ARM: relax the atomic release barrier to "dmb ishst"

I believe the full "dmb ish" barrier is not required to guarantee release
semantics for atomic operations. The weaker "dmb ishst" prevents previous
operations being reordered with a store executed afterwards, which is enough.

A key point to note (fortunately already correct) is that this barrier alone is
*insufficient* for sequential consistency, no matter how liberally placed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185339 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 95 insertions, 57 deletions

diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index ff8571b..370962d 100644
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -2557,8 +2557,12 @@ static SDValue LowerATOMIC_FENCE(SDValue Op, SelectionDAG &DAG,
                        DAG.getConstant(0, MVT::i32));
   }
 
+  ConstantSDNode *OrdN = cast<ConstantSDNode>(Op.getOperand(1));
+  AtomicOrdering Ord = static_cast<AtomicOrdering>(OrdN->getZExtValue());
+  unsigned Domain = Ord == Release ? ARM_MB::ISHST : ARM_MB::ISH;
+
   return DAG.getNode(ARMISD::MEMBARRIER, dl, MVT::Other, Op.getOperand(0),
-                     DAG.getConstant(ARM_MB::ISH, MVT::i32));
+                     DAG.getConstant(Domain, MVT::i32));
 }
 
 static SDValue LowerPREFETCH(SDValue Op, SelectionDAG &DAG,
diff --git a/test/CodeGen/ARM/atomic-64bit.ll b/test/CodeGen/ARM/atomic-64bit.ll
index f2c7305..878119b 100644
--- a/test/CodeGen/ARM/atomic-64bit.ll
+++ b/test/CodeGen/ARM/atomic-64bit.ll
@@ -3,24 +3,24 @@
 
 define i64 @test1(i64* %ptr, i64 %val) {
 ; CHECK: test1:
-; CHECK: dmb ish
+; CHECK: dmb ishst
 ; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
 ; CHECK: adds [[REG3:(r[0-9]?[02468])]], [[REG1]]
 ; CHECK: adc [[REG4:(r[0-9]?[13579])]], [[REG2]]
 ; CHECK: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK: cmp
 ; CHECK: bne
-; CHECK: dmb ish
+; CHECK: dmb {{ish$}}
 
 ; CHECK-THUMB: test1:
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb ishst
 ; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
 ; CHECK-THUMB: adds.w [[REG3:[a-z0-9]+]], [[REG1]]
 ; CHECK-THUMB: adc.w [[REG4:[a-z0-9]+]], [[REG2]]
 ; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK-THUMB: cmp
 ; CHECK-THUMB: bne
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb {{ish$}}
 
   %r = atomicrmw add i64* %ptr, i64 %val seq_cst
   ret i64 %r
@@ -28,24 +28,24 @@ define i64 @test1(i64* %ptr, i64 %val) {
 
 define i64 @test2(i64* %ptr, i64 %val) {
 ; CHECK: test2:
-; CHECK: dmb ish
+; CHECK: dmb ishst
 ; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
 ; CHECK: subs [[REG3:(r[0-9]?[02468])]], [[REG1]]
 ; CHECK: sbc [[REG4:(r[0-9]?[13579])]], [[REG2]]
 ; CHECK: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK: cmp
 ; CHECK: bne
-; CHECK: dmb ish
+; CHECK: dmb {{ish$}}
 
 ; CHECK-THUMB: test2:
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb ishst
 ; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
 ; CHECK-THUMB: subs.w [[REG3:[a-z0-9]+]], [[REG1]]
 ; CHECK-THUMB: sbc.w [[REG4:[a-z0-9]+]], [[REG2]]
 ; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK-THUMB: cmp
 ; CHECK-THUMB: bne
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb {{ish$}}
 
   %r = atomicrmw sub i64* %ptr, i64 %val seq_cst
   ret i64 %r
@@ -53,24 +53,24 @@ define i64 @test2(i64* %ptr, i64 %val) {
 
 define i64 @test3(i64* %ptr, i64 %val) {
 ; CHECK: test3:
-; CHECK: dmb ish
+; CHECK: dmb ishst
 ; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
 ; CHECK: and [[REG3:(r[0-9]?[02468])]], [[REG1]]
 ; CHECK: and [[REG4:(r[0-9]?[13579])]], [[REG2]]
 ; CHECK: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK: cmp
 ; CHECK: bne
-; CHECK: dmb ish
+; CHECK: dmb {{ish$}}
 
