core/crypto: More improvements

core/crypto/_aes/aes.odin

@@ -25,4 +25,5 @@ GHASH_BLOCK_SIZE :: 16
 GHASH_TAG_SIZE :: 16
 
 // RCON is the AES keyschedule round constants.
+@(rodata)
 RCON := [10]byte{0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80, 0x1B, 0x36}

core/crypto/_aes/ct64/ct64.odin

@@ -22,8 +22,6 @@
 
 package aes_ct64
 
-import "base:intrinsics"
-
 // Bitsliced AES for 64-bit general purpose (integer) registers.  Each
 // invocation will process up to 4 blocks at a time.  This implementation
 // is derived from the BearSSL ct64 code, and distributed under a 1-clause
@@ -212,11 +210,8 @@ orthogonalize :: proc "contextless" (q: ^[8]u64) {
 }
 
 @(require_results)
-interleave_in :: proc "contextless" (w: []u32) -> (q0, q1: u64) #no_bounds_check {
-	if len(w) < 4 {
-		intrinsics.trap()
-	}
-	x0, x1, x2, x3 := u64(w[0]), u64(w[1]), u64(w[2]), u64(w[3])
+interleave_in :: proc "contextless" (w0, w1, w2, w3: u32) -> (q0, q1: u64) #no_bounds_check {
+	x0, x1, x2, x3 := u64(w0), u64(w1), u64(w2), u64(w3)
 	x0 |= (x0 << 16)
 	x1 |= (x1 << 16)
 	x2 |= (x2 << 16)

core/crypto/_aes/ct64/ct64_enc.odin

@@ -22,12 +22,8 @@
 
 package aes_ct64
 
-import "base:intrinsics"
-
 add_round_key :: proc "contextless" (q: ^[8]u64, sk: []u64) #no_bounds_check {
-	if len(sk) < 8 {
-		intrinsics.trap()
-	}
+	ensure_contextless(len(sk) >= 8, "aes/ct64: invalid round key size")
 
 	q[0] ~= sk[0]
 	q[1] ~= sk[1]

core/crypto/_aes/ct64/ct64_keysched.odin

@@ -22,7 +22,6 @@
 
 package aes_ct64
 
-import "base:intrinsics"
 import "core:crypto/_aes"
 import "core:encoding/endian"
 import "core:mem"
@@ -42,7 +41,7 @@ sub_word :: proc "contextless" (x: u32) -> u32 {
 }
 
 @(private, require_results)
-keysched :: proc(comp_skey: []u64, key: []byte) -> int {
+keysched :: proc "contextless" (comp_skey: []u64, key: []byte) -> int {
 	num_rounds, key_len := 0, len(key)
 	switch key_len {
 	case _aes.KEY_SIZE_128:
@@ -52,7 +51,7 @@ keysched :: proc(comp_skey: []u64, key: []byte) -> int {
 	case _aes.KEY_SIZE_256:
 		num_rounds = _aes.ROUNDS_256
 	case:
-		panic("crypto/aes: invalid AES key size")
+		panic_contextless("crypto/aes: invalid AES key size")
 	}
 
 	skey: [60]u32 = ---
@@ -78,7 +77,7 @@ keysched :: proc(comp_skey: []u64, key: []byte) -> int {
 
