow2k(&tmp1, fe_relax_cast(&tmp1), 1)
+	fe_carry_mul(&tmp1, fe_relax_cast(&tmp2), fe_relax_cast(&tmp1))
+	fe_carry_pow2k(&tmp2, fe_relax_cast(&tmp1), 5)
+	fe_carry_mul(&tmp1, fe_relax_cast(&tmp2), fe_relax_cast(&tmp1))
+	fe_carry_pow2k(&tmp2, fe_relax_cast(&tmp1), 10)
+	fe_carry_mul(&tmp2, fe_relax_cast(&tmp2), fe_relax_cast(&tmp1))
+	fe_carry_pow2k(&tmp3, fe_relax_cast(&tmp2), 20)
+	fe_carry_mul(&tmp2, fe_relax_cast(&tmp3), fe_relax_cast(&tmp2))
+	fe_carry_pow2k(&tmp2, fe_relax_cast(&tmp2), 10)
+	fe_carry_mul(&tmp1, fe_relax_cast(&tmp2), fe_relax_cast(&tmp1))
+	fe_carry_pow2k(&tmp2, fe_relax_cast(&tmp1), 50)
+	fe_carry_mul(&tmp2, fe_relax_cast(&tmp2), fe_relax_cast(&tmp1))
+	fe_carry_pow2k(&tmp3, fe_relax_cast(&tmp2), 100)
+	fe_carry_mul(&tmp2, fe_relax_cast(&tmp3), fe_relax_cast(&tmp2))
+	fe_carry_pow2k(&tmp2, fe_relax_cast(&tmp2), 50)
+	fe_carry_mul(&tmp1, fe_relax_cast(&tmp2), fe_relax_cast(&tmp1))
+	fe_carry_pow2k(&tmp1, fe_relax_cast(&tmp1), 2)
+	fe_carry_mul(&tmp1, fe_relax_cast(&tmp1), arg1)
+
+	// quartic = x^((p-1)/4)
+	quartic := &tmp2
+	fe_carry_square(quartic, fe_relax_cast(&tmp1))
+	fe_carry_mul(quartic, fe_relax_cast(quartic), arg1)
+
+	// Serialize quartic once to save on repeated serialization/sanitization.
+	quartic_buf: [32]byte = ---
+	fe_to_bytes(&quartic_buf, quartic)
+	check := &tmp3
+
+	fe_one(check)
+	p1 := fe_equal_bytes(check, &quartic_buf)
+	fe_carry_opp(check, check)
+	m1 := fe_equal_bytes(check, &quartic_buf)
+	fe_carry_opp(check, &SQRT_M1)
+	ms := fe_equal_bytes(check, &quartic_buf)
+
+	// if quartic == -1 or sqrt(-1)
+	// then  isr = x^((p-1)/4) * sqrt(-1)
+	// else  isr = x^((p-1)/4)
+	fe_carry_mul(out1, fe_relax_cast(&tmp1), fe_relax_cast(&SQRT_M1))
+	fe_cond_assign(out1, &tmp1, (m1|ms) ~ 1)
+
+	mem.zero_explicit(&tmp1, size_of(tmp1))
+	mem.zero_explicit(&tmp2, size_of(tmp2))
+	mem.zero_explicit(&tmp3, size_of(tmp3))
+	mem.zero_explicit(&quartic_buf, size_of(quartic_buf))
+
+	return p1 | m1
+}
+
+fe_carry_inv :: proc (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element) {
+	tmp1: Tight_Field_Element
+
+	fe_carry_square(&tmp1, arg1)
+	_ = fe_carry_invsqrt(&tmp1, fe_relax_cast(&tmp1))
+	fe_carry_square(&tmp1, fe_relax_cast(&tmp1))
+	fe_carry_mul(out1, fe_relax_cast(&tmp1), arg1)
+
+	mem.zero_explicit(&tmp1, size_of(tmp1))
+}
diff --git a/core/crypto/_fiat/field_curve25519/field51.odin b/core/crypto/_fiat/field_curve25519/field51.odin
new file mode 100644
index 000000000..e4ca98b57
--- /dev/null
+++ b/core/crypto/_fiat/field_curve25519/field51.odin
@@ -0,0 +1,616 @@
+// The BSD 1-Clause License (BSD-1-Clause)
+//
+// Copyright (c) 2015-2020 the fiat-crypto authors (see the AUTHORS file)
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     1. Redistributions of source code must retain the above copyright
+//        notice, this list of conditions and the following disclaimer.
+//
+// THIS SOFTWARE IS PROVIDED BY the fiat-crypto authors "AS IS"
+// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
+// THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+// PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL Berkeley Software Design,
+// Inc. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+// LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+package field_curve25519
+
+// The file provides arithmetic on the field Z/(2^255-19) using
+// unsaturated 64-bit integer arithmetic.  It is derived primarily
+// from the machine generated Golang output from the fiat-crypto project.
+//
+// While the base implementation is provably correct, this implementation
+// makes no such claims as the port and optimizations were done by hand.
+// At some point, it may be worth adding support to fiat-crypto for
+// generating Odin output.
+//
+// TODO:
+//  * When fiat-crypto supports it, using a saturated 64-bit limbs
+//    instead of 51-bit limbs will be faster, though the gains are
+//    minimal unless adcx/adox/mulx are used.
+
+import fiat "core:crypto/_fiat"
+import "core:math/bits"
+
+Loose_Field_Element :: distinct [5]u64
+Tight_Field_Element :: distinct [5]u64
+
+SQRT_M1 := Tight_Field_Element{
+	1718705420411056,
+	234908883556509,
+	2233514472574048,
+	2117202627021982,
+	765476049583133,
+}
+
+_addcarryx_u51 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3: u64) -> (out1: u64, out2: fiat.u1) {
+	x1 := ((u64(arg1) + arg2) + arg3)
+	x2 := (x1 & 0x7ffffffffffff)
+	x3 := fiat.u1((x1 >> 51))
+	out1 = x2
+	out2 = x3
+	return
+}
+
+_subborrowx_u51 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3: u64) -> (out1: u64, out2: fiat.u1) {
+	x1 := ((i64(arg2) - i64(arg1)) - i64(arg3))
+	x2 := fiat.i1((x1 >> 51))
+	x3 := (u64(x1) & 0x7ffffffffffff)
+	out1 = x3
+	out2 = (0x0 - fiat.u1(x2))
+	return
+}
+
+fe_carry_mul :: proc (out1: ^Tight_Field_Element, arg1, arg2: ^Loose_Field_Element) {
+	x2, x1 := bits.mul_u64(arg1[4], (arg2[4] * 0x13))
+	x4, x3 := bits.mul_u64(arg1[4], (arg2[3] * 0x13))
+	x6, x5 := bits.mul_u64(arg1[4], (arg2[2] * 0x13))
+	x8, x7 := bits.mul_u64(arg1[4], (arg2[1] * 0x13))
+	x10, x9 := bits.mul_u64(arg1[3], (arg2[4] * 0x13))
+	x12, x11 := bits.mul_u64(arg1[3], (arg2[3] * 0x13))
+	x14, x13 := bits.mul_u64(arg1[3], (arg2[2] * 0x13))
+	x16, x15 := bits.mul_u64(arg1[2], (arg2[4] * 0x13))
+	x18, x17 := bits.mul_u64(arg1[2], (arg2[3] * 0x13))
+	x20, x19 := bits.mul_u64(arg1[1], (arg2[4] * 0x13))
+	x22, x21 := bits.mul_u64(arg1[4], arg2[0])
+	x24, x23 := bits.mul_u64(arg1[3], arg2[1])
+	x26, x25 := bits.mul_u64(arg1[3], arg2[0])
+	x28, x27 := bits.mul_u64(arg1[2], arg2[2])
+	x30, x29 := bits.mul_u64(arg1[2], arg2[1])
+	x32, x31 := bits.mul_u64(arg1[2], arg2[0])
+	x34, x33 := bits.mul_u64(arg1[1], arg2[3])
+	x36, x35 := bits.mul_u64(arg1[1], arg2[2])
+	x38, x37 := bits.mul_u64(arg1[1], arg2[1])
+	x40, x39 := bits.mul_u64(arg1[1], arg2[0])
+	x42, x41 := bits.mul_u64(arg1[0], arg2[4])
+	x44, x43 := bits.mul_u64(arg1[0], arg2[3])
+	x46, x45 := bits.mul_u64(arg1[0], arg2[2])
+	x48, x47 := bits.mul_u64(arg1[0], arg2[1])
+	x50, x49 := bits.mul_u64(arg1[0], arg2[0])
+	x51, x52 := bits.add_u64(x13, x7, u64(0x0))
+	x53, _ := bits.add_u64(x14, x8, u64(fiat.u1(x52)))
+	x55, x56 := bits.add_u64(x17, x51, u64(0x0))
+	x57, _ := bits.add_u64(x18, x53, u64(fiat.u1(x56)))
+	x59, x60 := bits.add_u64(x19, x55, u64(0x0))
+	x61, _ := bits.add_u64(x20, x57, u64(fiat.u1(x60)))
+	x63, x64 := bits.add_u64(x49, x59, u64(0x0))
+	x65, _ := bits.add_u64(x50, x61, u64(fiat.u1(x64)))
+	x67 := ((x63 >> 51) | ((x65 << 13) & 0xffffffffffffffff))
+	x68 := (x63 & 0x7ffffffffffff)
+	x69, x70 := bits.add_u64(x23, x21, u64(0x0))
+	x71, _ := bits.add_u64(x24, x22, u64(fiat.u1(x70)))
+	x73, x74 := bits.add_u64(x27, x69, u64(0x0))
+	x75, _ := bits.add_u64(x28, x71, u64(fiat.u1(x74)))
+	x77, x78 := bits.add_u64(x33, x73, u64(0x0))
+	x79, _ := bits.add_u64(x34, x75, u64(fiat.u1(x78)))
+	x81, x82 := bits.add_u64(x41, x77, u64(0x0))
+	x83, _ := bits.add_u64(x42, x79, u64(fiat.u1(x82)))
+	x85, x86 := bits.add_u64(x25, x1, u64(0x0))
+	x87, _ := bits.add_u64(x26, x2, u64(fiat.u1(x86)))
+	x89, x90 := bits.add_u64(x29, x85, u64(0x0))
+	x91, _ := bits.add_u64(x30, x87, u64(fiat.u1(x90)))
+	x93, x94 := bits.add_u64(x35, x89, u64(0x0))
+	x95, _ := bits.add_u64(x36, x91, u64(fiat.u1(x94)))
+	x97, x98 := bits.add_u64(x43, x93, u64(0x0))
+	x99, _ := bits.add_u64(x44, x95, u64(fiat.u1(x98)))
+	x101, x102 := bits.add_u64(x9, x3, u64(0x0))
+	x103, _ := bits.add_u64(x10, x4, u64(fiat.u1(x102)))
+	x105, x106 := bits.add_u64(x31, x101, u64(0x0))
+	x107, _ := bits.add_u64(x32, x103, u64(fiat.u1(x106)))
+	x109, x110 := bits.add_u64(x37, x105, u64(0x0))
+	x111, _ := bits.add_u64(x38, x107, u64(fiat.u1(x110)))
+	x113, x114 := bits.add_u64(x45, x109, u64(0x0))
+	x115, _ := bits.add_u64(x46, x111, u64(fiat.u1(x114)))
+	x117, x118 := bits.add_u64(x11, x5, u64(0x0))
+	x119, _ := bits.add_u64(x12, x6, u64(fiat.u1(x118)))
+	x121, x122 := bits.add_u64(x15, x117, u64(0x0))
+	x123, _ := bits.add_u64(x16, x119, u64(fiat.u1(x122)))
+	x125, x126 := bits.add_u64(x39, x121, u64(0x0))
+	x127, _ := bits.add_u64(x40, x123, u64(fiat.u1(x126)))
+	x129, x130 := bits.add_u64(x47, x125, u64(0x0))
+	x131, _ := bits.add_u64(x48, x127, u64(fiat.u1(x130)))
+	x133, x134 := bits.add_u64(x67, x129, u64(0x0))
+	x135 := (u64(fiat.u1(x134)) + x131)
+	x136 := ((x133 >> 51) | ((x135 << 13) & 0xffffffffffffffff))
+	x137 := (x133 & 0x7ffffffffffff)
+	x138, x139 := bits.add_u64(x136, x113, u64(0x0))
+	x140 := (u64(fiat.u1(x139)) + x115)
+	x141 := ((x138 >> 51) | ((x140 << 13) & 0xffffffffffffffff))
+	x142 := (x138 & 0x7ffffffffffff)
+	x143, x144 := bits.add_u64(x141, x97, u64(0x0))
+	x145 := (u64(fiat.u1(x144)) + x99)
+	x146 := ((x143 >> 51) | ((x145 << 13) & 0xffffffffffffffff))
+	x147 := (x143 & 0x7ffffffffffff)
+	x148, x149 := bits.add_u64(x146, x81, u64(0x0))
+	x150 := (u64(fiat.u1(x149)) + x83)
+	x151 := ((x148 >> 51) | ((x150 << 13) & 0xffffffffffffffff))
+	x152 := (x148 & 0x7ffffffffffff)
+	x153 := (x151 * 0x13)
+	x154 := (x68 + x153)
+	x155 := (x154 >> 51)
+	x156 := (x154 & 0x7ffffffffffff)
+	x157 := (x155 + x137)
+	x158 := fiat.u1((x157 >> 51))
+	x159 := (x157 & 0x7ffffffffffff)
+	x160 := (u64(x158) + x142)
+	out1[0] = x156
+	out1[1] = x159
+	out1[2] = x160
+	out1[3] = x147
+	out1[4] = x152
+}
+
+fe_carry_square :: proc (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element) {
+	x1 := (arg1[4] * 0x13)
+	x2 := (x1 * 0x2)
+	x3 := (arg1[4] * 0x2)
+	x4 := (arg1[3] * 0x13)
+	x5 := (x4 * 0x2)
+	x6 := (arg1[3] * 0x2)
+	x7 := (arg1[2] * 0x2)
+	x8 := (arg1[1] * 0x2)
+	x10, x9 := bits.mul_u64(arg1[4], x1)
+	x12, x11 := bits.mul_u64(arg1[3], x2)
+	x14, x13 := bits.mul_u64(arg1[3], x4)
+	x16, x15 := bits.mul_u64(arg1[2], x2)
+	x18, x17 := bits.mul_u64(arg1[2], x5)
+	x20, x19 := bits.mul_u64(arg1[2], arg1[2])
+	x22, x21 := bits.mul_u64(arg1[1], x2)
+	x24, x23 := bits.mul_u64(arg1[1], x6)
+	x26, x25 := bits.mul_u64(arg1[1], x7)
+	x28, x27 := bits.mul_u64(arg1[1], arg1[1])
+	x30, x29 := bits.mul_u64(arg1[0], x3)
+	x32, x31 := bits.mul_u64(arg1[0], x6)
+	x34, x33 := bits.mul_u64(arg1[0], x7)
+	x36, x35 := bits.mul_u64(arg1[0], x8)
+	x38, x37 := bits.mul_u64(arg1[0], arg1[0])
+	x39, x40 := bits.add_u64(x21, x17, u64(0x0))
+	x41, _ := bits.add_u64(x22, x18, u64(fiat.u1(x40)))
+	x43, x44 := bits.add_u64(x37, x39, u64(0x0))
+	x45, _ := bits.add_u64(x38, x41, u64(fiat.u1(x44)))
+	x47 := ((x43 >> 51) | ((x45 << 13) & 0xffffffffffffffff))
+	x48 := (x43 & 0x7ffffffffffff)
+	x49, x50 := bits.add_u64(x23, x19, u64(0x0))
+	x51, _ := bits.add_u64(x24, x20, u64(fiat.u1(x50)))
+	x53, x54 := bits.add_u64(x29, x49, u64(0x0))
+	x55, _ := bits.add_u64(x30, x51, u64(fiat.u1(x54)))
+	x57, x58 := bits.add_u64(x25, x9, u64(0x0))
+	x59, _ := bits.add_u64(x26, x10, u64(fiat.u1(x58)))
+	x61, x62 := bits.add_u64(x31, x57, u64(0x0))
+	x63, _ := bits.add_u64(x32, x59, u64(fiat.u1(x62)))
+	x65, x66 := bits.add_u64(x27, x11, u64(0x0))
+	x67, _ := bits.add_u64(x28, x12, u64(fiat.u1(x66)))
+	x69, x70 := bits.add_u64(x33, x65, u64(0x0))
+	x71, _ := bits.add_u64(x34, x67, u64(fiat.u1(x70)))
+	x73, x74 := bits.add_u64(x15, x13, u64(0x0))
+	x75, _ := bits.add_u64(x16, x14, u64(fiat.u1(x74)))
+	x77, x78 := bits.add_u64(x35, x73, u64(0x0))
+	x79, _ := bits.add_u64(x36, x75, u64(fiat.u1(x78)))
+	x81, x82 := bits.add_u64(x47, x77, u64(0x0))
+	x83 := (u64(fiat.u1(x82)) + x79)
+	x84 := ((x81 >> 51) | ((x83 << 13) & 0xffffffffffffffff))
+	x85 := (x81 & 0x7ffffffffffff)
+	x86, x87 := bits.add_u64(x84, x69, u64(0x0))
+	x88 := (u64(fiat.u1(x87)) + x71)
+	x89 := ((x86 >> 51) | ((x88 << 13) & 0xffffffffffffffff))
+	x90 := (x86 & 0x7ffffffffffff)
+	x91, x92 := bits.add_u64(x89, x61, u64(0x0))
+	x93 := (u64(fiat.u1(x92)) + x63)
+	x94 := ((x91 >> 51) | ((x93 << 13) & 0xffffffffffffffff))
+	x95 := (x91 & 0x7ffffffffffff)
+	x96, x97 := bits.add_u64(x94, x53, u64(0x0))
+	x98 := (u64(fiat.u1(x97)) + x55)
+	x99 := ((x96 >> 51) | ((x98 << 13) & 0xffffffffffffffff))
+	x100 := (x96 & 0x7ffffffffffff)
+	x101 := (x99 * 0x13)
+	x102 := (x48 + x101)
+	x103 := (x102 >> 51)
+	x104 := (x102 & 0x7ffffffffffff)
+	x105 := (x103 + x85)
+	x106 := fiat.u1((x105 >> 51))
+	x107 := (x105 & 0x7ffffffffffff)
+	x108 := (u64(x106) + x90)
+	out1[0] = x104
+	out1[1] = x107
+	out1[2] = x108
+	out1[3] = x95
+	out1[4] = x100
+}
+
+fe_carry :: proc "contextless" (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element) {
+	x1 := arg1[0]
+	x2 := ((x1 >> 51) + arg1[1])
+	x3 := ((x2 >> 51) + arg1[2])
+	x4 := ((x3 >> 51) + arg1[3])
+	x5 := ((x4 >> 51) + arg1[4])
+	x6 := ((x1 & 0x7ffffffffffff) + ((x5 >> 51) * 0x13))
+	x7 := (u64(fiat.u1((x6 >> 51))) + (x2 & 0x7ffffffffffff))
+	x8 := (x6 & 0x7ffffffffffff)
+	x9 := (x7 & 0x7ffffffffffff)
+	x10 := (u64(fiat.u1((x7 >> 51))) + (x3 & 0x7ffffffffffff))
+	x11 := (x4 & 0x7ffffffffffff)
+	x12 := (x5 & 0x7ffffffffffff)
+	out1[0] = x8
+	out1[1] = x9
+	out1[2] = x10
+	out1[3] = x11
+	out1[4] = x12
+}
+
+fe_add :: proc "contextless" (out1: ^Loose_Field_Element, arg1, arg2: ^Tight_Field_Element) {
+	x1 := (arg1[0] + arg2[0])
+	x2 := (arg1[1] + arg2[1])
+	x3 := (arg1[2] + arg2[2])
+	x4 := (arg1[3] + arg2[3])
+	x5 := (arg1[4] + arg2[4])
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+	out1[3] = x4
+	out1[4] = x5
+}
+
+fe_sub :: proc "contextless" (out1: ^Loose_Field_Element, arg1, arg2: ^Tight_Field_Element) {
+	x1 := ((0xfffffffffffda + arg1[0]) - arg2[0])
+	x2 := ((0xffffffffffffe + arg1[1]) - arg2[1])
+	x3 := ((0xffffffffffffe + arg1[2]) - arg2[2])
+	x4 := ((0xffffffffffffe + arg1[3]) - arg2[3])
+	x5 := ((0xffffffffffffe + arg1[4]) - arg2[4])
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+	out1[3] = x4
+	out1[4] = x5
+}
+
+fe_opp :: proc "contextless" (out1: ^Loose_Field_Element, arg1: ^Tight_Field_Element) {
+	x1 := (0xfffffffffffda - arg1[0])
+	x2 := (0xffffffffffffe - arg1[1])
+	x3 := (0xffffffffffffe - arg1[2])
+	x4 := (0xffffffffffffe - arg1[3])
+	x5 := (0xffffffffffffe - arg1[4])
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+	out1[3] = x4
+	out1[4] = x5
+}
+
+fe_cond_assign :: proc "contextless" (out1, arg1: ^Tight_Field_Element, arg2: int) {
+	x1 := fiat.cmovznz_u64(fiat.u1(arg2), out1[0], arg1[0])
+	x2 := fiat.cmovznz_u64(fiat.u1(arg2), out1[1], arg1[1])
+	x3 := fiat.cmovznz_u64(fiat.u1(arg2), out1[2], arg1[2])
+	x4 := fiat.cmovznz_u64(fiat.u1(arg2), out1[3], arg1[3])
+	x5 := fiat.cmovznz_u64(fiat.u1(arg2), out1[4], arg1[4])
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+	out1[3] = x4
+	out1[4] = x5
+}
+
+fe_to_bytes :: proc "contextless" (out1: ^[32]byte, arg1: ^Tight_Field_Element) {
+	x1, x2 := _subborrowx_u51(0x0, arg1[0], 0x7ffffffffffed)
+	x3, x4 := _subborrowx_u51(x2, arg1[1], 0x7ffffffffffff)
+	x5, x6 := _subborrowx_u51(x4, arg1[2], 0x7ffffffffffff)
+	x7, x8 := _subborrowx_u51(x6, arg1[3], 0x7ffffffffffff)
+	x9, x10 := _subborrowx_u51(x8, arg1[4], 0x7ffffffffffff)
+	x11 := fiat.cmovznz_u64(x10, u64(0x0), 0xffffffffffffffff)
+	x12, x13 := _addcarryx_u51(0x0, x1, (x11 & 0x7ffffffffffed))
+	x14, x15 := _addcarryx_u51(x13, x3, (x11 & 0x7ffffffffffff))
+	x16, x17 := _addcarryx_u51(x15, x5, (x11 & 0x7ffffffffffff))
+	x18, x19 := _addcarryx_u51(x17, x7, (x11 & 0x7ffffffffffff))
+	x20, _ := _addcarryx_u51(x19, x9, (x11 & 0x7ffffffffffff))
+	x22 := (x20 << 4)
+	x23 := (x18 * u64(0x2))
+	x24 := (x16 << 6)
+	x25 := (x14 << 3)
+	x26 := (u8(x12) & 0xff)
+	x27 := (x12 >> 8)
+	x28 := (u8(x27) & 0xff)
+	x29 := (x27 >> 8)
+	x30 := (u8(x29) & 0xff)
+	x31 := (x29 >> 8)
+	x32 := (u8(x31) & 0xff)
+	x33 := (x31 >> 8)
+	x34 := (u8(x33) & 0xff)
+	x35 := (x33 >> 8)
+	x36 := (u8(x35) & 0xff)
+	x37 := u8((x35 >> 8))
+	x38 := (x25 + u64(x37))
+	x39 := (u8(x38) & 0xff)
+	x40 := (x38 >> 8)
+	x41 := (u8(x40) & 0xff)
+	x42 := (x40 >> 8)
+	x43 := (u8(x42) & 0xff)
+	x44 := (x42 >> 8)
+	x45 := (u8(x44) & 0xff)
+	x46 := (x44 >> 8)
+	x47 := (u8(x46) & 0xff)
+	x48 := (x46 >> 8)
+	x49 := (u8(x48) & 0xff)
+	x50 := u8((x48 >> 8))
+	x51 := (x24 + u64(x50))
+	x52 := (u8(x51) & 0xff)
+	x53 := (x51 >> 8)
+	x54 := (u8(x53) & 0xff)
+	x55 := (x53 >> 8)
+	x56 := (u8(x55) & 0xff)
+	x57 := (x55 >> 8)
+	x58 := (u8(x57) & 0xff)
+	x59 := (x57 >> 8)
+	x60 := (u8(x59) & 0xff)
+	x61 := (x59 >> 8)
+	x62 := (u8(x61) & 0xff)
+	x63 := (x61 >> 8)
+	x64 := (u8(x63) & 0xff)
+	x65 := fiat.u1((x63 >> 8))
+	x66 := (x23 + u64(x65))
+	x67 := (u8(x66) & 0xff)
+	x68 := (x66 >> 8)
+	x69 := (u8(x68) & 0xff)
+	x70 := (x68 >> 8)
+	x71 := (u8(x70) & 0xff)
+	x72 := (x70 >> 8)
+	x73 := (u8(x72) & 0xff)
+	x74 := (x72 >> 8)
+	x75 := (u8(x74) & 0xff)
+	x76 := (x74 >> 8)
+	x77 := (u8(x76) & 0xff)
+	x78 := u8((x76 >> 8))
+	x79 := (x22 + u64(x78))
+	x80 := (u8(x79) & 0xff)
+	x81 := (x79 >> 8)
+	x82 := (u8(x81) & 0xff)
+	x83 := (x81 >> 8)
+	x84 := (u8(x83) & 0xff)
+	x85 := (x83 >> 8)
+	x86 := (u8(x85) & 0xff)
+	x87 := (x85 >> 8)
+	x88 := (u8(x87) & 0xff)
+	x89 := (x87 >> 8)
+	x90 := (u8(x89) & 0xff)
+	x91 := u8((x89 >> 8))
+	out1[0] = x26
+	out1[1] = x28
+	out1[2] = x30
+	out1[3] = x32
+	out1[4] = x34
+	out1[5] = x36
+	out1[6] = x39
+	out1[7] = x41
+	out1[8] = x43
+	out1[9] = x45
+	out1[10] = x47
+	out1[11] = x49
+	out1[12] = x52
+	out1[13] = x54
+	out1[14] = x56
+	out1[15] = x58
+	out1[16] = x60
