[arm-fast-isel] Add support for -arm-long-calls.

Patch by Jush Lu <jush.msn@gmail.com>.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@158368 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 121 insertions, 41 deletions

diff --git a/lib/Target/ARM/ARMFastISel.cpp b/lib/Target/ARM/ARMFastISel.cpp
index e5d5157dd5..a9b482ae2d 100644
--- a/lib/Target/ARM/ARMFastISel.cpp
+++ b/lib/Target/ARM/ARMFastISel.cpp
@@ -190,7 +190,7 @@ class ARMFastISel : public FastISel {
     unsigned ARMMaterializeGV(const GlobalValue *GV, EVT VT);
     unsigned ARMMoveToFPReg(EVT VT, unsigned SrcReg);
     unsigned ARMMoveToIntReg(EVT VT, unsigned SrcReg);
-    unsigned ARMSelectCallOp(const GlobalValue *GV);
+    unsigned ARMSelectCallOp(bool UseReg);
 
     // Call handling routines.
   private:
@@ -202,6 +202,7 @@ class ARMFastISel : public FastISel {
                          SmallVectorImpl<unsigned> &RegArgs,
                          CallingConv::ID CC,
                          unsigned &NumBytes);
+    unsigned getLibcallReg(const Twine &Name);
     bool FinishCall(MVT RetVT, SmallVectorImpl<unsigned> &UsedRegs,
                     const Instruction *I, CallingConv::ID CC,
                     unsigned &NumBytes);
@@ -2108,8 +2109,17 @@ bool ARMFastISel::SelectRet(const Instruction *I) {
   return true;
 }
 
-unsigned ARMFastISel::ARMSelectCallOp(const GlobalValue *GV) {
-  return isThumb2 ? ARM::tBL : ARM::BL;
+unsigned ARMFastISel::ARMSelectCallOp(bool UseReg) {
+  if (UseReg)
+    return isThumb2 ? ARM::tBLXr : ARM::BLX;
+  else
+    return isThumb2 ? ARM::tBL : ARM::BL;
+}
+
+unsigned ARMFastISel::getLibcallReg(const Twine &Name) {
+  GlobalValue *GV = new GlobalVariable(Type::getInt32Ty(*Context), false,
+                                       GlobalValue::ExternalLinkage, 0, Name);
+  return ARMMaterializeGV(GV, TLI.getValueType(GV->getType()));
 }
 
 // A quick function that will emit a call for a named libcall in F with the
@@ -2130,9 +2140,6 @@ bool ARMFastISel::ARMEmitLibcall(const Instruction *I, RTLIB::Libcall Call) {
   else if (!isTypeLegal(RetTy, RetVT))
     return false;
 
-  // TODO: For now if we have long calls specified we don't handle the call.
-  if (EnableARMLongCalls) return false;
-
   // Can't handle non-double multi-reg retvals.
   if (RetVT != MVT::isVoid && RetVT != MVT::i32) {  
     SmallVector<CCValAssign, 16> RVLocs;
@@ -2176,20 +2183,32 @@ bool ARMFastISel::ARMEmitLibcall(const Instruction *I, RTLIB::Libcall Call) {
   if (!ProcessCallArgs(Args, ArgRegs, ArgVTs, ArgFlags, RegArgs, CC, NumBytes))
     return false;
 
+  unsigned CalleeReg = 0;
+  if (EnableARMLongCalls) {
+    CalleeReg = getLibcallReg(TLI.getLibcallName(Call));
+    if (CalleeReg == 0) return false;
+  }
+
   // Issue the call.
-  MachineInstrBuilder MIB;
-  unsigned CallOpc = ARMSelectCallOp(NULL);
-  if (isThumb2)
-    // Explicitly adding the predicate here.
-    MIB = AddDefaultPred(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
-                         TII.get(CallOpc)))
-                         .addExternalSymbol(TLI.getLibcallName(Call));
-  else
+  unsigned CallOpc = ARMSelectCallOp(EnableARMLongCalls);
+  MachineInstrBuilder MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt,
+                                    DL, TII.get(CallOpc));
+  if (isThumb2) {
     // Explicitly adding the predicate here.
-    MIB = AddDefaultPred(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
-                         TII.get(CallOpc))
-          .addExternalSymbol(TLI.getLibcallName(Call)));
+    AddDefaultPred(MIB);
+    if (EnableARMLongCalls)
+      MIB.addReg(CalleeReg);
+    else
+      MIB.addExternalSymbol(TLI.getLibcallName(Call));
+  } else {
+    if (EnableARMLongCalls)
+      MIB.addReg(CalleeReg);
+    else
+      MIB.addExternalSymbol(TLI.getLibcallName(Call));
 
