Merged master:0eab9d582381 into amd-gfx:9a9ec94ec7f5

Local branch amd-gfx 9a9ec94 Merged master:8c05c7c8d87c into amd-gfx:1cc7c9fb5e62 Remote branch master 0eab9d5 [SCEV] Verify that all mapped SCEV AddRecs refer to valid loops.
jaebaek · Sep 30, 2020 · 26656ef · 26656ef
2 parents 9a9ec94 + 0eab9d5
commit 26656ef
Show file tree

Hide file tree

Showing 59 changed files with 4,785 additions and 115 deletions.
diff --git a/clang/include/clang/Basic/BuiltinsX86.def b/clang/include/clang/Basic/BuiltinsX86.def
@@ -1900,6 +1900,25 @@ TARGET_BUILTIN(__builtin_ia32_invpcid, "vUiv*", "nc", "invpcid")
 TARGET_BUILTIN(__builtin_ia32_enqcmd, "Ucv*vC*", "n", "enqcmd")
 TARGET_BUILTIN(__builtin_ia32_enqcmds, "Ucv*vC*", "n", "enqcmd")
 
+// KEY LOCKER
+TARGET_BUILTIN(__builtin_ia32_loadiwkey, "vUiV2OiV2OiV2Oi", "nV:128:", "kl")
+TARGET_BUILTIN(__builtin_ia32_encodekey128,
+               "UiUiV2OiV2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*", "nV:128:", "kl")
+TARGET_BUILTIN(__builtin_ia32_encodekey256,
+               "UiUiV2OiV2OiV2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*", "nV:128:", "kl")
+TARGET_BUILTIN(__builtin_ia32_aesenc128kl, "UcV2Oi*V2OivC*", "nV:128:", "kl")
+TARGET_BUILTIN(__builtin_ia32_aesenc256kl, "UcV2Oi*V2OivC*", "nV:128:", "kl")
+TARGET_BUILTIN(__builtin_ia32_aesdec128kl, "UcV2Oi*V2OivC*", "nV:128:", "kl")
+TARGET_BUILTIN(__builtin_ia32_aesdec256kl, "UcV2Oi*V2OivC*", "nV:128:", "kl")
+TARGET_BUILTIN(__builtin_ia32_aesencwide128kl,
+               "UcvC*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2OiV2OiV2OiV2OiV2OiV2OiV2OiV2Oi", "nV:128:", "kl,widekl")
+TARGET_BUILTIN(__builtin_ia32_aesencwide256kl,
+               "UcvC*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2OiV2OiV2OiV2OiV2OiV2OiV2OiV2Oi", "nV:128:", "kl,widekl")
+TARGET_BUILTIN(__builtin_ia32_aesdecwide128kl,
+               "UcvC*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2OiV2OiV2OiV2OiV2OiV2OiV2OiV2Oi", "nV:128:", "kl,widekl")
+TARGET_BUILTIN(__builtin_ia32_aesdecwide256kl,
+               "UcvC*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2Oi*V2OiV2OiV2OiV2OiV2OiV2OiV2OiV2Oi", "nV:128:", "kl,widekl")
+
 // SERIALIZE
 TARGET_BUILTIN(__builtin_ia32_serialize, "v", "n", "serialize")
 

diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
@@ -3253,6 +3253,10 @@ def minvpcid : Flag<["-"], "minvpcid">, Group<m_x86_Features_Group>;
 def mno_invpcid : Flag<["-"], "mno-invpcid">, Group<m_x86_Features_Group>;
 def mgfni : Flag<["-"], "mgfni">, Group<m_x86_Features_Group>;
 def mno_gfni : Flag<["-"], "mno-gfni">, Group<m_x86_Features_Group>;
+def mkl : Flag<["-"], "mkl">, Group<m_x86_Features_Group>;
+def mno_kl : Flag<["-"], "mno-kl">, Group<m_x86_Features_Group>;
+def mwidekl : Flag<["-"], "mwidekl">, Group<m_x86_Features_Group>;
+def mno_widekl : Flag<["-"], "mno-widekl">, Group<m_x86_Features_Group>;
 def mlwp : Flag<["-"], "mlwp">, Group<m_x86_Features_Group>;
 def mno_lwp : Flag<["-"], "mno-lwp">, Group<m_x86_Features_Group>;
 def mlzcnt : Flag<["-"], "mlzcnt">, Group<m_x86_Features_Group>;