 	q: [8]u64 = ---
 	for i, j := 0, 0; i < nkf; i, j = i + 4, j + 2 {
-		q[0], q[4] = interleave_in(skey[i:])
+		q[0], q[4] = interleave_in(skey[i], skey[i+1], skey[i+2], skey[i+3])
 		q[1] = q[0]
 		q[2] = q[0]
 		q[3] = q[0]
@@ -123,57 +122,3 @@ skey_expand :: proc "contextless" (skey, comp_skey: []u64, num_rounds: int) {
 		skey[v + 3] = (x3 << 4) - x3
 	}
 }
-
-orthogonalize_roundkey :: proc "contextless" (qq: []u64, key: []byte) {
-	if len(qq) < 8 || len(key) != 16 {
-		intrinsics.trap()
-	}
-
-	skey: [4]u32 = ---
-	skey[0] = endian.unchecked_get_u32le(key[0:])
-	skey[1] = endian.unchecked_get_u32le(key[4:])
-	skey[2] = endian.unchecked_get_u32le(key[8:])
-	skey[3] = endian.unchecked_get_u32le(key[12:])
-
-	q: [8]u64 = ---
-	q[0], q[4] = interleave_in(skey[:])
-	q[1] = q[0]
-	q[2] = q[0]
-	q[3] = q[0]
-	q[5] = q[4]
-	q[6] = q[4]
-	q[7] = q[4]
-	orthogonalize(&q)
-
-	comp_skey: [2]u64 = ---
-	comp_skey[0] =
-		(q[0] & 0x1111111111111111) |
-		(q[1] & 0x2222222222222222) |
-		(q[2] & 0x4444444444444444) |
-		(q[3] & 0x8888888888888888)
-	comp_skey[1] =
-		(q[4] & 0x1111111111111111) |
-		(q[5] & 0x2222222222222222) |
-		(q[6] & 0x4444444444444444) |
-		(q[7] & 0x8888888888888888)
-
-	for x, u in comp_skey {
-		x0 := x
-		x1, x2, x3 := x0, x0, x0
-		x0 &= 0x1111111111111111
-		x1 &= 0x2222222222222222
-		x2 &= 0x4444444444444444
-		x3 &= 0x8888888888888888
-		x1 >>= 1
-		x2 >>= 2
-		x3 >>= 3
-		qq[u * 4 + 0] = (x0 << 4) - x0
-		qq[u * 4 + 1] = (x1 << 4) - x1
-		qq[u * 4 + 2] = (x2 << 4) - x2
-		qq[u * 4 + 3] = (x3 << 4) - x3
-	}
-
-	mem.zero_explicit(&skey, size_of(skey))
-	mem.zero_explicit(&q, size_of(q))
-	mem.zero_explicit(&comp_skey, size_of(comp_skey))
-}

core/crypto/_aes/ct64/ghash.odin

@@ -22,7 +22,6 @@
 
 package aes_ct64
 
-import "base:intrinsics"
 import "core:crypto/_aes"
 import "core:encoding/endian"
 
@@ -64,9 +63,8 @@ rev64 :: proc "contextless" (x: u64) -> u64 {
 // Note: `dst` is both an input and an output, to support easy implementation
 // of GCM.
 ghash :: proc "contextless" (dst, key, data: []byte) {
-	if len(dst) != _aes.GHASH_BLOCK_SIZE || len(key) != _aes.GHASH_BLOCK_SIZE {
-		intrinsics.trap()
-	}
+	ensure_contextless(len(dst) == _aes.GHASH_BLOCK_SIZE)
+	ensure_contextless(len(key) == _aes.GHASH_BLOCK_SIZE)
 
 	buf := data
 	l := len(buf)

core/crypto/_aes/ct64/helpers.odin

@@ -1,75 +1,69 @@
 package aes_ct64
 
-import "base:intrinsics"
 import "core:crypto/_aes"
 import "core:encoding/endian"
 
+@(require_results)
+load_interleaved :: proc "contextless" (src: []byte) -> (u64, u64) #no_bounds_check {
+	w0 := endian.unchecked_get_u32le(src[0:])
+	w1 := endian.unchecked_get_u32le(src[4:])
+	w2 := endian.unchecked_get_u32le(src[8:])
+	w3 := endian.unchecked_get_u32le(src[12:])
+	return interleave_in(w0, w1, w2, w3)
+}
+
+store_interleaved :: proc "contextless" (dst: []byte, a0, a1: u64) #no_bounds_check {
+	w0, w1, w2, w3 := interleave_out(a0, a1)
+	endian.unchecked_put_u32le(dst[0:], w0)
+	endian.unchecked_put_u32le(dst[4:], w1)
+	endian.unchecked_put_u32le(dst[8:], w2)
+	endian.unchecked_put_u32le(dst[12:], w3)
+}
+
+@(require_results)
+xor_interleaved :: #force_inline proc "contextless" (a0, a1, b0, b1: u64) -> (u64, u64) {
+	return a0 ~ b0, a1 ~ b1
+}
+
+@(require_results)
+and_interleaved :: #force_inline proc "contextless" (a0, a1, b0, b1: u64) -> (u64, u64) {
+	return a0 & b0, a1 & b1
+}
+
 load_blockx1 :: proc "contextless" (q: ^[8]u64, src: []byte) {
-	if len(src) != _aes.BLOCK_SIZE {
-		intrinsics.trap()
-	}
+	ensure_contextless(len(src) == _aes.BLOCK_SIZE, "aes/ct64: invalid block size")
 
