test(verify): prove the amd64 sse families through the link-parity kernel

Assisted-by: GLM 5.3 Flash

verify/testdata/linkfold_amd64.s

@@ -0,0 +1,58 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+#include "textflag.h"
+
+// The link-regression kernel for the amd64 SSE families: fold XOR-folds
+// every 16-byte block of a buffer into one octa with MOVOU and PXOR,
+// folds any remaining bytes with a scalar loop, combines the two octa
+// lanes with PEXTRQ, passes the total to mixq through the ABI0 stack slot
+// and XORs the seed into the returned mix.  mixq multiplies by the odd
+// golden-ratio constant, the same arithmetic the Go side of the regression
+// mirrors bit for bit.  The run claim is native: the host is amd64.
+
+// func mixq(x uint64) uint64
+TEXT ·mixq(SB), NOSPLIT, $0-16
+	MOVQ  x+0(FP), AX
+	MOVQ  $0x9e3779b97f4a7c15, CX
+	IMULQ CX, AX
+	MOVQ  AX, ret+8(FP)
+	RET
+
+// func fold(buf []byte, seed uint64) uint64
+TEXT ·fold(SB), NOSPLIT, $16-40
+	MOVQ    buf+0(FP), SI
+	MOVQ    buf_len+8(FP), CX
+	MOVQ    seed+24(FP), BX
+	PXOR    X1, X1
+	XORQ    AX, AX
+
+blocks:
+	CMPQ    CX, $16
+	JB      tail
+	MOVOU   (SI), X2
+	PXOR    X2, X1
+	ADDQ    $16, SI
+	SUBQ    $16, CX
+	JMP     blocks
+
+tail:
+	TESTQ   CX, CX
+	JE      gathered
+	MOVBQZX (SI), DX
+	ADDQ    DX, AX
+	ADDQ    $1, SI
+	SUBQ    $1, CX
+	JMP     tail
+
+gathered:
+	MOVQ    X1, DX
+	PEXTRQ  $1, X1, CX
+	ADDQ    CX, DX
+	ADDQ    AX, DX
+	MOVQ    DX, 0(SP)
+	CALL    ·mixq(SB)
+	MOVQ    8(SP), DX
+	XORQ    BX, DX
+	MOVQ    DX, ret+32(FP)
+	RET