+	out1[17] = x62
+	out1[18] = x64
+	out1[19] = x67
+	out1[20] = x69
+	out1[21] = x71
+	out1[22] = x73
+	out1[23] = x75
+	out1[24] = x77
+	out1[25] = x80
+	out1[26] = x82
+	out1[27] = x84
+	out1[28] = x86
+	out1[29] = x88
+	out1[30] = x90
+	out1[31] = x91
+}
+
+_fe_from_bytes :: proc "contextless" (out1: ^Tight_Field_Element, arg1: ^[32]byte) {
+	x1 := (u64(arg1[31]) << 44)
+	x2 := (u64(arg1[30]) << 36)
+	x3 := (u64(arg1[29]) << 28)
+	x4 := (u64(arg1[28]) << 20)
+	x5 := (u64(arg1[27]) << 12)
+	x6 := (u64(arg1[26]) << 4)
+	x7 := (u64(arg1[25]) << 47)
+	x8 := (u64(arg1[24]) << 39)
+	x9 := (u64(arg1[23]) << 31)
+	x10 := (u64(arg1[22]) << 23)
+	x11 := (u64(arg1[21]) << 15)
+	x12 := (u64(arg1[20]) << 7)
+	x13 := (u64(arg1[19]) << 50)
+	x14 := (u64(arg1[18]) << 42)
+	x15 := (u64(arg1[17]) << 34)
+	x16 := (u64(arg1[16]) << 26)
+	x17 := (u64(arg1[15]) << 18)
+	x18 := (u64(arg1[14]) << 10)
+	x19 := (u64(arg1[13]) << 2)
+	x20 := (u64(arg1[12]) << 45)
+	x21 := (u64(arg1[11]) << 37)
+	x22 := (u64(arg1[10]) << 29)
+	x23 := (u64(arg1[9]) << 21)
+	x24 := (u64(arg1[8]) << 13)
+	x25 := (u64(arg1[7]) << 5)
+	x26 := (u64(arg1[6]) << 48)
+	x27 := (u64(arg1[5]) << 40)
+	x28 := (u64(arg1[4]) << 32)
+	x29 := (u64(arg1[3]) << 24)
+	x30 := (u64(arg1[2]) << 16)
+	x31 := (u64(arg1[1]) << 8)
+	x32 := arg1[0]
+	x33 := (x31 + u64(x32))
+	x34 := (x30 + x33)
+	x35 := (x29 + x34)
+	x36 := (x28 + x35)
+	x37 := (x27 + x36)
+	x38 := (x26 + x37)
+	x39 := (x38 & 0x7ffffffffffff)
+	x40 := u8((x38 >> 51))
+	x41 := (x25 + u64(x40))
+	x42 := (x24 + x41)
+	x43 := (x23 + x42)
+	x44 := (x22 + x43)
+	x45 := (x21 + x44)
+	x46 := (x20 + x45)
+	x47 := (x46 & 0x7ffffffffffff)
+	x48 := u8((x46 >> 51))
+	x49 := (x19 + u64(x48))
+	x50 := (x18 + x49)
+	x51 := (x17 + x50)
+	x52 := (x16 + x51)
+	x53 := (x15 + x52)
+	x54 := (x14 + x53)
+	x55 := (x13 + x54)
+	x56 := (x55 & 0x7ffffffffffff)
+	x57 := u8((x55 >> 51))
+	x58 := (x12 + u64(x57))
+	x59 := (x11 + x58)
+	x60 := (x10 + x59)
+	x61 := (x9 + x60)
+	x62 := (x8 + x61)
+	x63 := (x7 + x62)
+	x64 := (x63 & 0x7ffffffffffff)
+	x65 := u8((x63 >> 51))
+	x66 := (x6 + u64(x65))
+	x67 := (x5 + x66)
+	x68 := (x4 + x67)
+	x69 := (x3 + x68)
+	x70 := (x2 + x69)
+	x71 := (x1 + x70)
+	out1[0] = x39
+	out1[1] = x47
+	out1[2] = x56
+	out1[3] = x64
+	out1[4] = x71
+}
+
+fe_relax :: proc "contextless" (out1: ^Loose_Field_Element, arg1: ^Tight_Field_Element) {
+	x1 := arg1[0]
+	x2 := arg1[1]
+	x3 := arg1[2]
+	x4 := arg1[3]
+	x5 := arg1[4]
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+	out1[3] = x4
+	out1[4] = x5
+}
+
+fe_carry_scmul_121666 :: proc (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element) {
+	x2, x1 := bits.mul_u64(0x1db42, arg1[4])
+	x4, x3 := bits.mul_u64(0x1db42, arg1[3])
+	x6, x5 := bits.mul_u64(0x1db42, arg1[2])
+	x8, x7 := bits.mul_u64(0x1db42, arg1[1])
+	x10, x9 := bits.mul_u64(0x1db42, arg1[0])
+	x11 := ((x9 >> 51) | ((x10 << 13) & 0xffffffffffffffff))
+	x12 := (x9 & 0x7ffffffffffff)
+	x13, x14 := bits.add_u64(x11, x7, u64(0x0))
+	x15 := (u64(fiat.u1(x14)) + x8)
+	x16 := ((x13 >> 51) | ((x15 << 13) & 0xffffffffffffffff))
+	x17 := (x13 & 0x7ffffffffffff)
+	x18, x19 := bits.add_u64(x16, x5, u64(0x0))
+	x20 := (u64(fiat.u1(x19)) + x6)
+	x21 := ((x18 >> 51) | ((x20 << 13) & 0xffffffffffffffff))
+	x22 := (x18 & 0x7ffffffffffff)
+	x23, x24 := bits.add_u64(x21, x3, u64(0x0))
+	x25 := (u64(fiat.u1(x24)) + x4)
+	x26 := ((x23 >> 51) | ((x25 << 13) & 0xffffffffffffffff))
+	x27 := (x23 & 0x7ffffffffffff)
+	x28, x29 := bits.add_u64(x26, x1, u64(0x0))
+	x30 := (u64(fiat.u1(x29)) + x2)
+	x31 := ((x28 >> 51) | ((x30 << 13) & 0xffffffffffffffff))
+	x32 := (x28 & 0x7ffffffffffff)
+	x33 := (x31 * 0x13)
+	x34 := (x12 + x33)
+	x35 := fiat.u1((x34 >> 51))
+	x36 := (x34 & 0x7ffffffffffff)
+	x37 := (u64(x35) + x17)
+	x38 := fiat.u1((x37 >> 51))
+	x39 := (x37 & 0x7ffffffffffff)
+	x40 := (u64(x38) + x22)
+	out1[0] = x36
+	out1[1] = x39
+	out1[2] = x40
+	out1[3] = x27
+	out1[4] = x32
+}
+
+// The following routines were added by hand, and do not come from fiat-crypto.
+
+fe_zero :: proc "contextless" (out1: ^Tight_Field_Element) {
+	out1[0] = 0
+	out1[1] = 0
+	out1[2] = 0
+	out1[3] = 0
+	out1[4] = 0
+}
+
+fe_one :: proc "contextless" (out1: ^Tight_Field_Element) {
+	out1[0] = 1
+	out1[1] = 0
+	out1[2] = 0
+	out1[3] = 0
+	out1[4] = 0
+}
+
+fe_set :: proc "contextless" (out1, arg1: ^Tight_Field_Element) {
+	x1 := arg1[0]
+	x2 := arg1[1]
+	x3 := arg1[2]
+	x4 := arg1[3]
+	x5 := arg1[4]
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+	out1[3] = x4
+	out1[4] = x5
+}
+
+fe_cond_swap :: proc "contextless" (out1, out2: ^Tight_Field_Element, arg1: int) {
+	mask := -u64(arg1)
+	x := (out1[0] ~ out2[0]) & mask
+	x1, y1 := out1[0] ~ x, out2[0] ~ x
+	x = (out1[1] ~ out2[1]) & mask
+	x2, y2 := out1[1] ~ x, out2[1] ~ x
+	x = (out1[2] ~ out2[2]) & mask
+	x3, y3 := out1[2] ~ x, out2[2] ~ x
+	x = (out1[3] ~ out2[3]) & mask
+	x4, y4 := out1[3] ~ x, out2[3] ~ x
+	x = (out1[4] ~ out2[4]) & mask
+	x5, y5 := out1[4] ~ x, out2[4] ~ x
+	out1[0], out2[0] = x1, y1
+	out1[1], out2[1] = x2, y2
+	out1[2], out2[2] = x3, y3
+	out1[3], out2[3] = x4, y4
+	out1[4], out2[4] = x5, y5
+}
diff --git a/core/crypto/_fiat/field_poly1305/field.odin b/core/crypto/_fiat/field_poly1305/field.odin
new file mode 100644
index 000000000..bfb7cf1f9
--- /dev/null
+++ b/core/crypto/_fiat/field_poly1305/field.odin
@@ -0,0 +1,66 @@
+package field_poly1305
+
+import "core:crypto/util"
+import "core:mem"
+
+fe_relax_cast :: #force_inline proc "contextless" (arg1: ^Tight_Field_Element) -> ^Loose_Field_Element {
+	return transmute(^Loose_Field_Element)(arg1)
+}
+
+fe_tighten_cast :: #force_inline proc "contextless" (arg1: ^Loose_Field_Element) -> ^Tight_Field_Element {
+	return transmute(^Tight_Field_Element)(arg1)
+}
+
+fe_from_bytes :: #force_inline proc (out1: ^Tight_Field_Element, arg1: []byte, arg2: byte, sanitize: bool = true) {
+	// fiat-crypto's deserialization routine effectively processes a
+	// single byte at a time, and wants 256-bits of input for a value
+	// that will be 128-bits or 129-bits.
+	//
+	// This is somewhat cumbersome to use, so at a minimum a wrapper
+	// makes implementing the actual MAC block processing considerably
+	// neater.
+
+	assert(len(arg1) == 16)
+
+	when ODIN_ARCH == "386" || ODIN_ARCH == "amd64" {
+		// While it may be unwise to do deserialization here on our
+		// own when fiat-crypto provides equivalent functionality,
+		// doing it this way provides a little under 3x performance
+		// improvement when optimization is enabled.
+		src_p := transmute(^[2]u64)(&arg1[0])
+		lo := src_p[0]
+		hi := src_p[1]
+
+		// This is inspired by poly1305-donna, though adjustments were
+		// made since a Tight_Field_Element's limbs are 44-bits, 43-bits,
+		// and 43-bits wide.
+		//
+		// Note: This could be transplated into fe_from_u64s, but that
+		// code is called once per MAC, and is non-criticial path.
+		hibit := u64(arg2) << 41 // arg2 << 128
+		out1[0] = lo & 0xfffffffffff
+		out1[1] = ((lo >> 44) | (hi << 20)) & 0x7ffffffffff
+		out1[2] = ((hi >> 23) & 0x7ffffffffff) | hibit
+	} else {
+		tmp: [32]byte
+		copy_slice(tmp[0:16], arg1[:])
+		tmp[16] = arg2
+
+		_fe_from_bytes(out1, &tmp)
+		if sanitize {
+			// This is used to deserialize `s` which is confidential.
+			mem.zero_explicit(&tmp, size_of(tmp))
+		}
+	}
+}
+
+fe_from_u64s :: proc "contextless" (out1: ^Tight_Field_Element, lo, hi: u64) {
+	tmp: [32]byte
+	util.PUT_U64_LE(tmp[0:8], lo)
+	util.PUT_U64_LE(tmp[8:16], hi)
+
+	_fe_from_bytes(out1, &tmp)
+
+	// This routine is only used to deserialize `r` which is confidential.
+	mem.zero_explicit(&tmp, size_of(tmp))
+}
diff --git a/core/crypto/_fiat/field_poly1305/field4344.odin b/core/crypto/_fiat/field_poly1305/field4344.odin
new file mode 100644
index 000000000..ba9bc2694
--- /dev/null
+++ b/core/crypto/_fiat/field_poly1305/field4344.odin
@@ -0,0 +1,356 @@
+// The BSD 1-Clause License (BSD-1-Clause)
+//
+// Copyright (c) 2015-2020 the fiat-crypto authors (see the AUTHORS file)
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are
+// met:
+//
+//     1. Redistributions of source code must retain the above copyright
+//        notice, this list of conditions and the following disclaimer.
+//
+// THIS SOFTWARE IS PROVIDED BY the fiat-crypto authors "AS IS"
+// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
+// THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+// PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL Berkeley Software Design,
+// Inc. BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+// LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+package field_poly1305
+
+// This file provides arithmetic on the field Z/(2^130 - 5) using
+// unsaturated 64-bit integer arithmetic.  It is derived primarily
+// from the machine generate Golang output from the fiat-crypto project.
+//
+// While the base implementation is provably correct, this implementation
+// makes no such claims as the port and optimizations were done by hand.
+// At some point, it may be worth adding support to fiat-crypto for
+// generating Odin output.
+
+import fiat "core:crypto/_fiat"
+import "core:math/bits"
+
+Loose_Field_Element :: distinct [3]u64
+Tight_Field_Element :: distinct [3]u64
+
+_addcarryx_u44 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3: u64) -> (out1: u64, out2: fiat.u1) {
+	x1 := ((u64(arg1) + arg2) + arg3)
+	x2 := (x1 & 0xfffffffffff)
+	x3 := fiat.u1((x1 >> 44))
+	out1 = x2
+	out2 = x3
+	return
+}
+
+_subborrowx_u44 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3: u64) -> (out1: u64, out2: fiat.u1) {
+	x1 := ((i64(arg2) - i64(arg1)) - i64(arg3))
+	x2 := fiat.i1((x1 >> 44))
+	x3 := (u64(x1) & 0xfffffffffff)
+	out1 = x3
+	out2 = (0x0 - fiat.u1(x2))
+	return
+}
+
+_addcarryx_u43 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3: u64) -> (out1: u64, out2: fiat.u1) {
+	x1 := ((u64(arg1) + arg2) + arg3)
+	x2 := (x1 & 0x7ffffffffff)
+	x3 := fiat.u1((x1 >> 43))
+	out1 = x2
+	out2 = x3
+	return
+}
+
+_subborrowx_u43 :: #force_inline proc "contextless" (arg1: fiat.u1, arg2, arg3: u64) -> (out1: u64, out2: fiat.u1) {
+	x1 := ((i64(arg2) - i64(arg1)) - i64(arg3))
+	x2 := fiat.i1((x1 >> 43))
+	x3 := (u64(x1) & 0x7ffffffffff)
+	out1 = x3
+	out2 = (0x0 - fiat.u1(x2))
+	return
+}
+
+fe_carry_mul :: proc (out1: ^Tight_Field_Element, arg1, arg2: ^Loose_Field_Element) {
+	x2, x1 := bits.mul_u64(arg1[2], (arg2[2] * 0x5))
+	x4, x3 := bits.mul_u64(arg1[2], (arg2[1] * 0xa))
+	x6, x5 := bits.mul_u64(arg1[1], (arg2[2] * 0xa))
+	x8, x7 := bits.mul_u64(arg1[2], arg2[0])
+	x10, x9 := bits.mul_u64(arg1[1], (arg2[1] * 0x2))
+	x12, x11 := bits.mul_u64(arg1[1], arg2[0])
+	x14, x13 := bits.mul_u64(arg1[0], arg2[2])
+	x16, x15 := bits.mul_u64(arg1[0], arg2[1])
+	x18, x17 := bits.mul_u64(arg1[0], arg2[0])
+	x19, x20 := bits.add_u64(x5, x3, u64(0x0))
+	x21, _ := bits.add_u64(x6, x4, u64(fiat.u1(x20)))
+	x23, x24 := bits.add_u64(x17, x19, u64(0x0))
+	x25, _ := bits.add_u64(x18, x21, u64(fiat.u1(x24)))
+	x27 := ((x23 >> 44) | ((x25 << 20) & 0xffffffffffffffff))
+	x28 := (x23 & 0xfffffffffff)
+	x29, x30 := bits.add_u64(x9, x7, u64(0x0))
+	x31, _ := bits.add_u64(x10, x8, u64(fiat.u1(x30)))
+	x33, x34 := bits.add_u64(x13, x29, u64(0x0))
+	x35, _ := bits.add_u64(x14, x31, u64(fiat.u1(x34)))
+	x37, x38 := bits.add_u64(x11, x1, u64(0x0))
+	x39, _ := bits.add_u64(x12, x2, u64(fiat.u1(x38)))
+	x41, x42 := bits.add_u64(x15, x37, u64(0x0))
+	x43, _ := bits.add_u64(x16, x39, u64(fiat.u1(x42)))
+	x45, x46 := bits.add_u64(x27, x41, u64(0x0))
+	x47 := (u64(fiat.u1(x46)) + x43)
+	x48 := ((x45 >> 43) | ((x47 << 21) & 0xffffffffffffffff))
+	x49 := (x45 & 0x7ffffffffff)
+	x50, x51 := bits.add_u64(x48, x33, u64(0x0))
+	x52 := (u64(fiat.u1(x51)) + x35)
+	x53 := ((x50 >> 43) | ((x52 << 21) & 0xffffffffffffffff))
+	x54 := (x50 & 0x7ffffffffff)
+	x55 := (x53 * 0x5)
+	x56 := (x28 + x55)
+	x57 := (x56 >> 44)
+	x58 := (x56 & 0xfffffffffff)
+	x59 := (x57 + x49)
+	x60 := fiat.u1((x59 >> 43))
+	x61 := (x59 & 0x7ffffffffff)
+	x62 := (u64(x60) + x54)
+	out1[0] = x58
+	out1[1] = x61
+	out1[2] = x62
+}
+
+fe_carry_square :: proc (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element) {
+	x1 := (arg1[2] * 0x5)
+	x2 := (x1 * 0x2)
+	x3 := (arg1[2] * 0x2)
+	x4 := (arg1[1] * 0x2)
+	x6, x5 := bits.mul_u64(arg1[2], x1)
+	x8, x7 := bits.mul_u64(arg1[1], (x2 * 0x2))
+	x10, x9 := bits.mul_u64(arg1[1], (arg1[1] * 0x2))
+	x12, x11 := bits.mul_u64(arg1[0], x3)
+	x14, x13 := bits.mul_u64(arg1[0], x4)
+	x16, x15 := bits.mul_u64(arg1[0], arg1[0])
+	x17, x18 := bits.add_u64(x15, x7, u64(0x0))
+	x19, _ := bits.add_u64(x16, x8, u64(fiat.u1(x18)))
+	x21 := ((x17 >> 44) | ((x19 << 20) & 0xffffffffffffffff))
+	x22 := (x17 & 0xfffffffffff)
+	x23, x24 := bits.add_u64(x11, x9, u64(0x0))
+	x25, _ := bits.add_u64(x12, x10, u64(fiat.u1(x24)))
+	x27, x28 := bits.add_u64(x13, x5, u64(0x0))
+	x29, _ := bits.add_u64(x14, x6, u64(fiat.u1(x28)))
+	x31, x32 := bits.add_u64(x21, x27, u64(0x0))
+	x33 := (u64(fiat.u1(x32)) + x29)
+	x34 := ((x31 >> 43) | ((x33 << 21) & 0xffffffffffffffff))
+	x35 := (x31 & 0x7ffffffffff)
+	x36, x37 := bits.add_u64(x34, x23, u64(0x0))
+	x38 := (u64(fiat.u1(x37)) + x25)
+	x39 := ((x36 >> 43) | ((x38 << 21) & 0xffffffffffffffff))
+	x40 := (x36 & 0x7ffffffffff)
+	x41 := (x39 * 0x5)
+	x42 := (x22 + x41)
+	x43 := (x42 >> 44)
+	x44 := (x42 & 0xfffffffffff)
+	x45 := (x43 + x35)
+	x46 := fiat.u1((x45 >> 43))
+	x47 := (x45 & 0x7ffffffffff)
+	x48 := (u64(x46) + x40)
+	out1[0] = x44
+	out1[1] = x47
+	out1[2] = x48
+}
+
+fe_carry :: proc "contextless" (out1: ^Tight_Field_Element, arg1: ^Loose_Field_Element) {
+	x1 := arg1[0]
+	x2 := ((x1 >> 44) + arg1[1])
+	x3 := ((x2 >> 43) + arg1[2])
+	x4 := ((x1 & 0xfffffffffff) + ((x3 >> 43) * 0x5))
+	x5 := (u64(fiat.u1((x4 >> 44))) + (x2 & 0x7ffffffffff))
+	x6 := (x4 & 0xfffffffffff)
+	x7 := (x5 & 0x7ffffffffff)
+	x8 := (u64(fiat.u1((x5 >> 43))) + (x3 & 0x7ffffffffff))
+	out1[0] = x6
+	out1[1] = x7
+	out1[2] = x8
+}
+
+fe_add :: proc "contextless" (out1: ^Loose_Field_Element, arg1, arg2: ^Tight_Field_Element) {
+	x1 := (arg1[0] + arg2[0])
+	x2 := (arg1[1] + arg2[1])
+	x3 := (arg1[2] + arg2[2])
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+}
+
+fe_sub :: proc "contextless" (out1: ^Loose_Field_Element, arg1, arg2: ^Tight_Field_Element) {
+	x1 := ((0x1ffffffffff6 + arg1[0]) - arg2[0])
+	x2 := ((0xffffffffffe + arg1[1]) - arg2[1])
+	x3 := ((0xffffffffffe + arg1[2]) - arg2[2])
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+}
+
+fe_opp :: proc "contextless" (out1: ^Loose_Field_Element, arg1: ^Tight_Field_Element) {
+	x1 := (0x1ffffffffff6 - arg1[0])
+	x2 := (0xffffffffffe - arg1[1])
+	x3 := (0xffffffffffe - arg1[2])
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+}
+
+fe_cond_assign :: proc "contextless" (out1, arg1: ^Tight_Field_Element, arg2: bool) {
+	x1 := fiat.cmovznz_u64(fiat.u1(arg2), out1[0], arg1[0])
+	x2 := fiat.cmovznz_u64(fiat.u1(arg2), out1[1], arg1[1])
+	x3 := fiat.cmovznz_u64(fiat.u1(arg2), out1[2], arg1[2])
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+}
+
+fe_to_bytes :: proc "contextless" (out1: ^[32]byte, arg1: ^Tight_Field_Element) {
+	x1, x2 := _subborrowx_u44(0x0, arg1[0], 0xffffffffffb)
+	x3, x4 := _subborrowx_u43(x2, arg1[1], 0x7ffffffffff)
+	x5, x6 := _subborrowx_u43(x4, arg1[2], 0x7ffffffffff)
+	x7 := fiat.cmovznz_u64(x6, u64(0x0), 0xffffffffffffffff)
+	x8, x9 := _addcarryx_u44(0x0, x1, (x7 & 0xffffffffffb))
+	x10, x11 := _addcarryx_u43(x9, x3, (x7 & 0x7ffffffffff))
+	x12, _ := _addcarryx_u43(x11, x5, (x7 & 0x7ffffffffff))
+	x14 := (x12 << 7)
+	x15 := (x10 << 4)
+	x16 := (u8(x8) & 0xff)
+	x17 := (x8 >> 8)
+	x18 := (u8(x17) & 0xff)
+	x19 := (x17 >> 8)
+	x20 := (u8(x19) & 0xff)
+	x21 := (x19 >> 8)
+	x22 := (u8(x21) & 0xff)
+	x23 := (x21 >> 8)
+	x24 := (u8(x23) & 0xff)
+	x25 := u8((x23 >> 8))
+	x26 := (x15 + u64(x25))
+	x27 := (u8(x26) & 0xff)
+	x28 := (x26 >> 8)
+	x29 := (u8(x28) & 0xff)
+	x30 := (x28 >> 8)
+	x31 := (u8(x30) & 0xff)
+	x32 := (x30 >> 8)
+	x33 := (u8(x32) & 0xff)
+	x34 := (x32 >> 8)
+	x35 := (u8(x34) & 0xff)
+	x36 := u8((x34 >> 8))
+	x37 := (x14 + u64(x36))
+	x38 := (u8(x37) & 0xff)
+	x39 := (x37 >> 8)
+	x40 := (u8(x39) & 0xff)
+	x41 := (x39 >> 8)
+	x42 := (u8(x41) & 0xff)
+	x43 := (x41 >> 8)
+	x44 := (u8(x43) & 0xff)
+	x45 := (x43 >> 8)
+	x46 := (u8(x45) & 0xff)
+	x47 := (x45 >> 8)
+	x48 := (u8(x47) & 0xff)
+	x49 := u8((x47 >> 8))
+	out1[0] = x16
+	out1[1] = x18
+	out1[2] = x20
+	out1[3] = x22
+	out1[4] = x24
+	out1[5] = x27
+	out1[6] = x29
+	out1[7] = x31
+	out1[8] = x33
+	out1[9] = x35
+	out1[10] = x38
+	out1[11] = x40
+	out1[12] = x42
+	out1[13] = x44
+	out1[14] = x46
+	out1[15] = x48
+	out1[16] = x49
+}
+
+_fe_from_bytes :: proc "contextless" (out1: ^Tight_Field_Element, arg1: ^[32]byte) {
+	x1 := (u64(arg1[16]) << 41)
+	x2 := (u64(arg1[15]) << 33)
+	x3 := (u64(arg1[14]) << 25)
+	x4 := (u64(arg1[13]) << 17)
+	x5 := (u64(arg1[12]) << 9)
+	x6 := (u64(arg1[11]) * u64(0x2))
+	x7 := (u64(arg1[10]) << 36)
+	x8 := (u64(arg1[9]) << 28)
+	x9 := (u64(arg1[8]) << 20)
+	x10 := (u64(arg1[7]) << 12)
+	x11 := (u64(arg1[6]) << 4)
+	x12 := (u64(arg1[5]) << 40)
