diff --git a/clang/include/clang/Basic/Builtins.def b/clang/include/clang/Basic/Builtins.def
--- a/clang/include/clang/Basic/Builtins.def
+++ b/clang/include/clang/Basic/Builtins.def
@@ -910,6 +910,7 @@
 ATOMIC_BUILTIN(__hip_atomic_compare_exchange_strong, "v.", "t")
 ATOMIC_BUILTIN(__hip_atomic_exchange, "v.", "t")
 ATOMIC_BUILTIN(__hip_atomic_fetch_add, "v.", "t")
+ATOMIC_BUILTIN(__hip_atomic_fetch_sub, "v.", "t")
 ATOMIC_BUILTIN(__hip_atomic_fetch_and, "v.", "t")
 ATOMIC_BUILTIN(__hip_atomic_fetch_or, "v.", "t")
 ATOMIC_BUILTIN(__hip_atomic_fetch_xor, "v.", "t")
diff --git a/clang/lib/AST/Expr.cpp b/clang/lib/AST/Expr.cpp
--- a/clang/lib/AST/Expr.cpp
+++ b/clang/lib/AST/Expr.cpp
@@ -4857,6 +4857,7 @@
 
   case AO__hip_atomic_exchange:
   case AO__hip_atomic_fetch_add:
+  case AO__hip_atomic_fetch_sub:
   case AO__hip_atomic_fetch_and:
   case AO__hip_atomic_fetch_or:
   case AO__hip_atomic_fetch_xor:
diff --git a/clang/lib/CodeGen/CGAtomic.cpp b/clang/lib/CodeGen/CGAtomic.cpp
--- a/clang/lib/CodeGen/CGAtomic.cpp
+++ b/clang/lib/CodeGen/CGAtomic.cpp
@@ -623,6 +623,7 @@
                                                  : llvm::Instruction::Sub;
     [[fallthrough]];
   case AtomicExpr::AO__c11_atomic_fetch_sub:
+  case AtomicExpr::AO__hip_atomic_fetch_sub:
   case AtomicExpr::AO__opencl_atomic_fetch_sub:
   case AtomicExpr::AO__atomic_fetch_sub:
     Op = E->getValueType()->isFloatingType() ? llvm::AtomicRMWInst::FSub
@@ -897,6 +898,7 @@
   case AtomicExpr::AO__c11_atomic_fetch_add:
   case AtomicExpr::AO__c11_atomic_fetch_sub:
   case AtomicExpr::AO__hip_atomic_fetch_add:
+  case AtomicExpr::AO__hip_atomic_fetch_sub:
   case AtomicExpr::AO__opencl_atomic_fetch_add:
   case AtomicExpr::AO__opencl_atomic_fetch_sub:
     if (MemTy->isPointerType()) {
@@ -1013,6 +1015,7 @@
     case AtomicExpr::AO__c11_atomic_fetch_sub:
     case AtomicExpr::AO__opencl_atomic_fetch_sub:
     case AtomicExpr::AO__atomic_fetch_sub:
+    case AtomicExpr::AO__hip_atomic_fetch_sub:
     case AtomicExpr::AO__c11_atomic_fetch_xor:
     case AtomicExpr::AO__opencl_atomic_fetch_xor:
     case AtomicExpr::AO__opencl_atomic_fetch_min:
@@ -1218,6 +1221,7 @@
       [[fallthrough]];
     case AtomicExpr::AO__c11_atomic_fetch_sub:
     case AtomicExpr::AO__opencl_atomic_fetch_sub:
+    case AtomicExpr::AO__hip_atomic_fetch_sub:
     case AtomicExpr::AO__atomic_fetch_sub:
       LibCallName = "__atomic_fetch_sub";
       AddDirectArgument(*this, Args, UseOptimizedLibcall, Val1.getPointer(),
diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
--- a/clang/lib/Sema/SemaChecking.cpp
+++ b/clang/lib/Sema/SemaChecking.cpp
@@ -6438,6 +6438,7 @@
     Form = Copy;
     break;
   case AtomicExpr::AO__hip_atomic_fetch_add:
+  case AtomicExpr::AO__hip_atomic_fetch_sub:
   case AtomicExpr::AO__hip_atomic_fetch_min:
   case AtomicExpr::AO__hip_atomic_fetch_max:
   case AtomicExpr::AO__c11_atomic_fetch_add:
diff --git a/clang/test/CodeGenCUDA/atomic-ops.cu b/clang/test/CodeGenCUDA/atomic-ops.cu
--- a/clang/test/CodeGenCUDA/atomic-ops.cu
+++ b/clang/test/CodeGenCUDA/atomic-ops.cu
@@ -6,6 +6,7 @@
 // CHECK: cmpxchg weak ptr {{%[0-9]+}}, i32 {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("singlethread-one-as") monotonic monotonic, align 4
 // CHECK: atomicrmw xchg ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("singlethread-one-as")
 // CHECK: atomicrmw add ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("singlethread-one-as")
+// CHECK: atomicrmw sub ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("singlethread-one-as")
 // CHECK: atomicrmw and ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("singlethread-one-as")
 // CHECK: atomicrmw or ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("singlethread-one-as")
 // CHECK: atomicrmw xor ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("singlethread-one-as")
@@ -18,6 +19,7 @@
   flag = __hip_atomic_compare_exchange_weak(ptr, &val, desired, __ATOMIC_RELAXED, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SINGLETHREAD);
   val = __hip_atomic_exchange(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SINGLETHREAD);
   val = __hip_atomic_fetch_add(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SINGLETHREAD);
+  val = __hip_atomic_fetch_sub(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SINGLETHREAD);
   val = __hip_atomic_fetch_and(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SINGLETHREAD);