 ; CHECK-THUMB: test3:
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb ishst
 ; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
 ; CHECK-THUMB: and.w [[REG3:[a-z0-9]+]], [[REG1]]
 ; CHECK-THUMB: and.w [[REG4:[a-z0-9]+]], [[REG2]]
 ; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK-THUMB: cmp
 ; CHECK-THUMB: bne
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb {{ish$}}
 
   %r = atomicrmw and i64* %ptr, i64 %val seq_cst
   ret i64 %r
@@ -78,24 +78,24 @@ define i64 @test3(i64* %ptr, i64 %val) {
 
 define i64 @test4(i64* %ptr, i64 %val) {
 ; CHECK: test4:
-; CHECK: dmb ish
+; CHECK: dmb ishst
 ; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
 ; CHECK: orr [[REG3:(r[0-9]?[02468])]], [[REG1]]
 ; CHECK: orr [[REG4:(r[0-9]?[13579])]], [[REG2]]
 ; CHECK: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK: cmp
 ; CHECK: bne
-; CHECK: dmb ish
+; CHECK: dmb {{ish$}}
 
 ; CHECK-THUMB: test4:
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb ishst
 ; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
 ; CHECK-THUMB: orr.w [[REG3:[a-z0-9]+]], [[REG1]]
 ; CHECK-THUMB: orr.w [[REG4:[a-z0-9]+]], [[REG2]]
 ; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK-THUMB: cmp
 ; CHECK-THUMB: bne
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb {{ish$}}
 
   %r = atomicrmw or i64* %ptr, i64 %val seq_cst
   ret i64 %r
@@ -103,24 +103,24 @@ define i64 @test4(i64* %ptr, i64 %val) {
 
 define i64 @test5(i64* %ptr, i64 %val) {
 ; CHECK: test5:
-; CHECK: dmb ish
+; CHECK: dmb ishst
 ; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
 ; CHECK: eor [[REG3:(r[0-9]?[02468])]], [[REG1]]
 ; CHECK: eor [[REG4:(r[0-9]?[13579])]], [[REG2]]
 ; CHECK: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK: cmp
 ; CHECK: bne
-; CHECK: dmb ish
+; CHECK: dmb {{ish$}}
 
 ; CHECK-THUMB: test5:
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb ishst
 ; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
 ; CHECK-THUMB: eor.w [[REG3:[a-z0-9]+]], [[REG1]]
 ; CHECK-THUMB: eor.w [[REG4:[a-z0-9]+]], [[REG2]]
 ; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK-THUMB: cmp
 ; CHECK-THUMB: bne
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb {{ish$}}
 
   %r = atomicrmw xor i64* %ptr, i64 %val seq_cst
   ret i64 %r
@@ -128,20 +128,20 @@ define i64 @test5(i64* %ptr, i64 %val) {
 
 define i64 @test6(i64* %ptr, i64 %val) {
 ; CHECK: test6:
-; CHECK: dmb ish
+; CHECK: dmb ishst
 ; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
 ; CHECK: strexd {{[a-z0-9]+}}, {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}
 ; CHECK: cmp
 ; CHECK: bne
-; CHECK: dmb ish
+; CHECK: dmb {{ish$}}
 
 ; CHECK-THUMB: test6:
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb ishst
 ; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
 ; CHECK-THUMB: strexd {{[a-z0-9]+}}, {{[a-z0-9]+}}, {{[a-z0-9]+}}
 ; CHECK-THUMB: cmp
 ; CHECK-THUMB: bne
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb {{ish$}}
 
