Index: llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
+++ llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp
@@ -404,6 +404,11 @@
   }
 }
 
+static bool isConstantOne(const Value *V) {
+  const ConstantInt *CI = dyn_cast<ConstantInt>(V);
+  return CI && CI->isOne();
+}
+
 void AMDGPUAtomicOptimizer::optimizeAtomic(Instruction &I,
                                            AtomicRMWInst::BinOp Op,
                                            unsigned ValIdx,
@@ -523,7 +528,7 @@
       // old value times the number of active lanes.
       Value *const Ctpop = B.CreateIntCast(
           B.CreateUnaryIntrinsic(Intrinsic::ctpop, Ballot), Ty, false);
-      NewV = B.CreateMul(V, Ctpop);
+      NewV = isConstantOne(V) ? Ctpop : B.CreateMul(V, Ctpop);
       break;
     }
 
@@ -543,7 +548,8 @@
       // old value times the parity of the number of active lanes.
       Value *const Ctpop = B.CreateIntCast(
           B.CreateUnaryIntrinsic(Intrinsic::ctpop, Ballot), Ty, false);
-      NewV = B.CreateMul(V, B.CreateAnd(Ctpop, 1));
+      NewV = isConstantOne(V) ? B.CreateAnd(Ctpop, 1)
+                              : B.CreateMul(V, B.CreateAnd(Ctpop, 1));
       break;
     }
   }
@@ -622,7 +628,7 @@
         llvm_unreachable("Unhandled atomic op");
       case AtomicRMWInst::Add:
       case AtomicRMWInst::Sub:
-        LaneOffset = B.CreateMul(V, Mbcnt);
+        LaneOffset = isConstantOne(V) ? Mbcnt : B.CreateMul(V, Mbcnt);
         break;
       case AtomicRMWInst::And:
       case AtomicRMWInst::Or:
@@ -633,7 +639,8 @@
         LaneOffset = B.CreateSelect(Cond, Identity, V);
         break;
       case AtomicRMWInst::Xor:
-        LaneOffset = B.CreateMul(V, B.CreateAnd(Mbcnt, 1));
+        LaneOffset = isConstantOne(V) ? B.CreateAnd(Mbcnt, 1)
+                                      : B.CreateMul(V, B.CreateAnd(Mbcnt, 1));
         break;
       }
     }
Index: llvm/test/CodeGen/AMDGPU/GlobalISel/atomic_optimizations_mul_one.ll
===================================================================
--- /dev/null
+++ llvm/test/CodeGen/AMDGPU/GlobalISel/atomic_optimizations_mul_one.ll
@@ -0,0 +1,54 @@
+; RUN: opt -S -mtriple=amdgcn-- -amdgpu-atomic-optimizer -verify-machineinstrs %s | FileCheck -check-prefix=IR %s
+; RUN: llc -global-isel -mtriple=amdgcn-- -amdgpu-atomic-optimizations -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
+
+; IR-NOT: mul i32 1, %[[REG:[0-9]+]]
+; IR-NOT: mul i32 %[[REG:[0-9]+]], 1
+
+; GCN-NOT: s_mul_i32 s[[REG1:[0-9]+]], s[[REG2:[0-9]+]], 1
+; GCN-NOT: s_mul_i32 s[[REG1:[0-9]+]], 1, s[[REG2:[0-9]+]]
+; GCN-NOT: v_mul_lo_u32 v[[REG1:[0-9]+]], v[[REG2:[0-9]+]], 1
+; GCN-NOT: v_mul_lo_u32 v[[REG1:[0-9]+]], 1, v[[REG2:[0-9]+]]
+
+declare i32 @llvm.amdgcn.struct.buffer.atomic.add.i32(i32, <4 x i32>, i32, i32, i32, i32 immarg)
+declare i32 @llvm.amdgcn.struct.buffer.atomic.sub.i32(i32, <4 x i32>, i32, i32, i32, i32 immarg)
+declare i32 @llvm.amdgcn.struct.buffer.atomic.xor.i32(i32, <4 x i32>, i32, i32, i32, i32 immarg)
+declare void @llvm.amdgcn.struct.buffer.store.format.v4i32(<4 x i32>, <4 x i32>, i32, i32, i32, i32 immarg)
+
+define amdgpu_cs void @atomic_add(<4 x i32> inreg %arg)  {
+.entry:
+  call i32 @llvm.amdgcn.struct.buffer.atomic.add.i32(i32 1, <4 x i32> %arg, i32 0, i32 0, i32 0, i32 0)
+  ret void
+}
+
+define amdgpu_cs void @atomic_add_and_format(<4 x i32> inreg %arg) {
+.entry:
+  %a = call i32 @llvm.amdgcn.struct.buffer.atomic.add.i32(i32 1, <4 x i32> %arg, i32 0, i32 0, i32 0, i32 0)
+  call void @llvm.amdgcn.struct.buffer.store.format.v4i32(<4 x i32> %arg, <4 x i32> %arg, i32 %a, i32 0, i32 0, i32 0)
+  ret void
+}
+
+define amdgpu_cs void @atomic_sub(<4 x i32> inreg %arg)  {
+.entry:
+  call i32 @llvm.amdgcn.struct.buffer.atomic.sub.i32(i32 1, <4 x i32> %arg, i32 0, i32 0, i32 0, i32 0)
+  ret void
+}
+
+define amdgpu_cs void @atomic_sub_and_format(<4 x i32> inreg %arg) {
+.entry:
+  %a = call i32 @llvm.amdgcn.struct.buffer.atomic.sub.i32(i32 1, <4 x i32> %arg, i32 0, i32 0, i32 0, i32 0)
+  call void @llvm.amdgcn.struct.buffer.store.format.v4i32(<4 x i32> %arg, <4 x i32> %arg, i32 %a, i32 0, i32 0, i32 0)
+  ret void
+}
+
+define amdgpu_cs void @atomic_xor(<4 x i32> inreg %arg)  {
+.entry:
+  call i32 @llvm.amdgcn.struct.buffer.atomic.xor.i32(i32 1, <4 x i32> %arg, i32 0, i32 0, i32 0, i32 0)
+  ret void
+}
+
+define amdgpu_cs void @atomic_xor_and_format(<4 x i32> inreg %arg) {
+.entry:
+  %a = call i32 @llvm.amdgcn.struct.buffer.atomic.xor.i32(i32 1, <4 x i32> %arg, i32 0, i32 0, i32 0, i32 0)
+  call void @llvm.amdgcn.struct.buffer.store.format.v4i32(<4 x i32> %arg, <4 x i32> %arg, i32 %a, i32 0, i32 0, i32 0)
+  ret void
+}