From 4a93cd8f662b0d3c7b45163e3c165f9cef9a33c8 Mon Sep 17 00:00:00 2001
From: Quentin Colombet <qcolombet@apple.com>
Date: Thu, 27 Aug 2015 23:47:10 +0000
Subject: [PATCH] [AArch64][CollectLOH] Fix a regression that prevented us to
 detect chains of more than 2 instructions.

I introduced this regression a while back and did not noticed it because I
somehow forgot to push the initial test cases for the pass!

Fix that as well!


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@246239 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/AArch64/AArch64CollectLOH.cpp      |   8 +-
 .../AArch64/arm64-addr-type-promotion.ll      |   5 +-
 test/CodeGen/AArch64/arm64-collect-loh.ll     | 571 ++++++++++++++++++
 3 files changed, 580 insertions(+), 4 deletions(-)

diff --git a/lib/Target/AArch64/AArch64CollectLOH.cpp b/lib/Target/AArch64/AArch64CollectLOH.cpp
index 315aee76158..68d4a55fee4 100644
--- a/lib/Target/AArch64/AArch64CollectLOH.cpp
+++ b/lib/Target/AArch64/AArch64CollectLOH.cpp
@@ -523,6 +523,8 @@ static bool isCandidateStore(const MachineInstr *Instr) {
   switch (Instr->getOpcode()) {
   default:
     return false;
+  case AArch64::STRBBui:
+  case AArch64::STRHHui:
   case AArch64::STRBui:
   case AArch64::STRHui:
   case AArch64::STRWui:
@@ -884,7 +886,8 @@ static void computeOthers(const InstrToInstrs &UseToDefs,
     bool IsL2Add = (ImmediateDefOpc == AArch64::ADDXri);
     // If the chain is three instructions long and ldr is the second element,
     // then this ldr must load form GOT, otherwise this is not a correct chain.
-    if (L2 && !IsL2Add && L2->getOperand(2).getTargetFlags() != AArch64II::MO_GOT)
+    if (L2 && !IsL2Add &&
+        !(L2->getOperand(2).getTargetFlags() & AArch64II::MO_GOT))
       continue;
     SmallVector<const MachineInstr *, 3> Args;
     MCLOHType Kind;
@@ -1000,7 +1003,8 @@ static void collectInvolvedReg(const MachineFunction &MF, MapRegToId &RegToId,
   DEBUG(dbgs() << "** Collect Involved Register\n");
   for (const auto &MBB : MF) {
     for (const MachineInstr &MI : MBB) {
-      if (!canDefBePartOfLOH(&MI))
+      if (!canDefBePartOfLOH(&MI) &&
+          !isCandidateLoad(&MI) && !isCandidateStore(&MI))
         continue;
 
       // Process defs
diff --git a/test/CodeGen/AArch64/arm64-addr-type-promotion.ll b/test/CodeGen/AArch64/arm64-addr-type-promotion.ll
index 1a9074beb40..d46800d34ca 100644
--- a/test/CodeGen/AArch64/arm64-addr-type-promotion.ll
+++ b/test/CodeGen/AArch64/arm64-addr-type-promotion.ll
@@ -1,6 +1,7 @@
-; RUN: llc -march arm64 < %s | FileCheck %s
+; RUN: llc -march arm64 < %s -aarch64-collect-loh=false | FileCheck %s
 ; rdar://13452552
-; ModuleID = 'reduced_test.ll'
+; Disable the collecting of LOH so that the labels do not get in the
+; way of the NEXT patterns.
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-n32:64-S128"
 target triple = "arm64-apple-ios3.0.0"
 
diff --git a/test/CodeGen/AArch64/arm64-collect-loh.ll b/test/CodeGen/AArch64/arm64-collect-loh.ll
index c0aa63cc433..e737861ae94 100644
--- a/test/CodeGen/AArch64/arm64-collect-loh.ll
+++ b/test/CodeGen/AArch64/arm64-collect-loh.ll
@@ -51,3 +51,574 @@ if.end4:                                          ; preds = %if.then2, %if.then,
   %add6 = add nsw i32 %tmp3, %t.addr.0
   ret i32 %add6
 }
+
+@C = common global i32 0, align 4
+
+; Check that we catch AdrpLdrGotLdr case when we have a simple chain:
+; adrp -> ldrgot -> ldr.