   %r = atomicrmw xchg i64* %ptr, i64 %val seq_cst
   ret i64 %r
@@ -149,7 +149,7 @@ define i64 @test6(i64* %ptr, i64 %val) {
 
 define i64 @test7(i64* %ptr, i64 %val1, i64 %val2) {
 ; CHECK: test7:
-; CHECK: dmb ish
+; CHECK: dmb ishst
 ; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
 ; CHECK: cmp [[REG1]]
 ; CHECK: cmpeq [[REG2]]
@@ -157,10 +157,10 @@ define i64 @test7(i64* %ptr, i64 %val1, i64 %val2) {
 ; CHECK: strexd {{[a-z0-9]+}}, {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}
 ; CHECK: cmp
 ; CHECK: bne
-; CHECK: dmb ish
+; CHECK: dmb {{ish$}}
 
 ; CHECK-THUMB: test7:
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb ishst
 ; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
 ; CHECK-THUMB: cmp [[REG1]]
 ; CHECK-THUMB: it eq
@@ -169,7 +169,7 @@ define i64 @test7(i64* %ptr, i64 %val1, i64 %val2) {
 ; CHECK-THUMB: strexd {{[a-z0-9]+}}, {{[a-z0-9]+}}, {{[a-z0-9]+}}
 ; CHECK-THUMB: cmp
 ; CHECK-THUMB: bne
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb {{ish$}}
 
   %r = cmpxchg i64* %ptr, i64 %val1, i64 %val2 seq_cst
   ret i64 %r
@@ -186,7 +186,7 @@ define i64 @test8(i64* %ptr) {
 ; CHECK: strexd {{[a-z0-9]+}}, {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}
 ; CHECK: cmp
 ; CHECK: bne
-; CHECK: dmb ish
+; CHECK: dmb {{ish$}}
 
 ; CHECK-THUMB: test8:
 ; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
@@ -197,7 +197,7 @@ define i64 @test8(i64* %ptr) {
 ; CHECK-THUMB: strexd {{[a-z0-9]+}}, {{[a-z0-9]+}}, {{[a-z0-9]+}}
 ; CHECK-THUMB: cmp
 ; CHECK-THUMB: bne
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb {{ish$}}
 
   %r = load atomic i64* %ptr seq_cst, align 8
   ret i64 %r
@@ -207,20 +207,20 @@ define i64 @test8(i64* %ptr) {
 ; way to write it.
 define void @test9(i64* %ptr, i64 %val) {
 ; CHECK: test9:
-; CHECK: dmb ish
+; CHECK: dmb ishst
 ; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
 ; CHECK: strexd {{[a-z0-9]+}}, {{r[0-9]?[02468]}}, {{r[0-9]?[13579]}}
 ; CHECK: cmp
 ; CHECK: bne
-; CHECK: dmb ish
+; CHECK: dmb {{ish$}}
 
 ; CHECK-THUMB: test9:
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb ishst
 ; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
 ; CHECK-THUMB: strexd {{[a-z0-9]+}}, {{[a-z0-9]+}}, {{[a-z0-9]+}}
 ; CHECK-THUMB: cmp
 ; CHECK-THUMB: bne
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb {{ish$}}
 
   store atomic i64 %val, i64* %ptr seq_cst, align 8
   ret void
@@ -228,7 +228,7 @@ define void @test9(i64* %ptr, i64 %val) {
 
 define i64 @test10(i64* %ptr, i64 %val) {
 ; CHECK: test10:
-; CHECK: dmb ish
+; CHECK: dmb ishst
 ; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
 ; CHECK: subs {{[a-z0-9]+}}, [[REG1]], [[REG3:(r[0-9]?[02468])]]
 ; CHECK: sbcs {{[a-z0-9]+}}, [[REG2]], [[REG4:(r[0-9]?[13579])]]
@@ -236,10 +236,10 @@ define i64 @test10(i64* %ptr, i64 %val) {
 ; CHECK: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK: cmp
 ; CHECK: bne
-; CHECK: dmb ish
+; CHECK: dmb {{ish$}}
 
 ; CHECK-THUMB: test10:
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb ishst
 ; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
 ; CHECK-THUMB: subs.w {{[a-z0-9]+}}, [[REG1]], [[REG3:[a-z0-9]+]]
 ; CHECK-THUMB: sbcs.w {{[a-z0-9]+}}, [[REG2]], [[REG4:[a-z0-9]+]]
@@ -247,7 +247,7 @@ define i64 @test10(i64* %ptr, i64 %val) {
 ; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK-THUMB: cmp
 ; CHECK-THUMB: bne
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb {{ish$}}
 
