[llvm] r190078 - R600: Fix incorrect LDS size calculation

[Tom Stellard]

Author: tstellar
Date: Thu Sep  5 13:37:57 2013
New Revision: 190078

URL: http://llvm.org/viewvc/llvm-project?rev=190078&view=rev
Log:
R600: Fix incorrect LDS size calculation

GlobalAdderss nodes that appeared in more than one basic block were
being counted twice.

Added:
    llvm/trunk/test/CodeGen/R600/lds-size.ll
Modified:
    llvm/trunk/lib/Target/R600/AMDGPUISelLowering.cpp
    llvm/trunk/lib/Target/R600/AMDGPUMachineFunction.h

Modified: llvm/trunk/lib/Target/R600/AMDGPUISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/R600/AMDGPUISelLowering.cpp?rev=190078&r1=190077&r2=190078&view=diff
==============================================================================
--- llvm/trunk/lib/Target/R600/AMDGPUISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/R600/AMDGPUISelLowering.cpp Thu Sep  5 13:37:57 2013
@@ -246,12 +246,18 @@ SDValue AMDGPUTargetLowering::LowerGloba
   assert(G->getOffset() == 0 &&
          "Do not know what to do with an non-zero offset");
 
-  unsigned Offset = MFI->LDSSize;
   const GlobalValue *GV = G->getGlobal();
-  uint64_t Size = TD->getTypeAllocSize(GV->getType()->getElementType());
 
-  // XXX: Account for alignment?
-  MFI->LDSSize += Size;
+  unsigned Offset;
+  if (MFI->LocalMemoryObjects.count(GV) == 0) {
+    uint64_t Size = TD->getTypeAllocSize(GV->getType()->getElementType());
+    Offset = MFI->LDSSize;
+    MFI->LocalMemoryObjects[GV] = Offset;
+    // XXX: Account for alignment?
+    MFI->LDSSize += Size;
+  } else {
+    Offset = MFI->LocalMemoryObjects[GV];
+  }
 
   return DAG.getConstant(Offset, getPointerTy(G->getAddressSpace()));
 }

Modified: llvm/trunk/lib/Target/R600/AMDGPUMachineFunction.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/R600/AMDGPUMachineFunction.h?rev=190078&r1=190077&r2=190078&view=diff
==============================================================================
--- llvm/trunk/lib/Target/R600/AMDGPUMachineFunction.h (original)
+++ llvm/trunk/lib/Target/R600/AMDGPUMachineFunction.h Thu Sep  5 13:37:57 2013
@@ -14,6 +14,7 @@
 #define AMDGPUMACHINEFUNCTION_H
 
 #include "llvm/CodeGen/MachineFunction.h"
+#include <map>
 
 namespace llvm {
 
@@ -21,6 +22,9 @@ class AMDGPUMachineFunction : public Mac
 public:
   AMDGPUMachineFunction(const MachineFunction &MF);
   unsigned ShaderType;
+  /// A map to keep track of local memory objects and their offsets within
+  /// the local memory space.
+  std::map<const GlobalValue *, unsigned> LocalMemoryObjects;
   /// Number of bytes in the LDS that are being used.
   unsigned LDSSize;
 };

Added: llvm/trunk/test/CodeGen/R600/lds-size.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/R600/lds-size.ll?rev=190078&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/R600/lds-size.ll (added)
+++ llvm/trunk/test/CodeGen/R600/lds-size.ll Thu Sep  5 13:37:57 2013
@@ -0,0 +1,26 @@
+; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
+
+; This test makes sure we do not double count global values when they are
+; used in different basic blocks.
+
+; CHECK-LABEL: @test
+; CHECK: .long   166120
+; CHECK-NEXT: .long   1
+ at lds = internal addrspace(3) unnamed_addr global i32 zeroinitializer, align 4
+
+define void @test(i32 addrspace(1)* %out, i32 %cond) {
+entry:
+  %0 = icmp eq i32 %cond, 0
+  br i1 %0, label %if, label %else
+
+if:
+  store i32 1, i32 addrspace(3)* @lds
+  br label %endif
+
+else:
+  store i32 2, i32 addrspace(3)* @lds
+  br label %endif
+
+endif:
+  ret void
+}