[HLSL] Implicit resource binding for cbuffers

clang/include/clang/Basic/Attr.td

@@ -4789,6 +4789,7 @@ def HLSLResourceBinding: InheritableAttr {
       RegisterType RegType;
       std::optional<unsigned> SlotNumber;
       unsigned SpaceNumber;
+      std::optional<unsigned> ImplicitBindingOrderID;
 
   public:
       void setBinding(RegisterType RT, std::optional<unsigned> SlotNum, unsigned SpaceNum) {
@@ -4810,6 +4811,16 @@ def HLSLResourceBinding: InheritableAttr {
       unsigned getSpaceNumber() const {
         return SpaceNumber;
       }
+      void setImplicitBindingOrderID(uint32_t Value) {
+        ImplicitBindingOrderID = Value;
+      }
+      bool hasImplicitBindingOrderID() const { 
+        return ImplicitBindingOrderID.has_value();
+      }
+      uint32_t getImplicitBindingOrderID() const {
+        assert(hasImplicitBindingOrderID() && "attribute does not have implicit binding order id");
+        return ImplicitBindingOrderID.value();
+      }
   }];
 }
 

clang/lib/CodeGen/CGHLSLRuntime.cpp

@@ -23,6 +23,7 @@
 #include "clang/AST/Type.h"
 #include "clang/Basic/TargetOptions.h"
 #include "llvm/ADT/SmallVector.h"
+#include "llvm/IR/Constants.h"
 #include "llvm/IR/DerivedTypes.h"
 #include "llvm/IR/GlobalVariable.h"
 #include "llvm/IR/LLVMContext.h"
@@ -42,8 +43,8 @@ using namespace llvm;
 using llvm::hlsl::CBufferRowSizeInBytes;
 
 static void initializeBufferFromBinding(CodeGenModule &CGM,
-                                        llvm::GlobalVariable *GV, unsigned Slot,
-                                        unsigned Space);
+                                        llvm::GlobalVariable *GV,
+                                        HLSLResourceBindingAttr *RBA);
 
 namespace {
 
@@ -257,13 +258,10 @@ void CGHLSLRuntime::addBuffer(const HLSLBufferDecl *BufDecl) {
   emitBufferGlobalsAndMetadata(BufDecl, BufGV);
 
   // Initialize cbuffer from binding (implicit or explicit)
-  const HLSLResourceBindingAttr *RBA =
-      BufDecl->getAttr<HLSLResourceBindingAttr>();
-  // FIXME: handle implicit binding if no binding attribute is found
-  // (llvm/llvm-project#110722)
-  if (RBA && RBA->hasRegisterSlot())
-    initializeBufferFromBinding(CGM, BufGV, RBA->getSlotNumber(),
-                                RBA->getSpaceNumber());
+  HLSLResourceBindingAttr *RBA = BufDecl->getAttr<HLSLResourceBindingAttr>();
+  assert(RBA &&
+         "cbuffer/tbuffer should always have resource binding attribute");
+  initializeBufferFromBinding(CGM, BufGV, RBA);
 }
 
 llvm::TargetExtType *
@@ -539,19 +537,29 @@ static void initializeBuffer(CodeGenModule &CGM, llvm::GlobalVariable *GV,
 }
 
