Vendor import of llvm trunk r304222:

https://llvm.org/svn/llvm-project/llvm/trunk@304222

docs/Proposals/VectorizationPlan.rst

@@ -0,0 +1,182 @@
+==================
+Vectorization Plan
+==================
+
+.. contents::
+   :local:
+
+Abstract
+========
+The vectorization transformation can be rather complicated, involving several
+potential alternatives, especially for outer-loops [1]_ but also possibly for
+innermost loops. These alternatives may have significant performance impact,
+both positive and negative. A cost model is therefore employed to identify the
+best alternative, including the alternative of avoiding any transformation
+altogether.
+
+The Vectorization Plan is an explicit model for describing vectorization
+candidates. It serves for both optimizing candidates including estimating their
+cost reliably, and for performing their final translation into IR. This
+facilitates dealing with multiple vectorization candidates.
+
+High-level Design
+=================
+
+Vectorization Workflow
+----------------------
+VPlan-based vectorization involves three major steps, taking a "scenario-based
+approach" to vectorization planning:
+
+1. Legal Step: check if a loop can be legally vectorized; encode contraints and
+   artifacts if so.
+2. Plan Step:
+
+   a. Build initial VPlans following the constraints and decisions taken by
+      Legal Step 1, and compute their cost.
+   b. Apply optimizations to the VPlans, possibly forking additional VPlans.
+      Prune sub-optimal VPlans having relatively high cost.
+3. Execute Step: materialize the best VPlan. Note that this is the only step
+   that modifies the IR.
+
+Design Guidelines
+-----------------
+In what follows, the term "input IR" refers to code that is fed into the
+vectorizer whereas the term "output IR" refers to code that is generated by the
+vectorizer. The output IR contains code that has been vectorized or "widened"
+according to a loop Vectorization Factor (VF), and/or loop unroll-and-jammed
+according to an Unroll Factor (UF).
+The design of VPlan follows several high-level guidelines:
+
+1. Analysis-like: building and manipulating VPlans must not modify the input IR.
+   In particular, if the best option is not to vectorize at all, the
+   vectorization process terminates before reaching Step 3, and compilation
+   should proceed as if VPlans had not been built.
+
+2. Align Cost & Execute: each VPlan must support both estimating the cost and
+   generating the output IR code, such that the cost estimation evaluates the
+   to-be-generated code reliably.
+
+3. Support vectorizing additional constructs:
+
+   a. Outer-loop vectorization. In particular, VPlan must be able to model the
+      control-flow of the output IR which may include multiple basic-blocks and
+      nested loops.
+   b. SLP vectorization.
+   c. Combinations of the above, including nested vectorization: vectorizing
+      both an inner loop and an outer-loop at the same time (each with its own
+      VF and UF), mixed vectorization: vectorizing a loop with SLP patterns
+      inside [4]_, (re)vectorizing input IR containing vector code.
+   d. Function vectorization [2]_.
+
+4. Support multiple candidates efficiently. In particular, similar candidates
+   related to a range of possible VF's and UF's must be represented efficiently.
+   Potential versioning needs to be supported efficiently.
+
+5. Support vectorizing idioms, such as interleaved groups of strided loads or
+   stores. This is achieved by modeling a sequence of output instructions using
+   a "Recipe", which is responsible for computing its cost and generating its
+   code.
+
+6. Encapsulate Single-Entry Single-Exit regions (SESE). During vectorization
+   such regions may need to be, for example, predicated and linearized, or
+   replicated VF*UF times to handle scalarized and predicated instructions.
+   Innerloops are also modelled as SESE regions.
+
+Low-level Design
+================
+The low-level design of VPlan comprises of the following classes.
+
+:LoopVectorizationPlanner:
+  A LoopVectorizationPlanner is designed to handle the vectorization of a loop
+  or a loop nest. It can construct, optimize and discard one or more VPlans,
+  each VPlan modelling a distinct way to vectorize the loop or the loop nest.
+  Once the best VPlan is determined, including the best VF and UF, this VPlan
+  drives the generation of output IR.
+
+:VPlan:
+  A model of a vectorized candidate for a given input IR loop or loop nest. This
+  candidate is represented using a Hierarchical CFG. VPlan supports estimating
+  the cost and driving the generation of the output IR code it represents.
+
+:Hierarchical CFG:
+  A control-flow graph whose nodes are basic-blocks or Hierarchical CFG's. The
+  Hierarchical CFG data structure is similar to the Tile Tree [5]_, where
+  cross-Tile edges are lifted to connect Tiles instead of the original
+  basic-blocks as in Sharir [6]_, promoting the Tile encapsulation. The terms
+  Region and Block are used rather than Tile [5]_ to avoid confusion with loop
+  tiling.
+
+:VPBlockBase:
+  The building block of the Hierarchical CFG. A pure-virtual base-class of
+  VPBasicBlock and VPRegionBlock, see below. VPBlockBase models the hierarchical
+  control-flow relations with other VPBlocks. Note that in contrast to the IR
+  BasicBlock, a VPBlockBase models its control-flow successors and predecessors
+  directly, rather than through a Terminator branch or through predecessor
+  branches that "use" the VPBlockBase.
+
+:VPBasicBlock:
+  VPBasicBlock is a subclass of VPBlockBase, and serves as the leaves of the
+  Hierarchical CFG. It represents a sequence of output IR instructions that will
+  appear consecutively in an output IR basic-block. The instructions of this
+  basic-block originate from one or more VPBasicBlocks. VPBasicBlock holds a
+  sequence of zero or more VPRecipes that model the cost and generation of the
+  output IR instructions.
+
+:VPRegionBlock:
+  VPRegionBlock is a subclass of VPBlockBase. It models a collection of
+  VPBasicBlocks and VPRegionBlocks which form a SESE subgraph of the output IR
+  CFG. A VPRegionBlock may indicate that its contents are to be replicated a
+  constant number of times when output IR is generated, effectively representing
+  a loop with constant trip-count that will be completely unrolled. This is used
+  to support scalarized and predicated instructions with a single model for
+  multiple candidate VF's and UF's.
+
+:VPRecipeBase:
+  A pure-virtual base class modeling a sequence of one or more output IR
+  instructions, possibly based on one or more input IR instructions. These
+  input IR instructions are referred to as "Ingredients" of the Recipe. A Recipe
+  may specify how its ingredients are to be transformed to produce the output IR
+  instructions; e.g., cloned once, replicated multiple times or widened
+  according to selected VF.
+
+:VPTransformState:
+  Stores information used for generating output IR, passed from
+  LoopVectorizationPlanner to its selected VPlan for execution, and used to pass
+  additional information down to VPBlocks and VPRecipes.
+
+Related LLVM components
+-----------------------
+1. SLP Vectorizer: one can compare the VPlan model with LLVM's existing SLP
+   tree, where TSLP [3]_ adds Plan Step 2.b.
+
+2. RegionInfo: one can compare VPlan's H-CFG with the Region Analysis as used by
+   Polly [7]_.
+
+References
+----------
+.. [1] "Outer-loop vectorization: revisited for short SIMD architectures", Dorit
+    Nuzman and Ayal Zaks, PACT 2008.
+
+.. [2] "Proposal for function vectorization and loop vectorization with function
+    calls", Xinmin Tian, [`cfe-dev
+    <http://lists.llvm.org/pipermail/cfe-dev/2016-March/047732.html>`_].,
+    March 2, 2016.
+    See also `review <https://reviews.llvm.org/D22792>`_.
+
+.. [3] "Throttling Automatic Vectorization: When Less is More", Vasileios
+    Porpodas and Tim Jones, PACT 2015 and LLVM Developers' Meeting 2015.
+
+.. [4] "Exploiting mixed SIMD parallelism by reducing data reorganization
+    overhead", Hao Zhou and Jingling Xue, CGO 2016.
+
+.. [5] "Register Allocation via Hierarchical Graph Coloring", David Callahan and
+    Brian Koblenz, PLDI 1991
+
+.. [6] "Structural analysis: A new approach to flow analysis in optimizing
+    compilers", M. Sharir, Journal of Computer Languages, Jan. 1980
+
+.. [7] "Enabling Polyhedral Optimizations in LLVM", Tobias Grosser, Diploma
+    thesis, 2011.
+
+.. [8] "Introducing VPlan to the Loop Vectorizer", Gil Rapaport and Ayal Zaks,
+    European LLVM Developers' Meeting 2017.

docs/Vectorizers.rst

@@ -382,6 +382,17 @@ And Linpack-pc with the same configuration. Result is Mflops, higher is better.
 
 .. image:: linpack-pc.png
 
+Ongoing Development Directions
+------------------------------
+
+.. toctree::
+   :hidden:
+
+   Proposals/VectorizationPlan
+
+:doc:`Proposals/VectorizationPlan`
+   Modeling the process and upgrading the infrastructure of LLVM's Loop Vectorizer.
+
 .. _slp-vectorizer:
 
 The SLP Vectorizer

docs/index.rst

@@ -528,6 +528,7 @@ can be better.
 
    CodeOfConduct
    Proposals/GitHubMove
+   Proposals/VectorizationPlan
 
 :doc:`CodeOfConduct`
    Proposal to adopt a code of conduct on the LLVM social spaces (lists, events,
@@ -536,6 +537,8 @@ can be better.
 :doc:`Proposals/GitHubMove`
    Proposal to move from SVN/Git to GitHub.
 
+:doc:`Proposals/VectorizationPlan`
+   Proposal to model the process and upgrade the infrastructure of LLVM's Loop Vectorizer.
 
 Indices and tables
 ==================

include/llvm/Analysis/ScalarEvolution.h

@@ -1536,8 +1536,7 @@ class ScalarEvolution {
   /// Determine if the SCEV can be evaluated at loop's entry. It is true if it
   /// doesn't depend on a SCEVUnknown of an instruction which is dominated by
   /// the header of loop L.
-  bool isAvailableAtLoopEntry(const SCEV *S, const Loop *L, DominatorTree &DT,
-                              LoopInfo &LI);
+  bool isAvailableAtLoopEntry(const SCEV *S, const Loop *L);
 
   /// Return true if the given SCEV changes value in a known way in the
   /// specified loop.  This property being true implies that the value is

include/llvm/CodeGen/DIE.h

@@ -383,11 +383,11 @@ class DIEValue {
       return;
 #define HANDLE_DIEVALUE_SMALL(T)                                               \
   case is##T:                                                                  \
-    destruct<DIE##T>();
+    destruct<DIE##T>();                                                        \
     return;
 #define HANDLE_DIEVALUE_LARGE(T)                                               \
   case is##T:                                                                  \
-    destruct<const DIE##T *>();
+    destruct<const DIE##T *>();                                                \
     return;
 #include "llvm/CodeGen/DIEValue.def"
     }

include/llvm/DebugInfo/CodeView/CodeView.h

@@ -13,6 +13,8 @@
 #include <cinttypes>
 #include <type_traits>
 
+#include "llvm/Support/Endian.h"
+
 namespace llvm {
 namespace codeview {
 
@@ -291,7 +293,7 @@ enum class ModifierOptions : uint16_t {
 };
 CV_DEFINE_ENUM_CLASS_FLAGS_OPERATORS(ModifierOptions)
 
-enum class ModuleDebugFragmentKind : uint32_t {
+enum class DebugSubsectionKind : uint32_t {
   None = 0,
   Symbols = 0xf1,
   Lines = 0xf2,
@@ -550,6 +552,24 @@ enum LineFlags : uint16_t {
   LF_None = 0,
   LF_HaveColumns = 1, // CV_LINES_HAVE_COLUMNS
 };
+
+/// Data in the the SUBSEC_FRAMEDATA subection.
+struct FrameData {
+  support::ulittle32_t RvaStart;
+  support::ulittle32_t CodeSize;
+  support::ulittle32_t LocalSize;
+  support::ulittle32_t ParamsSize;
+  support::ulittle32_t MaxStackSize;
+  support::ulittle32_t FrameFunc;
+  support::ulittle16_t PrologSize;
+  support::ulittle16_t SavedRegsSize;
+  support::ulittle32_t Flags;
+  enum : uint32_t {
+    HasSEH = 1 << 0,
+    HasEH = 1 << 1,
+    IsFunctionStart = 1 << 2,
+  };
+};
 }
 }
 

include/llvm/DebugInfo/CodeView/DebugChecksumsSubsection.h

@@ -1,4 +1,4 @@
-//===- ModuleDebugFileChecksumFragment.h ------------------------*- C++ -*-===//
+//===- DebugChecksumsSubsection.h -------------------------------*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -7,12 +7,12 @@
 //
 //===----------------------------------------------------------------------===//
 
-#ifndef LLVM_DEBUGINFO_CODEVIEW_MODULEDEBUGFILECHECKSUMFRAGMENT_H
-#define LLVM_DEBUGINFO_CODEVIEW_MODULEDEBUGFILECHECKSUMFRAGMENT_H
+#ifndef LLVM_DEBUGINFO_CODEVIEW_DEBUGCHECKSUMSSUBSECTION_H
+#define LLVM_DEBUGINFO_CODEVIEW_DEBUGCHECKSUMSSUBSECTION_H
 
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/DenseMap.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugFragment.h"
+#include "llvm/DebugInfo/CodeView/DebugSubsection.h"
 #include "llvm/Support/Allocator.h"
 #include "llvm/Support/BinaryStreamArray.h"
 #include "llvm/Support/BinaryStreamReader.h"
@@ -21,7 +21,7 @@
 namespace llvm {
 namespace codeview {
 
-class StringTable;
+class DebugStringTableSubsection;
 
 struct FileChecksumEntry {
   uint32_t FileNameOffset;    // Byte offset of filename in global stringtable.
@@ -43,19 +43,22 @@ template <> struct VarStreamArrayExtractor<codeview::FileChecksumEntry> {
 
 namespace llvm {
 namespace codeview {
-class ModuleDebugFileChecksumFragmentRef final : public ModuleDebugFragmentRef {
+class DebugChecksumsSubsectionRef final : public DebugSubsectionRef {
   typedef VarStreamArray<codeview::FileChecksumEntry> FileChecksumArray;
   typedef FileChecksumArray::Iterator Iterator;
 
 public:
-  ModuleDebugFileChecksumFragmentRef()
-      : ModuleDebugFragmentRef(ModuleDebugFragmentKind::FileChecksums) {}
+  DebugChecksumsSubsectionRef()
+      : DebugSubsectionRef(DebugSubsectionKind::FileChecksums) {}
 
-  static bool classof(const ModuleDebugFragmentRef *S) {
-    return S->kind() == ModuleDebugFragmentKind::FileChecksums;
+  static bool classof(const DebugSubsectionRef *S) {
+    return S->kind() == DebugSubsectionKind::FileChecksums;
   }
 
+  bool valid() const { return Checksums.valid(); }
+
   Error initialize(BinaryStreamReader Reader);
+  Error initialize(BinaryStreamRef Stream);
 
   Iterator begin() { return Checksums.begin(); }
   Iterator end() { return Checksums.end(); }
@@ -66,23 +69,23 @@ class ModuleDebugFileChecksumFragmentRef final : public ModuleDebugFragmentRef {
   FileChecksumArray Checksums;
 };
 
-class ModuleDebugFileChecksumFragment final : public ModuleDebugFragment {
+class DebugChecksumsSubsection final : public DebugSubsection {
 public:
-  explicit ModuleDebugFileChecksumFragment(StringTable &Strings);
+  explicit DebugChecksumsSubsection(DebugStringTableSubsection &Strings);
 
-  static bool classof(const ModuleDebugFragment *S) {
-    return S->kind() == ModuleDebugFragmentKind::FileChecksums;
+  static bool classof(const DebugSubsection *S) {
+    return S->kind() == DebugSubsectionKind::FileChecksums;
   }
 
   void addChecksum(StringRef FileName, FileChecksumKind Kind,
                    ArrayRef<uint8_t> Bytes);
 
-  uint32_t calculateSerializedLength() override;
-  Error commit(BinaryStreamWriter &Writer) override;
+  uint32_t calculateSerializedSize() const override;
+  Error commit(BinaryStreamWriter &Writer) const override;
   uint32_t mapChecksumOffset(StringRef FileName) const;
 
 private:
-  StringTable &Strings;
+  DebugStringTableSubsection &Strings;
 
   DenseMap<uint32_t, uint32_t> OffsetMap;
   uint32_t SerializedSize = 0;

include/llvm/DebugInfo/CodeView/DebugFrameDataSubsection.h

@@ -0,0 +1,59 @@
+//===- DebugFrameDataSubsection.h ------------------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_DEBUGINFO_CODEVIEW_DEBUGFRAMEDATASUBSECTION_H
+#define LLVM_DEBUGINFO_CODEVIEW_DEBUGFRAMEDATASUBSECTION_H
+
+#include "llvm/DebugInfo/CodeView/CodeView.h"
+#include "llvm/DebugInfo/CodeView/DebugSubsection.h"
+#include "llvm/Support/BinaryStreamReader.h"
+#include "llvm/Support/Error.h"
+
+namespace llvm {
+namespace codeview {
+class DebugFrameDataSubsectionRef final : public DebugSubsectionRef {
+public:
+  DebugFrameDataSubsectionRef()
+      : DebugSubsectionRef(DebugSubsectionKind::FrameData) {}
+  static bool classof(const DebugSubsection *S) {
+    return S->kind() == DebugSubsectionKind::FrameData;
+  }
+
+  Error initialize(BinaryStreamReader Reader);
+
+  FixedStreamArray<FrameData>::Iterator begin() const { return Frames.begin(); }
+  FixedStreamArray<FrameData>::Iterator end() const { return Frames.end(); }
+
+  const void *getRelocPtr() const { return RelocPtr; }
+
+private:
+  const uint32_t *RelocPtr = nullptr;
+  FixedStreamArray<FrameData> Frames;
+};
+
+class DebugFrameDataSubsection final : public DebugSubsection {
+public:
+  DebugFrameDataSubsection()
+      : DebugSubsection(DebugSubsectionKind::FrameData) {}
+  static bool classof(const DebugSubsection *S) {
+    return S->kind() == DebugSubsectionKind::FrameData;
+  }
+
+  uint32_t calculateSerializedSize() const override;
+  Error commit(BinaryStreamWriter &Writer) const override;
+
+  void addFrameData(const FrameData &Frame);
+
+private:
+  std::vector<FrameData> Frames;
+};
+}
+}
+
+#endif

include/llvm/DebugInfo/CodeView/DebugInlineeLinesSubsection.h

@@ -1,4 +1,4 @@
-//===- ModuleDebugInlineeLinesFragment.h ------------------------*- C++ -*-===//
+//===- DebugInlineeLinesSubsection.h ----------------------------*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -7,11 +7,11 @@
 //
 //===----------------------------------------------------------------------===//
 
-#ifndef LLVM_DEBUGINFO_CODEVIEW_MODULEDEBUGINLINEELINESFRAGMENT_H
-#define LLVM_DEBUGINFO_CODEVIEW_MODULEDEBUGINLINEELINESFRAGMENT_H
+#ifndef LLVM_DEBUGINFO_CODEVIEW_BUGINLINEELINESSUBSECTION_H
+#define LLVM_DEBUGINFO_CODEVIEW_BUGINLINEELINESSUBSECTION_H
 
+#include "llvm/DebugInfo/CodeView/DebugSubsection.h"
 #include "llvm/DebugInfo/CodeView/Line.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugFragment.h"
 #include "llvm/Support/BinaryStreamArray.h"
 #include "llvm/Support/BinaryStreamReader.h"
 #include "llvm/Support/Error.h"
@@ -19,9 +19,8 @@
 namespace llvm {
 namespace codeview {
 
-class ModuleDebugInlineeLineFragmentRef;
-class ModuleDebugFileChecksumFragment;
-class StringTable;
+class DebugInlineeLinesSubsectionsRef;
+class DebugChecksumsSubsection;
 
 enum class InlineeLinesSignature : uint32_t {
   Normal,    // CV_INLINEE_SOURCE_LINE_SIGNATURE
@@ -51,15 +50,15 @@ template <> struct VarStreamArrayExtractor<codeview::InlineeSourceLine> {
 };
 
 namespace codeview {
-class ModuleDebugInlineeLineFragmentRef final : public ModuleDebugFragmentRef {
+class DebugInlineeLinesSubsectionRef final : public DebugSubsectionRef {
   typedef VarStreamArray<InlineeSourceLine> LinesArray;
   typedef LinesArray::Iterator Iterator;
 
 public:
-  ModuleDebugInlineeLineFragmentRef();
+  DebugInlineeLinesSubsectionRef();
 
-  static bool classof(const ModuleDebugFragmentRef *S) {
-    return S->kind() == ModuleDebugFragmentKind::InlineeLines;
+  static bool classof(const DebugSubsectionRef *S) {
+    return S->kind() == DebugSubsectionKind::InlineeLines;
   }
 
   Error initialize(BinaryStreamReader Reader);
@@ -73,23 +72,23 @@ class ModuleDebugInlineeLineFragmentRef final : public ModuleDebugFragmentRef {
   VarStreamArray<InlineeSourceLine> Lines;
 };
 
-class ModuleDebugInlineeLineFragment final : public ModuleDebugFragment {
+class DebugInlineeLinesSubsection final : public DebugSubsection {
 public:
-  ModuleDebugInlineeLineFragment(ModuleDebugFileChecksumFragment &Checksums,
-                                 bool HasExtraFiles);
+  DebugInlineeLinesSubsection(DebugChecksumsSubsection &Checksums,
+                              bool HasExtraFiles);
 
-  static bool classof(const ModuleDebugFragment *S) {
-    return S->kind() == ModuleDebugFragmentKind::InlineeLines;
+  static bool classof(const DebugSubsection *S) {
+    return S->kind() == DebugSubsectionKind::InlineeLines;
   }
 
-  Error commit(BinaryStreamWriter &Writer) override;
-  uint32_t calculateSerializedLength() override;
+  Error commit(BinaryStreamWriter &Writer) const override;
+  uint32_t calculateSerializedSize() const override;
 
   void addInlineSite(TypeIndex FuncId, StringRef FileName, uint32_t SourceLine);
   void addExtraFile(StringRef FileName);
 
 private:
-  ModuleDebugFileChecksumFragment &Checksums;
+  DebugChecksumsSubsection &Checksums;
 
   bool HasExtraFiles = false;
   uint32_t ExtraFileCount = 0;

include/llvm/DebugInfo/CodeView/DebugLinesSubsection.h

@@ -1,4 +1,4 @@
-//===- ModuleDebugLineFragment.h --------------------------------*- C++ -*-===//
+//===- DebugLinesSubsection.h --------------------------------*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -10,8 +10,8 @@
 #ifndef LLVM_DEBUGINFO_CODEVIEW_MODULEDEBUGLINEFRAGMENT_H
 #define LLVM_DEBUGINFO_CODEVIEW_MODULEDEBUGLINEFRAGMENT_H
 
+#include "llvm/DebugInfo/CodeView/DebugSubsection.h"
 #include "llvm/DebugInfo/CodeView/Line.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugFragment.h"
 #include "llvm/Support/BinaryStreamArray.h"
 #include "llvm/Support/BinaryStreamReader.h"
 #include "llvm/Support/Error.h"
@@ -19,8 +19,8 @@
 namespace llvm {
 namespace codeview {
 
-class ModuleDebugFileChecksumFragment;
-class StringTable;
+class DebugChecksumsSubsection;
+class DebugStringTableSubsection;
 
 // Corresponds to the `CV_DebugSLinesHeader_t` structure.
 struct LineFragmentHeader {
@@ -70,16 +70,16 @@ class LineColumnExtractor {
                        LineColumnEntry &Item, const LineFragmentHeader *Ctx);
 };
 
-class ModuleDebugLineFragmentRef final : public ModuleDebugFragmentRef {
+class DebugLinesSubsectionRef final : public DebugSubsectionRef {
   friend class LineColumnExtractor;
   typedef VarStreamArray<LineColumnEntry, LineColumnExtractor> LineInfoArray;
   typedef LineInfoArray::Iterator Iterator;
 
 public:
-  ModuleDebugLineFragmentRef();
+  DebugLinesSubsectionRef();
 
-  static bool classof(const ModuleDebugFragmentRef *S) {
-    return S->kind() == ModuleDebugFragmentKind::Lines;
+  static bool classof(const DebugSubsectionRef *S) {
+    return S->kind() == DebugSubsectionKind::Lines;
   }
 
   Error initialize(BinaryStreamReader Reader);
@@ -96,7 +96,7 @@ class ModuleDebugLineFragmentRef final : public ModuleDebugFragmentRef {
   LineInfoArray LinesAndColumns;
 };
 
-class ModuleDebugLineFragment final : public ModuleDebugFragment {
+class DebugLinesSubsection final : public DebugSubsection {
   struct Block {
     Block(uint32_t ChecksumBufferOffset)
         : ChecksumBufferOffset(ChecksumBufferOffset) {}
@@ -107,11 +107,11 @@ class ModuleDebugLineFragment final : public ModuleDebugFragment {
   };
 
 public:
-  ModuleDebugLineFragment(ModuleDebugFileChecksumFragment &Checksums,
-                          StringTable &Strings);
+  DebugLinesSubsection(DebugChecksumsSubsection &Checksums,
+                       DebugStringTableSubsection &Strings);
 
-  static bool classof(const ModuleDebugFragment *S) {
-    return S->kind() == ModuleDebugFragmentKind::Lines;
+  static bool classof(const DebugSubsection *S) {
+    return S->kind() == DebugSubsectionKind::Lines;
   }
 
   void createBlock(StringRef FileName);
@@ -119,8 +119,8 @@ class ModuleDebugLineFragment final : public ModuleDebugFragment {
   void addLineAndColumnInfo(uint32_t Offset, const LineInfo &Line,
                             uint32_t ColStart, uint32_t ColEnd);
 
-  uint32_t calculateSerializedLength() override;
-  Error commit(BinaryStreamWriter &Writer) override;
+  uint32_t calculateSerializedSize() const override;
+  Error commit(BinaryStreamWriter &Writer) const override;
 
   void setRelocationAddress(uint16_t Segment, uint16_t Offset);
   void setCodeSize(uint32_t Size);
@@ -129,7 +129,7 @@ class ModuleDebugLineFragment final : public ModuleDebugFragment {
   bool hasColumnInfo() const;
 
 private:
-  ModuleDebugFileChecksumFragment &Checksums;
+  DebugChecksumsSubsection &Checksums;
 
   uint16_t RelocOffset = 0;
   uint16_t RelocSegment = 0;

include/llvm/DebugInfo/CodeView/DebugStringTableSubsection.h

@@ -1,4 +1,4 @@
-//===- StringTable.h - CodeView String Table Reader/Writer ------*- C++ -*-===//
+//===- DebugStringTableSubsection.h - CodeView String Table -----*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -7,12 +7,12 @@
 //
 //===----------------------------------------------------------------------===//
 
-#ifndef LLVM_DEBUGINFO_CODEVIEW_STRINGTABLE_H
-#define LLVM_DEBUGINFO_CODEVIEW_STRINGTABLE_H
+#ifndef LLVM_DEBUGINFO_CODEVIEW_DEBUGSTRINGTABLESUBSECTION_H
+#define LLVM_DEBUGINFO_CODEVIEW_DEBUGSTRINGTABLESUBSECTION_H
 
 #include "llvm/ADT/StringMap.h"
 #include "llvm/ADT/StringRef.h"
-
+#include "llvm/DebugInfo/CodeView/DebugSubsection.h"
 #include "llvm/Support/BinaryStreamRef.h"
 #include "llvm/Support/Error.h"
 
@@ -28,11 +28,15 @@ namespace codeview {
 
 /// Represents a read-only view of a CodeView string table.  This is a very
 /// simple flat buffer consisting of null-terminated strings, where strings
-/// are retrieved by their offset in the buffer.  StringTableRef does not own
-/// the underlying storage for the buffer.
-class StringTableRef {
+/// are retrieved by their offset in the buffer.  DebugStringTableSubsectionRef
+/// does not own the underlying storage for the buffer.
+class DebugStringTableSubsectionRef : public DebugSubsectionRef {
 public:
-  StringTableRef();
+  DebugStringTableSubsectionRef();
+
+  static bool classof(const DebugSubsectionRef *S) {
+    return S->kind() == DebugSubsectionKind::StringTable;
+  }
 
   Error initialize(BinaryStreamRef Contents);
 
@@ -44,11 +48,18 @@ class StringTableRef {
   BinaryStreamRef Stream;
 };
 
-/// Represents a read-write view of a CodeView string table.  StringTable owns
-/// the underlying storage for the table, and is capable of serializing the
-/// string table into a format understood by StringTableRef.
-class StringTable {
+/// Represents a read-write view of a CodeView string table.
+/// DebugStringTableSubsection owns the underlying storage for the table, and is
+/// capable of serializing the string table into a format understood by
+/// DebugStringTableSubsectionRef.
+class DebugStringTableSubsection : public DebugSubsection {
 public:
+  DebugStringTableSubsection();
+
+  static bool classof(const DebugSubsection *S) {
+    return S->kind() == DebugSubsectionKind::StringTable;
+  }
+
   // If string S does not exist in the string table, insert it.
   // Returns the ID for S.
   uint32_t insert(StringRef S);
@@ -56,8 +67,8 @@ class StringTable {
   // Return the ID for string S.  Assumes S exists in the table.
   uint32_t getStringId(StringRef S) const;
 
-  uint32_t calculateSerializedSize() const;
-  Error commit(BinaryStreamWriter &Writer) const;
+  uint32_t calculateSerializedSize() const override;
+  Error commit(BinaryStreamWriter &Writer) const override;
 
   uint32_t size() const;
 

include/llvm/DebugInfo/CodeView/DebugSubsection.h

@@ -0,0 +1,52 @@
+//===- DebugSubsection.h ------------------------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_DEBUGINFO_CODEVIEW_MODULEDEBUGFRAGMENT_H
+#define LLVM_DEBUGINFO_CODEVIEW_MODULEDEBUGFRAGMENT_H
+
+#include "llvm/DebugInfo/CodeView/CodeView.h"
+#include "llvm/Support/BinaryStreamWriter.h"
+#include "llvm/Support/Casting.h"
+
+namespace llvm {
+namespace codeview {
+
+class DebugSubsectionRef {
+public:
+  explicit DebugSubsectionRef(DebugSubsectionKind Kind) : Kind(Kind) {}
+  virtual ~DebugSubsectionRef();
+
+  static bool classof(const DebugSubsectionRef *S) { return true; }
+
+  DebugSubsectionKind kind() const { return Kind; }
+
+protected:
+  DebugSubsectionKind Kind;
+};
+
+class DebugSubsection {
+public:
+  explicit DebugSubsection(DebugSubsectionKind Kind) : Kind(Kind) {}
+  virtual ~DebugSubsection();
+
+  static bool classof(const DebugSubsection *S) { return true; }
+
+  DebugSubsectionKind kind() const { return Kind; }
+
+  virtual Error commit(BinaryStreamWriter &Writer) const = 0;
+  virtual uint32_t calculateSerializedSize() const = 0;
+
+protected:
+  DebugSubsectionKind Kind;
+};
+
+} // namespace codeview
+} // namespace llvm
+
+#endif // LLVM_DEBUGINFO_CODEVIEW_MODULEDEBUGFRAGMENT_H

include/llvm/DebugInfo/CodeView/DebugSubsectionRecord.h

@@ -1,4 +1,4 @@
-//===- ModuleDebugFragment.h ------------------------------------*- C++ -*-===//
+//===- DebugSubsection.h ------------------------------------*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -20,52 +20,49 @@
 namespace llvm {
 namespace codeview {
 
-class ModuleDebugFragment;
+class DebugSubsection;
 
 // Corresponds to the `CV_DebugSSubsectionHeader_t` structure.
-struct ModuleDebugFragmentHeader {
-  support::ulittle32_t Kind;   // codeview::ModuleDebugFragmentKind enum
+struct DebugSubsectionHeader {
+  support::ulittle32_t Kind;   // codeview::DebugSubsectionKind enum
   support::ulittle32_t Length; // number of bytes occupied by this record.
 };
 
-class ModuleDebugFragmentRecord {
+class DebugSubsectionRecord {
 public:
-  ModuleDebugFragmentRecord();
-  ModuleDebugFragmentRecord(ModuleDebugFragmentKind Kind, BinaryStreamRef Data);
+  DebugSubsectionRecord();
+  DebugSubsectionRecord(DebugSubsectionKind Kind, BinaryStreamRef Data);
 
-  static Error initialize(BinaryStreamRef Stream,
-                          ModuleDebugFragmentRecord &Info);
+  static Error initialize(BinaryStreamRef Stream, DebugSubsectionRecord &Info);
 
   uint32_t getRecordLength() const;
-  ModuleDebugFragmentKind kind() const;
+  DebugSubsectionKind kind() const;
   BinaryStreamRef getRecordData() const;
 
 private:
-  ModuleDebugFragmentKind Kind;
+  DebugSubsectionKind Kind;
   BinaryStreamRef Data;
 };
 
-class ModuleDebugFragmentRecordBuilder {
+class DebugSubsectionRecordBuilder {
 public:
-  ModuleDebugFragmentRecordBuilder(ModuleDebugFragmentKind Kind,
-                                   ModuleDebugFragment &Frag);
+  DebugSubsectionRecordBuilder(DebugSubsectionKind Kind, DebugSubsection &Frag);
   uint32_t calculateSerializedLength();
   Error commit(BinaryStreamWriter &Writer);
 
 private:
-  ModuleDebugFragmentKind Kind;
-  ModuleDebugFragment &Frag;
+  DebugSubsectionKind Kind;
+  DebugSubsection &Frag;
 };
 
 } // namespace codeview
 
-template <>
-struct VarStreamArrayExtractor<codeview::ModuleDebugFragmentRecord> {
+template <> struct VarStreamArrayExtractor<codeview::DebugSubsectionRecord> {
   typedef void ContextType;
 
   static Error extract(BinaryStreamRef Stream, uint32_t &Length,
-                       codeview::ModuleDebugFragmentRecord &Info) {
-    if (auto EC = codeview::ModuleDebugFragmentRecord::initialize(Stream, Info))
+                       codeview::DebugSubsectionRecord &Info) {
+    if (auto EC = codeview::DebugSubsectionRecord::initialize(Stream, Info))
       return EC;
     Length = Info.getRecordLength();
     return Error::success();
@@ -73,7 +70,7 @@ struct VarStreamArrayExtractor<codeview::ModuleDebugFragmentRecord> {
 };
 
 namespace codeview {
-typedef VarStreamArray<ModuleDebugFragmentRecord> ModuleDebugFragmentArray;
+typedef VarStreamArray<DebugSubsectionRecord> DebugSubsectionArray;
 }
 } // namespace llvm
 

include/llvm/DebugInfo/CodeView/DebugSubsectionVisitor.h

@@ -1,4 +1,4 @@
-//===- ModuleDebugFragmentVisitor.h -----------------------------*- C++ -*-===//
+//===- DebugSubsectionVisitor.h -----------------------------*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -17,43 +17,41 @@ namespace llvm {
 
 namespace codeview {
 
-class ModuleDebugFileChecksumFragmentRef;
-class ModuleDebugFragmentRecord;
-class ModuleDebugInlineeLineFragmentRef;
-class ModuleDebugLineFragmentRef;
-class ModuleDebugUnknownFragmentRef;
+class DebugChecksumsSubsectionRef;
+class DebugSubsectionRecord;
+class DebugInlineeLinesSubsectionRef;
+class DebugLinesSubsectionRef;
+class DebugUnknownSubsectionRef;
 
-class ModuleDebugFragmentVisitor {
+class DebugSubsectionVisitor {
 public:
-  virtual ~ModuleDebugFragmentVisitor() = default;
+  virtual ~DebugSubsectionVisitor() = default;
 
-  virtual Error visitUnknown(ModuleDebugUnknownFragmentRef &Unknown) {
+  virtual Error visitUnknown(DebugUnknownSubsectionRef &Unknown) {
     return Error::success();
   }
-  virtual Error visitLines(ModuleDebugLineFragmentRef &Lines) {
+  virtual Error visitLines(DebugLinesSubsectionRef &Lines) {
     return Error::success();
   }
 
-  virtual Error
-  visitFileChecksums(ModuleDebugFileChecksumFragmentRef &Checksums) {
+  virtual Error visitFileChecksums(DebugChecksumsSubsectionRef &Checksums) {
     return Error::success();
   }
 
-  virtual Error visitInlineeLines(ModuleDebugInlineeLineFragmentRef &Inlinees) {
+  virtual Error visitInlineeLines(DebugInlineeLinesSubsectionRef &Inlinees) {
     return Error::success();
   }
 
   virtual Error finished() { return Error::success(); }
 };
 
