Merge branch 'release/rocm-rel-7.0' into amd/dev/chfang/opsel-cp

Author: vamovsik

clang/lib/Basic/Targets/AMDGPU.cpp

@@ -253,8 +253,11 @@ AMDGPUTargetInfo::AMDGPUTargetInfo(const llvm::Triple &Triple,
 
   MaxAtomicPromoteWidth = MaxAtomicInlineWidth = 64;
   CUMode = !(GPUFeatures & llvm::AMDGPU::FEATURE_WGP);
-  for (auto F : {"image-insts", "gws", "vmem-to-lds-load-insts"})
-    ReadOnlyFeatures.insert(F);
+
+  for (auto F : {"image-insts", "gws", "vmem-to-lds-load-insts"}) {
+    if (GPUKind != llvm::AMDGPU::GK_NONE)
+      ReadOnlyFeatures.insert(F);
+  }
   HalfArgsAndReturns = true;
 }
 

clang/test/CodeGenOpenCL/amdgpu-features-readonly.cl

@@ -1,4 +1,4 @@
-// RUN: %clang_cc1 -triple amdgcn -target-feature +gws -o /dev/null %s 2>&1 \
+// RUN: %clang_cc1 -triple amdgcn -target-cpu gfx942 -target-feature +gws -o /dev/null %s 2>&1 \
 // RUN:   | FileCheck --check-prefix=GWS %s
 
 // GWS: warning: feature flag '+gws' is ignored since the feature is read only [-Winvalid-command-line-argument]

clang/test/CodeGenOpenCL/amdgpu-readonly-features-written-with-no-target.cl

@@ -0,0 +1,16 @@
+// REQUIRES: amdgpu-registered-target
+
+// Check the readonly feature will can be written to the IR
+// if there is no target specified.
+
+// RUN: %clang_cc1 -triple amdgcn -emit-llvm -o - %s | FileCheck --check-prefix=NOCPU %s
+// RUN: %clang_cc1 -triple amdgcn -target-cpu gfx942 -emit-llvm -o - %s | FileCheck --check-prefix=GFX942 %s
+// RUN: %clang_cc1 -triple amdgcn -target-cpu gfx1100 -emit-llvm -o - %s | FileCheck --check-prefix=GFX1100 %s
+// RUN: %clang_cc1 -triple amdgcn -target-cpu gfx1200 -emit-llvm -o - %s | FileCheck --check-prefix=GFX1200 %s
+
+__attribute__((target("gws,image-insts,vmem-to-lds-load-insts"))) void test() {}
+
+// NOCPU: "target-features"="+gws,+image-insts,+vmem-to-lds-load-insts"
+// GFX942: "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-ds-pk-add-16-insts,+atomic-fadd-rtn-insts,+atomic-flat-pk-add-16-insts,+atomic-global-pk-add-bf16-inst,+ci-insts,+dl-insts,+dot1-insts,+dot10-insts,+dot2-insts,+dot3-insts,+dot4-insts,+dot5-insts,+dot6-insts,+dot7-insts,+dpp,+fp8-conversion-insts,+fp8-insts,+gfx8-insts,+gfx9-insts,+gfx90a-insts,+gfx940-insts,+mai-insts,+s-memrealtime,+s-memtime-inst,+wavefrontsize64,+xf32-insts"
+// GFX1100: "target-features"="+16-bit-insts,+atomic-fadd-rtn-insts,+ci-insts,+dl-insts,+dot10-insts,+dot12-insts,+dot5-insts,+dot7-insts,+dot8-insts,+dot9-insts,+dpp,+gfx10-3-insts,+gfx10-insts,+gfx11-insts,+gfx8-insts,+gfx9-insts,+wavefrontsize32"
+// GFX1200: "target-features"="+16-bit-insts,+atomic-buffer-global-pk-add-f16-insts,+atomic-buffer-pk-add-bf16-inst,+atomic-ds-pk-add-16-insts,+atomic-fadd-rtn-insts,+atomic-flat-pk-add-16-insts,+atomic-global-pk-add-bf16-inst,+ci-insts,+dl-insts,+dot10-insts,+dot11-insts,+dot12-insts,+dot7-insts,+dot8-insts,+dot9-insts,+dpp,+fp8-conversion-insts,+gfx10-3-insts,+gfx10-insts,+gfx11-insts,+gfx12-insts,+gfx8-insts,+gfx9-insts,+wavefrontsize32"

llvm/include/llvm/Object/OffloadBundle.h

@@ -31,42 +31,56 @@ namespace llvm {
 
 namespace object {
 
+// CompressedOffloadBundle represents the format for the compressed offload
+// bundles.
+//
+// The format is as follows:
+// - Magic Number (4 bytes) - A constant "CCOB".
+// - Version (2 bytes)
+// - Compression Method (2 bytes) - Uses the values from
+// llvm::compression::Format.
+// - Total file size (4 bytes in V2, 8 bytes in V3).
+// - Uncompressed Size (4 bytes in V1/V2, 8 bytes in V3).
+// - Truncated MD5 Hash (8 bytes).
+// - Compressed Data (variable length).
 class CompressedOffloadBundle {
 private:
-  static inline const size_t MagicSize = 4;
-  static inline const size_t VersionFieldSize = sizeof(uint16_t);
-  static inline const size_t MethodFieldSize = sizeof(uint16_t);
-  static inline const size_t FileSizeFieldSize = sizeof(uint32_t);
-  static inline const size_t UncompressedSizeFieldSize = sizeof(uint32_t);
-  static inline const size_t HashFieldSize = sizeof(uint64_t);
-  static inline const size_t V1HeaderSize =
-      MagicSize + VersionFieldSize + MethodFieldSize +
-      UncompressedSizeFieldSize + HashFieldSize;
-  static inline const size_t V2HeaderSize =
-      MagicSize + VersionFieldSize + FileSizeFieldSize + MethodFieldSize +
-      UncompressedSizeFieldSize + HashFieldSize;
   static inline const llvm::StringRef MagicNumber = "CCOB";
-  static inline const uint16_t Version = 2;
 