diff --git a/clang/lib/Basic/Targets/X86.cpp b/clang/lib/Basic/Targets/X86.cpp
@@ -276,6 +276,10 @@ bool X86TargetInfo::handleTargetFeatures(std::vector<std::string> &Features,
       HasCLDEMOTE = true;
     } else if (Feature == "+rdpid") {
       HasRDPID = true;
+    } else if (Feature == "+kl") {
+      HasKL = true;
+    } else if (Feature == "+widekl") {
+      HasWIDEKL = true;
     } else if (Feature == "+retpoline-external-thunk") {
       HasRetpolineExternalThunk = true;
     } else if (Feature == "+sahf") {
@@ -678,6 +682,10 @@ void X86TargetInfo::getTargetDefines(const LangOptions &Opts,
     Builder.defineMacro("__PREFETCHWT1__");
   if (HasCLZERO)
     Builder.defineMacro("__CLZERO__");
+  if (HasKL)
+    Builder.defineMacro("__KL__");
+  if (HasWIDEKL)
+    Builder.defineMacro("__WIDEKL__");
   if (HasRDPID)
     Builder.defineMacro("__RDPID__");
   if (HasCLDEMOTE)
@@ -833,6 +841,8 @@ bool X86TargetInfo::isValidFeatureName(StringRef Name) const {
       .Case("fxsr", true)
       .Case("gfni", true)
       .Case("invpcid", true)
+      .Case("kl", true)
+      .Case("widekl", true)
       .Case("lwp", true)
       .Case("lzcnt", true)
       .Case("mmx", true)
@@ -919,6 +929,8 @@ bool X86TargetInfo::hasFeature(StringRef Feature) const {
       .Case("fxsr", HasFXSR)
       .Case("gfni", HasGFNI)
       .Case("invpcid", HasINVPCID)
+      .Case("kl", HasKL)
+      .Case("widekl", HasWIDEKL)
       .Case("lwp", HasLWP)
       .Case("lzcnt", HasLZCNT)
       .Case("mm3dnow", MMX3DNowLevel >= AMD3DNow)

diff --git a/clang/lib/Basic/Targets/X86.h b/clang/lib/Basic/Targets/X86.h
@@ -127,6 +127,8 @@ class LLVM_LIBRARY_VISIBILITY X86TargetInfo : public TargetInfo {
   bool HasPTWRITE = false;
   bool HasINVPCID = false;
   bool HasENQCMD = false;
+  bool HasKL = false;      // For key locker
+  bool HasWIDEKL = false; // For wide key locker
   bool HasAMXTILE = false;
   bool HasAMXINT8 = false;
   bool HasAMXBF16 = false;

