return len(p), nil
}
+func (h *sha1Hash) WriteString(s string) (int, error) {
+ if len(s) > 0 && C._goboringcrypto_SHA1_Update(h.noescapeCtx(), unsafe.Pointer(unsafe.StringData(s)), C.size_t(len(s))) == 0 {
+ panic("boringcrypto: SHA1_Update failed")
+ }
+ return len(s), nil
+}
+
+func (h *sha1Hash) WriteByte(c byte) error {
+ if C._goboringcrypto_SHA1_Update(h.noescapeCtx(), unsafe.Pointer(&c), 1) == 0 {
+ panic("boringcrypto: SHA1_Update failed")
+ }
+ return nil
+}
+
func (h0 *sha1Hash) sum(dst []byte) []byte {
h := *h0 // make copy so future Write+Sum is valid
if C._goboringcrypto_SHA1_Final((*C.uint8_t)(noescape(unsafe.Pointer(&h.out[0]))), h.noescapeCtx()) == 0 {
func (d *digest) BlockSize() int { return BlockSize }
func (d *digest) Write(p []byte) (nn int, err error) {
- boringUnreachable()
nn = len(p)
d.len += uint64(nn)
- if d.nx > 0 {
- n := copy(d.x[d.nx:], p)
- d.nx += n
- if d.nx == chunk {
- block(d, d.x[:])
- d.nx = 0
- }
- p = p[n:]
- }
+ n := fillChunk(d, p)
+ p = p[n:]
if len(p) >= chunk {
n := len(p) &^ (chunk - 1)
block(d, p[:n])
return
}
+func (d *digest) WriteString(s string) (nn int, err error) {
+ nn = len(s)
+ d.len += uint64(nn)
+ n := fillChunk(d, s)
+
+ // This duplicates the code in Write, except that it calls
+ // blockString rather than block. It would be nicer to pass
+ // in a func, but as of this writing (Go 1.20) that causes
+ // memory allocations that we want to avoid.
+
+ s = s[n:]
+ if len(s) >= chunk {
+ n := len(s) &^ (chunk - 1)
+ blockString(d, s[:n])
+ s = s[n:]
+ }
+ if len(s) > 0 {
+ d.nx = copy(d.x[:], s)
+ }
+ return
+}
+
+// fillChunk fills the remainder of the current chunk, if any.
+func fillChunk[S []byte | string](d *digest, p S) int {
+ boringUnreachable()
+ if d.nx == 0 {
+ return 0
+ }
+ n := copy(d.x[d.nx:], p)
+ d.nx += n
+ if d.nx == chunk {
+ block(d, d.x[:])
+ d.nx = 0
+ }
+ return n
+}
+
+func (d *digest) WriteByte(c byte) error {
+ boringUnreachable()
+ d.len++
+ d.x[d.nx] = c
+ d.nx++
+ if d.nx == chunk {
+ block(d, d.x[:])
+ d.nx = 0
+ }
+ return nil
+}
+
func (d *digest) Sum(in []byte) []byte {
boringUnreachable()
// Make a copy of d so that caller can keep writing and summing.
}
c.Reset()
}
+ bw := c.(io.ByteWriter)
+ for i := 0; i < len(g.in); i++ {
+ bw.WriteByte(g.in[i])
+ }
+ s = fmt.Sprintf("%x", c.Sum(nil))
+ if s != g.out {
+ t.Errorf("sha1[WriteByte](%s) = %s want %s", g.in, s, g.out)
+ }
}
}
if boring.Enabled {
t.Skip("BoringCrypto doesn't allocate the same way as stdlib")
}
- in := []byte("hello, world!")
+ const ins = "hello, world!"
+ in := []byte(ins)
out := make([]byte, 0, Size)
h := New()
n := int(testing.AllocsPerRun(10, func() {
if n > 0 {
t.Errorf("allocs = %d, want 0", n)
}
+
+ sw := h.(io.StringWriter)
+ n = int(testing.AllocsPerRun(10, func() {
+ h.Reset()
+ sw.WriteString(ins)
+ out = h.Sum(out[:0])
+ }))
+ if n > 0 {
+ t.Errorf("string allocs = %d, want 0", n)
+ }
+
+ bw := h.(io.ByteWriter)
+ n = int(testing.AllocsPerRun(10, func() {
+ h.Reset()
+ for _, b := range in {
+ bw.WriteByte(b)
+ }
+ out = h.Sum(out[:0])
+ }))
+ if n > 0 {
+ t.Errorf("byte allocs = %d, want 0", n)
+ }
}
var bench = New()
// blockGeneric is a portable, pure Go version of the SHA-1 block step.
