cmd/compile: mark some AMD64 atomic ops as clobberFlags

Fixes #16985.

Change-Id: I5954db28f7b70dd3ac7768e471d5df871a5b20f9
Reviewed-on: https://go-review.googlesource.com/28510
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

src/cmd/compile/internal/ssa/gen/AMD64Ops.go

@@ -526,8 +526,8 @@ func init() {
 		// *(arg1+auxint+aux) += arg0.  arg2=mem.
 		// Returns a tuple of <old contents of *(arg1+auxint+aux), memory>.
 		// Note: arg0 and arg1 are backwards compared to MOVLstore (to facilitate resultInArg0)!
-		{name: "XADDLlock", argLength: 3, reg: gpstorexchg, asm: "XADDL", typ: "(UInt32,Mem)", aux: "SymOff", resultInArg0: true},
+		{name: "XADDLlock", argLength: 3, reg: gpstorexchg, asm: "XADDL", typ: "(UInt32,Mem)", aux: "SymOff", resultInArg0: true, clobberFlags: true},
-		{name: "XADDQlock", argLength: 3, reg: gpstorexchg, asm: "XADDQ", typ: "(UInt64,Mem)", aux: "SymOff", resultInArg0: true},
+		{name: "XADDQlock", argLength: 3, reg: gpstorexchg, asm: "XADDQ", typ: "(UInt64,Mem)", aux: "SymOff", resultInArg0: true, clobberFlags: true},
 		{name: "AddTupleFirst32", argLength: 2}, // arg0=tuple <x,y>.  Returns <x+arg1,y>.
 		{name: "AddTupleFirst64", argLength: 2}, // arg0=tuple <x,y>.  Returns <x+arg1,y>.
 
@@ -550,12 +550,12 @@ func init() {
 		//    JEQ ...
 		// but we can't do that because memory-using ops can't generate flags yet
 		// (flagalloc wants to move flag-generating instructions around).
-		{name: "CMPXCHGLlock", argLength: 4, reg: cmpxchg, asm: "CMPXCHGL", aux: "SymOff"},
+		{name: "CMPXCHGLlock", argLength: 4, reg: cmpxchg, asm: "CMPXCHGL", aux: "SymOff", clobberFlags: true},
-		{name: "CMPXCHGQlock", argLength: 4, reg: cmpxchg, asm: "CMPXCHGQ", aux: "SymOff"},
+		{name: "CMPXCHGQlock", argLength: 4, reg: cmpxchg, asm: "CMPXCHGQ", aux: "SymOff", clobberFlags: true},
 
 		// Atomic memory updates.
-		{name: "ANDBlock", argLength: 3, reg: gpstore, asm: "ANDB", aux: "SymOff"}, // *(arg0+auxint+aux) &= arg1
+		{name: "ANDBlock", argLength: 3, reg: gpstore, asm: "ANDB", aux: "SymOff", clobberFlags: true}, // *(arg0+auxint+aux) &= arg1
-		{name: "ORBlock", argLength: 3, reg: gpstore, asm: "ORB", aux: "SymOff"},   // *(arg0+auxint+aux) |= arg1
+		{name: "ORBlock", argLength: 3, reg: gpstore, asm: "ORB", aux: "SymOff", clobberFlags: true},   // *(arg0+auxint+aux) |= arg1
 	}
 
 	var AMD64blocks = []blockData{

src/cmd/compile/internal/ssa/opGen.go

@@ -6872,6 +6872,7 @@ var opcodeTable = [...]opInfo{
 		auxType:      auxSymOff,
 		argLen:       3,
 		resultInArg0: true,
+		clobberFlags: true,
 		asm:          x86.AXADDL,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -6888,6 +6889,7 @@ var opcodeTable = [...]opInfo{
 		auxType:      auxSymOff,
 		argLen:       3,
 		resultInArg0: true,
+		clobberFlags: true,
 		asm:          x86.AXADDQ,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -6913,6 +6915,7 @@ var opcodeTable = [...]opInfo{
 		name:         "CMPXCHGLlock",
 		auxType:      auxSymOff,
 		argLen:       4,
+		clobberFlags: true,
 		asm:          x86.ACMPXCHGL,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -6931,6 +6934,7 @@ var opcodeTable = [...]opInfo{
 		name:         "CMPXCHGQlock",
 		auxType:      auxSymOff,
 		argLen:       4,
+		clobberFlags: true,
 		asm:          x86.ACMPXCHGQ,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -6949,6 +6953,7 @@ var opcodeTable = [...]opInfo{
 		name:         "ANDBlock",
 		auxType:      auxSymOff,
 		argLen:       3,
+		clobberFlags: true,
 		asm:          x86.AANDB,
 		reg: regInfo{
 			inputs: []inputInfo{
@@ -6961,6 +6966,7 @@ var opcodeTable = [...]opInfo{
 		name:         "ORBlock",
 		auxType:      auxSymOff,
 		argLen:       3,
+		clobberFlags: true,
 		asm:          x86.AORB,
 		reg: regInfo{
 			inputs: []inputInfo{

test/fixedbugs/issue16985.go

@@ -0,0 +1,37 @@
+// run
+
+// Copyright 2016 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// issue 16985: intrinsified AMD64 atomic ops should clobber flags
+
+package main
+
+import "sync/atomic"
+
+var count uint32
+
+func main() {
+	buffer := []byte("T")
+	for i := 0; i < len(buffer); {
+		atomic.AddUint32(&count, 1)
+		_ = buffer[i]
+		i++
+		i++
+	}
+
+	for i := 0; i < len(buffer); {
+		atomic.CompareAndSwapUint32(&count, 0, 1)
+		_ = buffer[i]
+		i++
+		i++
+	}
+
+	for i := 0; i < len(buffer); {
+		atomic.SwapUint32(&count, 1)
+		_ = buffer[i]
+		i++
+		i++
+	}
+}