XCore target: Fix eliminateFrameIndex() to handle large frames

Large frame offsets are loaded from the ConstantPool.
Where possible, offsets are encoded using the smaller MKMSK instruction.
Large frame offsets can only be used when there is a frame-pointer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@196085 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/XCore/XCoreRegisterInfo.cpp b/lib/Target/XCore/XCoreRegisterInfo.cpp
index dbd2f52a5b1ae1ef1ca356e3133a44606c8d351b..607e512b3ec69e77368659cb23d15514c684ee8d 100644 (file)
--- a/lib/Target/XCore/XCoreRegisterInfo.cpp
+++ b/lib/Target/XCore/XCoreRegisterInfo.cpp
@@ -16,16 +16,19 @@
 #include "XCoreMachineFunctionInfo.h"
 #include "llvm/ADT/BitVector.h"
 #include "llvm/ADT/STLExtras.h"
+#include "llvm/CodeGen/MachineConstantPool.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/CodeGen/RegisterScavenging.h"
+#include "llvm/IR/Constants.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/Type.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/TargetFrameLowering.h"
 #include "llvm/Target/TargetInstrInfo.h"
@@ -244,14 +247,22 @@ XCoreRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II,
 void XCoreRegisterInfo::
 loadConstant(MachineBasicBlock &MBB, MachineBasicBlock::iterator I,
             unsigned DstReg, int64_t Value, DebugLoc dl) const {
-  // TODO use mkmsk if possible.
-  if (!isImmU16(Value)) {
-    // TODO use constant pool.
-    report_fatal_error("loadConstant value too big " + Twine(Value));
-  }
-  int Opcode = isImmU6(Value) ? XCore::LDC_ru6 : XCore::LDC_lru6;
   const TargetInstrInfo &TII = *MBB.getParent()->getTarget().getInstrInfo();
-  BuildMI(MBB, I, dl, TII.get(Opcode), DstReg).addImm(Value);
+  if (isMask_32(Value)) {
+    int N = Log2_32(Value) + 1;
+    BuildMI(MBB, I, dl, TII.get(XCore::MKMSK_rus), DstReg).addImm(N);
+  } else if (isImmU16(Value)) {
+    int Opcode = isImmU6(Value) ? XCore::LDC_ru6 : XCore::LDC_lru6;
+    BuildMI(MBB, I, dl, TII.get(Opcode), DstReg).addImm(Value);
+    return;
+  } else {
+    MachineConstantPool *ConstantPool = MBB.getParent()->getConstantPool();
+    const Constant *C = ConstantInt::get(
+        Type::getInt32Ty(MBB.getParent()->getFunction()->getContext()), Value);
+    unsigned Idx = ConstantPool->getConstantPoolIndex(C, 4);
+    BuildMI(MBB, I, dl, TII.get(XCore::LDWCP_lru6), DstReg)
+        .addConstantPoolIndex(Idx);
+  }
 }
 
 unsigned XCoreRegisterInfo::getFrameRegister(const MachineFunction &MF) const {

diff --git a/test/CodeGen/XCore/epilogue_prologue_fp.ll b/test/CodeGen/XCore/epilogue_prologue_fp.ll
new file mode 100644 (file)
index 0000000..9b9837c
--- /dev/null
+++ b/test/CodeGen/XCore/epilogue_prologue_fp.ll
@@ -0,0 +1,42 @@
+; Functions with frames > 256K bytes require a frame pointer to access the stack.
+; At present, functions must be compiled using '-fno-omit-frame-pointer'.
+; RUN: llc < %s -march=xcore -disable-fp-elim | FileCheck %s
+
+declare void @f0(i32*)
+
+; CHECK: .section .cp.rodata.cst4,"aMc",@progbits,4
+; CHECK: .LCPI[[NUM:[0-9_]+]]:
+; CHECK: .long   99999
+; CHECK: .text
+; CHECK-LABEL:f1
+; CHECK:      entsp 65535
+; CHECK-NEXT: extsp 34465
+; CHECK-NEXT: stw r10, sp[1]
+; CHECK-NEXT: ldaw r10, sp[0]
+; CHECK-NEXT: ldw r1, cp[.LCPI[[NUM]]]
+; CHECK-NEXT: ldaw r0, r10[r1]
+; CHECK-NEXT: extsp 1
+; CHECK-NEXT: bl f0
+; CHECK-NEXT: ldaw sp, sp[1]
+; CHECK-NEXT: set sp, r10
+; CHECK-NEXT: ldw r10, sp[1]
+; CHECK-NEXT: ldaw sp, sp[65535]
+; CHECK-NEXT: retsp 34465
+define void @f1() nounwind {
+entry:
+  %0 = alloca [99998 x i32]
+  %1 = getelementptr inbounds [99998 x i32]* %0, i32 0, i32 99997
+  call void @f0(i32* %1)
+  ret void
+}
+
+; CHECK-LABEL:f2
+; CHECK:      mkmsk  [[REG:r[0-9]+]], 15
+; CHECK-NEXT: ldaw r0, r10{{\[}}[[REG]]{{\]}}
+define void @f2() nounwind {
+entry:
+  %0 = alloca [32768 x i32]
+  %1 = getelementptr inbounds [32768 x i32]* %0, i32 0, i32 32765
+  call void @f0(i32* %1)
+  ret void
+}