+	x13 := (u64(arg1[4]) << 32)
+	x14 := (u64(arg1[3]) << 24)
+	x15 := (u64(arg1[2]) << 16)
+	x16 := (u64(arg1[1]) << 8)
+	x17 := arg1[0]
+	x18 := (x16 + u64(x17))
+	x19 := (x15 + x18)
+	x20 := (x14 + x19)
+	x21 := (x13 + x20)
+	x22 := (x12 + x21)
+	x23 := (x22 & 0xfffffffffff)
+	x24 := u8((x22 >> 44))
+	x25 := (x11 + u64(x24))
+	x26 := (x10 + x25)
+	x27 := (x9 + x26)
+	x28 := (x8 + x27)
+	x29 := (x7 + x28)
+	x30 := (x29 & 0x7ffffffffff)
+	x31 := fiat.u1((x29 >> 43))
+	x32 := (x6 + u64(x31))
+	x33 := (x5 + x32)
+	x34 := (x4 + x33)
+	x35 := (x3 + x34)
+	x36 := (x2 + x35)
+	x37 := (x1 + x36)
+	out1[0] = x23
+	out1[1] = x30
+	out1[2] = x37
+}
+
+fe_relax :: proc "contextless" (out1: ^Loose_Field_Element, arg1: ^Tight_Field_Element) {
+	x1 := arg1[0]
+	x2 := arg1[1]
+	x3 := arg1[2]
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+}
+
+// The following routines were added by hand, and do not come from fiat-crypto.
+
+fe_zero :: proc "contextless" (out1: ^Tight_Field_Element) {
+	out1[0] = 0
+	out1[1] = 0
+	out1[2] = 0
+}
+
+fe_set :: #force_inline proc "contextless" (out1, arg1: ^Tight_Field_Element) {
+	x1 := arg1[0]
+	x2 := arg1[1]
+	x3 := arg1[2]
+	out1[0] = x1
+	out1[1] = x2
+	out1[2] = x3
+}
+
+fe_cond_swap :: proc "contextless" (out1, out2: ^Tight_Field_Element, arg1: bool) {
+	mask := -u64(arg1)
+	x := (out1[0] ~ out2[0]) & mask
+	x1, y1 := out1[0] ~ x, out2[0] ~ x
+	x = (out1[1] ~ out2[1]) & mask
+	x2, y2 := out1[1] ~ x, out2[1] ~ x
+	x = (out1[2] ~ out2[2]) & mask
+	x3, y3 := out1[2] ~ x, out2[2] ~ x
+	out1[0], out2[0] = x1, y1
+	out1[1], out2[1] = x2, y2
+	out1[2], out2[2] = x3, y3
+}
diff --git a/core/crypto/chacha20/chacha20.odin b/core/crypto/chacha20/chacha20.odin
new file mode 100644
index 000000000..f6f551692
--- /dev/null
+++ b/core/crypto/chacha20/chacha20.odin
@@ -0,0 +1,581 @@
+package chacha20
+
+import "core:crypto/util"
+import "core:math/bits"
+import "core:mem"
+
+KEY_SIZE :: 32
+NONCE_SIZE :: 12
+XNONCE_SIZE :: 24
+
+_MAX_CTR_IETF :: 0xffffffff
+
+_BLOCK_SIZE :: 64
+_STATE_SIZE_U32 :: 16
+_ROUNDS :: 20
+
+_SIGMA_0 : u32 : 0x61707865
+_SIGMA_1 : u32 : 0x3320646e
+_SIGMA_2 : u32 : 0x79622d32
+_SIGMA_3 : u32 : 0x6b206574
+
+Context :: struct {
+	_s: [_STATE_SIZE_U32]u32,
+
+	_buffer: [_BLOCK_SIZE]byte,
+	_off: int,
+
+	_is_ietf_flavor: bool,
+	_is_initialized: bool,
+}
+
+init :: proc (ctx: ^Context, key, nonce: []byte) {
+	if len(key) != KEY_SIZE {
+		panic("crypto/chacha20: invalid ChaCha20 key size")
+	}
+	if n_len := len(nonce); n_len != NONCE_SIZE && n_len != XNONCE_SIZE {
+		panic("crypto/chacha20: invalid (X)ChaCha20 nonce size")
+	}
+
+	k, n := key, nonce
+
+	// Derive the XChaCha20 subkey and sub-nonce via HChaCha20.
+	is_xchacha := len(nonce) == XNONCE_SIZE
+	if is_xchacha {
+		sub_key := ctx._buffer[:KEY_SIZE]
+		_hchacha20(sub_key, k, n)
+		k = sub_key
+		n = n[16:24]
+	}
+
+	ctx._s[0] = _SIGMA_0
+	ctx._s[1] = _SIGMA_1
+	ctx._s[2] = _SIGMA_2
+	ctx._s[3] = _SIGMA_3
+	ctx._s[4] = util.U32_LE(k[0:4])
+	ctx._s[5] = util.U32_LE(k[4:8])
+	ctx._s[6] = util.U32_LE(k[8:12])
+	ctx._s[7] = util.U32_LE(k[12:16])
+	ctx._s[8] = util.U32_LE(k[16:20])
+	ctx._s[9] = util.U32_LE(k[20:24])
+	ctx._s[10] = util.U32_LE(k[24:28])
+	ctx._s[11] = util.U32_LE(k[28:32])
+	ctx._s[12] = 0
+	if !is_xchacha {
+		ctx._s[13] = util.U32_LE(n[0:4])
+		ctx._s[14] = util.U32_LE(n[4:8])
+		ctx._s[15] = util.U32_LE(n[8:12])
+	} else {
+		ctx._s[13] = 0
+		ctx._s[14] = util.U32_LE(n[0:4])
+		ctx._s[15] = util.U32_LE(n[4:8])
+
+		// The sub-key is stored in the keystream buffer.  While
+		// this will be overwritten in most circumstances, explicitly
+		// clear it out early.
+		mem.zero_explicit(&ctx._buffer, KEY_SIZE)
+	}
+
+	ctx._off = _BLOCK_SIZE
+	ctx._is_ietf_flavor = !is_xchacha
+	ctx._is_initialized = true
+}
+
+seek :: proc (ctx: ^Context, block_nr: u64) {
+	assert(ctx._is_initialized)
+
+	if ctx._is_ietf_flavor {
+		if block_nr > _MAX_CTR_IETF {
+			panic("crypto/chacha20: attempted to seek past maximum counter")
+		}
+	} else {
+		ctx._s[13] = u32(block_nr >> 32)
+	}
+	ctx._s[12] = u32(block_nr)
+	ctx._off = _BLOCK_SIZE
+}
+
+xor_bytes :: proc (ctx: ^Context, dst, src: []byte) {
+	assert(ctx._is_initialized)
+
+	// TODO: Enforcing that dst and src alias exactly or not at all
+	// is a good idea, though odd aliasing should be extremely uncommon.
+
+	src, dst := src, dst
+	if dst_len := len(dst); dst_len < len(src) {
+		src = src[:dst_len]
+	}
+
+	for remaining := len(src); remaining > 0; {
+		// Process multiple blocks at once
+		if ctx._off == _BLOCK_SIZE {
+			if nr_blocks := remaining / _BLOCK_SIZE; nr_blocks > 0 {
+				direct_bytes := nr_blocks * _BLOCK_SIZE
+				_do_blocks(ctx, dst, src, nr_blocks)
+				remaining -= direct_bytes
+				if remaining == 0 {
+					return
+				}
+				dst = dst[direct_bytes:]
+				src = src[direct_bytes:]
+			}
+
+			// If there is a partial block, generate and buffer 1 block
+			// worth of keystream.
+			_do_blocks(ctx, ctx._buffer[:], nil, 1)
+			ctx._off = 0
+		}
+
+		// Process partial blocks from the buffered keystream.
+		to_xor := min(_BLOCK_SIZE - ctx._off, remaining)
+		buffered_keystream := ctx._buffer[ctx._off:]
+		for i := 0; i < to_xor; i = i + 1 {
+			dst[i] = buffered_keystream[i] ~ src[i]
+		}
+		ctx._off += to_xor
+		dst = dst[to_xor:]
+		src = src[to_xor:]
+		remaining -= to_xor
+	}
+}
+
+keystream_bytes :: proc (ctx: ^Context, dst: []byte) {
+	assert(ctx._is_initialized)
+
+	dst := dst
+	for remaining := len(dst); remaining > 0; {
+		// Process multiple blocks at once
+		if ctx._off == _BLOCK_SIZE {
+			if nr_blocks := remaining / _BLOCK_SIZE; nr_blocks > 0 {
+				direct_bytes := nr_blocks * _BLOCK_SIZE
+				_do_blocks(ctx, dst, nil, nr_blocks)
+				remaining -= direct_bytes
+				if remaining == 0 {
+					return
+				}
+				dst = dst[direct_bytes:]
+			}
+
+			// If there is a partial block, generate and buffer 1 block
+			// worth of keystream.
+			_do_blocks(ctx, ctx._buffer[:], nil, 1)
+			ctx._off = 0
+		}
+
+		// Process partial blocks from the buffered keystream.
+		to_copy := min(_BLOCK_SIZE - ctx._off, remaining)
+		buffered_keystream := ctx._buffer[ctx._off:]
+		copy(dst[:to_copy], buffered_keystream[:to_copy])
+		ctx._off += to_copy
+		dst = dst[to_copy:]
+		remaining -= to_copy
+	}
+}
+
+reset :: proc (ctx: ^Context) {
+	mem.zero_explicit(&ctx._s, size_of(ctx._s))
+	mem.zero_explicit(&ctx._buffer, size_of(ctx._buffer))
+
+	ctx._is_initialized = false
+}
+
+_do_blocks :: proc (ctx: ^Context, dst, src: []byte, nr_blocks: int) {
+	// Enforce the maximum consumed keystream per nonce.
+	//
+	// While all modern "standard" definitions of ChaCha20 use
+	// the IETF 32-bit counter, for XChaCha20 most common
+	// implementations allow for a 64-bit counter.
+	//
+	// Honestly, the answer here is "use a MRAE primitive", but
+	// go with common practice in the case of XChaCha20.
+	if ctx._is_ietf_flavor {
+		if u64(ctx._s[12]) + u64(nr_blocks) > 0xffffffff {
+			panic("crypto/chacha20: maximum ChaCha20 keystream per nonce reached")
+		}
+	} else {
+		ctr := (u64(ctx._s[13]) << 32) | u64(ctx._s[12])
+		if _, carry := bits.add_u64(ctr, u64(nr_blocks), 0); carry != 0 {
+			panic("crypto/chacha20: maximum XChaCha20 keystream per nonce reached")
+		}
+	}
+
+	dst, src := dst, src
+	x := &ctx._s
+	for n := 0; n < nr_blocks; n = n + 1 {
+		x0, x1, x2, x3 := _SIGMA_0, _SIGMA_1, _SIGMA_2, _SIGMA_3
+		x4, x5, x6, x7, x8, x9, x10, x11, x12, x13, x14, x15 := x[4], x[5], x[6], x[7], x[8], x[9], x[10], x[11], x[12], x[13], x[14], x[15]
+
+		for i := _ROUNDS; i > 0; i = i - 2 {
+			// Even when forcing inlining manually inlining all of
+			// these is decently faster.
+
+			// quarterround(x, 0, 4, 8, 12)
+			x0 += x4
+			x12 ~= x0
+			x12 = util.ROTL32(x12, 16)
+			x8 += x12
+			x4 ~= x8
+			x4 = util.ROTL32(x4, 12)
+			x0 += x4
+			x12 ~= x0
+			x12 = util.ROTL32(x12, 8)
+			x8 += x12
+			x4 ~= x8
+			x4 = util.ROTL32(x4, 7)
+
+			// quarterround(x, 1, 5, 9, 13)
+			x1 += x5
+			x13 ~= x1
+			x13 = util.ROTL32(x13, 16)
+			x9 += x13
+			x5 ~= x9
+			x5 = util.ROTL32(x5, 12)
+			x1 += x5
+			x13 ~= x1
+			x13 = util.ROTL32(x13, 8)
+			x9 += x13
+			x5 ~= x9
+			x5 = util.ROTL32(x5, 7)
+
+			// quarterround(x, 2, 6, 10, 14)
+			x2 += x6
+			x14 ~= x2
+			x14 = util.ROTL32(x14, 16)
+			x10 += x14
+			x6 ~= x10
+			x6 = util.ROTL32(x6, 12)
+			x2 += x6
+			x14 ~= x2
+			x14 = util.ROTL32(x14, 8)
+			x10 += x14
+			x6 ~= x10
+			x6 = util.ROTL32(x6, 7)
+
+			// quarterround(x, 3, 7, 11, 15)
+			x3 += x7
+			x15 ~= x3
+			x15 = util.ROTL32(x15, 16)
+			x11 += x15
+			x7 ~= x11
+			x7 = util.ROTL32(x7, 12)
+			x3 += x7
+			x15 ~= x3
+			x15 = util.ROTL32(x15, 8)
+			x11 += x15
+			x7 ~= x11
+			x7 = util.ROTL32(x7, 7)
+
+			// quarterround(x, 0, 5, 10, 15)
+			x0 += x5
+			x15 ~= x0
+			x15 = util.ROTL32(x15, 16)
+			x10 += x15
+			x5 ~= x10
+			x5 = util.ROTL32(x5, 12)
+			x0 += x5
+			x15 ~= x0
+			x15 = util.ROTL32(x15, 8)
+			x10 += x15
+			x5 ~= x10
+			x5 = util.ROTL32(x5, 7)
+
+			// quarterround(x, 1, 6, 11, 12)
+			x1 += x6
+			x12 ~= x1
+			x12 = util.ROTL32(x12, 16)
+			x11 += x12
+			x6 ~= x11
+			x6 = util.ROTL32(x6, 12)
+			x1 += x6
+			x12 ~= x1
+			x12 = util.ROTL32(x12, 8)
+			x11 += x12
+			x6 ~= x11
+			x6 = util.ROTL32(x6, 7)
+
+			// quarterround(x, 2, 7, 8, 13)
+			x2 += x7
+			x13 ~= x2
+			x13 = util.ROTL32(x13, 16)
+			x8 += x13
+			x7 ~= x8
+			x7 = util.ROTL32(x7, 12)
+			x2 += x7
+			x13 ~= x2
+			x13 = util.ROTL32(x13, 8)
+			x8 += x13
+			x7 ~= x8
+			x7 = util.ROTL32(x7, 7)
+
+			// quarterround(x, 3, 4, 9, 14)
+			x3 += x4
+			x14 ~= x3
+			x14 = util.ROTL32(x14, 16)
+			x9 += x14
+			x4 ~= x9
+			x4 = util.ROTL32(x4, 12)
+			x3 += x4
+			x14 ~= x3
+			x14 = util.ROTL32(x14, 8)
+			x9 += x14
+			x4 ~= x9
+			x4 = util.ROTL32(x4, 7)
+		}
+
+		x0 += _SIGMA_0
+		x1 += _SIGMA_1
+		x2 += _SIGMA_2
+		x3 += _SIGMA_3
+		x4 += x[4]
+		x5 += x[5]
+		x6 += x[6]
+		x7 += x[7]
+		x8 += x[8]
+		x9 += x[9]
+		x10 += x[10]
+		x11 += x[11]
+		x12 += x[12]
+		x13 += x[13]
+		x14 += x[14]
+		x15 += x[15]
+
+		// While the "correct" answer to getting more performance out of
+		// this is "use vector operations", support for that is currently
+		// a work in progress/to be designed.
+		//
+		// Until dedicated assembly can be written leverage the fact that
+		// the callers of this routine ensure that src/dst are valid.
+
+		when ODIN_ARCH == "386" || ODIN_ARCH == "amd64" {
+			// util.PUT_U32_LE/util.U32_LE are not required on little-endian
+			// systems that also happen to not be strict about aligned
+			// memory access.
+
+			dst_p := transmute(^[16]u32)(&dst[0])
+			if src != nil {
+				src_p := transmute(^[16]u32)(&src[0])
+				dst_p[0] = src_p[0] ~ x0
+				dst_p[1] = src_p[1] ~ x1
+				dst_p[2] = src_p[2] ~ x2
+				dst_p[3] = src_p[3] ~ x3
+				dst_p[4] = src_p[4] ~ x4
+				dst_p[5] = src_p[5] ~ x5
+				dst_p[6] = src_p[6] ~ x6
+				dst_p[7] = src_p[7] ~ x7
+				dst_p[8] = src_p[8] ~ x8
+				dst_p[9] = src_p[9] ~ x9
+				dst_p[10] = src_p[10] ~ x10
+				dst_p[11] = src_p[11] ~ x11
+				dst_p[12] = src_p[12] ~ x12
+				dst_p[13] = src_p[13] ~ x13
+				dst_p[14] = src_p[14] ~ x14
+				dst_p[15] = src_p[15] ~ x15
+				src = src[_BLOCK_SIZE:]
+			} else {
+				dst_p[0] = x0
+				dst_p[1] = x1
+				dst_p[2] = x2
+				dst_p[3] = x3
+				dst_p[4] = x4
+				dst_p[5] = x5
+				dst_p[6] = x6
+				dst_p[7] = x7
+				dst_p[8] = x8
+				dst_p[9] = x9
+				dst_p[10] = x10
+				dst_p[11] = x11
+				dst_p[12] = x12
+				dst_p[13] = x13
+				dst_p[14] = x14
+				dst_p[15] = x15
+			}
+			dst = dst[_BLOCK_SIZE:]
+		} else {
+			#no_bounds_check {
+				if src != nil {
+					util.PUT_U32_LE(dst[0:4], util.U32_LE(src[0:4]) ~ x0)
+					util.PUT_U32_LE(dst[4:8], util.U32_LE(src[4:8]) ~ x1)
+					util.PUT_U32_LE(dst[8:12], util.U32_LE(src[8:12]) ~ x2)
+					util.PUT_U32_LE(dst[12:16], util.U32_LE(src[12:16]) ~ x3)
+					util.PUT_U32_LE(dst[16:20], util.U32_LE(src[16:20]) ~ x4)
+					util.PUT_U32_LE(dst[20:24], util.U32_LE(src[20:24]) ~ x5)
+					util.PUT_U32_LE(dst[24:28], util.U32_LE(src[24:28]) ~ x6)
+					util.PUT_U32_LE(dst[28:32], util.U32_LE(src[28:32]) ~ x7)
+					util.PUT_U32_LE(dst[32:36], util.U32_LE(src[32:36]) ~ x8)
+					util.PUT_U32_LE(dst[36:40], util.U32_LE(src[36:40]) ~ x9)
+					util.PUT_U32_LE(dst[40:44], util.U32_LE(src[40:44]) ~ x10)
+					util.PUT_U32_LE(dst[44:48], util.U32_LE(src[44:48]) ~ x11)
+					util.PUT_U32_LE(dst[48:52], util.U32_LE(src[48:52]) ~ x12)
+					util.PUT_U32_LE(dst[52:56], util.U32_LE(src[52:56]) ~ x13)
+					util.PUT_U32_LE(dst[56:60], util.U32_LE(src[56:60]) ~ x14)
+					util.PUT_U32_LE(dst[60:64], util.U32_LE(src[60:64]) ~ x15)
+					src = src[_BLOCK_SIZE:]
+				} else {
+					util.PUT_U32_LE(dst[0:4], x0)
+					util.PUT_U32_LE(dst[4:8], x1)
+					util.PUT_U32_LE(dst[8:12], x2)
+					util.PUT_U32_LE(dst[12:16], x3)
+					util.PUT_U32_LE(dst[16:20], x4)
+					util.PUT_U32_LE(dst[20:24], x5)
+					util.PUT_U32_LE(dst[24:28], x6)
+					util.PUT_U32_LE(dst[28:32], x7)
+					util.PUT_U32_LE(dst[32:36], x8)
+					util.PUT_U32_LE(dst[36:40], x9)
+					util.PUT_U32_LE(dst[40:44], x10)
+					util.PUT_U32_LE(dst[44:48], x11)
+					util.PUT_U32_LE(dst[48:52], x12)
+					util.PUT_U32_LE(dst[52:56], x13)
+					util.PUT_U32_LE(dst[56:60], x14)
+					util.PUT_U32_LE(dst[60:64], x15)
+				}
+				dst = dst[_BLOCK_SIZE:]
+			}
+		}
+
+		// Increment the counter.  Overflow checking is done upon
+		// entry into the routine, so a 64-bit increment safely
+		// covers both cases.
+		new_ctr := ((u64(ctx._s[13]) << 32) | u64(ctx._s[12])) + 1
+		x[12] = u32(new_ctr)
+		x[13] = u32(new_ctr >> 32)
+	}
+}
+
+_hchacha20 :: proc (dst, key, nonce: []byte) {
+	x0, x1, x2, x3 := _SIGMA_0, _SIGMA_1, _SIGMA_2, _SIGMA_3
+	x4 := util.U32_LE(key[0:4])
+	x5 := util.U32_LE(key[4:8])
+	x6 := util.U32_LE(key[8:12])
+	x7 := util.U32_LE(key[12:16])
+	x8 := util.U32_LE(key[16:20])
+	x9 := util.U32_LE(key[20:24])
+	x10 := util.U32_LE(key[24:28])
+	x11 := util.U32_LE(key[28:32])
+	x12 := util.U32_LE(nonce[0:4])
+	x13 := util.U32_LE(nonce[4:8])
+	x14 := util.U32_LE(nonce[8:12])
+	x15 := util.U32_LE(nonce[12:16])
+
+	for i := _ROUNDS; i > 0; i = i - 2 {
+		// quarterround(x, 0, 4, 8, 12)
+		x0 += x4
+		x12 ~= x0
+		x12 = util.ROTL32(x12, 16)
+		x8 += x12
+		x4 ~= x8
+		x4 = util.ROTL32(x4, 12)
+		x0 += x4
+		x12 ~= x0
+		x12 = util.ROTL32(x12, 8)
+		x8 += x12
+		x4 ~= x8
+		x4 = util.ROTL32(x4, 7)
+
+		// quarterround(x, 1, 5, 9, 13)
+		x1 += x5
+		x13 ~= x1
+		x13 = util.ROTL32(x13, 16)
+		x9 += x13
+		x5 ~= x9
+		x5 = util.ROTL32(x5, 12)
+		x1 += x5
+		x13 ~= x1
+		x13 = util.ROTL32(x13, 8)
+		x9 += x13
+		x5 ~= x9
+		x5 = util.ROTL32(x5, 7)
+
+		// quarterround(x, 2, 6, 10, 14)
+		x2 += x6
+		x14 ~= x2
+		x14 = util.ROTL32(x14, 16)
+		x10 += x14
+		x6 ~= x10
+		x6 = util.ROTL32(x6, 12)
+		x2 += x6
+		x14 ~= x2
+		x14 = util.ROTL32(x14, 8)
+		x10 += x14
+		x6 ~= x10
+		x6 = util.ROTL32(x6, 7)
+
+		// quarterround(x, 3, 7, 11, 15)
+		x3 += x7
+		x15 ~= x3
+		x15 = util.ROTL32(x15, 16)
+		x11 += x15
+		x7 ~= x11
+		x7 = util.ROTL32(x7, 12)
+		x3 += x7
+		x15 ~= x3
+		x15 = util.ROTL32(x15, 8)
+		x11 += x15
+		x7 ~= x11
+		x7 = util.ROTL32(x7, 7)
+
+		// quarterround(x, 0, 5, 10, 15)
+		x0 += x5
+		x15 ~= x0
+		x15 = util.ROTL32(x15, 16)
+		x10 += x15
+		x5 ~= x10
+		x5 = util.ROTL32(x5, 12)
+		x0 += x5
+		x15 ~= x0
+		x15 = util.ROTL32(x15, 8)
+		x10 += x15
+		x5 ~= x10
+		x5 = util.ROTL32(x5, 7)
+
+		// quarterround(x, 1, 6, 11, 12)
+		x1 += x6
+		x12 ~= x1
+		x12 = util.ROTL32(x12, 16)
+		x11 += x12
+		x6 ~= x11
+		x6 = util.ROTL32(x6, 12)
+		x1 += x6
+		x12 ~= x1
+		x12 = util.ROTL32(x12, 8)
+		x11 += x12
+		x6 ~= x11
+		x6 = util.ROTL32(x6, 7)
+
+		// quarterround(x, 2, 7, 8, 13)
+		x2 += x7
+		x13 ~= x2
+		x13 = util.ROTL32(x13, 16)
+		x8 += x13
+		x7 ~= x8
+		x7 = util.ROTL32(x7, 12)
+		x2 += x7
+		x13 ~= x2
+		x13 = util.ROTL32(x13, 8)
+		x8 += x13
+		x7 ~= x8
+		x7 = util.ROTL32(x7, 7)
+
+		// quarterround(x, 3, 4, 9, 14)
+		x3 += x4
+		x14 ~= x3
+		x14 = util.ROTL32(x14, 16)
+		x9 += x14
+		x4 ~= x9
+		x4 = util.ROTL32(x4, 12)
+		x3 += x4
+		x14 ~= x3
+		x14 = util.ROTL32(x14, 8)
+		x9 += x14
+		x4 ~= x9
+		x4 = util.ROTL32(x4, 7)
+	}
+
+	util.PUT_U32_LE(dst[0:4], x0)
+	util.PUT_U32_LE(dst[4:8], x1)
+	util.PUT_U32_LE(dst[8:12], x2)
+	util.PUT_U32_LE(dst[12:16], x3)
+	util.PUT_U32_LE(dst[16:20], x12)
+	util.PUT_U32_LE(dst[20:24], x13)
+	util.PUT_U32_LE(dst[24:28], x14)
+	util.PUT_U32_LE(dst[28:32], x15)
+}
diff --git a/core/crypto/chacha20poly1305/chacha20poly1305.odin b/core/crypto/chacha20poly1305/chacha20poly1305.odin
new file mode 100644
index 000000000..67d89df56
--- /dev/null
+++ b/core/crypto/chacha20poly1305/chacha20poly1305.odin
@@ -0,0 +1,146 @@
+package chacha20poly1305
+
+import "core:crypto"
+import "core:crypto/chacha20"
+import "core:crypto/poly1305"
+import "core:crypto/util"
+import "core:mem"
+
+KEY_SIZE :: chacha20.KEY_SIZE
+NONCE_SIZE :: chacha20.NONCE_SIZE
+TAG_SIZE :: poly1305.TAG_SIZE
+
+_P_MAX :: 64 * 0xffffffff // 64 * (2^32-1)
+
+_validate_common_slice_sizes :: proc (tag, key, nonce, aad, text: []byte) {
