AMDGPU: Prepare for explicit absolute relocations in code generation

Summary:
We will use absolute relocations for LDS symbols.

Change-Id: I9a32795ed0ea835e433a787129cfe3c57ee9a325

Reviewers: arsenm, rampitec

Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D61492

llvm-svn: 363517

diff --git a/llvm/include/llvm/MC/MCExpr.h b/llvm/include/llvm/MC/MCExpr.h
index 706247746fba5f3bd3612fe4d729544d4f2fd5c2..7d9b265b9d590d4003413ffbec11db91950ef720 100644 (file)
--- a/llvm/include/llvm/MC/MCExpr.h
+++ b/llvm/include/llvm/MC/MCExpr.h
@@ -293,6 +293,8 @@ public:
     VK_AMDGPU_REL32_LO,      // symbol@rel32@lo
     VK_AMDGPU_REL32_HI,      // symbol@rel32@hi
     VK_AMDGPU_REL64,         // symbol@rel64
+    VK_AMDGPU_ABS32_LO,      // symbol@abs32@lo
+    VK_AMDGPU_ABS32_HI,      // symbol@abs32@hi
 
     VK_TPREL,
     VK_DTPREL

diff --git a/llvm/lib/MC/MCExpr.cpp b/llvm/lib/MC/MCExpr.cpp
index fe14272a189b63ef0f0a73c30ff14712ef9220e4..b3384599635ecee77b7dc75f43ecbf9d5c35d10a 100644 (file)
--- a/llvm/lib/MC/MCExpr.cpp
+++ b/llvm/lib/MC/MCExpr.cpp
@@ -310,6 +310,8 @@ StringRef MCSymbolRefExpr::getVariantKindName(VariantKind Kind) {
   case VK_AMDGPU_REL32_LO: return "rel32@lo";
   case VK_AMDGPU_REL32_HI: return "rel32@hi";
   case VK_AMDGPU_REL64: return "rel64";
+  case VK_AMDGPU_ABS32_LO: return "abs32@lo";
+  case VK_AMDGPU_ABS32_HI: return "abs32@hi";
   }
   llvm_unreachable("Invalid variant kind");
 }
@@ -425,6 +427,8 @@ MCSymbolRefExpr::getVariantKindForName(StringRef Name) {
     .Case("rel32@lo", VK_AMDGPU_REL32_LO)
     .Case("rel32@hi", VK_AMDGPU_REL32_HI)
     .Case("rel64", VK_AMDGPU_REL64)
+    .Case("abs32@lo", VK_AMDGPU_ABS32_LO)
+    .Case("abs32@hi", VK_AMDGPU_ABS32_HI)
     .Default(VK_Invalid);
 }
 

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUMCInstLower.cpp b/llvm/lib/Target/AMDGPU/AMDGPUMCInstLower.cpp
index 017d4ad1625f2fb5e39d7df1bc7d2d6f81ebf3c7..ae4c32c258a7493a240a3521b87ef7f8bdf4edd9 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AMDGPUMCInstLower.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUMCInstLower.cpp
@@ -90,6 +90,10 @@ static MCSymbolRefExpr::VariantKind getVariantKind(unsigned MOFlags) {
     return MCSymbolRefExpr::VK_AMDGPU_REL32_LO;
   case SIInstrInfo::MO_REL32_HI:
     return MCSymbolRefExpr::VK_AMDGPU_REL32_HI;
+  case SIInstrInfo::MO_ABS32_LO:
+    return MCSymbolRefExpr::VK_AMDGPU_ABS32_LO;
+  case SIInstrInfo::MO_ABS32_HI:
+    return MCSymbolRefExpr::VK_AMDGPU_ABS32_HI;
   }
 }
 
@@ -146,10 +150,13 @@ bool AMDGPUMCInstLower::lowerOperand(const MachineOperand &MO,
     SmallString<128> SymbolName;
     AP.getNameWithPrefix(SymbolName, GV);
     MCSymbol *Sym = Ctx.getOrCreateSymbol(SymbolName);
-    const MCExpr *SymExpr =
+    const MCExpr *Expr =
       MCSymbolRefExpr::create(Sym, getVariantKind(MO.getTargetFlags()),Ctx);
-    const MCExpr *Expr = MCBinaryExpr::createAdd(SymExpr,
-      MCConstantExpr::create(MO.getOffset(), Ctx), Ctx);
+    int64_t Offset = MO.getOffset();
+    if (Offset != 0) {
+      Expr = MCBinaryExpr::createAdd(Expr,
+                                     MCConstantExpr::create(Offset, Ctx), Ctx);
+    }
     MCOp = MCOperand::createExpr(Expr);
     return true;
   }