+    // Explicitly adding the predicate here.
+    AddDefaultPred(MIB);
+  }
   // Add implicit physical register uses to the call.
   for (unsigned i = 0, e = RegArgs.size(); i != e; ++i)
     MIB.addReg(RegArgs[i]);
@@ -2237,9 +2256,6 @@ bool ARMFastISel::SelectCall(const Instruction *I,
            RetVT != MVT::i8  && RetVT != MVT::i1)
     return false;
 
-  // TODO: For now if we have long calls specified we don't handle the call.
-  if (EnableARMLongCalls) return false;
-
   // Can't handle non-double multi-reg retvals.
   if (RetVT != MVT::isVoid && RetVT != MVT::i1 && RetVT != MVT::i8 &&
       RetVT != MVT::i16 && RetVT != MVT::i32) {
@@ -2306,43 +2322,43 @@ bool ARMFastISel::SelectCall(const Instruction *I,
   if (!ProcessCallArgs(Args, ArgRegs, ArgVTs, ArgFlags, RegArgs, CC, NumBytes))
     return false;
 
-  // Issue the call.
-  MachineInstrBuilder MIB;
+  bool UseReg = false;
   const GlobalValue *GV = dyn_cast<GlobalValue>(Callee);
-  unsigned CallOpc = ARMSelectCallOp(GV);
+  if (!GV || EnableARMLongCalls) UseReg = true;
+
   unsigned CalleeReg = 0;
+  if (UseReg) {
+    if (IntrMemName)
+      CalleeReg = getLibcallReg(IntrMemName);
+    else
+      CalleeReg = getRegForValue(Callee);
 
-  if (!GV){
-    CallOpc = isThumb2 ? ARM::tBLXr : ARM::BLX;
-    CalleeReg = getRegForValue(Callee);
     if (CalleeReg == 0) return false;
   }
 
-  // Explicitly adding the predicate here.
+  // Issue the call.
+  unsigned CallOpc = ARMSelectCallOp(UseReg);
+  MachineInstrBuilder MIB = BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt,
+                                    DL, TII.get(CallOpc));
   if(isThumb2) {
     // Explicitly adding the predicate here.
-    MIB = AddDefaultPred(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
-                                 TII.get(CallOpc)));
-    if (!GV)
+    AddDefaultPred(MIB);
+    if (UseReg)
       MIB.addReg(CalleeReg);
     else if (!IntrMemName)
       MIB.addGlobalAddress(GV, 0, 0);
     else 
       MIB.addExternalSymbol(IntrMemName, 0);
   } else {
-    if (!GV)
-      MIB = AddDefaultPred(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
-                                   TII.get(CallOpc))
-            .addReg(CalleeReg));
+    if (UseReg)
+      MIB.addReg(CalleeReg);
     else if (!IntrMemName)
-      // Explicitly adding the predicate here.
-      MIB = AddDefaultPred(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
-                                   TII.get(CallOpc))
-            .addGlobalAddress(GV, 0, 0));
+      MIB.addGlobalAddress(GV, 0, 0);
     else
-      MIB = AddDefaultPred(BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DL,
-                                   TII.get(CallOpc))
-            .addExternalSymbol(IntrMemName, 0));
+      MIB.addExternalSymbol(IntrMemName, 0);
+
+    // Explicitly adding the predicate here.
+    AddDefaultPred(MIB);
   }
   
   // Add implicit physical register uses to the call.
diff --git a/test/CodeGen/ARM/fast-isel-call.ll b/test/CodeGen/ARM/fast-isel-call.ll
index 10d6746acf..0f2475651c 100644
--- a/test/CodeGen/ARM/fast-isel-call.ll
+++ b/test/CodeGen/ARM/fast-isel-call.ll
@@ -1,5 +1,7 @@
 ; RUN: llc < %s -O0 -fast-isel-abort -relocation-model=dynamic-no-pic -mtriple=armv7-apple-ios | FileCheck %s --check-prefix=ARM
 ; RUN: llc < %s -O0 -fast-isel-abort -relocation-model=dynamic-no-pic -mtriple=thumbv7-apple-ios | FileCheck %s --check-prefix=THUMB
+; RUN: llc < %s -O0 -fast-isel-abort -relocation-model=dynamic-no-pic -mtriple=armv7-apple-ios -arm-long-calls | FileCheck %s --check-prefix=ARM-LONG
+; RUN: llc < %s -O0 -fast-isel-abort -relocation-model=dynamic-no-pic -mtriple=thumbv7-apple-ios -arm-long-calls | FileCheck %s --check-prefix=THUMB-LONG
 