-Error visitModuleDebugFragment(const ModuleDebugFragmentRecord &R,
-                               ModuleDebugFragmentVisitor &V);
+Error visitDebugSubsection(const DebugSubsectionRecord &R,
+                           DebugSubsectionVisitor &V);
 
 template <typename T>
-Error visitModuleDebugFragments(T &&FragmentRange,
-                                ModuleDebugFragmentVisitor &V) {
+Error visitDebugSubsections(T &&FragmentRange, DebugSubsectionVisitor &V) {
   for (const auto &L : FragmentRange) {
-    if (auto EC = visitModuleDebugFragment(L, V))
+    if (auto EC = visitDebugSubsection(L, V))
       return EC;
   }
   if (auto EC = V.finished())

include/llvm/DebugInfo/CodeView/DebugSymbolsSubsection.h

@@ -0,0 +1,53 @@
+//===- DebugSymbolsSubsection.h --------------------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_DEBUGINFO_CODEVIEW_DEBUGSYMBOLSSUBSECTION_H
+#define LLVM_DEBUGINFO_CODEVIEW_DEBUGSYMBOLSSUBSECTION_H
+
+#include "llvm/DebugInfo/CodeView/DebugSubsection.h"
+#include "llvm/DebugInfo/CodeView/SymbolRecord.h"
+#include "llvm/Support/Error.h"
+
+namespace llvm {
+namespace codeview {
+class DebugSymbolsSubsectionRef final : public DebugSubsectionRef {
+public:
+  DebugSymbolsSubsectionRef()
+      : DebugSubsectionRef(DebugSubsectionKind::Symbols) {}
+
+  static bool classof(const DebugSubsectionRef *S) {
+    return S->kind() == DebugSubsectionKind::Symbols;
+  }
+
+  Error initialize(BinaryStreamReader Reader);
+
+private:
+  CVSymbolArray Records;
+};
+
+class DebugSymbolsSubsection final : public DebugSubsection {
+public:
+  DebugSymbolsSubsection() : DebugSubsection(DebugSubsectionKind::Symbols) {}
+  static bool classof(const DebugSubsection *S) {
+    return S->kind() == DebugSubsectionKind::Symbols;
+  }
+
+  uint32_t calculateSerializedSize() const override;
+  Error commit(BinaryStreamWriter &Writer) const override;
+
+  void addSymbol(CVSymbol Symbol);
+
+private:
+  uint32_t Length = 0;
+  std::vector<CVSymbol> Records;
+};
+}
+}
+
+#endif

include/llvm/DebugInfo/CodeView/DebugUnknownSubsection.h

@@ -1,4 +1,4 @@
-//===- ModuleDebugUnknownFragment.h -----------------------------*- C++ -*-===//
+//===- DebugUnknownSubsection.h -----------------------------*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -10,17 +10,16 @@
 #ifndef LLVM_DEBUGINFO_CODEVIEW_MODULEDEBUGUNKNOWNFRAGMENT_H
 #define LLVM_DEBUGINFO_CODEVIEW_MODULEDEBUGUNKNOWNFRAGMENT_H
 
-#include "llvm/DebugInfo/CodeView/ModuleDebugFragment.h"
+#include "llvm/DebugInfo/CodeView/DebugSubsection.h"
 #include "llvm/Support/BinaryStreamRef.h"
 
 namespace llvm {
 namespace codeview {
 
-class ModuleDebugUnknownFragmentRef final : public ModuleDebugFragmentRef {
+class DebugUnknownSubsectionRef final : public DebugSubsectionRef {
 public:
-  ModuleDebugUnknownFragmentRef(ModuleDebugFragmentKind Kind,
-                                BinaryStreamRef Data)
-      : ModuleDebugFragmentRef(Kind), Data(Data) {}
+  DebugUnknownSubsectionRef(DebugSubsectionKind Kind, BinaryStreamRef Data)
+      : DebugSubsectionRef(Kind), Data(Data) {}
 
   BinaryStreamRef getData() const { return Data; }
 

include/llvm/DebugInfo/CodeView/ModuleDebugFragment.h

@@ -1,48 +0,0 @@
-//===- ModuleDebugFragment.h ------------------------------------*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-#ifndef LLVM_DEBUGINFO_CODEVIEW_MODULEDEBUGFRAGMENT_H
-#define LLVM_DEBUGINFO_CODEVIEW_MODULEDEBUGFRAGMENT_H
-
-#include "llvm/DebugInfo/CodeView/CodeView.h"
-#include "llvm/Support/BinaryStreamWriter.h"
-#include "llvm/Support/Casting.h"
-
-namespace llvm {
-namespace codeview {
-
-class ModuleDebugFragmentRef {
-public:
-  explicit ModuleDebugFragmentRef(ModuleDebugFragmentKind Kind) : Kind(Kind) {}
-  virtual ~ModuleDebugFragmentRef();
-
-  ModuleDebugFragmentKind kind() const { return Kind; }
-
-protected:
-  ModuleDebugFragmentKind Kind;
-};
-
-class ModuleDebugFragment {
-public:
-  explicit ModuleDebugFragment(ModuleDebugFragmentKind Kind) : Kind(Kind) {}
-  virtual ~ModuleDebugFragment();
-
-  ModuleDebugFragmentKind kind() const { return Kind; }
-
-  virtual Error commit(BinaryStreamWriter &Writer) = 0;
-  virtual uint32_t calculateSerializedLength() = 0;
-
-protected:
-  ModuleDebugFragmentKind Kind;
-};
-
-} // namespace codeview
-} // namespace llvm
-
-#endif // LLVM_DEBUGINFO_CODEVIEW_MODULEDEBUGFRAGMENT_H

include/llvm/DebugInfo/CodeView/SymbolVisitorDelegate.h

@@ -19,7 +19,7 @@ class BinaryStreamReader;
 
 namespace codeview {
 
-class StringTableRef;
+class DebugStringTableSubsectionRef;
 
 class SymbolVisitorDelegate {
 public:
@@ -27,7 +27,7 @@ class SymbolVisitorDelegate {
 
   virtual uint32_t getRecordOffset(BinaryStreamReader Reader) = 0;
   virtual StringRef getFileNameForFileOffset(uint32_t FileOffset) = 0;
-  virtual StringTableRef getStringTable() = 0;
+  virtual DebugStringTableSubsectionRef getStringTable() = 0;
 };
 
 } // end namespace codeview

include/llvm/DebugInfo/PDB/Native/DbiModuleDescriptorBuilder.h

@@ -11,9 +11,9 @@
 #define LLVM_DEBUGINFO_PDB_RAW_DBIMODULEDESCRIPTORBUILDER_H
 
 #include "llvm/ADT/StringRef.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugFileChecksumFragment.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugInlineeLinesFragment.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugLineFragment.h"
+#include "llvm/DebugInfo/CodeView/DebugChecksumsSubsection.h"
+#include "llvm/DebugInfo/CodeView/DebugInlineeLinesSubsection.h"
+#include "llvm/DebugInfo/CodeView/DebugLinesSubsection.h"
 #include "llvm/DebugInfo/CodeView/SymbolRecord.h"
 #include "llvm/DebugInfo/PDB/Native/RawTypes.h"
 #include "llvm/Support/Error.h"
@@ -25,7 +25,7 @@ namespace llvm {
 class BinaryStreamWriter;
 
 namespace codeview {
-class ModuleDebugFragmentRecordBuilder;
+class DebugSubsectionRecordBuilder;
 }
 
 namespace msf {
@@ -49,11 +49,11 @@ class DbiModuleDescriptorBuilder {
   void setObjFileName(StringRef Name);
   void addSymbol(codeview::CVSymbol Symbol);
 
-  void addC13Fragment(std::unique_ptr<codeview::ModuleDebugLineFragment> Lines);
+  void addC13Fragment(std::unique_ptr<codeview::DebugLinesSubsection> Lines);
   void addC13Fragment(
-      std::unique_ptr<codeview::ModuleDebugInlineeLineFragment> Inlinees);
+      std::unique_ptr<codeview::DebugInlineeLinesSubsection> Inlinees);
   void setC13FileChecksums(
-      std::unique_ptr<codeview::ModuleDebugFileChecksumFragment> Checksums);
+      std::unique_ptr<codeview::DebugChecksumsSubsection> Checksums);
 
   uint16_t getStreamIndex() const;
   StringRef getModuleName() const { return ModuleName; }
@@ -83,12 +83,11 @@ class DbiModuleDescriptorBuilder {
   std::vector<std::string> SourceFiles;
   std::vector<codeview::CVSymbol> Symbols;
 
-  std::unique_ptr<codeview::ModuleDebugFileChecksumFragment> ChecksumInfo;
-  std::vector<std::unique_ptr<codeview::ModuleDebugLineFragment>> LineInfo;
-  std::vector<std::unique_ptr<codeview::ModuleDebugInlineeLineFragment>>
-      Inlinees;
+  std::unique_ptr<codeview::DebugChecksumsSubsection> ChecksumInfo;
+  std::vector<std::unique_ptr<codeview::DebugLinesSubsection>> LineInfo;
+  std::vector<std::unique_ptr<codeview::DebugInlineeLinesSubsection>> Inlinees;
 
-  std::vector<std::unique_ptr<codeview::ModuleDebugFragmentRecordBuilder>>
+  std::vector<std::unique_ptr<codeview::DebugSubsectionRecordBuilder>>
       C13Builders;
 
   ModuleInfoHeader Layout;

include/llvm/DebugInfo/PDB/Native/DbiStream.h

@@ -10,7 +10,7 @@
 #ifndef LLVM_DEBUGINFO_PDB_RAW_PDBDBISTREAM_H
 #define LLVM_DEBUGINFO_PDB_RAW_PDBDBISTREAM_H
 
-#include "llvm/DebugInfo/CodeView/ModuleDebugFragment.h"
+#include "llvm/DebugInfo/CodeView/DebugSubsection.h"
 #include "llvm/DebugInfo/MSF/MappedBlockStream.h"
 #include "llvm/DebugInfo/PDB/Native/DbiModuleDescriptor.h"
 #include "llvm/DebugInfo/PDB/Native/DbiModuleList.h"
@@ -19,8 +19,6 @@
 #include "llvm/DebugInfo/PDB/Native/RawTypes.h"
 #include "llvm/DebugInfo/PDB/PDBTypes.h"
 #include "llvm/Support/BinaryStreamArray.h"
-#include "llvm/Support/BinaryStreamArray.h"
-#include "llvm/Support/BinaryStreamRef.h"
 #include "llvm/Support/BinaryStreamRef.h"
 #include "llvm/Support/Endian.h"
 #include "llvm/Support/Error.h"

include/llvm/DebugInfo/PDB/Native/ModuleDebugStream.h

@@ -12,7 +12,7 @@
 
 #include "llvm/ADT/iterator_range.h"
 #include "llvm/DebugInfo/CodeView/CVRecord.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugFragmentRecord.h"
+#include "llvm/DebugInfo/CodeView/DebugSubsectionRecord.h"
 #include "llvm/DebugInfo/CodeView/SymbolRecord.h"
 #include "llvm/DebugInfo/MSF/MappedBlockStream.h"
 #include "llvm/Support/BinaryStreamArray.h"
@@ -25,8 +25,7 @@ class PDBFile;
 class DbiModuleDescriptor;
 
 class ModuleDebugStreamRef {
-  typedef codeview::ModuleDebugFragmentArray::Iterator
-      LinesAndChecksumsIterator;
+  typedef codeview::DebugSubsectionArray::Iterator LinesAndChecksumsIterator;
 
 public:
   ModuleDebugStreamRef(const DbiModuleDescriptor &Module,
@@ -58,7 +57,7 @@ class ModuleDebugStreamRef {
   BinaryStreamRef C13LinesSubstream;
   BinaryStreamRef GlobalRefsSubstream;
 
-  codeview::ModuleDebugFragmentArray LinesAndChecksums;
+  codeview::DebugSubsectionArray LinesAndChecksums;
 };
 }
 }

include/llvm/DebugInfo/PDB/Native/PDBStringTable.h

@@ -12,7 +12,7 @@
 
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/StringRef.h"
-#include "llvm/DebugInfo/CodeView/StringTable.h"
+#include "llvm/DebugInfo/CodeView/DebugStringTableSubsection.h"
 #include "llvm/Support/BinaryStreamArray.h"
 #include "llvm/Support/BinaryStreamRef.h"
 #include "llvm/Support/Endian.h"
@@ -52,7 +52,7 @@ class PDBStringTable {
   Error readEpilogue(BinaryStreamReader &Reader);
 
   const PDBStringTableHeader *Header = nullptr;
-  codeview::StringTableRef Strings;
+  codeview::DebugStringTableSubsectionRef Strings;
   FixedStreamArray<support::ulittle32_t> IDs;
   uint32_t ByteSize = 0;
   uint32_t NameCount = 0;

include/llvm/DebugInfo/PDB/Native/PDBStringTableBuilder.h

@@ -16,7 +16,7 @@
 
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/StringRef.h"
-#include "llvm/DebugInfo/CodeView/StringTable.h"
+#include "llvm/DebugInfo/CodeView/DebugStringTableSubsection.h"
 #include "llvm/Support/Error.h"
 #include <vector>
 
@@ -41,8 +41,10 @@ class PDBStringTableBuilder {
   uint32_t calculateSerializedSize() const;
   Error commit(BinaryStreamWriter &Writer) const;
 
-  codeview::StringTable &getStrings() { return Strings; }
-  const codeview::StringTable &getStrings() const { return Strings; }
+  codeview::DebugStringTableSubsection &getStrings() { return Strings; }
+  const codeview::DebugStringTableSubsection &getStrings() const {
+    return Strings;
+  }
 
 private:
   uint32_t calculateHashTableSize() const;
@@ -51,7 +53,7 @@ class PDBStringTableBuilder {
   Error writeHashTable(BinaryStreamWriter &Writer) const;
   Error writeEpilogue(BinaryStreamWriter &Writer) const;
 
-  codeview::StringTable Strings;
+  codeview::DebugStringTableSubsection Strings;
 };
 
 } // end namespace pdb

include/llvm/MC/ConstantPools.h

@@ -19,6 +19,7 @@
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/Support/SMLoc.h"
 #include <cstdint>
+#include <map>
 
 namespace llvm {
 
@@ -44,7 +45,7 @@ struct ConstantPoolEntry {
 class ConstantPool {
   using EntryVecTy = SmallVector<ConstantPoolEntry, 4>;
   EntryVecTy Entries;
-  DenseMap<int64_t, const MCSymbolRefExpr *> CachedEntries;
+  std::map<int64_t, const MCSymbolRefExpr *> CachedEntries;
 
 public:
   // Initialize a new empty constant pool

include/llvm/Support/ManagedStatic.h

@@ -14,25 +14,22 @@
 #ifndef LLVM_SUPPORT_MANAGEDSTATIC_H
 #define LLVM_SUPPORT_MANAGEDSTATIC_H
 
-#include "llvm/Support/Compiler.h"
 #include <atomic>
 #include <cstddef>
 
 namespace llvm {
 
 /// object_creator - Helper method for ManagedStatic.
-template<class C>
-LLVM_LIBRARY_VISIBILITY void* object_creator() {
-  return new C();
-}
+template <class C> struct object_creator {
+  static void *call() { return new C(); }
+};
 
 /// object_deleter - Helper method for ManagedStatic.
 ///
-template <typename T> struct LLVM_LIBRARY_VISIBILITY object_deleter {
+template <typename T> struct object_deleter {
   static void call(void *Ptr) { delete (T *)Ptr; }
 };
-template <typename T, size_t N>
-struct LLVM_LIBRARY_VISIBILITY object_deleter<T[N]> {
+template <typename T, size_t N> struct object_deleter<T[N]> {
   static void call(void *Ptr) { delete[](T *)Ptr; }
 };
 
@@ -59,14 +56,15 @@ class ManagedStaticBase {
 /// libraries that link in LLVM components) and for making destruction be
 /// explicit through the llvm_shutdown() function call.
 ///
-template<class C>
+template <class C, class Creator = object_creator<C>,
+          class Deleter = object_deleter<C>>
 class ManagedStatic : public ManagedStaticBase {
 public:
   // Accessors.
   C &operator*() {
     void *Tmp = Ptr.load(std::memory_order_acquire);
     if (!Tmp)
-      RegisterManagedStatic(object_creator<C>, object_deleter<C>::call);
+      RegisterManagedStatic(Creator::call, Deleter::call);
 
     return *static_cast<C *>(Ptr.load(std::memory_order_relaxed));
   }
@@ -76,7 +74,7 @@ class ManagedStatic : public ManagedStaticBase {
   const C &operator*() const {
     void *Tmp = Ptr.load(std::memory_order_acquire);
     if (!Tmp)
-      RegisterManagedStatic(object_creator<C>, object_deleter<C>::call);
+      RegisterManagedStatic(Creator::call, Deleter::call);
 
     return *static_cast<C *>(Ptr.load(std::memory_order_relaxed));
   }

include/llvm/TableGen/Record.h

@@ -1189,6 +1189,9 @@ class DagInit final : public TypedInit, public FoldingSetNode,
     return Init ? Init->getValue() : StringRef();
   }
 
+  ArrayRef<Init *> getArgs() const {
+    return makeArrayRef(getTrailingObjects<Init *>(), NumArgs);
+  }
   ArrayRef<StringInit *> getArgNames() const {
     return makeArrayRef(getTrailingObjects<StringInit *>(), NumArgNames);
   }
@@ -1200,19 +1203,16 @@ class DagInit final : public TypedInit, public FoldingSetNode,
   typedef SmallVectorImpl<Init*>::const_iterator       const_arg_iterator;
   typedef SmallVectorImpl<StringInit*>::const_iterator const_name_iterator;
 
-  inline const_arg_iterator  arg_begin() const { return getTrailingObjects<Init *>(); }
-  inline const_arg_iterator  arg_end  () const { return arg_begin() + NumArgs;   }
-  inline iterator_range<const_arg_iterator> args() const {
-    return llvm::make_range(arg_begin(), arg_end());
-  }
+  inline const_arg_iterator  arg_begin() const { return getArgs().begin(); }
+  inline const_arg_iterator  arg_end  () const { return getArgs().end(); }
 
-  inline size_t              arg_size () const { return NumArgs;  }
+  inline size_t              arg_size () const { return NumArgs; }
   inline bool                arg_empty() const { return NumArgs == 0; }
 
-  inline const_name_iterator name_begin() const { return getTrailingObjects<StringInit *>(); }
-  inline const_name_iterator name_end  () const { return name_begin() + NumArgNames;   }
+  inline const_name_iterator name_begin() const { return getArgNames().begin();}
+  inline const_name_iterator name_end  () const { return getArgNames().end(); }
 
-  inline size_t              name_size () const { return NumArgNames;  }
+  inline size_t              name_size () const { return NumArgNames; }
   inline bool                name_empty() const { return NumArgNames == 0; }
 
   Init *getBit(unsigned Bit) const override {

include/llvm/Transforms/Scalar/GVNExpression.h

@@ -58,10 +58,11 @@ class Expression {
 private:
   ExpressionType EType;
   unsigned Opcode;
+  mutable hash_code HashVal;
 
 public:
   Expression(ExpressionType ET = ET_Base, unsigned O = ~2U)
-      : EType(ET), Opcode(O) {}
+      : EType(ET), Opcode(O), HashVal(0) {}
   Expression(const Expression &) = delete;
   Expression &operator=(const Expression &) = delete;
   virtual ~Expression();
@@ -82,6 +83,14 @@ class Expression {
 
     return equals(Other);
   }
+  hash_code getComputedHash() const {
+    // It's theoretically possible for a thing to hash to zero.  In that case,
+    // we will just compute the hash a few extra times, which is no worse that
+    // we did before, which was to compute it always.
+    if (static_cast<unsigned>(HashVal) == 0)
+      HashVal = getHashValue();
+    return HashVal;
+  }
 
   virtual bool equals(const Expression &Other) const { return true; }
 

lib/Analysis/ScalarEvolution.cpp

@@ -2178,8 +2178,7 @@ StrengthenNoWrapFlags(ScalarEvolution *SE, SCEVTypes Type,
   return Flags;
 }
 
-bool ScalarEvolution::isAvailableAtLoopEntry(const SCEV *S, const Loop *L,
-                                             DominatorTree &DT, LoopInfo &LI) {
+bool ScalarEvolution::isAvailableAtLoopEntry(const SCEV *S, const Loop *L) {
   if (!isLoopInvariant(S, L))
     return false;
   // If a value depends on a SCEVUnknown which is defined after the loop, we
@@ -2516,7 +2515,7 @@ const SCEV *ScalarEvolution::getAddExpr(SmallVectorImpl<const SCEV *> &Ops,
     const SCEVAddRecExpr *AddRec = cast<SCEVAddRecExpr>(Ops[Idx]);
     const Loop *AddRecLoop = AddRec->getLoop();
     for (unsigned i = 0, e = Ops.size(); i != e; ++i)
-      if (isAvailableAtLoopEntry(Ops[i], AddRecLoop, DT, LI)) {
+      if (isAvailableAtLoopEntry(Ops[i], AddRecLoop)) {
         LIOps.push_back(Ops[i]);
         Ops.erase(Ops.begin()+i);
         --i; --e;
@@ -2791,7 +2790,7 @@ const SCEV *ScalarEvolution::getMulExpr(SmallVectorImpl<const SCEV *> &Ops,
     const SCEVAddRecExpr *AddRec = cast<SCEVAddRecExpr>(Ops[Idx]);
     const Loop *AddRecLoop = AddRec->getLoop();
     for (unsigned i = 0, e = Ops.size(); i != e; ++i)
-      if (isAvailableAtLoopEntry(Ops[i], AddRecLoop, DT, LI)) {
+      if (isAvailableAtLoopEntry(Ops[i], AddRecLoop)) {
         LIOps.push_back(Ops[i]);
         Ops.erase(Ops.begin()+i);
         --i; --e;

lib/CodeGen/AsmPrinter/CodeViewDebug.cpp

@@ -15,8 +15,8 @@
 #include "llvm/ADT/TinyPtrVector.h"
 #include "llvm/DebugInfo/CodeView/CVTypeVisitor.h"
 #include "llvm/DebugInfo/CodeView/CodeView.h"
+#include "llvm/DebugInfo/CodeView/DebugInlineeLinesSubsection.h"
 #include "llvm/DebugInfo/CodeView/Line.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugInlineeLinesFragment.h"
 #include "llvm/DebugInfo/CodeView/SymbolRecord.h"
 #include "llvm/DebugInfo/CodeView/TypeDatabase.h"
 #include "llvm/DebugInfo/CodeView/TypeDumpVisitor.h"
@@ -393,7 +393,7 @@ void CodeViewDebug::endModule() {
   // subprograms.
   switchToDebugSectionForSymbol(nullptr);
 
-  MCSymbol *CompilerInfo = beginCVSubsection(ModuleDebugFragmentKind::Symbols);
+  MCSymbol *CompilerInfo = beginCVSubsection(DebugSubsectionKind::Symbols);
   emitCompilerInformation();
   endCVSubsection(CompilerInfo);
 
@@ -417,7 +417,7 @@ void CodeViewDebug::endModule() {
 
   // Emit UDT records for any types used by global variables.
   if (!GlobalUDTs.empty()) {
-    MCSymbol *SymbolsEnd = beginCVSubsection(ModuleDebugFragmentKind::Symbols);
+    MCSymbol *SymbolsEnd = beginCVSubsection(DebugSubsectionKind::Symbols);
     emitDebugInfoForUDTs(GlobalUDTs);
     endCVSubsection(SymbolsEnd);
   }
@@ -630,8 +630,7 @@ void CodeViewDebug::emitInlineeLinesSubsection() {
     return;
 
   OS.AddComment("Inlinee lines subsection");
-  MCSymbol *InlineEnd =
-      beginCVSubsection(ModuleDebugFragmentKind::InlineeLines);
+  MCSymbol *InlineEnd = beginCVSubsection(DebugSubsectionKind::InlineeLines);
 
   // We don't provide any extra file info.
   // FIXME: Find out if debuggers use this info.
@@ -756,7 +755,7 @@ void CodeViewDebug::emitDebugInfoForFunction(const Function *GV,
 
   // Emit a symbol subsection, required by VS2012+ to find function boundaries.
   OS.AddComment("Symbol subsection for " + Twine(FuncName));
-  MCSymbol *SymbolsEnd = beginCVSubsection(ModuleDebugFragmentKind::Symbols);
+  MCSymbol *SymbolsEnd = beginCVSubsection(DebugSubsectionKind::Symbols);
   {
     MCSymbol *ProcRecordBegin = MMI->getContext().createTempSymbol(),
              *ProcRecordEnd = MMI->getContext().createTempSymbol();
@@ -2111,7 +2110,7 @@ void CodeViewDebug::beginInstruction(const MachineInstr *MI) {
   maybeRecordLocation(DL, Asm->MF);
 }
 
-MCSymbol *CodeViewDebug::beginCVSubsection(ModuleDebugFragmentKind Kind) {
+MCSymbol *CodeViewDebug::beginCVSubsection(DebugSubsectionKind Kind) {
   MCSymbol *BeginLabel = MMI->getContext().createTempSymbol(),
            *EndLabel = MMI->getContext().createTempSymbol();
   OS.EmitIntValue(unsigned(Kind), 4);
@@ -2171,7 +2170,7 @@ void CodeViewDebug::emitDebugInfoForGlobals() {
         if (!GV->hasComdat() && !GV->isDeclarationForLinker()) {
           if (!EndLabel) {
             OS.AddComment("Symbol subsection for globals");
-            EndLabel = beginCVSubsection(ModuleDebugFragmentKind::Symbols);
+            EndLabel = beginCVSubsection(DebugSubsectionKind::Symbols);
           }
           // FIXME: emitDebugInfoForGlobal() doesn't handle DIExpressions.
           emitDebugInfoForGlobal(GVE->getVariable(), GV, Asm->getSymbol(GV));
@@ -2189,7 +2188,7 @@ void CodeViewDebug::emitDebugInfoForGlobals() {
           OS.AddComment("Symbol subsection for " +
                         Twine(GlobalValue::dropLLVMManglingEscape(GV->getName())));
           switchToDebugSectionForSymbol(GVSym);
-          EndLabel = beginCVSubsection(ModuleDebugFragmentKind::Symbols);
+          EndLabel = beginCVSubsection(DebugSubsectionKind::Symbols);
           // FIXME: emitDebugInfoForGlobal() doesn't handle DIExpressions.
           emitDebugInfoForGlobal(GVE->getVariable(), GV, GVSym);
           endCVSubsection(EndLabel);

lib/CodeGen/AsmPrinter/CodeViewDebug.h

@@ -216,7 +216,7 @@ class LLVM_LIBRARY_VISIBILITY CodeViewDebug : public DebugHandlerBase {
   /// Opens a subsection of the given kind in a .debug$S codeview section.
   /// Returns an end label for use with endCVSubsection when the subsection is
   /// finished.
-  MCSymbol *beginCVSubsection(codeview::ModuleDebugFragmentKind Kind);
+  MCSymbol *beginCVSubsection(codeview::DebugSubsectionKind Kind);
 
   void endCVSubsection(MCSymbol *EndLabel);
 

lib/CodeGen/GlobalISel/Localizer.cpp

@@ -23,7 +23,7 @@ using namespace llvm;
 char Localizer::ID = 0;
 INITIALIZE_PASS(Localizer, DEBUG_TYPE,
                 "Move/duplicate certain instructions close to their use", false,
-                false);
+                false)
 
 Localizer::Localizer() : MachineFunctionPass(ID) {
   initializeLocalizerPass(*PassRegistry::getPassRegistry());

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

@@ -14567,7 +14567,8 @@ static SDValue narrowExtractedVectorLoad(SDNode *Extract, SelectionDAG &DAG) {
   // extract instead or remove that condition entirely.
   auto *Ld = dyn_cast<LoadSDNode>(Extract->getOperand(0));
   auto *ExtIdx = dyn_cast<ConstantSDNode>(Extract->getOperand(1));
-  if (!Ld || !Ld->hasOneUse() || Ld->isVolatile() || !ExtIdx)
+  if (!Ld || !Ld->hasOneUse() || Ld->getExtensionType() || Ld->isVolatile() ||
+      !ExtIdx)
     return SDValue();
 
   // The narrow load will be offset from the base address of the old load if

lib/CodeGen/SelectionDAG/LegalizeDAG.cpp

@@ -925,10 +925,6 @@ getStrictFPOpcodeAction(const TargetLowering &TLI, unsigned Opcode, EVT VT) {
   if (Action != TargetLowering::Legal)
     Action = TargetLowering::Expand;
 
-  // ISD::FPOWI returns 'Legal' even though it should be expanded.
-  if (Opcode == ISD::STRICT_FPOWI && Action == TargetLowering::Legal)
-    Action = TargetLowering::Expand;
-
   return Action;
 }
 
@@ -1027,7 +1023,6 @@ void SelectionDAGLegalize::LegalizeOp(SDNode *Node) {
     break;
   case ISD::EXTRACT_ELEMENT:
   case ISD::FLT_ROUNDS_:
-  case ISD::FPOWI:
   case ISD::MERGE_VALUES:
   case ISD::EH_RETURN:
   case ISD::FRAME_TO_ARGS_OFFSET:

lib/CodeGen/TargetLoweringBase.cpp

@@ -935,6 +935,7 @@ void TargetLoweringBase::initActions() {
     
     // These library functions default to expand.
     setOperationAction(ISD::FROUND, VT, Expand);
+    setOperationAction(ISD::FPOWI, VT, Expand);
 
     // These operations default to expand for vector types.
     if (VT.isVector()) {

lib/DebugInfo/CodeView/CMakeLists.txt

@@ -7,14 +7,16 @@ add_llvm_library(LLVMDebugInfoCodeView
   Formatters.cpp
   LazyRandomTypeCollection.cpp
   Line.cpp
-  ModuleDebugFileChecksumFragment.cpp
-  ModuleDebugFragment.cpp
-  ModuleDebugFragmentRecord.cpp
-  ModuleDebugFragmentVisitor.cpp
-  ModuleDebugInlineeLinesFragment.cpp
-  ModuleDebugLineFragment.cpp
+  DebugChecksumsSubsection.cpp
+  DebugFrameDataSubsection.cpp
+  DebugInlineeLinesSubsection.cpp
+  DebugLinesSubsection.cpp
+  DebugStringTableSubsection.cpp
+  DebugSubsection.cpp
+  DebugSubsectionRecord.cpp
+  DebugSubsectionVisitor.cpp
+  DebugSymbolsSubsection.cpp
   RecordSerialization.cpp
-  StringTable.cpp
   SymbolRecordMapping.cpp
   SymbolDumper.cpp
   SymbolSerializer.cpp

lib/DebugInfo/CodeView/DebugChecksumsSubsection.cpp

@@ -1,4 +1,4 @@
-//===- ModuleDebugFileChecksumFragment.cpp ----------------------*- C++ -*-===//
+//===- DebugChecksumsSubsection.cpp ----------------------*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -7,10 +7,10 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "llvm/DebugInfo/CodeView/ModuleDebugFileChecksumFragment.h"
+#include "llvm/DebugInfo/CodeView/DebugChecksumsSubsection.h"
 
 #include "llvm/DebugInfo/CodeView/CodeViewError.h"
-#include "llvm/DebugInfo/CodeView/StringTable.h"
+#include "llvm/DebugInfo/CodeView/DebugStringTableSubsection.h"
 #include "llvm/Support/BinaryStreamReader.h"
 
 using namespace llvm;
@@ -42,22 +42,24 @@ Error llvm::VarStreamArrayExtractor<FileChecksumEntry>::extract(
   return Error::success();
 }
 
-Error ModuleDebugFileChecksumFragmentRef::initialize(
-    BinaryStreamReader Reader) {
+Error DebugChecksumsSubsectionRef::initialize(BinaryStreamReader Reader) {
   if (auto EC = Reader.readArray(Checksums, Reader.bytesRemaining()))
     return EC;
 
   return Error::success();
 }
+Error DebugChecksumsSubsectionRef::initialize(BinaryStreamRef Section) {
+  BinaryStreamReader Reader(Section);
+  return initialize(Reader);
+}
 
-ModuleDebugFileChecksumFragment::ModuleDebugFileChecksumFragment(
-    StringTable &Strings)
-    : ModuleDebugFragment(ModuleDebugFragmentKind::FileChecksums),
-      Strings(Strings) {}
+DebugChecksumsSubsection::DebugChecksumsSubsection(
+    DebugStringTableSubsection &Strings)
+    : DebugSubsection(DebugSubsectionKind::FileChecksums), Strings(Strings) {}
 
-void ModuleDebugFileChecksumFragment::addChecksum(StringRef FileName,
-                                                  FileChecksumKind Kind,
-                                                  ArrayRef<uint8_t> Bytes) {
+void DebugChecksumsSubsection::addChecksum(StringRef FileName,
+                                           FileChecksumKind Kind,
+                                           ArrayRef<uint8_t> Bytes) {
   FileChecksumEntry Entry;
   if (!Bytes.empty()) {
     uint8_t *Copy = Storage.Allocate<uint8_t>(Bytes.size());
@@ -78,11 +80,11 @@ void ModuleDebugFileChecksumFragment::addChecksum(StringRef FileName,
   SerializedSize += Len;
 }
 
-uint32_t ModuleDebugFileChecksumFragment::calculateSerializedLength() {
+uint32_t DebugChecksumsSubsection::calculateSerializedSize() const {
   return SerializedSize;
 }
 
-Error ModuleDebugFileChecksumFragment::commit(BinaryStreamWriter &Writer) {
+Error DebugChecksumsSubsection::commit(BinaryStreamWriter &Writer) const {
   for (const auto &FC : Checksums) {
     FileChecksumEntryHeader Header;
     Header.ChecksumKind = uint8_t(FC.Kind);
@@ -98,8 +100,7 @@ Error ModuleDebugFileChecksumFragment::commit(BinaryStreamWriter &Writer) {
   return Error::success();
 }
 
-uint32_t
-ModuleDebugFileChecksumFragment::mapChecksumOffset(StringRef FileName) const {
+uint32_t DebugChecksumsSubsection::mapChecksumOffset(StringRef FileName) const {
   uint32_t Offset = Strings.getStringId(FileName);
   auto Iter = OffsetMap.find(Offset);
   assert(Iter != OffsetMap.end());

lib/DebugInfo/CodeView/DebugFrameDataSubsection.cpp

@@ -0,0 +1,44 @@
+//===- DebugFrameDataSubsection.cpp -----------------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/DebugInfo/CodeView/DebugFrameDataSubsection.h"
+#include "llvm/DebugInfo/CodeView/CodeViewError.h"
+
+using namespace llvm;
+using namespace llvm::codeview;
+
+Error DebugFrameDataSubsectionRef::initialize(BinaryStreamReader Reader) {
+  if (auto EC = Reader.readObject(RelocPtr))
+    return EC;
+  if (Reader.bytesRemaining() % sizeof(FrameData) != 0)
+    return make_error<CodeViewError>(cv_error_code::corrupt_record,
+                                     "Invalid frame data record format!");
+
+  uint32_t Count = Reader.bytesRemaining() / sizeof(FrameData);
+  if (auto EC = Reader.readArray(Frames, Count))
+    return EC;
+  return Error::success();
+}
+
+uint32_t DebugFrameDataSubsection::calculateSerializedSize() const {
+  return 4 + sizeof(FrameData) * Frames.size();
+}
+
+Error DebugFrameDataSubsection::commit(BinaryStreamWriter &Writer) const {
+  if (auto EC = Writer.writeInteger<uint32_t>(0))
+    return EC;
+
+  if (auto EC = Writer.writeArray(makeArrayRef(Frames)))
+    return EC;
+  return Error::success();
+}
+
+void DebugFrameDataSubsection::addFrameData(const FrameData &Frame) {
+  Frames.push_back(Frame);
+}

lib/DebugInfo/CodeView/DebugInlineeLinesSubsection.cpp

@@ -1,4 +1,4 @@
-//===- ModuleDebugInlineeLineFragment.cpp ------------------------*- C++-*-===//
+//===- DebugInlineeLinesSubsection.cpp ------------------------*- C++-*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -7,12 +7,12 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "llvm/DebugInfo/CodeView/ModuleDebugInlineeLinesFragment.h"
+#include "llvm/DebugInfo/CodeView/DebugInlineeLinesSubsection.h"
 
 #include "llvm/DebugInfo/CodeView/CodeViewError.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugFileChecksumFragment.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugFragmentRecord.h"
-#include "llvm/DebugInfo/CodeView/StringTable.h"
+#include "llvm/DebugInfo/CodeView/DebugChecksumsSubsection.h"
+#include "llvm/DebugInfo/CodeView/DebugStringTableSubsection.h"
+#include "llvm/DebugInfo/CodeView/DebugSubsectionRecord.h"
 
 using namespace llvm;
 using namespace llvm::codeview;
@@ -37,10 +37,10 @@ Error VarStreamArrayExtractor<InlineeSourceLine>::extract(
   return Error::success();
 }
 
-ModuleDebugInlineeLineFragmentRef::ModuleDebugInlineeLineFragmentRef()
-    : ModuleDebugFragmentRef(ModuleDebugFragmentKind::InlineeLines) {}
+DebugInlineeLinesSubsectionRef::DebugInlineeLinesSubsectionRef()
+    : DebugSubsectionRef(DebugSubsectionKind::InlineeLines) {}
 
-Error ModuleDebugInlineeLineFragmentRef::initialize(BinaryStreamReader Reader) {
+Error DebugInlineeLinesSubsectionRef::initialize(BinaryStreamReader Reader) {
   if (auto EC = Reader.readEnum(Signature))
     return EC;
 
@@ -52,16 +52,16 @@ Error ModuleDebugInlineeLineFragmentRef::initialize(BinaryStreamReader Reader) {
   return Error::success();
 }
 
-bool ModuleDebugInlineeLineFragmentRef::hasExtraFiles() const {
+bool DebugInlineeLinesSubsectionRef::hasExtraFiles() const {
   return Signature == InlineeLinesSignature::ExtraFiles;
 }
 
-ModuleDebugInlineeLineFragment::ModuleDebugInlineeLineFragment(
-    ModuleDebugFileChecksumFragment &Checksums, bool HasExtraFiles)
-    : ModuleDebugFragment(ModuleDebugFragmentKind::InlineeLines),
-      Checksums(Checksums), HasExtraFiles(HasExtraFiles) {}
+DebugInlineeLinesSubsection::DebugInlineeLinesSubsection(
+    DebugChecksumsSubsection &Checksums, bool HasExtraFiles)
+    : DebugSubsection(DebugSubsectionKind::InlineeLines), Checksums(Checksums),
+      HasExtraFiles(HasExtraFiles) {}
 