+; CHECK-LABEL: _getC
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _C@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _C@GOTPAGEOFF]
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr w0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotLdr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[LDR_LABEL]]
+define i32 @getC() {
+  %res = load i32, i32* @C, align 4
+  ret i32 %res
+}
+
+; LDRSW supports loading from a literal.
+; Make sure we emit AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getSExtC
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _C@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _C@GOTPAGEOFF]
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldrsw x0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotLdr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[LDR_LABEL]]
+define i64 @getSExtC() {
+  %res = load i32, i32* @C, align 4
+  %sextres = sext i32 %res to i64
+  ret i64 %sextres
+}
+
+; It may not be safe to fold the literal in the load if the address is
+; used several times.
+; Make sure we emit AdrpLdrGot for those.
+; CHECK-LABEL: _getSeveralC
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _C@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _C@GOTPAGEOFF]
+; CHECK-NEXT: ldr [[LOAD:w[0-9]+]], {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: add [[ADD:w[0-9]+]], [[LOAD]], w0
+; CHECK-NEXT: str [[ADD]], {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGot [[ADRP_LABEL]], [[LDRGOT_LABEL]]
+define void @getSeveralC(i32 %t) {
+entry:
+  %tmp = load i32, i32* @C, align 4
+  %add = add nsw i32 %tmp, %t
+  store i32 %add, i32* @C, align 4
+  ret void
+}
+
+; Make sure we catch that:
+; adrp -> ldrgot -> str.
+; CHECK-LABEL: _setC
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _C@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _C@GOTPAGEOFF]
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: str w0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotStr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[LDR_LABEL]]
+define void @setC(i32 %t) {
+entry:
+  store i32 %t, i32* @C, align 4
+  ret void
+}
+
+; Perform the same tests for internal global and a displacement
+; in the addressing mode.
+; Indeed we will get an ADD for those instead of LOADGot.
+@InternalC = internal global i32 0, align 4
+
+; Check that we catch AdrpAddLdr case when we have a simple chain:
+; adrp -> add -> ldr.
+; CHECK-LABEL: _getInternalCPlus4
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _InternalC@PAGE
+; CHECK-NEXT: [[ADDGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: add [[ADDGOT_REG:x[0-9]+]], [[ADRP_REG]], _InternalC@PAGEOFF
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr w0, {{\[}}[[ADDGOT_REG]], #16]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpAddLdr [[ADRP_LABEL]], [[ADDGOT_LABEL]], [[LDR_LABEL]]
+define i32 @getInternalCPlus4() {
+  %addr = getelementptr i32, i32* @InternalC, i32 4
+  %res = load i32, i32* %addr, align 4
+  ret i32 %res
+}
+
+; LDRSW supports loading from a literal.
+; Make sure we emit AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getSExtInternalCPlus4
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _InternalC@PAGE
+; CHECK-NEXT: [[ADDGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: add [[ADDGOT_REG:x[0-9]+]], [[ADRP_REG]], _InternalC@PAGEOFF
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldrsw x0, {{\[}}[[ADDGOT_REG]], #16]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpAddLdr [[ADRP_LABEL]], [[ADDGOT_LABEL]], [[LDR_LABEL]]
+define i64 @getSExtInternalCPlus4() {
+  %addr = getelementptr i32, i32* @InternalC, i32 4
+  %res = load i32, i32* %addr, align 4
+  %sextres = sext i32 %res to i64
+  ret i64 %sextres
+}
+
+; It may not be safe to fold the literal in the load if the address is
+; used several times.
+; Make sure we emit AdrpAdd for those.