+	if len(tag) != TAG_SIZE {
+		panic("crypto/chacha20poly1305: invalid destination tag size")
+	}
+	if len(key) != KEY_SIZE {
+		panic("crypto/chacha20poly1305: invalid key size")
+	}
+	if len(nonce) != NONCE_SIZE {
+		panic("crypto/chacha20poly1305: invalid nonce size")
+	}
+
+	#assert(size_of(int) == 8 || size_of(int) <= 4)
+	when size_of(int) == 8 {
+		// A_MAX = 2^64 - 1 due to the length field limit.
+		// P_MAX = 64 * (2^32 - 1) due to the IETF ChaCha20 counter limit.
+		//
+		// A_MAX is limited by size_of(int), so there is no need to
+		// enforce it. P_MAX only needs to be checked on 64-bit targets,
+		// for reasons that should be obvious.
+		if text_len := len(text); text_len > _P_MAX {
+			panic("crypto/chacha20poly1305: oversized src data")
+		}
+	}
+}
+
+_PAD: [16]byte
+_update_mac_pad16 :: #force_inline proc (ctx: ^poly1305.Context, x_len: int) {
+	if pad_len := 16 - (x_len & (16-1)); pad_len != 16 {
+		poly1305.update(ctx, _PAD[:pad_len])
+	}
+}
+
+encrypt :: proc (ciphertext, tag, key, nonce, aad, plaintext: []byte) {
+	_validate_common_slice_sizes(tag, key, nonce, aad, plaintext)
+	if len(ciphertext) != len(plaintext) {
+		panic("crypto/chacha20poly1305: invalid destination ciphertext size")
+	}
+
+	stream_ctx: chacha20.Context = ---
+	chacha20.init(&stream_ctx, key, nonce)
+
+	// otk = poly1305_key_gen(key, nonce)
+	otk: [poly1305.KEY_SIZE]byte = ---
+	chacha20.keystream_bytes(&stream_ctx, otk[:])
+	mac_ctx: poly1305.Context = ---
+	poly1305.init(&mac_ctx, otk[:])
+	mem.zero_explicit(&otk, size_of(otk))
+
+	aad_len, ciphertext_len := len(aad), len(ciphertext)
+
+	// There is nothing preventing aad and ciphertext from overlapping
+	// so auth the AAD before encrypting (slightly different from the
+	// RFC, since the RFC encrypts into a new buffer).
+	//
+	// mac_data = aad | pad16(aad)
+	poly1305.update(&mac_ctx, aad)
+	_update_mac_pad16(&mac_ctx, aad_len)
+
+	// ciphertext = chacha20_encrypt(key, 1, nonce, plaintext)
+	chacha20.seek(&stream_ctx, 1)
+	chacha20.xor_bytes(&stream_ctx, ciphertext, plaintext)
+	chacha20.reset(&stream_ctx) // Don't need the stream context anymore.
+
+	// mac_data |= ciphertext | pad16(ciphertext)
+	poly1305.update(&mac_ctx, ciphertext)
+	_update_mac_pad16(&mac_ctx, ciphertext_len)
+
+	// mac_data |= num_to_8_le_bytes(aad.length)
+	// mac_data |= num_to_8_le_bytes(ciphertext.length)
+	l_buf := otk[0:16] // Reuse the scratch buffer.
+	util.PUT_U64_LE(l_buf[0:8], u64(aad_len))
+	util.PUT_U64_LE(l_buf[8:16], u64(ciphertext_len))
+	poly1305.update(&mac_ctx, l_buf)
+
+	// tag = poly1305_mac(mac_data, otk)
+	poly1305.final(&mac_ctx, tag) // Implicitly sanitizes context.
+}
+
+decrypt :: proc (plaintext, tag, key, nonce, aad, ciphertext: []byte) -> bool {
+	_validate_common_slice_sizes(tag, key, nonce, aad, ciphertext)
+	if len(ciphertext) != len(plaintext) {
+		panic("crypto/chacha20poly1305: invalid destination plaintext size")
+	}
+
+	// Note: Unlike encrypt, this can fail early, so use defer for
+	// sanitization rather than assuming control flow reaches certain
+	// points where needed.
+
+	stream_ctx: chacha20.Context = ---
+	chacha20.init(&stream_ctx, key, nonce)
+
+	// otk = poly1305_key_gen(key, nonce)
+	otk: [poly1305.KEY_SIZE]byte = ---
+	chacha20.keystream_bytes(&stream_ctx, otk[:])
+	defer chacha20.reset(&stream_ctx)
+
+	mac_ctx: poly1305.Context = ---
+	poly1305.init(&mac_ctx, otk[:])
+	defer mem.zero_explicit(&otk, size_of(otk))
+
+	aad_len, ciphertext_len := len(aad), len(ciphertext)
+
+	// mac_data = aad | pad16(aad)
+	// mac_data |= ciphertext | pad16(ciphertext)
+	// mac_data |= num_to_8_le_bytes(aad.length)
+	// mac_data |= num_to_8_le_bytes(ciphertext.length)
+	poly1305.update(&mac_ctx, aad)
+	_update_mac_pad16(&mac_ctx, aad_len)
+	poly1305.update(&mac_ctx, ciphertext)
+	_update_mac_pad16(&mac_ctx, ciphertext_len)
+	l_buf := otk[0:16] // Reuse the scratch buffer.
+	util.PUT_U64_LE(l_buf[0:8], u64(aad_len))
+	util.PUT_U64_LE(l_buf[8:16], u64(ciphertext_len))
+	poly1305.update(&mac_ctx, l_buf)
+
+	// tag = poly1305_mac(mac_data, otk)
+	derived_tag := otk[0:poly1305.TAG_SIZE] // Reuse the scratch buffer again.
+	poly1305.final(&mac_ctx, derived_tag) // Implicitly sanitizes context.
+
+	// Validate the tag in constant time.
+	if crypto.compare_constant_time(tag, derived_tag) != 1 {
+		// Zero out the plaintext, as a defense in depth measure.
+		mem.zero_explicit(raw_data(plaintext), ciphertext_len)
+		return false
+	}
+
+	// plaintext = chacha20_decrypt(key, 1, nonce, ciphertext)
+	chacha20.seek(&stream_ctx, 1)
+	chacha20.xor_bytes(&stream_ctx, plaintext, ciphertext)
+
+	return true
+}
diff --git a/core/crypto/crypto.odin b/core/crypto/crypto.odin
new file mode 100644
index 000000000..35e88c5ed
--- /dev/null
+++ b/core/crypto/crypto.odin
@@ -0,0 +1,52 @@
+package crypto
+
+import "core:mem"
+
+// compare_constant_time returns 1 iff a and b are equal, 0 otherwise.
+//
+// The execution time of this routine is constant regardless of the contents
+// of the slices being compared, as long as the length of the slices is equal.
+// If the length of the two slices is different, it will early-return 0.
+compare_constant_time :: proc "contextless" (a, b: []byte) -> int {
+	// If the length of the slices is different, early return.
+	//
+	// This leaks the fact that the slices have a different length,
+	// but the routine is primarily intended for comparing things
+	// like MACS and password digests.
+	n := len(a)
+	if n != len(b) {
+		return 0
+	}
+
+	return compare_byte_ptrs_constant_time(raw_data(a), raw_data(b), n)
+}
+
+// compare_byte_ptrs_constant_time returns 1 iff the bytes pointed to by
+// a and b are equal, 0 otherwise.
+//
+// The execution time of this routine is constant regardless of the
+// contents of the memory being compared.
+compare_byte_ptrs_constant_time :: proc "contextless" (a, b: ^byte, n: int) -> int {
+	x := mem.slice_ptr(a, n)
+	y := mem.slice_ptr(b, n)
+
+	v: byte
+	for i in 0..<n {
+		v |= x[i] ~ y[i]
+	}
+
+	// After the loop, v == 0 iff a == b.  The subtraction will underflow
+	// iff v == 0, setting the sign-bit, which gets returned.
+	return int((u32(v)-1) >> 31)
+}
+
+// rand_bytes fills the dst buffer with cryptographic entropy taken from
+// the system entropy source.  This routine will block if the system entropy
+// source is not ready yet.  All system entropy source failures are treated
+// as catastrophic, resulting in a panic.
+rand_bytes :: proc (dst: []byte) {
+	// zero-fill the buffer first
+	mem.zero_explicit(raw_data(dst), len(dst))
+
+	_rand_bytes(dst)
+}
diff --git a/core/crypto/poly1305/poly1305.odin b/core/crypto/poly1305/poly1305.odin
new file mode 100644
index 000000000..8986be879
--- /dev/null
+++ b/core/crypto/poly1305/poly1305.odin
@@ -0,0 +1,163 @@
+package poly1305
+
+import "core:crypto"
+import "core:crypto/util"
+import field "core:crypto/_fiat/field_poly1305"
+import "core:mem"
+
+KEY_SIZE :: 32
+TAG_SIZE :: 16
+
+_BLOCK_SIZE :: 16
+
+sum :: proc (dst, msg, key: []byte) {
+	ctx: Context = ---
+
+	init(&ctx, key)
+	update(&ctx, msg)
+	final(&ctx, dst)
+}
+
+verify :: proc (tag, msg, key: []byte) -> bool {
+	ctx: Context = ---
+	derived_tag: [16]byte = ---
+
+	if len(tag) != TAG_SIZE {
+		panic("crypto/poly1305: invalid tag size")
+	}
+
+	init(&ctx, key)
+	update(&ctx, msg)
+	final(&ctx, derived_tag[:])
+
+	return crypto.compare_constant_time(derived_tag[:], tag) == 1
+}
+
+Context :: struct {
+	_r: field.Tight_Field_Element,
+	_a: field.Tight_Field_Element,
+	_s: field.Tight_Field_Element,
+
+	_buffer: [_BLOCK_SIZE]byte,
+	_leftover: int,
+
+	_is_initialized: bool,
+}
+
+init :: proc (ctx: ^Context, key: []byte) {
+	if len(key) != KEY_SIZE {
+		panic("crypto/poly1305: invalid key size")
+	}
+
+	// r = le_bytes_to_num(key[0..15])
+	// r = clamp(r) (r &= 0xffffffc0ffffffc0ffffffc0fffffff)
+	tmp_lo := util.U64_LE(key[0:8]) & 0x0ffffffc0fffffff
+	tmp_hi := util.U64_LE(key[8:16]) & 0xffffffc0ffffffc
+	field.fe_from_u64s(&ctx._r, tmp_lo, tmp_hi)
+
+	// s = le_bytes_to_num(key[16..31])
+	field.fe_from_bytes(&ctx._s, key[16:32], 0)
+
+	// a = 0
+	field.fe_zero(&ctx._a)
+
+	// No leftover in buffer
+	ctx._leftover = 0
+
+	ctx._is_initialized = true
+}
+
+update :: proc (ctx: ^Context, data: []byte) {
+	assert(ctx._is_initialized)
+
+	msg := data
+	msg_len := len(data)
+
+	// Handle leftover
+	if ctx._leftover > 0 {
+		want := min(_BLOCK_SIZE - ctx._leftover, msg_len)
+		copy_slice(ctx._buffer[ctx._leftover:], msg[:want])
+		msg_len = msg_len - want
+		msg = msg[want:]
+		ctx._leftover = ctx._leftover + want
+		if ctx._leftover < _BLOCK_SIZE {
+			return
+		}
+		_blocks(ctx, ctx._buffer[:])
+		ctx._leftover = 0
+	}
+
+	// Process full blocks
+	if msg_len >= _BLOCK_SIZE {
+		want := msg_len & (~int(_BLOCK_SIZE - 1))
+		_blocks(ctx, msg[:want])
+		msg = msg[want:]
+		msg_len = msg_len - want
+	}
+
+	// Store leftover
+	if msg_len > 0 {
+		// TODO: While -donna does it this way, I'm fairly sure that
+		// `ctx._leftover == 0` is an invariant at this point.
+		copy(ctx._buffer[ctx._leftover:], msg)
+		ctx._leftover = ctx._leftover + msg_len
+	}
+}
+
+final :: proc (ctx: ^Context, dst: []byte) {
+	assert(ctx._is_initialized)
+
+	if len(dst) != TAG_SIZE {
+		panic("poly1305: invalid destination tag size")
+	}
+
+	// Process remaining block
+	if ctx._leftover > 0 {
+		ctx._buffer[ctx._leftover] = 1
+		for i := ctx._leftover + 1; i < _BLOCK_SIZE; i = i + 1 {
+			ctx._buffer[i] = 0
+		}
+		_blocks(ctx, ctx._buffer[:], true)
+	}
+
+	// a += s
+	field.fe_add(field.fe_relax_cast(&ctx._a), &ctx._a, &ctx._s) // _a unreduced
+	field.fe_carry(&ctx._a, field.fe_relax_cast(&ctx._a)) // _a reduced
+
+	// return num_to_16_le_bytes(a)
+	tmp: [32]byte = ---
+	field.fe_to_bytes(&tmp, &ctx._a)
+	copy_slice(dst, tmp[0:16])
+
+	reset(ctx)
+}
+
+reset :: proc (ctx: ^Context) {
+	mem.zero_explicit(&ctx._r, size_of(ctx._r))
+	mem.zero_explicit(&ctx._a, size_of(ctx._a))
+	mem.zero_explicit(&ctx._s, size_of(ctx._s))
+	mem.zero_explicit(&ctx._buffer, size_of(ctx._buffer))
+
+	ctx._is_initialized = false
+}
+
+_blocks :: proc (ctx: ^Context, msg: []byte, final := false) {
+	n: field.Tight_Field_Element = ---
+	final_byte := byte(!final)
+
+	data := msg
+	data_len := len(data)
+	for data_len >= _BLOCK_SIZE {
+		// n = le_bytes_to_num(msg[((i-1)*16)..*i*16] | [0x01])
+		field.fe_from_bytes(&n, data[:_BLOCK_SIZE], final_byte, false)
+
+		// a += n
+		field.fe_add(field.fe_relax_cast(&ctx._a), &ctx._a, &n) // _a unreduced
+
+		// a = (r * a) % p
+		field.fe_carry_mul(&ctx._a, field.fe_relax_cast(&ctx._a), field.fe_relax_cast(&ctx._r)) // _a reduced
+
+		data = data[_BLOCK_SIZE:]
+		data_len = data_len - _BLOCK_SIZE
+	}
+}
diff --git a/core/crypto/rand_generic.odin b/core/crypto/rand_generic.odin
new file mode 100644
index 000000000..98890b5b1
--- /dev/null
+++ b/core/crypto/rand_generic.odin
@@ -0,0 +1,7 @@
+package crypto
+
+when ODIN_OS != "linux" {
+	_rand_bytes :: proc (dst: []byte) {
+		unimplemented("crypto: rand_bytes not supported on this OS")
+	}
+}
diff --git a/core/crypto/rand_linux.odin b/core/crypto/rand_linux.odin
new file mode 100644
index 000000000..4d1183757
--- /dev/null
+++ b/core/crypto/rand_linux.odin
@@ -0,0 +1,37 @@
+package crypto
+
+import "core:fmt"
+import "core:os"
+import "core:sys/unix"
+
+_MAX_PER_CALL_BYTES :: 33554431 // 2^25 - 1
+
+_rand_bytes :: proc (dst: []byte) {
+	dst := dst
+	l := len(dst)
+
+	for l > 0 {
+		to_read := min(l, _MAX_PER_CALL_BYTES)
+		ret := unix.sys_getrandom(raw_data(dst), to_read, 0)
+		if ret < 0 {
+			switch os.Errno(-ret) {
+			case os.EINTR:
+				// Call interupted by a signal handler, just retry the
+				// request.
+				continue
+			case os.ENOSYS:
+				// The kernel is apparently prehistoric (< 3.17 circa 2014)
+				// and does not support getrandom.
+				panic("crypto: getrandom not available in kernel")
+			case:
+				// All other failures are things that should NEVER happen
+				// unless the kernel interface changes (ie: the Linux
+				// developers break userland).
+				panic(fmt.tprintf("crypto: getrandom failed: %d", ret))
+			}
+		}
+
+		l -= ret
+		dst = dst[ret:]
+	}
+}
diff --git a/core/crypto/x25519/x25519.odin b/core/crypto/x25519/x25519.odin
new file mode 100644
index 000000000..dfc8daa47
--- /dev/null
+++ b/core/crypto/x25519/x25519.odin
@@ -0,0 +1,126 @@
+package x25519
+
+import field "core:crypto/_fiat/field_curve25519"
+import "core:mem"
+
+SCALAR_SIZE :: 32
+POINT_SIZE :: 32
+
+_BASE_POINT: [32]byte = {9, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}
+
+_scalar_bit :: #force_inline proc "contextless" (s: ^[32]byte, i: int) -> u8 {
+	if i < 0 {
+		return 0
+	}
+	return (s[i>>3] >> uint(i&7)) & 1
+}
+
+_scalarmult :: proc (out, scalar, point: ^[32]byte) {
+	// Montgomery pseduo-multiplication taken from Monocypher.
+
+	// computes the scalar product
+	x1: field.Tight_Field_Element = ---
+	field.fe_from_bytes(&x1, point)
+
+	// computes the actual scalar product (the result is in x2 and z2)
+	x2, x3, z2, z3: field.Tight_Field_Element =  ---, ---, ---, ---
+	t0, t1: field.Loose_Field_Element = ---, ---
+
+	// Montgomery ladder
+	// In projective coordinates, to avoid divisions: x = X / Z
+	// We don't care about the y coordinate, it's only 1 bit of information
+	field.fe_one(&x2) // "zero" point
+	field.fe_zero(&z2)
+	field.fe_set(&x3, &x1) // "one" point
+	field.fe_one(&z3)
+
+	swap: int
+	for pos := 255-1; pos >= 0; pos = pos - 1 	{
+		// constant time conditional swap before ladder step
+		b := int(_scalar_bit(scalar, pos))
+		swap ~= b // xor trick avoids swapping at the end of the loop
+		field.fe_cond_swap(&x2, &x3, swap)
+		field.fe_cond_swap(&z2, &z3, swap)
+		swap = b // anticipates one last swap after the loop
+
+		// Montgomery ladder step: replaces (P2, P3) by (P2*2, P2+P3)
+		// with differential addition
+		//
+		// Note: This deliberately omits reductions after add/sub operations
+		// if the result is only ever used as the input to a mul/square since
+		// the implementations of those can deal with non-reduced inputs.
+		//
+		// fe_tighten_cast is only used to store a fully reduced
+		// output in a Loose_Field_Element, or to provide such a
+		// Loose_Field_Element as a Tight_Field_Element argument.
+		field.fe_sub(&t0, &x3, &z3)
+		field.fe_sub(&t1, &x2, &z2)
+		field.fe_add(field.fe_relax_cast(&x2), &x2, &z2) // x2 - unreduced
+		field.fe_add(field.fe_relax_cast(&z2), &x3, &z3) // z2 - unreduced
+		field.fe_carry_mul(&z3, &t0, field.fe_relax_cast(&x2))
+		field.fe_carry_mul(&z2, field.fe_relax_cast(&z2), &t1) // z2 - reduced
+		field.fe_carry_square(field.fe_tighten_cast(&t0), &t1) // t0 - reduced
+		field.fe_carry_square(field.fe_tighten_cast(&t1), field.fe_relax_cast(&x2)) // t1 - reduced
+		field.fe_add(field.fe_relax_cast(&x3), &z3, &z2) // x3 - unreduced
+		field.fe_sub(field.fe_relax_cast(&z2), &z3, &z2) // z2 - unreduced
+		field.fe_carry_mul(&x2, &t1, &t0) // x2 - reduced
+		field.fe_sub(&t1, field.fe_tighten_cast(&t1), field.fe_tighten_cast(&t0)) // safe - t1/t0 is reduced
+		field.fe_carry_square(&z2, field.fe_relax_cast(&z2)) // z2 - reduced
+		field.fe_carry_scmul_121666(&z3, &t1)
+		field.fe_carry_square(&x3, field.fe_relax_cast(&x3)) // x3 - reduced
+		field.fe_add(&t0, field.fe_tighten_cast(&t0), &z3) // safe - t0 is reduced
+		field.fe_carry_mul(&z3, field.fe_relax_cast(&x1), field.fe_relax_cast(&z2))
+		field.fe_carry_mul(&z2, &t1, &t0)
+	}
+	// last swap is necessary to compensate for the xor trick
+	// Note: after this swap, P3 == P2 + P1.
+	field.fe_cond_swap(&x2, &x3, swap)
+	field.fe_cond_swap(&z2, &z3, swap)
+
+	// normalises the coordinates: x == X / Z
+	field.fe_carry_inv(&z2, field.fe_relax_cast(&z2))
+	field.fe_carry_mul(&x2, field.fe_relax_cast(&x2), field.fe_relax_cast(&z2))
+	field.fe_to_bytes(out, &x2)
+
+	mem.zero_explicit(&x1, size_of(x1))
+	mem.zero_explicit(&x2, size_of(x2))
+	mem.zero_explicit(&x3, size_of(x3))
+	mem.zero_explicit(&z2, size_of(z2))
+	mem.zero_explicit(&z3, size_of(z3))
+	mem.zero_explicit(&t0, size_of(t0))
+	mem.zero_explicit(&t1, size_of(t1))
+}
+
+scalarmult :: proc (dst, scalar, point: []byte) {
+	if len(scalar) != SCALAR_SIZE {
+		panic("crypto/x25519: invalid scalar size")
+	}
+	if len(point) != POINT_SIZE {
+		panic("crypto/x25519: invalid point size")
+	}
+	if len(dst) != POINT_SIZE {
+		panic("crypto/x25519: invalid destination point size")
+	}
+
+	// "clamp" the scalar
+	e: [32]byte = ---
+	copy_slice(e[:], scalar)
+	e[0] &= 248
+	e[31] &= 127
+	e[31] |= 64
+
+	p: [32]byte = ---
+	copy_slice(p[:], point)
+
+	d: [32]byte = ---
+	_scalarmult(&d, &e, &p)
+	copy_slice(dst, d[:])
+
+	mem.zero_explicit(&e, size_of(e))
+	mem.zero_explicit(&d, size_of(d))
+}
+
+scalarmult_basepoint :: proc (dst, scalar: []byte) {
+	// TODO/perf: Switch to using a precomputed table.
+	scalarmult(dst, scalar, _BASE_POINT[:])
+}
diff --git a/core/mem/virtual/virtual_linux.odin b/core/mem/virtual/virtual_linux.odin
index c4dd564ee..71a56e499 100644
--- a/core/mem/virtual/virtual_linux.odin
+++ b/core/mem/virtual/virtual_linux.odin
@@ -4,64 +4,56 @@ package mem_virtual
 
