2 files changed, 43 insertions, 8 deletions
diff --git a/src/internal/runtime/atomic/atomic_loong64.go b/src/internal/runtime/atomic/atomic_loong64.go
index a362628323..6586ad2f6c 100644
--- a/src/internal/runtime/atomic/atomic_loong64.go
+++ b/src/internal/runtime/atomic/atomic_loong64.go
@@ -12,6 +12,7 @@ import (
 )
 
 const (
+	offsetLOONG64HasLAMCAS = unsafe.Offsetof(cpu.Loong64.HasLAMCAS)
 	offsetLoong64HasLAM_BH = unsafe.Offsetof(cpu.Loong64.HasLAM_BH)
 )
 
diff --git a/src/internal/runtime/atomic/atomic_loong64.s b/src/internal/runtime/atomic/atomic_loong64.s
index 60741a23c2..d67300afc4 100644
--- a/src/internal/runtime/atomic/atomic_loong64.s
+++ b/src/internal/runtime/atomic/atomic_loong64.s
@@ -16,18 +16,32 @@ TEXT ·Cas(SB), NOSPLIT, $0-17
 	MOVV	ptr+0(FP), R4
 	MOVW	old+8(FP), R5
 	MOVW	new+12(FP), R6
-	DBAR
+
+	MOVBU	internal∕cpu·Loong64+const_offsetLOONG64HasLAMCAS(SB), R8
+	BEQ	R8, cas_again
+	MOVV	R5, R7  // backup old value
+	AMCASDBW	R6, (R4), R5
+	BNE	R7, R5, cas_fail0
+	MOVV	$1, R4
+	MOVB	R4, ret+16(FP)
+	RET
+cas_fail0:
+	MOVB	R0, ret+16(FP)
+	RET
+
+	// Implemented using the ll-sc instruction pair
+	DBAR	$0x14	// LoadAcquire barrier
 cas_again:
 	MOVV	R6, R7
 	LL	(R4), R8
-	BNE	R5, R8, cas_fail
+	BNE	R5, R8, cas_fail1
 	SC	R7, (R4)
 	BEQ	R7, cas_again
 	MOVV	$1, R4
 	MOVB	R4, ret+16(FP)
-	DBAR
+	DBAR	$0x12	// StoreRelease barrier
 	RET
-cas_fail:
+cas_fail1:
 	MOVV	$0, R4
 	JMP	-4(PC)
 
@@ -43,21 +57,41 @@ TEXT ·Cas64(SB), NOSPLIT, $0-25
 	MOVV	ptr+0(FP), R4
 	MOVV	old+8(FP), R5
 	MOVV	new+16(FP), R6
-	DBAR
+
+	MOVBU	internal∕cpu·Loong64+const_offsetLOONG64HasLAMCAS(SB), R8
+	BEQ	R8, cas64_again
+	MOVV	R5, R7  // backup old value
+	AMCASDBV	R6, (R4), R5
+	BNE	R7, R5, cas64_fail0
+	MOVV	$1, R4
+	MOVB	R4, ret+24(FP)
+	RET
+cas64_fail0:
+	MOVB	R0, ret+24(FP)
+	RET
+
+	// Implemented using the ll-sc instruction pair
+	DBAR	$0x14
 cas64_again:
 	MOVV	R6, R7
 	LLV	(R4), R8
-	BNE	R5, R8, cas64_fail
+	BNE	R5, R8, cas64_fail1
 	SCV	R7, (R4)
 	BEQ	R7, cas64_again
 	MOVV	$1, R4
 	MOVB	R4, ret+24(FP)
-	DBAR
+	DBAR	$0x12
 	RET
-cas64_fail:
+cas64_fail1:
 	MOVV	$0, R4
 	JMP	-4(PC)
 
+TEXT ·Casint32(SB),NOSPLIT,$0-17
+	JMP	·Cas(SB)
+
+TEXT ·Casint64(SB),NOSPLIT,$0-25
+	JMP	·Cas64(SB)
+
 TEXT ·Casuintptr(SB), NOSPLIT, $0-25
 	JMP	·Cas64(SB)