From: Lynn Boger <laboger@linux.vnet.ibm.com>
Date: Mon, 28 Oct 2019 18:40:27 +0000 (-0400)
Subject: crypto/elliptic: clean up ppc64le implementation slightly
X-Git-Tag: go1.14beta1~520
X-Git-Url: http://www.git.cypherpunks.su/?a=commitdiff_plain;h=ac346a5fd99711e94996cae030cad1064da1daf2;p=gostls13.git

crypto/elliptic: clean up ppc64le implementation slightly

As suggested by comments from the review of CL 168478, this adds
Go code to do reverse bytes and removes the asm code, as well
as making a few cosmetic changes.

Change-Id: I08276a11222e03c3b42f4c9dc0d10a371a418be7
Reviewed-on: https://go-review.googlesource.com/c/go/+/203937
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Michael Munday <mike.munday@ibm.com>
---

diff --git a/src/crypto/elliptic/p256_asm_ppc64le.s b/src/crypto/elliptic/p256_asm_ppc64le.s
index 4428a18260..924e365c6c 100644
--- a/src/crypto/elliptic/p256_asm_ppc64le.s
+++ b/src/crypto/elliptic/p256_asm_ppc64le.s
@@ -105,10 +105,10 @@ GLOBL byteswap<>+0(SB), RODATA, $16
 #define VMULT_LOW(x1, x2, out_low) \
 	VMULUWM x1, x2, out_low
 
-	//
-	// Vector multiply high word
-	//
-	//	VMLHF x0, x1, out_hi
+//
+// Vector multiply high word
+//
+//	VMLHF x0, x1, out_hi
 #define VMULT_HI(x1, x2, out_hi) \
 	VMULEUW x1, x2, TMP1;                  \
 	VMULOUW x1, x2, TMP2;                  \
@@ -167,27 +167,6 @@ GLOBL byteswap<>+0(SB), RODATA, $16
 #define res_ptr R3
 #define a_ptr R4
 
-// func p256ReverseBytes(res, in []byte)
-// Reuse of target and destination OK
-TEXT ·p256ReverseBytes(SB), NOSPLIT, $0-48
-	MOVD res+0(FP), res_ptr
-	MOVD in+24(FP), a_ptr
-
-	MOVD $8, R5
-	MOVD $16, R6
-	MOVD $24, R7
-
-	MOVDBR (R0+a_ptr), R8
-	MOVDBR (R5+a_ptr), R9
-	MOVDBR (R6+a_ptr), R10
-	MOVDBR (R7+a_ptr), R11
-
-	MOVD R11, (R0+res_ptr)
-	MOVD R10, (R5+res_ptr)
-	MOVD R9, (R6+res_ptr)
-	MOVD R8, (R7+res_ptr)
-	RET
-
 #undef res_ptr
 #undef a_ptr
 
@@ -1624,8 +1603,6 @@ TEXT ·p256PointAddAffineAsm(SB), NOSPLIT, $16-48
 	VPERM  X1, X1, SWAP, X1
 	CALL   p256MulInternal<>(SB)
 
-	// VST T1, 64(P3ptr)
-	// VST T0, 80(P3ptr)
 	VOR T0, T0, Z3L
 	VOR T1, T1, Z3H
 
diff --git a/src/crypto/elliptic/p256_ppc64le.go b/src/crypto/elliptic/p256_ppc64le.go
index 4b41fb99af..160bdb12e3 100644
--- a/src/crypto/elliptic/p256_ppc64le.go
+++ b/src/crypto/elliptic/p256_ppc64le.go
@@ -8,6 +8,7 @@ package elliptic
 
 import (
 	"crypto/subtle"
+	"encoding/binary"
 	"math/big"
 )
 
@@ -77,10 +78,6 @@ func p256Select(point *p256Point, table []p256Point, idx int)
 //go:noescape
 func p256SelectBase(point *p256Point, table []p256Point, idx int)
 
-// Reverse the bytes (endianness)
-//go:noescape
-func p256ReverseBytes(res, in []byte)
-
 // Point add with P2 being affine point
 // If sign == 1 -> P2 = -P2
 // If sel == 0 -> P3 = P1
@@ -145,6 +142,25 @@ func maybeReduceModP(in *big.Int) *big.Int {
 	return new(big.Int).Mod(in, p256Params.P)
 }
 
+// p256ReverseBytes copies the first 32 bytes from in to res in reverse order.
+func p256ReverseBytes(res, in []byte) {
+	// remove bounds check
+	in = in[:32]
+	res = res[:32]
+
+	// Load in reverse order
+	a := binary.BigEndian.Uint64(in[0:])
+	b := binary.BigEndian.Uint64(in[8:])
+	c := binary.BigEndian.Uint64(in[16:])
+	d := binary.BigEndian.Uint64(in[24:])
+
+	// Store in normal order
+	binary.LittleEndian.PutUint64(res[0:], d)
+	binary.LittleEndian.PutUint64(res[8:], c)
+	binary.LittleEndian.PutUint64(res[16:], b)
+	binary.LittleEndian.PutUint64(res[24:], a)
+}
+
 func (curve p256CurveFast) CombinedMult(bigX, bigY *big.Int, baseScalar, scalar []byte) (x, y *big.Int) {
 	var r1, r2 p256Point