// It's used by sha1block_generic.go and tests.
-func blockGeneric(dig *digest, p []byte) {
+func blockGeneric[S []byte | string](dig *digest, p S) {
var w [16]uint32
h0, h1, h2, h3, h4 := dig.h[0], dig.h[1], dig.h[2], dig.h[3], dig.h[4]
FUNC4(a, b, c, d, e); \
MIX(a, b, c, d, e, 0xCA62C1D6)
-// func block(dig *digest, p []byte)
-TEXT ·block(SB),NOSPLIT,$92-16
+// func doBlock(dig *digest, p *byte, n int)
+TEXT ·doBlock(SB),NOSPLIT,$92-12
MOVL dig+0(FP), BP
MOVL p+4(FP), SI
- MOVL p_len+8(FP), DX
+ MOVL n+8(FP), DX
SHRL $6, DX
SHLL $6, DX
package sha1
-import "internal/cpu"
+import (
+ "internal/cpu"
+ "unsafe"
+)
//go:noescape
-func blockAVX2(dig *digest, p []byte)
+func blockAVX2(dig *digest, p *byte, n int)
//go:noescape
-func blockAMD64(dig *digest, p []byte)
+func blockAMD64(dig *digest, p *byte, n int)
var useAVX2 = cpu.X86.HasAVX2 && cpu.X86.HasBMI1 && cpu.X86.HasBMI2
func block(dig *digest, p []byte) {
if useAVX2 && len(p) >= 256 {
- // blockAVX2 calculates sha1 for 2 block per iteration
+ // blockAVX2 calculates sha1 for 2 blocks per iteration
// it also interleaves precalculation for next block.
// So it may read up-to 192 bytes past end of p
// We may add checks inside blockAVX2, but this will
if safeLen%128 != 0 {
safeLen -= 64
}
- blockAVX2(dig, p[:safeLen])
- blockAMD64(dig, p[safeLen:])
+ blockAVX2(dig, unsafe.SliceData(p), safeLen)
+ pRem := p[safeLen:]
+ blockAMD64(dig, unsafe.SliceData(pRem), len(pRem))
} else {
- blockAMD64(dig, p)
+ blockAMD64(dig, unsafe.SliceData(p), len(p))
+ }
+}
+
+// blockString is a duplicate of block that takes a string.
+func blockString(dig *digest, s string) {
+ if useAVX2 && len(s) >= 256 {
+ safeLen := len(s) - 128
+ if safeLen%128 != 0 {
+ safeLen -= 64
+ }
+ blockAVX2(dig, unsafe.StringData(s), safeLen)
+ sRem := s[safeLen:]
+ blockAMD64(dig, unsafe.StringData(sRem), len(sRem))
+ } else {
+ blockAMD64(dig, unsafe.StringData(s), len(s))
}
}
FUNC4(a, b, c, d, e); \
MIX(a, b, c, d, e, 0xCA62C1D6)
-TEXT ·blockAMD64(SB),NOSPLIT,$64-32
+TEXT ·blockAMD64(SB),NOSPLIT,$64-24
MOVQ dig+0(FP), BP
- MOVQ p_base+8(FP), SI
- MOVQ p_len+16(FP), DX
+ MOVQ p+8(FP), SI
+ MOVQ n+16(FP), DX
SHRQ $6, DX
SHLQ $6, DX
-TEXT ·blockAVX2(SB),$1408-32
+TEXT ·blockAVX2(SB),$1408-24
MOVQ dig+0(FP), DI
- MOVQ p_base+8(FP), SI
- MOVQ p_len+16(FP), DX
+ MOVQ p+8(FP), SI
+ MOVQ n+16(FP), DX
SHRQ $6, DX
SHLQ $6, DX
#define Rctr R12 // loop counter
#define Rw R14 // point to w buffer
-// func block(dig *digest, p []byte)
+// func doBlock(dig *digest, p *byte, n int)
// 0(FP) is *digest
// 4(FP) is p.array (struct Slice)
// 8(FP) is p.len
-//12(FP) is p.cap
//
