[PowerPC][Future] Initial support for PC Relative addressing for global values

This patch adds PC Relative support for global values that are known at link
time. If a global value requires access through the global offset table (GOT)
it is not covered in this patch.

Differential Revision: https://reviews.llvm.org/D75280

Author: stefanp-synopsys

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

@@ -2588,17 +2588,19 @@ bool PPCTargetLowering::SelectAddressRegRegOnly(SDValue N, SDValue &Base,
 }
 
 /// Returns true if this address is a PC Relative address.
-/// PC Relative addresses are marked with the flag PPCII::MO_PCREL_FLAG.
+/// PC Relative addresses are marked with the flag PPCII::MO_PCREL_FLAG
+/// or if the node opcode is PPCISD::MAT_PCREL_ADDR.
 bool PPCTargetLowering::SelectAddressPCRel(SDValue N, SDValue &Base) const {
-  ConstantPoolSDNode *ConstPoolNode =
-      dyn_cast<ConstantPoolSDNode>(N.getNode());
-  bool HasFlag = ConstPoolNode &&
-                 ConstPoolNode->getTargetFlags() == PPCII::MO_PCREL_FLAG;
-  bool HasNode = N.getOpcode() == PPCISD::MAT_PCREL_ADDR;
-  if (HasFlag || HasNode) {
-    Base = N;
+  // This is a materialize PC Relative node. Always select this as PC Relative.
+  Base = N;
+  if (N.getOpcode() == PPCISD::MAT_PCREL_ADDR)
     return true;
-  }
+  if (ConstantPoolSDNode *CPN = dyn_cast<ConstantPoolSDNode>(N))
+    if (CPN->getTargetFlags() & PPCII::MO_PCREL_FLAG)
+      return true;
+  if (GlobalAddressSDNode *GAN = dyn_cast<GlobalAddressSDNode>(N))
+    if (GAN->getTargetFlags() & PPCII::MO_PCREL_FLAG)
+      return true;
   return false;
 }
 
@@ -3049,6 +3051,12 @@ SDValue PPCTargetLowering::LowerGlobalAddress(SDValue Op,
   // 64-bit SVR4 ABI & AIX ABI code is always position-independent.
   // The actual address of the GlobalValue is stored in the TOC.
   if (Subtarget.is64BitELFABI() || Subtarget.isAIXABI()) {
+    if (!isAccessedAsGotIndirect(Op) && Subtarget.isUsingPCRelativeCalls()) {
+      EVT Ty = getPointerTy(DAG.getDataLayout());
+      SDValue GA = DAG.getTargetGlobalAddress(GV, DL, Ty, GSDN->getOffset(),
+                                              PPCII::MO_PCREL_FLAG);
+      return DAG.getNode(PPCISD::MAT_PCREL_ADDR, DL, Ty, GA);
+    }
     setUsesTOCBasePtr(DAG);
     SDValue GA = DAG.getTargetGlobalAddress(GV, DL, PtrVT, GSDN->getOffset());
     return getTOCEntry(DAG, DL, GA);