-uint32_t ModuleDebugInlineeLineFragment::calculateSerializedLength() {
+uint32_t DebugInlineeLinesSubsection::calculateSerializedSize() const {
   // 4 bytes for the signature
   uint32_t Size = sizeof(InlineeLinesSignature);
 
@@ -78,7 +78,7 @@ uint32_t ModuleDebugInlineeLineFragment::calculateSerializedLength() {
   return Size;
 }
 
-Error ModuleDebugInlineeLineFragment::commit(BinaryStreamWriter &Writer) {
+Error DebugInlineeLinesSubsection::commit(BinaryStreamWriter &Writer) const {
   InlineeLinesSignature Sig = InlineeLinesSignature::Normal;
   if (HasExtraFiles)
     Sig = InlineeLinesSignature::ExtraFiles;
@@ -102,7 +102,7 @@ Error ModuleDebugInlineeLineFragment::commit(BinaryStreamWriter &Writer) {
   return Error::success();
 }
 
-void ModuleDebugInlineeLineFragment::addExtraFile(StringRef FileName) {
+void DebugInlineeLinesSubsection::addExtraFile(StringRef FileName) {
   uint32_t Offset = Checksums.mapChecksumOffset(FileName);
 
   auto &Entry = Entries.back();
@@ -110,9 +110,9 @@ void ModuleDebugInlineeLineFragment::addExtraFile(StringRef FileName) {
   ++ExtraFileCount;
 }
 
-void ModuleDebugInlineeLineFragment::addInlineSite(TypeIndex FuncId,
-                                                   StringRef FileName,
-                                                   uint32_t SourceLine) {
+void DebugInlineeLinesSubsection::addInlineSite(TypeIndex FuncId,
+                                                StringRef FileName,
+                                                uint32_t SourceLine) {
   uint32_t Offset = Checksums.mapChecksumOffset(FileName);
 
   Entries.emplace_back();

lib/DebugInfo/CodeView/DebugLinesSubsection.cpp

@@ -1,4 +1,4 @@
-//===- ModuleDebugLineFragment.cpp -------------------------------*- C++-*-===//
+//===- DebugLinesSubsection.cpp -------------------------------*- C++-*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -7,12 +7,12 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "llvm/DebugInfo/CodeView/ModuleDebugLineFragment.h"
+#include "llvm/DebugInfo/CodeView/DebugLinesSubsection.h"
 
 #include "llvm/DebugInfo/CodeView/CodeViewError.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugFileChecksumFragment.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugFragmentRecord.h"
-#include "llvm/DebugInfo/CodeView/StringTable.h"
+#include "llvm/DebugInfo/CodeView/DebugChecksumsSubsection.h"
+#include "llvm/DebugInfo/CodeView/DebugStringTableSubsection.h"
+#include "llvm/DebugInfo/CodeView/DebugSubsectionRecord.h"
 
 using namespace llvm;
 using namespace llvm::codeview;
@@ -49,10 +49,10 @@ Error LineColumnExtractor::extract(BinaryStreamRef Stream, uint32_t &Len,
   return Error::success();
 }
 
-ModuleDebugLineFragmentRef::ModuleDebugLineFragmentRef()
-    : ModuleDebugFragmentRef(ModuleDebugFragmentKind::Lines) {}
+DebugLinesSubsectionRef::DebugLinesSubsectionRef()
+    : DebugSubsectionRef(DebugSubsectionKind::Lines) {}
 
-Error ModuleDebugLineFragmentRef::initialize(BinaryStreamReader Reader) {
+Error DebugLinesSubsectionRef::initialize(BinaryStreamReader Reader) {
   if (auto EC = Reader.readObject(Header))
     return EC;
 
@@ -63,23 +63,21 @@ Error ModuleDebugLineFragmentRef::initialize(BinaryStreamReader Reader) {
   return Error::success();
 }
 
-bool ModuleDebugLineFragmentRef::hasColumnInfo() const {
+bool DebugLinesSubsectionRef::hasColumnInfo() const {
   return !!(Header->Flags & LF_HaveColumns);
 }
 
-ModuleDebugLineFragment::ModuleDebugLineFragment(
-    ModuleDebugFileChecksumFragment &Checksums, StringTable &Strings)
-    : ModuleDebugFragment(ModuleDebugFragmentKind::Lines),
-      Checksums(Checksums) {}
+DebugLinesSubsection::DebugLinesSubsection(DebugChecksumsSubsection &Checksums,
+                                           DebugStringTableSubsection &Strings)
+    : DebugSubsection(DebugSubsectionKind::Lines), Checksums(Checksums) {}
 
-void ModuleDebugLineFragment::createBlock(StringRef FileName) {
+void DebugLinesSubsection::createBlock(StringRef FileName) {
   uint32_t Offset = Checksums.mapChecksumOffset(FileName);
 
   Blocks.emplace_back(Offset);
 }
 
-void ModuleDebugLineFragment::addLineInfo(uint32_t Offset,
-                                          const LineInfo &Line) {
+void DebugLinesSubsection::addLineInfo(uint32_t Offset, const LineInfo &Line) {
   Block &B = Blocks.back();
   LineNumberEntry LNE;
   LNE.Flags = Line.getRawData();
@@ -87,10 +85,10 @@ void ModuleDebugLineFragment::addLineInfo(uint32_t Offset,
   B.Lines.push_back(LNE);
 }
 
-void ModuleDebugLineFragment::addLineAndColumnInfo(uint32_t Offset,
-                                                   const LineInfo &Line,
-                                                   uint32_t ColStart,
-                                                   uint32_t ColEnd) {
+void DebugLinesSubsection::addLineAndColumnInfo(uint32_t Offset,
+                                                const LineInfo &Line,
+                                                uint32_t ColStart,
+                                                uint32_t ColEnd) {
   Block &B = Blocks.back();
   assert(B.Lines.size() == B.Columns.size());
 
@@ -101,7 +99,7 @@ void ModuleDebugLineFragment::addLineAndColumnInfo(uint32_t Offset,
   B.Columns.push_back(CNE);
 }
 
-Error ModuleDebugLineFragment::commit(BinaryStreamWriter &Writer) {
+Error DebugLinesSubsection::commit(BinaryStreamWriter &Writer) const {
   LineFragmentHeader Header;
   Header.CodeSize = CodeSize;
   Header.Flags = hasColumnInfo() ? LF_HaveColumns : 0;
@@ -135,7 +133,7 @@ Error ModuleDebugLineFragment::commit(BinaryStreamWriter &Writer) {
   return Error::success();
 }
 
-uint32_t ModuleDebugLineFragment::calculateSerializedLength() {
+uint32_t DebugLinesSubsection::calculateSerializedSize() const {
   uint32_t Size = sizeof(LineFragmentHeader);
   for (const auto &B : Blocks) {
     Size += sizeof(LineBlockFragmentHeader);
@@ -146,16 +144,16 @@ uint32_t ModuleDebugLineFragment::calculateSerializedLength() {
   return Size;
 }
 
-void ModuleDebugLineFragment::setRelocationAddress(uint16_t Segment,
-                                                   uint16_t Offset) {
+void DebugLinesSubsection::setRelocationAddress(uint16_t Segment,
+                                                uint16_t Offset) {
   RelocOffset = Offset;
   RelocSegment = Segment;
 }
 
-void ModuleDebugLineFragment::setCodeSize(uint32_t Size) { CodeSize = Size; }
+void DebugLinesSubsection::setCodeSize(uint32_t Size) { CodeSize = Size; }
 
-void ModuleDebugLineFragment::setFlags(LineFlags Flags) { this->Flags = Flags; }
+void DebugLinesSubsection::setFlags(LineFlags Flags) { this->Flags = Flags; }
 
-bool ModuleDebugLineFragment::hasColumnInfo() const {
+bool DebugLinesSubsection::hasColumnInfo() const {
   return Flags & LF_HaveColumns;
 }

lib/DebugInfo/CodeView/DebugStringTableSubsection.cpp

@@ -1,4 +1,4 @@
-//===- StringTable.cpp - CodeView String Table Reader/Writer ----*- C++ -*-===//
+//===- DebugStringTableSubsection.cpp - CodeView String Table ---*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -7,7 +7,7 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "llvm/DebugInfo/CodeView/StringTable.h"
+#include "llvm/DebugInfo/CodeView/DebugStringTableSubsection.h"
 
 #include "llvm/Support/BinaryStream.h"
 #include "llvm/Support/BinaryStreamReader.h"
@@ -16,14 +16,16 @@
 using namespace llvm;
 using namespace llvm::codeview;
 
-StringTableRef::StringTableRef() {}
+DebugStringTableSubsectionRef::DebugStringTableSubsectionRef()
+    : DebugSubsectionRef(DebugSubsectionKind::StringTable) {}
 
-Error StringTableRef::initialize(BinaryStreamRef Contents) {
+Error DebugStringTableSubsectionRef::initialize(BinaryStreamRef Contents) {
   Stream = Contents;
   return Error::success();
 }
 
-Expected<StringRef> StringTableRef::getString(uint32_t Offset) const {
+Expected<StringRef>
+DebugStringTableSubsectionRef::getString(uint32_t Offset) const {
   BinaryStreamReader Reader(Stream);
   Reader.setOffset(Offset);
   StringRef Result;
@@ -32,7 +34,10 @@ Expected<StringRef> StringTableRef::getString(uint32_t Offset) const {
   return Result;
 }
 
-uint32_t StringTable::insert(StringRef S) {
+DebugStringTableSubsection::DebugStringTableSubsection()
+    : DebugSubsection(DebugSubsectionKind::StringTable) {}
+
+uint32_t DebugStringTableSubsection::insert(StringRef S) {
   auto P = Strings.insert({S, StringSize});
 
   // If a given string didn't exist in the string table, we want to increment
@@ -42,9 +47,11 @@ uint32_t StringTable::insert(StringRef S) {
   return P.first->second;
 }
 
-uint32_t StringTable::calculateSerializedSize() const { return StringSize; }
+uint32_t DebugStringTableSubsection::calculateSerializedSize() const {
+  return StringSize;
+}
 
-Error StringTable::commit(BinaryStreamWriter &Writer) const {
+Error DebugStringTableSubsection::commit(BinaryStreamWriter &Writer) const {
   assert(Writer.bytesRemaining() == StringSize);
   uint32_t MaxOffset = 1;
 
@@ -62,9 +69,9 @@ Error StringTable::commit(BinaryStreamWriter &Writer) const {
   return Error::success();
 }
 
-uint32_t StringTable::size() const { return Strings.size(); }
+uint32_t DebugStringTableSubsection::size() const { return Strings.size(); }
 
-uint32_t StringTable::getStringId(StringRef S) const {
+uint32_t DebugStringTableSubsection::getStringId(StringRef S) const {
   auto P = Strings.find(S);
   assert(P != Strings.end());
   return P->second;

lib/DebugInfo/CodeView/DebugSubsection.cpp

@@ -1,4 +1,4 @@
-//===- ModuleDebugFragment.cpp -----------------------------------*- C++-*-===//
+//===- DebugSubsection.cpp -----------------------------------*- C++-*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -7,10 +7,10 @@
 //
 //===----------------------------------------------------------------------===//
 
-#include "llvm/DebugInfo/CodeView/ModuleDebugFragment.h"
+#include "llvm/DebugInfo/CodeView/DebugSubsection.h"
 
 using namespace llvm::codeview;
 
-ModuleDebugFragmentRef::~ModuleDebugFragmentRef() {}
+DebugSubsectionRef::~DebugSubsectionRef() {}
 
-ModuleDebugFragment::~ModuleDebugFragment() {}
+DebugSubsection::~DebugSubsection() {}

lib/DebugInfo/CodeView/DebugSubsectionRecord.cpp

@@ -0,0 +1,81 @@
+//===- DebugSubsectionRecord.cpp -----------------------------*- C++-*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/DebugInfo/CodeView/DebugSubsectionRecord.h"
+#include "llvm/DebugInfo/CodeView/DebugSubsection.h"
+
+#include "llvm/Support/BinaryStreamReader.h"
+
+using namespace llvm;
+using namespace llvm::codeview;
+
+DebugSubsectionRecord::DebugSubsectionRecord()
+    : Kind(DebugSubsectionKind::None) {}
+
+DebugSubsectionRecord::DebugSubsectionRecord(DebugSubsectionKind Kind,
+                                             BinaryStreamRef Data)
+    : Kind(Kind), Data(Data) {}
+
+Error DebugSubsectionRecord::initialize(BinaryStreamRef Stream,
+                                        DebugSubsectionRecord &Info) {
+  const DebugSubsectionHeader *Header;
+  BinaryStreamReader Reader(Stream);
+  if (auto EC = Reader.readObject(Header))
+    return EC;
+
+  DebugSubsectionKind Kind =
+      static_cast<DebugSubsectionKind>(uint32_t(Header->Kind));
+  switch (Kind) {
+  case DebugSubsectionKind::FileChecksums:
+  case DebugSubsectionKind::Lines:
+  case DebugSubsectionKind::InlineeLines:
+    break;
+  default:
+    llvm_unreachable("Unexpected debug fragment kind!");
+  }
+  if (auto EC = Reader.readStreamRef(Info.Data, Header->Length))
+    return EC;
+  Info.Kind = Kind;
+  return Error::success();
+}
+
+uint32_t DebugSubsectionRecord::getRecordLength() const {
+  uint32_t Result = sizeof(DebugSubsectionHeader) + Data.getLength();
+  assert(Result % 4 == 0);
+  return Result;
+}
+
+DebugSubsectionKind DebugSubsectionRecord::kind() const { return Kind; }
+
+BinaryStreamRef DebugSubsectionRecord::getRecordData() const { return Data; }
+
+DebugSubsectionRecordBuilder::DebugSubsectionRecordBuilder(
+    DebugSubsectionKind Kind, DebugSubsection &Frag)
+    : Kind(Kind), Frag(Frag) {}
+
+uint32_t DebugSubsectionRecordBuilder::calculateSerializedLength() {
+  uint32_t Size = sizeof(DebugSubsectionHeader) +
+                  alignTo(Frag.calculateSerializedSize(), 4);
+  return Size;
+}
+
+Error DebugSubsectionRecordBuilder::commit(BinaryStreamWriter &Writer) {
+  DebugSubsectionHeader Header;
+  Header.Kind = uint32_t(Kind);
+  Header.Length = calculateSerializedLength() - sizeof(DebugSubsectionHeader);
+
+  if (auto EC = Writer.writeObject(Header))
+    return EC;
+  if (auto EC = Frag.commit(Writer))
+    return EC;
+  if (auto EC = Writer.padToAlignment(4))
+    return EC;
+
+  return Error::success();
+}

lib/DebugInfo/CodeView/DebugSubsectionVisitor.cpp

@@ -0,0 +1,52 @@
+//===- DebugSubsectionVisitor.cpp ---------------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/DebugInfo/CodeView/DebugSubsectionVisitor.h"
+
+#include "llvm/DebugInfo/CodeView/DebugChecksumsSubsection.h"
+#include "llvm/DebugInfo/CodeView/DebugInlineeLinesSubsection.h"
+#include "llvm/DebugInfo/CodeView/DebugLinesSubsection.h"
+#include "llvm/DebugInfo/CodeView/DebugSubsectionRecord.h"
+#include "llvm/DebugInfo/CodeView/DebugUnknownSubsection.h"
+#include "llvm/Support/BinaryStreamReader.h"
+#include "llvm/Support/BinaryStreamRef.h"
+
+using namespace llvm;
+using namespace llvm::codeview;
+
+Error llvm::codeview::visitDebugSubsection(const DebugSubsectionRecord &R,
+                                           DebugSubsectionVisitor &V) {
+  BinaryStreamReader Reader(R.getRecordData());
+  switch (R.kind()) {
+  case DebugSubsectionKind::Lines: {
+    DebugLinesSubsectionRef Fragment;
+    if (auto EC = Fragment.initialize(Reader))
+      return EC;
+
+    return V.visitLines(Fragment);
+  }
+  case DebugSubsectionKind::FileChecksums: {
+    DebugChecksumsSubsectionRef Fragment;
+    if (auto EC = Fragment.initialize(Reader))
+      return EC;
+
+    return V.visitFileChecksums(Fragment);
+  }
+  case DebugSubsectionKind::InlineeLines: {
+    DebugInlineeLinesSubsectionRef Fragment;
+    if (auto EC = Fragment.initialize(Reader))
+      return EC;
+    return V.visitInlineeLines(Fragment);
+  }
+  default: {
+    DebugUnknownSubsectionRef Fragment(R.kind(), R.getRecordData());
+    return V.visitUnknown(Fragment);
+  }
+  }
+}

lib/DebugInfo/CodeView/DebugSymbolsSubsection.cpp

@@ -0,0 +1,34 @@
+//===- DebugSymbolsSubsection.cpp -------------------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/DebugInfo/CodeView/DebugSymbolsSubsection.h"
+
+using namespace llvm;
+using namespace llvm::codeview;
+
+Error DebugSymbolsSubsectionRef::initialize(BinaryStreamReader Reader) {
+  return Reader.readArray(Records, Reader.getLength());
+}
+
+uint32_t DebugSymbolsSubsection::calculateSerializedSize() const {
+  return Length;
+}
+
+Error DebugSymbolsSubsection::commit(BinaryStreamWriter &Writer) const {
+  for (const auto &Record : Records) {
+    if (auto EC = Writer.writeBytes(Record.RecordData))
+      return EC;
+  }
+  return Error::success();
+}
+
+void DebugSymbolsSubsection::addSymbol(CVSymbol Symbol) {
+  Records.push_back(Symbol);
+  Length += Symbol.length();
+}

lib/DebugInfo/CodeView/EnumTables.cpp

@@ -245,20 +245,20 @@ static const EnumEntry<uint32_t> FrameProcSymFlagNames[] = {
 };
 
 static const EnumEntry<uint32_t> ModuleSubstreamKindNames[] = {
-    CV_ENUM_CLASS_ENT(ModuleDebugFragmentKind, None),
-    CV_ENUM_CLASS_ENT(ModuleDebugFragmentKind, Symbols),
-    CV_ENUM_CLASS_ENT(ModuleDebugFragmentKind, Lines),
-    CV_ENUM_CLASS_ENT(ModuleDebugFragmentKind, StringTable),
-    CV_ENUM_CLASS_ENT(ModuleDebugFragmentKind, FileChecksums),
-    CV_ENUM_CLASS_ENT(ModuleDebugFragmentKind, FrameData),
-    CV_ENUM_CLASS_ENT(ModuleDebugFragmentKind, InlineeLines),
-    CV_ENUM_CLASS_ENT(ModuleDebugFragmentKind, CrossScopeImports),
-    CV_ENUM_CLASS_ENT(ModuleDebugFragmentKind, CrossScopeExports),
-    CV_ENUM_CLASS_ENT(ModuleDebugFragmentKind, ILLines),
-    CV_ENUM_CLASS_ENT(ModuleDebugFragmentKind, FuncMDTokenMap),
-    CV_ENUM_CLASS_ENT(ModuleDebugFragmentKind, TypeMDTokenMap),
-    CV_ENUM_CLASS_ENT(ModuleDebugFragmentKind, MergedAssemblyInput),
-    CV_ENUM_CLASS_ENT(ModuleDebugFragmentKind, CoffSymbolRVA),
+    CV_ENUM_CLASS_ENT(DebugSubsectionKind, None),
+    CV_ENUM_CLASS_ENT(DebugSubsectionKind, Symbols),
+    CV_ENUM_CLASS_ENT(DebugSubsectionKind, Lines),
+    CV_ENUM_CLASS_ENT(DebugSubsectionKind, StringTable),
+    CV_ENUM_CLASS_ENT(DebugSubsectionKind, FileChecksums),
+    CV_ENUM_CLASS_ENT(DebugSubsectionKind, FrameData),
+    CV_ENUM_CLASS_ENT(DebugSubsectionKind, InlineeLines),
+    CV_ENUM_CLASS_ENT(DebugSubsectionKind, CrossScopeImports),
+    CV_ENUM_CLASS_ENT(DebugSubsectionKind, CrossScopeExports),
+    CV_ENUM_CLASS_ENT(DebugSubsectionKind, ILLines),
+    CV_ENUM_CLASS_ENT(DebugSubsectionKind, FuncMDTokenMap),
+    CV_ENUM_CLASS_ENT(DebugSubsectionKind, TypeMDTokenMap),
+    CV_ENUM_CLASS_ENT(DebugSubsectionKind, MergedAssemblyInput),
+    CV_ENUM_CLASS_ENT(DebugSubsectionKind, CoffSymbolRVA),
 };
 
 static const EnumEntry<uint16_t> ExportSymFlagNames[] = {

lib/DebugInfo/CodeView/ModuleDebugFragmentRecord.cpp

@@ -1,84 +0,0 @@
-//===- ModuleDebugFragmentRecord.cpp -----------------------------*- C++-*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-#include "llvm/DebugInfo/CodeView/ModuleDebugFragmentRecord.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugFragment.h"
-
-#include "llvm/Support/BinaryStreamReader.h"
-
-using namespace llvm;
-using namespace llvm::codeview;
-
-ModuleDebugFragmentRecord::ModuleDebugFragmentRecord()
-    : Kind(ModuleDebugFragmentKind::None) {}
-
-ModuleDebugFragmentRecord::ModuleDebugFragmentRecord(
-    ModuleDebugFragmentKind Kind, BinaryStreamRef Data)
-    : Kind(Kind), Data(Data) {}
-
-Error ModuleDebugFragmentRecord::initialize(BinaryStreamRef Stream,
-                                            ModuleDebugFragmentRecord &Info) {
-  const ModuleDebugFragmentHeader *Header;
-  BinaryStreamReader Reader(Stream);
-  if (auto EC = Reader.readObject(Header))
-    return EC;
-
-  ModuleDebugFragmentKind Kind =
-      static_cast<ModuleDebugFragmentKind>(uint32_t(Header->Kind));
-  switch (Kind) {
-  case ModuleDebugFragmentKind::FileChecksums:
-  case ModuleDebugFragmentKind::Lines:
-  case ModuleDebugFragmentKind::InlineeLines:
-    break;
-  default:
-    llvm_unreachable("Unexpected debug fragment kind!");
-  }
-  if (auto EC = Reader.readStreamRef(Info.Data, Header->Length))
-    return EC;
-  Info.Kind = Kind;
-  return Error::success();
-}
-
-uint32_t ModuleDebugFragmentRecord::getRecordLength() const {
-  uint32_t Result = sizeof(ModuleDebugFragmentHeader) + Data.getLength();
-  assert(Result % 4 == 0);
-  return Result;
-}
-
-ModuleDebugFragmentKind ModuleDebugFragmentRecord::kind() const { return Kind; }
-
-BinaryStreamRef ModuleDebugFragmentRecord::getRecordData() const {
-  return Data;
-}
-
-ModuleDebugFragmentRecordBuilder::ModuleDebugFragmentRecordBuilder(
-    ModuleDebugFragmentKind Kind, ModuleDebugFragment &Frag)
-    : Kind(Kind), Frag(Frag) {}
-
-uint32_t ModuleDebugFragmentRecordBuilder::calculateSerializedLength() {
-  uint32_t Size = sizeof(ModuleDebugFragmentHeader) +
-                  alignTo(Frag.calculateSerializedLength(), 4);
-  return Size;
-}
-
-Error ModuleDebugFragmentRecordBuilder::commit(BinaryStreamWriter &Writer) {
-  ModuleDebugFragmentHeader Header;
-  Header.Kind = uint32_t(Kind);
-  Header.Length =
-      calculateSerializedLength() - sizeof(ModuleDebugFragmentHeader);
-
-  if (auto EC = Writer.writeObject(Header))
-    return EC;
-  if (auto EC = Frag.commit(Writer))
-    return EC;
-  if (auto EC = Writer.padToAlignment(4))
-    return EC;
-
-  return Error::success();
-}

lib/DebugInfo/CodeView/ModuleDebugFragmentVisitor.cpp

@@ -1,52 +0,0 @@
-//===- ModuleDebugFragmentVisitor.cpp ---------------------------*- C++ -*-===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-
-#include "llvm/DebugInfo/CodeView/ModuleDebugFragmentVisitor.h"
-
-#include "llvm/DebugInfo/CodeView/ModuleDebugFileChecksumFragment.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugFragmentRecord.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugInlineeLinesFragment.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugLineFragment.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugUnknownFragment.h"
-#include "llvm/Support/BinaryStreamReader.h"
-#include "llvm/Support/BinaryStreamRef.h"
-
-using namespace llvm;
-using namespace llvm::codeview;
-
-Error llvm::codeview::visitModuleDebugFragment(
-    const ModuleDebugFragmentRecord &R, ModuleDebugFragmentVisitor &V) {
-  BinaryStreamReader Reader(R.getRecordData());
-  switch (R.kind()) {
-  case ModuleDebugFragmentKind::Lines: {
-    ModuleDebugLineFragmentRef Fragment;
-    if (auto EC = Fragment.initialize(Reader))
-      return EC;
-
-    return V.visitLines(Fragment);
-  }
-  case ModuleDebugFragmentKind::FileChecksums: {
-    ModuleDebugFileChecksumFragmentRef Fragment;
-    if (auto EC = Fragment.initialize(Reader))
-      return EC;
-
-    return V.visitFileChecksums(Fragment);
-  }
-  case ModuleDebugFragmentKind::InlineeLines: {
-    ModuleDebugInlineeLineFragmentRef Fragment;
-    if (auto EC = Fragment.initialize(Reader))
-      return EC;
-    return V.visitInlineeLines(Fragment);
-  }
-  default: {
-    ModuleDebugUnknownFragmentRef Fragment(R.kind(), R.getRecordData());
-    return V.visitUnknown(Fragment);
-  }
-  }
-}

lib/DebugInfo/CodeView/SymbolDumper.cpp

@@ -11,8 +11,8 @@
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/DebugInfo/CodeView/CVSymbolVisitor.h"
+#include "llvm/DebugInfo/CodeView/DebugStringTableSubsection.h"
 #include "llvm/DebugInfo/CodeView/EnumTables.h"
-#include "llvm/DebugInfo/CodeView/StringTable.h"
 #include "llvm/DebugInfo/CodeView/SymbolDeserializer.h"
 #include "llvm/DebugInfo/CodeView/SymbolDumpDelegate.h"
 #include "llvm/DebugInfo/CodeView/SymbolRecord.h"
@@ -369,7 +369,7 @@ Error CVSymbolDumperImpl::visitKnownRecord(
   DictScope S(W, "DefRangeSubfield");
 
   if (ObjDelegate) {
-    StringTableRef Strings = ObjDelegate->getStringTable();
+    DebugStringTableSubsectionRef Strings = ObjDelegate->getStringTable();
     auto ExpectedProgram = Strings.getString(DefRangeSubfield.Program);
     if (!ExpectedProgram) {
       consumeError(ExpectedProgram.takeError());
@@ -390,7 +390,7 @@ Error CVSymbolDumperImpl::visitKnownRecord(CVSymbol &CVR,
   DictScope S(W, "DefRange");
 
   if (ObjDelegate) {
-    StringTableRef Strings = ObjDelegate->getStringTable();
+    DebugStringTableSubsectionRef Strings = ObjDelegate->getStringTable();
     auto ExpectedProgram = Strings.getString(DefRange.Program);
     if (!ExpectedProgram) {
       consumeError(ExpectedProgram.takeError());

lib/DebugInfo/PDB/Native/DbiModuleDescriptorBuilder.cpp

@@ -10,7 +10,7 @@
 #include "llvm/DebugInfo/PDB/Native/DbiModuleDescriptorBuilder.h"
 
 #include "llvm/ADT/ArrayRef.h"
-#include "llvm/DebugInfo/CodeView/ModuleDebugFragmentRecord.h"
+#include "llvm/DebugInfo/CodeView/DebugSubsectionRecord.h"
 #include "llvm/DebugInfo/MSF/MSFBuilder.h"
 #include "llvm/DebugInfo/MSF/MSFCommon.h"
 #include "llvm/DebugInfo/MSF/MappedBlockStream.h"
@@ -170,8 +170,8 @@ Error DbiModuleDescriptorBuilder::commit(BinaryStreamWriter &ModiWriter,
 }
 
 void DbiModuleDescriptorBuilder::addC13Fragment(
-    std::unique_ptr<ModuleDebugLineFragment> Lines) {
-  ModuleDebugLineFragment &Frag = *Lines;
+    std::unique_ptr<DebugLinesSubsection> Lines) {
+  DebugLinesSubsection &Frag = *Lines;
 
   // File Checksums have to come first, so push an empty entry on if this
   // is the first.
@@ -180,12 +180,12 @@ void DbiModuleDescriptorBuilder::addC13Fragment(
 
   this->LineInfo.push_back(std::move(Lines));
   C13Builders.push_back(
-      llvm::make_unique<ModuleDebugFragmentRecordBuilder>(Frag.kind(), Frag));
+      llvm::make_unique<DebugSubsectionRecordBuilder>(Frag.kind(), Frag));
 }
 
 void DbiModuleDescriptorBuilder::addC13Fragment(
-    std::unique_ptr<codeview::ModuleDebugInlineeLineFragment> Inlinees) {
-  ModuleDebugInlineeLineFragment &Frag = *Inlinees;
+    std::unique_ptr<codeview::DebugInlineeLinesSubsection> Inlinees) {
+  DebugInlineeLinesSubsection &Frag = *Inlinees;
 
   // File Checksums have to come first, so push an empty entry on if this
   // is the first.
@@ -194,17 +194,17 @@ void DbiModuleDescriptorBuilder::addC13Fragment(
 
   this->Inlinees.push_back(std::move(Inlinees));
   C13Builders.push_back(
-      llvm::make_unique<ModuleDebugFragmentRecordBuilder>(Frag.kind(), Frag));
+      llvm::make_unique<DebugSubsectionRecordBuilder>(Frag.kind(), Frag));
 }
 
 void DbiModuleDescriptorBuilder::setC13FileChecksums(
-    std::unique_ptr<ModuleDebugFileChecksumFragment> Checksums) {
+    std::unique_ptr<DebugChecksumsSubsection> Checksums) {
   assert(!ChecksumInfo && "Can't have more than one checksum info!");
 
   if (C13Builders.empty())
     C13Builders.push_back(nullptr);
 
   ChecksumInfo = std::move(Checksums);
-  C13Builders[0] = llvm::make_unique<ModuleDebugFragmentRecordBuilder>(
+  C13Builders[0] = llvm::make_unique<DebugSubsectionRecordBuilder>(
       ChecksumInfo->kind(), *ChecksumInfo);
 }

lib/MC/MCCodeView.cpp

@@ -145,7 +145,7 @@ void CodeViewContext::emitStringTable(MCObjectStreamer &OS) {
   MCSymbol *StringBegin = Ctx.createTempSymbol("strtab_begin", false),
            *StringEnd = Ctx.createTempSymbol("strtab_end", false);
 
-  OS.EmitIntValue(unsigned(ModuleDebugFragmentKind::StringTable), 4);
+  OS.EmitIntValue(unsigned(DebugSubsectionKind::StringTable), 4);
   OS.emitAbsoluteSymbolDiff(StringEnd, StringBegin, 4);
   OS.EmitLabel(StringBegin);
 
@@ -172,7 +172,7 @@ void CodeViewContext::emitFileChecksums(MCObjectStreamer &OS) {
   MCSymbol *FileBegin = Ctx.createTempSymbol("filechecksums_begin", false),
            *FileEnd = Ctx.createTempSymbol("filechecksums_end", false);
 
-  OS.EmitIntValue(unsigned(ModuleDebugFragmentKind::FileChecksums), 4);
+  OS.EmitIntValue(unsigned(DebugSubsectionKind::FileChecksums), 4);
   OS.emitAbsoluteSymbolDiff(FileEnd, FileBegin, 4);
   OS.EmitLabel(FileBegin);
 
@@ -197,7 +197,7 @@ void CodeViewContext::emitLineTableForFunction(MCObjectStreamer &OS,
   MCSymbol *LineBegin = Ctx.createTempSymbol("linetable_begin", false),
            *LineEnd = Ctx.createTempSymbol("linetable_end", false);
 
-  OS.EmitIntValue(unsigned(ModuleDebugFragmentKind::Lines), 4);
+  OS.EmitIntValue(unsigned(DebugSubsectionKind::Lines), 4);
   OS.emitAbsoluteSymbolDiff(LineEnd, LineBegin, 4);
   OS.EmitLabel(LineBegin);
   OS.EmitCOFFSecRel32(FuncBegin, /*Offset=*/0);

lib/Support/APFloat.cpp

@@ -1559,11 +1559,13 @@ IEEEFloat::opStatus IEEEFloat::divideSpecials(const IEEEFloat &rhs) {
   case PackCategoriesIntoKey(fcInfinity, fcNaN):
     category = fcNaN;
     copySignificand(rhs);
+    LLVM_FALLTHROUGH;
   case PackCategoriesIntoKey(fcNaN, fcZero):
   case PackCategoriesIntoKey(fcNaN, fcNormal):
   case PackCategoriesIntoKey(fcNaN, fcInfinity):
   case PackCategoriesIntoKey(fcNaN, fcNaN):
     sign = false;
+    LLVM_FALLTHROUGH;
   case PackCategoriesIntoKey(fcInfinity, fcZero):
   case PackCategoriesIntoKey(fcInfinity, fcNormal):
   case PackCategoriesIntoKey(fcZero, fcInfinity):

lib/Support/Timer.cpp

@@ -72,10 +72,15 @@ std::unique_ptr<raw_fd_ostream> llvm::CreateInfoOutputFile() {
   return llvm::make_unique<raw_fd_ostream>(2, false); // stderr.
 }
 
-static TimerGroup *getDefaultTimerGroup() {
-  static TimerGroup DefaultTimerGroup("misc", "Miscellaneous Ungrouped Timers");
-  return &DefaultTimerGroup;
-}
+namespace {
+struct CreateDefaultTimerGroup {
+  static void *call() {
+    return new TimerGroup("misc", "Miscellaneous Ungrouped Timers");
+  }
+};
+} // namespace
+static ManagedStatic<TimerGroup, CreateDefaultTimerGroup> DefaultTimerGroup;
+static TimerGroup *getDefaultTimerGroup() { return &*DefaultTimerGroup; }
 
 //===----------------------------------------------------------------------===//
 // Timer Implementation

lib/TableGen/Record.cpp

@@ -405,27 +405,21 @@ IntInit::convertInitializerBitRange(ArrayRef<unsigned> Bits) const {
 }
 
 CodeInit *CodeInit::get(StringRef V) {
-  static DenseMap<StringRef, CodeInit*> ThePool;
+  static StringMap<CodeInit*, BumpPtrAllocator &> ThePool(Allocator);
 
-  auto I = ThePool.insert(std::make_pair(V, nullptr));
-  if (I.second) {
-    StringRef VCopy = V.copy(Allocator);
-    I.first->first = VCopy;
-    I.first->second = new(Allocator) CodeInit(VCopy);
-  }
-  return I.first->second;
+  auto &Entry = *ThePool.insert(std::make_pair(V, nullptr)).first;
+  if (!Entry.second)
+    Entry.second = new(Allocator) CodeInit(Entry.getKey());
+  return Entry.second;
 }
 
 StringInit *StringInit::get(StringRef V) {
-  static DenseMap<StringRef, StringInit*> ThePool;
+  static StringMap<StringInit*, BumpPtrAllocator &> ThePool(Allocator);
 
-  auto I = ThePool.insert(std::make_pair(V, nullptr));
-  if (I.second) {
-    StringRef VCopy = V.copy(Allocator);
-    I.first->first = VCopy;
-    I.first->second = new(Allocator) StringInit(VCopy);
-  }
-  return I.first->second;
+  auto &Entry = *ThePool.insert(std::make_pair(V, nullptr)).first;
+  if (!Entry.second)
+    Entry.second = new(Allocator) StringInit(Entry.getKey());
+  return Entry.second;
 }
 
 Init *StringInit::convertInitializerTo(RecTy *Ty) const {
@@ -1540,7 +1534,7 @@ Init *DagInit::resolveReferences(Record &R, const RecordVal *RV) const {
   SmallVector<Init*, 8> NewArgs;
   NewArgs.reserve(arg_size());
   bool ArgsChanged = false;
-  for (const Init *Arg : args()) {
+  for (const Init *Arg : getArgs()) {
     Init *NewArg = Arg->resolveReferences(R, RV);
     NewArgs.push_back(NewArg);
     ArgsChanged |= NewArg != Arg;

lib/Target/AArch64/AArch64FrameLowering.cpp

@@ -137,6 +137,34 @@ static cl::opt<bool> EnableRedZone("aarch64-redzone",
 
 STATISTIC(NumRedZoneFunctions, "Number of functions using red zone");
 
+/// Look at each instruction that references stack frames and return the stack
+/// size limit beyond which some of these instructions will require a scratch
+/// register during their expansion later.
+static unsigned estimateRSStackSizeLimit(MachineFunction &MF) {
+  // FIXME: For now, just conservatively guestimate based on unscaled indexing
+  // range. We'll end up allocating an unnecessary spill slot a lot, but
+  // realistically that's not a big deal at this stage of the game.
+  for (MachineBasicBlock &MBB : MF) {
+    for (MachineInstr &MI : MBB) {
+      if (MI.isDebugValue() || MI.isPseudo() ||
+          MI.getOpcode() == AArch64::ADDXri ||
+          MI.getOpcode() == AArch64::ADDSXri)
+        continue;
+
+      for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
+        if (!MI.getOperand(i).isFI())
+          continue;
+
+        int Offset = 0;
+        if (isAArch64FrameOffsetLegal(MI, Offset, nullptr, nullptr, nullptr) ==
+            AArch64FrameOffsetCannotUpdate)
+          return 0;
+      }
+    }
+  }
+  return 255;
+}
+
 bool AArch64FrameLowering::canUseRedZone(const MachineFunction &MF) const {
   if (!EnableRedZone)
     return false;
@@ -1169,16 +1197,13 @@ void AArch64FrameLowering::determineCalleeSaves(MachineFunction &MF,
   unsigned NumRegsSpilled = SavedRegs.count();
   bool CanEliminateFrame = NumRegsSpilled == 0;
 