 define i32 @t0(i1 zeroext %a) nounwind {
   %1 = zext i1 %a to i32
@@ -99,6 +101,11 @@ entry:
 ; ARM: uxtb r9, r12
 ; ARM: str r9, [sp, #4]
 ; ARM: bl _bar
+; ARM-LONG: @t10
+; ARM-LONG: movw lr, :lower16:L_bar$non_lazy_ptr
+; ARM-LONG: movt lr, :upper16:L_bar$non_lazy_ptr
+; ARM-LONG: ldr lr, [lr]
+; ARM-LONG: blx lr
 ; THUMB: @t10
 ; THUMB: movs r0, #0
 ; THUMB: movt r0, #0
@@ -121,6 +128,11 @@ entry:
 ; THUMB: uxtb.w r9, r12
 ; THUMB: str.w r9, [sp, #4]
 ; THUMB: bl _bar
+; THUMB-LONG: @t10
+; THUMB-LONG: movw lr, :lower16:L_bar$non_lazy_ptr
+; THUMB-LONG: movt lr, :upper16:L_bar$non_lazy_ptr
+; THUMB-LONG: ldr.w lr, [lr]
+; THUMB-LONG: blx lr
   %call = call i32 @bar(i8 zeroext 0, i8 zeroext -8, i8 zeroext -69, i8 zeroext 28, i8 zeroext 40, i8 zeroext -70)
   ret i32 0
 }
@@ -146,3 +158,23 @@ define void @foo3() uwtable {
   %call = call i32 %1(i32 0)
   ret void
 }
+
+define i32 @LibCall(i32 %a, i32 %b) {
+entry:
+; ARM: LibCall
+; ARM: bl ___udivsi3
+; ARM-LONG: LibCall
+; ARM-LONG: movw r2, :lower16:L___udivsi3$non_lazy_ptr
+; ARM-LONG: movt r2, :upper16:L___udivsi3$non_lazy_ptr
+; ARM-LONG: ldr r2, [r2]
+; ARM-LONG: blx r2
+; THUMB: LibCall
+; THUMB: bl ___udivsi3
+; THUMB-LONG: LibCall
+; THUMB-LONG: movw r2, :lower16:L___udivsi3$non_lazy_ptr
+; THUMB-LONG: movt r2, :upper16:L___udivsi3$non_lazy_ptr
+; THUMB-LONG: ldr r2, [r2]
+; THUMB-LONG: blx r2
+        %tmp1 = udiv i32 %a, %b         ; <i32> [#uses=1]
+        ret i32 %tmp1
+}
diff --git a/test/CodeGen/ARM/fast-isel-intrinsic.ll b/test/CodeGen/ARM/fast-isel-intrinsic.ll
index e6bdfa78d4..b73fceff6c 100644
--- a/test/CodeGen/ARM/fast-isel-intrinsic.ll
+++ b/test/CodeGen/ARM/fast-isel-intrinsic.ll
@@ -1,5 +1,7 @@
 ; RUN: llc < %s -O0 -fast-isel-abort -relocation-model=dynamic-no-pic -mtriple=armv7-apple-ios | FileCheck %s --check-prefix=ARM
 ; RUN: llc < %s -O0 -fast-isel-abort -relocation-model=dynamic-no-pic -mtriple=thumbv7-apple-ios | FileCheck %s --check-prefix=THUMB
+; RUN: llc < %s -O0 -fast-isel-abort -relocation-model=dynamic-no-pic -mtriple=armv7-apple-ios -arm-long-calls | FileCheck %s --check-prefix=ARM-LONG
+; RUN: llc < %s -O0 -fast-isel-abort -relocation-model=dynamic-no-pic -mtriple=thumbv7-apple-ios -arm-long-calls | FileCheck %s --check-prefix=THUMB-LONG
 