 import "core:c"
 import "core:intrinsics"
+import "core:sys/unix"
 
-when ODIN_ARCH == "amd64" {
-	SYS_mmap     :: 9
-	SYS_mprotect :: 10
-	SYS_munmap   :: 11
-	SYS_madvise  :: 28
-	
-	PROT_NONE  :: 0x0
-	PROT_READ  :: 0x1
-	PROT_WRITE :: 0x2
-	PROT_EXEC  :: 0x4
-	PROT_GROWSDOWN :: 0x01000000
-	PROT_GROWSUP :: 0x02000000
+PROT_NONE  :: 0x0
+PROT_READ  :: 0x1
+PROT_WRITE :: 0x2
+PROT_EXEC  :: 0x4
+PROT_GROWSDOWN :: 0x01000000
+PROT_GROWSUP :: 0x02000000
 
-	MAP_FIXED     :: 0x1
-	MAP_PRIVATE   :: 0x2
-	MAP_SHARED    :: 0x4
-	MAP_ANONYMOUS :: 0x20
-	
-	MADV_NORMAL      :: 0
-	MADV_RANDOM      :: 1
-	MADV_SEQUENTIAL  :: 2
-	MADV_WILLNEED    :: 3
-	MADV_DONTNEED    :: 4
-	MADV_FREE        :: 8
-	MADV_REMOVE      :: 9
-	MADV_DONTFORK    :: 10
-	MADV_DOFORK      :: 11
-	MADV_MERGEABLE   :: 12
-	MADV_UNMERGEABLE :: 13
-	MADV_HUGEPAGE    :: 14
-	MADV_NOHUGEPAGE  :: 15
-	MADV_DONTDUMP    :: 16
-	MADV_DODUMP      :: 17
-	MADV_WIPEONFORK  :: 18
-	MADV_KEEPONFORK  :: 19
-	MADV_HWPOISON    :: 100
-} else {
-	#panic("Unsupported architecture")
-}
+MAP_FIXED     :: 0x1
+MAP_PRIVATE   :: 0x2
+MAP_SHARED    :: 0x4
+MAP_ANONYMOUS :: 0x20
+
+MADV_NORMAL      :: 0
+MADV_RANDOM      :: 1
+MADV_SEQUENTIAL  :: 2
+MADV_WILLNEED    :: 3
+MADV_DONTNEED    :: 4
+MADV_FREE        :: 8
+MADV_REMOVE      :: 9
+MADV_DONTFORK    :: 10
+MADV_DOFORK      :: 11
+MADV_MERGEABLE   :: 12
+MADV_UNMERGEABLE :: 13
+MADV_HUGEPAGE    :: 14
+MADV_NOHUGEPAGE  :: 15
+MADV_DONTDUMP    :: 16
+MADV_DODUMP      :: 17
+MADV_WIPEONFORK  :: 18
+MADV_KEEPONFORK  :: 19
+MADV_HWPOISON    :: 100
 