-  // FIXME: Set BigStack if any stack slot references may be out of range.
-  // For now, just conservatively guestimate based on unscaled indexing
-  // range. We'll end up allocating an unnecessary spill slot a lot, but
-  // realistically that's not a big deal at this stage of the game.
   // The CSR spill slots have not been allocated yet, so estimateStackSize
   // won't include them.
   MachineFrameInfo &MFI = MF.getFrameInfo();
   unsigned CFSize = MFI.estimateStackSize(MF) + 8 * NumRegsSpilled;
   DEBUG(dbgs() << "Estimated stack frame size: " << CFSize << " bytes.\n");
-  bool BigStack = (CFSize >= 256);
+  unsigned EstimatedStackSizeLimit = estimateRSStackSizeLimit(MF);
+  bool BigStack = (CFSize > EstimatedStackSizeLimit);
   if (BigStack || !CanEliminateFrame || RegInfo->cannotEliminateFrame(MF))
     AFI->setHasStackFrame(true);
 

lib/Target/AArch64/AArch64ISelLowering.cpp

@@ -381,7 +381,6 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
   setOperationAction(ISD::FNEARBYINT, MVT::v4f16, Expand);
   setOperationAction(ISD::FNEG, MVT::v4f16, Expand);
   setOperationAction(ISD::FPOW, MVT::v4f16, Expand);
-  setOperationAction(ISD::FPOWI, MVT::v4f16, Expand);
   setOperationAction(ISD::FREM, MVT::v4f16, Expand);
   setOperationAction(ISD::FROUND, MVT::v4f16, Expand);
   setOperationAction(ISD::FRINT, MVT::v4f16, Expand);
@@ -413,7 +412,6 @@ AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM,
   setOperationAction(ISD::FNEARBYINT, MVT::v8f16, Expand);
   setOperationAction(ISD::FNEG, MVT::v8f16, Expand);
   setOperationAction(ISD::FPOW, MVT::v8f16, Expand);
-  setOperationAction(ISD::FPOWI, MVT::v8f16, Expand);
   setOperationAction(ISD::FREM, MVT::v8f16, Expand);
   setOperationAction(ISD::FROUND, MVT::v8f16, Expand);
   setOperationAction(ISD::FRINT, MVT::v8f16, Expand);
@@ -726,7 +724,6 @@ void AArch64TargetLowering::addTypeForNEON(MVT VT, MVT PromotedBitwiseVT) {
   if (VT == MVT::v2f32 || VT == MVT::v4f32 || VT == MVT::v2f64) {
     setOperationAction(ISD::FSIN, VT, Expand);
     setOperationAction(ISD::FCOS, VT, Expand);
-    setOperationAction(ISD::FPOWI, VT, Expand);
     setOperationAction(ISD::FPOW, VT, Expand);
     setOperationAction(ISD::FLOG, VT, Expand);
     setOperationAction(ISD::FLOG2, VT, Expand);

lib/Target/AMDGPU/AMDGPUSubtarget.h

@@ -730,7 +730,7 @@ class SISubtarget final : public AMDGPUSubtarget {
   /// \returns True if waitcnt instruction is needed before barrier instruction,
   /// false otherwise.
   bool needWaitcntBeforeBarrier() const {
-    return getGeneration() < GFX9;
+    return true;
   }
 
   /// \returns true if the flat_scratch register should be initialized with the

lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

@@ -736,6 +736,9 @@ void GCNPassConfig::addMachineSSAOptimization() {
   addPass(createSIShrinkInstructionsPass());
   if (EnableSDWAPeephole) {
     addPass(&SIPeepholeSDWAID);
+    addPass(&MachineLICMID);
+    addPass(&MachineCSEID);
+    addPass(&SIFoldOperandsID);
     addPass(&DeadMachineInstructionElimID);
   }
 }

lib/Target/AMDGPU/SIFoldOperands.cpp

@@ -247,9 +247,10 @@ static bool tryAddToFoldList(SmallVectorImpl<FoldCandidate> &FoldList,
 
 // If the use operand doesn't care about the value, this may be an operand only
 // used for register indexing, in which case it is unsafe to fold.
-static bool isUseSafeToFold(const MachineInstr &MI,
+static bool isUseSafeToFold(const SIInstrInfo *TII,
+                            const MachineInstr &MI,
                             const MachineOperand &UseMO) {
-  return !UseMO.isUndef();
+  return !UseMO.isUndef() && !TII->isSDWA(MI);
   //return !MI.hasRegisterImplicitUseOperand(UseMO.getReg());
 }
 
@@ -261,7 +262,7 @@ void SIFoldOperands::foldOperand(
   SmallVectorImpl<MachineInstr *> &CopiesToReplace) const {
   const MachineOperand &UseOp = UseMI->getOperand(UseOpIdx);
 
-  if (!isUseSafeToFold(*UseMI, UseOp))
+  if (!isUseSafeToFold(TII, *UseMI, UseOp))
     return;
 
   // FIXME: Fold operands with subregs.

lib/Target/AMDGPU/SIPeepholeSDWA.cpp

@@ -55,6 +55,7 @@ class SIPeepholeSDWA : public MachineFunctionPass {
 
   std::unordered_map<MachineInstr *, std::unique_ptr<SDWAOperand>> SDWAOperands;
   std::unordered_map<MachineInstr *, SDWAOperandsVector> PotentialMatches;
+  SmallVector<MachineInstr *, 8> ConvertedInstructions;
 
   Optional<int64_t> foldToImm(const MachineOperand &Op) const;
 
@@ -69,6 +70,7 @@ class SIPeepholeSDWA : public MachineFunctionPass {
   void matchSDWAOperands(MachineFunction &MF);
   bool isConvertibleToSDWA(const MachineInstr &MI) const;
   bool convertToSDWA(MachineInstr &MI, const SDWAOperandsVector &SDWAOperands);
+  void legalizeScalarOperands(MachineInstr &MI) const;
 
   StringRef getPassName() const override { return "SI Peephole SDWA"; }
 
@@ -289,7 +291,7 @@ bool SDWASrcOperand::convertToSDWA(MachineInstr &MI, const SIInstrInfo *TII) {
   MachineOperand *SrcSel = TII->getNamedOperand(MI, AMDGPU::OpName::src0_sel);
   MachineOperand *SrcMods =
       TII->getNamedOperand(MI, AMDGPU::OpName::src0_modifiers);
-  assert(Src && Src->isReg());
+  assert(Src && (Src->isReg() || Src->isImm()));
   if (!isSameReg(*Src, *getReplacedOperand())) {
     // If this is not src0 then it should be src1
     Src = TII->getNamedOperand(MI, AMDGPU::OpName::src1);
@@ -580,18 +582,8 @@ void SIPeepholeSDWA::matchSDWAOperands(MachineFunction &MF) {
 }
 
 bool SIPeepholeSDWA::isConvertibleToSDWA(const MachineInstr &MI) const {
-  // Check if this instruction can be converted to SDWA:
-  // 1. Does this opcode support SDWA
-  if (AMDGPU::getSDWAOp(MI.getOpcode()) == -1)
-    return false;
-
-  // 2. Are all operands - VGPRs
-  for (const MachineOperand &Operand : MI.explicit_operands()) {
-    if (!Operand.isReg() || !TRI->isVGPR(*MRI, Operand.getReg()))
-      return false;
-  }
-
-  return true;
+  // Check if this instruction has opcode that supports SDWA
+  return AMDGPU::getSDWAOp(MI.getOpcode()) != -1;
 }
 
 bool SIPeepholeSDWA::convertToSDWA(MachineInstr &MI,
@@ -685,7 +677,9 @@ bool SIPeepholeSDWA::convertToSDWA(MachineInstr &MI,
     if (PotentialMatches.count(Operand->getParentInst()) == 0)
       Converted |= Operand->convertToSDWA(*SDWAInst, TII);
   }
-  if (!Converted) {
+  if (Converted) {
+    ConvertedInstructions.push_back(SDWAInst);
+  } else {
     SDWAInst->eraseFromParent();
     return false;
   }
@@ -698,6 +692,29 @@ bool SIPeepholeSDWA::convertToSDWA(MachineInstr &MI,
   return true;
 }
 
+// If an instruction was converted to SDWA it should not have immediates or SGPR
+// operands. Copy its scalar operands into VGPRs.
+void SIPeepholeSDWA::legalizeScalarOperands(MachineInstr &MI) const {
+  const MCInstrDesc &Desc = TII->get(MI.getOpcode());
+  for (unsigned I = 0, E = MI.getNumExplicitOperands(); I != E; ++I) {
+    MachineOperand &Op = MI.getOperand(I);
+    if (!Op.isImm() && !(Op.isReg() && !TRI->isVGPR(*MRI, Op.getReg())))
+      continue;
+    if (Desc.OpInfo[I].RegClass == -1 ||
+       !TRI->hasVGPRs(TRI->getRegClass(Desc.OpInfo[I].RegClass)))
+      continue;
+    unsigned VGPR = MRI->createVirtualRegister(&AMDGPU::VGPR_32RegClass);
+    auto Copy = BuildMI(*MI.getParent(), MI.getIterator(), MI.getDebugLoc(),
+                        TII->get(AMDGPU::V_MOV_B32_e32), VGPR);
+    if (Op.isImm())
+      Copy.addImm(Op.getImm());
+    else if (Op.isReg())
+      Copy.addReg(Op.getReg(), Op.isKill() ? RegState::Kill : 0,
+                  Op.getSubReg());
+    Op.ChangeToRegister(VGPR, false);
+  }
+}
+
 bool SIPeepholeSDWA::runOnMachineFunction(MachineFunction &MF) {
   const SISubtarget &ST = MF.getSubtarget<SISubtarget>();
 
@@ -728,5 +745,9 @@ bool SIPeepholeSDWA::runOnMachineFunction(MachineFunction &MF) {
 
   PotentialMatches.clear();
   SDWAOperands.clear();
+
+  while (!ConvertedInstructions.empty())
+    legalizeScalarOperands(*ConvertedInstructions.pop_back_val());
+
   return false;
 }

lib/Target/ARM/ARMISelLowering.cpp

@@ -585,7 +585,6 @@ ARMTargetLowering::ARMTargetLowering(const TargetMachine &TM,
     setOperationAction(ISD::FSQRT, MVT::v2f64, Expand);
     setOperationAction(ISD::FSIN, MVT::v2f64, Expand);
     setOperationAction(ISD::FCOS, MVT::v2f64, Expand);
-    setOperationAction(ISD::FPOWI, MVT::v2f64, Expand);
     setOperationAction(ISD::FPOW, MVT::v2f64, Expand);
     setOperationAction(ISD::FLOG, MVT::v2f64, Expand);
     setOperationAction(ISD::FLOG2, MVT::v2f64, Expand);
@@ -603,7 +602,6 @@ ARMTargetLowering::ARMTargetLowering(const TargetMachine &TM,
     setOperationAction(ISD::FSQRT, MVT::v4f32, Expand);
     setOperationAction(ISD::FSIN, MVT::v4f32, Expand);
     setOperationAction(ISD::FCOS, MVT::v4f32, Expand);
-    setOperationAction(ISD::FPOWI, MVT::v4f32, Expand);
     setOperationAction(ISD::FPOW, MVT::v4f32, Expand);
     setOperationAction(ISD::FLOG, MVT::v4f32, Expand);
     setOperationAction(ISD::FLOG2, MVT::v4f32, Expand);
@@ -620,7 +618,6 @@ ARMTargetLowering::ARMTargetLowering(const TargetMachine &TM,
     setOperationAction(ISD::FSQRT, MVT::v2f32, Expand);
     setOperationAction(ISD::FSIN, MVT::v2f32, Expand);
     setOperationAction(ISD::FCOS, MVT::v2f32, Expand);
-    setOperationAction(ISD::FPOWI, MVT::v2f32, Expand);
     setOperationAction(ISD::FPOW, MVT::v2f32, Expand);
     setOperationAction(ISD::FLOG, MVT::v2f32, Expand);
     setOperationAction(ISD::FLOG2, MVT::v2f32, Expand);
@@ -743,7 +740,6 @@ ARMTargetLowering::ARMTargetLowering(const TargetMachine &TM,
     setOperationAction(ISD::FSQRT,      MVT::f64, Expand);
     setOperationAction(ISD::FSIN,       MVT::f64, Expand);
     setOperationAction(ISD::FCOS,       MVT::f64, Expand);
-    setOperationAction(ISD::FPOWI,      MVT::f64, Expand);
     setOperationAction(ISD::FPOW,       MVT::f64, Expand);
     setOperationAction(ISD::FLOG,       MVT::f64, Expand);
     setOperationAction(ISD::FLOG2,      MVT::f64, Expand);

lib/Target/Hexagon/HexagonISelLowering.cpp

@@ -2003,7 +2003,7 @@ HexagonTargetLowering::HexagonTargetLowering(const TargetMachine &TM,
     // Floating point arithmetic/math functions:
     ISD::FADD,    ISD::FSUB,    ISD::FMUL,    ISD::FMA,     ISD::FDIV,
     ISD::FREM,    ISD::FNEG,    ISD::FABS,    ISD::FSQRT,   ISD::FSIN,
-    ISD::FCOS,    ISD::FPOWI,   ISD::FPOW,    ISD::FLOG,    ISD::FLOG2,
+    ISD::FCOS,    ISD::FPOW,    ISD::FLOG,    ISD::FLOG2,
     ISD::FLOG10,  ISD::FEXP,    ISD::FEXP2,   ISD::FCEIL,   ISD::FTRUNC,
     ISD::FRINT,   ISD::FNEARBYINT,            ISD::FROUND,  ISD::FFLOOR,
     ISD::FMINNUM, ISD::FMAXNUM, ISD::FSINCOS,

lib/Target/Mips/AsmParser/MipsAsmParser.cpp

@@ -13,6 +13,7 @@
 #include "MCTargetDesc/MipsMCTargetDesc.h"
 #include "MipsTargetStreamer.h"
 #include "MCTargetDesc/MipsBaseInfo.h"
+#include "llvm/ADT/APFloat.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/StringSwitch.h"
@@ -216,9 +217,15 @@ class MipsAsmParser : public MCTargetAsmParser {
                                unsigned SrcReg, bool Is32BitSym, SMLoc IDLoc,
                                MCStreamer &Out, const MCSubtargetInfo *STI);
 
+  bool emitPartialAddress(MipsTargetStreamer &TOut, SMLoc IDLoc, MCSymbol *Sym);
+
   bool expandLoadImm(MCInst &Inst, bool Is32BitImm, SMLoc IDLoc,
                      MCStreamer &Out, const MCSubtargetInfo *STI);
 
+  bool expandLoadImmReal(MCInst &Inst, bool IsSingle, bool IsGPR, bool Is64FPU,
+                         SMLoc IDLoc, MCStreamer &Out,
+                         const MCSubtargetInfo *STI);
+
   bool expandLoadAddress(unsigned DstReg, unsigned BaseReg,
                          const MCOperand &Offset, bool Is32BitAddress,
                          SMLoc IDLoc, MCStreamer &Out,
@@ -1011,6 +1018,16 @@ class MipsOperand : public MCParsedAsmOperand {
     Inst.addOperand(MCOperand::createReg(getAFGR64Reg()));
   }
 
+  void addStrictlyAFGR64AsmRegOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+    Inst.addOperand(MCOperand::createReg(getAFGR64Reg()));
+  }
+
+  void addStrictlyFGR64AsmRegOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+    Inst.addOperand(MCOperand::createReg(getFGR64Reg()));
+  }
+
   void addFGR64AsmRegOperands(MCInst &Inst, unsigned N) const {
     assert(N == 1 && "Invalid number of operands!");
     Inst.addOperand(MCOperand::createReg(getFGR64Reg()));
@@ -1027,6 +1044,15 @@ class MipsOperand : public MCParsedAsmOperand {
                     "registers");
   }
 
+  void addStrictlyFGR32AsmRegOperands(MCInst &Inst, unsigned N) const {
+    assert(N == 1 && "Invalid number of operands!");
+    Inst.addOperand(MCOperand::createReg(getFGR32Reg()));
+    // FIXME: We ought to do this for -integrated-as without -via-file-asm too.
+    if (!AsmParser.useOddSPReg() && RegIdx.Index & 1)
+      AsmParser.Error(StartLoc, "-mno-odd-spreg prohibits the use of odd FPU "
+                                "registers");
+  }
+
   void addFGRH32AsmRegOperands(MCInst &Inst, unsigned N) const {
     assert(N == 1 && "Invalid number of operands!");
     Inst.addOperand(MCOperand::createReg(getFGRH32Reg()));
@@ -1574,6 +1600,11 @@ class MipsOperand : public MCParsedAsmOperand {
     return isRegIdx() && RegIdx.Kind & RegKind_FGR && RegIdx.Index <= 31;
   }
 
+  bool isStrictlyFGRAsmReg() const {
+    // AFGR64 is $0-$15 but we handle this in getAFGR64()
+    return isRegIdx() && RegIdx.Kind == RegKind_FGR && RegIdx.Index <= 31;
+  }
+
   bool isHWRegsAsmReg() const {
     return isRegIdx() && RegIdx.Kind & RegKind_HWRegs && RegIdx.Index <= 31;
   }
@@ -2368,6 +2399,27 @@ MipsAsmParser::tryExpandInstruction(MCInst &Inst, SMLoc IDLoc, MCStreamer &Out,
   case Mips::PseudoTRUNC_W_D:
     return expandTrunc(Inst, true, true, IDLoc, Out, STI) ? MER_Fail
                                                           : MER_Success;
+
+  case Mips::LoadImmSingleGPR:
+    return expandLoadImmReal(Inst, true, true, false, IDLoc, Out, STI)
+               ? MER_Fail
+               : MER_Success;
+  case Mips::LoadImmSingleFGR:
+    return expandLoadImmReal(Inst, true, false, false, IDLoc, Out, STI)
+               ? MER_Fail
+               : MER_Success;
+  case Mips::LoadImmDoubleGPR:
+    return expandLoadImmReal(Inst, false, true, false, IDLoc, Out, STI)
+               ? MER_Fail
+               : MER_Success;
+  case Mips::LoadImmDoubleFGR:
+      return expandLoadImmReal(Inst, false, false, true, IDLoc, Out, STI)
+               ? MER_Fail
+               : MER_Success;
+  case Mips::LoadImmDoubleFGR_32:
+    return expandLoadImmReal(Inst, false, false, false, IDLoc, Out, STI)
+               ? MER_Fail
+               : MER_Success;
   case Mips::Ulh:
     return expandUlh(Inst, true, IDLoc, Out, STI) ? MER_Fail : MER_Success;
   case Mips::Ulhu:
@@ -2952,6 +3004,302 @@ bool MipsAsmParser::loadAndAddSymbolAddress(const MCExpr *SymExpr,
   return false;
 }
 
+// Each double-precision register DO-D15 overlaps with two of the single
+// precision registers F0-F31. As an example, all of the following hold true:
+// D0 + 1 == F1, F1 + 1 == D1, F1 + 1 == F2, depending on the context.
+static unsigned nextReg(unsigned Reg) {
+  if (MipsMCRegisterClasses[Mips::FGR32RegClassID].contains(Reg))
+    return Reg == (unsigned)Mips::F31 ? (unsigned)Mips::F0 : Reg + 1;
+  switch (Reg) {
+  default: llvm_unreachable("Unknown register in assembly macro expansion!");
+  case Mips::ZERO: return Mips::AT;
+  case Mips::AT:   return Mips::V0;
+  case Mips::V0:   return Mips::V1;
+  case Mips::V1:   return Mips::A0;
+  case Mips::A0:   return Mips::A1;
+  case Mips::A1:   return Mips::A2;
+  case Mips::A2:   return Mips::A3;
+  case Mips::A3:   return Mips::T0;
+  case Mips::T0:   return Mips::T1;
+  case Mips::T1:   return Mips::T2;
+  case Mips::T2:   return Mips::T3;
+  case Mips::T3:   return Mips::T4;
+  case Mips::T4:   return Mips::T5;
+  case Mips::T5:   return Mips::T6;
+  case Mips::T6:   return Mips::T7;
+  case Mips::T7:   return Mips::S0;
+  case Mips::S0:   return Mips::S1;
+  case Mips::S1:   return Mips::S2;
+  case Mips::S2:   return Mips::S3;
+  case Mips::S3:   return Mips::S4;
+  case Mips::S4:   return Mips::S5;
+  case Mips::S5:   return Mips::S6;
+  case Mips::S6:   return Mips::S7;
+  case Mips::S7:   return Mips::T8;
+  case Mips::T8:   return Mips::T9;
+  case Mips::T9:   return Mips::K0;
+  case Mips::K0:   return Mips::K1;
+  case Mips::K1:   return Mips::GP;
+  case Mips::GP:   return Mips::SP;
+  case Mips::SP:   return Mips::FP;
+  case Mips::FP:   return Mips::RA;
+  case Mips::RA:   return Mips::ZERO;
+  case Mips::D0:   return Mips::F1;
+  case Mips::D1:   return Mips::F3;
+  case Mips::D2:   return Mips::F5;
+  case Mips::D3:   return Mips::F7;
+  case Mips::D4:   return Mips::F9;
+  case Mips::D5:   return Mips::F11;
+  case Mips::D6:   return Mips::F13;
+  case Mips::D7:   return Mips::F15;
+  case Mips::D8:   return Mips::F17;
+  case Mips::D9:   return Mips::F19;
+  case Mips::D10:   return Mips::F21;
+  case Mips::D11:   return Mips::F23;
+  case Mips::D12:   return Mips::F25;
+  case Mips::D13:   return Mips::F27;
+  case Mips::D14:   return Mips::F29;
+  case Mips::D15:   return Mips::F31;
+  }
+}
+
+// FIXME: This method is too general. In principle we should compute the number
+// of instructions required to synthesize the immediate inline compared to
+// synthesizing the address inline and relying on non .text sections.
+// For static O32 and N32 this may yield a small benefit, for static N64 this is
+// likely to yield a much larger benefit as we have to synthesize a 64bit
+// address to load a 64 bit value.
+bool MipsAsmParser::emitPartialAddress(MipsTargetStreamer &TOut, SMLoc IDLoc,
+                                       MCSymbol *Sym) {
+  unsigned ATReg = getATReg(IDLoc);
+  if (!ATReg)
+    return true;
+
+  if(IsPicEnabled) {
+    const MCExpr *GotSym =
+        MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None, getContext());
+    const MipsMCExpr *GotExpr =
+        MipsMCExpr::create(MipsMCExpr::MEK_GOT, GotSym, getContext());
+
+    if(isABI_O32() || isABI_N32()) {
+      TOut.emitRRX(Mips::LW, ATReg, Mips::GP, MCOperand::createExpr(GotExpr),
+                   IDLoc, STI);
+    } else { //isABI_N64()
+      TOut.emitRRX(Mips::LD, ATReg, Mips::GP, MCOperand::createExpr(GotExpr),
+                   IDLoc, STI);
+    }
+  } else { //!IsPicEnabled
+    const MCExpr *HiSym =
+        MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None, getContext());
+    const MipsMCExpr *HiExpr =
+        MipsMCExpr::create(MipsMCExpr::MEK_HI, HiSym, getContext());
+
+    // FIXME: This is technically correct but gives a different result to gas,
+    // but gas is incomplete there (it has a fixme noting it doesn't work with
+    // 64-bit addresses).
+    // FIXME: With -msym32 option, the address expansion for N64 should probably
+    // use the O32 / N32 case. It's safe to use the 64 address expansion as the
+    // symbol's value is considered sign extended.
+    if(isABI_O32() || isABI_N32()) {
+      TOut.emitRX(Mips::LUi, ATReg, MCOperand::createExpr(HiExpr), IDLoc, STI);
+    } else { //isABI_N64()
+      const MCExpr *HighestSym =
+          MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None, getContext());
+      const MipsMCExpr *HighestExpr =
+          MipsMCExpr::create(MipsMCExpr::MEK_HIGHEST, HighestSym, getContext());
+      const MCExpr *HigherSym =
+          MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None, getContext());
+      const MipsMCExpr *HigherExpr =
+          MipsMCExpr::create(MipsMCExpr::MEK_HIGHER, HigherSym, getContext());
+
+      TOut.emitRX(Mips::LUi, ATReg, MCOperand::createExpr(HighestExpr), IDLoc,
+                  STI);
+      TOut.emitRRX(Mips::DADDiu, ATReg, ATReg,
+                   MCOperand::createExpr(HigherExpr), IDLoc, STI);
+      TOut.emitRRI(Mips::DSLL, ATReg, ATReg, 16, IDLoc, STI);
+      TOut.emitRRX(Mips::DADDiu, ATReg, ATReg, MCOperand::createExpr(HiExpr),
+                   IDLoc, STI);
+      TOut.emitRRI(Mips::DSLL, ATReg, ATReg, 16, IDLoc, STI);
+    }
+  }
+  return false;
+}
+
+bool MipsAsmParser::expandLoadImmReal(MCInst &Inst, bool IsSingle, bool IsGPR,
+                                      bool Is64FPU, SMLoc IDLoc,
+                                      MCStreamer &Out,
+                                      const MCSubtargetInfo *STI) {
+  MipsTargetStreamer &TOut = getTargetStreamer();
+  assert(Inst.getNumOperands() == 2 && "Invalid operand count");
+  assert(Inst.getOperand(0).isReg() && Inst.getOperand(1).isImm() &&
+         "Invalid instruction operand.");
+
+  unsigned FirstReg = Inst.getOperand(0).getReg();
+  uint64_t ImmOp64 = Inst.getOperand(1).getImm();
+
+  uint32_t HiImmOp64 = (ImmOp64 & 0xffffffff00000000) >> 32;
+  // If ImmOp64 is AsmToken::Integer type (all bits set to zero in the
+  // exponent field), convert it to double (e.g. 1 to 1.0)
+  if ((HiImmOp64 & 0x7ff00000) == 0) {
+    APFloat RealVal(APFloat::IEEEdouble(), ImmOp64);
+    ImmOp64 = RealVal.bitcastToAPInt().getZExtValue();
+  }
+
+  uint32_t LoImmOp64 = ImmOp64 & 0xffffffff;
+  HiImmOp64 = (ImmOp64 & 0xffffffff00000000) >> 32;
+
+  if (IsSingle) {
+    // Conversion of a double in an uint64_t to a float in a uint32_t,
+    // retaining the bit pattern of a float.
+    uint32_t ImmOp32;
+    double doubleImm = BitsToDouble(ImmOp64);
+    float tmp_float = static_cast<float>(doubleImm);
+    ImmOp32 = FloatToBits(tmp_float);
+
+    if (IsGPR) {
+      if (loadImmediate(ImmOp32, FirstReg, Mips::NoRegister, true, true, IDLoc,
+                        Out, STI))
+        return true;
+      return false;
+    } else {
+      unsigned ATReg = getATReg(IDLoc);
+      if (!ATReg)
+        return true;
+      if (LoImmOp64 == 0) {
+        if (loadImmediate(ImmOp32, ATReg, Mips::NoRegister, true, true, IDLoc,
+                          Out, STI))
+          return true;
+        TOut.emitRR(Mips::MTC1, FirstReg, ATReg, IDLoc, STI);
+        return false;
+      }
+
+      MCSection *CS = getStreamer().getCurrentSectionOnly();
+      // FIXME: Enhance this expansion to use the .lit4 & .lit8 sections
+      // where appropriate.
+      MCSection *ReadOnlySection = getContext().getELFSection(
+          ".rodata", ELF::SHT_PROGBITS, ELF::SHF_ALLOC);
+
+      MCSymbol *Sym = getContext().createTempSymbol();
+      const MCExpr *LoSym =
+          MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None, getContext());
+      const MipsMCExpr *LoExpr =
+          MipsMCExpr::create(MipsMCExpr::MEK_LO, LoSym, getContext());
+
+      getStreamer().SwitchSection(ReadOnlySection);
+      getStreamer().EmitLabel(Sym, IDLoc);
+      getStreamer().EmitIntValue(ImmOp32, 4);
+      getStreamer().SwitchSection(CS);
+
+      if(emitPartialAddress(TOut, IDLoc, Sym))
+        return true;
+      TOut.emitRRX(Mips::LWC1, FirstReg, ATReg,
+                   MCOperand::createExpr(LoExpr), IDLoc, STI);
+    }
+    return false;
+  }
+
+  // if(!IsSingle)
+  unsigned ATReg = getATReg(IDLoc);
+  if (!ATReg)
+    return true;
+
+  if (IsGPR) {
+    if (LoImmOp64 == 0) {
+      if(isABI_N32() || isABI_N64()) {
+        if (loadImmediate(HiImmOp64, FirstReg, Mips::NoRegister, false, true,
+                          IDLoc, Out, STI))
+          return true;
+        return false;
+      } else {
+        if (loadImmediate(HiImmOp64, FirstReg, Mips::NoRegister, true, true,
+                        IDLoc, Out, STI))
+          return true;
+
+        if (loadImmediate(0, nextReg(FirstReg), Mips::NoRegister, true, true,
+                        IDLoc, Out, STI))
+          return true;
+        return false;
+      }
+    }
+
+    MCSection *CS = getStreamer().getCurrentSectionOnly();
+    MCSection *ReadOnlySection = getContext().getELFSection(
+        ".rodata", ELF::SHT_PROGBITS, ELF::SHF_ALLOC);
+
+    MCSymbol *Sym = getContext().createTempSymbol();
+    const MCExpr *LoSym =
+        MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None, getContext());
+    const MipsMCExpr *LoExpr =
+        MipsMCExpr::create(MipsMCExpr::MEK_LO, LoSym, getContext());
+
+    getStreamer().SwitchSection(ReadOnlySection);
+    getStreamer().EmitLabel(Sym, IDLoc);
+    getStreamer().EmitIntValue(HiImmOp64, 4);
+    getStreamer().EmitIntValue(LoImmOp64, 4);
+    getStreamer().SwitchSection(CS);
+
+    if(emitPartialAddress(TOut, IDLoc, Sym))
+      return true;
+    if(isABI_N64())
+      TOut.emitRRX(Mips::DADDiu, ATReg, ATReg,
+                   MCOperand::createExpr(LoExpr), IDLoc, STI);
+    else
+      TOut.emitRRX(Mips::ADDiu, ATReg, ATReg,
+                   MCOperand::createExpr(LoExpr), IDLoc, STI);
+
+    if(isABI_N32() || isABI_N64())
+      TOut.emitRRI(Mips::LD, FirstReg, ATReg, 0, IDLoc, STI);
+    else {
+      TOut.emitRRI(Mips::LW, FirstReg, ATReg, 0, IDLoc, STI);
+      TOut.emitRRI(Mips::LW, nextReg(FirstReg), ATReg, 4, IDLoc, STI);
+    }
+    return false;
+  } else { // if(!IsGPR && !IsSingle)
+    if ((LoImmOp64 == 0) &&
+        !((HiImmOp64 & 0xffff0000) && (HiImmOp64 & 0x0000ffff))) {
+      // FIXME: In the case where the constant is zero, we can load the
+      // register directly from the zero register.
+      if (loadImmediate(HiImmOp64, ATReg, Mips::NoRegister, true, true, IDLoc,
+                        Out, STI))
+        return true;
+      if (isABI_N32() || isABI_N64())
+        TOut.emitRR(Mips::DMTC1, FirstReg, ATReg, IDLoc, STI);
+      else if (hasMips32r2()) {
+        TOut.emitRR(Mips::MTC1, FirstReg, Mips::ZERO, IDLoc, STI);
+        TOut.emitRRR(Mips::MTHC1_D32, FirstReg, FirstReg, ATReg, IDLoc, STI);
+      } else {
+        TOut.emitRR(Mips::MTC1, nextReg(FirstReg), ATReg, IDLoc, STI);
+        TOut.emitRR(Mips::MTC1, FirstReg, Mips::ZERO, IDLoc, STI);
+      }
+      return false;
+    }
+
+    MCSection *CS = getStreamer().getCurrentSectionOnly();
+    // FIXME: Enhance this expansion to use the .lit4 & .lit8 sections
+    // where appropriate.
+    MCSection *ReadOnlySection = getContext().getELFSection(
+        ".rodata", ELF::SHT_PROGBITS, ELF::SHF_ALLOC);
+
+    MCSymbol *Sym = getContext().createTempSymbol();
+    const MCExpr *LoSym =
+        MCSymbolRefExpr::create(Sym, MCSymbolRefExpr::VK_None, getContext());
+    const MipsMCExpr *LoExpr =
+        MipsMCExpr::create(MipsMCExpr::MEK_LO, LoSym, getContext());
+
+    getStreamer().SwitchSection(ReadOnlySection);
+    getStreamer().EmitLabel(Sym, IDLoc);
+    getStreamer().EmitIntValue(HiImmOp64, 4);
+    getStreamer().EmitIntValue(LoImmOp64, 4);
+    getStreamer().SwitchSection(CS);
+
+    if(emitPartialAddress(TOut, IDLoc, Sym))
+      return true;
+    TOut.emitRRX(Is64FPU ? Mips::LDC164 : Mips::LDC1, FirstReg, ATReg,
+                 MCOperand::createExpr(LoExpr), IDLoc, STI);
+  }
+  return false;
+}
+
 bool MipsAsmParser::expandUncondBranchMMPseudo(MCInst &Inst, SMLoc IDLoc,
                                                MCStreamer &Out,
                                                const MCSubtargetInfo *STI) {
@@ -4318,45 +4666,6 @@ bool MipsAsmParser::expandDMULMacro(MCInst &Inst, SMLoc IDLoc, MCStreamer &Out,
   return false;
 }
 
-static unsigned nextReg(unsigned Reg) {
-  switch (Reg) {
-  case Mips::ZERO: return Mips::AT;
-  case Mips::AT:   return Mips::V0;
-  case Mips::V0:   return Mips::V1;
-  case Mips::V1:   return Mips::A0;
-  case Mips::A0:   return Mips::A1;
-  case Mips::A1:   return Mips::A2;
-  case Mips::A2:   return Mips::A3;
-  case Mips::A3:   return Mips::T0;
-  case Mips::T0:   return Mips::T1;
-  case Mips::T1:   return Mips::T2;
-  case Mips::T2:   return Mips::T3;
-  case Mips::T3:   return Mips::T4;
-  case Mips::T4:   return Mips::T5;
-  case Mips::T5:   return Mips::T6;
-  case Mips::T6:   return Mips::T7;
-  case Mips::T7:   return Mips::S0;
-  case Mips::S0:   return Mips::S1;
-  case Mips::S1:   return Mips::S2;
-  case Mips::S2:   return Mips::S3;
-  case Mips::S3:   return Mips::S4;
-  case Mips::S4:   return Mips::S5;
-  case Mips::S5:   return Mips::S6;
-  case Mips::S6:   return Mips::S7;
-  case Mips::S7:   return Mips::T8;
-  case Mips::T8:   return Mips::T9;
-  case Mips::T9:   return Mips::K0;
-  case Mips::K0:   return Mips::K1;
-  case Mips::K1:   return Mips::GP;
-  case Mips::GP:   return Mips::SP;
-  case Mips::SP:   return Mips::FP;
-  case Mips::FP:   return Mips::RA;
-  case Mips::RA:   return Mips::ZERO;
-  default:         return 0;
-  }
-
-}
-
 // Expand 'ld $<reg> offset($reg2)' to 'lw $<reg>, offset($reg2);
 //                                      lw $<reg+1>>, offset+4($reg2)'
 // or expand 'sd $<reg> offset($reg2)' to 'sw $<reg>, offset($reg2);

lib/Target/Mips/MipsISelLowering.cpp

@@ -362,7 +362,6 @@ MipsTargetLowering::MipsTargetLowering(const MipsTargetMachine &TM,
   setOperationAction(ISD::FCOS,              MVT::f64,   Expand);
   setOperationAction(ISD::FSINCOS,           MVT::f32,   Expand);
   setOperationAction(ISD::FSINCOS,           MVT::f64,   Expand);
-  setOperationAction(ISD::FPOWI,             MVT::f32,   Expand);
   setOperationAction(ISD::FPOW,              MVT::f32,   Expand);
   setOperationAction(ISD::FPOW,              MVT::f64,   Expand);
   setOperationAction(ISD::FLOG,              MVT::f32,   Expand);

lib/Target/Mips/MipsInstrFPU.td

@@ -681,6 +681,29 @@ def PseudoTRUNC_W_D : MipsAsmPseudoInst<(outs FGR32Opnd:$fd),
                                         "trunc.w.d\t$fd, $fs, $rs">,
                       FGR_64, HARDFLOAT;
 