   val = __hip_atomic_fetch_or(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SINGLETHREAD);
   val = __hip_atomic_fetch_xor(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SINGLETHREAD);
@@ -42,6 +44,7 @@
 // CHECK: cmpxchg weak ptr {{%[0-9]+}}, i32 {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("wavefront-one-as") monotonic monotonic, align 4
 // CHECK: atomicrmw xchg ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("wavefront-one-as")
 // CHECK: atomicrmw add ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("wavefront-one-as")
+// CHECK: atomicrmw sub ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("wavefront-one-as")
 // CHECK: atomicrmw and ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("wavefront-one-as")
 // CHECK: atomicrmw or ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("wavefront-one-as")
 // CHECK: atomicrmw xor ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("wavefront-one-as")
@@ -54,6 +57,7 @@
   flag = __hip_atomic_compare_exchange_weak(ptr, &val, desired, __ATOMIC_RELAXED, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WAVEFRONT);
   val = __hip_atomic_exchange(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WAVEFRONT);
   val = __hip_atomic_fetch_add(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WAVEFRONT);
+  val = __hip_atomic_fetch_sub(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WAVEFRONT);
   val = __hip_atomic_fetch_and(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WAVEFRONT);
   val = __hip_atomic_fetch_or(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WAVEFRONT);
   val = __hip_atomic_fetch_xor(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WAVEFRONT);
@@ -78,6 +82,7 @@
 // CHECK: cmpxchg weak ptr {{%[0-9]+}}, i32 {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("workgroup-one-as") monotonic monotonic, align 4
 // CHECK: atomicrmw xchg ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("workgroup-one-as")
 // CHECK: atomicrmw add ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("workgroup-one-as")
+// CHECK: atomicrmw sub ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("workgroup-one-as")
 // CHECK: atomicrmw and ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("workgroup-one-as")
 // CHECK: atomicrmw or ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("workgroup-one-as")
 // CHECK: atomicrmw xor ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("workgroup-one-as")
@@ -89,6 +94,7 @@
   flag = __hip_atomic_compare_exchange_weak(ptr, &val, desired, __ATOMIC_RELAXED, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WORKGROUP);
   val = __hip_atomic_exchange(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WORKGROUP);
   val = __hip_atomic_fetch_add(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WORKGROUP);
+  val = __hip_atomic_fetch_sub(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WORKGROUP);
   val = __hip_atomic_fetch_and(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WORKGROUP);
   val = __hip_atomic_fetch_or(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WORKGROUP);
   val = __hip_atomic_fetch_xor(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WORKGROUP);
@@ -112,6 +118,7 @@
 // CHECK: cmpxchg weak ptr {{%[0-9]+}}, i32 {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("agent-one-as") monotonic monotonic, align 4
 // CHECK: atomicrmw xchg ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("agent-one-as")
 // CHECK: atomicrmw add ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("agent-one-as")
+// CHECK: atomicrmw sub ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("agent-one-as")
 // CHECK: atomicrmw and ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("agent-one-as")
 // CHECK: atomicrmw or ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("agent-one-as")
 // CHECK: atomicrmw xor ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("agent-one-as")
@@ -123,6 +130,7 @@
   flag = __hip_atomic_compare_exchange_weak(ptr, &val, desired, __ATOMIC_RELAXED, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
   val = __hip_atomic_exchange(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
   val = __hip_atomic_fetch_add(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
+  val = __hip_atomic_fetch_sub(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