-	w: [4]u32 = ---
-	w[0] = endian.unchecked_get_u32le(src[0:])
-	w[1] = endian.unchecked_get_u32le(src[4:])
-	w[2] = endian.unchecked_get_u32le(src[8:])
-	w[3] = endian.unchecked_get_u32le(src[12:])
-	q[0], q[4] = interleave_in(w[:])
+	q[0], q[4] = #force_inline load_interleaved(src)
 	orthogonalize(q)
 }
 
 store_blockx1 :: proc "contextless" (dst: []byte, q: ^[8]u64) {
-	if len(dst) != _aes.BLOCK_SIZE {
-		intrinsics.trap()
-	}
+	ensure_contextless(len(dst) == _aes.BLOCK_SIZE, "aes/ct64: invalid block size")
 
 	orthogonalize(q)
-	w0, w1, w2, w3 := interleave_out(q[0], q[4])
-	endian.unchecked_put_u32le(dst[0:], w0)
-	endian.unchecked_put_u32le(dst[4:], w1)
-	endian.unchecked_put_u32le(dst[8:], w2)
-	endian.unchecked_put_u32le(dst[12:], w3)
+	#force_inline store_interleaved(dst, q[0], q[4])
 }
 
 load_blocks :: proc "contextless" (q: ^[8]u64, src: [][]byte) {
-	if n := len(src); n > STRIDE || n == 0 {
-		intrinsics.trap()
-	}
+	ensure_contextless(len(src) == 0 || len(src) <= STRIDE, "aes/ct64: invalid block(s) size")
 
-	w: [4]u32 = ---
 	for s, i in src {
-		if len(s) != _aes.BLOCK_SIZE {
-			intrinsics.trap()
-		}
-
-		w[0] = endian.unchecked_get_u32le(s[0:])
-		w[1] = endian.unchecked_get_u32le(s[4:])
-		w[2] = endian.unchecked_get_u32le(s[8:])
-		w[3] = endian.unchecked_get_u32le(s[12:])
-		q[i], q[i + 4] = interleave_in(w[:])
+		ensure_contextless(len(s) == _aes.BLOCK_SIZE, "aes/ct64: invalid block size")
+		q[i], q[i + 4] = #force_inline load_interleaved(s)
 	}
 	orthogonalize(q)
 }
 
 store_blocks :: proc "contextless" (dst: [][]byte, q: ^[8]u64) {
-	if n := len(dst); n > STRIDE || n == 0 {
-		intrinsics.trap()
-	}
+	ensure_contextless(len(dst) == 0 || len(dst) <= STRIDE, "aes/ct64: invalid block(s) size")
 
 	orthogonalize(q)
 	for d, i in dst {
 		// Allow storing [0,4] blocks.
 		if d == nil {
 			break
 		}
-		if len(d) != _aes.BLOCK_SIZE {
-			intrinsics.trap()
-		}
-
-		w0, w1, w2, w3 := interleave_out(q[i], q[i + 4])
-		endian.unchecked_put_u32le(d[0:], w0)
-		endian.unchecked_put_u32le(d[4:], w1)
-		endian.unchecked_put_u32le(d[8:], w2)
-		endian.unchecked_put_u32le(d[12:], w3)
+		ensure_contextless(len(d) == _aes.BLOCK_SIZE, "aes/ct64: invalid block size")
+		#force_inline store_interleaved(d, q[i], q[i + 4])
 	}
 }

core/crypto/_aes/hw_intel/ghash.odin

@@ -52,7 +52,7 @@ GHASH_STRIDE_BYTES_HW :: GHASH_STRIDE_HW * _aes.GHASH_BLOCK_SIZE
 // that it is right-shifted by 1 bit. The left-shift is relatively
 // inexpensive, and it can be mutualised.
 //
-// Since SSE2 opcodes do not have facilities for shitfting full 128-bit
+// Since SSE2 opcodes do not have facilities for shifting full 128-bit
 // values with bit precision, we have to break down values into 64-bit
 // chunks. We number chunks from 0 to 3 in left to right order.
 
