Merge branch 'main' into staging-sm6.9

.github/CODEOWNERS

@@ -1 +1,2 @@
-* @microsoft/hlsl-release
+# Uncomment the next line in release branches after ask-mode begins
+# * @microsoft/hlsl-release

CMakeLists.txt

@@ -17,15 +17,6 @@ if(POLICY CMP0022)
   cmake_policy(SET CMP0022 NEW) # automatic when 2.8.12 is required
 endif()
 
-if (POLICY CMP0051)
-  # CMake 3.1 and higher include generator expressions of the form
-  # $<TARGETLIB:obj> in the SOURCES property.  These need to be
-  # stripped everywhere that access the SOURCES property, so we just
-  # defer to the OLD behavior of not including generator expressions
-  # in the output for now.
-  cmake_policy(SET CMP0051 OLD)
-endif()
-
 if(CMAKE_VERSION VERSION_LESS 3.1.20141117)
   set(cmake_3_2_USES_TERMINAL)
 else()

CONTRIBUTING.md

@@ -40,10 +40,32 @@ Before submitting a feature or substantial code contribution please discuss it w
 
 ### Coding guidelines
 
-The coding, style, and general engineering guidelines follow those described in the docs/CodingStandards.rst. For additional guidelines in code specific to HLSL, see the docs/HLSLChanges.rst file.
+The coding, style, and general engineering guidelines follow those described in the [LLVM Coding Standards](docs/CodingStandards.rst). For additional guidelines in code specific to HLSL, see the [HLSL Changes](docs/HLSLChanges.rst) docs.
 
 DXC has adopted a clang-format requirement for all incoming changes to C and C++ files. PRs to DXC should have the *changed code* clang formatted to the LLVM style, and leave the remaining portions of the file unchanged. This can be done using the `git-clang-format` tool or IDE driven workflows. A GitHub action will run on all PRs to validate that the change is properly formatted.
 
+#### Applying LLVM Standards
+
+All new code contributed to DXC should follow the LLVM coding standards.
+
+Note that the LLVM Coding Standards have a golden rule:
+
+> **If you are extending, enhancing, or bug fixing already implemented code, use the style that is already being used so that the source is uniform and easy to follow.**
+
+The golden rule should continue to be applied to places where DXC is self-consistent. A good example is DXC's common use of `PascalCase` instead of `camelCase` for APIs in some parts of the HLSL implementation. In any place where DXC is not self-consistent new code should follow the LLVM Coding Standard.
+
+A good secondary rule to follow is:
+
+> **When in doubt, follow LLVM.**
+
+Adopting LLVM's coding standards provides a consistent set of rules and guidelines to hold all contributions to. This allows patch authors to clearly understand the expectations placed on contributions, and allows reviewers to have a bar to measure contributions against. Aligning with LLVM by default ensures the path of least resistance for everyone.
+
+Since many of the LLVM Coding Standards are not enforced automatically we rely on code reviews to provide feedback and ensure contributions align with the expected coding standards. Since we rely on reviewers for enforcement and humans make mistakes, please keep in mind:
+
+> **Code review is a conversation.**
+
+It is completely reasonable for a patch author to question feedback and provide additional context about why something was done the way it was. Reviewers often see narrow slices in diffs rather than the full context of a file or part of the compiler, so they may not always provide perfect feedback. This is especially true with the application of the "golden rule" since it depends on understanding a wider context.
+
 ### Documenting Pull Requests
 
 Pull request descriptions should have the following format:

docs/SPIR-V.rst

@@ -320,6 +320,7 @@ Supported extensions
 * SPV_KHR_maximal_reconvergence
 * SPV_KHR_float_controls
 * SPV_NV_shader_subgroup_partitioned
+* SPV_KHR_quad_control
 
 Vulkan specific attributes
 --------------------------
@@ -4008,6 +4009,8 @@ Quad          ``QuadReadAcrossX()``        ``OpGroupNonUniformQuadSwap``
 Quad          ``QuadReadAcrossY()``        ``OpGroupNonUniformQuadSwap``
 Quad          ``QuadReadAcrossDiagonal()`` ``OpGroupNonUniformQuadSwap``
 Quad          ``QuadReadLaneAt()``         ``OpGroupNonUniformQuadBroadcast``
+Quad          ``QuadAny()``                ``OpGroupNonUniformQuadAnyKHR``
+Quad          ``QuadAll()``                ``OpGroupNonUniformQuadAllKHR``
 N/A           ``WaveMatch()``              ``OpGroupNonUniformPartitionNV``
 Multiprefix   ``WaveMultiPrefixSum()``     ``OpGroupNonUniform*Add``           ``PartitionedExclusiveScanNV``
 Multiprefix   ``WaveMultiPrefixProduct()`` ``OpGroupNonUniform*Mul``           ``PartitionedExclusiveScanNV``
@@ -4016,6 +4019,11 @@ Multiprefix   ``WaveMultiPrefixBitOr()``   ``OpGroupNonUniformLogicalOr``      `
 Multiprefix   ``WaveMultiPrefixBitXor()``  ``OpGroupNonUniformLogicalXor``     ``PartitionedExclusiveScanNV``
 ============= ============================ =================================== ==============================
 
