common/bitutil: optimize TestBytes

Author: cuiweixie

common/bitutil/bitutil.go

@@ -120,31 +120,7 @@ func safeORBytes(dst, a, b []byte) int {
 
 // TestBytes tests whether any bit is set in the input byte slice.
 func TestBytes(p []byte) bool {
-	if supportsUnaligned {
-		return fastTestBytes(p)
-	}
-	return safeTestBytes(p)
-}
-
-// fastTestBytes tests for set bits in bulk. It only works on architectures that
-// support unaligned read/writes.
-func fastTestBytes(p []byte) bool {
-	n := len(p)
-	w := n / wordSize
-	if w > 0 {
-		pw := *(*[]uintptr)(unsafe.Pointer(&p))
-		for i := 0; i < w; i++ {
-			if pw[i] != 0 {
-				return true
-			}
-		}
-	}
-	for i := n - n%wordSize; i < n; i++ {
-		if p[i] != 0 {
-			return true
-		}
-	}
-	return false
+	return testBytes(p)
 }
 
 // safeTestBytes tests for set bits one byte at a time. It works on all

common/bitutil/test_amd64.s

@@ -0,0 +1,63 @@
+// Copyright 2025 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+// inspired by: https://github.com/golang/go/blob/4a3cef2036097d323b6cc0bbe90fc4d8c7588660/src/crypto/internal/fips140/subtle/xor_amd64.s
+
+//go:build !purego
+
+#include "textflag.h"
+
+// func testBytesASM(p *byte, n int) bool
+TEXT ·testBytesASM(SB), NOSPLIT, $0
+	MOVQ  p+0(FP), SI
+	MOVQ  n+8(FP), DX
+	TESTQ DX, DX            // if len is 0, return false
+	JZ    not_found
+	TESTQ $15, DX            // AND 15 & len, if not zero jump to not_aligned.
+	JNZ   not_aligned
+
+aligned:
+	MOVQ $0, AX // position in slice
+
+	PCALIGN $16
+loop16b:
+	MOVOU (SI)(AX*1), X0   // Load 16 bytes
+	PTEST X0, X0           // Test if all bits are zero (ZF=1 if all zero)
+	JNZ   found            // If any bit is set (ZF=0), jump to found
+	ADDQ  $16, AX
+	CMPQ  DX, AX
+	JNE   loop16b
+	JMP   not_found
+
+	PCALIGN $16
+loop_1b:
+	SUBQ  $1, DX           // Test 1 byte backwards.
+	MOVB  (SI)(DX*1), DI
+	TESTB DI, DI           // Test if byte is non-zero
+	JNZ   found
+	TESTQ $7, DX           // AND 7 & len, if not zero jump to loop_1b.
+	JNZ   loop_1b
+	CMPQ  DX, $0           // if len is 0, ret.
+	JE    not_found
+	TESTQ $15, DX          // AND 15 & len, if zero jump to aligned.
+	JZ    aligned
+
+not_aligned:
+	TESTQ $7, DX           // AND $7 & len, if not zero jump to loop_1b.
+	JNE   loop_1b
+	SUBQ  $8, DX           // Test 8 bytes backwards.
+	MOVQ  (SI)(DX*1), DI
+	TESTQ DI, DI           // Test if 8 bytes are non-zero
+	JNZ   found
+	CMPQ  DX, $16          // if len is greater or equal 16 here, it must be aligned.
+	JGE   aligned
+	JMP   not_found
+
+not_found:
+	MOVB $0, ret+16(FP)
+	RET
+
+found:
+	MOVB $1, ret+16(FP)
+	RET
+