@@ -155,7 +155,7 @@ square_f128 :: #force_inline proc "contextless" (kw: x86.__m128i) -> (x86.__m128
 @(enable_target_feature = "sse2,ssse3,pclmul")
 ghash :: proc "contextless" (dst, key, data: []byte) #no_bounds_check {
 	if len(dst) != _aes.GHASH_BLOCK_SIZE || len(key) != _aes.GHASH_BLOCK_SIZE {
-		intrinsics.trap()
+		panic_contextless("aes/ghash: invalid dst or key size")
 	}
 
 	// Note: BearSSL opts to copy the remainder into a zero-filled

core/crypto/_blake2/blake2.odin

@@ -18,6 +18,8 @@ BLAKE2S_SIZE :: 32
 BLAKE2B_BLOCK_SIZE :: 128
 BLAKE2B_SIZE :: 64
 
+MAX_SIZE :: 255
+
 Blake2s_Context :: struct {
 	h:            [8]u32,
 	t:            [2]u32,
@@ -68,30 +70,27 @@ Blake2_Tree :: struct {
 	is_last_node:    bool,
 }
 
-@(private)
+@(private, rodata)
 BLAKE2S_IV := [8]u32 {
 	0x6a09e667, 0xbb67ae85, 0x3c6ef372, 0xa54ff53a,
 	0x510e527f, 0x9b05688c, 0x1f83d9ab, 0x5be0cd19,
 }
 
-@(private)
+@(private, rodata)
 BLAKE2B_IV := [8]u64 {
 	0x6a09e667f3bcc908, 0xbb67ae8584caa73b,
 	0x3c6ef372fe94f82b, 0xa54ff53a5f1d36f1,
 	0x510e527fade682d1, 0x9b05688c2b3e6c1f,
 	0x1f83d9abfb41bd6b, 0x5be0cd19137e2179,
 }
 
-init :: proc(ctx: ^$T, cfg: ^Blake2_Config) {
+init :: proc "contextless" (ctx: ^$T, cfg: ^Blake2_Config) {
 	when T == Blake2s_Context {
 		max_size :: BLAKE2S_SIZE
 	} else when T == Blake2b_Context {
 		max_size :: BLAKE2B_SIZE
 	}
-
-	if cfg.size > max_size {
-		panic("blake2: requested output size exceeeds algorithm max")
-	}
+	ensure_contextless(cfg.size <= max_size, "blake2: requested output size exceeeds algorithm max")
 
 	// To save having to allocate a scratch buffer, use the internal
 	// data buffer (`ctx.x`), as it is exactly the correct size.
@@ -167,8 +166,8 @@ init :: proc(ctx: ^$T, cfg: ^Blake2_Config) {
 	ctx.is_initialized = true
 }
 
-update :: proc(ctx: ^$T, p: []byte) {
-	assert(ctx.is_initialized)
+update :: proc "contextless" (ctx: ^$T, p: []byte) {
+	ensure_contextless(ctx.is_initialized)
 
 	p := p
 	when T == Blake2s_Context {
@@ -195,8 +194,8 @@ update :: proc(ctx: ^$T, p: []byte) {
 	ctx.nx += copy(ctx.x[ctx.nx:], p)
 }
 
-final :: proc(ctx: ^$T, hash: []byte, finalize_clone: bool = false) {
-	assert(ctx.is_initialized)
+final :: proc "contextless" (ctx: ^$T, hash: []byte, finalize_clone: bool = false) {
+	ensure_contextless(ctx.is_initialized)
 
 	ctx := ctx
 	if finalize_clone {
@@ -206,24 +205,19 @@ final :: proc(ctx: ^$T, hash: []byte, finalize_clone: bool = false) {
 	}
 	defer(reset(ctx))
 
+	ensure_contextless(len(hash) >= int(ctx.size), "crypto/blake2: invalid destination digest size")
 	when T == Blake2s_Context {
-		if len(hash) < int(ctx.size) {
-			panic("crypto/blake2s: invalid destination digest size")
-		}
 		blake2s_final(ctx, hash)
 	} else when T == Blake2b_Context {
-		if len(hash) < int(ctx.size) {
-			panic("crypto/blake2b: invalid destination digest size")
-		}
 		blake2b_final(ctx, hash)
 	}
 }
 
-clone :: proc(ctx, other: ^$T) {
+clone :: proc "contextless" (ctx, other: ^$T) {
 	ctx^ = other^
 }
 
-reset :: proc(ctx: ^$T) {
+reset :: proc "contextless" (ctx: ^$T) {
 	if !ctx.is_initialized {
 		return
 	}