+``QuadAny`` and ``QuadAll`` will use the ``OpGroupNonUniformQuadAnyKHR`` and
+``OpGroupNonUniformQuadAllKHR`` instructions if the ``SPV_KHR_quad_control``
+extension is enabled. If it is not, they will fall back to constructing the
+value using multiple calls to ``OpGroupNonUniformQuadBroadcast``.
+
 The Implicit ``vk`` Namespace
 =============================
 

include/dxc/DXIL/DxilConstants.h

@@ -155,6 +155,11 @@ const float kMinMipLodBias = -16.0f;
 
 const unsigned kResRetStatusIndex = 4;
 
+/* <py::lines('OLOAD_DIMS-TEXT')>hctdb_instrhelp.get_max_oload_dims()</py>*/
+// OLOAD_DIMS-TEXT:BEGIN
+const unsigned kDxilMaxOloadDims = 2;
+// OLOAD_DIMS-TEXT:END
+
 enum class ComponentType : uint32_t {
   Invalid = 0,
   I1,
@@ -893,8 +898,11 @@ enum class OpCode : unsigned {
   GetDimensions = 72,   // gets texture size information
   RawBufferLoad = 139,  // reads from a raw buffer and structured buffer
   RawBufferStore = 140, // writes to a RWByteAddressBuffer or RWStructuredBuffer
-  TextureLoad = 66,     // reads texel data without any filtering or sampling
-  TextureStore = 67,    // reads texel data without any filtering or sampling
+  RawBufferVectorLoad = 303, // reads from a raw buffer and structured buffer
+  RawBufferVectorStore =
+      304,           // writes to a RWByteAddressBuffer or RWStructuredBuffer
+  TextureLoad = 66,  // reads texel data without any filtering or sampling
+  TextureStore = 67, // reads texel data without any filtering or sampling
   TextureStoreSample = 225, // stores texel data at specified sample index
 
   // Sampler Feedback
@@ -1039,7 +1047,7 @@ enum class OpCode : unsigned {
   NumOpCodes_Dxil_1_7 = 226,
   NumOpCodes_Dxil_1_8 = 258,
 
-  NumOpCodes = 303 // exclusive last value of enumeration
+  NumOpCodes = 305 // exclusive last value of enumeration
 };
 // OPCODE-ENUM:END
 
@@ -1273,6 +1281,8 @@ enum class OpCodeClass : unsigned {
   GetDimensions,
   RawBufferLoad,
   RawBufferStore,
+  RawBufferVectorLoad,
+  RawBufferVectorStore,
   TextureLoad,
   TextureStore,
   TextureStoreSample,
@@ -1351,7 +1361,7 @@ enum class OpCodeClass : unsigned {
   NumOpClasses_Dxil_1_7 = 153,
   NumOpClasses_Dxil_1_8 = 174,
 
-  NumOpClasses = 177 // exclusive last value of enumeration
+  NumOpClasses = 179 // exclusive last value of enumeration
 };
 // OPCODECLASS-ENUM:END
 
@@ -1410,6 +1420,12 @@ const unsigned kRawBufferLoadElementOffsetOpIdx = 3;
 const unsigned kRawBufferLoadMaskOpIdx = 4;
 const unsigned kRawBufferLoadAlignmentOpIdx = 5;
 
+// RawBufferVectorLoad.
+const unsigned kRawBufferVectorLoadHandleOpIdx = 1;
+const unsigned kRawBufferVectorLoadIndexOpIdx = 2;
+const unsigned kRawBufferVectorLoadElementOffsetOpIdx = 3;
+const unsigned kRawBufferVectorLoadAlignmentOpIdx = 4;
+
 // RawBufferStore
 const unsigned kRawBufferStoreHandleOpIdx = 1;
 const unsigned kRawBufferStoreIndexOpIdx = 2;
@@ -1419,7 +1435,14 @@ const unsigned kRawBufferStoreVal1OpIdx = 5;
 const unsigned kRawBufferStoreVal2OpIdx = 6;
 const unsigned kRawBufferStoreVal3OpIdx = 7;
 const unsigned kRawBufferStoreMaskOpIdx = 8;
-const unsigned kRawBufferStoreAlignmentOpIdx = 8;
+const unsigned kRawBufferStoreAlignmentOpIdx = 9;
+
+// RawBufferVectorStore
+const unsigned kRawBufferVectorStoreHandleOpIdx = 1;
+const unsigned kRawBufferVectorStoreIndexOpIdx = 2;
+const unsigned kRawBufferVectorStoreElementOffsetOpIdx = 3;
+const unsigned kRawBufferVectorStoreValOpIdx = 4;
+const unsigned kRawBufferVectorStoreAlignmentOpIdx = 5;
 