 mmap :: proc "contextless" (addr: rawptr, length: uint, prot: c.int, flags: c.int, fd: c.int, offset: uintptr) -> rawptr {
-	res := intrinsics.syscall(SYS_mmap, uintptr(addr), uintptr(length), uintptr(prot), uintptr(flags), uintptr(fd), offset)
+	res := intrinsics.syscall(unix.SYS_mmap, uintptr(addr), uintptr(length), uintptr(prot), uintptr(flags), uintptr(fd), offset)
 	return rawptr(res)
 }
 
 munmap :: proc "contextless" (addr: rawptr, length: uint) -> c.int {
-	res := intrinsics.syscall(SYS_munmap, uintptr(addr), uintptr(length))
+	res := intrinsics.syscall(unix.SYS_munmap, uintptr(addr), uintptr(length))
 	return c.int(res)
 }
 
 mprotect :: proc "contextless" (addr: rawptr, length: uint, prot: c.int) -> c.int {
-	res := intrinsics.syscall(SYS_mprotect, uintptr(addr), uintptr(length), uint(prot))
+	res := intrinsics.syscall(unix.SYS_mprotect, uintptr(addr), uintptr(length), uint(prot))
 	return c.int(res)
 }
 
 madvise :: proc "contextless" (addr: rawptr, length: uint, advice: c.int) -> c.int {
-	res := intrinsics.syscall(SYS_madvise, uintptr(addr), uintptr(length), uintptr(advice))
+	res := intrinsics.syscall(unix.SYS_madvise, uintptr(addr), uintptr(length), uintptr(advice))
 	return c.int(res)
 }
 
diff --git a/core/os/os_linux.odin b/core/os/os_linux.odin
index bc4717b44..260a051ce 100644
--- a/core/os/os_linux.odin
+++ b/core/os/os_linux.odin
@@ -8,6 +8,7 @@ import "core:strings"
 import "core:c"
 import "core:strconv"
 import "core:intrinsics"
+import "core:sys/unix"
 
