feat(disasm): name the CLDEMOTE encoding the decoder refuses

The hint NOP opcode 0F 1C /r with a memory operand is CLDEMOTE, a
memory-only instruction the toolchain's own table carries; the decoder
rejects the encoding instead of naming it.  The rejected-encoding side
of the supplementary table names it from the bytes, and the corpus row
0f1c03 pins the text in the unlisted fixture, round trip byte exact.

Assisted-by: GLM 5.3 Flash

disasm/disasm.go

@@ -68,6 +68,12 @@ func Decode(a arch.Arch, code []byte, addr uint64) (Instruction, error) {
 	default: // amd64
 		inst, err := x86asm.Decode(code, 64)
 		if err != nil {
+			// A few named families the decoder refuses outright live in
+			// the same supplementary table; anything else keeps the
+			// placeholder.
+			if text, n, ok := nameAMD64Rejected(code); ok {
+				return Instruction{Addr: addr, Text: text, Len: n}, nil
+			}
 			return Instruction{Addr: addr, Text: "???", Len: 1}, nil
 		}
 		if inst.Op == 0 {

disasm/naming_amd64.go

@@ -318,3 +318,24 @@ func nameAMD64Endbr(p amd64Prefixes, tail []byte) (string, int, bool) {
 	}
 	return "", 0, false
 }
+
+// nameAMD64Rejected names an amd64 encoding the decoder refuses outright,
+// one family at a time as the corpus rows land.  CLDEMOTE, NP 0F 1C /r
+// with a memory operand, is the first: the toolchain's own table carries
+// it as a memory-only instruction, and the decoder rejects the encoding
+// instead of naming it.  The register forms of the same opcode are the
+// hint NOPs the corpus does not spell, and they stay rejected.
+func nameAMD64Rejected(code []byte) (string, int, bool) {
+	p, ok := scanAMD64Prefixes(code)
+	if !ok || p.osz || p.rep || p.repne {
+		return "", 0, false
+	}
+	if len(code) < p.n+3 || code[p.n] != 0x0f || code[p.n+1] != 0x1c {
+		return "", 0, false
+	}
+	rm, ok := decodeAMD64RM(code[p.n+2:], p.rexR, p.rexX, p.rexB)
+	if !ok || rm.regForm {
+		return "", 0, false
+	}
+	return "CLDEMOTE " + rm.text(), p.n + 2 + rm.n, true
+}

disasm/naming_amd64_test.go

@@ -76,6 +76,10 @@ func TestDegenerateNaming(t *testing.T) {
 		// toolchain spelling.
 		{[]byte{0xf3, 0x0f, 0x1e, 0xfa}, "ENDBR64"},
 		{[]byte{0xf3, 0x0f, 0x1e, 0xfb}, "ENDBR32"},
+		// amd64enc_extra.s: CLDEMOTE (BX) // 0f1c03.  The decoder
+		// rejects the encoding outright; the table names it from the
+		// bytes.
+		{[]byte{0x0f, 0x1c, 0x03}, "CLDEMOTE 0(BX)"},
 	} {
 		ins, err := Decode(arch.AMD64, tt.code, 0)
 		if err != nil {
@@ -102,12 +106,13 @@ func TestDegenerateNamingBoundaries(t *testing.T) {
 		text string
 	}{
 		// Rejected outright: RDSEED without the operand-size override
-		// (the bare 0F C7 /7 register form), MONITORX and MWAITX, the
-		// hint NOP CLDEMOTE.
+		// (the bare 0F C7 /7 register form), MONITORX and MWAITX, and
+		// the register form of the hint NOP opcode, which the corpus
+		// does not spell.
 		{[]byte{0x0f, 0xc7, 0xfa}, "???"},
 		{[]byte{0x0f, 0x01, 0xfa}, "???"},
 		{[]byte{0x0f, 0x01, 0xfb}, "???"},
-		{[]byte{0x0f, 0x1c, 0x03}, "???"},
+		{[]byte{0x0f, 0x1c, 0xc3}, "???"},
 		// Degenerate but outside the table's prefix gates: repne ADCX is
 		// no instruction the corpus names.
 		{[]byte{0xf2, 0x0f, 0x38, 0xf6, 0xd2}, "REPNE; Op(0)"},

disasm/testdata/parity_amd64_unlisted.txt

@@ -1228,3 +1228,4 @@ f3410fc7fb	RDPID R11
 f30faef3	UMONITOR BX
 f20faef3	UMWAIT BX
 f30f1efa	ENDBR64
+0f1c03	CLDEMOTE 0(BX)