   val = __hip_atomic_fetch_and(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
   val = __hip_atomic_fetch_or(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
   val = __hip_atomic_fetch_xor(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
@@ -146,6 +154,7 @@
 // CHECK: cmpxchg weak ptr {{%[0-9]+}}, i32 {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("one-as") monotonic monotonic, align 4
 // CHECK: atomicrmw xchg ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("one-as")
 // CHECK: atomicrmw add ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("one-as")
+// CHECK: atomicrmw sub ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("one-as")
 // CHECK: atomicrmw and ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("one-as")
 // CHECK: atomicrmw or ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("one-as")
 // CHECK: atomicrmw xor ptr {{%[0-9]+}}, i32 {{%[0-9]+}} syncscope("one-as")
@@ -158,6 +167,7 @@
   flag = __hip_atomic_compare_exchange_weak(ptr, &val, desired, __ATOMIC_RELAXED, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
   val = __hip_atomic_exchange(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
   val = __hip_atomic_fetch_add(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
+  val = __hip_atomic_fetch_sub(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
   val = __hip_atomic_fetch_and(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
   val = __hip_atomic_fetch_or(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
   val = __hip_atomic_fetch_xor(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
@@ -182,6 +192,7 @@
 // CHECK: cmpxchg weak ptr {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("singlethread-one-as") monotonic monotonic, align 8
 // CHECK: atomicrmw xchg ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("singlethread-one-as")
 // CHECK: atomicrmw add ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("singlethread-one-as")
+// CHECK: atomicrmw sub ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("singlethread-one-as")
 // CHECK: atomicrmw and ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("singlethread-one-as")
 // CHECK: atomicrmw or ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("singlethread-one-as")
 // CHECK: atomicrmw xor ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("singlethread-one-as")
@@ -193,6 +204,7 @@
   flag = __hip_atomic_compare_exchange_weak(ptr, &val, desired, __ATOMIC_RELAXED, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SINGLETHREAD);
   val = __hip_atomic_exchange(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SINGLETHREAD);
   val = __hip_atomic_fetch_add(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SINGLETHREAD);
+  val = __hip_atomic_fetch_sub(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SINGLETHREAD);
   val = __hip_atomic_fetch_and(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SINGLETHREAD);
   val = __hip_atomic_fetch_or(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SINGLETHREAD);
   val = __hip_atomic_fetch_xor(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SINGLETHREAD);
@@ -220,6 +232,7 @@
 // CHECK: cmpxchg weak ptr {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("wavefront-one-as") monotonic monotonic, align 8
 // CHECK: atomicrmw xchg ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("wavefront-one-as")
 // CHECK: atomicrmw add ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("wavefront-one-as")
+// CHECK: atomicrmw sub ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("wavefront-one-as")
 // CHECK: atomicrmw and ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("wavefront-one-as")
 // CHECK: atomicrmw or ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("wavefront-one-as")
 // CHECK: atomicrmw xor ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("wavefront-one-as")
@@ -232,6 +245,7 @@
   flag = __hip_atomic_compare_exchange_weak(ptr, &val, desired, __ATOMIC_RELAXED, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WAVEFRONT);
   val = __hip_atomic_exchange(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WAVEFRONT);
   val = __hip_atomic_fetch_add(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WAVEFRONT);
+  val = __hip_atomic_fetch_sub(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WAVEFRONT);
   val = __hip_atomic_fetch_and(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WAVEFRONT);
