Merge pull request #17 from rakitzis/arm-asm-fixes

fixes for issue #15 -- ARM assembly alignment bug

blocks.go

@@ -1,4 +1,4 @@
-// +build !amd64 appengine gccgo
+// +build !arm,!amd64 appengine gccgo
 
 package siphash
 

blocks_arm.s

@@ -0,0 +1,144 @@
+#include "textflag.h"
+
+#define ROUND()\
+	ADD.S	R2,R0,R0;\
+	ADC	R3,R1,R1;\
+	EOR	R2<<13,R0,R8;\
+	EOR	R3>>19,R8,R8;\
+	EOR	R2>>19,R1,R11;\
+	EOR	R3<<13,R11,R11;\
+	ADD.S	R6,R4,R4;\
+	ADC	R7,R5,R5;\
+	EOR	R6<<16,R4,R2;\
+	EOR	R7>>16,R2,R2;\
+	EOR	R6>>16,R5,R3;\
+	EOR	R7<<16,R3,R3;\
+	ADD.S	R2,R1,R1;\
+	ADC	R3,R0,R0;\
+	EOR	R2<<21,R1,R6;\
+	EOR	R3>>11,R6,R6;\
+	EOR	R2>>11,R0,R7;\
+	EOR	R3<<21,R7,R7;\
+	ADD.S	R8,R4,R4;\
+	ADC	R11,R5,R5;\
+	EOR	R8<<17,R4,R2;\
+	EOR	R11>>15,R2,R2;\
+	EOR	R8>>15,R5,R3;\
+	EOR	R11<<17,R3,R3;\
+	ADD.S	R2,R1,R1;\
+	ADC	R3,R0,R0;\
+	EOR	R2<<13,R1,R8;\
+	EOR	R3>>19,R8,R8;\
+	EOR	R2>>19,R0,R11;\
+	EOR	R3<<13,R11,R11;\
+	ADD.S	R6,R5,R5;\
+	ADC	R7,R4,R4;\
+	EOR	R6<<16,R5,R2;\
+	EOR	R7>>16,R2,R2;\
+	EOR	R6>>16,R4,R3;\
+	EOR	R7<<16,R3,R3;\
+	ADD.S	R2,R0,R0;\
+	ADC	R3,R1,R1;\
+	EOR	R2<<21,R0,R6;\
+	EOR	R3>>11,R6,R6;\
+	EOR	R2>>11,R1,R7;\
+	EOR	R3<<21,R7,R7;\
+	ADD.S	R8,R5,R5;\
+	ADC	R11,R4,R4;\
+	EOR	R8<<17,R5,R2;\
+	EOR	R11>>15,R2,R2;\
+	EOR	R8>>15,R4,R3;\
+	EOR	R11<<17,R3,R3;
+
+// once(d *digest)
+TEXT ·once(SB),NOSPLIT,$4-4
+	MOVW	d+0(FP),R8
+	MOVM.IA	(R8),[R0,R1,R2,R3,R4,R5,R6,R7]
+	MOVW	48(R8),R12
+	MOVW	52(R8),R14
+	EOR	R12,R6,R6
+	EOR	R14,R7,R7
+	ROUND()
+	EOR	R12,R0,R0
+	EOR	R14,R1,R1
+	MOVW	d+0(FP),R8
+	MOVM.IA [R0,R1,R2,R3,R4,R5,R6,R7],(R8)
+	RET
+
+// finalize(d *digest) uint64
+TEXT ·finalize(SB),NOSPLIT,$4-12
+	MOVW	d+0(FP),R8
+	MOVM.IA	(R8),[R0,R1,R2,R3,R4,R5,R6,R7]
+	MOVW	48(R8),R12
+	MOVW	52(R8),R14
+	EOR	R12,R6,R6
+	EOR	R14,R7,R7
+	ROUND()
+	EOR	R12,R0,R0
+	EOR	R14,R1,R1
+	EOR	$255,R4
+	ROUND()
+	ROUND()
+	EOR	R2,R0,R0
+	EOR	R3,R1,R1
+	EOR	R6,R4,R4
+	EOR	R7,R5,R5
+	EOR	R4,R0,R0
+	EOR	R5,R1,R1
+	MOVW	R0,ret_lo+4(FP)
+	MOVW	R1,ret_hi+8(FP)
+	RET
+
+// blocks(d *digest, data []uint8)
+TEXT ·blocks(SB),NOSPLIT,$8-16
+	MOVW	R9,sav-8(SP)
+	MOVW	d+0(FP),R8
+	MOVM.IA	(R8),[R0,R1,R2,R3,R4,R5,R6,R7]
+	MOVW	p+4(FP),R9
+	MOVW	p_len+8(FP),R11
+	ADD	R9,R11,R11
+	MOVW	R11,endp-4(SP)
+	AND.S   $3,R9,R8
+	BNE     blocksunaligned
+blocksloop:
+	MOVM.IA.W (R9),[R12,R14]
+	EOR	R12,R6,R6
+	EOR	R14,R7,R7
+	ROUND()
+	EOR	R12,R0,R0
+	EOR	R14,R1,R1
+	MOVW	endp-4(SP),R11
+	CMP	R11,R9
+	BLO	blocksloop
+	MOVW	d+0(FP),R8
+	MOVM.IA [R0,R1,R2,R3,R4,R5,R6,R7],(R8)
+	MOVW	sav-8(SP),R9
+	RET
+blocksunaligned:
+	MOVBU    (R9),R12
+	MOVBU    1(R9),R11
+	ORR     R11<<8,R12,R12
+	MOVBU    2(R9),R11
+	ORR     R11<<16,R12,R12
+	MOVBU    3(R9),R11
+	ORR     R11<<24,R12,R12
+	MOVBU    4(R9),R14
+	MOVBU    5(R9),R11
+	ORR     R11<<8,R14,R14
+	MOVBU    6(R9),R11
+	ORR     R11<<16,R14,R14
+	MOVBU    7(R9),R11
+	ORR     R11<<24,R14,R14
+	ADD     $8,R9,R9
+	EOR     R12,R6,R6
+	EOR     R14,R7,R7
+	ROUND()
+	EOR     R12,R0,R0
+	EOR     R14,R1,R1
+	MOVW    endp-4(SP),R11
+	CMP     R11,R9
+	BLO     blocksunaligned
+	MOVW    d+0(FP),R8
+	MOVM.IA [R0,R1,R2,R3,R4,R5,R6,R7],(R8)
+	MOVW    sav-8(SP),R9
+	RET

