feat(arch): add the {1toN} broadcast to the packed amd64 memory sources

Assisted-by: GLM 5.3 Flash

arch/arm64_ext.go

@@ -170,6 +170,11 @@ type ExtOperand struct {
 	// operand (the encoder may derive the sh bit from the value).
 	Shift    int
 	HasShift bool
+	// Broadcast spells the {1toN} broadcast on an amd64 memory operand: the
+	// base-relative location holds one element the hardware splats across
+	// every lane of the destination, which the encoder lays down as EVEX.b.
+	// Only the memory positions of the entries that carry Bcast accept it.
+	Broadcast bool
 }
 
 // ExtVector builds a scalable vector operand, ADD Z1.S style.
@@ -200,6 +205,14 @@ func ExtMemory(base int, disp int64) ExtOperand {
 	return ExtOperand{Kind: ExtMem, Reg: base, Imm: disp}
 }
 
+// ExtBroadcast builds the {1toN} broadcast spelling of a base-relative memory
+// operand, the amd64 packed forms' m16bcst shape: the base is a 64-bit general
+// register number, 0..15, the displacement keeps the plain ModR/M semantics,
+// and the encoder sets EVEX.b so the single element splats across the lanes.
+func ExtBroadcast(base int, disp int64) ExtOperand {
+	return ExtOperand{Kind: ExtMem, Reg: base, Imm: disp, Broadcast: true}
+}
+
 // ExtField is one named field of the 32-bit encoding word: a bit offset from
 // the least significant end and the field's width.
 type ExtField struct {
@@ -467,6 +480,13 @@ type ExtInstr struct {
 	// of their own, ExtFormAmdMemVec and ExtFormAmdVecMem, and need no
 	// flag.  The arm64 entries all carry the zero value.
 	Mem int
+	// Bcast records that the entry's memory position takes the {1toN}
+	// broadcast beside the plain vector source, the SDM's m16bcst and
+	// m32bcst spellings: one element the hardware splats across the lanes,
+	// encoded as EVEX.b.  The packed arithmetic carries it; the scalar
+	// forms and the full-width sources do not.  The arm64 entries all
+	// carry the zero value.
+	Bcast bool
 }
 
 // Encode assembles the operands into the 4 little-endian bytes of the