+def LoadImmSingleGPR : MipsAsmPseudoInst<(outs GPR32Opnd:$rd),
+                                         (ins imm64:$fpimm),
+                                         "li.s\t$rd, $fpimm">;
+
+def LoadImmSingleFGR : MipsAsmPseudoInst<(outs StrictlyFGR32Opnd:$rd),
+                                         (ins imm64:$fpimm),
+                                         "li.s\t$rd, $fpimm">,
+                       HARDFLOAT;
+
+def LoadImmDoubleGPR : MipsAsmPseudoInst<(outs GPR32Opnd:$rd),
+                                         (ins imm64:$fpimm),
+                                         "li.d\t$rd, $fpimm">;
+
+def LoadImmDoubleFGR_32 : MipsAsmPseudoInst<(outs StrictlyAFGR64Opnd:$rd),
+                                            (ins imm64:$fpimm),
+                                            "li.d\t$rd, $fpimm">,
+                          FGR_32, HARDFLOAT;
+
+def LoadImmDoubleFGR : MipsAsmPseudoInst<(outs StrictlyFGR64Opnd:$rd),
+                                         (ins imm64:$fpimm),
+                                         "li.d\t$rd, $fpimm">,
+                       FGR_64, HARDFLOAT;
+
 //===----------------------------------------------------------------------===//
 // InstAliases.
 //===----------------------------------------------------------------------===//

lib/Target/Mips/MipsRegisterInfo.td

@@ -552,16 +552,31 @@ def AFGR64AsmOperand : MipsAsmRegOperand {
   let PredicateMethod = "isFGRAsmReg";
 }
 
+def StrictlyAFGR64AsmOperand : MipsAsmRegOperand {
+  let Name = "StrictlyAFGR64AsmReg";
+  let PredicateMethod = "isStrictlyFGRAsmReg";
+}
+
 def FGR64AsmOperand : MipsAsmRegOperand {
   let Name = "FGR64AsmReg";
   let PredicateMethod = "isFGRAsmReg";
 }
 
+def StrictlyFGR64AsmOperand : MipsAsmRegOperand {
+  let Name = "StrictlyFGR64AsmReg";
+  let PredicateMethod = "isStrictlyFGRAsmReg";
+}
+
 def FGR32AsmOperand : MipsAsmRegOperand {
   let Name = "FGR32AsmReg";
   let PredicateMethod = "isFGRAsmReg";
 }
 
+def StrictlyFGR32AsmOperand : MipsAsmRegOperand {
+  let Name = "StrictlyFGR32AsmReg";
+  let PredicateMethod = "isStrictlyFGRAsmReg";
+}
+
 def FGRH32AsmOperand : MipsAsmRegOperand {
   let Name = "FGRH32AsmReg";
   let PredicateMethod = "isFGRAsmReg";
@@ -639,14 +654,26 @@ def AFGR64Opnd : RegisterOperand<AFGR64> {
   let ParserMatchClass = AFGR64AsmOperand;
 }
 
+def StrictlyAFGR64Opnd : RegisterOperand<AFGR64> {
+  let ParserMatchClass = StrictlyAFGR64AsmOperand;
+}
+
 def FGR64Opnd : RegisterOperand<FGR64> {
   let ParserMatchClass = FGR64AsmOperand;
 }
 
+def StrictlyFGR64Opnd : RegisterOperand<FGR64> {
+  let ParserMatchClass = StrictlyFGR64AsmOperand;
+}
+
 def FGR32Opnd : RegisterOperand<FGR32> {
   let ParserMatchClass = FGR32AsmOperand;
 }
 
+def StrictlyFGR32Opnd : RegisterOperand<FGR32> {
+  let ParserMatchClass = StrictlyFGR32AsmOperand;
+}
+
 def FGRCCOpnd : RegisterOperand<FGRCC> {
   // The assembler doesn't use register classes so we can re-use
   // FGR32AsmOperand.

lib/Target/PowerPC/PPCISelLowering.cpp

@@ -539,7 +539,6 @@ PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM,
       setOperationAction(ISD::FSIN, VT, Expand);
       setOperationAction(ISD::FCOS, VT, Expand);
       setOperationAction(ISD::FABS, VT, Expand);
-      setOperationAction(ISD::FPOWI, VT, Expand);
       setOperationAction(ISD::FFLOOR, VT, Expand);
       setOperationAction(ISD::FCEIL,  VT, Expand);
       setOperationAction(ISD::FTRUNC, VT, Expand);
@@ -798,7 +797,6 @@ PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM,
     setOperationAction(ISD::FABS , MVT::v4f64, Legal);
     setOperationAction(ISD::FSIN , MVT::v4f64, Expand);
     setOperationAction(ISD::FCOS , MVT::v4f64, Expand);
-    setOperationAction(ISD::FPOWI , MVT::v4f64, Expand);
     setOperationAction(ISD::FPOW , MVT::v4f64, Expand);
     setOperationAction(ISD::FLOG , MVT::v4f64, Expand);
     setOperationAction(ISD::FLOG2 , MVT::v4f64, Expand);
@@ -844,7 +842,6 @@ PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM,
     setOperationAction(ISD::FABS , MVT::v4f32, Legal);
     setOperationAction(ISD::FSIN , MVT::v4f32, Expand);
     setOperationAction(ISD::FCOS , MVT::v4f32, Expand);
-    setOperationAction(ISD::FPOWI , MVT::v4f32, Expand);
     setOperationAction(ISD::FPOW , MVT::v4f32, Expand);
     setOperationAction(ISD::FLOG , MVT::v4f32, Expand);
     setOperationAction(ISD::FLOG2 , MVT::v4f32, Expand);

lib/Target/SystemZ/SystemZ.td

@@ -54,6 +54,8 @@ include "SystemZInstrFormats.td"
 include "SystemZInstrInfo.td"
 include "SystemZInstrVector.td"
 include "SystemZInstrFP.td"
+include "SystemZInstrHFP.td"
+include "SystemZInstrDFP.td"
 
 def SystemZInstrInfo : InstrInfo {}
 

lib/Target/SystemZ/SystemZFeatures.td

@@ -115,12 +115,18 @@ def FeatureTransactionalExecution : SystemZFeature<
   "Assume that the transactional-execution facility is installed"
 >;
 
+def FeatureDFPZonedConversion : SystemZFeature<
+  "dfp-zoned-conversion", "DFPZonedConversion",
+  "Assume that the DFP zoned-conversion facility is installed"
+>;
+
 def Arch10NewFeatures : SystemZFeatureList<[
     FeatureExecutionHint,
     FeatureLoadAndTrap,
     FeatureMiscellaneousExtensions,
     FeatureProcessorAssist,
-    FeatureTransactionalExecution
+    FeatureTransactionalExecution,
+    FeatureDFPZonedConversion
 ]>;
 
 //===----------------------------------------------------------------------===//
@@ -144,6 +150,11 @@ def FeatureMessageSecurityAssist5 : SystemZFeature<
   "Assume that the message-security-assist extension facility 5 is installed"
 >;
 
+def FeatureDFPPackedConversion : SystemZFeature<
+  "dfp-packed-conversion", "DFPPackedConversion",
+  "Assume that the DFP packed-conversion facility is installed"
+>;
+
 def FeatureVector : SystemZFeature<
   "vector", "Vector",
   "Assume that the vectory facility is installed"
@@ -154,6 +165,7 @@ def Arch11NewFeatures : SystemZFeatureList<[
     FeatureLoadAndZeroRightmostByte,
     FeatureLoadStoreOnCond2,
     FeatureMessageSecurityAssist5,
+    FeatureDFPPackedConversion,
     FeatureVector
 ]>;
 

lib/Target/SystemZ/SystemZISelLowering.cpp

@@ -4189,12 +4189,20 @@ static SDValue buildVector(SelectionDAG &DAG, const SDLoc &DL, EVT VT,
   if (Single.getNode() && (Count > 1 || Single.getOpcode() == ISD::LOAD))
     return DAG.getNode(SystemZISD::REPLICATE, DL, VT, Single);
 
+  // If all elements are loads, use VLREP/VLEs (below).
+  bool AllLoads = true;
+  for (auto Elem : Elems)
+    if (Elem.getOpcode() != ISD::LOAD || cast<LoadSDNode>(Elem)->isIndexed()) {
+      AllLoads = false;
+      break;
+    }
+
   // The best way of building a v2i64 from two i64s is to use VLVGP.
-  if (VT == MVT::v2i64)
+  if (VT == MVT::v2i64 && !AllLoads)
     return joinDwords(DAG, DL, Elems[0], Elems[1]);
 
   // Use a 64-bit merge high to combine two doubles.
-  if (VT == MVT::v2f64)
+  if (VT == MVT::v2f64 && !AllLoads)
     return buildMergeScalars(DAG, DL, VT, Elems[0], Elems[1]);
 
   // Build v4f32 values directly from the FPRs:
@@ -4204,7 +4212,7 @@ static SDValue buildVector(SelectionDAG &DAG, const SDLoc &DL, EVT VT,
   //      <ABxx>         <CDxx>
   //                V                 VMRHG
   //              <ABCD>
-  if (VT == MVT::v4f32) {
+  if (VT == MVT::v4f32 && !AllLoads) {
     SDValue Op01 = buildMergeScalars(DAG, DL, VT, Elems[0], Elems[1]);
     SDValue Op23 = buildMergeScalars(DAG, DL, VT, Elems[2], Elems[3]);
     // Avoid unnecessary undefs by reusing the other operand.
@@ -4246,23 +4254,37 @@ static SDValue buildVector(SelectionDAG &DAG, const SDLoc &DL, EVT VT,
         Constants[I] = DAG.getUNDEF(Elems[I].getValueType());
     Result = DAG.getBuildVector(VT, DL, Constants);
   } else {
-    // Otherwise try to use VLVGP to start the sequence in order to
+    // Otherwise try to use VLREP or VLVGP to start the sequence in order to
     // avoid a false dependency on any previous contents of the vector
-    // register.  This only makes sense if one of the associated elements
-    // is defined.
-    unsigned I1 = NumElements / 2 - 1;
-    unsigned I2 = NumElements - 1;
-    bool Def1 = !Elems[I1].isUndef();
-    bool Def2 = !Elems[I2].isUndef();
-    if (Def1 || Def2) {
-      SDValue Elem1 = Elems[Def1 ? I1 : I2];
-      SDValue Elem2 = Elems[Def2 ? I2 : I1];
-      Result = DAG.getNode(ISD::BITCAST, DL, VT,
-                           joinDwords(DAG, DL, Elem1, Elem2));
-      Done[I1] = true;
-      Done[I2] = true;
-    } else
-      Result = DAG.getUNDEF(VT);
+    // register.
+
+    // Use a VLREP if at least one element is a load.
+    unsigned LoadElIdx = UINT_MAX;
+    for (unsigned I = 0; I < NumElements; ++I)
+      if (Elems[I].getOpcode() == ISD::LOAD &&
+          cast<LoadSDNode>(Elems[I])->isUnindexed()) {
+        LoadElIdx = I;
+        break;
+      }
+    if (LoadElIdx != UINT_MAX) {
+      Result = DAG.getNode(SystemZISD::REPLICATE, DL, VT, Elems[LoadElIdx]);
+      Done[LoadElIdx] = true;
+    } else {
+      // Try to use VLVGP.
+      unsigned I1 = NumElements / 2 - 1;
+      unsigned I2 = NumElements - 1;
+      bool Def1 = !Elems[I1].isUndef();
+      bool Def2 = !Elems[I2].isUndef();
+      if (Def1 || Def2) {
+        SDValue Elem1 = Elems[Def1 ? I1 : I2];
+        SDValue Elem2 = Elems[Def2 ? I2 : I1];
+        Result = DAG.getNode(ISD::BITCAST, DL, VT,
+                             joinDwords(DAG, DL, Elem1, Elem2));
+        Done[I1] = true;
+        Done[I2] = true;
+      } else
+        Result = DAG.getUNDEF(VT);
+    }
   }
 
   // Use VLVGx to insert the other elements.

lib/Target/SystemZ/SystemZInstrDFP.td

@@ -0,0 +1,231 @@
+//==- SystemZInstrDFP.td - Floating-point SystemZ instructions -*- tblgen-*-==//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// The instructions in this file implement SystemZ decimal floating-point
+// arithmetic.  These instructions are inot currently used for code generation,
+// are provided for use with the assembler and disassembler only.  If LLVM
+// ever supports decimal floating-point types (_Decimal64 etc.), they can
+// also be used for code generation for those types.
+//
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// Move instructions
+//===----------------------------------------------------------------------===//
+
+// Load and test.
+let Defs = [CC] in {
+  def LTDTR : UnaryRRE<"ltdtr", 0xB3D6, null_frag, FP64,  FP64>;
+  def LTXTR : UnaryRRE<"ltxtr", 0xB3DE, null_frag, FP128, FP128>;
+}
+
+
+//===----------------------------------------------------------------------===//
+// Conversion instructions
+//===----------------------------------------------------------------------===//
+
+// Convert floating-point values to narrower representations.  The destination
+// of LDXTR is a 128-bit value, but only the first register of the pair is used.
+def LEDTR : TernaryRRFe<"ledtr", 0xB3D5, FP32,  FP64>;
+def LDXTR : TernaryRRFe<"ldxtr", 0xB3DD, FP128, FP128>;
+
+// Extend floating-point values to wider representations.
+def LDETR : BinaryRRFd<"ldetr", 0xB3D4, FP64,  FP32>;
+def LXDTR : BinaryRRFd<"lxdtr", 0xB3DC, FP128, FP64>;
+
+// Convert a signed integer value to a floating-point one.
+def CDGTR : UnaryRRE<"cdgtr", 0xB3F1, null_frag, FP64,  GR64>;
+def CXGTR : UnaryRRE<"cxgtr", 0xB3F9, null_frag, FP128, GR64>;
+let Predicates = [FeatureFPExtension] in {
+  def CDGTRA : TernaryRRFe<"cdgtra", 0xB3F1, FP64,  GR64>;
+  def CXGTRA : TernaryRRFe<"cxgtra", 0xB3F9, FP128, GR64>;
+  def CDFTR : TernaryRRFe<"cdftr", 0xB951, FP64,  GR32>;
+  def CXFTR : TernaryRRFe<"cxftr", 0xB959, FP128, GR32>;
+}
+
+// Convert an unsigned integer value to a floating-point one.
+let Predicates = [FeatureFPExtension] in {
+  def CDLGTR : TernaryRRFe<"cdlgtr", 0xB952, FP64,  GR64>;
+  def CXLGTR : TernaryRRFe<"cxlgtr", 0xB95A, FP128, GR64>;
+  def CDLFTR : TernaryRRFe<"cdlftr", 0xB953, FP64,  GR32>;
+  def CXLFTR : TernaryRRFe<"cxlftr", 0xB95B, FP128, GR32>;
+}
+
+// Convert a floating-point value to a signed integer value.
+let Defs = [CC] in {
+  def CGDTR : BinaryRRFe<"cgdtr", 0xB3E1, GR64, FP64>;
+  def CGXTR : BinaryRRFe<"cgxtr", 0xB3E9, GR64, FP128>;
+  let Predicates = [FeatureFPExtension] in {
+    def CGDTRA : TernaryRRFe<"cgdtra", 0xB3E1, GR64, FP64>;
+    def CGXTRA : TernaryRRFe<"cgxtra", 0xB3E9, GR64, FP128>;
+    def CFDTR : TernaryRRFe<"cfdtr", 0xB941, GR32, FP64>;
+    def CFXTR : TernaryRRFe<"cfxtr", 0xB949, GR32, FP128>;
+  }
+}
+
+// Convert a floating-point value to an unsigned integer value.
+let Defs = [CC] in {
+  let Predicates = [FeatureFPExtension] in {
+    def CLGDTR : TernaryRRFe<"clgdtr", 0xB942, GR64, FP64>;
+    def CLGXTR : TernaryRRFe<"clgxtr", 0xB94A, GR64, FP128>;
+    def CLFDTR : TernaryRRFe<"clfdtr", 0xB943, GR32, FP64>;
+    def CLFXTR : TernaryRRFe<"clfxtr", 0xB94B, GR32, FP128>;
+  }
+}
+
+// Convert a packed value to a floating-point one.
+def CDSTR : UnaryRRE<"cdstr", 0xB3F3, null_frag, FP64,  GR64>;
+def CXSTR : UnaryRRE<"cxstr", 0xB3FB, null_frag, FP128, GR128>;
+def CDUTR : UnaryRRE<"cdutr", 0xB3F2, null_frag, FP64,  GR64>;
+def CXUTR : UnaryRRE<"cxutr", 0xB3FA, null_frag, FP128, GR128>;
+
+// Convert a floating-point value to a packed value.
+def CSDTR : BinaryRRFd<"csdtr", 0xB3E3, GR64,  FP64>;
+def CSXTR : BinaryRRFd<"csxtr", 0xB3EB, GR128, FP128>;
+def CUDTR : UnaryRRE<"cudtr", 0xB3E2, null_frag, GR64,  FP64>;
+def CUXTR : UnaryRRE<"cuxtr", 0xB3EA, null_frag, GR128, FP128>;
+
+// Convert from/to memory values in the zoned format.
+let Predicates = [FeatureDFPZonedConversion] in {
+  def CDZT : BinaryRSL<"cdzt", 0xEDAA, FP64>;
+  def CXZT : BinaryRSL<"cxzt", 0xEDAB, FP128>;
+  def CZDT : StoreBinaryRSL<"czdt", 0xEDA8, FP64>;
+  def CZXT : StoreBinaryRSL<"czxt", 0xEDA9, FP128>;
+}
+
+// Convert from/to memory values in the packed format.
+let Predicates = [FeatureDFPPackedConversion] in {
+  def CDPT : BinaryRSL<"cdpt", 0xEDAE, FP64>;
+  def CXPT : BinaryRSL<"cxpt", 0xEDAF, FP128>;
+  def CPDT : StoreBinaryRSL<"cpdt", 0xEDAC, FP64>;
+  def CPXT : StoreBinaryRSL<"cpxt", 0xEDAD, FP128>;
+}
+
+// Perform floating-point operation.
+let Defs = [CC, R1L, F0Q], Uses = [R0L, F4Q] in
+  def PFPO : SideEffectInherentE<"pfpo", 0x010A>;
+
+
+//===----------------------------------------------------------------------===//
+// Unary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Round to an integer, with the second operand (M3) specifying the rounding
+// mode.  M4 can be set to 4 to suppress detection of inexact conditions.
+def FIDTR : TernaryRRFe<"fidtr", 0xB3D7, FP64,  FP64>;
+def FIXTR : TernaryRRFe<"fixtr", 0xB3DF, FP128, FP128>;
+
+// Extract biased exponent.
+def EEDTR : UnaryRRE<"eedtr", 0xB3E5, null_frag, FP64,  FP64>;
+def EEXTR : UnaryRRE<"eextr", 0xB3ED, null_frag, FP128, FP128>;
+
+// Extract significance.
+def ESDTR : UnaryRRE<"esdtr", 0xB3E7, null_frag, FP64,  FP64>;
+def ESXTR : UnaryRRE<"esxtr", 0xB3EF, null_frag, FP128, FP128>;
+
+
+//===----------------------------------------------------------------------===//
+// Binary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Addition.
+let Defs = [CC] in {
+  let isCommutable = 1 in {
+    def ADTR : BinaryRRFa<"adtr", 0xB3D2, null_frag, FP64,  FP64,  FP64>;
+    def AXTR : BinaryRRFa<"axtr", 0xB3DA, null_frag, FP128, FP128, FP128>;
+  }
+  let Predicates = [FeatureFPExtension] in {
+    def ADTRA : TernaryRRFa<"adtra", 0xB3D2, FP64,  FP64,  FP64>;
+    def AXTRA : TernaryRRFa<"axtra", 0xB3DA, FP128, FP128, FP128>;
+  }
+}
+
+// Subtraction.
+let Defs = [CC] in {
+  def SDTR : BinaryRRFa<"sdtr", 0xB3D3, null_frag, FP64,  FP64,  FP64>;
+  def SXTR : BinaryRRFa<"sxtr", 0xB3DB, null_frag, FP128, FP128, FP128>;
+  let Predicates = [FeatureFPExtension] in {
+    def SDTRA : TernaryRRFa<"sdtra", 0xB3D3, FP64,  FP64,  FP64>;
+    def SXTRA : TernaryRRFa<"sxtra", 0xB3DB, FP128, FP128, FP128>;
+  }
+}
+
+// Multiplication.
+let isCommutable = 1 in {
+  def MDTR : BinaryRRFa<"mdtr", 0xB3D0, null_frag, FP64,  FP64,  FP64>;
+  def MXTR : BinaryRRFa<"mxtr", 0xB3D8, null_frag, FP128, FP128, FP128>;
+}
+let Predicates = [FeatureFPExtension] in {
+  def MDTRA : TernaryRRFa<"mdtra", 0xB3D0, FP64,  FP64,  FP64>;
+  def MXTRA : TernaryRRFa<"mxtra", 0xB3D8, FP128, FP128, FP128>;
+}
+
+// Division.
+def DDTR : BinaryRRFa<"ddtr", 0xB3D1, null_frag, FP64,  FP64,  FP64>;
+def DXTR : BinaryRRFa<"dxtr", 0xB3D9, null_frag, FP128, FP128, FP128>;
+let Predicates = [FeatureFPExtension] in {
+  def DDTRA : TernaryRRFa<"ddtra", 0xB3D1, FP64,  FP64,  FP64>;
+  def DXTRA : TernaryRRFa<"dxtra", 0xB3D9, FP128, FP128, FP128>;
+}
+
+// Quantize.
+def QADTR : TernaryRRFb<"qadtr", 0xB3F5, FP64,  FP64,  FP64>;
+def QAXTR : TernaryRRFb<"qaxtr", 0xB3FD, FP128, FP128, FP128>;
+
+// Reround.
+def RRDTR : TernaryRRFb<"rrdtr", 0xB3F7, FP64,  FP64,  FP64>;
+def RRXTR : TernaryRRFb<"rrxtr", 0xB3FF, FP128, FP128, FP128>;
+
+// Shift significand left/right.
+def SLDT : BinaryRXF<"sldt", 0xED40, null_frag, FP64,  FP64,  null_frag, 0>;
+def SLXT : BinaryRXF<"slxt", 0xED48, null_frag, FP128, FP128, null_frag, 0>;
+def SRDT : BinaryRXF<"srdt", 0xED41, null_frag, FP64,  FP64,  null_frag, 0>;
+def SRXT : BinaryRXF<"srxt", 0xED49, null_frag, FP128, FP128, null_frag, 0>;
+
+// Insert biased exponent.
+def IEDTR : BinaryRRFb<"iedtr", 0xB3F6, null_frag, FP64,  FP64,   FP64>;
+def IEXTR : BinaryRRFb<"iextr", 0xB3FE, null_frag, FP128, FP128, FP128>;
+
+
+//===----------------------------------------------------------------------===//
+// Comparisons
+//===----------------------------------------------------------------------===//
+
+// Compare.
+let Defs = [CC] in {
+  def CDTR : CompareRRE<"cdtr", 0xB3E4, null_frag, FP64,  FP64>;
+  def CXTR : CompareRRE<"cxtr", 0xB3EC, null_frag, FP128, FP128>;
+}
+
+// Compare and signal.
+let Defs = [CC] in {
+  def KDTR : CompareRRE<"kdtr", 0xB3E0, null_frag, FP64,  FP64>;
+  def KXTR : CompareRRE<"kxtr", 0xB3E8, null_frag, FP128, FP128>;
+}
+
+// Compare biased exponent.
+let Defs = [CC] in {
+  def CEDTR : CompareRRE<"cedtr", 0xB3F4, null_frag, FP64,  FP64>;
+  def CEXTR : CompareRRE<"cextr", 0xB3FC, null_frag, FP128, FP128>;
+}
+
+// Test Data Class.
+let Defs = [CC] in {
+  def TDCET : TestRXE<"tdcet", 0xED50, null_frag, FP32>;
+  def TDCDT : TestRXE<"tdcdt", 0xED54, null_frag, FP64>;
+  def TDCXT : TestRXE<"tdcxt", 0xED58, null_frag, FP128>;
+}
+
+// Test Data Group.
+let Defs = [CC] in {
+  def TDGET : TestRXE<"tdget", 0xED51, null_frag, FP32>;
+  def TDGDT : TestRXE<"tdgdt", 0xED55, null_frag, FP64>;
+  def TDGXT : TestRXE<"tdgxt", 0xED59, null_frag, FP128>;
+}
+

lib/Target/SystemZ/SystemZInstrFP.td

@@ -121,7 +121,8 @@ let canFoldAsLoad = 1, SimpleBDXLoad = 1 in {
   defm LD : UnaryRXPair<"ld", 0x68, 0xED65, load, FP64, 8>;
 
   // For z13 we prefer LDE over LE to avoid partial register dependencies.
-  def LDE32 : UnaryRXE<"lde", 0xED24, null_frag, FP32, 4>;
+  let isCodeGenOnly = 1 in
+    def LDE32 : UnaryRXE<"lde", 0xED24, null_frag, FP32, 4>;
 
   // These instructions are split after register allocation, so we don't
   // want a custom inserter.
@@ -437,18 +438,18 @@ def : Pat<(fmul (f128 (fpextend FP64:$src1)),
                 bdxaddr12only:$addr)>;
 
 // Fused multiply-add.
-def MAEBR : TernaryRRD<"maebr", 0xB30E, z_fma, FP32>;
-def MADBR : TernaryRRD<"madbr", 0xB31E, z_fma, FP64>;
+def MAEBR : TernaryRRD<"maebr", 0xB30E, z_fma, FP32, FP32>;
+def MADBR : TernaryRRD<"madbr", 0xB31E, z_fma, FP64, FP64>;
 
-def MAEB : TernaryRXF<"maeb", 0xED0E, z_fma, FP32, load, 4>;
-def MADB : TernaryRXF<"madb", 0xED1E, z_fma, FP64, load, 8>;
+def MAEB : TernaryRXF<"maeb", 0xED0E, z_fma, FP32, FP32, load, 4>;
+def MADB : TernaryRXF<"madb", 0xED1E, z_fma, FP64, FP64, load, 8>;
 
 // Fused multiply-subtract.
-def MSEBR : TernaryRRD<"msebr", 0xB30F, z_fms, FP32>;
-def MSDBR : TernaryRRD<"msdbr", 0xB31F, z_fms, FP64>;
+def MSEBR : TernaryRRD<"msebr", 0xB30F, z_fms, FP32, FP32>;
+def MSDBR : TernaryRRD<"msdbr", 0xB31F, z_fms, FP64, FP64>;
 
-def MSEB : TernaryRXF<"mseb", 0xED0F, z_fms, FP32, load, 4>;
-def MSDB : TernaryRXF<"msdb", 0xED1F, z_fms, FP64, load, 8>;
+def MSEB : TernaryRXF<"mseb", 0xED0F, z_fms, FP32, FP32, load, 4>;
+def MSDB : TernaryRXF<"msdb", 0xED1F, z_fms, FP64, FP64, load, 8>;
 
 // Division.
 def DEBR : BinaryRRE<"debr", 0xB30D, fdiv, FP32,  FP32>;

lib/Target/SystemZ/SystemZInstrFormats.td

@@ -527,6 +527,22 @@ class InstRRFc<bits<16> op, dag outs, dag ins, string asmstr, list<dag> pattern>
   let Inst{3-0}   = R2;
 }
 
+class InstRRFd<bits<16> op, dag outs, dag ins, string asmstr, list<dag> pattern>
+  : InstSystemZ<4, outs, ins, asmstr, pattern> {
+  field bits<32> Inst;
+  field bits<32> SoftFail = 0;
+
+  bits<4> R1;
+  bits<4> R2;
+  bits<4> M4;
+
+  let Inst{31-16} = op;
+  let Inst{15-12} = 0;
+  let Inst{11-8}  = M4;
+  let Inst{7-4}   = R1;
+  let Inst{3-0}   = R2;
+}
+
 class InstRRFe<bits<16> op, dag outs, dag ins, string asmstr, list<dag> pattern>
   : InstSystemZ<4, outs, ins, asmstr, pattern> {
   field bits<32> Inst;
@@ -725,6 +741,22 @@ class InstRSLa<bits<16> op, dag outs, dag ins, string asmstr, list<dag> pattern>
   let Inst{7-0}   = op{7-0};
 }
 
+class InstRSLb<bits<16> op, dag outs, dag ins, string asmstr, list<dag> pattern>
+  : InstSystemZ<6, outs, ins, asmstr, pattern> {
+  field bits<48> Inst;
+  field bits<48> SoftFail = 0;
+
+  bits<4> R1;
+  bits<24> BDL2;
+  bits<4> M3;
+
+  let Inst{47-40} = op{15-8};
+  let Inst{39-16} = BDL2;
+  let Inst{15-12} = R1;
+  let Inst{11-8}  = M3;
+  let Inst{7-0}   = op{7-0};
+}
+
 class InstRSYa<bits<16> op, dag outs, dag ins, string asmstr, list<dag> pattern>
   : InstSystemZ<6, outs, ins, asmstr, pattern> {
   field bits<48> Inst;
@@ -2752,6 +2784,15 @@ class BinaryRRE<string mnemonic, bits<16> opcode, SDPatternOperator operator,
   let DisableEncoding = "$R1src";
 }
 
+class BinaryRRD<string mnemonic, bits<16> opcode, SDPatternOperator operator,
+                RegisterOperand cls1, RegisterOperand cls2>
+  : InstRRD<opcode, (outs cls1:$R1), (ins cls2:$R3, cls2:$R2),
+            mnemonic#"\t$R1, $R3, $R2",
+            [(set cls1:$R1, (operator cls2:$R3, cls2:$R2))]> {
+  let OpKey = mnemonic#cls;
+  let OpType = "reg";
+}
+
 class BinaryRRFa<string mnemonic, bits<16> opcode, SDPatternOperator operator,
                  RegisterOperand cls1, RegisterOperand cls2,
                  RegisterOperand cls3>
@@ -2808,6 +2849,11 @@ multiclass BinaryMemRRFcOpt<string mnemonic, bits<16> opcode,
   def Opt : UnaryMemRRFc<mnemonic, opcode, cls1, cls2>;
 }
 
+class BinaryRRFd<string mnemonic, bits<16> opcode, RegisterOperand cls1,
+                RegisterOperand cls2>
+  : InstRRFd<opcode, (outs cls1:$R1), (ins cls2:$R2, imm32zx4:$M4),
+             mnemonic#"\t$R1, $R2, $M4", []>;
+
 class BinaryRRFe<string mnemonic, bits<16> opcode, RegisterOperand cls1,
                 RegisterOperand cls2>
   : InstRRFe<opcode, (outs cls1:$R1), (ins imm32zx4:$M3, cls2:$R2),
@@ -2958,6 +3004,13 @@ multiclass BinaryRSAndK<string mnemonic, bits<8> opcode1, bits<16> opcode2,
   }
 }
 
+class BinaryRSL<string mnemonic, bits<16> opcode, RegisterOperand cls>
+  : InstRSLb<opcode, (outs cls:$R1),
+             (ins bdladdr12onlylen8:$BDL2, imm32zx4:$M3),
+             mnemonic#"\t$R1, $BDL2, $M3", []> {
+  let mayLoad = 1;
+}
+
 class BinaryRX<string mnemonic, bits<8> opcode, SDPatternOperator operator,
                RegisterOperand cls, SDPatternOperator load, bits<5> bytes,
                AddressingMode mode = bdxaddr12only>
@@ -2987,6 +3040,18 @@ class BinaryRXE<string mnemonic, bits<16> opcode, SDPatternOperator operator,
   let M3 = 0;
 }
 
+class BinaryRXF<string mnemonic, bits<16> opcode, SDPatternOperator operator,
+                RegisterOperand cls1, RegisterOperand cls2,
+                SDPatternOperator load, bits<5> bytes>
+  : InstRXF<opcode, (outs cls1:$R1), (ins cls2:$R3, bdxaddr12only:$XBD2),
+            mnemonic#"\t$R1, $R3, $XBD2",
+            [(set cls1:$R1, (operator cls2:$R3, (load bdxaddr12only:$XBD2)))]> {
+  let OpKey = mnemonic#"r"#cls;
+  let OpType = "mem";
+  let mayLoad = 1;
+  let AccessBytes = bytes;
+}
+
 class BinaryRXY<string mnemonic, bits<16> opcode, SDPatternOperator operator,
                 RegisterOperand cls, SDPatternOperator load, bits<5> bytes,
                 AddressingMode mode = bdxaddr20only>
@@ -3294,6 +3359,13 @@ multiclass StoreBinaryRSPair<string mnemonic, bits<8> rsOpcode,
   }
 }
 
+class StoreBinaryRSL<string mnemonic, bits<16> opcode, RegisterOperand cls>
+  : InstRSLb<opcode, (outs),
+             (ins cls:$R1, bdladdr12onlylen8:$BDL2, imm32zx4:$M3),
+             mnemonic#"\t$R1, $BDL2, $M3", []> {
+  let mayStore = 1;
+}
+
 class StoreBinaryVRV<string mnemonic, bits<16> opcode, bits<5> bytes,
                      Immediate index>
   : InstVRV<opcode, (outs), (ins VR128:$V1, bdvaddr12only:$VBD2, index:$M3),
@@ -3581,6 +3653,12 @@ class SideEffectTernarySSF<string mnemonic, bits<12> opcode,
             (ins bdaddr12only:$BD1, bdaddr12only:$BD2, cls:$R3),
             mnemonic#"\t$BD1, $BD2, $R3", []>;
 
+class TernaryRRFa<string mnemonic, bits<16> opcode,
+                 RegisterOperand cls1, RegisterOperand cls2,
+                 RegisterOperand cls3>
+  : InstRRFa<opcode, (outs cls1:$R1), (ins cls2:$R2, cls3:$R3, imm32zx4:$M4),
+             mnemonic#"\t$R1, $R2, $R3, $M4", []>;
+
 class TernaryRRFb<string mnemonic, bits<16> opcode,
                   RegisterOperand cls1, RegisterOperand cls2,
                   RegisterOperand cls3>
@@ -3597,11 +3675,11 @@ class TernaryRRFe<string mnemonic, bits<16> opcode, RegisterOperand cls1,
              (ins imm32zx4:$M3, cls2:$R2, imm32zx4:$M4),
              mnemonic#"\t$R1, $M3, $R2, $M4", []>;
 