+; CHECK-LABEL: _getSeveralInternalCPlus4
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _InternalC@PAGE
+; CHECK-NEXT: [[ADDGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: add [[ADDGOT_REG:x[0-9]+]], [[ADRP_REG]], _InternalC@PAGEOFF
+; CHECK-NEXT: ldr [[LOAD:w[0-9]+]], {{\[}}[[ADDGOT_REG]], #16]
+; CHECK-NEXT: add [[ADD:w[0-9]+]], [[LOAD]], w0
+; CHECK-NEXT: str [[ADD]], {{\[}}[[ADDGOT_REG]], #16]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpAdd [[ADRP_LABEL]], [[ADDGOT_LABEL]]
+define void @getSeveralInternalCPlus4(i32 %t) {
+entry:
+  %addr = getelementptr i32, i32* @InternalC, i32 4
+  %tmp = load i32, i32* %addr, align 4
+  %add = add nsw i32 %tmp, %t
+  store i32 %add, i32* %addr, align 4
+  ret void
+}
+
+; Make sure we catch that:
+; adrp -> add -> str.
+; CHECK-LABEL: _setInternalCPlus4
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _InternalC@PAGE
+; CHECK-NEXT: [[ADDGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: add [[ADDGOT_REG:x[0-9]+]], [[ADRP_REG]], _InternalC@PAGEOFF
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: str w0, {{\[}}[[ADDGOT_REG]], #16]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpAddStr [[ADRP_LABEL]], [[ADDGOT_LABEL]], [[LDR_LABEL]]
+define void @setInternalCPlus4(i32 %t) {
+entry:
+  %addr = getelementptr i32, i32* @InternalC, i32 4
+  store i32 %t, i32* %addr, align 4
+  ret void
+}
+
+; Check that we catch AdrpAddLdr case when we have a simple chain:
+; adrp -> ldr.
+; CHECK-LABEL: _getInternalC
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _InternalC@PAGE
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr w0, {{\[}}[[ADRP_REG]], _InternalC@PAGEOFF]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdr [[ADRP_LABEL]], [[LDR_LABEL]]
+define i32 @getInternalC() {
+  %res = load i32, i32* @InternalC, align 4
+  ret i32 %res
+}
+
+; LDRSW supports loading from a literal.
+; Make sure we emit AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getSExtInternalC
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _InternalC@PAGE
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldrsw x0, {{\[}}[[ADRP_REG]], _InternalC@PAGEOFF]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdr [[ADRP_LABEL]], [[LDR_LABEL]]
+define i64 @getSExtInternalC() {
+  %res = load i32, i32* @InternalC, align 4
+  %sextres = sext i32 %res to i64
+  ret i64 %sextres
+}
+
+; It may not be safe to fold the literal in the load if the address is
+; used several times.
+; Make sure we do not catch anything here. We have a adrp alone,
+; there is not much we can do about it.
+; CHECK-LABEL: _getSeveralInternalC
+; CHECK: adrp [[ADRP_REG:x[0-9]+]], _InternalC@PAGE
+; CHECK-NEXT: ldr [[LOAD:w[0-9]+]], {{\[}}[[ADRP_REG]], _InternalC@PAGEOFF]
+; CHECK-NEXT: add [[ADD:w[0-9]+]], [[LOAD]], w0
+; CHECK-NEXT: str [[ADD]], {{\[}}[[ADRP_REG]], _InternalC@PAGEOFF]
+; CHECK-NEXT: ret
+define void @getSeveralInternalC(i32 %t) {
+entry:
+  %tmp = load i32, i32* @InternalC, align 4
+  %add = add nsw i32 %tmp, %t
+  store i32 %add, i32* @InternalC, align 4
+  ret void
+}
+
+; Make sure we do not catch anything when:
+; adrp -> str.
+; We cannot fold anything in the str at this point.
+; Indeed, strs do not support litterals.
+; CHECK-LABEL: _setInternalC
+; CHECK: adrp [[ADRP_REG:x[0-9]+]], _InternalC@PAGE
+; CHECK-NEXT: str w0, {{\[}}[[ADRP_REG]], _InternalC@PAGEOFF]
+; CHECK-NEXT: ret
+define void @setInternalC(i32 %t) {
+entry:
+  store i32 %t, i32* @InternalC, align 4
+  ret void
+}
+
+; Now check other variant of loads/stores.
+
+@D = common global i8 0, align 4
+
+; LDRB does not support loading from a literal.
