Update llpc from commit 9570efcc (#2776)

* [Continuations] Fix tablegen in cmake
* Fix unstable barycentric
* Fix the OpDPdy
* Add the missing symbol usage information in ShaderModule
* [Continuations] Rewrite continuation stack lowering in CleanupContinuations
* Remove unnecessary buffer coherency settings
* [Continuations] Use lgc.rt.shaderstage metadata instead of analyzeShaderKinds
* Fix a type mismatch in reserved field
* Fix the dir create name error.
* Fix integer alpha
* Fix dbShaderControl for color export shader
* Disable uber-fetch lit test
* llpcfe dxil: Fix some arithmetic instruction failure in HLK
* [Continuations] Clear payload with poison
* [Continuations] Replace lgc.rt.shader.index calls.
* [Continuations] Insert lgc.cps.jump and lgc.cps.await operations in LowerRaytracingPipelinePass.
* [Continuations] Enable Continuations transform in LLPC
* [Continuations] lgc/cps: use a backend-friendlier lowering of as.continuation.reference
* lgc: add a buffer flag to convert dx buffer descritpor for oob
* Add support for SPV_KHR_float_controls2
* [Continuations] Lower GetCurrentFuncAddr and GetShaderKind in lgc
* [Continuations] Merge the lgc.* dialects into the Continuations CMake project.
* Disable CPS failing lit test
* Re-enable raytracing tests
* Bump CMake minimum version to 3.21
* [ContinuationsUtil]- Remove Operations.h header

CMakeLists.txt

@@ -23,7 +23,7 @@
  #
  #######################################################################################################################
 
-cmake_minimum_required(VERSION 3.13.4)
+cmake_minimum_required(VERSION 3.21)
 
 project(VKGC LANGUAGES C CXX)
 
@@ -206,7 +206,8 @@ if(CMAKE_CXX_COMPILER_ID STREQUAL "MSVC")
             endif()
             set_property(TARGET check-amdllpc PROPERTY FOLDER "LLPC Tests")
             set_property(TARGET check-amdllpc-units PROPERTY FOLDER "LLPC Tests")
-            set_property(TARGET check-lgccps-units PROPERTY FOLDER "LgcCps Tests")
+            set_property(TARGET check-continuations PROPERTY FOLDER "Continuations Tests")
+            set_property(TARGET check-continuations-units PROPERTY FOLDER "Continuations Tests")
             set_property(TARGET check-lgc-units PROPERTY FOLDER "LGC Tests")
         endif()
     endif()

cmake/continuations.cmake

@@ -25,19 +25,16 @@
 
 set(LLPC_SOURCE_DIR "${CMAKE_CURRENT_LIST_DIR}/..")
 
-# Function to add continuations and its dependencies as LLVM external projects.
+# Macro to add continuations and its dependencies as LLVM external projects.
 # This appends the project names to LLVM_EXTERNAL_PROJECTS and sets each LLVM_EXTERNAL_*_SOURCE_DIR,
 # all in the caller's scope.
-function(add_continuations_projects)
-  if (NOT "${LLVM_EXTERNAL_CONTINUATIONS_SOURCE_DIR}")
-    if (NOT "${LLVM_EXTERNAL_LLVM_DIALECTS_SOURCE_DIR}")
+macro(add_continuations_projects)
+  if (NOT continuations IN_LIST LLVM_EXTERNAL_PROJECTS)
+    if (NOT llvm_dialects IN_LIST LLVM_EXTERNAL_PROJECTS)
       list(APPEND LLVM_EXTERNAL_PROJECTS llvm_dialects)
-      set(LLVM_EXTERNAL_LLVM_DIALECTS_SOURCE_DIR "${LLPC_SOURCE_DIR}/imported/llvm-dialects" PARENT_SCOPE)
+      set(LLVM_EXTERNAL_LLVM_DIALECTS_SOURCE_DIR "${LLPC_SOURCE_DIR}/imported/llvm-dialects")
     endif()
-    list(APPEND LLVM_EXTERNAL_PROJECTS lgccps lgcrt continuations)
-    set(LLVM_EXTERNAL_CONTINUATIONS_SOURCE_DIR "${LLPC_SOURCE_DIR}/shared/continuations" PARENT_SCOPE)
-    set(LLVM_EXTERNAL_LGCRT_SOURCE_DIR "${LLPC_SOURCE_DIR}/shared/lgcrt" PARENT_SCOPE)
-    set(LLVM_EXTERNAL_LGCCPS_SOURCE_DIR "${LLPC_SOURCE_DIR}/shared/lgccps" PARENT_SCOPE)
-    set(LLVM_EXTERNAL_PROJECTS "${LLVM_EXTERNAL_PROJECTS}" PARENT_SCOPE)
+    list(APPEND LLVM_EXTERNAL_PROJECTS Continuations)
+    set(LLVM_EXTERNAL_CONTINUATIONS_SOURCE_DIR "${LLPC_SOURCE_DIR}/shared/continuations")
   endif()
-endfunction()
+endmacro()