common/bitutil/test_arm64.s

@@ -0,0 +1,68 @@
+// Copyright 2025 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+// inspired by: https://github.com/golang/go/blob/4a3cef2036097d323b6cc0bbe90fc4d8c7588660/src/crypto/internal/fips140/subtle/xor_arm64.s
+
+//go:build !purego
+
+#include "textflag.h"
+
+// func testBytesASM(p *byte, n int) bool
+TEXT ·testBytesASM(SB), NOSPLIT|NOFRAME, $0
+	MOVD	p+0(FP), R0
+	MOVD	n+8(FP), R1
+	CMP	$64, R1
+	BLT	tail
+loop_64:
+	VLD1.P	64(R0), [V0.B16, V1.B16, V2.B16, V3.B16]
+	// OR all vectors together to check if any byte is non-zero
+	VORR	V0.B16, V1.B16, V4.B16
+	VORR	V2.B16, V3.B16, V5.B16
+	VORR	V4.B16, V5.B16, V6.B16
+	// Check if any byte in V6 is non-zero by checking both 64-bit halves
+	VMOV	V6.D[0], R2
+	VMOV	V6.D[1], R3
+	ORR	R2, R3, R2
+	CBNZ	R2, found
+	SUBS	$64, R1
+	CMP	$64, R1
+	BGE	loop_64
+tail:
+	// quick end
+	CBZ	R1, not_found
+	TBZ	$5, R1, less_than32
+	VLD1.P	32(R0), [V0.B16, V1.B16]
+	VORR	V0.B16, V1.B16, V2.B16
+	VMOV	V2.D[0], R2
+	VMOV	V2.D[1], R3
+	ORR	R2, R3, R2
+	CBNZ	R2, found
+less_than32:
+	TBZ	$4, R1, less_than16
+	LDP.P	16(R0), (R11, R12)
+	ORR	R11, R12, R2
+	CBNZ	R2, found
+less_than16:
+	TBZ	$3, R1, less_than8
+	MOVD.P	8(R0), R11
+	CBNZ	R11, found
+less_than8:
+	TBZ	$2, R1, less_than4
+	MOVWU.P	4(R0), R11
+	CBNZ	R11, found
+less_than4:
+	TBZ	$1, R1, less_than2
+	MOVHU.P	2(R0), R11
+	CBNZ	R11, found
+less_than2:
+	TBZ	$0, R1, not_found
+	MOVBU	(R0), R11
+	CBNZ	R11, found
+not_found:
+	MOVD	$0, R0
+	MOVB	R0, ret+16(FP)
+	RET
+found:
+	MOVD	$1, R0
+	MOVB	R0, ret+16(FP)
+	RET

common/bitutil/test_asm.go

@@ -0,0 +1,15 @@
+// Copyright 2025 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+// inspired by: https://github.com/golang/go/blob/4a3cef2036097d323b6cc0bbe90fc4d8c7588660/src/crypto/internal/fips140/subtle/xor_asm.go
+
+//go:build (amd64 || arm64) && !purego
+
+package bitutil
+
+func testBytes(p []byte) bool {
+	return testBytesASM(&p[0], len(p))
+}
+
+//go:noescape
+func testBytesASM(p *byte, n int) bool

common/bitutil/test_generic.go

@@ -0,0 +1,37 @@
+// Copyright 2025 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build (!amd64 && !arm64) || purego
+
+package bitutil
+
+import "unsafe"
+
+func testBytes(p []byte) bool {
+	if supportsUnaligned {
+		return fastTestBytes(p)
+	}
+	return safeTestBytes(p)
+}
+
+// fastTestBytes tests for set bits in bulk. It only works on architectures that
+// support unaligned read/writes.
+func fastTestBytes(p []byte) bool {
+	n := len(p)
+	w := n / wordSize
+	if w > 0 {
+		pw := *(*[]uintptr)(unsafe.Pointer(&p))
+		for i := 0; i < w; i++ {
+			if pw[i] != 0 {
+				return true
+			}
+		}
+	}
+	for i := n - n%wordSize; i < n; i++ {
+		if p[i] != 0 {
+			return true
+		}
+	}
+	return false
+}