-class TernaryRRD<string mnemonic, bits<16> opcode,
-                 SDPatternOperator operator, RegisterOperand cls>
-  : InstRRD<opcode, (outs cls:$R1), (ins cls:$R1src, cls:$R3, cls:$R2),
+class TernaryRRD<string mnemonic, bits<16> opcode, SDPatternOperator operator,
+                 RegisterOperand cls1, RegisterOperand cls2>
+  : InstRRD<opcode, (outs cls1:$R1), (ins cls2:$R1src, cls2:$R3, cls2:$R2),
             mnemonic#"\t$R1, $R3, $R2",
-            [(set cls:$R1, (operator cls:$R1src, cls:$R3, cls:$R2))]> {
+            [(set cls1:$R1, (operator cls2:$R1src, cls2:$R3, cls2:$R2))]> {
   let OpKey = mnemonic#cls;
   let OpType = "reg";
   let Constraints = "$R1 = $R1src";
@@ -3661,12 +3739,13 @@ class SideEffectTernaryMemMemRSY<string mnemonic, bits<16> opcode,
 }
 
 class TernaryRXF<string mnemonic, bits<16> opcode, SDPatternOperator operator,
-                 RegisterOperand cls, SDPatternOperator load, bits<5> bytes>
-  : InstRXF<opcode, (outs cls:$R1),
-            (ins cls:$R1src, cls:$R3, bdxaddr12only:$XBD2),
+                 RegisterOperand cls1, RegisterOperand cls2,
+                 SDPatternOperator load, bits<5> bytes>
+  : InstRXF<opcode, (outs cls1:$R1),
+            (ins cls2:$R1src, cls2:$R3, bdxaddr12only:$XBD2),
             mnemonic#"\t$R1, $R3, $XBD2",
-            [(set cls:$R1, (operator cls:$R1src, cls:$R3,
-                                     (load bdxaddr12only:$XBD2)))]> {
+            [(set cls1:$R1, (operator cls2:$R1src, cls2:$R3,
+                                      (load bdxaddr12only:$XBD2)))]> {
   let OpKey = mnemonic#"r"#cls;
   let OpType = "mem";
   let Constraints = "$R1 = $R1src";

lib/Target/SystemZ/SystemZInstrHFP.td

@@ -0,0 +1,240 @@
+//==- SystemZInstrHFP.td - Floating-point SystemZ instructions -*- tblgen-*-==//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// The instructions in this file implement SystemZ hexadecimal floating-point
+// arithmetic.  Since this format is not mapped to any source-language data
+// type, these instructions are not used for code generation, but are provided
+// for use with the assembler and disassembler only.
+//
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// Move instructions
+//===----------------------------------------------------------------------===//
+
+// Load and test.
+let Defs = [CC] in {
+  def LTER : UnaryRR <"lter", 0x32,   null_frag, FP32,  FP32>;
+  def LTDR : UnaryRR <"ltdr", 0x22,   null_frag, FP64,  FP64>;
+  def LTXR : UnaryRRE<"ltxr", 0xB362, null_frag, FP128, FP128>;
+}
+
+//===----------------------------------------------------------------------===//
+// Conversion instructions
+//===----------------------------------------------------------------------===//
+
+// Convert floating-point values to narrower representations.
+def LEDR : UnaryRR <"ledr", 0x35,   null_frag, FP32, FP64>;
+def LEXR : UnaryRRE<"lexr", 0xB366, null_frag, FP32, FP128>;
+def LDXR : UnaryRR <"ldxr", 0x25,   null_frag, FP64, FP128>;
+let isAsmParserOnly = 1 in {
+  def LRER : UnaryRR <"lrer", 0x35, null_frag, FP32, FP64>;
+  def LRDR : UnaryRR <"lrdr", 0x25, null_frag, FP64, FP128>;
+}
+
+// Extend floating-point values to wider representations.
+def LDER : UnaryRRE<"lder", 0xB324, null_frag, FP64,  FP32>;
+def LXER : UnaryRRE<"lxer", 0xB326, null_frag, FP128, FP32>;
+def LXDR : UnaryRRE<"lxdr", 0xB325, null_frag, FP128, FP64>;
+
+def LDE : UnaryRXE<"lde", 0xED24, null_frag, FP64,  4>;
+def LXE : UnaryRXE<"lxe", 0xED26, null_frag, FP128, 4>;
+def LXD : UnaryRXE<"lxd", 0xED25, null_frag, FP128, 8>;
+
+// Convert a signed integer register value to a floating-point one.
+def CEFR : UnaryRRE<"cefr", 0xB3B4, null_frag, FP32,  GR32>;
+def CDFR : UnaryRRE<"cdfr", 0xB3B5, null_frag, FP64,  GR32>;
+def CXFR : UnaryRRE<"cxfr", 0xB3B6, null_frag, FP128, GR32>;
+
+def CEGR : UnaryRRE<"cegr", 0xB3C4, null_frag, FP32,  GR64>;
+def CDGR : UnaryRRE<"cdgr", 0xB3C5, null_frag, FP64,  GR64>;
+def CXGR : UnaryRRE<"cxgr", 0xB3C6, null_frag, FP128, GR64>;
+
+// Convert a floating-point register value to a signed integer value,
+// with the second operand (modifier M3) specifying the rounding mode.
+let Defs = [CC] in {
+  def CFER : BinaryRRFe<"cfer", 0xB3B8, GR32, FP32>;
+  def CFDR : BinaryRRFe<"cfdr", 0xB3B9, GR32, FP64>;
+  def CFXR : BinaryRRFe<"cfxr", 0xB3BA, GR32, FP128>;
+
+  def CGER : BinaryRRFe<"cger", 0xB3C8, GR64, FP32>;
+  def CGDR : BinaryRRFe<"cgdr", 0xB3C9, GR64, FP64>;
+  def CGXR : BinaryRRFe<"cgxr", 0xB3CA, GR64, FP128>;
+}
+
+// Convert BFP to HFP.
+let Defs = [CC] in {
+  def THDER : UnaryRRE<"thder", 0xB358, null_frag, FP64, FP32>;
+  def THDR  : UnaryRRE<"thdr",  0xB359, null_frag, FP64, FP64>;
+}
+
+// Convert HFP to BFP.
+let Defs = [CC] in {
+  def TBEDR : BinaryRRFe<"tbedr", 0xB350, FP32, FP64>;
+  def TBDR  : BinaryRRFe<"tbdr",  0xB351, FP64, FP64>;
+}
+
+
+//===----------------------------------------------------------------------===//
+// Unary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Negation (Load Complement).
+let Defs = [CC] in {
+  def LCER : UnaryRR <"lcer", 0x33,   null_frag, FP32,  FP32>;
+  def LCDR : UnaryRR <"lcdr", 0x23,   null_frag, FP64,  FP64>;
+  def LCXR : UnaryRRE<"lcxr", 0xB363, null_frag, FP128, FP128>;
+}
+
+// Absolute value (Load Positive).
+let Defs = [CC] in {
+  def LPER : UnaryRR <"lper", 0x30,   null_frag, FP32,  FP32>;
+  def LPDR : UnaryRR <"lpdr", 0x20,   null_frag, FP64,  FP64>;
+  def LPXR : UnaryRRE<"lpxr", 0xB360, null_frag, FP128, FP128>;
+}
+
+// Negative absolute value (Load Negative).
+let Defs = [CC] in {
+  def LNER : UnaryRR <"lner", 0x31,   null_frag, FP32,  FP32>;
+  def LNDR : UnaryRR <"lndr", 0x21,   null_frag, FP64,  FP64>;
+  def LNXR : UnaryRRE<"lnxr", 0xB361, null_frag, FP128, FP128>;
+}
+
+// Halve.
+def HER : UnaryRR <"her", 0x34, null_frag, FP32, FP32>;
+def HDR : UnaryRR <"hdr", 0x24, null_frag, FP64, FP64>;
+
+// Square root.
+def SQER : UnaryRRE<"sqer", 0xB245, null_frag, FP32,  FP32>;
+def SQDR : UnaryRRE<"sqdr", 0xB244, null_frag, FP64,  FP64>;
+def SQXR : UnaryRRE<"sqxr", 0xB336, null_frag, FP128, FP128>;
+
+def SQE : UnaryRXE<"sqe", 0xED34, null_frag, FP32, 4>;
+def SQD : UnaryRXE<"sqd", 0xED35, null_frag, FP64, 8>;
+
+// Round to an integer (rounding towards zero).
+def FIER : UnaryRRE<"fier", 0xB377, null_frag, FP32,  FP32>;
+def FIDR : UnaryRRE<"fidr", 0xB37F, null_frag, FP64,  FP64>;
+def FIXR : UnaryRRE<"fixr", 0xB367, null_frag, FP128, FP128>;
+
+
+//===----------------------------------------------------------------------===//
+// Binary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Addition.
+let Defs = [CC] in {
+  let isCommutable = 1 in {
+    def AER : BinaryRR<"aer", 0x3A, null_frag, FP32,  FP32>;
+    def ADR : BinaryRR<"adr", 0x2A, null_frag, FP64,  FP64>;
+    def AXR : BinaryRR<"axr", 0x36, null_frag, FP128, FP128>;
+  }
+  def AE : BinaryRX<"ae", 0x7A, null_frag, FP32, load, 4>;
+  def AD : BinaryRX<"ad", 0x6A, null_frag, FP64, load, 8>;
+}
+
+// Addition (unnormalized).
+let Defs = [CC] in {
+  let isCommutable = 1 in {
+    def AUR : BinaryRR<"aur", 0x3E, null_frag, FP32, FP32>;
+    def AWR : BinaryRR<"awr", 0x2E, null_frag, FP64, FP64>;
+  }
+  def AU : BinaryRX<"au", 0x7E, null_frag, FP32, load, 4>;
+  def AW : BinaryRX<"aw", 0x6E, null_frag, FP64, load, 8>;
+}
+
+// Subtraction.
+let Defs = [CC] in {
+  def SER : BinaryRR<"ser", 0x3B, null_frag, FP32,  FP32>;
+  def SDR : BinaryRR<"sdr", 0x2B, null_frag, FP64,  FP64>;
+  def SXR : BinaryRR<"sxr", 0x37, null_frag, FP128, FP128>;
+
+  def SE : BinaryRX<"se", 0x7B, null_frag, FP32, load, 4>;
+  def SD : BinaryRX<"sd", 0x6B, null_frag, FP64, load, 8>;
+}
+
+// Subtraction (unnormalized).
+let Defs = [CC] in {
+  def SUR : BinaryRR<"sur", 0x3F, null_frag, FP32, FP32>;
+  def SWR : BinaryRR<"swr", 0x2F, null_frag, FP64, FP64>;
+
+  def SU : BinaryRX<"su", 0x7F, null_frag, FP32, load, 4>;
+  def SW : BinaryRX<"sw", 0x6F, null_frag, FP64, load, 8>;
+}
+
+// Multiplication.
+let isCommutable = 1 in {
+  def MEER : BinaryRRE<"meer", 0xB337, null_frag, FP32,  FP32>;
+  def MDR  : BinaryRR <"mdr",  0x2C,   null_frag, FP64,  FP64>;
+  def MXR  : BinaryRR <"mxr",  0x26,   null_frag, FP128, FP128>;
+}
+def MEE : BinaryRXE<"mee", 0xED37, null_frag, FP32, load, 4>;
+def MD  : BinaryRX <"md",  0x6C,   null_frag, FP64, load, 8>;
+
+// Extending multiplication (f32 x f32 -> f64).
+def MDER : BinaryRR<"mder", 0x3C, null_frag, FP64, FP32>;
+def MDE  : BinaryRX<"mde",  0x7C, null_frag, FP64, load, 4>;
+let isAsmParserOnly = 1 in {
+  def MER : BinaryRR<"mer", 0x3C, null_frag, FP64, FP32>;
+  def ME  : BinaryRX<"me",  0x7C, null_frag, FP64, load, 4>;
+}
+
+// Extending multiplication (f64 x f64 -> f128).
+def MXDR : BinaryRR<"mxdr", 0x27, null_frag, FP128, FP64>;
+def MXD  : BinaryRX<"mxd",  0x67, null_frag, FP128, load, 8>;
+
+// Fused multiply-add.
+def MAER : TernaryRRD<"maer", 0xB32E, null_frag, FP32, FP32>;
+def MADR : TernaryRRD<"madr", 0xB33E, null_frag, FP64, FP64>;
+def MAE  : TernaryRXF<"mae",  0xED2E, null_frag, FP32, FP32, load, 4>;
+def MAD  : TernaryRXF<"mad",  0xED3E, null_frag, FP64, FP64, load, 8>;
+
+// Fused multiply-subtract.
+def MSER : TernaryRRD<"mser", 0xB32F, null_frag, FP32, FP32>;
+def MSDR : TernaryRRD<"msdr", 0xB33F, null_frag, FP64, FP64>;
+def MSE  : TernaryRXF<"mse",  0xED2F, null_frag, FP32, FP32, load, 4>;
+def MSD  : TernaryRXF<"msd",  0xED3F, null_frag, FP64, FP64, load, 8>;
+
+// Multiplication (unnormalized).
+def MYR  : BinaryRRD<"myr",  0xB33B, null_frag, FP128, FP64>;
+def MYHR : BinaryRRD<"myhr", 0xB33D, null_frag, FP64,  FP64>;
+def MYLR : BinaryRRD<"mylr", 0xB339, null_frag, FP64,  FP64>;
+def MY   : BinaryRXF<"my",   0xED3B, null_frag, FP128, FP64, load, 8>;
+def MYH  : BinaryRXF<"myh",  0xED3D, null_frag, FP64,  FP64, load, 8>;
+def MYL  : BinaryRXF<"myl",  0xED39, null_frag, FP64,  FP64, load, 8>;
+
+// Fused multiply-add (unnormalized).
+def MAYR  : TernaryRRD<"mayr",  0xB33A, null_frag, FP128, FP64>;
+def MAYHR : TernaryRRD<"mayhr", 0xB33C, null_frag, FP64,  FP64>;
+def MAYLR : TernaryRRD<"maylr", 0xB338, null_frag, FP64,  FP64>;
+def MAY   : TernaryRXF<"may",   0xED3A, null_frag, FP128, FP64, load, 8>;
+def MAYH  : TernaryRXF<"mayh",  0xED3C, null_frag, FP64,  FP64, load, 8>;
+def MAYL  : TernaryRXF<"mayl",  0xED38, null_frag, FP64,  FP64, load, 8>;
+
+// Division.
+def DER : BinaryRR <"der", 0x3D,   null_frag, FP32,  FP32>;
+def DDR : BinaryRR <"ddr", 0x2D,   null_frag, FP64,  FP64>;
+def DXR : BinaryRRE<"dxr", 0xB22D, null_frag, FP128, FP128>;
+def DE  : BinaryRX <"de",  0x7D,   null_frag, FP32, load, 4>;
+def DD  : BinaryRX <"dd",  0x6D,   null_frag, FP64, load, 8>;
+
+
+//===----------------------------------------------------------------------===//
+// Comparisons
+//===----------------------------------------------------------------------===//
+
+let Defs = [CC] in {
+  def CER : CompareRR <"cer", 0x39,   null_frag, FP32,  FP32>;
+  def CDR : CompareRR <"cdr", 0x29,   null_frag, FP64,  FP64>;
+  def CXR : CompareRRE<"cxr", 0xB369, null_frag, FP128, FP128>;
+
+  def CE : CompareRX<"ce", 0x79, null_frag, FP32, load, 4>;
+  def CD : CompareRX<"cd", 0x69, null_frag, FP64, load, 8>;
+}
+

lib/Target/SystemZ/SystemZScheduleZ13.td

@@ -908,6 +908,238 @@ def : InstRW<[FXa, Lat30, GroupAlone], (instregex "SFASR$")>;
 def : InstRW<[FXa, LSU, Lat30, GroupAlone], (instregex "LFAS$")>;
 def : InstRW<[FXb, Lat3, GroupAlone], (instregex "SRNM(B|T)?$")>;
 
+
+// --------------------- Hexadecimal floating point ------------------------- //
+
+//===----------------------------------------------------------------------===//
+// HFP: Move instructions
+//===----------------------------------------------------------------------===//
+
+// Load and Test
+def : InstRW<[VecXsPm, Lat4], (instregex "LT(D|E)R$")>;
+def : InstRW<[VecDF2, VecDF2, Lat11, GroupAlone], (instregex "LTXR$")>;
+
+//===----------------------------------------------------------------------===//
+// HFP: Conversion instructions
+//===----------------------------------------------------------------------===//
+
+// Load rounded
+def : InstRW<[VecBF], (instregex "(LEDR|LRER)$")>;
+def : InstRW<[VecBF], (instregex "LEXR$")>;
+def : InstRW<[VecDF2, VecDF2], (instregex "(LDXR|LRDR)$")>;
+
+// Load lengthened
+def : InstRW<[LSU], (instregex "LDE$")>;
+def : InstRW<[FXb], (instregex "LDER$")>;
+def : InstRW<[VecBF2, VecBF2, LSU, Lat12, GroupAlone], (instregex "LX(D|E)$")>;
+def : InstRW<[VecBF2, VecBF2, GroupAlone], (instregex "LX(D|E)R$")>;
+
+// Convert from fixed
+def : InstRW<[FXb, VecBF, Lat9, BeginGroup], (instregex "CE(F|G)R$")>;
+def : InstRW<[FXb, VecBF, Lat9, BeginGroup], (instregex "CD(F|G)R$")>;
+def : InstRW<[FXb, VecDF2, VecDF2, Lat12, GroupAlone], (instregex "CX(F|G)R$")>;
+
+// Convert to fixed
+def : InstRW<[FXb, VecBF, Lat11, BeginGroup], (instregex "CF(E|D)R$")>;
+def : InstRW<[FXb, VecBF, Lat11, BeginGroup], (instregex "CG(E|D)R$")>;
+def : InstRW<[FXb, VecDF, VecDF, Lat20, BeginGroup], (instregex "C(F|G)XR$")>;
+
+// Convert BFP to HFP / HFP to BFP.
+def : InstRW<[VecBF], (instregex "THD(E)?R$")>;
+def : InstRW<[VecBF], (instregex "TB(E)?DR$")>;
+
+//===----------------------------------------------------------------------===//
+// HFP: Unary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Load Complement / Negative / Positive
+def : InstRW<[VecXsPm, Lat4], (instregex "L(C|N|P)DR$")>;
+def : InstRW<[VecXsPm, Lat4], (instregex "L(C|N|P)ER$")>;
+def : InstRW<[VecDF2, VecDF2, Lat11, GroupAlone], (instregex "L(C|N|P)XR$")>;
+
+// Halve
+def : InstRW<[VecBF], (instregex "H(E|D)R$")>;
+
+// Square root
+def : InstRW<[VecFPd, LSU], (instregex "SQ(E|D)$")>;
+def : InstRW<[VecFPd], (instregex "SQ(E|D)R$")>;
+def : InstRW<[VecFPd, VecFPd, GroupAlone], (instregex "SQXR$")>;
+
+// Load FP integer
+def : InstRW<[VecBF], (instregex "FIER$")>;
+def : InstRW<[VecBF], (instregex "FIDR$")>;
+def : InstRW<[VecDF2, VecDF2, Lat11, GroupAlone], (instregex "FIXR$")>;
+
+//===----------------------------------------------------------------------===//
+// HFP: Binary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Addition
+def : InstRW<[VecBF, LSU, Lat12], (instregex "A(E|D|U|W)$")>;
+def : InstRW<[VecBF], (instregex "A(E|D|U|W)R$")>;
+def : InstRW<[VecDF2, VecDF2, Lat11, GroupAlone], (instregex "AXR$")>;
+
+// Subtraction
+def : InstRW<[VecBF, LSU, Lat12], (instregex "S(E|D|U|W)$")>;
+def : InstRW<[VecBF], (instregex "S(E|D|U|W)R$")>;
+def : InstRW<[VecDF2, VecDF2, Lat11, GroupAlone], (instregex "SXR$")>;
+
+// Multiply
+def : InstRW<[VecBF, LSU, Lat12], (instregex "M(D|DE|E|EE)$")>;
+def : InstRW<[VecBF], (instregex "M(D|DE|E|EE)R$")>;
+def : InstRW<[VecBF2, VecBF2, LSU, Lat12, GroupAlone], (instregex "MXD$")>;
+def : InstRW<[VecBF2, VecBF2, GroupAlone], (instregex "MXDR$")>;
+def : InstRW<[VecDF2, VecDF2, Lat20, GroupAlone], (instregex "MXR$")>;
+def : InstRW<[VecBF2, VecBF2, LSU, Lat12, GroupAlone], (instregex "MY(H|L)?$")>;
+def : InstRW<[VecBF2, VecBF2, GroupAlone], (instregex "MY(H|L)?R$")>;
+
+// Multiply and add / subtract
+def : InstRW<[VecBF, LSU, Lat12, GroupAlone], (instregex "M(A|S)E$")>;
+def : InstRW<[VecBF, GroupAlone], (instregex "M(A|S)ER$")>;
+def : InstRW<[VecBF, LSU, Lat12, GroupAlone], (instregex "M(A|S)D$")>;
+def : InstRW<[VecBF], (instregex "M(A|S)DR$")>;
+def : InstRW<[VecBF2, VecBF2, LSU, Lat12, GroupAlone], (instregex "MAY(H|L)?$")>;
+def : InstRW<[VecBF2, VecBF2, GroupAlone], (instregex "MAY(H|L)?R$")>;
+
+// Division
+def : InstRW<[VecFPd, LSU], (instregex "D(E|D)$")>;
+def : InstRW<[VecFPd], (instregex "D(E|D)R$")>;
+def : InstRW<[VecFPd, VecFPd, GroupAlone], (instregex "DXR$")>;
+
+//===----------------------------------------------------------------------===//
+// HFP: Comparisons
+//===----------------------------------------------------------------------===//
+
+// Compare
+def : InstRW<[VecXsPm, LSU, Lat8], (instregex "C(E|D)$")>;
+def : InstRW<[VecXsPm, Lat4], (instregex "C(E|D)R$")>;
+def : InstRW<[VecDF, VecDF, Lat20, GroupAlone], (instregex "CXR$")>;
+
+
+// ------------------------ Decimal floating point -------------------------- //
+
+//===----------------------------------------------------------------------===//
+// DFP: Move instructions
+//===----------------------------------------------------------------------===//
+
+// Load and Test
+def : InstRW<[VecDF], (instregex "LTDTR$")>;
+def : InstRW<[VecDF2, VecDF2, Lat11, GroupAlone], (instregex "LTXTR$")>;
+
+//===----------------------------------------------------------------------===//
+// DFP: Conversion instructions
+//===----------------------------------------------------------------------===//
+
+// Load rounded
+def : InstRW<[VecDF, Lat15], (instregex "LEDTR$")>;
+def : InstRW<[VecDF, VecDF, Lat20], (instregex "LDXTR$")>;
+
+// Load lengthened
+def : InstRW<[VecDF], (instregex "LDETR$")>;
+def : InstRW<[VecDF2, VecDF2, Lat11, GroupAlone], (instregex "LXDTR$")>;
+
+// Convert from fixed / logical
+def : InstRW<[FXb, VecDF, Lat30, BeginGroup], (instregex "CD(F|G)TR(A)?$")>;
+def : InstRW<[FXb, VecDF2, VecDF2, Lat30, GroupAlone], (instregex "CX(F|G)TR(A)?$")>;
+def : InstRW<[FXb, VecDF, Lat30, BeginGroup], (instregex "CDL(F|G)TR$")>;
+def : InstRW<[FXb, VecDF2, VecDF2, Lat30, GroupAlone], (instregex "CXL(F|G)TR$")>;
+
+// Convert to fixed / logical
+def : InstRW<[FXb, VecDF, Lat30, BeginGroup], (instregex "C(F|G)DTR(A)?$")>;
+def : InstRW<[FXb, VecDF, VecDF, Lat30, BeginGroup], (instregex "C(F|G)XTR(A)?$")>;
+def : InstRW<[FXb, VecDF, Lat30, BeginGroup], (instregex "CL(F|G)DTR$")>;
+def : InstRW<[FXb, VecDF, VecDF, Lat30, BeginGroup], (instregex "CL(F|G)XTR$")>;
+
+// Convert from / to signed / unsigned packed
+def : InstRW<[FXb, VecDF, Lat9, BeginGroup], (instregex "CD(S|U)TR$")>;
+def : InstRW<[FXb, FXb, VecDF2, VecDF2, Lat15, GroupAlone], (instregex "CX(S|U)TR$")>;
+def : InstRW<[FXb, VecDF, Lat12, BeginGroup], (instregex "C(S|U)DTR$")>;
+def : InstRW<[FXb, FXb, VecDF2, VecDF2, Lat15, BeginGroup], (instregex "C(S|U)XTR$")>;
+
+// Convert from / to zoned
+def : InstRW<[LSU, VecDF, Lat11, BeginGroup], (instregex "CDZT$")>;
+def : InstRW<[LSU, LSU, VecDF2, VecDF2, Lat15, GroupAlone], (instregex "CXZT$")>;
+def : InstRW<[FXb, LSU, VecDF, Lat11, BeginGroup], (instregex "CZDT$")>;
+def : InstRW<[FXb, LSU, VecDF, VecDF, Lat15, GroupAlone], (instregex "CZXT$")>;
+
+// Convert from / to packed
+def : InstRW<[LSU, VecDF, Lat11, BeginGroup], (instregex "CDPT$")>;
+def : InstRW<[LSU, LSU, VecDF2, VecDF2, Lat15, GroupAlone], (instregex "CXPT$")>;
+def : InstRW<[FXb, LSU, VecDF, Lat11, BeginGroup], (instregex "CPDT$")>;
+def : InstRW<[FXb, LSU, VecDF, VecDF, Lat15, GroupAlone], (instregex "CPXT$")>;
+
+// Perform floating-point operation
+def : InstRW<[LSU, Lat30, GroupAlone], (instregex "PFPO$")>;
+
+//===----------------------------------------------------------------------===//
+// DFP: Unary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Load FP integer
+def : InstRW<[VecDF], (instregex "FIDTR$")>;
+def : InstRW<[VecDF2, VecDF2, Lat11, GroupAlone], (instregex "FIXTR$")>;
+
+// Extract biased exponent
+def : InstRW<[FXb, VecDF, Lat12, BeginGroup], (instregex "EEDTR$")>;
+def : InstRW<[FXb, VecDF, Lat12, BeginGroup], (instregex "EEXTR$")>;
+
+// Extract significance
+def : InstRW<[FXb, VecDF, Lat12, BeginGroup], (instregex "ESDTR$")>;
+def : InstRW<[FXb, VecDF, VecDF, Lat15, BeginGroup], (instregex "ESXTR$")>;
+
+//===----------------------------------------------------------------------===//
+// DFP: Binary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Addition
+def : InstRW<[VecDF], (instregex "ADTR(A)?$")>;
+def : InstRW<[VecDF2, VecDF2, Lat11, GroupAlone], (instregex "AXTR(A)?$")>;
+
+// Subtraction
+def : InstRW<[VecDF], (instregex "SDTR(A)?$")>;
+def : InstRW<[VecDF2, VecDF2, Lat11, GroupAlone], (instregex "SXTR(A)?$")>;
+
+// Multiply
+def : InstRW<[VecDF, Lat30], (instregex "MDTR(A)?$")>;
+def : InstRW<[VecDF2, VecDF2, Lat30, GroupAlone], (instregex "MXTR(A)?$")>;
+
+// Division
+def : InstRW<[VecDF, Lat30], (instregex "DDTR(A)?$")>;
+def : InstRW<[VecDF2, VecDF2, Lat30, GroupAlone], (instregex "DXTR(A)?$")>;
+
+// Quantize
+def : InstRW<[VecDF], (instregex "QADTR$")>;
+def : InstRW<[VecDF2, VecDF2, Lat11, GroupAlone], (instregex "QAXTR$")>;
+
+// Reround
+def : InstRW<[FXb, VecDF, Lat11], (instregex "RRDTR$")>;
+def : InstRW<[FXb, VecDF2, VecDF2, Lat15, GroupAlone], (instregex "RRXTR$")>;
+
+// Shift significand left/right
+def : InstRW<[LSU, VecDF, Lat11], (instregex "S(L|R)DT$")>;
+def : InstRW<[LSU, VecDF2, VecDF2, Lat15, GroupAlone], (instregex "S(L|R)XT$")>;
+
+// Insert biased exponent
+def : InstRW<[FXb, VecDF, Lat11], (instregex "IEDTR$")>;
+def : InstRW<[FXb, VecDF2, VecDF2, Lat15, GroupAlone], (instregex "IEXTR$")>;
+
+//===----------------------------------------------------------------------===//
+// DFP: Comparisons
+//===----------------------------------------------------------------------===//
+
+// Compare
+def : InstRW<[VecDF], (instregex "(K|C)DTR$")>;
+def : InstRW<[VecDF, VecDF, Lat11, GroupAlone], (instregex "(K|C)XTR$")>;
+
+// Compare biased exponent
+def : InstRW<[VecDF], (instregex "CEDTR$")>;
+def : InstRW<[VecDF], (instregex "CEXTR$")>;
+
+// Test Data Class/Group
+def : InstRW<[LSU, VecDF, Lat11], (instregex "TD(C|G)(E|D)T$")>;
+def : InstRW<[LSU, VecDF2, VecDF2, Lat15, GroupAlone], (instregex "TD(C|G)XT$")>;
+
+
 // --------------------------------- Vector --------------------------------- //
 
 //===----------------------------------------------------------------------===//

lib/Target/SystemZ/SystemZScheduleZ196.td

@@ -839,5 +839,224 @@ def : InstRW<[FXU, Lat30, GroupAlone], (instregex "SFASR$")>;
 def : InstRW<[FXU, LSU, Lat30, GroupAlone], (instregex "LFAS$")>;
 def : InstRW<[FXU, Lat2, GroupAlone], (instregex "SRNM(B|T)?$")>;
 
+
+// --------------------- Hexadecimal floating point ------------------------- //
+
+//===----------------------------------------------------------------------===//
+// HFP: Move instructions
+//===----------------------------------------------------------------------===//
+
+// Load and Test
+def : InstRW<[FPU], (instregex "LT(D|E)R$")>;
+def : InstRW<[FPU2, FPU2, Lat9, GroupAlone], (instregex "LTXR$")>;
+
+//===----------------------------------------------------------------------===//
+// HFP: Conversion instructions
+//===----------------------------------------------------------------------===//
+
+// Load rounded
+def : InstRW<[FPU], (instregex "(LEDR|LRER)$")>;
+def : InstRW<[FPU], (instregex "LEXR$")>;
+def : InstRW<[FPU], (instregex "(LDXR|LRDR)$")>;
+
+// Load lengthened
+def : InstRW<[LSU], (instregex "LDE$")>;
+def : InstRW<[FXU], (instregex "LDER$")>;
+def : InstRW<[FPU2, FPU2, LSU, Lat15, GroupAlone], (instregex "LX(D|E)$")>;
+def : InstRW<[FPU2, FPU2, Lat10, GroupAlone], (instregex "LX(D|E)R$")>;
+
+// Convert from fixed
+def : InstRW<[FXU, FPU, Lat9, GroupAlone], (instregex "CE(F|G)R$")>;
+def : InstRW<[FXU, FPU, Lat9, GroupAlone], (instregex "CD(F|G)R$")>;
+def : InstRW<[FXU, FPU2, FPU2, Lat11, GroupAlone], (instregex "CX(F|G)R$")>;
+
+// Convert to fixed
+def : InstRW<[FXU, FPU, Lat12, GroupAlone], (instregex "CF(E|D)R$")>;
+def : InstRW<[FXU, FPU, Lat12, GroupAlone], (instregex "CG(E|D)R$")>;
+def : InstRW<[FXU, FPU, FPU, Lat20, GroupAlone], (instregex "C(F|G)XR$")>;
+
+// Convert BFP to HFP / HFP to BFP.
+def : InstRW<[FPU], (instregex "THD(E)?R$")>;
+def : InstRW<[FPU], (instregex "TB(E)?DR$")>;
+
+//===----------------------------------------------------------------------===//
+// HFP: Unary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Load Complement / Negative / Positive
+def : InstRW<[FPU], (instregex "L(C|N|P)DR$")>;
+def : InstRW<[FPU], (instregex "L(C|N|P)ER$")>;
+def : InstRW<[FPU2, FPU2, Lat9, GroupAlone], (instregex "L(C|N|P)XR$")>;
+
+// Halve
+def : InstRW<[FPU], (instregex "H(E|D)R$")>;
+
+// Square root
+def : InstRW<[FPU, LSU, Lat30], (instregex "SQ(E|D)$")>;
+def : InstRW<[FPU, Lat30], (instregex "SQ(E|D)R$")>;
+def : InstRW<[FPU2, FPU2, Lat30, GroupAlone], (instregex "SQXR$")>;
+
+// Load FP integer
+def : InstRW<[FPU], (instregex "FIER$")>;
+def : InstRW<[FPU], (instregex "FIDR$")>;
+def : InstRW<[FPU2, FPU2, Lat15, GroupAlone], (instregex "FIXR$")>;
+
+//===----------------------------------------------------------------------===//
+// HFP: Binary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Addition
+def : InstRW<[FPU, LSU, Lat12], (instregex "A(E|D|U|W)$")>;
+def : InstRW<[FPU], (instregex "A(E|D|U|W)R$")>;
+def : InstRW<[FPU2, FPU2, Lat20, GroupAlone], (instregex "AXR$")>;
+
+// Subtraction
+def : InstRW<[FPU, LSU, Lat12], (instregex "S(E|D|U|W)$")>;
+def : InstRW<[FPU], (instregex "S(E|D|U|W)R$")>;
+def : InstRW<[FPU2, FPU2, Lat20, GroupAlone], (instregex "SXR$")>;
+
+// Multiply
+def : InstRW<[FPU, LSU, Lat12], (instregex "M(D|DE|E|EE)$")>;
+def : InstRW<[FPU], (instregex "M(D|DE|E|EE)R$")>;
+def : InstRW<[FPU2, FPU2, LSU, Lat15, GroupAlone], (instregex "MXD$")>;
+def : InstRW<[FPU2, FPU2, Lat10, GroupAlone], (instregex "MXDR$")>;
+def : InstRW<[FPU2, FPU2, Lat30, GroupAlone], (instregex "MXR$")>;
+def : InstRW<[FPU2, FPU2, LSU, Lat15, GroupAlone], (instregex "MY(H|L)?$")>;
+def : InstRW<[FPU2, FPU2, Lat10, GroupAlone], (instregex "MY(H|L)?R$")>;
+
+// Multiply and add / subtract
+def : InstRW<[FPU, LSU, Lat12, GroupAlone], (instregex "M(A|S)E$")>;
+def : InstRW<[FPU, GroupAlone], (instregex "M(A|S)ER$")>;
+def : InstRW<[FPU, LSU, Lat12, GroupAlone], (instregex "M(A|S)D$")>;
+def : InstRW<[FPU, GroupAlone], (instregex "M(A|S)DR$")>;
+def : InstRW<[FPU2, FPU2, LSU, Lat12, GroupAlone], (instregex "MAY(H|L)?$")>;
+def : InstRW<[FPU2, FPU2, GroupAlone], (instregex "MAY(H|L)?R$")>;
+
+// Division
+def : InstRW<[FPU, LSU, Lat30], (instregex "D(E|D)$")>;
+def : InstRW<[FPU, Lat30], (instregex "D(E|D)R$")>;
+def : InstRW<[FPU2, FPU2, Lat30, GroupAlone], (instregex "DXR$")>;
+
+//===----------------------------------------------------------------------===//
+// HFP: Comparisons
+//===----------------------------------------------------------------------===//
+
+// Compare
+def : InstRW<[FPU, LSU, Lat12], (instregex "C(E|D)$")>;
+def : InstRW<[FPU], (instregex "C(E|D)R$")>;
+def : InstRW<[FPU, FPU, Lat15], (instregex "CXR$")>;
+
+
+// ------------------------ Decimal floating point -------------------------- //
+
+//===----------------------------------------------------------------------===//
+// DFP: Move instructions
+//===----------------------------------------------------------------------===//
+
+// Load and Test
+def : InstRW<[DFU, Lat20], (instregex "LTDTR$")>;
+def : InstRW<[DFU2, DFU2, Lat20, GroupAlone], (instregex "LTXTR$")>;
+
+//===----------------------------------------------------------------------===//
+// DFP: Conversion instructions
+//===----------------------------------------------------------------------===//
+
+// Load rounded
+def : InstRW<[DFU, Lat30], (instregex "LEDTR$")>;
+def : InstRW<[DFU, DFU, Lat30], (instregex "LDXTR$")>;
+
+// Load lengthened
+def : InstRW<[DFU, Lat20], (instregex "LDETR$")>;
+def : InstRW<[DFU2, DFU2, Lat20, GroupAlone], (instregex "LXDTR$")>;
+
+// Convert from fixed / logical
+def : InstRW<[FXU, DFU, Lat30, GroupAlone], (instregex "CD(F|G)TR(A)?$")>;
+def : InstRW<[FXU, DFU2, DFU2, Lat30, GroupAlone], (instregex "CX(F|G)TR(A)?$")>;
+def : InstRW<[FXU, DFU, Lat11, GroupAlone], (instregex "CDL(F|G)TR$")>;
+def : InstRW<[FXU, DFU2, DFU2, Lat11, GroupAlone], (instregex "CXL(F|G)TR$")>;
+
+// Convert to fixed / logical
+def : InstRW<[FXU, DFU, Lat30, GroupAlone], (instregex "C(F|G)DTR(A)?$")>;
+def : InstRW<[FXU, DFU, DFU, Lat30, GroupAlone], (instregex "C(F|G)XTR(A)?$")>;
+def : InstRW<[FXU, DFU, Lat30, GroupAlone], (instregex "CL(F|G)DTR$")>;
+def : InstRW<[FXU, DFU, DFU, Lat30, GroupAlone], (instregex "CL(F|G)XTR$")>;
+
+// Convert from / to signed / unsigned packed
+def : InstRW<[FXU, DFU, Lat12, GroupAlone], (instregex "CD(S|U)TR$")>;
+def : InstRW<[FXU, FXU, DFU2, DFU2, Lat20, GroupAlone], (instregex "CX(S|U)TR$")>;
+def : InstRW<[FXU, DFU, Lat12, GroupAlone], (instregex "C(S|U)DTR$")>;
+def : InstRW<[FXU, FXU, DFU2, DFU2, Lat20, GroupAlone], (instregex "C(S|U)XTR$")>;
+
+// Perform floating-point operation
+def : InstRW<[LSU, Lat30, GroupAlone], (instregex "PFPO$")>;
+
+//===----------------------------------------------------------------------===//
+// DFP: Unary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Load FP integer
+def : InstRW<[DFU, Lat20], (instregex "FIDTR$")>;
+def : InstRW<[DFU2, DFU2, Lat20, GroupAlone], (instregex "FIXTR$")>;
+
+// Extract biased exponent
+def : InstRW<[FXU, DFU, Lat15, GroupAlone], (instregex "EEDTR$")>;
+def : InstRW<[FXU, DFU, Lat15, GroupAlone], (instregex "EEXTR$")>;
+
+// Extract significance
+def : InstRW<[FXU, DFU, Lat15, GroupAlone], (instregex "ESDTR$")>;
+def : InstRW<[FXU, DFU, DFU, Lat20, GroupAlone], (instregex "ESXTR$")>;
+
+//===----------------------------------------------------------------------===//
+// DFP: Binary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Addition
+def : InstRW<[DFU, Lat30], (instregex "ADTR(A)?$")>;
+def : InstRW<[DFU2, DFU2, Lat30, GroupAlone], (instregex "AXTR(A)?$")>;
+
+// Subtraction
+def : InstRW<[DFU, Lat30], (instregex "SDTR(A)?$")>;
+def : InstRW<[DFU2, DFU2, Lat30, GroupAlone], (instregex "SXTR(A)?$")>;
+
+// Multiply
+def : InstRW<[DFU, Lat30], (instregex "MDTR(A)?$")>;
+def : InstRW<[DFU2, DFU2, Lat30, GroupAlone], (instregex "MXTR(A)?$")>;
+
+// Division
+def : InstRW<[DFU, Lat30], (instregex "DDTR(A)?$")>;
+def : InstRW<[DFU2, DFU2, Lat30, GroupAlone], (instregex "DXTR(A)?$")>;
+
+// Quantize
+def : InstRW<[DFU, Lat30], (instregex "QADTR$")>;
+def : InstRW<[DFU2, DFU2, Lat30, GroupAlone], (instregex "QAXTR$")>;
+
+// Reround
+def : InstRW<[FXU, DFU, Lat30], (instregex "RRDTR$")>;
+def : InstRW<[FXU, DFU2, DFU2, Lat30, GroupAlone], (instregex "RRXTR$")>;
+
+// Shift significand left/right
+def : InstRW<[LSU, DFU, Lat11], (instregex "S(L|R)DT$")>;
+def : InstRW<[LSU, DFU2, DFU2, Lat15, GroupAlone], (instregex "S(L|R)XT$")>;
+
+// Insert biased exponent
+def : InstRW<[FXU, DFU, Lat11], (instregex "IEDTR$")>;
+def : InstRW<[FXU, DFU2, DFU2, Lat15, GroupAlone], (instregex "IEXTR$")>;
+
+//===----------------------------------------------------------------------===//
+// DFP: Comparisons
+//===----------------------------------------------------------------------===//
+
+// Compare
+def : InstRW<[DFU, Lat11], (instregex "(K|C)DTR$")>;
+def : InstRW<[DFU, DFU, Lat15, GroupAlone], (instregex "(K|C)XTR$")>;
+
+// Compare biased exponent
+def : InstRW<[DFU, Lat8], (instregex "CEDTR$")>;
+def : InstRW<[DFU, Lat9], (instregex "CEXTR$")>;
+
+// Test Data Class/Group
+def : InstRW<[LSU, DFU, Lat15], (instregex "TD(C|G)(E|D)T$")>;
+def : InstRW<[LSU, DFU2, DFU2, Lat15, GroupAlone], (instregex "TD(C|G)XT$")>;
+
 }
 

lib/Target/SystemZ/SystemZScheduleZEC12.td

@@ -877,5 +877,230 @@ def : InstRW<[FXU, Lat30, GroupAlone], (instregex "SFASR$")>;
 def : InstRW<[FXU, LSU, Lat30, GroupAlone], (instregex "LFAS$")>;
 def : InstRW<[FXU, Lat2, GroupAlone], (instregex "SRNM(B|T)?$")>;
 