 static void initializeBufferFromBinding(CodeGenModule &CGM,
-                                        llvm::GlobalVariable *GV, unsigned Slot,
-                                        unsigned Space) {
+                                        llvm::GlobalVariable *GV,
+                                        HLSLResourceBindingAttr *RBA) {
   llvm::Type *Int1Ty = llvm::Type::getInt1Ty(CGM.getLLVMContext());
-  llvm::Value *Args[] = {
-      llvm::ConstantInt::get(CGM.IntTy, Space), /* reg_space */
-      llvm::ConstantInt::get(CGM.IntTy, Slot),  /* lower_bound */
-      llvm::ConstantInt::get(CGM.IntTy, 1),     /* range_size */
-      llvm::ConstantInt::get(CGM.IntTy, 0),     /* index */
-      llvm::ConstantInt::get(Int1Ty, false)     /* non-uniform */
-  };
-  initializeBuffer(CGM, GV,
-                   CGM.getHLSLRuntime().getCreateHandleFromBindingIntrinsic(),
-                   Args);
+  auto *NonUniform = llvm::ConstantInt::get(Int1Ty, false);
+  auto *Index = llvm::ConstantInt::get(CGM.IntTy, 0);
+  auto *RangeSize = llvm::ConstantInt::get(CGM.IntTy, 1);
+  auto *Space =
+      llvm::ConstantInt::get(CGM.IntTy, RBA ? RBA->getSpaceNumber() : 0);
+
+  if (RBA->hasRegisterSlot()) {
+    auto *RegSlot = llvm::ConstantInt::get(CGM.IntTy, RBA->getSlotNumber());
+    Intrinsic::ID Intr =
+        CGM.getHLSLRuntime().getCreateHandleFromBindingIntrinsic();
+    initializeBuffer(CGM, GV, Intr,
+                     {Space, RegSlot, RangeSize, Index, NonUniform});
+  } else {
+    auto *OrderID =
+        llvm::ConstantInt::get(CGM.IntTy, RBA->getImplicitBindingOrderID());
+    Intrinsic::ID Intr =
+        CGM.getHLSLRuntime().getCreateHandleFromImplicitBindingIntrinsic();
+    initializeBuffer(CGM, GV, Intr,
+                     {OrderID, Space, RangeSize, Index, NonUniform});
+  }
 }
 
 llvm::Instruction *CGHLSLRuntime::getConvergenceToken(BasicBlock &BB) {

clang/lib/Sema/SemaHLSL.cpp

@@ -537,6 +537,18 @@ void createHostLayoutStructForBuffer(Sema &S, HLSLBufferDecl *BufDecl) {
   BufDecl->addLayoutStruct(LS);
 }
 
+static void addImplicitBindingAttrToBuffer(Sema &S, HLSLBufferDecl *BufDecl,
+                                           uint32_t ImplicitBindingOrderID) {
+  RegisterType RT =
+      BufDecl->isCBuffer() ? RegisterType::CBuffer : RegisterType::SRV;
+  auto *Attr =
+      HLSLResourceBindingAttr::CreateImplicit(S.getASTContext(), "", "0", {});
+  std::optional<unsigned> RegSlot;
+  Attr->setBinding(RT, RegSlot, 0);
+  Attr->setImplicitBindingOrderID(ImplicitBindingOrderID);
+  BufDecl->addAttr(Attr);
+}
+
 // Handle end of cbuffer/tbuffer declaration
 void SemaHLSL::ActOnFinishBuffer(Decl *Dcl, SourceLocation RBrace) {
   auto *BufDecl = cast<HLSLBufferDecl>(Dcl);
@@ -547,9 +559,17 @@ void SemaHLSL::ActOnFinishBuffer(Decl *Dcl, SourceLocation RBrace) {
   // create buffer layout struct
   createHostLayoutStructForBuffer(SemaRef, BufDecl);
 
-  if (std::none_of(Dcl->attr_begin(), Dcl->attr_end(),
-                   [](Attr *A) { return isa<HLSLResourceBindingAttr>(A); }))
+  HLSLResourceBindingAttr *RBA = Dcl->getAttr<HLSLResourceBindingAttr>();
+  if (!RBA || !RBA->hasRegisterSlot()) {
     SemaRef.Diag(Dcl->getLocation(), diag::warn_hlsl_implicit_binding);
+    // Use HLSLResourceBindingAttr to transfer implicit binding order_ID
+    // to codegen. If it does not exist, create an implicit attribute.
+    uint32_t OrderID = getNextImplicitBindingOrderID();
+    if (RBA)
+      RBA->setImplicitBindingOrderID(OrderID);
+    else
+      addImplicitBindingAttrToBuffer(SemaRef, BufDecl, OrderID);
+  }
 