cmake/lgc.cmake

@@ -25,20 +25,19 @@
 
 set(LLPC_SOURCE_DIR "${CMAKE_CURRENT_LIST_DIR}/..")
 
-# Function to add LGC and its dependencies as LLVM external projects.
+include("${LLPC_SOURCE_DIR}/cmake/continuations.cmake")
+
+# Macro to add LGC and its dependencies as LLVM external projects.
 # This appends the project names to LLVM_EXTERNAL_PROJECTS and sets each LLVM_EXTERNAL_*_SOURCE_DIR,
 # all in the caller's scope.
-function(add_lgc_projects)
-    if (NOT "${LLVM_EXTERNAL_LGC_SOURCE_DIR}")
-        if (NOT "${LLVM_EXTERNAL_LLVM_DIALECTS_SOURCE_DIR}")
+macro(add_lgc_projects)
+    add_continuations_projects()
+    if (NOT lgc IN_LIST LLVM_EXTERNAL_PROJECTS)
+        if (NOT llvm_dialects IN_LIST LLVM_EXTERNAL_PROJECTS)
             list(APPEND LLVM_EXTERNAL_PROJECTS llvm_dialects)
-            set(LLVM_EXTERNAL_LLVM_DIALECTS_SOURCE_DIR "${LLPC_SOURCE_DIR}/imported/llvm-dialects" PARENT_SCOPE)
+            set(LLVM_EXTERNAL_LLVM_DIALECTS_SOURCE_DIR "${LLPC_SOURCE_DIR}/imported/llvm-dialects")
         endif()
-        list(APPEND LLVM_EXTERNAL_PROJECTS LgcCps LgcRt Continuations lgc)
-        set(LLVM_EXTERNAL_PROJECTS "${LLVM_EXTERNAL_PROJECTS}" PARENT_SCOPE)
-        set(LLVM_EXTERNAL_LGCCPS_SOURCE_DIR "${LLPC_SOURCE_DIR}/shared/lgccps" PARENT_SCOPE)
-        set(LLVM_EXTERNAL_LGCRT_SOURCE_DIR "${LLPC_SOURCE_DIR}/shared/lgcrt" PARENT_SCOPE)
-        set(LLVM_EXTERNAL_CONTINUATIONS_SOURCE_DIR "${LLPC_SOURCE_DIR}/shared/continuations" PARENT_SCOPE)
-        set(LLVM_EXTERNAL_LGC_SOURCE_DIR "${LLPC_SOURCE_DIR}/lgc" PARENT_SCOPE)
+        list(APPEND LLVM_EXTERNAL_PROJECTS lgc)
+        set(LLVM_EXTERNAL_LGC_SOURCE_DIR "${LLPC_SOURCE_DIR}/lgc")
     endif()
-endfunction()
+endmacro()

include/vkgcDefs.h

@@ -49,7 +49,7 @@
 #define LLPC_INTERFACE_MAJOR_VERSION 69
 
 /// LLPC minor interface version.
-#define LLPC_INTERFACE_MINOR_VERSION 0
+#define LLPC_INTERFACE_MINOR_VERSION 2
 
 #ifndef LLPC_CLIENT_INTERFACE_MAJOR_VERSION
 #error LLPC client version is not defined
@@ -80,7 +80,9 @@
 //  %Version History
 //  | %Version | Change Description                                                                                    |
 //  | -------- | ----------------------------------------------------------------------------------------------------- |
-//  |     69.0 | Add enablePrimGeneratedQuery to PipelineOptions                                                       |
+//  |     69.2 | Add enablePrimGeneratedQuery to PipelineOptions                                                       |
+//  |     69.1 | Add useBarycentric to ShaderModuleUsage                                                               |
+//  |     69.0 | Enable continuations transform in LLPC                                                                |
 //  |     68.0 | Remove ICache *cache in all PipelineBuildInfo                                                         |
 //  |     67.0 | Modify the uber fetch shader. Adds locationMask(64bit) at the beginning of uber fetch shader internal |
 //  |          | buffer which flags whether the related attribute data is valid.                                       |
@@ -686,6 +688,11 @@ struct ShaderModuleUsage {
   bool pixelCenterInteger;     ///< Whether pixel coord is Integer
   bool useGenericBuiltIn;      ///< Whether to use builtIn inputs that include gl_PointCoord, gl_PrimitiveId,
                                ///  gl_Layer, gl_ClipDistance or gl_CullDistance.
+  bool enableXfb;              ///< Whether transform feedback is enabled
+  unsigned localSizeX;         ///< Compute shader work-group size in the X dimension
+  unsigned localSizeY;         ///< Compute shader work-group size in the Y dimension
+  unsigned localSizeZ;         ///< Compute shader work-group size in the Z dimension
+  bool useBarycentric;         ///< Whether to use gl_BarycentricXX or pervertexEXT decoration
 };
 