+
+// --------------------- Hexadecimal floating point ------------------------- //
+
+//===----------------------------------------------------------------------===//
+// HFP: Move instructions
+//===----------------------------------------------------------------------===//
+
+// Load and Test
+def : InstRW<[FPU], (instregex "LT(D|E)R$")>;
+def : InstRW<[FPU2, FPU2, Lat9, GroupAlone], (instregex "LTXR$")>;
+
+//===----------------------------------------------------------------------===//
+// HFP: Conversion instructions
+//===----------------------------------------------------------------------===//
+
+// Load rounded
+def : InstRW<[FPU], (instregex "(LEDR|LRER)$")>;
+def : InstRW<[FPU], (instregex "LEXR$")>;
+def : InstRW<[FPU], (instregex "(LDXR|LRDR)$")>;
+
+// Load lengthened
+def : InstRW<[LSU], (instregex "LDE$")>;
+def : InstRW<[FXU], (instregex "LDER$")>;
+def : InstRW<[FPU2, FPU2, LSU, Lat15, GroupAlone], (instregex "LX(D|E)$")>;
+def : InstRW<[FPU2, FPU2, Lat10, GroupAlone], (instregex "LX(D|E)R$")>;
+
+// Convert from fixed
+def : InstRW<[FXU, FPU, Lat9, GroupAlone], (instregex "CE(F|G)R$")>;
+def : InstRW<[FXU, FPU, Lat9, GroupAlone], (instregex "CD(F|G)R$")>;
+def : InstRW<[FXU, FPU2, FPU2, Lat11, GroupAlone], (instregex "CX(F|G)R$")>;
+
+// Convert to fixed
+def : InstRW<[FXU, FPU, Lat12, GroupAlone], (instregex "CF(E|D)R$")>;
+def : InstRW<[FXU, FPU, Lat12, GroupAlone], (instregex "CG(E|D)R$")>;
+def : InstRW<[FXU, FPU, FPU, Lat20, GroupAlone], (instregex "C(F|G)XR$")>;
+
+// Convert BFP to HFP / HFP to BFP.
+def : InstRW<[FPU], (instregex "THD(E)?R$")>;
+def : InstRW<[FPU], (instregex "TB(E)?DR$")>;
+
+//===----------------------------------------------------------------------===//
+// HFP: Unary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Load Complement / Negative / Positive
+def : InstRW<[FPU], (instregex "L(C|N|P)DR$")>;
+def : InstRW<[FPU], (instregex "L(C|N|P)ER$")>;
+def : InstRW<[FPU2, FPU2, Lat9, GroupAlone], (instregex "L(C|N|P)XR$")>;
+
+// Halve
+def : InstRW<[FPU], (instregex "H(E|D)R$")>;
+
+// Square root
+def : InstRW<[FPU, LSU, Lat30], (instregex "SQ(E|D)$")>;
+def : InstRW<[FPU, Lat30], (instregex "SQ(E|D)R$")>;
+def : InstRW<[FPU2, FPU2, Lat30, GroupAlone], (instregex "SQXR$")>;
+
+// Load FP integer
+def : InstRW<[FPU], (instregex "FIER$")>;
+def : InstRW<[FPU], (instregex "FIDR$")>;
+def : InstRW<[FPU2, FPU2, Lat15, GroupAlone], (instregex "FIXR$")>;
+
+//===----------------------------------------------------------------------===//
+// HFP: Binary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Addition
+def : InstRW<[FPU, LSU, Lat12], (instregex "A(E|D|U|W)$")>;
+def : InstRW<[FPU], (instregex "A(E|D|U|W)R$")>;
+def : InstRW<[FPU2, FPU2, Lat20, GroupAlone], (instregex "AXR$")>;
+
+// Subtraction
+def : InstRW<[FPU, LSU, Lat12], (instregex "S(E|D|U|W)$")>;
+def : InstRW<[FPU], (instregex "S(E|D|U|W)R$")>;
+def : InstRW<[FPU2, FPU2, Lat20, GroupAlone], (instregex "SXR$")>;
+
+// Multiply
+def : InstRW<[FPU, LSU, Lat12], (instregex "M(D|DE|E|EE)$")>;
+def : InstRW<[FPU], (instregex "M(D|DE|E|EE)R$")>;
+def : InstRW<[FPU2, FPU2, LSU, Lat15, GroupAlone], (instregex "MXD$")>;
+def : InstRW<[FPU2, FPU2, Lat10, GroupAlone], (instregex "MXDR$")>;
+def : InstRW<[FPU2, FPU2, Lat30, GroupAlone], (instregex "MXR$")>;
+def : InstRW<[FPU2, FPU2, LSU, Lat15, GroupAlone], (instregex "MY(H|L)?$")>;
+def : InstRW<[FPU2, FPU2, Lat10, GroupAlone], (instregex "MY(H|L)?R$")>;
+
+// Multiply and add / subtract
+def : InstRW<[FPU, LSU, Lat12, GroupAlone], (instregex "M(A|S)E$")>;
+def : InstRW<[FPU, GroupAlone], (instregex "M(A|S)ER$")>;
+def : InstRW<[FPU, LSU, Lat12, GroupAlone], (instregex "M(A|S)D$")>;
+def : InstRW<[FPU, GroupAlone], (instregex "M(A|S)DR$")>;
+def : InstRW<[FPU2, FPU2, LSU, Lat12, GroupAlone], (instregex "MAY(H|L)?$")>;
+def : InstRW<[FPU2, FPU2, GroupAlone], (instregex "MAY(H|L)?R$")>;
+
+// Division
+def : InstRW<[FPU, LSU, Lat30], (instregex "D(E|D)$")>;
+def : InstRW<[FPU, Lat30], (instregex "D(E|D)R$")>;
+def : InstRW<[FPU2, FPU2, Lat30, GroupAlone], (instregex "DXR$")>;
+
+//===----------------------------------------------------------------------===//
+// HFP: Comparisons
+//===----------------------------------------------------------------------===//
+
+// Compare
+def : InstRW<[FPU, LSU, Lat12], (instregex "C(E|D)$")>;
+def : InstRW<[FPU], (instregex "C(E|D)R$")>;
+def : InstRW<[FPU, FPU, Lat15], (instregex "CXR$")>;
+
+
+// ------------------------ Decimal floating point -------------------------- //
+
+//===----------------------------------------------------------------------===//
+// DFP: Move instructions
+//===----------------------------------------------------------------------===//
+
+// Load and Test
+def : InstRW<[DFU, Lat20], (instregex "LTDTR$")>;
+def : InstRW<[DFU2, DFU2, Lat20, GroupAlone], (instregex "LTXTR$")>;
+
+//===----------------------------------------------------------------------===//
+// DFP: Conversion instructions
+//===----------------------------------------------------------------------===//
+
+// Load rounded
+def : InstRW<[DFU, Lat30], (instregex "LEDTR$")>;
+def : InstRW<[DFU, DFU, Lat30], (instregex "LDXTR$")>;
+
+// Load lengthened
+def : InstRW<[DFU, Lat20], (instregex "LDETR$")>;
+def : InstRW<[DFU2, DFU2, Lat20, GroupAlone], (instregex "LXDTR$")>;
+
+// Convert from fixed / logical
+def : InstRW<[FXU, DFU, Lat30, GroupAlone], (instregex "CD(F|G)TR(A)?$")>;
+def : InstRW<[FXU, DFU2, DFU2, Lat30, GroupAlone], (instregex "CX(F|G)TR(A)?$")>;
+def : InstRW<[FXU, DFU, Lat11, GroupAlone], (instregex "CDL(F|G)TR$")>;
+def : InstRW<[FXU, DFU2, DFU2, Lat11, GroupAlone], (instregex "CXL(F|G)TR$")>;
+
+// Convert to fixed / logical
+def : InstRW<[FXU, DFU, Lat30, GroupAlone], (instregex "C(F|G)DTR(A)?$")>;
+def : InstRW<[FXU, DFU, DFU, Lat30, GroupAlone], (instregex "C(F|G)XTR(A)?$")>;
+def : InstRW<[FXU, DFU, Lat30, GroupAlone], (instregex "CL(F|G)DTR$")>;
+def : InstRW<[FXU, DFU, DFU, Lat30, GroupAlone], (instregex "CL(F|G)XTR$")>;
+
+// Convert from / to signed / unsigned packed
+def : InstRW<[FXU, DFU, Lat12, GroupAlone], (instregex "CD(S|U)TR$")>;
+def : InstRW<[FXU, FXU, DFU2, DFU2, Lat20, GroupAlone], (instregex "CX(S|U)TR$")>;
+def : InstRW<[FXU, DFU, Lat12, GroupAlone], (instregex "C(S|U)DTR$")>;
+def : InstRW<[FXU, FXU, DFU2, DFU2, Lat20, GroupAlone], (instregex "C(S|U)XTR$")>;
+
+// Convert from / to zoned
+def : InstRW<[LSU, DFU2, Lat7, GroupAlone], (instregex "CDZT$")>;
+def : InstRW<[LSU, LSU, DFU2, DFU2, Lat10, GroupAlone], (instregex "CXZT$")>;
+def : InstRW<[FXU, LSU, DFU, Lat11, GroupAlone], (instregex "CZDT$")>;
+def : InstRW<[FXU, LSU, DFU, DFU, Lat15, GroupAlone], (instregex "CZXT$")>;
+
+// Perform floating-point operation
+def : InstRW<[LSU, Lat30, GroupAlone], (instregex "PFPO$")>;
+
+//===----------------------------------------------------------------------===//
+// DFP: Unary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Load FP integer
+def : InstRW<[DFU, Lat20], (instregex "FIDTR$")>;
+def : InstRW<[DFU2, DFU2, Lat20, GroupAlone], (instregex "FIXTR$")>;
+
+// Extract biased exponent
+def : InstRW<[FXU, DFU, Lat15, GroupAlone], (instregex "EEDTR$")>;
+def : InstRW<[FXU, DFU, Lat15, GroupAlone], (instregex "EEXTR$")>;
+
+// Extract significance
+def : InstRW<[FXU, DFU, Lat15, GroupAlone], (instregex "ESDTR$")>;
+def : InstRW<[FXU, DFU, DFU, Lat20, GroupAlone], (instregex "ESXTR$")>;
+
+//===----------------------------------------------------------------------===//
+// DFP: Binary arithmetic
+//===----------------------------------------------------------------------===//
+
+// Addition
+def : InstRW<[DFU, Lat30], (instregex "ADTR(A)?$")>;
+def : InstRW<[DFU2, DFU2, Lat30, GroupAlone], (instregex "AXTR(A)?$")>;
+
+// Subtraction
+def : InstRW<[DFU, Lat30], (instregex "SDTR(A)?$")>;
+def : InstRW<[DFU2, DFU2, Lat30, GroupAlone], (instregex "SXTR(A)?$")>;
+
+// Multiply
+def : InstRW<[DFU, Lat30], (instregex "MDTR(A)?$")>;
+def : InstRW<[DFU2, DFU2, Lat30, GroupAlone], (instregex "MXTR(A)?$")>;
+
+// Division
+def : InstRW<[DFU, Lat30], (instregex "DDTR(A)?$")>;
+def : InstRW<[DFU2, DFU2, Lat30, GroupAlone], (instregex "DXTR(A)?$")>;
+
+// Quantize
+def : InstRW<[DFU, Lat30], (instregex "QADTR$")>;
+def : InstRW<[DFU2, DFU2, Lat30, GroupAlone], (instregex "QAXTR$")>;
+
+// Reround
+def : InstRW<[FXU, DFU, Lat30], (instregex "RRDTR$")>;
+def : InstRW<[FXU, DFU2, DFU2, Lat30, GroupAlone], (instregex "RRXTR$")>;
+
+// Shift significand left/right
+def : InstRW<[LSU, DFU, Lat11], (instregex "S(L|R)DT$")>;
+def : InstRW<[LSU, DFU2, DFU2, Lat15, GroupAlone], (instregex "S(L|R)XT$")>;
+
+// Insert biased exponent
+def : InstRW<[FXU, DFU, Lat11], (instregex "IEDTR$")>;
+def : InstRW<[FXU, DFU2, DFU2, Lat15, GroupAlone], (instregex "IEXTR$")>;
+
+//===----------------------------------------------------------------------===//
+// DFP: Comparisons
+//===----------------------------------------------------------------------===//
+
+// Compare
+def : InstRW<[DFU, Lat11], (instregex "(K|C)DTR$")>;
+def : InstRW<[DFU, DFU, Lat15, GroupAlone], (instregex "(K|C)XTR$")>;
+
+// Compare biased exponent
+def : InstRW<[DFU, Lat8], (instregex "CEDTR$")>;
+def : InstRW<[DFU, Lat9], (instregex "CEXTR$")>;
+
+// Test Data Class/Group
+def : InstRW<[LSU, DFU, Lat15], (instregex "TD(C|G)(E|D)T$")>;
+def : InstRW<[LSU, DFU2, DFU2, Lat15, GroupAlone], (instregex "TD(C|G)XT$")>;
+
 }
 

lib/Target/SystemZ/SystemZSubtarget.cpp

@@ -42,8 +42,10 @@ SystemZSubtarget::SystemZSubtarget(const Triple &TT, const std::string &CPU,
       HasMiscellaneousExtensions(false),
       HasExecutionHint(false), HasLoadAndTrap(false),
       HasTransactionalExecution(false), HasProcessorAssist(false),
+      HasDFPZonedConversion(false),
       HasVector(false), HasLoadStoreOnCond2(false),
       HasLoadAndZeroRightmostByte(false), HasMessageSecurityAssist5(false),
+      HasDFPPackedConversion(false),
       TargetTriple(TT), InstrInfo(initializeSubtargetDependencies(CPU, FS)),
       TLInfo(TM, *this), TSInfo(), FrameLowering() {}
 

lib/Target/SystemZ/SystemZSubtarget.h

@@ -47,10 +47,12 @@ class SystemZSubtarget : public SystemZGenSubtargetInfo {
   bool HasLoadAndTrap;
   bool HasTransactionalExecution;
   bool HasProcessorAssist;
+  bool HasDFPZonedConversion;
   bool HasVector;
   bool HasLoadStoreOnCond2;
   bool HasLoadAndZeroRightmostByte;
   bool HasMessageSecurityAssist5;
+  bool HasDFPPackedConversion;
 
 private:
   Triple TargetTriple;
@@ -133,6 +135,9 @@ class SystemZSubtarget : public SystemZGenSubtargetInfo {
   // Return true if the target has the processor-assist facility.
   bool hasProcessorAssist() const { return HasProcessorAssist; }
 
+  // Return true if the target has the DFP zoned-conversion facility.
+  bool hasDFPZonedConversion() const { return HasDFPZonedConversion; }
+
   // Return true if the target has the load-and-zero-rightmost-byte facility.
   bool hasLoadAndZeroRightmostByte() const {
     return HasLoadAndZeroRightmostByte;
@@ -142,6 +147,9 @@ class SystemZSubtarget : public SystemZGenSubtargetInfo {
   // extension facility 5.
   bool hasMessageSecurityAssist5() const { return HasMessageSecurityAssist5; }
 
+  // Return true if the target has the DFP packed-conversion facility.
+  bool hasDFPPackedConversion() const { return HasDFPPackedConversion; }
+
   // Return true if the target has the vector facility.
   bool hasVector() const { return HasVector; }
 

lib/Target/WebAssembly/WebAssemblyISelLowering.cpp

@@ -84,8 +84,8 @@ WebAssemblyTargetLowering::WebAssemblyTargetLowering(
                     ISD::SETULT, ISD::SETULE, ISD::SETUGT, ISD::SETUGE})
       setCondCodeAction(CC, T, Expand);
     // Expand floating-point library function operators.
-    for (auto Op : {ISD::FSIN, ISD::FCOS, ISD::FSINCOS, ISD::FPOWI, ISD::FPOW,
-                    ISD::FREM, ISD::FMA})
+    for (auto Op : {ISD::FSIN, ISD::FCOS, ISD::FSINCOS, ISD::FPOW, ISD::FREM,
+                    ISD::FMA})
       setOperationAction(Op, T, Expand);
     // Note supported floating-point library function operators that otherwise
     // default to expand.

lib/Target/X86/X86ISelLowering.cpp

@@ -80,6 +80,12 @@ static cl::opt<int> ExperimentalPrefLoopAlignment(
              " of the loop header PC will be 0)."),
     cl::Hidden);
 
+static cl::opt<bool> MulConstantOptimization(
+    "mul-constant-optimization", cl::init(true),
+    cl::desc("Replace 'mul x, Const' with more effective instructions like "
+             "SHIFT, LEA, etc."),
+    cl::Hidden);
+
 /// Call this when the user attempts to do something unsupported, like
 /// returning a double without SSE2 enabled on x86_64. This is not fatal, unlike
 /// report_fatal_error, so calling code should attempt to recover without
@@ -670,7 +676,6 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
     setOperationAction(ISD::FSINCOS,   VT, Expand);
     setOperationAction(ISD::FCOS,      VT, Expand);
     setOperationAction(ISD::FREM,      VT, Expand);
-    setOperationAction(ISD::FPOWI,     VT, Expand);
     setOperationAction(ISD::FCOPYSIGN, VT, Expand);
     setOperationAction(ISD::FPOW,      VT, Expand);
     setOperationAction(ISD::FLOG,      VT, Expand);
@@ -30928,6 +30933,75 @@ static SDValue reduceVMULWidth(SDNode *N, SelectionDAG &DAG,
   }
 }
 
+static SDValue combineMulSpecial(uint64_t MulAmt, SDNode *N, SelectionDAG &DAG,
+                                 EVT VT, SDLoc DL) {
+
+  auto combineMulShlAddOrSub = [&](int Mult, int Shift, bool isAdd) {
+    SDValue Result = DAG.getNode(X86ISD::MUL_IMM, DL, VT, N->getOperand(0),
+                                 DAG.getConstant(Mult, DL, VT));
+    Result = DAG.getNode(ISD::SHL, DL, VT, Result,
+                         DAG.getConstant(Shift, DL, MVT::i8));
+    Result = DAG.getNode(isAdd ? ISD::ADD : ISD::SUB, DL, VT, N->getOperand(0),
+                         Result);
+    return Result;
+  };
+
+  auto combineMulMulAddOrSub = [&](bool isAdd) {
+    SDValue Result = DAG.getNode(X86ISD::MUL_IMM, DL, VT, N->getOperand(0),
+                                 DAG.getConstant(9, DL, VT));
+    Result = DAG.getNode(ISD::MUL, DL, VT, Result, DAG.getConstant(3, DL, VT));
+    Result = DAG.getNode(isAdd ? ISD::ADD : ISD::SUB, DL, VT, N->getOperand(0),
+                         Result);
+    return Result;
+  };
+
+  switch (MulAmt) {
+  default:
+    break;
+  case 11:
+    // mul x, 11 => add ((shl (mul x, 5), 1), x)
+    return combineMulShlAddOrSub(5, 1, /*isAdd*/ true);
+  case 21:
+    // mul x, 21 => add ((shl (mul x, 5), 2), x)
+    return combineMulShlAddOrSub(5, 2, /*isAdd*/ true);
+  case 22:
+    // mul x, 22 => add (add ((shl (mul x, 5), 2), x), x)
+    return DAG.getNode(ISD::ADD, DL, VT, N->getOperand(0),
+                       combineMulShlAddOrSub(5, 2, /*isAdd*/ true));
+  case 19:
+    // mul x, 19 => sub ((shl (mul x, 5), 2), x)
+    return combineMulShlAddOrSub(5, 2, /*isAdd*/ false);
+  case 13:
+    // mul x, 13 => add ((shl (mul x, 3), 2), x)
+    return combineMulShlAddOrSub(3, 2, /*isAdd*/ true);
+  case 23:
+    // mul x, 13 => sub ((shl (mul x, 3), 3), x)
+    return combineMulShlAddOrSub(3, 3, /*isAdd*/ false);
+  case 14:
+    // mul x, 14 => add (add ((shl (mul x, 3), 2), x), x)
+    return DAG.getNode(ISD::ADD, DL, VT, N->getOperand(0),
+                       combineMulShlAddOrSub(3, 2, /*isAdd*/ true));
+  case 26:
+    // mul x, 26 => sub ((mul (mul x, 9), 3), x)
+    return combineMulMulAddOrSub(/*isAdd*/ false);
+  case 28:
+    // mul x, 28 => add ((mul (mul x, 9), 3), x)
+    return combineMulMulAddOrSub(/*isAdd*/ true);
+  case 29:
+    // mul x, 29 => add (add ((mul (mul x, 9), 3), x), x)
+    return DAG.getNode(ISD::ADD, DL, VT, N->getOperand(0),
+                       combineMulMulAddOrSub(/*isAdd*/ true));
+  case 30:
+    // mul x, 30 => sub (sub ((shl x, 5), x), x)
+    return DAG.getNode(
+        ISD::SUB, DL, VT, N->getOperand(0),
+        DAG.getNode(ISD::SUB, DL, VT, N->getOperand(0),
+                    DAG.getNode(ISD::SHL, DL, VT, N->getOperand(0),
+                                DAG.getConstant(5, DL, MVT::i8))));
+  }
+  return SDValue();
+}
+
 /// Optimize a single multiply with constant into two operations in order to
 /// implement it with two cheaper instructions, e.g. LEA + SHL, LEA + LEA.
 static SDValue combineMul(SDNode *N, SelectionDAG &DAG,
@@ -30937,6 +31011,8 @@ static SDValue combineMul(SDNode *N, SelectionDAG &DAG,
   if (DCI.isBeforeLegalize() && VT.isVector())
     return reduceVMULWidth(N, DAG, Subtarget);
 
+  if (!MulConstantOptimization)
+    return SDValue();
   // An imul is usually smaller than the alternative sequence.
   if (DAG.getMachineFunction().getFunction()->optForMinSize())
     return SDValue();
@@ -30992,7 +31068,8 @@ static SDValue combineMul(SDNode *N, SelectionDAG &DAG,
     else
       NewMul = DAG.getNode(X86ISD::MUL_IMM, DL, VT, NewMul,
                            DAG.getConstant(MulAmt2, DL, VT));
-  }
+  } else if (!Subtarget.slowLEA())
+    NewMul = combineMulSpecial(MulAmt, N, DAG, VT, DL);
 
   if (!NewMul) {
     assert(MulAmt != 0 &&

lib/Transforms/Scalar/NewGVN.cpp

@@ -377,7 +377,6 @@ class CongruenceClass {
   int StoreCount = 0;
 };
 
-struct HashedExpression;
 namespace llvm {
 template <> struct DenseMapInfo<const Expression *> {
   static const Expression *getEmptyKey() {
@@ -391,41 +390,25 @@ template <> struct DenseMapInfo<const Expression *> {
     return reinterpret_cast<const Expression *>(Val);
   }
   static unsigned getHashValue(const Expression *E) {
-    return static_cast<unsigned>(E->getHashValue());
+    return static_cast<unsigned>(E->getComputedHash());
   }
-  static unsigned getHashValue(const HashedExpression &HE);
-  static bool isEqual(const HashedExpression &LHS, const Expression *RHS);
   static bool isEqual(const Expression *LHS, const Expression *RHS) {
     if (LHS == RHS)
       return true;
     if (LHS == getTombstoneKey() || RHS == getTombstoneKey() ||
         LHS == getEmptyKey() || RHS == getEmptyKey())
       return false;
+    // Compare hashes before equality.  This is *not* what the hashtable does,
+    // since it is computing it modulo the number of buckets, whereas we are
+    // using the full hash keyspace.  Since the hashes are precomputed, this
+    // check is *much* faster than equality.
+    if (LHS->getComputedHash() != RHS->getComputedHash())
+      return false;
     return *LHS == *RHS;
   }
 };
 } // end namespace llvm
 
-// This is just a wrapper around Expression that computes the hash value once at
-// creation time.  Hash values for an Expression can't change once they are
-// inserted into the DenseMap (it breaks DenseMap), so they must be immutable at
-// that point anyway.
-struct HashedExpression {
-  const Expression *E;
-  unsigned HashVal;
-  HashedExpression(const Expression *E)
-      : E(E), HashVal(DenseMapInfo<const Expression *>::getHashValue(E)) {}
-};
-
-unsigned
-DenseMapInfo<const Expression *>::getHashValue(const HashedExpression &HE) {
-  return HE.HashVal;
-}
-bool DenseMapInfo<const Expression *>::isEqual(const HashedExpression &LHS,
-                                               const Expression *RHS) {
-  return isEqual(LHS.E, RHS);
-}
-
 namespace {
 class NewGVN {
   Function &F;
@@ -707,7 +690,7 @@ class NewGVN {
   void markPredicateUsersTouched(Instruction *);
   void markValueLeaderChangeTouched(CongruenceClass *CC);
   void markMemoryLeaderChangeTouched(CongruenceClass *CC);
-  void markPhiOfOpsChanged(const HashedExpression &HE);
+  void markPhiOfOpsChanged(const Expression *E);
   void addPredicateUsers(const PredicateBase *, Instruction *) const;
   void addMemoryUsers(const MemoryAccess *To, MemoryAccess *U) const;
   void addAdditionalUsers(Value *To, Value *User) const;
@@ -956,8 +939,12 @@ const Expression *NewGVN::checkSimplificationResults(Expression *E,
   if (CC && CC->getDefiningExpr()) {
     // If we simplified to something else, we need to communicate
     // that we're users of the value we simplified to.
-    if (I != V)
-      addAdditionalUsers(V, I);
+    if (I != V) {
+      // Don't add temporary instructions to the user lists.
+      if (!AllTempInstructions.count(I))
+        addAdditionalUsers(V, I);
+    }
+
     if (I)
       DEBUG(dbgs() << "Simplified " << *I << " to "
                    << " expression " << *CC->getDefiningExpr() << "\n");
@@ -2195,8 +2182,8 @@ void NewGVN::moveValueToNewCongruenceClass(Instruction *I, const Expression *E,
 
 // For a given expression, mark the phi of ops instructions that could have
 // changed as a result.
-void NewGVN::markPhiOfOpsChanged(const HashedExpression &HE) {
-  touchAndErase(ExpressionToPhiOfOps, HE);
+void NewGVN::markPhiOfOpsChanged(const Expression *E) {
+  touchAndErase(ExpressionToPhiOfOps, E);
 }
 
 // Perform congruence finding on a given value numbering expression.
@@ -2210,14 +2197,13 @@ void NewGVN::performCongruenceFinding(Instruction *I, const Expression *E) {
   assert(!IClass->isDead() && "Found a dead class");
 
   CongruenceClass *EClass = nullptr;
-  HashedExpression HE(E);
   if (const auto *VE = dyn_cast<VariableExpression>(E)) {
     EClass = ValueToClass.lookup(VE->getVariableValue());
   } else if (isa<DeadExpression>(E)) {
     EClass = TOPClass;
   }
   if (!EClass) {
-    auto lookupResult = ExpressionToClass.insert_as({E, nullptr}, HE);
+    auto lookupResult = ExpressionToClass.insert({E, nullptr});
 
     // If it's not in the value table, create a new congruence class.
     if (lookupResult.second) {
@@ -2268,7 +2254,7 @@ void NewGVN::performCongruenceFinding(Instruction *I, const Expression *E) {
                  << "\n");
     if (ClassChanged) {
       moveValueToNewCongruenceClass(I, E, IClass, EClass);
-      markPhiOfOpsChanged(HE);
+      markPhiOfOpsChanged(E);
     }
 
     markUsersTouched(I);
@@ -2502,9 +2488,8 @@ NewGVN::makePossiblePhiOfOps(Instruction *I, bool HasBackedge,
         // Clone the instruction, create an expression from it, and see if we
         // have a leader.
         Instruction *ValueOp = I->clone();
-        auto Iter = TempToMemory.end();
         if (MemAccess)
-          Iter = TempToMemory.insert({ValueOp, MemAccess}).first;
+          TempToMemory.insert({ValueOp, MemAccess});
 
         for (auto &Op : ValueOp->operands()) {
           Op = Op->DoPHITranslation(PHIBlock, PredBB);
@@ -2523,7 +2508,7 @@ NewGVN::makePossiblePhiOfOps(Instruction *I, bool HasBackedge,
         AllTempInstructions.erase(ValueOp);
         ValueOp->deleteValue();
         if (MemAccess)
-          TempToMemory.erase(Iter);
+          TempToMemory.erase(ValueOp);
         if (!E)
           return nullptr;
         FoundVal = findPhiOfOpsLeader(E, PredBB);

lib/Transforms/Vectorize/LoopVectorize.cpp

@@ -7173,7 +7173,7 @@ LoopVectorizationCostModel::getInstructionCost(Instruction *I, unsigned VF) {
   // Note: Even if all instructions are scalarized, return true if any memory
   // accesses appear in the loop to get benefits from address folding etc.
   bool TypeNotScalarized =
-      VF > 1 && VectorTy->isVectorTy() && TTI.getNumberOfParts(VectorTy) < VF;
+      VF > 1 && !VectorTy->isVoidTy() && TTI.getNumberOfParts(VectorTy) < VF;
   return VectorizationCostTy(C, TypeNotScalarized);
 }
 
@@ -7312,7 +7312,7 @@ unsigned LoopVectorizationCostModel::getInstructionCost(Instruction *I,
   Type *RetTy = I->getType();
   if (canTruncateToMinimalBitwidth(I, VF))
     RetTy = IntegerType::get(RetTy->getContext(), MinBWs[I]);
-  VectorTy = isScalarAfterVectorization(I, VF) ? RetTy : ToVectorTy(RetTy, VF);
+  VectorTy = ToVectorTy(RetTy, VF);
   auto SE = PSE.getSE();
 
   // TODO: We need to estimate the cost of intrinsic calls.
@@ -7445,10 +7445,9 @@ unsigned LoopVectorizationCostModel::getInstructionCost(Instruction *I,
     } else if (Legal->isUniform(Op2)) {
       Op2VK = TargetTransformInfo::OK_UniformValue;
     }
-    SmallVector<const Value *, 4> Operands(I->operand_values());
-    unsigned N = isScalarAfterVectorization(I, VF) ? VF : 1;
-    return N * TTI.getArithmeticInstrCost(I->getOpcode(), VectorTy, Op1VK,
-                                          Op2VK, Op1VP, Op2VP, Operands);
+    SmallVector<const Value *, 4> Operands(I->operand_values()); 
+    return TTI.getArithmeticInstrCost(I->getOpcode(), VectorTy, Op1VK,
+                                      Op2VK, Op1VP, Op2VP, Operands);
   }
   case Instruction::Select: {
     SelectInst *SI = cast<SelectInst>(I);
@@ -7471,15 +7470,7 @@ unsigned LoopVectorizationCostModel::getInstructionCost(Instruction *I,
   }
   case Instruction::Store:
   case Instruction::Load: {
-    unsigned Width = VF;
-    if (Width > 1) {
-      InstWidening Decision = getWideningDecision(I, Width);
-      assert(Decision != CM_Unknown &&
-             "CM decision should be taken at this point");
-      if (Decision == CM_Scalarize)
-        Width = 1;
-    }
-    VectorTy = ToVectorTy(getMemInstValueType(I), Width);
+    VectorTy = ToVectorTy(getMemInstValueType(I), VF);
     return getMemoryInstructionCost(I, VF);
   }
   case Instruction::ZExt:

test/CodeGen/AArch64/reg-scavenge-frame.mir

@@ -0,0 +1,52 @@
+# RUN: llc -run-pass=prologepilog -verify-machineinstrs %s -o - | FileCheck %s
+
+--- |
+  target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+  target triple = "aarch64-linux-gnu"
+  define void @ScavengeForFrameWithoutOffset() { ret void }
+...
+---
+name:            ScavengeForFrameWithoutOffset
+tracksRegLiveness: true
+stack:
+  - { id: 0, type: spill-slot, offset: 0, size: 32, alignment: 8 }
+body:             |
+  bb.0:
+    liveins: %d16_d17_d18_d19
+    %x0 = COPY %xzr
+    %x1 = COPY %xzr
+    %x2 = COPY %xzr
+    %x3 = COPY %xzr
+    %x4 = COPY %xzr
+    %x5 = COPY %xzr
+    %x6 = COPY %xzr
+    %x7 = COPY %xzr
+    %x8 = COPY %xzr
+    %x9 = COPY %xzr
+    %x10 = COPY %xzr
+    %x11 = COPY %xzr
+    %x12 = COPY %xzr
+    %x13 = COPY %xzr
+    %x14 = COPY %xzr
+    %x15 = COPY %xzr
+    %x16 = COPY %xzr
+    %x17 = COPY %xzr
+    %x18 = COPY %xzr
+    %x19 = COPY %xzr
+    %x20 = COPY %xzr
+    %x21 = COPY %xzr
+    %x22 = COPY %xzr
+    %x23 = COPY %xzr
+    %x24 = COPY %xzr
+    %x25 = COPY %xzr
+    %x26 = COPY %xzr
+    %x27 = COPY %xzr
+    %x28 = COPY %xzr
+    %fp = COPY %xzr
+    %lr = COPY %xzr
+    ST1Fourv1d killed %d16_d17_d18_d19, %stack.0 :: (store 32 into %stack.0, align 8)
+# CHECK:  STRXui killed %[[SCAVREG:x[0-9]+|fp|lr]], %sp, [[SPOFFSET:[0-9]+]] :: (store 8 into %stack.1)
+# CHECK-NEXT:  %[[SCAVREG]] = ADDXri %sp, {{[0-9]+}}, 0
+# CHECK-NEXT:  ST1Fourv1d killed %d16_d17_d18_d19, killed %[[SCAVREG]] :: (store 32 into %stack.0, align 8)
+# CHECK-NEXT:  %[[SCAVREG]] = LDRXui %sp, [[SPOFFSET]] :: (load 8 from %stack.1)
+...

test/CodeGen/AMDGPU/add.v2i16.ll

@@ -23,7 +23,7 @@ define amdgpu_kernel void @v_test_add_v2i16(<2 x i16> addrspace(1)* %out, <2 x i
 ; GFX9: s_load_dword [[VAL0:s[0-9]+]]
 ; GFX9: s_load_dword [[VAL1:s[0-9]+]]
 ; GFX9: v_mov_b32_e32 [[VVAL1:v[0-9]+]]
-; GFX9: v_pk_add_u16 v{{[0-9]+}}, [[VVAL1]], [[VAL0]]
+; GFX9: v_pk_add_u16 v{{[0-9]+}}, [[VAL0]], [[VVAL1]]
 
 ; VI: s_add_i32
 ; VI: s_add_i32
@@ -50,7 +50,7 @@ define amdgpu_kernel void @s_test_add_self_v2i16(<2 x i16> addrspace(1)* %out, <
 
 ; FIXME: VI should not scalarize arg access.
 ; GCN-LABEL: {{^}}s_test_add_v2i16_kernarg:
-; GFX9: v_pk_add_u16 v{{[0-9]+}}, v{{[0-9]+}}, s{{[0-9]+}}
+; GFX9: v_pk_add_u16 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}}
 
 ; VI: v_add_i32
 ; VI: v_add_i32_sdwa
@@ -62,10 +62,11 @@ define amdgpu_kernel void @s_test_add_v2i16_kernarg(<2 x i16> addrspace(1)* %out
 
 ; GCN-LABEL: {{^}}v_test_add_v2i16_constant:
 ; GFX9: s_mov_b32 [[CONST:s[0-9]+]], 0x1c8007b{{$}}
-; GFX9: v_pk_add_u16 v{{[0-9]+}}, [[CONST]], v{{[0-9]+}}
+; GFX9: v_pk_add_u16 v{{[0-9]+}}, v{{[0-9]+}}, [[CONST]]
 
 ; VI-DAG: v_add_u16_e32 v{{[0-9]+}}, 0x7b, v{{[0-9]+}}
-; VI-DAG: v_add_u16_e32 v{{[0-9]+}}, 0x1c8, v{{[0-9]+}}
+; VI-DAG: v_mov_b32_e32 v[[SCONST:[0-9]+]], 0x1c8
+; VI-DAG: v_add_u16_sdwa v{{[0-9]+}}, v[[SCONST]], v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
 define amdgpu_kernel void @v_test_add_v2i16_constant(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in0) #1 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.out = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i32 %tid
@@ -79,10 +80,11 @@ define amdgpu_kernel void @v_test_add_v2i16_constant(<2 x i16> addrspace(1)* %ou
 ; FIXME: Need to handle non-uniform case for function below (load without gep).
 ; GCN-LABEL: {{^}}v_test_add_v2i16_neg_constant:
 ; GFX9: s_mov_b32 [[CONST:s[0-9]+]], 0xfc21fcb3{{$}}
-; GFX9: v_pk_add_u16 v{{[0-9]+}}, [[CONST]], v{{[0-9]+}}
+; GFX9: v_pk_add_u16 v{{[0-9]+}}, v{{[0-9]+}}, [[CONST]]
 
