cipher: xts ppc64x

cipher/xts_asm.go

@@ -1,4 +1,4 @@
-//go:build (amd64 || arm64 || s390x) && !purego
+//go:build (amd64 || arm64 || s390x || ppc64 || ppc64le) && !purego
 
 package cipher
 

cipher/xts_asm_test.go

@@ -1,4 +1,4 @@
-//go:build (amd64 || arm64 || s390x) && !purego
+//go:build (amd64 || arm64 || s390x || ppc64 || ppc64le) && !purego
 
 package cipher
 

cipher/xts_generic.go

@@ -1,4 +1,4 @@
-//go:build purego || !(amd64 || arm64 || s390x)
+//go:build purego || !(amd64 || arm64 || s390x || ppc64 || ppc64le)
 
 package cipher
 

cipher/xts_ppc64x.s

@@ -0,0 +1,166 @@
+// Copyright 2024 Sun Yimin. All rights reserved.
+// Use of this source code is governed by a MIT-style
+// license that can be found in the LICENSE file.
+
+//go:build (ppc64 || ppc64le) && !purego
+
+#include "textflag.h"
+
+DATA xtsMask<>+0x00(SB)/8, $0x0f0e0d0c0b0a0908 // byte swap BE -> LE
+DATA xtsMask<>+0x08(SB)/8, $0x0706050403020100
+DATA xtsMask<>+0x10(SB)/8, $0x0000000000000000
+DATA xtsMask<>+0x18(SB)/8, $0x0000000000000087
+DATA xtsMask<>+0x20(SB)/8, $0xe100000000000000
+DATA xtsMask<>+0x28(SB)/8, $0x0000000000000000
+GLOBL xtsMask<>(SB), (NOPTR+RODATA), $48
+
+#define ESPERM  V21  // Endian swapping permute into BE
+
+#define POLY V0
+#define B0 V1
+#define T0 V2
+#define T1 V3
+#define CPOOL R7
+
+#define doubleTweak(B0, POLY, T0, T1) \
+	\ // Multiply by 2
+	VSPLTB $0, B0, T0    \
+	VSPLTISB $7, T1      \
+	VSRAB    T0, T1, T0  \
+	VAND    POLY, T0, T0 \// T0 for reduction
+	\
+	VSPLTISB $1, T1      \
+	VSL B0, T1, T1       \
+	VXOR T0, T1, B0
+
+#define gbDoubleTweak(B0, POLY, T0, T1) \
+	VSPLTB $15, B0, T0   \
+	VSPLTISB $7, T1      \
+	VSLB T0, T1, T0      \
+	VSRAB T0, T1, T0     \
+	VAND POLY, T0, T0    \ // T0 for reduction
+	VSPLTISB $1, T1      \
+	VSR B0, T1, B0       \
+	VXOR T0, B0, B0
+
+// func mul2(tweak *[blockSize]byte, isGB bool)
+TEXT ·mul2(SB),NOSPLIT,$0
+	MOVD tweak+0(FP), R3
+	MOVBZ isGB+8(FP), R4
+
+	MOVD $xtsMask<>(SB), CPOOL
+
+	CMPB R4, $1
+	BEQ gb_alg
+	
+	// Load polynomial for reduction
+	MOVD $16, R5
+	LXVD2X (CPOOL)(R5), POLY
+
+	// Load tweak
+	LXVD2X (R3), B0
+#ifdef GOARCH_ppc64le
+	XXPERMDI B0, B0, $2, B0
+	doubleTweak(B0, POLY, T0, T1)
+	XXPERMDI B0, B0, $2, B0
+#else
+	LXVD2X (CPOOL), ESPERM
+	
+	VPERM B0, B0, ESPERM, B0
+	doubleTweak(B0, POLY, T0, T1)
+	VPERM B0, B0, ESPERM, B0
+#endif
+	STXVD2X B0, (R3)
+
+	RET
+
+gb_alg:	
+	// Load polynomial for reduction
+	MOVD $32, R5
+	LXVD2X (CPOOL)(R5), POLY
+
+	// Load tweak
+	LXVD2X (R3), B0
+#ifdef GOARCH_ppc64le
+	LVX (CPOOL), ESPERM
+	VPERM B0, B0, ESPERM, B0
+	gbDoubleTweak(B0, POLY, T0, T1)
+	VPERM B0, B0, ESPERM, B0
+#else
+	gbDoubleTweak(B0, POLY, T0, T1)
+#endif
+	STXVD2X B0, (R3)
+	RET
+
+// func doubleTweaks(tweak *[blockSize]byte, tweaks []byte, isGB bool)
+TEXT ·doubleTweaks(SB),NOSPLIT,$0
+	MOVD tweak+0(FP), R3
+	MOVD tweaks+8(FP), R4
+	MOVD tweaks_len+16(FP), R5
+	MOVBZ isGB+32(FP), R6
+
+	MOVD $xtsMask<>(SB), CPOOL
+
+	// Load tweak
+	LXVD2X (R3), B0
+
+	CMPB R6, $1
+	BEQ gb_alg
+
+	SRD	$4, R5
+	MOVD R5, CTR
+
+#ifndef GOARCH_ppc64le
+	LXVD2X (CPOOL), ESPERM
+#endif	
+	// Load polynomial for reduction
+	MOVD $16, R5
+	LXVD2X (CPOOL)(R5), POLY
+
+loop:
+		STXVD2X B0, (R4)
+		ADD $16, R4
+
+#ifdef GOARCH_ppc64le
+		XXPERMDI B0, B0, $2, B0
+		doubleTweak(B0, POLY, T0, T1)
+		XXPERMDI B0, B0, $2, B0
+#else
+		VPERM B0, B0, ESPERM, B0
+		doubleTweak(B0, POLY, T0, T1)
+		VPERM B0, B0, ESPERM, B0
+#endif
+
+		BDNZ	loop
+
+	STXVD2X B0, (R3)
+	RET
+
+gb_alg:	
+	SRD	$4, R5
+	MOVD R5, CTR
+
+	// Load polynomial for reduction
+	MOVD $32, R5
+	LXVD2X (CPOOL)(R5), POLY
+
+#ifdef GOARCH_ppc64le
+	LVX (CPOOL), ESPERM
+#endif
+
+gbLoop:
+	STXVD2X B0, (R4)
+	ADD $16, R4
+
+#ifdef GOARCH_ppc64le
+		VPERM B0, B0, ESPERM, B0
+		gbDoubleTweak(B0, POLY, T0, T1)
+		VPERM B0, B0, ESPERM, B0
+#else
+		gbDoubleTweak(B0, POLY, T0, T1)
+#endif
+
+		BDNZ	gbLoop
+
+	STXVD2X B0, (R3)
+	RET