 /// Represents common part of shader module data
@@ -862,7 +869,7 @@ struct PipelineShaderOptions {
   /// Threshold to use for loops with "DontUnroll" hint (0 = use llvm.loop.unroll.disable).
   unsigned dontUnrollHintThreshold;
 
-  /// Whether fastmath contract could be disabled on Dot operations.
+  /// Whether fast math contract could be disabled on Dot operations.
   bool noContractOpDot;
 
   /// The enabled fast math flags (0 = depends on input language).
@@ -1170,9 +1177,13 @@ enum DispatchDimSwizzleMode : unsigned {
 };
 
 enum class LlpcRaytracingMode : unsigned {
-  None = 0,      // Not goto any raytracing compiling path
-  Legacy,        // LLpc Legacy compiling path
-  Gpurt2,        // Raytracing lowering at the end of spirvLower.
+  None = 0, // Not goto any raytracing compiling path
+  Legacy,   // LLpc Legacy compiling path
+#if LLPC_CLIENT_INTERFACE_MAJOR_VERSION < 69
+  Gpurt2, // Raytracing lowering at the end of spirvLower.
+#else
+  Continufy, // Enable continuation with the continufy path.
+#endif
   Continuations, // Enable continuation in the new raytracing path
 };
 

lgc/CMakeLists.txt

@@ -53,7 +53,7 @@ add_llvm_library(LLVMlgc LINK_COMPONENTS
     Vectorize
 )
 
-target_link_libraries(LLVMlgc PUBLIC llvm_dialects)
+target_link_libraries(LLVMlgc PUBLIC LLVMContinuations)
 
 ### Cached Project Options #############################################################################################
 option(LLPC_BUILD_NAVI12 "LLPC support for NAVI12?" ON)
@@ -91,14 +91,6 @@ tablegen(LGC state/LgcDialect.cpp.inc -gen-dialect-defs --dialect lgc
     EXTRA_INCLUDES ${CMAKE_CURRENT_SOURCE_DIR}/../imported/llvm-dialects/include
     )
 
-set(LLVM_TARGET_DEFINITIONS interface/lgc/GpurtDialect.td)
-tablegen(LGC interface/lgc/GpurtDialect.h.inc -gen-dialect-decls --dialect lgc.gpurt
-    EXTRA_INCLUDES ${CMAKE_CURRENT_SOURCE_DIR}/../imported/llvm-dialects/include
-    )
-tablegen(LGC state/GpurtDialect.cpp.inc -gen-dialect-defs --dialect lgc.gpurt
-    EXTRA_INCLUDES ${CMAKE_CURRENT_SOURCE_DIR}/../imported/llvm-dialects/include
-    )
-
 add_public_tablegen_target(LgcDialectTableGen)
 
 add_dependencies(LLVMlgc LgcDialectTableGen)
@@ -254,7 +246,6 @@ target_sources(LLVMlgc PRIVATE
     state/Compiler.cpp
     state/LgcContext.cpp
     state/LgcDialect.cpp
-    state/GpurtDialect.cpp
     state/PalMetadata.cpp
     state/PassManagerCache.cpp
     state/PipelineShaders.cpp
@@ -288,4 +279,4 @@ if (LLPC_BUILD_TESTS)
   add_subdirectory(unittests)
 endif()
 