+; Make sure we emit AdrpLdrGot and not AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getD
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _D@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _D@GOTPAGEOFF]
+; CHECK-NEXT: ldrb w0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGot [[ADRP_LABEL]], [[LDRGOT_LABEL]]
+define i8 @getD() {
+  %res = load i8, i8* @D, align 4
+  ret i8 %res
+}
+
+; CHECK-LABEL: _setD
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _D@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _D@GOTPAGEOFF]
+; CHECK-NEXT: [[STR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: strb w0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotStr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[STR_LABEL]]
+define void @setD(i8 %t) {
+  store i8 %t, i8* @D, align 4
+  ret void
+}
+
+; LDRSB supports loading from a literal.
+; Make sure we emit AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getSExtD
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _D@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _D@GOTPAGEOFF]
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldrsb w0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotLdr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[LDR_LABEL]]
+define i32 @getSExtD() {
+  %res = load i8, i8* @D, align 4
+  %sextres = sext i8 %res to i32
+  ret i32 %sextres
+}
+
+; LDRSB supports loading from a literal.
+; Make sure we emit AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getSExt64D
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _D@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _D@GOTPAGEOFF]
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldrsb x0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotLdr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[LDR_LABEL]]
+define i64 @getSExt64D() {
+  %res = load i8, i8* @D, align 4
+  %sextres = sext i8 %res to i64
+  ret i64 %sextres
+}
+
+@E = common global i16 0, align 4
+
+; LDRH does not support loading from a literal.
+; Make sure we emit AdrpLdrGot and not AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getE
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _E@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _E@GOTPAGEOFF]
+; CHECK-NEXT: ldrh w0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGot [[ADRP_LABEL]], [[LDRGOT_LABEL]]
+define i16 @getE() {
+  %res = load i16, i16* @E, align 4
+  ret i16 %res
+}
+
+; LDRSH supports loading from a literal.
+; Make sure we emit AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getSExtE
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _E@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _E@GOTPAGEOFF]
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldrsh w0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotLdr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[LDR_LABEL]]
+define i32 @getSExtE() {
+  %res = load i16, i16* @E, align 4
+  %sextres = sext i16 %res to i32
+  ret i32 %sextres
+}
+
+; CHECK-LABEL: _setE
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _E@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _E@GOTPAGEOFF]
+; CHECK-NEXT: [[STR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: strh w0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotStr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[STR_LABEL]]
+define void @setE(i16 %t) {
+  store i16 %t, i16* @E, align 4
+  ret void
+}
+
+; LDRSH supports loading from a literal.
+; Make sure we emit AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getSExt64E
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _E@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _E@GOTPAGEOFF]
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldrsh x0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotLdr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[LDR_LABEL]]
+define i64 @getSExt64E() {
+  %res = load i16, i16* @E, align 4
+  %sextres = sext i16 %res to i64
+  ret i64 %sextres
+}
+
+@F = common global i64 0, align 4
+
+; LDR supports loading from a literal.
+; Make sure we emit AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getF
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _F@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _F@GOTPAGEOFF]
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr x0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotLdr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[LDR_LABEL]]
+define i64 @getF() {
+  %res = load i64, i64* @F, align 4
+  ret i64 %res
+}
+
+; CHECK-LABEL: _setF
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _F@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _F@GOTPAGEOFF]
+; CHECK-NEXT: [[STR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: str x0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotStr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[STR_LABEL]]
+define void @setF(i64 %t) {
+  store i64 %t, i64* @F, align 4
+  ret void
+}
+
+@G = common global float 0.0, align 4
+
+; LDR float supports loading from a literal.
+; Make sure we emit AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getG
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _G@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _G@GOTPAGEOFF]
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr s0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotLdr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[LDR_LABEL]]
+define float @getG() {
+  %res = load float, float* @G, align 4
+  ret float %res
+}
+
+; CHECK-LABEL: _setG
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _G@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _G@GOTPAGEOFF]
+; CHECK-NEXT: [[STR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: str s0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotStr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[STR_LABEL]]
+define void @setG(float %t) {
+  store float %t, float* @G, align 4
+  ret void
+}
+
+@H = common global half 0.0, align 4
+
+; LDR half supports loading from a literal.
