fix(asm): encode BSF, BSR and POPCNT

Assisted-by: GLM 5.3

asm/encode.go

@@ -95,7 +95,7 @@ func (e *enc) encode(mnem string, ops []Operand) error {
 		return e.encodePushPop(ops, true)
 	case "POP":
 		return e.encodePushPop(ops, false)
-	case "LZCNT", "TZCNT":
+	case "BSF", "BSR", "LZCNT", "TZCNT", "POPCNT":
 		return e.encodeCount(base, ops, size)
 	case "MOVBLZX", "MOVBQZX", "MOVWLZX", "MOVWQZX", "MOVWLSX", "MOVLQSX":
 		return e.encodeMovExtend(base, ops)

asm/encode_test.go

@@ -203,6 +203,14 @@ func TestScalarGroundTruth(t *testing.T) {
 		{"LZCNTQ R8,R9", "LZCNTQ", []Operand{r8, r9}, "f34d0fbdc8", "LZCNT"},
 		{"LZCNTW AX,CX", "LZCNTW", []Operand{AX, CX}, "66f30fbdc8", "LZCNT"},
 		{"TZCNTL AX,CX", "TZCNTL", []Operand{AX, CX}, "f30fbcc8", "TZCNT"},
+		// Bit scan: BSF/BSR are the unprefixed forms of TZCNT/LZCNT's map.
+		{"BSFL AX,CX", "BSFL", []Operand{AX, CX}, "0fbcc8", "BSF"},
+		{"BSFQ R8,R9", "BSFQ", []Operand{r8, r9}, "4d0fbcc8", "BSF"},
+		{"BSFW AX,CX", "BSFW", []Operand{AX, CX}, "660fbcc8", "BSF"},
+		{"BSRL AX,CX", "BSRL", []Operand{AX, CX}, "0fbdc8", "BSR"},
+		{"BSRQ AX,CX", "BSRQ", []Operand{AX, CX}, "480fbdc8", "BSR"},
+		{"POPCNTL AX,CX", "POPCNTL", []Operand{AX, CX}, "f30fb8c8", "POPCNT"},
+		{"POPCNTQ R8,R9", "POPCNTQ", []Operand{r8, r9}, "f34d0fb8c8", "POPCNT"},
 		{"CMOVLGT CX,AX", "CMOVLGT", []Operand{CX, AX}, "0f4fc1", "CMOVG"},
 		{"CMOVLEQ CX,AX", "CMOVLEQ", []Operand{CX, AX}, "0f44c1", "CMOVE"},
 		{"CMOVQGT R9,R8", "CMOVQGT", []Operand{r9, r8}, "4d0f4fc1", "CMOVG"},

asm/instrs.go

@@ -597,25 +597,39 @@ func (e *enc) encodeSet(upper string, ops []Operand) error {
 	return e.emit(i)
 }
 
-// --- LZCNT / TZCNT ----------------------------------------------------------
+// --- bit scan / bit count ----------------------------------------------------
 
-// encodeCount encodes LZCNT/TZCNT (leading / trailing zero count): F3 0F BD
-// or F3 0F BC, with reg = dst and rm = src.  The size suffix selects the
-// operand width (LZCNTW/LZCNTL/LZCNTQ).
+// countOp maps the bit-scan and bit-count mnemonics to their opcode byte and
+// mandatory prefix.  TZCNT/LZCNT/POPCNT are the F3-prefixed forms of the
+// same map as BSF/BSR's 0F BC/BD; POPCNT is F3 0F B8.
+var countOp = map[string]struct {
+	op     byte
+	prefix byte
+}{
+	"BSF":    {0xBC, 0},
+	"BSR":    {0xBD, 0},
+	"TZCNT":  {0xBC, 0xF3},
+	"LZCNT":  {0xBD, 0xF3},
+	"POPCNT": {0xB8, 0xF3},
+}
+
+// encodeCount encodes the bit-scan and bit-count family — BSF (0F BC),
+// BSR (0F BD), TZCNT (F3 0F BC), LZCNT (F3 0F BD) and POPCNT (F3 0F B8) —
+// with reg = dst and rm = src.  The size suffix selects the operand width
+// (BSFQ, TZCNTL, …).  Note BSF/BSR leave the destination undefined when the
+// source is zero (unlike their F3-prefixed counterparts); callers must
+// guard non-zero inputs themselves.
 func (e *enc) encodeCount(base string, ops []Operand, size int) error {
 	if len(ops) != 2 {
 		return fmt.Errorf("%s expects 2 operands, got %d", base, len(ops))
 	}
-	op := byte(0xBD)
-	if base == "TZCNT" {
-		op = 0xBC
-	}
+	spec := countOp[base]
 	dstReg, ok := ops[1].(Reg)
 	if !ok {
 		return fmt.Errorf("%s destination must be a register", base)
 	}
-	i := newInstr(size, []byte{0x0F, op})
-	i.prefix = 0xF3
+	i := newInstr(size, []byte{0x0F, spec.op})
+	i.prefix = spec.prefix
 	if err := setRM(i, dstReg, ops[0], size); err != nil {
 		return err
 	}