   val = __hip_atomic_fetch_or(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WAVEFRONT);
   val = __hip_atomic_fetch_xor(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WAVEFRONT);
@@ -260,6 +274,7 @@
 // CHECK: cmpxchg weak ptr {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("workgroup-one-as") monotonic monotonic, align 8
 // CHECK: atomicrmw xchg ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("workgroup-one-as")
 // CHECK: atomicrmw add ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("workgroup-one-as")
+// CHECK: atomicrmw sub ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("workgroup-one-as")
 // CHECK: atomicrmw and ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("workgroup-one-as")
 // CHECK: atomicrmw or ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("workgroup-one-as")
 // CHECK: atomicrmw xor ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("workgroup-one-as")
@@ -271,6 +286,7 @@
   flag = __hip_atomic_compare_exchange_weak(ptr, &val, desired, __ATOMIC_RELAXED, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WORKGROUP);
   val = __hip_atomic_exchange(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WORKGROUP);
   val = __hip_atomic_fetch_add(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WORKGROUP);
+  val = __hip_atomic_fetch_sub(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WORKGROUP);
   val = __hip_atomic_fetch_and(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WORKGROUP);
   val = __hip_atomic_fetch_or(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WORKGROUP);
   val = __hip_atomic_fetch_xor(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_WORKGROUP);
@@ -296,6 +312,7 @@
 // CHECK: cmpxchg weak ptr {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("agent-one-as") monotonic monotonic, align 8
 // CHECK: atomicrmw xchg ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("agent-one-as")
 // CHECK: atomicrmw add ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("agent-one-as")
+// CHECK: atomicrmw sub ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("agent-one-as")
 // CHECK: atomicrmw and ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("agent-one-as")
 // CHECK: atomicrmw or ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("agent-one-as")
 // CHECK: atomicrmw xor ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("agent-one-as")
@@ -307,6 +324,7 @@
   flag = __hip_atomic_compare_exchange_weak(ptr, &val, desired, __ATOMIC_RELAXED, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
   val = __hip_atomic_exchange(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
   val = __hip_atomic_fetch_add(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
+  val = __hip_atomic_fetch_sub(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
   val = __hip_atomic_fetch_and(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
   val = __hip_atomic_fetch_or(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
   val = __hip_atomic_fetch_xor(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_AGENT);
@@ -332,6 +350,7 @@
 // CHECK: cmpxchg weak ptr {{%[0-9]+}}, i64 {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("one-as") monotonic monotonic, align 8
 // CHECK: atomicrmw xchg ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("one-as")
 // CHECK: atomicrmw add ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("one-as")
+// CHECK: atomicrmw sub ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("one-as")
 // CHECK: atomicrmw and ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("one-as")
 // CHECK: atomicrmw or ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("one-as")
 // CHECK: atomicrmw xor ptr {{%[0-9]+}}, i64 {{%[0-9]+}} syncscope("one-as")
@@ -344,6 +363,7 @@
   flag = __hip_atomic_compare_exchange_weak(ptr, &val, desired, __ATOMIC_RELAXED, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
   val = __hip_atomic_exchange(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
   val = __hip_atomic_fetch_add(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
+  val = __hip_atomic_fetch_sub(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
   val = __hip_atomic_fetch_and(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
   val = __hip_atomic_fetch_or(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);
   val = __hip_atomic_fetch_xor(ptr, val, __ATOMIC_RELAXED, __HIP_MEMORY_SCOPE_SYSTEM);