+; Make sure we emit AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getH
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _H@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _H@GOTPAGEOFF]
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr h0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotLdr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[LDR_LABEL]]
+define half @getH() {
+  %res = load half, half* @H, align 4
+  ret half %res
+}
+
+; CHECK-LABEL: _setH
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _H@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _H@GOTPAGEOFF]
+; CHECK-NEXT: [[STR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: str h0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotStr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[STR_LABEL]]
+define void @setH(half %t) {
+  store half %t, half* @H, align 4
+  ret void
+}
+
+@I = common global double 0.0, align 4
+
+; LDR double supports loading from a literal.
+; Make sure we emit AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getI
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _I@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _I@GOTPAGEOFF]
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr d0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotLdr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[LDR_LABEL]]
+define double @getI() {
+  %res = load double, double* @I, align 4
+  ret double %res
+}
+
+; CHECK-LABEL: _setI
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _I@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _I@GOTPAGEOFF]
+; CHECK-NEXT: [[STR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: str d0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotStr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[STR_LABEL]]
+define void @setI(double %t) {
+  store double %t, double* @I, align 4
+  ret void
+}
+
+@J = common global <2 x i32> <i32 0, i32 0>, align 4
+
+; LDR 64-bit vector supports loading from a literal.
+; Make sure we emit AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getJ
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _J@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _J@GOTPAGEOFF]
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr d0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotLdr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[LDR_LABEL]]
+define <2 x i32> @getJ() {
+  %res = load <2 x i32>, <2 x i32>* @J, align 4
+  ret <2 x i32> %res
+}
+
+; CHECK-LABEL: _setJ
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _J@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _J@GOTPAGEOFF]
+; CHECK-NEXT: [[STR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: str d0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotStr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[STR_LABEL]]
+define void @setJ(<2 x i32> %t) {
+  store <2 x i32> %t, <2 x i32>* @J, align 4
+  ret void
+}
+
+@K = common global <4 x i32> <i32 0, i32 0, i32 0, i32 0>, align 4
+
+; LDR 128-bit vector supports loading from a literal.
+; Make sure we emit AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getK
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _K@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _K@GOTPAGEOFF]
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr q0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotLdr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[LDR_LABEL]]
+define <4 x i32> @getK() {
+  %res = load <4 x i32>, <4 x i32>* @K, align 4
+  ret <4 x i32> %res
+}
+
+; CHECK-LABEL: _setK
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _K@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _K@GOTPAGEOFF]
+; CHECK-NEXT: [[STR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: str q0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotStr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[STR_LABEL]]
+define void @setK(<4 x i32> %t) {
+  store <4 x i32> %t, <4 x i32>* @K, align 4
+  ret void
+}
+
+@L = common global <1 x i8> <i8 0>, align 4
+
+; LDR 8-bit vector supports loading from a literal.
+; Make sure we emit AdrpLdrGotLdr for those.
+; CHECK-LABEL: _getL
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _L@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _L@GOTPAGEOFF]
+; CHECK-NEXT: [[LDR_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr b0, {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGotLdr [[ADRP_LABEL]], [[LDRGOT_LABEL]], [[LDR_LABEL]]
+define <1 x i8> @getL() {
+  %res = load <1 x i8>, <1 x i8>* @L, align 4
+  ret <1 x i8> %res
+}
+
+; CHECK-LABEL: _setL
+; CHECK: [[ADRP_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: adrp [[ADRP_REG:x[0-9]+]], _L@GOTPAGE
+; CHECK-NEXT: [[LDRGOT_LABEL:Lloh[0-9]+]]:
+; CHECK-NEXT: ldr [[LDRGOT_REG:x[0-9]+]], {{\[}}[[ADRP_REG]], _L@GOTPAGEOFF]
+; Ultimately we should generate str b0, but right now, we match the vector
+; variant which does not allow to fold the immediate into the store.
+; CHECK-NEXT: st1.b { v0 }[0], {{\[}}[[LDRGOT_REG]]]
+; CHECK-NEXT: ret
+; CHECK: .loh AdrpLdrGot [[ADRP_LABEL]], [[LDRGOT_LABEL]]
+define void @setL(<1 x i8> %t) {
+  store <1 x i8> %t, <1 x i8>* @L, align 4
+  ret void
+}
-- 
2.34.1