 @message1 = global [60 x i8] c"The LLVM Compiler Infrastructure\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00\00", align 1
 @temp = common global [60 x i8] zeroinitializer, align 1
@@ -13,6 +15,11 @@ define void @t1() nounwind ssp {
 ; ARM: movw r2, #10
 ; ARM: uxtb r1, r1
 ; ARM: bl _memset
+; ARM-LONG: t1
+; ARM-LONG: movw r3, :lower16:L_memset$non_lazy_ptr
+; ARM-LONG: movt r3, :upper16:L_memset$non_lazy_ptr
+; ARM-LONG: ldr r3, [r3]
+; ARM-LONG: blx r3
 ; THUMB: t1
 ; THUMB: movw r0, :lower16:_message1
 ; THUMB: movt r0, :upper16:_message1
@@ -23,6 +30,11 @@ define void @t1() nounwind ssp {
 ; THUMB: movt r2, #0
 ; THUMB: uxtb r1, r1
 ; THUMB: bl _memset
+; THUMB-LONG: t1
+; THUMB-LONG: movw r3, :lower16:L_memset$non_lazy_ptr
+; THUMB-LONG: movt r3, :upper16:L_memset$non_lazy_ptr
+; THUMB-LONG: ldr r3, [r3]
+; THUMB-LONG: blx r3
   call void @llvm.memset.p0i8.i32(i8* getelementptr inbounds ([60 x i8]* @message1, i32 0, i32 5), i8 64, i32 10, i32 1, i1 false)
   ret void
 }
@@ -41,6 +53,11 @@ define void @t2() nounwind ssp {
 ; ARM: mov r0, r1
 ; ARM: ldr r1, [sp]                @ 4-byte Reload
 ; ARM: bl _memcpy
+; ARM-LONG: t2
+; ARM-LONG: movw r3, :lower16:L_memcpy$non_lazy_ptr
+; ARM-LONG: movt r3, :upper16:L_memcpy$non_lazy_ptr
+; ARM-LONG: ldr r3, [r3]
+; ARM-LONG: blx r3
 ; THUMB: t2
 ; THUMB: movw r0, :lower16:L_temp$non_lazy_ptr
 ; THUMB: movt r0, :upper16:L_temp$non_lazy_ptr
@@ -51,6 +68,11 @@ define void @t2() nounwind ssp {
 ; THUMB: movt r2, #0
 ; THUMB: mov r0, r1
 ; THUMB: bl _memcpy
+; THUMB-LONG: t2
+; THUMB-LONG: movw r3, :lower16:L_memcpy$non_lazy_ptr
+; THUMB-LONG: movt r3, :upper16:L_memcpy$non_lazy_ptr
+; THUMB-LONG: ldr r3, [r3]
+; THUMB-LONG: blx r3
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* getelementptr inbounds ([60 x i8]* @temp, i32 0, i32 4), i8* getelementptr inbounds ([60 x i8]* @temp, i32 0, i32 16), i32 17, i32 1, i1 false)
   ret void
 }
@@ -67,6 +89,11 @@ define void @t3() nounwind ssp {
 ; ARM: movw r2, #10
 ; ARM: mov r0, r1
 ; ARM: bl _memmove
+; ARM-LONG: t3
+; ARM-LONG: movw r3, :lower16:L_memmove$non_lazy_ptr
+; ARM-LONG: movt r3, :upper16:L_memmove$non_lazy_ptr
+; ARM-LONG: ldr r3, [r3]
+; ARM-LONG: blx r3
 ; THUMB: t3
 ; THUMB: movw r0, :lower16:L_temp$non_lazy_ptr
 ; THUMB: movt r0, :upper16:L_temp$non_lazy_ptr
@@ -77,6 +104,11 @@ define void @t3() nounwind ssp {
 ; THUMB: movt r2, #0
 ; THUMB: mov r0, r1
 ; THUMB: bl _memmove
+; THUMB-LONG: t3
+; THUMB-LONG: movw r3, :lower16:L_memmove$non_lazy_ptr
+; THUMB-LONG: movt r3, :upper16:L_memmove$non_lazy_ptr
+; THUMB-LONG: ldr r3, [r3]
+; THUMB-LONG: blx r3
   call void @llvm.memmove.p0i8.p0i8.i32(i8* getelementptr inbounds ([60 x i8]* @temp, i32 0, i32 4), i8* getelementptr inbounds ([60 x i8]* @temp, i32 0, i32 16), i32 10, i32 1, i1 false)
   ret void
 }