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -14037,6 +14037,93 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
   case X86::BI__builtin_ia32_psubusb128:
   case X86::BI__builtin_ia32_psubusw128:
     return EmitX86BinaryIntrinsic(*this, Ops, Intrinsic::usub_sat);
+  case X86::BI__builtin_ia32_encodekey128:
+  case X86::BI__builtin_ia32_encodekey256:
+  case X86::BI__builtin_ia32_aesenc128kl:
+  case X86::BI__builtin_ia32_aesdec128kl:
+  case X86::BI__builtin_ia32_aesenc256kl:
+  case X86::BI__builtin_ia32_aesdec256kl:
+  case X86::BI__builtin_ia32_aesencwide128kl:
+  case X86::BI__builtin_ia32_aesdecwide128kl:
+  case X86::BI__builtin_ia32_aesencwide256kl:
+  case X86::BI__builtin_ia32_aesdecwide256kl: {
+    int FirstReturnOp;
+    int ResultCount;
+    SmallVector<Value*, 9> InOps;
+    unsigned ID;
+
+    switch (BuiltinID) {
+    default: llvm_unreachable("Unsupported intrinsic!");
+    case X86::BI__builtin_ia32_encodekey128:
+      ID = Intrinsic::x86_encodekey128;
+      InOps = {Ops[0], Ops[1]};
+      FirstReturnOp = 2;
+      ResultCount = 6;
+      break;
+    case X86::BI__builtin_ia32_encodekey256:
+      ID = Intrinsic::x86_encodekey256;
+      InOps = {Ops[0], Ops[1], Ops[2]};
+      FirstReturnOp = 3;
+      ResultCount = 7;
+      break;
+    case X86::BI__builtin_ia32_aesenc128kl:
+    case X86::BI__builtin_ia32_aesdec128kl:
+    case X86::BI__builtin_ia32_aesenc256kl:
+    case X86::BI__builtin_ia32_aesdec256kl: {
+      InOps = {Ops[1], Ops[2]};
+      FirstReturnOp = 0;
+      ResultCount = 1;
+      switch (BuiltinID) {
+      case X86::BI__builtin_ia32_aesenc128kl:
+        ID = Intrinsic::x86_aesenc128kl;
+        break;
+      case X86::BI__builtin_ia32_aesdec128kl:
+        ID = Intrinsic::x86_aesdec128kl;
+        break;
+      case X86::BI__builtin_ia32_aesenc256kl:
+        ID = Intrinsic::x86_aesenc256kl;
+        break;
+      case X86::BI__builtin_ia32_aesdec256kl:
+        ID = Intrinsic::x86_aesdec256kl;
+        break;
+      }
+      break;
+    }
+    case X86::BI__builtin_ia32_aesencwide128kl:
+    case X86::BI__builtin_ia32_aesdecwide128kl:
+    case X86::BI__builtin_ia32_aesencwide256kl:
+    case X86::BI__builtin_ia32_aesdecwide256kl: {
+      InOps = {Ops[0], Ops[9], Ops[10], Ops[11], Ops[12], Ops[13],
+               Ops[14], Ops[15], Ops[16]};
+      FirstReturnOp = 1;
+      ResultCount = 8;
+      switch (BuiltinID) {
+      case X86::BI__builtin_ia32_aesencwide128kl:
+        ID = Intrinsic::x86_aesencwide128kl;
+        break;
+      case X86::BI__builtin_ia32_aesdecwide128kl:
+        ID = Intrinsic::x86_aesdecwide128kl;
+        break;
+      case X86::BI__builtin_ia32_aesencwide256kl:
+        ID = Intrinsic::x86_aesencwide256kl;
+        break;
+      case X86::BI__builtin_ia32_aesdecwide256kl:
+        ID = Intrinsic::x86_aesdecwide256kl;
+        break;
+      }
+      break;
+    }
+    }
+
+    Value *Call = Builder.CreateCall(CGM.getIntrinsic(ID), InOps);
+
+    for (int i = 0; i < ResultCount; ++i) {
+      Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, i + 1),
+                                        Ops[FirstReturnOp + i]);
+    }
+
+    return Builder.CreateExtractValue(Call, 0);
+  }
   }
 }
 

diff --git a/clang/lib/Headers/CMakeLists.txt b/clang/lib/Headers/CMakeLists.txt
@@ -72,6 +72,8 @@ set(files
   inttypes.h
   invpcidintrin.h
   iso646.h
+  keylockerintrin.h
+  keylocker_wide_intrin.h
   limits.h
   lwpintrin.h
   lzcntintrin.h

diff --git a/clang/lib/Headers/immintrin.h b/clang/lib/Headers/immintrin.h
@@ -471,6 +471,16 @@ _storebe_i64(void * __P, long long __D) {
 #include <invpcidintrin.h>
 #endif
 
+#if !(defined(_MSC_VER) || defined(__SCE__)) || __has_feature(modules) ||      \
+    defined(__KL__)
+#include <keylockerintrin.h>
+#endif
+
+#if !(defined(_MSC_VER) || defined(__SCE__)) || __has_feature(modules) ||      \
+    defined(__WIDEKL__)
+#include <keylocker_wide_intrin.h>
+#endif
+
 #if !(defined(_MSC_VER) || defined(__SCE__)) || __has_feature(modules) ||      \
     defined(__AMXTILE__) || defined(__AMXINT8__) || defined(__AMXBF16__)
 #include <amxintrin.h>