[WIP] Extend data layout to add non zero null value for address space

Author: ranapratap55

llvm/include/llvm/IR/DataLayout.h

@@ -21,6 +21,7 @@
 
 #include "llvm/ADT/APInt.h"
 #include "llvm/ADT/ArrayRef.h"
+#include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringRef.h"
@@ -164,6 +165,8 @@ class DataLayout {
   /// well-defined bitwise representation.
   SmallVector<unsigned, 8> NonIntegralAddressSpaces;
 
+  DenseMap<unsigned, int64_t> AddrSpaceToNonZeroValueMap;
+
   /// Attempts to set the alignment of the given type. Returns an error
   /// description on failure.
   Error setAlignment(AlignTypeEnum AlignType, Align ABIAlign, Align PrefAlign,
@@ -299,6 +302,17 @@ class DataLayout {
     return ManglingMode == MM_WinCOFFX86;
   }
 
+  int64_t getNullPointerValue(unsigned AddrSpace) {
+    auto It = AddrSpaceToNonZeroValueMap.find(AddrSpace);
+    if (It == AddrSpaceToNonZeroValueMap.end())
+      return 0;
+    return It->second;
+  }
+
+  void setNullPointerValue(unsigned AddrSpace, int64_t Value) {
+    AddrSpaceToNonZeroValueMap[AddrSpace] = Value;
+  }
+
   /// Returns true if symbols with leading question marks should not receive IR
   /// mangling. True for Windows mangling modes.
   bool doNotMangleLeadingQuestionMark() const {

llvm/lib/IR/DataLayout.cpp

@@ -216,6 +216,8 @@ void DataLayout::reset(StringRef Desc) {
   if (Error Err = setPointerAlignmentInBits(0, Align(8), Align(8), 64, 64))
     return report_fatal_error(std::move(Err));
 
+  setNullPointerValue(INT_MAX, 0);
+
   if (Error Err = parseSpecifier(Desc))
     return report_fatal_error(std::move(Err));
 }
@@ -251,6 +253,15 @@ template <typename IntTy> static Error getInt(StringRef R, IntTy &Result) {
   return Error::success();
 }
 
+template <typename IntTy>
+static Error getIntForAddrSpace(StringRef R, IntTy &Result) {
+  if (R.equals("neg"))
+    Result = -1;
+  else
+    return getInt<IntTy>(R, Result);
+  return Error::success();
+}
+
 /// Get an unsigned integer representing the number of bits and convert it into
 /// bytes. Error out of not a byte width multiple.
 template <typename IntTy>
@@ -502,6 +513,31 @@ Error DataLayout::parseSpecifier(StringRef Desc) {
         return Err;
       break;
     }
+    case 'z': {
+      unsigned AddrSpace = 0;
+      if (!Tok.empty())
+        if (Error Err = getInt(Tok, AddrSpace))
+          return Err;
+      if (!isUInt<24>(AddrSpace))
+        return reportError("Invalid address space, must be a 24-bit integer");
+
+      if (Rest.empty())
+        return reportError("Missing address space specification for pointer in "
+                           "datalayout string");
+      if (Error Err = ::split(Rest, ':', Split))
+        return Err;
+      int64_t Value;
+      if (Error Err = getIntForAddrSpace(Tok, Value))
+        return Err;
+
+      // for default address space e.g., z:0
+      if (AddrSpace == 0) {
+        setNullPointerValue(INT_MAX, 0);
+        break;
+      }
+      setNullPointerValue(AddrSpace, Value);
+      break;
+    }
     case 'G': { // Default address space for global variables.
       if (Error Err = getAddrSpace(Tok, DefaultGlobalsAddrSpace))
         return Err;

llvm/test/CodeGen/AMDGPU/datalayout-non-zero-lds-value.ll

@@ -0,0 +1,20 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt < %s -S -mtriple=amdgcn-- | FileCheck %s
+
+; CHECK: target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128-p9:192:256:256:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-z:0-z2:neg-z3:neg-z5:neg-S32-A5-G1-ni:7:8:9"
+target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128-p9:192:256:256:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-z:0-z2:neg-z3:neg-z5:neg-S32-A5-G1-ni:7:8:9"
+@lds = addrspace(3) global [8 x i32] [i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8]
+
+define amdgpu_kernel void @load_init_lds_global(ptr addrspace(1) %out, i1 %p) {
+; CHECK-LABEL: define amdgpu_kernel void @load_init_lds_global(
+; CHECK-SAME: ptr addrspace(1) [[OUT:%.*]], i1 [[P:%.*]]) {
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr [8 x i32], ptr addrspace(3) @lds, i32 0, i32 10
+; CHECK-NEXT:    [[LD:%.*]] = load i32, ptr addrspace(3) [[GEP]], align 4
+; CHECK-NEXT:    store i32 [[LD]], ptr addrspace(1) [[OUT]], align 4
+; CHECK-NEXT:    ret void
+;
+  %gep = getelementptr [8 x i32], ptr addrspace(3) @lds, i32 0, i32 10
+  %ld = load i32, ptr addrspace(3) %gep
+  store i32 %ld, ptr addrspace(1) %out
+  ret void
+}