   %r = atomicrmw min i64* %ptr, i64 %val seq_cst
   ret i64 %r
@@ -255,7 +255,7 @@ define i64 @test10(i64* %ptr, i64 %val) {
 
 define i64 @test11(i64* %ptr, i64 %val) {
 ; CHECK: test11:
-; CHECK: dmb ish
+; CHECK: dmb ishst
 ; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
 ; CHECK: subs {{[a-z0-9]+}}, [[REG1]], [[REG3:(r[0-9]?[02468])]]
 ; CHECK: sbcs {{[a-z0-9]+}}, [[REG2]], [[REG4:(r[0-9]?[13579])]]
@@ -263,11 +263,11 @@ define i64 @test11(i64* %ptr, i64 %val) {
 ; CHECK: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK: cmp
 ; CHECK: bne
-; CHECK: dmb ish
+; CHECK: dmb {{ish$}}
 
 
 ; CHECK-THUMB: test11:
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb ishst
 ; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
 ; CHECK-THUMB: subs.w {{[a-z0-9]+}}, [[REG1]], [[REG3:[a-z0-9]+]]
 ; CHECK-THUMB: sbcs.w {{[a-z0-9]+}}, [[REG2]], [[REG4:[a-z0-9]+]]
@@ -275,7 +275,7 @@ define i64 @test11(i64* %ptr, i64 %val) {
 ; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK-THUMB: cmp
 ; CHECK-THUMB: bne
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb {{ish$}}
 
   %r = atomicrmw umin i64* %ptr, i64 %val seq_cst
   ret i64 %r
@@ -283,7 +283,7 @@ define i64 @test11(i64* %ptr, i64 %val) {
 
 define i64 @test12(i64* %ptr, i64 %val) {
 ; CHECK: test12:
-; CHECK: dmb ish
+; CHECK: dmb ishst
 ; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
 ; CHECK: subs {{[a-z0-9]+}}, [[REG1]], [[REG3:(r[0-9]?[02468])]]
 ; CHECK: sbcs {{[a-z0-9]+}}, [[REG2]], [[REG4:(r[0-9]?[13579])]]
@@ -291,10 +291,10 @@ define i64 @test12(i64* %ptr, i64 %val) {
 ; CHECK: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK: cmp
 ; CHECK: bne
-; CHECK: dmb ish
+; CHECK: dmb {{ish$}}
 
 ; CHECK-THUMB: test12:
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb ishst
 ; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
 ; CHECK-THUMB: subs.w {{[a-z0-9]+}}, [[REG1]], [[REG3:[a-z0-9]+]]
 ; CHECK-THUMB: sbcs.w {{[a-z0-9]+}}, [[REG2]], [[REG4:[a-z0-9]+]]
@@ -302,7 +302,7 @@ define i64 @test12(i64* %ptr, i64 %val) {
 ; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK-THUMB: cmp
 ; CHECK-THUMB: bne
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb {{ish$}}
 
   %r = atomicrmw max i64* %ptr, i64 %val seq_cst
   ret i64 %r
@@ -310,7 +310,7 @@ define i64 @test12(i64* %ptr, i64 %val) {
 
 define i64 @test13(i64* %ptr, i64 %val) {
 ; CHECK: test13:
-; CHECK: dmb ish
+; CHECK: dmb ishst
 ; CHECK: ldrexd [[REG1:(r[0-9]?[02468])]], [[REG2:(r[0-9]?[13579])]]
 ; CHECK: subs {{[a-z0-9]+}}, [[REG1]], [[REG3:(r[0-9]?[02468])]]
 ; CHECK: sbcs {{[a-z0-9]+}}, [[REG2]], [[REG4:(r[0-9]?[13579])]]
@@ -318,10 +318,10 @@ define i64 @test13(i64* %ptr, i64 %val) {
 ; CHECK: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK: cmp
 ; CHECK: bne
-; CHECK: dmb ish
+; CHECK: dmb {{ish$}}
 