 Handle    :: distinct i32
 File_Time :: distinct u64
@@ -265,8 +266,6 @@ X_OK :: 1 // Test for execute permission
 W_OK :: 2 // Test for write permission
 R_OK :: 4 // Test for read permission
 
-SYS_GETTID :: 186
-
 foreign libc {
 	@(link_name="__errno_location") __errno_location    :: proc() -> ^int ---
 
@@ -594,7 +593,7 @@ exit :: proc "contextless" (code: int) -> ! {
 }
 
 current_thread_id :: proc "contextless" () -> int {
-	return cast(int)intrinsics.syscall(SYS_GETTID)
+	return unix.sys_gettid()
 }
 
 dlopen :: proc(filename: string, flags: int) -> rawptr {
diff --git a/core/sync/sync2/futex_linux.odin b/core/sync/sync2/futex_linux.odin
index 1bd41c7cf..fca28cace 100644
--- a/core/sync/sync2/futex_linux.odin
+++ b/core/sync/sync2/futex_linux.odin
@@ -5,6 +5,7 @@ package sync2
 import "core:c"
 import "core:time"
 import "core:intrinsics"
+import "core:sys/unix"
 
 FUTEX_WAIT :: 0
 FUTEX_WAKE :: 1
@@ -34,7 +35,7 @@ get_errno :: proc(r: int) -> int {
 }
 
 internal_futex :: proc(f: ^Futex, op: c.int, val: u32, timeout: rawptr) -> int {
-	code := int(intrinsics.syscall(202, uintptr(f), uintptr(op), uintptr(val), uintptr(timeout), 0, 0))
+	code := int(intrinsics.syscall(unix.SYS_futex, uintptr(f), uintptr(op), uintptr(val), uintptr(timeout), 0, 0))
 	return get_errno(code)
 }
 
diff --git a/core/sync/sync2/primitives_linux.odin b/core/sync/sync2/primitives_linux.odin
index 4c81295bd..89ed97985 100644
--- a/core/sync/sync2/primitives_linux.odin
+++ b/core/sync/sync2/primitives_linux.odin
@@ -2,9 +2,8 @@
 //+private
 package sync2
 
-import "core:intrinsics"
+import "core:sys/unix"
 
 _current_thread_id :: proc "contextless" () -> int {
-	SYS_GETTID :: 186
-	return int(intrinsics.syscall(SYS_GETTID))
+	return unix.sys_gettid()
 }
diff --git a/core/sync/sync_linux.odin b/core/sync/sync_linux.odin
index fe856df94..340437c11 100644
--- a/core/sync/sync_linux.odin
+++ b/core/sync/sync_linux.odin
@@ -1,11 +1,9 @@
 package sync
 
 import "core:sys/unix"
-import "core:intrinsics"
 
 current_thread_id :: proc "contextless" () -> int {
-	SYS_GETTID :: 186
-	return int(intrinsics.syscall(SYS_GETTID))
+	return unix.sys_gettid()
 }
 
 
diff --git a/core/sys/unix/syscalls_linux.odin b/core/sys/unix/syscalls_linux.odin
new file mode 100644
index 000000000..659eedfbb
--- /dev/null
+++ b/core/sys/unix/syscalls_linux.odin
@@ -0,0 +1,60 @@
+package unix
+
+import "core:intrinsics"
+
+// Linux has inconsistent system call numbering across architectures,
+// for largely historical reasons.  This attempts to provide a unified
+// Odin-side interface for system calls that are required for the core
+// library to work.
+
+// For authorative system call numbers, the following files in the kernel
+// source can be used:
+//
+//  amd64: arch/x86/entry/syscalls/syscall_64.tbl
+//  arm64: include/uapi/asm-generic/unistd.h
+//  386: arch/x86/entry/syscalls/sycall_32.tbl
+//  arm: arch/arm/tools/syscall.tbl
+
+when ODIN_ARCH == "amd64" {
+	SYS_mmap : uintptr : 9
+	SYS_mprotect : uintptr : 10
+	SYS_munmap : uintptr : 11
+	SYS_madvise : uintptr : 28
+	SYS_futex : uintptr : 202
+	SYS_gettid : uintptr : 186
+	SYS_getrandom : uintptr : 318
+} else when ODIN_ARCH == "arm64" {
+	SYS_mmap : uintptr : 222
+	SYS_mprotect : uintptr : 226
+	SYS_munmap : uintptr : 215
+	SYS_madvise : uintptr : 233
+	SYS_futex : uintptr : 98
+	SYS_gettid : uintptr : 178
+	SYS_getrandom : uintptr : 278
+} else when ODIN_ARCH == "386" {
+	SYS_mmap : uintptr : 192 // 90 is "sys_old_mmap", we want mmap2
+	SYS_mprotect : uintptr : 125
+	SYS_munmap : uintptr : 91
+	SYS_madvise : uintptr : 219
+	SYS_futex : uintptr : 240
+	SYS_gettid : uintptr : 224
+	SYS_getrandom : uintptr : 355
+} else when ODIN_ARCH == "arm" {
+	SYS_mmap : uintptr : 192 // 90 is "sys_old_mmap", we want mmap2
+	SYS_mprotect : uintptr : 125
+	SYS_munmap: uintptr : 91
+	SYS_madvise: uintptr : 220
+	SYS_futex : uintptr : 240
+	SYS_gettid : uintptr: 224
+	SYS_getrandom : uintptr : 384
+} else {
+	#panic("Unsupported architecture")
+}
+
+sys_gettid :: proc "contextless" () -> int {
+	return cast(int)intrinsics.syscall(SYS_gettid)
+}
+
+sys_getrandom :: proc "contextless" (buf: ^byte, buflen: int, flags: uint) -> int {
+	return cast(int)intrinsics.syscall(SYS_getrandom, buf, cast(uintptr)(buflen), cast(uintptr)(flags))
+}
diff --git a/src/main.cpp b/src/main.cpp
index 5371393d1..99a55b2b6 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -545,8 +545,8 @@ void usage(String argv0) {
 	print_usage_line(1, "version   print version");
 	print_usage_line(1, "report    print information useful to reporting a bug");
 	print_usage_line(0, "");
-	print_usage_line(0, "For more information of flags, apply the flag to see what is possible");
-	print_usage_line(1, "-help");
+	print_usage_line(0, "For further details on a command, use -help after the command name");
+	print_usage_line(1, "e.g. odin build -help");
 }
 
 
diff --git a/tests/core/crypto/test_core_crypto.odin b/tests/core/crypto/test_core_crypto.odin
index df9920552..2ad00be66 100644
--- a/tests/core/crypto/test_core_crypto.odin
+++ b/tests/core/crypto/test_core_crypto.odin
@@ -115,6 +115,15 @@ main :: proc() {
     test_haval_224(&t)
     test_haval_256(&t)
 
+    // "modern" crypto tests
+    test_chacha20(&t)
+    test_poly1305(&t)
+    test_chacha20poly1305(&t)
+    test_x25519(&t)
+    test_rand_bytes(&t)
+
+    bench_modern(&t)
+
     fmt.printf("%v/%v tests successful.\n", TEST_count - TEST_fail, TEST_count)
 }
 