blocks_asm.go

@@ -0,0 +1,21 @@
+// +build arm amd64,!appengine,!gccgo
+
+// Written in 2012 by Dmitry Chestnykh.
+//
+// To the extent possible under law, the author have dedicated all copyright
+// and related and neighboring rights to this software to the public domain
+// worldwide. This software is distributed without any warranty.
+// http://creativecommons.org/publicdomain/zero/1.0/
+
+// This file contains a function definition for use with assembly implementations of Hash()
+
+package siphash
+
+//go:noescape
+func blocks(d *digest, p []uint8)
+
+//go:noescape
+func finalize(d *digest) uint64
+
+//go:noescape
+func once(d *digest)

hash.go

@@ -1,4 +1,4 @@
-// +build !amd64 appengine gccgo
+// +build !arm,!amd64 appengine gccgo
 
 // Written in 2012 by Dmitry Chestnykh.
 //

hash128.go

@@ -1,4 +1,4 @@
-// +build !amd64 appengine gccgo
+// +build !arm,!amd64 appengine gccgo
 // Written in 2012 by Dmitry Chestnykh.
 // Modifications 2014 for 128-bit hash function by Damian Gryski.
 //

hash_arm.go

@@ -0,0 +1,27 @@
+// +build arm
+
+package siphash
+
+// NB: ARM implementation of forgoes extra speed for Hash()
+// and Hash128() by simply reusing the same blocks() implementation
+// in assembly used by the streaming hash.
+
+func Hash(k0, k1 uint64, p []byte) uint64 {
+	var d digest
+	d.size = Size
+	d.k0 = k0
+	d.k1 = k1
+	d.Reset()
+	d.Write(p)
+	return d.Sum64()
+}
+
+func Hash128(k0, k1 uint64, p []byte) (uint64, uint64) {
+	var d digest
+	d.size = Size128
+	d.k0 = k0
+	d.k1 = k1
+	d.Reset()
+	d.Write(p)
+	return d.sum128()
+}

hash_asm.go

@@ -22,12 +22,3 @@ func Hash(k0, k1 uint64, b []byte) uint64
 // Hash128 returns the 128-bit SipHash-2-4 of the given byte slice with two
 // 64-bit parts of 128-bit key: k0 and k1.
 func Hash128(k0, k1 uint64, b []byte) (uint64, uint64)
-
-//go:noescape
-func blocks(d *digest, p []uint8)
-
-//go:noescape
-func finalize(d *digest) uint64
-
-//go:noescape
-func once(d *digest)

siphash_test.go

@@ -347,16 +347,60 @@ func TestAlign(t *testing.T) {
 	data := "0076a9143219adce9b6f0a21fd53cb17e2fd9b2b4fac40b388ac"
 	k0 := uint64(316665572293978160)
 	k1 := uint64(8573005253291875333)
-	want := uint64(16770526497674945769)
+
+	want := []uint64{
+		16380727507974277821,
+		16770526497674945769,
+		11373998677292870540,
+		10374222295991299613,
+	}
+	want128 := []uint64{
+		14802151199638645495,
+		13251497035884452880,
+		7034723853391616289,
+		16742813562040528752,
+		10468120447644272532,
+		10941274532208162335,
+		11293904790559355408,
+		15432350433573653068,
+	}
 
 	d, err := hex.DecodeString(data)
 	if err != nil {
 		t.Fatal(err)
 	}
 
-	res := Hash(k0, k1, d[1:])
-	if res != want {
-		t.Fatalf("Expected %v got %v", want, res)
+	var k [16]byte
+	binary.LittleEndian.PutUint64(k[0:], k0)
+	binary.LittleEndian.PutUint64(k[8:], k1)
+
+	for i := range want {
+		res := Hash(k0, k1, d[i:])
+		if res != want[i] {
+			t.Fatalf("Expected %v got %v", want[i], res)
+		}
+		reslo, reshi := Hash128(k0, k1, d[i:])
+		if reslo != want128[i*2] {
+			t.Fatalf("Expected %v got %v", want128[i*2], reslo)
+		}
+		if reshi != want128[i*2+1] {
+			t.Fatalf("Expected %v got %v", want128[i*2+1], reshi)
+		}
+		dig := newDigest(Size, k[:])
+		dig.Write(d[i:])
+		res = dig.Sum64()
+		if res != want[i] {
+			t.Fatalf("Expected %v got %v", want[i], res)
+		}
+		dig128 := newDigest(Size128, k[:])
+		dig128.Write(d[i:])
+		reslo, reshi = dig128.sum128()
+		if reslo != want128[i*2] {
+			t.Fatalf("Expected %v got %v", want128[i*2], reslo)
+		}
+		if reshi != want128[i*2+1] {
+			t.Fatalf("Expected %v got %v", want128[i*2+1], reshi)
+		}
 	}
 }