 ; CHECK-THUMB: test13:
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb ishst
 ; CHECK-THUMB: ldrexd [[REG1:[a-z0-9]+]], [[REG2:[a-z0-9]+]]
 ; CHECK-THUMB: subs.w {{[a-z0-9]+}}, [[REG1]], [[REG3:[a-z0-9]+]]
 ; CHECK-THUMB: sbcs.w {{[a-z0-9]+}}, [[REG2]], [[REG4:[a-z0-9]+]]
@@ -329,7 +329,7 @@ define i64 @test13(i64* %ptr, i64 %val) {
 ; CHECK-THUMB: strexd {{[a-z0-9]+}}, [[REG3]], [[REG4]]
 ; CHECK-THUMB: cmp
 ; CHECK-THUMB: bne
-; CHECK-THUMB: dmb ish
+; CHECK-THUMB: dmb {{ish$}}
   %r = atomicrmw umax i64* %ptr, i64 %val seq_cst
   ret i64 %r
 }
diff --git a/test/CodeGen/ARM/atomic-load-store.ll b/test/CodeGen/ARM/atomic-load-store.ll
index 66916a7..7ae7129 100644
--- a/test/CodeGen/ARM/atomic-load-store.ll
+++ b/test/CodeGen/ARM/atomic-load-store.ll
@@ -6,15 +6,15 @@
 
 define void @test1(i32* %ptr, i32 %val1) {
 ; ARM: test1
-; ARM: dmb ish
+; ARM: dmb ishst
 ; ARM-NEXT: str
-; ARM-NEXT: dmb ish
+; ARM-NEXT: dmb {{ish$}}
 ; THUMBONE: test1
 ; THUMBONE: __sync_lock_test_and_set_4
 ; THUMBTWO: test1
-; THUMBTWO: dmb ish
+; THUMBTWO: dmb ishst
 ; THUMBTWO-NEXT: str
-; THUMBTWO-NEXT: dmb ish
+; THUMBTWO-NEXT: dmb {{ish$}}
   store atomic i32 %val1, i32* %ptr seq_cst, align 4
   ret void
 }
@@ -22,12 +22,12 @@ define void @test1(i32* %ptr, i32 %val1) {
 define i32 @test2(i32* %ptr) {
 ; ARM: test2
 ; ARM: ldr
-; ARM-NEXT: dmb ish
+; ARM-NEXT: dmb {{ish$}}
 ; THUMBONE: test2
 ; THUMBONE: __sync_val_compare_and_swap_4
 ; THUMBTWO: test2
 ; THUMBTWO: ldr
-; THUMBTWO-NEXT: dmb ish
+; THUMBTWO-NEXT: dmb {{ish$}}
   %val = load atomic i32* %ptr seq_cst, align 4
   ret i32 %val
 }
@@ -69,3 +69,37 @@ define void @test_old_store_64bit(i64* %p, i64 %v) {
   store atomic i64 %v, i64* %p seq_cst, align 8
   ret void
 }
+
+; Release operations only need the store barrier provided by a "dmb ishst",
+
+define void @test_store_release(i32* %p, i32 %v) {
+; ARM: test_store_release:
+; ARM: dmb ishst
+; THUMBTWO: test_store_release:
+; THUMBTWO: dmb ishst
+
+  store atomic i32 %v, i32* %p release, align 4
+  ret void
+}
+
+; However, if sequential consistency is needed *something* must ensure a release
+; followed by an acquire does not get reordered. In that case a "dmb ishst" is
+; not adequate.
+define i32 @test_seq_cst(i32* %p, i32 %v) {
+; ARM: test_seq_cst:
+; ARM: dmb ishst
+; ARM: str
+; ARM: dmb {{ish$}}
+; ARM: ldr
+; ARM: dmb {{ish$}}
+
+; THUMBTWO: test_seq_cst:
+; THUMBTWO: dmb ishst
+; THUMBTWO: str
+; THUMBTWO: dmb {{ish$}}
+; THUMBTWO: ldr
+; THUMBTWO: dmb {{ish$}}
+  store atomic i32 %v, i32* %p seq_cst, align 4
+  %val = load atomic i32* %p seq_cst, align 4
+  ret i32 %val
+}