diff --git a/src/cmd/asm/internal/asm/testdata/loong64enc1.s b/src/cmd/asm/internal/asm/testdata/loong64enc1.s
index ad0ffa33d6..cfda9ba0b4 100644
--- a/src/cmd/asm/internal/asm/testdata/loong64enc1.s
+++ b/src/cmd/asm/internal/asm/testdata/loong64enc1.s
@@ -6,12 +6,16 @@
 
 TEXT asmtest(SB),DUPOK|NOSPLIT,$0
 lable1:
-	BFPT	1(PC)			// 00050048
-	BFPT	lable1	// BFPT 2	//1ffdff4b
+	BFPT	1(PC)				// 00050048
+	BFPT	lable1		// BFPT 2	// 1ffdff4b
+	BFPT	FCC0, lable1	// BFPT FCC0, 2	// 1ff9ff4b
+	BFPT	FCC7, lable1	// BFPT FCC7, 2	// fff5ff4b
 
 lable2:
-	BFPF	1(PC)			// 00040048
-	BFPF	lable2	// BFPF 4 	// 1ffcff4b
+	BFPF	1(PC)				// 00040048
+	BFPF	lable2		// BFPF 6 	// 1ffcff4b
+	BFPF	FCC0, lable2	// BFPF FCC0, 6	// 1ff8ff4b
+	BFPF	FCC7, lable2	// BFPF FCC7, 6	// fff4ff4b
 
 	// relocation in play so the assembled offset should be 0
 	JMP	foo(SB)			// 00000050
diff --git a/src/cmd/internal/obj/loong64/asm.go b/src/cmd/internal/obj/loong64/asm.go
index 1254695be8..7fdba0c6c2 100644
--- a/src/cmd/internal/obj/loong64/asm.go
+++ b/src/cmd/internal/obj/loong64/asm.go
@@ -308,7 +308,8 @@ var optab = []Optab{
 	{ABEQ, C_REG, C_REG, C_NONE, C_BRAN, C_NONE, 6, 4, 0, 0},
 	{ABEQ, C_REG, C_NONE, C_NONE, C_BRAN, C_NONE, 6, 4, 0, 0},
 	{ABLEZ, C_REG, C_NONE, C_NONE, C_BRAN, C_NONE, 6, 4, 0, 0},
-	{ABFPT, C_NONE, C_NONE, C_NONE, C_BRAN, C_NONE, 6, 4, 0, NOTUSETMP},
+	{ABFPT, C_NONE, C_NONE, C_NONE, C_BRAN, C_NONE, 6, 4, 0, 0},
+	{ABFPT, C_FCCREG, C_NONE, C_NONE, C_BRAN, C_NONE, 6, 4, 0, 0},
 
 	{AJMP, C_NONE, C_NONE, C_NONE, C_BRAN, C_NONE, 11, 4, 0, 0}, // b
 	{AJAL, C_NONE, C_NONE, C_NONE, C_BRAN, C_NONE, 11, 4, 0, 0}, // bl
@@ -1985,7 +1986,9 @@ func (c *ctxt0) asmout(p *obj.Prog, o *Optab, out []uint32) {
 			width = 21
 			// FCC0 is the implicit source operand, now that we
 			// don't register-allocate from the FCC bank.
-			rj = REG_FCC0
+			if rj == 0 {
+				rj = REG_FCC0
+			}
 		case ABEQ, ABNE:
 			if rd == 0 || rd == REGZERO || rj == REGZERO {
 				// BEQZ/BNEZ can be encoded with 21-bit offsets.