diff --git a/tests/core/crypto/test_core_crypto_modern.odin b/tests/core/crypto/test_core_crypto_modern.odin
new file mode 100644
index 000000000..71adad137
--- /dev/null
+++ b/tests/core/crypto/test_core_crypto_modern.odin
@@ -0,0 +1,535 @@
+package test_core_crypto
+
+import "core:testing"
+import "core:fmt"
+import "core:mem"
+import "core:time"
+import "core:crypto"
+
+import "core:crypto/chacha20"
+import "core:crypto/chacha20poly1305"
+import "core:crypto/poly1305"
+import "core:crypto/x25519"
+
+_digit_value :: proc(r: rune) -> int {
+	ri := int(r)
+	v: int = 16
+	switch r {
+	case '0'..='9': v = ri-'0'
+	case 'a'..='z': v = ri-'a'+10
+	case 'A'..='Z': v = ri-'A'+10
+	}
+	return v
+}
+
+_decode_hex32 :: proc(s: string) -> [32]byte{
+	b: [32]byte
+	for i := 0; i < len(s); i = i + 2 {
+		hi := _digit_value(rune(s[i]))
+		lo := _digit_value(rune(s[i+1]))
+		b[i/2] = byte(hi << 4 | lo)
+	}
+	return b
+}
+
+_PLAINTEXT_SUNSCREEN_STR := "Ladies and Gentlemen of the class of '99: If I could offer you only one tip for the future, sunscreen would be it."
+
+@(test)
+test_chacha20 :: proc(t: ^testing.T) {
+	log(t, "Testing (X)ChaCha20")
+
+	// Test cases taken from RFC 8439, and draft-irtf-cfrg-xchacha-03
+	plaintext := transmute([]byte)(_PLAINTEXT_SUNSCREEN_STR)
+
+	key := [chacha20.KEY_SIZE]byte{
+		0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
+		0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f,
+		0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
+		0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f,
+	}
+
+	nonce := [chacha20.NONCE_SIZE]byte{
+		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x4a,
+		0x00, 0x00, 0x00, 0x00,
+	}
+
+	ciphertext := [114]byte{
+		0x6e, 0x2e, 0x35, 0x9a, 0x25, 0x68, 0xf9, 0x80,
+		0x41, 0xba, 0x07, 0x28, 0xdd, 0x0d, 0x69, 0x81,
+		0xe9, 0x7e, 0x7a, 0xec, 0x1d, 0x43, 0x60, 0xc2,
+		0x0a, 0x27, 0xaf, 0xcc, 0xfd, 0x9f, 0xae, 0x0b,
+		0xf9, 0x1b, 0x65, 0xc5, 0x52, 0x47, 0x33, 0xab,
+		0x8f, 0x59, 0x3d, 0xab, 0xcd, 0x62, 0xb3, 0x57,
+		0x16, 0x39, 0xd6, 0x24, 0xe6, 0x51, 0x52, 0xab,
+		0x8f, 0x53, 0x0c, 0x35, 0x9f, 0x08, 0x61, 0xd8,
+		0x07, 0xca, 0x0d, 0xbf, 0x50, 0x0d, 0x6a, 0x61,
+		0x56, 0xa3, 0x8e, 0x08, 0x8a, 0x22, 0xb6, 0x5e,
+		0x52, 0xbc, 0x51, 0x4d, 0x16, 0xcc, 0xf8, 0x06,
+		0x81, 0x8c, 0xe9, 0x1a, 0xb7, 0x79, 0x37, 0x36,
+		0x5a, 0xf9, 0x0b, 0xbf, 0x74, 0xa3, 0x5b, 0xe6,
+		0xb4, 0x0b, 0x8e, 0xed, 0xf2, 0x78, 0x5e, 0x42,
+		0x87, 0x4d,
+	}
+	ciphertext_str := hex_string(ciphertext[:])
+
+	derived_ciphertext: [114]byte
+	ctx: chacha20.Context = ---
+	chacha20.init(&ctx, key[:], nonce[:])
+	chacha20.seek(&ctx, 1) // The test vectors start the counter at 1.
+	chacha20.xor_bytes(&ctx, derived_ciphertext[:], plaintext[:])
+
+	derived_ciphertext_str := hex_string(derived_ciphertext[:])
+	expect(t, derived_ciphertext_str == ciphertext_str, fmt.tprintf("Expected %s for xor_bytes(plaintext_str), but got %s instead", ciphertext_str, derived_ciphertext_str))
+
+	xkey := [chacha20.KEY_SIZE]byte{
+		0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87,
+		0x88, 0x89, 0x8a, 0x8b, 0x8c, 0x8d, 0x8e, 0x8f,
+		0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97,
+		0x98, 0x99, 0x9a, 0x9b, 0x9c, 0x9d, 0x9e, 0x9f,
+	}
+
+	xnonce := [chacha20.XNONCE_SIZE]byte{
+		0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47,
+		0x48, 0x49, 0x4a, 0x4b, 0x4c, 0x4d, 0x4e, 0x4f,
+		0x50, 0x51, 0x52, 0x53, 0x54, 0x55, 0x56, 0x57,
+	}
+
+	xciphertext := [114]byte{
+		0xbd, 0x6d, 0x17, 0x9d, 0x3e, 0x83, 0xd4, 0x3b,
+		0x95, 0x76, 0x57, 0x94, 0x93, 0xc0, 0xe9, 0x39,
+		0x57, 0x2a, 0x17, 0x00, 0x25, 0x2b, 0xfa, 0xcc,
+		0xbe, 0xd2, 0x90, 0x2c, 0x21, 0x39, 0x6c, 0xbb,
+		0x73, 0x1c, 0x7f, 0x1b, 0x0b, 0x4a, 0xa6, 0x44,
+		0x0b, 0xf3, 0xa8, 0x2f, 0x4e, 0xda, 0x7e, 0x39,
+		0xae, 0x64, 0xc6, 0x70, 0x8c, 0x54, 0xc2, 0x16,
+		0xcb, 0x96, 0xb7, 0x2e, 0x12, 0x13, 0xb4, 0x52,
+		0x2f, 0x8c, 0x9b, 0xa4, 0x0d, 0xb5, 0xd9, 0x45,
+		0xb1, 0x1b, 0x69, 0xb9, 0x82, 0xc1, 0xbb, 0x9e,
+		0x3f, 0x3f, 0xac, 0x2b, 0xc3, 0x69, 0x48, 0x8f,
+		0x76, 0xb2, 0x38, 0x35, 0x65, 0xd3, 0xff, 0xf9,
+		0x21, 0xf9, 0x66, 0x4c, 0x97, 0x63, 0x7d, 0xa9,
+		0x76, 0x88, 0x12, 0xf6, 0x15, 0xc6, 0x8b, 0x13,
+		0xb5, 0x2e,
+	}
+	xciphertext_str := hex_string(xciphertext[:])
+
+	chacha20.init(&ctx, xkey[:], xnonce[:])
+	chacha20.seek(&ctx, 1)
+	chacha20.xor_bytes(&ctx, derived_ciphertext[:], plaintext[:])
+
+	derived_ciphertext_str = hex_string(derived_ciphertext[:])
+	expect(t, derived_ciphertext_str == xciphertext_str, fmt.tprintf("Expected %s for xor_bytes(plaintext_str), but got %s instead", xciphertext_str, derived_ciphertext_str))
+}
+
+@(test)
+test_poly1305 :: proc(t: ^testing.T) {
+	log(t, "Testing poly1305")
+
+	// Test cases taken from poly1305-donna.
+	key := [poly1305.KEY_SIZE]byte{
+		0xee,0xa6,0xa7,0x25,0x1c,0x1e,0x72,0x91,
+		0x6d,0x11,0xc2,0xcb,0x21,0x4d,0x3c,0x25,
+		0x25,0x39,0x12,0x1d,0x8e,0x23,0x4e,0x65,
+		0x2d,0x65,0x1f,0xa4,0xc8,0xcf,0xf8,0x80,
+	}
+
+	msg := [131]byte{
+		0x8e,0x99,0x3b,0x9f,0x48,0x68,0x12,0x73,
+		0xc2,0x96,0x50,0xba,0x32,0xfc,0x76,0xce,
+		0x48,0x33,0x2e,0xa7,0x16,0x4d,0x96,0xa4,
+		0x47,0x6f,0xb8,0xc5,0x31,0xa1,0x18,0x6a,
+		0xc0,0xdf,0xc1,0x7c,0x98,0xdc,0xe8,0x7b,
+		0x4d,0xa7,0xf0,0x11,0xec,0x48,0xc9,0x72,
+		0x71,0xd2,0xc2,0x0f,0x9b,0x92,0x8f,0xe2,
+		0x27,0x0d,0x6f,0xb8,0x63,0xd5,0x17,0x38,
+		0xb4,0x8e,0xee,0xe3,0x14,0xa7,0xcc,0x8a,
+		0xb9,0x32,0x16,0x45,0x48,0xe5,0x26,0xae,
+		0x90,0x22,0x43,0x68,0x51,0x7a,0xcf,0xea,
+		0xbd,0x6b,0xb3,0x73,0x2b,0xc0,0xe9,0xda,
+		0x99,0x83,0x2b,0x61,0xca,0x01,0xb6,0xde,
+		0x56,0x24,0x4a,0x9e,0x88,0xd5,0xf9,0xb3,
+		0x79,0x73,0xf6,0x22,0xa4,0x3d,0x14,0xa6,
+		0x59,0x9b,0x1f,0x65,0x4c,0xb4,0x5a,0x74,
+		0xe3,0x55,0xa5,
+	}
+
+	tag := [poly1305.TAG_SIZE]byte{
+		0xf3,0xff,0xc7,0x70,0x3f,0x94,0x00,0xe5,
+		0x2a,0x7d,0xfb,0x4b,0x3d,0x33,0x05,0xd9,
+	}
+	tag_str := hex_string(tag[:])
+
+	// Verify - oneshot + compare
+	ok := poly1305.verify(tag[:], msg[:], key[:])
+	expect(t, ok, "oneshot verify call failed")
+
+	// Sum - oneshot
+	derived_tag: [poly1305.TAG_SIZE]byte
+	poly1305.sum(derived_tag[:], msg[:], key[:])
+	derived_tag_str := hex_string(derived_tag[:])
+	expect(t, derived_tag_str == tag_str, fmt.tprintf("Expected %s for sum(msg, key), but got %s instead", tag_str, derived_tag_str))
+
+	// Incremental
+	mem.zero(&derived_tag, size_of(derived_tag))
+	ctx: poly1305.Context = ---
+	poly1305.init(&ctx, key[:])
+	read_lengths := [11]int{32, 64, 16, 8, 4, 2, 1, 1, 1, 1, 1}
+	off := 0
+	for read_length in read_lengths {
+		to_read := msg[off:off+read_length]
+		poly1305.update(&ctx, to_read)
+		off = off + read_length
+	}
+	poly1305.final(&ctx, derived_tag[:])
+	derived_tag_str = hex_string(derived_tag[:])
+	expect(t, derived_tag_str == tag_str, fmt.tprintf("Expected %s for init/update/final - incremental, but got %s instead", tag_str, derived_tag_str))
+}
+
+@(test)
+test_chacha20poly1305 :: proc(t: ^testing.T) {
+	log(t, "Testing chacha20poly1205")
+
+	plaintext := transmute([]byte)(_PLAINTEXT_SUNSCREEN_STR)
+
+	aad := [12]byte{
+		0x50, 0x51, 0x52, 0x53, 0xc0, 0xc1, 0xc2, 0xc3,
+		0xc4, 0xc5, 0xc6, 0xc7,
+	}
+
+	key := [chacha20poly1305.KEY_SIZE]byte{
+		0x80, 0x81, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87,
+		0x88, 0x89, 0x8a, 0x8b, 0x8c, 0x8d, 0x8e, 0x8f,
+		0x90, 0x91, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97,
+		0x98, 0x99, 0x9a, 0x9b, 0x9c, 0x9d, 0x9e, 0x9f,
+	}
+
+	nonce := [chacha20poly1305.NONCE_SIZE]byte{
+		0x07, 0x00, 0x00, 0x00,
+		0x40, 0x41, 0x42, 0x43, 0x44, 0x45, 0x46, 0x47,
+	}
+
+	ciphertext := [114]byte{
+		0xd3, 0x1a, 0x8d, 0x34, 0x64, 0x8e, 0x60, 0xdb,
+		0x7b, 0x86, 0xaf, 0xbc, 0x53, 0xef, 0x7e, 0xc2,
+		0xa4, 0xad, 0xed, 0x51, 0x29, 0x6e, 0x08, 0xfe,
+		0xa9, 0xe2, 0xb5, 0xa7, 0x36, 0xee, 0x62, 0xd6,
+		0x3d, 0xbe, 0xa4, 0x5e, 0x8c, 0xa9, 0x67, 0x12,
+		0x82, 0xfa, 0xfb, 0x69, 0xda, 0x92, 0x72, 0x8b,
+		0x1a, 0x71, 0xde, 0x0a, 0x9e, 0x06, 0x0b, 0x29,
+		0x05, 0xd6, 0xa5, 0xb6, 0x7e, 0xcd, 0x3b, 0x36,
+		0x92, 0xdd, 0xbd, 0x7f, 0x2d, 0x77, 0x8b, 0x8c,
+		0x98, 0x03, 0xae, 0xe3, 0x28, 0x09, 0x1b, 0x58,
+		0xfa, 0xb3, 0x24, 0xe4, 0xfa, 0xd6, 0x75, 0x94,
+		0x55, 0x85, 0x80, 0x8b, 0x48, 0x31, 0xd7, 0xbc,
+		0x3f, 0xf4, 0xde, 0xf0, 0x8e, 0x4b, 0x7a, 0x9d,
+		0xe5, 0x76, 0xd2, 0x65, 0x86, 0xce, 0xc6, 0x4b,
+		0x61, 0x16,
+	}
+	ciphertext_str := hex_string(ciphertext[:])
+
+	tag := [chacha20poly1305.TAG_SIZE]byte{
+		0x1a, 0xe1, 0x0b, 0x59, 0x4f, 0x09, 0xe2, 0x6a,
+		0x7e, 0x90, 0x2e, 0xcb, 0xd0, 0x60, 0x06, 0x91,
+	}
+	tag_str := hex_string(tag[:])
+
+	derived_tag: [chacha20poly1305.TAG_SIZE]byte
+	derived_ciphertext: [114]byte
+
+	chacha20poly1305.encrypt(derived_ciphertext[:], derived_tag[:], key[:], nonce[:], aad[:], plaintext)
+
+	derived_ciphertext_str := hex_string(derived_ciphertext[:])
+	expect(t, derived_ciphertext_str == ciphertext_str, fmt.tprintf("Expected ciphertext %s for encrypt(aad, plaintext), but got %s instead", ciphertext_str, derived_ciphertext_str))
+
+	derived_tag_str := hex_string(derived_tag[:])
+	expect(t, derived_tag_str == tag_str, fmt.tprintf("Expected tag %s for encrypt(aad, plaintext), but got %s instead", tag_str, derived_tag_str))
+
+	derived_plaintext: [114]byte
+	ok := chacha20poly1305.decrypt(derived_plaintext[:], tag[:], key[:], nonce[:], aad[:], ciphertext[:])
+	derived_plaintext_str := string(derived_plaintext[:])
+	expect(t, ok, "Expected true for decrypt(tag, aad, ciphertext)")
+	expect(t, derived_plaintext_str == _PLAINTEXT_SUNSCREEN_STR, fmt.tprintf("Expected plaintext %s for decrypt(tag, aad, ciphertext), but got %s instead", _PLAINTEXT_SUNSCREEN_STR, derived_plaintext_str))
+
+	derived_ciphertext[0] ~= 0xa5
+	ok = chacha20poly1305.decrypt(derived_plaintext[:], tag[:], key[:], nonce[:], aad[:], derived_ciphertext[:])
+	expect(t, !ok, "Expected false for decrypt(tag, aad, corrupted_ciphertext)")
+
+	aad[0] ~= 0xa5
+	ok = chacha20poly1305.decrypt(derived_plaintext[:], tag[:], key[:], nonce[:], aad[:], ciphertext[:])
+	expect(t, !ok, "Expected false for decrypt(tag, corrupted_aad, ciphertext)")
+}
+
+TestECDH :: struct {
+	scalar:  string,
+	point:   string,
+	product: string,
+}
+
+@(test)
+test_x25519 :: proc(t: ^testing.T) {
+	log(t, "Testing X25519")
+
+	test_vectors := [?]TestECDH {
+		// Test vectors from RFC 7748
+		TestECDH{
+			"a546e36bf0527c9d3b16154b82465edd62144c0ac1fc5a18506a2244ba449ac4",
+			"e6db6867583030db3594c1a424b15f7c726624ec26b3353b10a903a6d0ab1c4c",
+			"c3da55379de9c6908e94ea4df28d084f32eccf03491c71f754b4075577a28552",
+		},
+		TestECDH{
+			"4b66e9d4d1b4673c5ad22691957d6af5c11b6421e0ea01d42ca4169e7918ba0d",
+			"e5210f12786811d3f4b7959d0538ae2c31dbe7106fc03c3efc4cd549c715a493",
+			"95cbde9476e8907d7aade45cb4b873f88b595a68799fa152e6f8f7647aac7957",
+		},
+	}
+	for v, _ in test_vectors {
+		scalar := _decode_hex32(v.scalar)
+		point := _decode_hex32(v.point)
+
+		derived_point: [x25519.POINT_SIZE]byte
+		x25519.scalarmult(derived_point[:], scalar[:], point[:])
+		derived_point_str := hex_string(derived_point[:])
+
+		expect(t, derived_point_str == v.product, fmt.tprintf("Expected %s for %s * %s, but got %s instead", v.product, v.scalar, v.point, derived_point_str))
+
+		// Abuse the test vectors to sanity-check the scalar-basepoint multiply.
+		p1, p2: [x25519.POINT_SIZE]byte
+		x25519.scalarmult_basepoint(p1[:], scalar[:])
+		x25519.scalarmult(p2[:], scalar[:], x25519._BASE_POINT[:])
+		p1_str, p2_str := hex_string(p1[:]), hex_string(p2[:])
+		expect(t, p1_str == p2_str, fmt.tprintf("Expected %s for %s * basepoint, but got %s instead", p2_str, v.scalar, p1_str))
+	}
+
+    // TODO/tests: Run the wycheproof test vectors, once I figure out
+    // how to work with JSON.
+}
+
+@(test)
+test_rand_bytes :: proc(t: ^testing.T) {
+	log(t, "Testing rand_bytes")
+
+	if ODIN_OS != "linux" {
+		log(t, "rand_bytes not supported - skipping")
+		return
+	}
+
+	allocator := context.allocator
+
+	buf := make([]byte, 1 << 25, allocator)
+	defer delete(buf)
+
+	// Testing a CSPRNG for correctness is incredibly involved and
+	// beyond the scope of an implementation that offloads
+	// responsibility for correctness to the OS.
+	//
+	// Just attempt to randomize a sufficiently large buffer, where
+	// sufficiently large is:
+	//  * Larger than the maximum getentropy request size (256 bytes).
+	//  * Larger than the maximum getrandom request size (2^25 - 1 bytes).
+	//
+	// While theoretically non-deterministic, if this fails, chances
+	// are the CSPRNG is busted.
+	seems_ok := false
+	for i := 0; i < 256; i = i + 1 {
+		mem.zero_explicit(raw_data(buf), len(buf))
+		crypto.rand_bytes(buf)
+
+		if buf[0] != 0 && buf[len(buf)-1] != 0 {
+			seems_ok = true
+			break
+		}
+	}
+
+	expect(t, seems_ok, "Expected to randomize the head and tail of the buffer within a handful of attempts")
+}
+
+@(test)
+bench_modern :: proc(t: ^testing.T) {
+	fmt.println("Starting benchmarks:")
+
+	bench_chacha20(t)
+	bench_poly1305(t)
+	bench_chacha20poly1305(t)
+	bench_x25519(t)
+}
+
+_setup_sized_buf :: proc(options: ^time.Benchmark_Options, allocator := context.allocator) -> (err: time.Benchmark_Error) {
+	assert(options != nil)
+
+	options.input = make([]u8, options.bytes, allocator)
+	return nil if len(options.input) == options.bytes else .Allocation_Error
+}
+
+_teardown_sized_buf :: proc(options: ^time.Benchmark_Options, allocator := context.allocator) -> (err: time.Benchmark_Error) {
+	assert(options != nil)
+
+	delete(options.input)
+	return nil
+}
+
+_benchmark_chacha20 :: proc(options: ^time.Benchmark_Options, allocator := context.allocator) -> (err: time.Benchmark_Error) {
+	buf := options.input
+	key := [chacha20.KEY_SIZE]byte{
+		0xde, 0xad, 0xbe, 0xef, 0xde, 0xad, 0xbe, 0xef,
+		0xde, 0xad, 0xbe, 0xef, 0xde, 0xad, 0xbe, 0xef,
+		0xde, 0xad, 0xbe, 0xef, 0xde, 0xad, 0xbe, 0xef,
+		0xde, 0xad, 0xbe, 0xef, 0xde, 0xad, 0xbe, 0xef,
+	}
+	nonce := [chacha20.NONCE_SIZE]byte{
+		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+		0x00, 0x00, 0x00, 0x00,
+	}
+
+	ctx: chacha20.Context = ---
+	chacha20.init(&ctx, key[:], nonce[:])
+
+	for _ in 0..=options.rounds {
+		chacha20.xor_bytes(&ctx, buf, buf)
+	}
+	options.count     = options.rounds
+	options.processed = options.rounds * options.bytes
+	return nil
+}
+
+_benchmark_poly1305 :: proc(options: ^time.Benchmark_Options, allocator := context.allocator) -> (err: time.Benchmark_Error) {
+	buf := options.input
+	key := [poly1305.KEY_SIZE]byte{
+		0xde, 0xad, 0xbe, 0xef, 0xde, 0xad, 0xbe, 0xef,
+		0xde, 0xad, 0xbe, 0xef, 0xde, 0xad, 0xbe, 0xef,
+		0xde, 0xad, 0xbe, 0xef, 0xde, 0xad, 0xbe, 0xef,
+		0xde, 0xad, 0xbe, 0xef, 0xde, 0xad, 0xbe, 0xef,
+	}
+
+	tag: [poly1305.TAG_SIZE]byte = ---
+	for _ in 0..=options.rounds {
+		poly1305.sum(tag[:], buf, key[:])
+	}
+	options.count     = options.rounds
+	options.processed = options.rounds * options.bytes
+	//options.hash      = u128(h)
+	return nil
+}
+
+_benchmark_chacha20poly1305 :: proc(options: ^time.Benchmark_Options, allocator := context.allocator) -> (err: time.Benchmark_Error) {
+	buf := options.input
+	key := [chacha20.KEY_SIZE]byte{
+		0xde, 0xad, 0xbe, 0xef, 0xde, 0xad, 0xbe, 0xef,
+		0xde, 0xad, 0xbe, 0xef, 0xde, 0xad, 0xbe, 0xef,
+		0xde, 0xad, 0xbe, 0xef, 0xde, 0xad, 0xbe, 0xef,
+		0xde, 0xad, 0xbe, 0xef, 0xde, 0xad, 0xbe, 0xef,
+	}
+	nonce := [chacha20.NONCE_SIZE]byte{
+		0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+		0x00, 0x00, 0x00, 0x00,
+	}
+
+	tag: [chacha20poly1305.TAG_SIZE]byte = ---
+
+	for _ in 0..=options.rounds {
+		chacha20poly1305.encrypt(buf,tag[:], key[:], nonce[:], nil, buf)
+	}
+	options.count     = options.rounds
+	options.processed = options.rounds * options.bytes
+	return nil
+}
+
+benchmark_print :: proc(name: string, options: ^time.Benchmark_Options) {
+	fmt.printf("\t[%v] %v rounds, %v bytes processed in %v ns\n\t\t%5.3f rounds/s, %5.3f MiB/s\n",
+		name,
+		options.rounds,
+		options.processed,
+		time.duration_nanoseconds(options.duration),
+		options.rounds_per_second,
+		options.megabytes_per_second,
+	)
+}
+
+bench_chacha20 :: proc(t: ^testing.T) {
+	name    := "ChaCha20 64 bytes"
+	options := &time.Benchmark_Options{
+		rounds   = 1_000,
+		bytes    = 64,
+		setup    = _setup_sized_buf,
+		bench    = _benchmark_chacha20,
+		teardown = _teardown_sized_buf,
+	}
+
+	err  := time.benchmark(options, context.allocator)
+	expect(t, err == nil, name)
+	benchmark_print(name, options)
+
+	name = "ChaCha20 1024 bytes"
+	options.bytes = 1024
+	err = time.benchmark(options, context.allocator)
+	expect(t, err == nil, name)
+	benchmark_print(name, options)
+
+	name = "ChaCha20 65536 bytes"
+	options.bytes = 65536
+	err = time.benchmark(options, context.allocator)
+	expect(t, err == nil, name)
+	benchmark_print(name, options)
+}
+
+bench_poly1305 :: proc(t: ^testing.T) {
+	name    := "Poly1305 64 zero bytes"
+	options := &time.Benchmark_Options{
+		rounds   = 1_000,
+		bytes    = 64,
+		setup    = _setup_sized_buf,
+		bench    = _benchmark_poly1305,
+		teardown = _teardown_sized_buf,
+	}
+
+	err  := time.benchmark(options, context.allocator)
+	expect(t, err == nil, name)
+	benchmark_print(name, options)
+
+	name = "Poly1305 1024 zero bytes"
+	options.bytes = 1024
+	err = time.benchmark(options, context.allocator)
+	expect(t, err == nil, name)
+	benchmark_print(name, options)
+}
+
+bench_chacha20poly1305 :: proc(t: ^testing.T) {
+	name    := "chacha20poly1305 64 bytes"
+	options := &time.Benchmark_Options{
+		rounds   = 1_000,
+		bytes    = 64,
+		setup    = _setup_sized_buf,
+		bench    = _benchmark_chacha20poly1305,
+		teardown = _teardown_sized_buf,
+	}
+
+	err  := time.benchmark(options, context.allocator)
+	expect(t, err == nil, name)
+	benchmark_print(name, options)
+
+	name = "chacha20poly1305 1024 bytes"
+	options.bytes = 1024
+	err = time.benchmark(options, context.allocator)
+	expect(t, err == nil, name)
+	benchmark_print(name, options)
+
+	name = "chacha20poly1305 65536 bytes"
+	options.bytes = 65536
+	err = time.benchmark(options, context.allocator)
+	expect(t, err == nil, name)
+	benchmark_print(name, options)
+}
+
+bench_x25519 :: proc(t: ^testing.T) {
+	point := _decode_hex32("deadbeefdeadbeefdeadbeefdeadbeefdeadbeefdeadbeefdeadbeefdeadbeef")
+	scalar := _decode_hex32("cafebabecafebabecafebabecafebabecafebabecafebabecafebabecafebabe")
+	out: [x25519.POINT_SIZE]byte = ---
+
+	iters :: 10000
+	start := time.now()
+	for i := 0; i < iters; i = i + 1 {
+		x25519.scalarmult(out[:], scalar[:], point[:])
+	}
+	elapsed := time.since(start)
+
+	log(t, fmt.tprintf("x25519.scalarmult: ~%f us/op", time.duration_microseconds(elapsed) / iters))
+}