llvm/lib/Target/PowerPC/PPCInstrPrefix.td

@@ -348,30 +348,153 @@ let Predicates = [PrefixInstrs] in {
 // follow-up patches will address this refactoring and the AddedComplexity will
 // be removed.
 let Predicates = [PCRelativeMemops], AddedComplexity = 500 in {
+  // Load i32
+  def : Pat<(i32 (zextloadi8  (PPCmatpcreladdr pcreladdr:$ga))),
+            (PLBZpc $ga, 0)>;
+  def : Pat<(i32 (extloadi8   (PPCmatpcreladdr pcreladdr:$ga))),
+            (PLBZpc $ga, 0)>;
+  def : Pat<(i32 (sextloadi16 (PPCmatpcreladdr pcreladdr:$ga))),
+            (PLHApc $ga, 0)>;
+  def : Pat<(i32 (zextloadi16 (PPCmatpcreladdr pcreladdr:$ga))),
+            (PLHZpc $ga, 0)>;
+  def : Pat<(i32 (extloadi16 (PPCmatpcreladdr pcreladdr:$ga))),
+            (PLHZpc $ga, 0)>;
+  def : Pat<(i32 (load (PPCmatpcreladdr pcreladdr:$ga))), (PLWZpc $ga, 0)>;
+
+  // Store i32
+  def : Pat<(truncstorei8 i32:$RS, (PPCmatpcreladdr pcreladdr:$ga)),
+            (PSTBpc $RS, $ga, 0)>;
+  def : Pat<(truncstorei16 i32:$RS, (PPCmatpcreladdr pcreladdr:$ga)),
+            (PSTHpc $RS, $ga, 0)>;
+  def : Pat<(store i32:$RS, (PPCmatpcreladdr pcreladdr:$ga)),
+            (PSTWpc $RS, $ga, 0)>;
+
+  // Load i64
+  def : Pat<(i64 (zextloadi8  (PPCmatpcreladdr pcreladdr:$ga))),
+            (PLBZ8pc $ga, 0)>;
+  def : Pat<(i64 (extloadi8   (PPCmatpcreladdr pcreladdr:$ga))),
+            (PLBZ8pc $ga, 0)>;
+  def : Pat<(i64 (sextloadi16 (PPCmatpcreladdr pcreladdr:$ga))),
+            (PLHA8pc $ga, 0)>;
+  def : Pat<(i64 (zextloadi16 (PPCmatpcreladdr pcreladdr:$ga))),
+            (PLHZ8pc $ga, 0)>;
+  def : Pat<(i64 (extloadi16 (PPCmatpcreladdr pcreladdr:$ga))),
+            (PLHZ8pc $ga, 0)>;
+  def : Pat<(i64 (zextloadi32 (PPCmatpcreladdr pcreladdr:$ga))),
+            (PLWZ8pc $ga, 0)>;
+  def : Pat<(i64 (sextloadi32 (PPCmatpcreladdr pcreladdr:$ga))),
+            (PLWA8pc $ga, 0)>;
+  def : Pat<(i64 (extloadi32 (PPCmatpcreladdr pcreladdr:$ga))),
+            (PLWZ8pc $ga, 0)>;
+  def : Pat<(i64 (load (PPCmatpcreladdr pcreladdr:$ga))), (PLDpc $ga, 0)>;
+
+  // Store i64
+  def : Pat<(truncstorei8 i64:$RS, (PPCmatpcreladdr pcreladdr:$ga)),
+            (PSTB8pc $RS, $ga, 0)>;
+  def : Pat<(truncstorei16 i64:$RS, (PPCmatpcreladdr pcreladdr:$ga)),
+            (PSTH8pc $RS, $ga, 0)>;
+  def : Pat<(truncstorei32 i64:$RS, (PPCmatpcreladdr pcreladdr:$ga)),
+            (PSTW8pc $RS, $ga, 0)>;
+  def : Pat<(store i64:$RS, (PPCmatpcreladdr pcreladdr:$ga)),
+            (PSTDpc $RS, $ga, 0)>;
+
   // Load f32
   def : Pat<(f32 (load (PPCmatpcreladdr pcreladdr:$addr))), (PLFSpc $addr, 0)>;
 
+  // Store f32
+  def : Pat<(store f32:$FRS, (PPCmatpcreladdr pcreladdr:$ga)),
+            (PSTFSpc $FRS, $ga, 0)>;
+
   // Load f64
   def : Pat<(f64 (extloadf32 (PPCmatpcreladdr pcreladdr:$addr))),
             (COPY_TO_REGCLASS (PLFSpc $addr, 0), VSFRC)>;
   def : Pat<(f64 (load (PPCmatpcreladdr pcreladdr:$addr))), (PLFDpc $addr, 0)>;
 
+  // Store f64
+  def : Pat<(store f64:$FRS, (PPCmatpcreladdr pcreladdr:$ga)),
+            (PSTFDpc $FRS, $ga, 0)>;
+
   // Load f128
   def : Pat<(f128 (load (PPCmatpcreladdr pcreladdr:$addr))),
             (COPY_TO_REGCLASS (PLXVpc $addr, 0), VRRC)>;
 
+  // Store f128
+  def : Pat<(store f128:$XS, (PPCmatpcreladdr pcreladdr:$ga)),
+            (PSTXVpc (COPY_TO_REGCLASS $XS, VSRC), $ga, 0)>;
+
   // Load v4i32
   def : Pat<(v4i32 (load (PPCmatpcreladdr pcreladdr:$addr))), (PLXVpc $addr, 0)>;
 
+  // Store v4i32
+  def : Pat<(store v4i32:$XS, (PPCmatpcreladdr pcreladdr:$ga)),
+            (PSTXVpc $XS, $ga, 0)>;
+
   // Load v2i64
   def : Pat<(v2i64 (load (PPCmatpcreladdr pcreladdr:$addr))), (PLXVpc $addr, 0)>;
 
