crypto/elliptic: reduce allocations in s390x P256 code

The Go compiler assumes that pointers escape when passed into assembly
functions. To override this behavior we can annotate assembly functions
with go:noescape, telling the compiler that we know pointers do not
escape from it.

By annotating the assembly functions in the s390x P256 code in this way
we enable more variables to be allocated on the stack rather than
the heap, reducing the number of heap allocations required to execute
this code:

name        old alloc/op   new alloc/op   delta
SignP256      3.66kB ± 0%    2.64kB ± 0%  -27.95%  (p=0.008 n=5+5)
VerifyP256    4.46kB ± 0%    1.23kB ± 0%  -72.40%  (p=0.008 n=5+5)

name        old allocs/op  new allocs/op  delta
SignP256        40.0 ± 0%      31.0 ± 0%  -22.50%  (p=0.008 n=5+5)
VerifyP256      41.0 ± 0%      24.0 ± 0%  -41.46%  (p=0.008 n=5+5)

Change-Id: Id526c30c9b04b2ad79a55d76cab0e30cc8d60402
Reviewed-on: https://go-review.googlesource.com/66230
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>

diff --git a/src/crypto/elliptic/p256_s390x.go b/src/crypto/elliptic/p256_s390x.go
index 44c0f41f072ed41782ac57879b4c7064c7c0f0ff..a36786e466d9f500d64f16be401a18408ce12556 100644 (file)
--- a/src/crypto/elliptic/p256_s390x.go
+++ b/src/crypto/elliptic/p256_s390x.go
@@ -50,6 +50,8 @@ func (curve p256CurveFast) Params() *CurveParams {
 
 // Functions implemented in p256_asm_s390x.s
 // Montgomery multiplication modulo P256
+//
+//go:noescape
 func p256MulAsm(res, in1, in2 []byte)
 
 // Montgomery square modulo P256
@@ -58,19 +60,31 @@ func p256Sqr(res, in []byte) {
 }
 
 // Montgomery multiplication by 1
+//
+//go:noescape
 func p256FromMont(res, in []byte)
 
 // iff cond == 1  val <- -val
+//
+//go:noescape
 func p256NegCond(val *p256Point, cond int)
 
 // if cond == 0 res <- b; else res <- a
+//
+//go:noescape
 func p256MovCond(res, a, b *p256Point, cond int)
 
 // Constant time table access
+//
+//go:noescape
 func p256Select(point *p256Point, table []p256Point, idx int)
+
+//go:noescape
 func p256SelectBase(point *p256Point, table []p256Point, idx int)
 
 // Montgomery multiplication modulo Ord(G)
+//
+//go:noescape
 func p256OrdMul(res, in1, in2 []byte)
 
 // Montgomery square modulo Ord(G), repeated n times
@@ -85,10 +99,16 @@ func p256OrdSqr(res, in []byte, n int) {
 // If sign == 1 -> P2 = -P2
 // If sel == 0 -> P3 = P1
 // if zero == 0 -> P3 = P2
+//
+//go:noescape
 func p256PointAddAffineAsm(P3, P1, P2 *p256Point, sign, sel, zero int)
 
 // Point add
+//
+//go:noescape
 func p256PointAddAsm(P3, P1, P2 *p256Point) int
+
+//go:noescape
 func p256PointDoubleAsm(P3, P1 *p256Point)
 
 func (curve p256CurveFast) Inverse(k *big.Int) *big.Int {