-target_link_libraries(LLVMlgc PRIVATE LLVMLgcCps LLVMLgcRt)
+target_link_libraries(LLVMlgc PRIVATE LLVMContinuations)

lgc/builder/ArithBuilder.cpp

@@ -407,7 +407,8 @@ Value *BuilderImpl::CreateATan(Value *yOverX, const Twine &instName) {
   Value *outsideBound = CreateSelect(CreateFCmpOGT(absX, one), one, zero);
   result = CreateFMul(outsideBound, result);
   result = CreateFAdd(partialResult, result);
-  return CreateFMul(result, CreateFSign(yOverX));
+  result = CreateFMul(result, CreateFSign(yOverX));
+  return CreateSelect(CreateIsNaN(yOverX), ConstantFP::getNaN(yOverX->getType()), result);
 }
 
 // =====================================================================================================================
@@ -452,9 +453,9 @@ Value *BuilderImpl::CreateATan2(Value *y, Value *x, const Twine &instName) {
   Value *result = CreateATan(yOverX);
   Value *addP1 = CreateFAdd(result, p1);
   result = CreateSelect(CreateFCmpOLT(x, zero), addP1, result);
-  result = CreateSelect(CreateFCmpONE(x, zero), result, p0);
+  result = CreateSelect(CreateFCmpUNE(x, zero), result, p0);
   Value *zeroOrPi = CreateSelect(CreateFCmpOGT(x, zero), zero, getPi(x->getType()));
-  result = CreateSelect(CreateFCmpONE(y, zero), result, zeroOrPi, instName);
+  result = CreateSelect(CreateFCmpUNE(y, zero), result, zeroOrPi, instName);
   return result;
 }
 

lgc/builder/BuilderRecorder.cpp

@@ -1814,7 +1814,7 @@ Value *Builder::CreateSubgroupShuffleDown(Value *const value, Value *const offse
 // @param value : The value to read from the chosen rotated lane to all active lanes.
 // @param delta : The delta/offset added to lane id.
 // @param clusterSize : The cluster size if exists.
-// @param instName : Name to give instruction.
+// @param instName : Name to give final instruction.
 Value *Builder::CreateSubgroupRotate(Value *const value, Value *const delta, Value *const clusterSize,
                                      const Twine &instName) {
   return record(BuilderOpcode::SubgroupRotate, value->getType(), {value, delta, clusterSize}, instName);

lgc/builder/DescBuilder.cpp

@@ -175,6 +175,22 @@ Value *BuilderImpl::CreateBufferDesc(uint64_t descSet, unsigned binding, Value *
     descPtr = CreateBitCast(descPtr, getDescPtrTy(resType));
     // Load the descriptor.
     desc = CreateLoad(getDescTy(resType), descPtr);
+
+    // Force convert the buffer view to raw view.
+    if (flags & BufferFlagForceRawView) {
+      Value *desc1 = CreateExtractElement(desc, 1);
+      Value *desc2 = CreateExtractElement(desc, 2);
+      Value *desc3 = CreateExtractElement(desc, 3);
+      // stride is 14 bits in dword1[29:16]
+      Value *stride = CreateAnd(CreateLShr(desc1, getInt32(16)), getInt32(0x3fff));
+      stride = CreateBinaryIntrinsic(Intrinsic::smax, stride, getInt32(1));
+      // set srd with new stride = 0 and new num_record = stride * num_record, num_record is dword2[31:0]
+      desc = CreateInsertElement(desc, CreateAnd(desc1, getInt32(0xc000ffff)), 1);
+      desc = CreateInsertElement(desc, CreateMul(stride, desc2), 2);
+      // gfx10 and gfx11 have oob fields with 2 bits in dword3[29:28] here force to set to 3 as OOB_COMPLETE mode.
+      if (getPipelineState()->getTargetInfo().getGfxIpVersion().major >= 10)
+        desc = CreateInsertElement(desc, CreateOr(desc3, getInt32(0x30000000)), 3);
+    }
   }
 
   if (node && (node->concreteType == ResourceNodeType::DescriptorBufferCompact ||
@@ -257,7 +273,8 @@ Value *BuilderImpl::CreateGetDescPtr(ResourceNodeType concreteType, ResourceNode
     // to that. For a simple non-variably-indexed immutable sampler not passed through a function call
     // or phi node, we rely on subsequent LLVM optimizations promoting the value back to a constant.
     StringRef startGlobalName = lgcName::ImmutableSamplerGlobal;
-    std::string globalName = (startGlobalName + Twine(node->set) + "_" + Twine(node->binding)).str();
+    std::string globalName =
+        (startGlobalName + Twine(node->set) + "_" + Twine(node->binding) + "_" + Twine(node->visibility)).str();
     Module *module = GetInsertPoint()->getModule();
     descPtr = module->getGlobalVariable(globalName, /*AllowInternal=*/true);
     if (!descPtr) {