+  // Store v2i64
+  def : Pat<(store v2i64:$XS, (PPCmatpcreladdr pcreladdr:$ga)),
+            (PSTXVpc $XS, $ga, 0)>;
+
   // Load v4f32
   def : Pat<(v4f32 (load (PPCmatpcreladdr pcreladdr:$addr))), (PLXVpc $addr, 0)>;
 
+  // Store v4f32
+  def : Pat<(store v4f32:$XS, (PPCmatpcreladdr pcreladdr:$ga)),
+            (PSTXVpc $XS, $ga, 0)>;
+
   // Load v2f64
   def : Pat<(v2f64 (load (PPCmatpcreladdr pcreladdr:$addr))), (PLXVpc $addr, 0)>;
 
+  // Store v2f64
+  def : Pat<(store v2f64:$XS, (PPCmatpcreladdr pcreladdr:$ga)),
+            (PSTXVpc $XS, $ga, 0)>;
+
+  // Atomic Load
+  def : Pat<(atomic_load_8 (PPCmatpcreladdr pcreladdr:$ga)),
+            (PLBZpc $ga, 0)>;
+  def : Pat<(atomic_load_16 (PPCmatpcreladdr pcreladdr:$ga)),
+            (PLHZpc $ga, 0)>;
+  def : Pat<(atomic_load_32 (PPCmatpcreladdr pcreladdr:$ga)),
+            (PLWZpc $ga, 0)>;
+  def : Pat<(atomic_load_64 (PPCmatpcreladdr pcreladdr:$ga)),
+            (PLDpc $ga, 0)>;
+
+  // Atomic Store
+  def : Pat<(atomic_store_8 (PPCmatpcreladdr pcreladdr:$ga), i32:$RS),
+            (PSTBpc $RS, $ga, 0)>;
+  def : Pat<(atomic_store_16 (PPCmatpcreladdr pcreladdr:$ga), i32:$RS),
+            (PSTHpc $RS, $ga, 0)>;
+  def : Pat<(atomic_store_32 (PPCmatpcreladdr pcreladdr:$ga), i32:$RS),
+            (PSTWpc $RS, $ga, 0)>;
+  def : Pat<(atomic_store_8 (PPCmatpcreladdr pcreladdr:$ga), i64:$RS),
+            (PSTB8pc $RS, $ga, 0)>;
+  def : Pat<(atomic_store_16 (PPCmatpcreladdr pcreladdr:$ga), i64:$RS),
+            (PSTH8pc $RS, $ga, 0)>;
+  def : Pat<(atomic_store_32 (PPCmatpcreladdr pcreladdr:$ga), i64:$RS),
+            (PSTW8pc $RS, $ga, 0)>;
+  def : Pat<(atomic_store_64 (PPCmatpcreladdr pcreladdr:$ga), i64:$RS),
+            (PSTDpc $RS, $ga, 0)>;
+
+  // Special Cases For PPCstore_scal_int_from_vsr
+  def : Pat<(PPCstore_scal_int_from_vsr
+              (f64 (PPCcv_fp_to_sint_in_vsr f64:$src)),
+              (PPCmatpcreladdr pcreladdr:$dst), 8),
+            (PSTXSDpc (XSCVDPSXDS f64:$src), $dst, 0)>;
+  def : Pat<(PPCstore_scal_int_from_vsr
+              (f64 (PPCcv_fp_to_sint_in_vsr f128:$src)),
+              (PPCmatpcreladdr pcreladdr:$dst), 8),
+            (PSTXSDpc (COPY_TO_REGCLASS (XSCVQPSDZ f128:$src), VFRC), $dst, 0)>;
+
+  def : Pat<(PPCstore_scal_int_from_vsr
+              (f64 (PPCcv_fp_to_uint_in_vsr f64:$src)),
+              (PPCmatpcreladdr pcreladdr:$dst), 8),
+            (PSTXSDpc (XSCVDPUXDS f64:$src), $dst, 0)>;
+  def : Pat<(PPCstore_scal_int_from_vsr
+              (f64 (PPCcv_fp_to_uint_in_vsr f128:$src)),
+              (PPCmatpcreladdr pcreladdr:$dst), 8),
+            (PSTXSDpc (COPY_TO_REGCLASS (XSCVQPUDZ f128:$src), VFRC), $dst, 0)>;
+
   // If the PPCmatpcreladdr node is not caught by any other pattern it should be
   // caught here and turned into a paddi instruction to materialize the address.
   def : Pat<(PPCmatpcreladdr pcreladdr:$addr), (PADDI8pc 0, $addr)>;