// Stack frame
#define p_end end-4(SP) // pointer to the end of data
MIX(Ra, Rb, Rc, Rd, Re)
-// func block(dig *digest, p []byte)
-TEXT ·block(SB), 0, $352-16
+// func doBlock(dig *digest, p *byte, n int)
+TEXT ·doBlock(SB), 0, $352-12
MOVW p+4(FP), Rdata // pointer to the data
- MOVW p_len+8(FP), Rt0 // number of bytes
+ MOVW n+8(FP), Rt0 // number of bytes
ADD Rdata, Rt0
MOVW Rt0, p_end // pointer to end of data
package sha1
-import "internal/cpu"
+import (
+ "internal/cpu"
+ "unsafe"
+)
var k = []uint32{
0x5A827999,
}
//go:noescape
-func sha1block(h []uint32, p []byte, k []uint32)
+func sha1block(h []uint32, p *byte, n int, k []uint32)
func block(dig *digest, p []byte) {
if !cpu.ARM64.HasSHA1 {
blockGeneric(dig, p)
} else {
h := dig.h[:]
- sha1block(h, p, k)
+ sha1block(h, unsafe.SliceData(p), len(p), k)
+ }
+}
+
+func blockString(dig *digest, s string) {
+ if !cpu.ARM64.HasSHA1 {
+ blockGeneric(dig, s)
+ } else {
+ h := dig.h[:]
+ sha1block(h, unsafe.StringData(s), len(s), k)
}
}
SHA1H V3, V1 \
VMOV V2.B16, V3.B16
-// func sha1block(h []uint32, p []byte, k []uint32)
+// func sha1block(h []uint32, p *byte, n int, k []uint32)
TEXT ·sha1block(SB),NOSPLIT,$0
MOVD h_base+0(FP), R0 // hash value first address
- MOVD p_base+24(FP), R1 // message first address
- MOVD k_base+48(FP), R2 // k constants first address
- MOVD p_len+32(FP), R3 // message length
+ MOVD p+24(FP), R1 // message first address
+ MOVD k_base+40(FP), R2 // k constants first address
+ MOVD n+32(FP), R3 // message length
VLD1.P 16(R0), [V0.S4]
FMOVS (R0), F20
SUB $16, R0, R0
package sha1
+import "unsafe"
+
//go:noescape
-func block(dig *digest, p []byte)
+func doBlock(dig *digest, p *byte, n int)
+
+func block(dig *digest, p []byte) {
+ doBlock(dig, unsafe.SliceData(p), len(p))
+}
+
+func blockString(dig *digest, s string) {
+ doBlock(dig, unsafe.StringData(s), len(s))
+}
func block(dig *digest, p []byte) {
blockGeneric(dig, p)
}
+
+func blockString(dig *digest, s string) {
+ blockGeneric(dig, s)
+}
package sha1
-import "internal/cpu"
+import (
+ "internal/cpu"
+ "unsafe"
+)
var useAsm = cpu.S390X.HasSHA1
+
+func doBlockGeneric(dig *digest, p *byte, n int) {
+ blockGeneric(dig, unsafe.String(p, n))
+}
#include "textflag.h"
-// func block(dig *digest, p []byte)
-TEXT ·block(SB), NOSPLIT|NOFRAME, $0-32
+// func doBlock(dig *digest, p *byte, n int)
+TEXT ·doBlock(SB), NOSPLIT|NOFRAME, $0-24
MOVBZ ·useAsm(SB), R4
- LMG dig+0(FP), R1, R3 // R2 = &p[0], R3 = len(p)
+ LMG dig+0(FP), R1, R3 // R2 = p, R3 = n
MOVBZ $1, R0 // SHA-1 function code
CMPBEQ R4, $0, generic
RET
generic:
- BR ·blockGeneric(SB)
+ BR ·doBlockGeneric(SB)