diff --git a/llvm/lib/Target/AMDGPU/MCTargetDesc/SIMCCodeEmitter.cpp b/llvm/lib/Target/AMDGPU/MCTargetDesc/SIMCCodeEmitter.cpp
index 40da1875ee8c355a9ec27e73592722fd8fa1178d..28ca47dce45c658fcd888f67ee13d655129dc51d 100644 (file)
--- a/llvm/lib/Target/AMDGPU/MCTargetDesc/SIMCCodeEmitter.cpp
+++ b/llvm/lib/Target/AMDGPU/MCTargetDesc/SIMCCodeEmitter.cpp
@@ -399,8 +399,12 @@ SIMCCodeEmitter::getSDWAVopcDstEncoding(const MCInst &MI, unsigned OpNo,
 
 static bool needsPCRel(const MCExpr *Expr) {
   switch (Expr->getKind()) {
-  case MCExpr::SymbolRef:
-    return true;
+  case MCExpr::SymbolRef: {
+    auto *SE = cast<MCSymbolRefExpr>(Expr);
+    MCSymbolRefExpr::VariantKind Kind = SE->getKind();
+    return Kind != MCSymbolRefExpr::VK_AMDGPU_ABS32_LO &&
+           Kind != MCSymbolRefExpr::VK_AMDGPU_ABS32_HI;
+  }
   case MCExpr::Binary: {
     auto *BE = cast<MCBinaryExpr>(Expr);
     if (BE->getOpcode() == MCBinaryExpr::Sub)

diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
index 554427826217346b7c002e559a90ce93c078fd6c..078c08a77f3579352248d6e9d4cb702f26567b7f 100644 (file)
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -5846,7 +5846,9 @@ SIInstrInfo::getSerializableDirectMachineOperandTargetFlags() const {
     { MO_GOTPCREL32_LO, "amdgpu-gotprel32-lo" },
     { MO_GOTPCREL32_HI, "amdgpu-gotprel32-hi" },
     { MO_REL32_LO, "amdgpu-rel32-lo" },
-    { MO_REL32_HI, "amdgpu-rel32-hi" }
+    { MO_REL32_HI, "amdgpu-rel32-hi" },
+    { MO_ABS32_LO, "amdgpu-abs32-lo" },
+    { MO_ABS32_HI, "amdgpu-abs32-hi" },
   };
 
   return makeArrayRef(TargetFlags);

diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.h b/llvm/lib/Target/AMDGPU/SIInstrInfo.h
index 9ce2f505a29dc31dc619a9516d26cd3028b7b4c3..2d31df5298d6716b4b64714f1c9924b6eaaafc0c 100644 (file)
--- a/llvm/lib/Target/AMDGPU/SIInstrInfo.h
+++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.h
@@ -143,7 +143,7 @@ protected:
 
 public:
   enum TargetOperandFlags {
-    MO_MASK = 0x7,
+    MO_MASK = 0xf,
 
     MO_NONE = 0,
     // MO_GOTPCREL -> symbol@GOTPCREL -> R_AMDGPU_GOTPCREL.
@@ -160,7 +160,10 @@ public:
     MO_REL32_HI = 5,
 
     MO_LONG_BRANCH_FORWARD = 6,
-    MO_LONG_BRANCH_BACKWARD = 7
+    MO_LONG_BRANCH_BACKWARD = 7,
+
+    MO_ABS32_LO = 8,
+    MO_ABS32_HI = 9,
   };
 
   explicit SIInstrInfo(const GCNSubtarget &ST);

diff --git a/llvm/test/CodeGen/MIR/AMDGPU/target-flags.mir b/llvm/test/CodeGen/MIR/AMDGPU/target-flags.mir
index 7b7eb6bc681daf9b32a734c042ac6d78762ca64e..e37bb7e6592ae84f013247c1fd13190883a6ea1a 100644 (file)
--- a/llvm/test/CodeGen/MIR/AMDGPU/target-flags.mir
+++ b/llvm/test/CodeGen/MIR/AMDGPU/target-flags.mir
@@ -26,7 +26,9 @@ body: |
     ; CHECK: [[S_MOV_B64_:%[0-9]+]]:sreg_64 = S_MOV_B64 target-flags(amdgpu-gotprel) @foo
     ; CHECK: S_ENDPGM 0
     %0 = SI_PC_ADD_REL_OFFSET target-flags(amdgpu-rel32-lo) @foo + 4, target-flags(amdgpu-rel32-hi) @foo + 4, implicit-def dead $scc
-  %1 = S_MOV_B64 target-flags(amdgpu-gotprel) @foo
+    %1 = S_MOV_B64 target-flags(amdgpu-gotprel) @foo
+    %2:sreg_32 = S_MOV_B32 target-flags(amdgpu-abs32-lo) @foo
+    %3:sreg_32 = S_MOV_B32 target-flags(amdgpu-abs32-hi) @foo
 
     S_ENDPGM 0
 ...