 // TextureStore.
 const unsigned kTextureStoreHandleOpIdx = 1;

include/dxc/DXIL/DxilInstructions.h

@@ -8923,5 +8923,98 @@ struct DxilInst_HitObject_MakeNop {
   // Metadata
   bool requiresUniformInputs() const { return false; }
 };
+
+/// This instruction reads from a raw buffer and structured buffer
+struct DxilInst_RawBufferVectorLoad {
+  llvm::Instruction *Instr;
+  // Construction and identification
+  DxilInst_RawBufferVectorLoad(llvm::Instruction *pInstr) : Instr(pInstr) {}
+  operator bool() const {
+    return hlsl::OP::IsDxilOpFuncCallInst(
+        Instr, hlsl::OP::OpCode::RawBufferVectorLoad);
+  }
+  // Validation support
+  bool isAllowed() const { return true; }
+  bool isArgumentListValid() const {
+    if (5 != llvm::dyn_cast<llvm::CallInst>(Instr)->getNumArgOperands())
+      return false;
+    return true;
+  }
+  // Metadata
+  bool requiresUniformInputs() const { return false; }
+  // Operand indexes
+  enum OperandIdx {
+    arg_buf = 1,
+    arg_index = 2,
+    arg_elementOffset = 3,
+    arg_alignment = 4,
+  };
+  // Accessors
+  llvm::Value *get_buf() const { return Instr->getOperand(1); }
+  void set_buf(llvm::Value *val) { Instr->setOperand(1, val); }
+  llvm::Value *get_index() const { return Instr->getOperand(2); }
+  void set_index(llvm::Value *val) { Instr->setOperand(2, val); }
+  llvm::Value *get_elementOffset() const { return Instr->getOperand(3); }
+  void set_elementOffset(llvm::Value *val) { Instr->setOperand(3, val); }
+  llvm::Value *get_alignment() const { return Instr->getOperand(4); }
+  void set_alignment(llvm::Value *val) { Instr->setOperand(4, val); }
+  int32_t get_alignment_val() const {
+    return (int32_t)(llvm::dyn_cast<llvm::ConstantInt>(Instr->getOperand(4))
+                         ->getZExtValue());
+  }
+  void set_alignment_val(int32_t val) {
+    Instr->setOperand(4, llvm::Constant::getIntegerValue(
+                             llvm::IntegerType::get(Instr->getContext(), 32),
+                             llvm::APInt(32, (uint64_t)val)));
+  }
+};
+
+/// This instruction writes to a RWByteAddressBuffer or RWStructuredBuffer
+struct DxilInst_RawBufferVectorStore {
+  llvm::Instruction *Instr;
+  // Construction and identification
+  DxilInst_RawBufferVectorStore(llvm::Instruction *pInstr) : Instr(pInstr) {}
+  operator bool() const {
+    return hlsl::OP::IsDxilOpFuncCallInst(
+        Instr, hlsl::OP::OpCode::RawBufferVectorStore);
+  }
+  // Validation support
+  bool isAllowed() const { return true; }
+  bool isArgumentListValid() const {
+    if (6 != llvm::dyn_cast<llvm::CallInst>(Instr)->getNumArgOperands())
+      return false;
+    return true;
+  }
+  // Metadata
+  bool requiresUniformInputs() const { return false; }
+  // Operand indexes
+  enum OperandIdx {
+    arg_uav = 1,
+    arg_index = 2,
+    arg_elementOffset = 3,
+    arg_value0 = 4,
+    arg_alignment = 5,
+  };
+  // Accessors
+  llvm::Value *get_uav() const { return Instr->getOperand(1); }
+  void set_uav(llvm::Value *val) { Instr->setOperand(1, val); }
+  llvm::Value *get_index() const { return Instr->getOperand(2); }
+  void set_index(llvm::Value *val) { Instr->setOperand(2, val); }
+  llvm::Value *get_elementOffset() const { return Instr->getOperand(3); }
+  void set_elementOffset(llvm::Value *val) { Instr->setOperand(3, val); }
+  llvm::Value *get_value0() const { return Instr->getOperand(4); }
+  void set_value0(llvm::Value *val) { Instr->setOperand(4, val); }
+  llvm::Value *get_alignment() const { return Instr->getOperand(5); }
+  void set_alignment(llvm::Value *val) { Instr->setOperand(5, val); }
+  int32_t get_alignment_val() const {
+    return (int32_t)(llvm::dyn_cast<llvm::ConstantInt>(Instr->getOperand(5))
+                         ->getZExtValue());
+  }
+  void set_alignment_val(int32_t val) {
+    Instr->setOperand(5, llvm::Constant::getIntegerValue(
+                             llvm::IntegerType::get(Instr->getContext(), 32),
+                             llvm::APInt(32, (uint64_t)val)));
+  }
+};
 // INSTR-HELPER:END
 } // namespace hlsl