 public:
+  struct CompressedBundleHeader {
+    unsigned Version;
+    llvm::compression::Format CompressionFormat;
+    std::optional<size_t> FileSize;
+    size_t UncompressedFileSize;
+    uint64_t Hash;
+
+    static llvm::Expected<CompressedBundleHeader> tryParse(llvm::StringRef);
+  };
+
+  static inline const uint16_t DefaultVersion = 2;
+
   static llvm::Expected<std::unique_ptr<llvm::MemoryBuffer>>
   compress(llvm::compression::Params P, const llvm::MemoryBuffer &Input,
-           bool Verbose = false);
+           uint16_t Version, bool Verbose = false);
   static llvm::Expected<std::unique_ptr<llvm::MemoryBuffer>>
-  decompress(llvm::MemoryBufferRef &Input, bool Verbose = false);
+  decompress(const llvm::MemoryBuffer &Input, bool Verbose = false);
 };
 
 /// Bundle entry in binary clang-offload-bundler format.
 struct OffloadBundleEntry {
   uint64_t Offset = 0u;
   uint64_t Size = 0u;
   uint64_t IDLength = 0u;
-  StringRef ID;
-  OffloadBundleEntry(uint64_t O, uint64_t S, uint64_t I, StringRef T)
-      : Offset(O), Size(S), IDLength(I), ID(T) {}
+  std::string ID;
+  OffloadBundleEntry(uint64_t O, uint64_t S, uint64_t I, std::string T)
+      : Offset(O), Size(S), IDLength(I) {
+    ID.reserve(T.size());
+    ID = T;
+  }
   void dumpInfo(raw_ostream &OS) {
     OS << "Offset = " << Offset << ", Size = " << Size
-       << ", ID Length = " << IDLength << ", ID = " << ID;
+       << ", ID Length = " << IDLength << ", ID = " << ID << "\n";
   }
   void dumpURI(raw_ostream &OS, StringRef FilePath) {
     OS << ID.data() << "\tfile://" << FilePath << "#offset=" << Offset
@@ -81,16 +95,21 @@ class OffloadBundleFatBin {
   StringRef FileName;
   uint64_t NumberOfEntries;
   SmallVector<OffloadBundleEntry> Entries;
+  bool Decompressed;
 
 public:
+  std::unique_ptr<MemoryBuffer> DecompressedBuffer;
+
   SmallVector<OffloadBundleEntry> getEntries() { return Entries; }
   uint64_t getSize() const { return Size; }
   StringRef getFileName() const { return FileName; }
   uint64_t getNumEntries() const { return NumberOfEntries; }
+  bool isDecompressed() const { return Decompressed; }
 
-  static Expected<std::unique_ptr<OffloadBundleFatBin>>
-  create(MemoryBufferRef, uint64_t SectionOffset, StringRef FileName);
-  Error extractBundle(const ObjectFile &Source);
+  LLVM_ABI static Expected<std::unique_ptr<OffloadBundleFatBin>>
+  create(MemoryBufferRef, uint64_t SectionOffset, StringRef FileName,
+         bool Decompress = false);
+  LLVM_ABI Error extractBundle(const ObjectFile &Source);
 
   Error dumpEntryToCodeObject();
 
@@ -105,9 +124,15 @@ class OffloadBundleFatBin {
       Entry.dumpURI(outs(), FileName);
   }
 
-  OffloadBundleFatBin(MemoryBufferRef Source, StringRef File)
-      : FileName(File), NumberOfEntries(0),
-        Entries(SmallVector<OffloadBundleEntry>()) {}
+  OffloadBundleFatBin(MemoryBufferRef Source, StringRef File,
+                      bool Decompress = false)
+      : FileName(File), Decompressed(Decompress), NumberOfEntries(0),
+        Entries(SmallVector<OffloadBundleEntry>()) {
+    if (Decompress) {
+      DecompressedBuffer =
+          MemoryBuffer::getMemBufferCopy(Source.getBuffer(), File);
+    }
+  }
 };
 
 enum UriTypeT { FILE_URI, MEMORY_URI };
@@ -190,6 +215,10 @@ Error extractOffloadBundleFatBinary(
 Error extractCodeObject(const ObjectFile &Source, int64_t Offset, int64_t Size,
                         StringRef OutputFileName);
 
+/// Extract code object memory from the given \p Source object file at \p Offset
+/// and of \p Size, and copy into \p OutputFileName.
+LLVM_ABI Error extractCodeObject(MemoryBufferRef Buffer, int64_t Offset,
+                                 int64_t Size, StringRef OutputFileName);
 /// Extracts an Offload Bundle Entry given by URI
 Error extractOffloadBundleByURI(StringRef URIstr);