 ; VI-DAG: v_add_u16_e32 v{{[0-9]+}}, 0xfffffcb3, v{{[0-9]+}}
-; VI-DAG: v_add_u16_e32 v{{[0-9]+}}, 0xfffffc21, v{{[0-9]+}}
+; VI-DAG: v_mov_b32_e32 v[[SCONST:[0-9]+]], 0xfffffc21
+; VI-DAG: v_add_u16_sdwa v{{[0-9]+}}, v[[SCONST]], v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
 define amdgpu_kernel void @v_test_add_v2i16_neg_constant(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in0) #1 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
   %gep.out = getelementptr inbounds <2 x i16>, <2 x i16> addrspace(1)* %out, i32 %tid
@@ -96,11 +98,11 @@ define amdgpu_kernel void @v_test_add_v2i16_neg_constant(<2 x i16> addrspace(1)*
 ; GCN-LABEL: {{^}}v_test_add_v2i16_inline_neg1:
 ; GFX9: v_pk_add_u16 v{{[0-9]+}}, v{{[0-9]+}}, -1{{$}}
 
+; VI: v_mov_b32_e32 v[[SCONST:[0-9]+]], -1
 ; VI: flat_load_ushort [[LOAD0:v[0-9]+]]
 ; VI: flat_load_ushort [[LOAD1:v[0-9]+]]
-; VI-DAG: v_add_u16_e32 v{{[0-9]+}}, -1, [[LOAD0]]
+; VI-DAG: v_add_u16_sdwa v{{[0-9]+}}, v[[SCONST]], [[LOAD0]] dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
 ; VI-DAG: v_add_u16_e32 v{{[0-9]+}}, -1, [[LOAD1]]
-; VI-DAG: v_lshlrev_b32_e32 v{{[0-9]+}}, 16,
 ; VI: v_or_b32_e32
 define amdgpu_kernel void @v_test_add_v2i16_inline_neg1(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in0) #1 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -114,7 +116,7 @@ define amdgpu_kernel void @v_test_add_v2i16_inline_neg1(<2 x i16> addrspace(1)*
 
 ; GCN-LABEL: {{^}}v_test_add_v2i16_inline_lo_zero_hi:
 ; GFX9: s_mov_b32 [[K:s[0-9]+]], 32{{$}}
-; GFX9: v_pk_add_u16 v{{[0-9]+}}, [[K]], v{{[0-9]+}}{{$}}
+; GFX9: v_pk_add_u16 v{{[0-9]+}}, v{{[0-9]+}}, [[K]]{{$}}
 
 ; VI-NOT: v_add_u16
 ; VI: v_add_u16_e32 v{{[0-9]+}}, 32, v{{[0-9]+}}
@@ -134,12 +136,12 @@ define amdgpu_kernel void @v_test_add_v2i16_inline_lo_zero_hi(<2 x i16> addrspac
 ; The high element gives fp
 ; GCN-LABEL: {{^}}v_test_add_v2i16_inline_fp_split:
 ; GFX9: s_mov_b32 [[K:s[0-9]+]], 1.0
-; GFX9: v_pk_add_u16 v{{[0-9]+}}, [[K]], v{{[0-9]+}}{{$}}
+; GFX9: v_pk_add_u16 v{{[0-9]+}}, v{{[0-9]+}}, [[K]]{{$}}
 
 ; VI-NOT: v_add_u16
-; VI: v_add_u16_e32 v{{[0-9]+}}, 0x3f80, v{{[0-9]+}}
+; VI: v_mov_b32_e32 v[[K:[0-9]+]], 0x3f80
+; VI: v_add_u16_sdwa v{{[0-9]+}}, v[[K]], v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
 ; VI-NOT: v_add_u16
-; VI: v_lshlrev_b32_e32 v{{[0-9]+}}, 16,
 ; VI: v_or_b32_e32
 define amdgpu_kernel void @v_test_add_v2i16_inline_fp_split(<2 x i16> addrspace(1)* %out, <2 x i16> addrspace(1)* %in0) #1 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x()
@@ -191,19 +193,17 @@ define amdgpu_kernel void @v_test_add_v2i16_zext_to_v2i32(<2 x i32> addrspace(1)
 ; GFX9: flat_load_dword [[A:v[0-9]+]]
 ; GFX9: flat_load_dword [[B:v[0-9]+]]
 
-; GFX9: v_mov_b32_e32 v{{[0-9]+}}, 0{{$}}
 ; GFX9: v_pk_add_u16 [[ADD:v[0-9]+]], [[A]], [[B]]
 ; GFX9-DAG: v_and_b32_e32 v[[ELT0:[0-9]+]], 0xffff, [[ADD]]
 ; GFX9-DAG: v_lshrrev_b32_e32 v[[ELT1:[0-9]+]], 16, [[ADD]]
 ; GFX9: buffer_store_dwordx4
 
+; VI-DAG: v_mov_b32_e32 v{{[0-9]+}}, 0{{$}}
 ; VI: flat_load_ushort v[[A_LO:[0-9]+]]
 ; VI: flat_load_ushort v[[A_HI:[0-9]+]]
 ; VI: flat_load_ushort v[[B_LO:[0-9]+]]
 ; VI: flat_load_ushort v[[B_HI:[0-9]+]]
 
-; VI-DAG: v_mov_b32_e32 v{{[0-9]+}}, 0{{$}}
-; VI-DAG: v_mov_b32_e32 v{{[0-9]+}}, 0{{$}}
 ; VI-DAG: v_add_u16_e32
 ; VI-DAG: v_add_u16_e32
 

test/CodeGen/AMDGPU/bfe-combine.ll

@@ -1,12 +1,16 @@
-; RUN: llc -march=amdgcn -mcpu=fiji < %s | FileCheck --check-prefix=GCN --check-prefix=VI %s
+; RUN: llc -march=amdgcn -mcpu=fiji -amdgpu-sdwa-peephole=0 < %s | FileCheck --check-prefix=GCN --check-prefix=VI %s
+; RUN: llc -march=amdgcn -mcpu=fiji < %s | FileCheck --check-prefix=GCN --check-prefix=VI-SDWA %s
 ; RUN: llc -march=amdgcn -mcpu=bonaire < %s | FileCheck --check-prefix=GCN --check-prefix=CI %s
 
 ; GCN-LABEL: {{^}}bfe_combine8:
 ; VI: v_bfe_u32 v[[BFE:[0-9]+]], v{{[0-9]+}}, 8, 8
 ; VI: v_lshlrev_b32_e32 v[[ADDRBASE:[0-9]+]], 2, v[[BFE]]
+; VI-SDWA: v_mov_b32_e32 v[[SHIFT:[0-9]+]], 2
+; VI-SDWA: v_lshlrev_b32_sdwa v[[ADDRBASE:[0-9]+]], v[[SHIFT]], v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_1
 ; CI: v_lshrrev_b32_e32 v[[SHR:[0-9]+]], 6, v{{[0-9]+}}
 ; CI: v_and_b32_e32 v[[ADDRLO:[0-9]+]], 0x3fc, v[[SHR]]
 ; VI: v_add_i32_e32 v[[ADDRLO:[0-9]+]], vcc, s{{[0-9]+}}, v[[ADDRBASE]]
+; VI-SDWA: v_add_i32_e32 v[[ADDRLO:[0-9]+]], vcc, s{{[0-9]+}}, v[[ADDRBASE]]
 ; GCN: load_dword v{{[0-9]+}}, v{{\[}}[[ADDRLO]]:
 define amdgpu_kernel void @bfe_combine8(i32 addrspace(1)* nocapture %arg, i32 %x) {
   %id = tail call i32 @llvm.amdgcn.workitem.id.x() #2
@@ -22,6 +26,10 @@ define amdgpu_kernel void @bfe_combine8(i32 addrspace(1)* nocapture %arg, i32 %x
 ; GCN-LABEL: {{^}}bfe_combine16:
 ; VI: v_bfe_u32 v[[BFE:[0-9]+]], v{{[0-9]+}}, 16, 16
 ; VI: v_lshlrev_b32_e32 v[[ADDRBASE:[0-9]+]], {{[^,]+}}, v[[BFE]]
+; VI-SDWA: v_mov_b32_e32 v[[SHIFT:[0-9]+]], 15
+; VI-SDWA: v_lshlrev_b32_sdwa v[[ADDRBASE1:[0-9]+]], v[[SHIFT]], v{{[0-9]+}} dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; VI-SDWA: v_lshlrev_b64 v{{\[}}[[ADDRBASE:[0-9]+]]:{{[^\]+}}], 2, v{{\[}}[[ADDRBASE1]]:{{[^\]+}}]
+; VI-SDWA: v_add_i32_e32 v[[ADDRLO:[0-9]+]], vcc, s{{[0-9]+}}, v[[ADDRBASE]]
 ; CI: v_lshrrev_b32_e32 v[[SHR:[0-9]+]], 1, v{{[0-9]+}}
 ; CI: v_and_b32_e32 v[[AND:[0-9]+]], 0x7fff8000, v[[SHR]]
 ; CI: v_lshl_b64 v{{\[}}[[ADDRLO:[0-9]+]]:{{[^\]+}}], v{{\[}}[[AND]]:{{[^\]+}}], 2

test/CodeGen/AMDGPU/commute-compares.ll

@@ -1,4 +1,4 @@
-; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
+; RUN: llc -march=amdgcn -amdgpu-sdwa-peephole=0 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s
 
 declare i32 @llvm.amdgcn.workitem.id.x() #0
 

test/CodeGen/AMDGPU/commute_modifiers.ll

@@ -51,7 +51,7 @@ define amdgpu_kernel void @commute_mul_imm_fneg_f32(float addrspace(1)* %out, fl
 ; FUNC-LABEL: @commute_add_lit_fabs_f32
 ; SI: buffer_load_dword [[X:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}}
 ; SI: v_mov_b32_e32 [[K:v[0-9]+]], 0x44800000
-; SI: v_add_f32_e64 [[REG:v[0-9]+]], [[K]], |[[X]]|
+; SI: v_add_f32_e64 [[REG:v[0-9]+]], |[[X]]|, [[K]]
 ; SI: buffer_store_dword [[REG]]
 define amdgpu_kernel void @commute_add_lit_fabs_f32(float addrspace(1)* %out, float addrspace(1)* %in) #0 {
   %tid = call i32 @llvm.amdgcn.workitem.id.x() #1

test/CodeGen/AMDGPU/copy-illegal-type.ll

@@ -1,5 +1,5 @@
 ; RUN: llc -march=amdgcn -mcpu=tahiti < %s | FileCheck -check-prefix=GCN -check-prefix=SI -check-prefix=FUNC %s
-; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global < %s | FileCheck -check-prefix=GCN -check-prefix=VI -check-prefix=FUNC %s
+; RUN: llc -march=amdgcn -mcpu=tonga -mattr=-flat-for-global -amdgpu-sdwa-peephole=0 < %s | FileCheck -check-prefix=GCN -check-prefix=VI -check-prefix=FUNC %s
 
 declare i32 @llvm.amdgcn.workitem.id.x() nounwind readnone
 declare i32 @llvm.amdgcn.workitem.id.y() nounwind readnone

test/CodeGen/AMDGPU/cvt_f32_ubyte.ll

@@ -94,7 +94,6 @@ define amdgpu_kernel void @load_v4i8_to_v4f32_unaligned(<4 x float> addrspace(1)
 ; GCN-DAG: v_cvt_f32_ubyte3_e32
 
 ; GCN-DAG: v_lshrrev_b32_e32 v{{[0-9]+}}, 24
-; GCN-DAG: v_lshrrev_b32_e32 v{{[0-9]+}}, 16
 
 ; SI-DAG: v_lshlrev_b32_e32 v{{[0-9]+}}, 16
 ; SI-DAG: v_lshlrev_b32_e32 v{{[0-9]+}}, 8

test/CodeGen/AMDGPU/fabs.f64.ll

@@ -55,7 +55,7 @@ define amdgpu_kernel void @fabs_v4f64(<4 x double> addrspace(1)* %out, <4 x doub
 ; SI-LABEL: {{^}}fabs_fold_f64:
 ; SI: s_load_dwordx2 [[ABS_VALUE:s\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, 0xb
 ; SI-NOT: and
-; SI: v_mul_f64 {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, |[[ABS_VALUE]]|
+; SI: v_mul_f64 {{v\[[0-9]+:[0-9]+\]}}, |[[ABS_VALUE]]|, {{v\[[0-9]+:[0-9]+\]}}
 ; SI: s_endpgm
 define amdgpu_kernel void @fabs_fold_f64(double addrspace(1)* %out, double %in0, double %in1) {
   %fabs = call double @llvm.fabs.f64(double %in0)
@@ -67,7 +67,7 @@ define amdgpu_kernel void @fabs_fold_f64(double addrspace(1)* %out, double %in0,
 ; SI-LABEL: {{^}}fabs_fn_fold_f64:
 ; SI: s_load_dwordx2 [[ABS_VALUE:s\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, 0xb
 ; SI-NOT: and
-; SI: v_mul_f64 {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, |[[ABS_VALUE]]|
+; SI: v_mul_f64 {{v\[[0-9]+:[0-9]+\]}}, |[[ABS_VALUE]]|, {{v\[[0-9]+:[0-9]+\]}}
 ; SI: s_endpgm
 define amdgpu_kernel void @fabs_fn_fold_f64(double addrspace(1)* %out, double %in0, double %in1) {
   %fabs = call double @fabs(double %in0)

test/CodeGen/AMDGPU/fabs.ll

@@ -75,7 +75,7 @@ define amdgpu_kernel void @fabs_v4f32(<4 x float> addrspace(1)* %out, <4 x float
 ; SI: s_load_dword [[ABS_VALUE:s[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0xb
 ; VI: s_load_dword [[ABS_VALUE:s[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0x2c
 ; GCN-NOT: and
-; GCN: v_mul_f32_e64 v{{[0-9]+}}, v{{[0-9]+}}, |[[ABS_VALUE]]|
+; GCN: v_mul_f32_e64 v{{[0-9]+}}, |[[ABS_VALUE]]|, v{{[0-9]+}}
 define amdgpu_kernel void @fabs_fn_fold(float addrspace(1)* %out, float %in0, float %in1) {
   %fabs = call float @fabs(float %in0)
   %fmul = fmul float %fabs, %in1
@@ -87,7 +87,7 @@ define amdgpu_kernel void @fabs_fn_fold(float addrspace(1)* %out, float %in0, fl
 ; SI: s_load_dword [[ABS_VALUE:s[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0xb
 ; VI: s_load_dword [[ABS_VALUE:s[0-9]+]], s[{{[0-9]+:[0-9]+}}], 0x2c
 ; GCN-NOT: and
-; GCN: v_mul_f32_e64 v{{[0-9]+}}, v{{[0-9]+}}, |[[ABS_VALUE]]|
+; GCN: v_mul_f32_e64 v{{[0-9]+}}, |[[ABS_VALUE]]|, v{{[0-9]+}}
 define amdgpu_kernel void @fabs_fold(float addrspace(1)* %out, float %in0, float %in1) {
   %fabs = call float @llvm.fabs.f32(float %in0)
   %fmul = fmul float %fabs, %in1

test/CodeGen/AMDGPU/fadd.f16.ll

@@ -96,9 +96,9 @@ entry:
 }
 
 ; GCN-LABEL: {{^}}fadd_v2f16_imm_a:
-; GCN: buffer_load_dword v[[B_V2_F16:[0-9]+]]
+; GCN-DAG: buffer_load_dword v[[B_V2_F16:[0-9]+]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+; SI:  v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
 ; SI:  v_add_f32_e32 v[[R_F32_0:[0-9]+]], 1.0, v[[B_F32_0]]
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
@@ -107,9 +107,9 @@ entry:
 ; SI-DAG:  v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
 ; SI:  v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 
-; VI-DAG: v_add_f16_e32 v[[R_F16_1:[0-9]+]], 2.0, v[[B_F16_1]]
+; VI-DAG: v_mov_b32_e32 v[[CONST2:[0-9]+]], 0x4000
+; VI-DAG: v_add_f16_sdwa v[[R_F16_HI:[0-9]+]], v[[CONST2]], v[[B_V2_F16]] dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; VI-DAG: v_add_f16_e32 v[[R_F16_0:[0-9]+]], 1.0, v[[B_V2_F16]]
-; VI-DAG: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
 ; VI:     v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 
 ; GCN: buffer_store_dword v[[R_V2_F16]]
@@ -125,9 +125,9 @@ entry:
 }
 
 ; GCN-LABEL: {{^}}fadd_v2f16_imm_b:
-; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
+; GCN-DAG: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; SI-DAG: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
 ; SI:  v_add_f32_e32 v[[R_F32_0:[0-9]+]], 2.0, v[[A_F32_0]]
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
@@ -136,10 +136,10 @@ entry:
 ; SI-DAG:  v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
 ; SI:  v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 
-; VI-DAG: v_add_f16_e32 v[[R_F16_0:[0-9]+]], 1.0, v[[A_F16_1]]
+; VI-DAG: v_mov_b32_e32 v[[CONST1:[0-9]+]], 0x3c00
+; VI-DAG: v_add_f16_sdwa v[[R_F16_0:[0-9]+]], v[[CONST1]], v[[A_V2_F16]] dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; VI-DAG: v_add_f16_e32 v[[R_F16_1:[0-9]+]], 2.0, v[[A_V2_F16]]
-; VI-DAG: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_0]]
-; VI:     v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_1]]
+; VI:     v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_0]], v[[R_F16_1]]
 
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm

test/CodeGen/AMDGPU/fadd64.ll

@@ -13,7 +13,7 @@ define amdgpu_kernel void @v_fadd_f64(double addrspace(1)* %out, double addrspac
 }
 
 ; CHECK-LABEL: {{^}}s_fadd_f64:
-; CHECK: v_add_f64 {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}
+; CHECK: v_add_f64 {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}
 define amdgpu_kernel void @s_fadd_f64(double addrspace(1)* %out, double %r0, double %r1) {
   %r2 = fadd double %r0, %r1
   store double %r2, double addrspace(1)* %out

test/CodeGen/AMDGPU/fcanonicalize.f16.ll

@@ -205,9 +205,9 @@ define amdgpu_kernel void @test_fold_canonicalize_snan3_value_f16(half addrspace
 }
 
 ; GCN-LABEL: {{^}}v_test_canonicalize_var_v2f16:
-; VI: v_mul_f16_e32 [[REG0:v[0-9]+]], 1.0, {{v[0-9]+}}
+; VI: v_mov_b32_e32 v[[CONST1:[0-9]+]], 0x3c00
+; VI-DAG: v_mul_f16_sdwa [[REG0:v[0-9]+]], v[[CONST1]], {{v[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; VI-DAG: v_mul_f16_e32 [[REG1:v[0-9]+]], 1.0, {{v[0-9]+}}
-; VI-DAG: v_lshlrev_b32_e32 v{{[0-9]+}}, 16,
 ; VI-NOT: v_and_b32
 
 ; GFX9: v_pk_mul_f16 [[REG:v[0-9]+]], 1.0, {{v[0-9]+$}}
@@ -223,7 +223,8 @@ define amdgpu_kernel void @v_test_canonicalize_var_v2f16(<2 x half> addrspace(1)
 ; GCN-LABEL: {{^}}v_test_canonicalize_fabs_var_v2f16:
 ; VI-DAG: v_bfe_u32
 ; VI-DAG: v_and_b32_e32 v{{[0-9]+}}, 0x7fff7fff, v{{[0-9]+}}
-; VI: v_mul_f16_e32 [[REG0:v[0-9]+]], 1.0, v{{[0-9]+}}
+; VI-DAG: v_mov_b32_e32 v[[CONST1:[0-9]+]], 0x3c00
+; VI: v_mul_f16_sdwa [[REG0:v[0-9]+]], v[[CONST1]], v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
 ; VI: v_mul_f16_e32 [[REG1:v[0-9]+]], 1.0, v{{[0-9]+}}
 ; VI-NOT: 0xffff
 ; VI: v_or_b32
@@ -240,9 +241,10 @@ define amdgpu_kernel void @v_test_canonicalize_fabs_var_v2f16(<2 x half> addrspa
 }
 
 ; GCN-LABEL: {{^}}v_test_canonicalize_fneg_fabs_var_v2f16:
-; VI: v_or_b32_e32 v{{[0-9]+}}, 0x80008000, v{{[0-9]+}}
-; VI: v_mul_f16_e32 [[REG0:v[0-9]+]], 1.0, v{{[0-9]+}}
-; VI: v_mul_f16_e32 [[REG1:v[0-9]+]], 1.0, v{{[0-9]+}}
+; VI-DAG: v_mov_b32_e32 v[[CONST1:[0-9]+]], 0x3c00
+; VI-DAG: v_or_b32_e32 v{{[0-9]+}}, 0x80008000, v{{[0-9]+}}
+; VI-DAG: v_mul_f16_sdwa [[REG0:v[0-9]+]], v[[CONST1]], v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
+; VI-DAG: v_mul_f16_e32 [[REG1:v[0-9]+]], 1.0, v{{[0-9]+}}
 ; VI: v_or_b32
 
 ; GFX9: v_and_b32_e32 [[ABS:v[0-9]+]], 0x7fff7fff, v{{[0-9]+}}
@@ -259,11 +261,10 @@ define amdgpu_kernel void @v_test_canonicalize_fneg_fabs_var_v2f16(<2 x half> ad
 
 ; FIXME: Fold modifier
 ; GCN-LABEL: {{^}}v_test_canonicalize_fneg_var_v2f16:
-; VI: v_xor_b32_e32 [[FNEG:v[0-9]+]], 0x80008000, v{{[0-9]+}}
-; VI-DAG: v_lshrrev_b32_e32 [[FNEG_HI:v[0-9]+]], 16, [[FNEG]]
-; VI-DAG: v_mul_f16_e32 [[REG1:v[0-9]+]], 1.0, [[FNEG_HI]]
+; VI-DAG: v_mov_b32_e32 v[[CONST1:[0-9]+]], 0x3c00
+; VI-DAG: v_xor_b32_e32 [[FNEG:v[0-9]+]], 0x80008000, v{{[0-9]+}}
+; VI-DAG: v_mul_f16_sdwa [[REG1:v[0-9]+]], v[[CONST1]], [[FNEG]] dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; VI-DAG: v_mul_f16_e32 [[REG0:v[0-9]+]], 1.0, [[FNEG]]
-; VI-DAG: v_lshlrev_b32_e32 v{{[0-9]+}}, 16,
 ; VI-NOT: 0xffff
 
 ; GFX9: v_pk_mul_f16 [[REG:v[0-9]+]], 1.0, {{v[0-9]+}} neg_lo:[0,1] neg_hi:[0,1]{{$}}

test/CodeGen/AMDGPU/fmul.f16.ll

@@ -96,17 +96,18 @@ entry:
 }
 
 ; GCN-LABEL: {{^}}fmul_v2f16_imm_a:
-; GCN: buffer_load_dword v[[B_V2_F16:[0-9]+]]
+; GCN-DAG: buffer_load_dword v[[B_V2_F16:[0-9]+]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_0:[0-9]+]], v[[B_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
+; SI:  v_lshrrev_b32_e32 v[[B_F16_1:[0-9]+]], 16, v[[B_V2_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[B_F32_1:[0-9]+]], v[[B_F16_1]]
 ; SI:  v_mul_f32_e32 v[[R_F32_0:[0-9]+]], 0x40400000, v[[B_F32_0]]
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
 ; SI:  v_mul_f32_e32 v[[R_F32_1:[0-9]+]], 4.0, v[[B_F32_1]]
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
-; VI-DAG:  v_mul_f16_e32 v[[R_F16_1:[0-9]+]], 4.0, v[[B_F16_1]]
+; VI-DAG:  v_mov_b32_e32 v[[CONST4:[0-9]+]], 0x4400
+; VI-DAG:  v_mul_f16_sdwa v[[R_F16_HI:[0-9]+]], v[[CONST4]], v[[B_V2_F16]] dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; VI-DAG:  v_mul_f16_e32 v[[R_F16_0:[0-9]+]], 0x4200, v[[B_V2_F16]]
-; GCN-DAG: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
+; SI-DAG: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
 ; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm
@@ -121,17 +122,18 @@ entry:
 }
 
 ; GCN-LABEL: {{^}}fmul_v2f16_imm_b:
-; GCN: buffer_load_dword v[[A_V2_F16:[0-9]+]]
+; GCN-DAG: buffer_load_dword v[[A_V2_F16:[0-9]+]]
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_0:[0-9]+]], v[[A_V2_F16]]
-; GCN: v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
+; SI:  v_lshrrev_b32_e32 v[[A_F16_1:[0-9]+]], 16, v[[A_V2_F16]]
 ; SI:  v_cvt_f32_f16_e32 v[[A_F32_1:[0-9]+]], v[[A_F16_1]]
 ; SI:  v_mul_f32_e32 v[[R_F32_0:[0-9]+]], 4.0, v[[A_F32_0]]
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_0:[0-9]+]], v[[R_F32_0]]
 ; SI:  v_mul_f32_e32 v[[R_F32_1:[0-9]+]], 0x40400000, v[[A_F32_1]]
 ; SI:  v_cvt_f16_f32_e32 v[[R_F16_1:[0-9]+]], v[[R_F32_1]]
-; VI-DAG:  v_mul_f16_e32 v[[R_F16_1:[0-9]+]], 0x4200, v[[A_F16_1]]
+; VI-DAG:  v_mov_b32_e32 v[[CONST3:[0-9]+]], 0x4200
+; VI-DAG:  v_mul_f16_sdwa v[[R_F16_HI:[0-9]+]], v[[CONST3]], v[[A_V2_F16]] dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 ; VI-DAG:  v_mul_f16_e32 v[[R_F16_0:[0-9]+]], 4.0, v[[A_V2_F16]]
-; GCN-DAG: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
+; SI-DAG: v_lshlrev_b32_e32 v[[R_F16_HI:[0-9]+]], 16, v[[R_F16_1]]
 ; GCN: v_or_b32_e32 v[[R_V2_F16:[0-9]+]], v[[R_F16_HI]], v[[R_F16_0]]
 ; GCN: buffer_store_dword v[[R_V2_F16]]
 ; GCN: s_endpgm

test/CodeGen/AMDGPU/fneg-fabs.f16.ll

@@ -71,7 +71,9 @@ define amdgpu_kernel void @v_fneg_fabs_f16(half addrspace(1)* %out, half addrspa
 ; FIXME: single bit op
 ; GCN-LABEL: {{^}}s_fneg_fabs_v2f16:
 ; CIVI: s_mov_b32 [[MASK:s[0-9]+]], 0x8000{{$}}
-; CIVI: v_or_b32_e32 v{{[0-9]+}}, [[MASK]],
+; VI: v_mov_b32_e32 [[VMASK:v[0-9]+]], [[MASK]]
+; CI: v_or_b32_e32 v{{[0-9]+}}, [[MASK]],
+; VI: v_or_b32_sdwa v{{[0-9]+}}, [[VMASK]], v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
 ; CIVI: v_or_b32_e32 v{{[0-9]+}}, [[MASK]],
 ; CIVI: flat_store_dword
 
@@ -85,10 +87,15 @@ define amdgpu_kernel void @s_fneg_fabs_v2f16(<2 x half> addrspace(1)* %out, <2 x
 
 ; GCN-LABEL: {{^}}fneg_fabs_v4f16:
 ; CIVI: s_mov_b32 [[MASK:s[0-9]+]], 0x8000{{$}}
-; CIVI: v_or_b32_e32 v{{[0-9]+}}, [[MASK]],
-; CIVI: v_or_b32_e32 v{{[0-9]+}}, [[MASK]],
-; CIVI: v_or_b32_e32 v{{[0-9]+}}, [[MASK]],
-; CIVI: v_or_b32_e32 v{{[0-9]+}}, [[MASK]],
+; CI: v_or_b32_e32 v{{[0-9]+}}, [[MASK]],
+; CI: v_or_b32_e32 v{{[0-9]+}}, [[MASK]],
+; CI: v_or_b32_e32 v{{[0-9]+}}, [[MASK]],
+; CI: v_or_b32_e32 v{{[0-9]+}}, [[MASK]],
+; VI: v_mov_b32_e32 [[VMASK:v[0-9]+]], [[MASK]]
+; VI: v_or_b32_sdwa v{{[0-9]+}}, [[VMASK]], v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI: v_or_b32_e32 v{{[0-9]+}}, [[MASK]],
+; VI: v_or_b32_sdwa v{{[0-9]+}}, [[VMASK]], v{{[0-9]+}} dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD
+; VI: v_or_b32_e32 v{{[0-9]+}}, [[MASK]],
 
 ; GFX9: s_mov_b32 [[MASK:s[0-9]+]], 0x80008000
 ; GFX9: s_or_b32 s{{[0-9]+}}, [[MASK]], s{{[0-9]+}}

test/CodeGen/AMDGPU/fneg-fabs.f64.ll

@@ -5,7 +5,7 @@
 ; into 2 modifiers, although theoretically that should work.
 
 ; GCN-LABEL: {{^}}fneg_fabs_fadd_f64:
-; GCN: v_add_f64 {{v\[[0-9]+:[0-9]+\]}}, -|v{{\[[0-9]+:[0-9]+\]}}|, {{s\[[0-9]+:[0-9]+\]}}
+; GCN: v_add_f64 {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, -|v{{\[[0-9]+:[0-9]+\]}}|
 define amdgpu_kernel void @fneg_fabs_fadd_f64(double addrspace(1)* %out, double %x, double %y) {
   %fabs = call double @llvm.fabs.f64(double %x)
   %fsub = fsub double -0.000000e+00, %fabs
@@ -25,7 +25,7 @@ define amdgpu_kernel void @v_fneg_fabs_fadd_f64(double addrspace(1)* %out, doubl
 }
 
 ; GCN-LABEL: {{^}}fneg_fabs_fmul_f64:
-; GCN: v_mul_f64 {{v\[[0-9]+:[0-9]+\]}}, -|{{v\[[0-9]+:[0-9]+\]}}|, {{s\[[0-9]+:[0-9]+\]}}
+; GCN: v_mul_f64 {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, -|v{{\[[0-9]+:[0-9]+\]}}|
 define amdgpu_kernel void @fneg_fabs_fmul_f64(double addrspace(1)* %out, double %x, double %y) {
   %fabs = call double @llvm.fabs.f64(double %x)
   %fsub = fsub double -0.000000e+00, %fabs

test/CodeGen/AMDGPU/fneg-fabs.ll

@@ -4,7 +4,7 @@
 
 ; FUNC-LABEL: {{^}}fneg_fabs_fadd_f32:
 ; SI-NOT: and
-; SI: v_subrev_f32_e64 {{v[0-9]+}}, |{{v[0-9]+}}|, {{s[0-9]+}}
+; SI: v_sub_f32_e64 {{v[0-9]+}}, {{s[0-9]+}}, |{{v[0-9]+}}|
 define amdgpu_kernel void @fneg_fabs_fadd_f32(float addrspace(1)* %out, float %x, float %y) {
   %fabs = call float @llvm.fabs.f32(float %x)
   %fsub = fsub float -0.000000e+00, %fabs
@@ -15,7 +15,7 @@ define amdgpu_kernel void @fneg_fabs_fadd_f32(float addrspace(1)* %out, float %x
 
 ; FUNC-LABEL: {{^}}fneg_fabs_fmul_f32:
 ; SI-NOT: and
-; SI: v_mul_f32_e64 {{v[0-9]+}}, -|{{v[0-9]+}}|, {{s[0-9]+}}
+; SI: v_mul_f32_e64 {{v[0-9]+}}, {{s[0-9]+}}, -|{{v[0-9]+}}|
 ; SI-NOT: and
 define amdgpu_kernel void @fneg_fabs_fmul_f32(float addrspace(1)* %out, float %x, float %y) {
   %fabs = call float @llvm.fabs.f32(float %x)

test/CodeGen/AMDGPU/fneg.f16.ll

@@ -130,13 +130,15 @@ define amdgpu_kernel void @v_fneg_fold_v2f16(<2 x half> addrspace(1)* %out, <2 x
 }
 
 ; GCN-LABEL: {{^}}v_extract_fneg_fold_v2f16:
-; GCN: flat_load_dword [[VAL:v[0-9]+]]
+; GCN-DAG: flat_load_dword [[VAL:v[0-9]+]]
 ; CI-DAG: v_mul_f32_e32 v{{[0-9]+}}, -4.0, v{{[0-9]+}}
 ; CI-DAG: v_sub_f32_e32 v{{[0-9]+}}, 2.0, v{{[0-9]+}}
 
-; GFX89: v_lshrrev_b32_e32 [[ELT1:v[0-9]+]], 16, [[VAL]]
+; GFX9: v_lshrrev_b32_e32 [[ELT1:v[0-9]+]], 16, [[VAL]]
 ; GFX89-DAG: v_mul_f16_e32 v{{[0-9]+}}, -4.0, [[VAL]]
-; GFX89-DAG: v_sub_f16_e32 v{{[0-9]+}}, 2.0, [[ELT1]]
+; GFX9-DAG: v_sub_f16_e32 v{{[0-9]+}}, 2.0, [[ELT1]]
+; VI-DAG: v_mov_b32_e32 [[CONST2:v[0-9]+]], 0x4000
+; VI-DAG: v_sub_f16_sdwa v{{[0-9]+}}, [[CONST2]], [[VAL]] dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:WORD_1
 define amdgpu_kernel void @v_extract_fneg_fold_v2f16(<2 x half> addrspace(1)* %in) #0 {
   %val = load <2 x half>, <2 x half> addrspace(1)* %in
   %fneg = fsub <2 x half> <half -0.0, half -0.0>, %val

test/CodeGen/AMDGPU/fract.f64.ll

@@ -12,7 +12,7 @@ declare double @llvm.floor.f64(double) #0
 ; SI-DAG: v_fract_f64_e32 [[FRC:v\[[0-9]+:[0-9]+\]]], v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]]
 ; SI-DAG: v_mov_b32_e32 v[[UPLO:[0-9]+]], -1
 ; SI-DAG: v_mov_b32_e32 v[[UPHI:[0-9]+]], 0x3fefffff
-; SI-DAG: v_min_f64 v{{\[}}[[MINLO:[0-9]+]]:[[MINHI:[0-9]+]]], v{{\[}}[[UPLO]]:[[UPHI]]], [[FRC]]
+; SI-DAG: v_min_f64 v{{\[}}[[MINLO:[0-9]+]]:[[MINHI:[0-9]+]]], [[FRC]], v{{\[}}[[UPLO]]:[[UPHI]]]
 ; SI-DAG: v_cmp_class_f64_e64 vcc, v{{\[}}[[LO]]:[[HI]]], 3
 ; SI: v_cndmask_b32_e32 v[[RESLO:[0-9]+]], v[[MINLO]], v[[LO]], vcc
 ; SI: v_cndmask_b32_e32 v[[RESHI:[0-9]+]], v[[MINHI]], v[[HI]], vcc
@@ -39,7 +39,7 @@ define amdgpu_kernel void @fract_f64(double addrspace(1)* %out, double addrspace
 ; SI-DAG: v_fract_f64_e64 [[FRC:v\[[0-9]+:[0-9]+\]]], -v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]]
 ; SI-DAG: v_mov_b32_e32 v[[UPLO:[0-9]+]], -1
 ; SI-DAG: v_mov_b32_e32 v[[UPHI:[0-9]+]], 0x3fefffff
-; SI-DAG: v_min_f64 v{{\[}}[[MINLO:[0-9]+]]:[[MINHI:[0-9]+]]], v{{\[}}[[UPLO]]:[[UPHI]]], [[FRC]]
+; SI-DAG: v_min_f64 v{{\[}}[[MINLO:[0-9]+]]:[[MINHI:[0-9]+]]], [[FRC]], v{{\[}}[[UPLO]]:[[UPHI]]]
 ; SI-DAG: v_cmp_class_f64_e64 vcc, v{{\[}}[[LO]]:[[HI]]], 3
 ; SI: v_cndmask_b32_e32 v[[RESLO:[0-9]+]], v[[MINLO]], v[[LO]], vcc
 ; SI: v_cndmask_b32_e32 v[[RESHI:[0-9]+]], v[[MINHI]], v[[HI]], vcc
@@ -67,7 +67,7 @@ define amdgpu_kernel void @fract_f64_neg(double addrspace(1)* %out, double addrs
 ; SI-DAG: v_fract_f64_e64 [[FRC:v\[[0-9]+:[0-9]+\]]], -|v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]]|
 ; SI-DAG: v_mov_b32_e32 v[[UPLO:[0-9]+]], -1
 ; SI-DAG: v_mov_b32_e32 v[[UPHI:[0-9]+]], 0x3fefffff
-; SI-DAG: v_min_f64 v{{\[}}[[MINLO:[0-9]+]]:[[MINHI:[0-9]+]]], v{{\[}}[[UPLO]]:[[UPHI]]], [[FRC]]
+; SI-DAG: v_min_f64 v{{\[}}[[MINLO:[0-9]+]]:[[MINHI:[0-9]+]]], [[FRC]], v{{\[}}[[UPLO]]:[[UPHI]]]
 ; SI-DAG: v_cmp_class_f64_e64 vcc, v{{\[}}[[LO]]:[[HI]]], 3
 ; SI: v_cndmask_b32_e32 v[[RESLO:[0-9]+]], v[[MINLO]], v[[LO]], vcc
 ; SI: v_cndmask_b32_e32 v[[RESHI:[0-9]+]], v[[MINHI]], v[[HI]], vcc