   SemaRef.PopDeclContext();
 }
@@ -1999,6 +2019,8 @@ void SemaHLSL::ActOnEndOfTranslationUnit(TranslationUnitDecl *TU) {
     HLSLBufferDecl *DefaultCBuffer = HLSLBufferDecl::CreateDefaultCBuffer(
         SemaRef.getASTContext(), SemaRef.getCurLexicalContext(),
         DefaultCBufferDecls);
+    addImplicitBindingAttrToBuffer(SemaRef, DefaultCBuffer,
+                                   getNextImplicitBindingOrderID());
     SemaRef.getCurLexicalContext()->addDecl(DefaultCBuffer);
     createHostLayoutStructForBuffer(SemaRef, DefaultCBuffer);
 

clang/test/AST/HLSL/ast-dump-comment-cbuffer.hlsl

@@ -21,6 +21,7 @@ cbuffer A {
 
 // AST: HLSLBufferDecl {{.*}} line:11:9 cbuffer A
 // AST-NEXT: HLSLResourceClassAttr {{.*}} Implicit CBuffer
+// AST-NEXT: HLSLResourceBindingAttr {{.*}} Implicit "" "0"
 // AST-NEXT: FullComment
 // AST-NEXT: ParagraphComment
 // AST-NEXT: TextComment {{.*}} Text=" CBuffer decl."

clang/test/AST/HLSL/packoffset.hlsl

@@ -5,6 +5,7 @@
 cbuffer A
 {
     // CHECK-NEXT:-HLSLResourceClassAttr {{.*}} Implicit CBuffer
+    // CHECK-NEXT: HLSLResourceBindingAttr {{.*}} Implicit "" "0"
     // CHECK-NEXT: VarDecl {{.*}} A1 'hlsl_constant float4'
     // CHECK-NEXT: HLSLPackOffsetAttr {{.*}} 0 0
     float4 A1 : packoffset(c);

clang/test/AST/HLSL/pch_hlsl_buffer.hlsl

@@ -16,12 +16,14 @@ float foo() {
 // Make sure cbuffer/tbuffer works for PCH.
 // CHECK: HLSLBufferDecl {{.*}} line:{{[0-9]+}}:9 imported <undeserialized declarations> cbuffer A
 // CHECK-NEXT: HLSLResourceClassAttr {{.*}} Implicit CBuffer
+// CHECK-NEXT: HLSLResourceBindingAttr {{.*}} Implicit "" "0"
 // CHECK-NEXT: VarDecl 0x[[A:[0-9a-f]+]] {{.*}} imported used a 'hlsl_constant float'
 // CHECK-NEXT: CXXRecordDecl {{.*}} imported implicit <undeserialized declarations> struct __cblayout_A definition
 // CHECK: FieldDecl {{.*}} imported a 'float'
 
 // CHECK: HLSLBufferDecl {{.*}} line:{{[0-9]+}}:9 imported <undeserialized declarations> tbuffer B
 // CHECK-NEXT: HLSLResourceClassAttr {{.*}} Implicit SRV
+// CHECK-NEXT: HLSLResourceBindingAttr {{.*}} Implicit "" "0"
 // CHECK-NEXT: VarDecl 0x[[B:[0-9a-f]+]] {{.*}} imported used b 'hlsl_constant float'
 // CHECK-NEXT: CXXRecordDecl 0x{{[0-9a-f]+}} {{.*}} imported implicit <undeserialized declarations> struct __cblayout_B definition
 // CHECK: FieldDecl 0x{{[0-9a-f]+}} {{.*}} imported b 'float'

clang/test/CodeGenHLSL/GlobalConstructorFunction.hlsl

@@ -27,6 +27,7 @@ void main(unsigned GI : SV_GroupIndex) {}
 // Verify function constructors are emitted
 // NOINLINE-NEXT:   call void @_Z13call_me_firstv()
 // NOINLINE-NEXT:   call void @_Z12then_call_mev()
+// NOINLINE-NEXT:   call void @_GLOBAL__sub_I_GlobalConstructorFunction.hlsl()
 // NOINLINE-NEXT:   %0 = call i32 @llvm.dx.flattened.thread.id.in.group()
 // NOINLINE-NEXT:   call void @_Z4mainj(i32 %0)
 // NOINLINE-NEXT:   call void @_Z12call_me_lastv(
@@ -36,6 +37,9 @@ void main(unsigned GI : SV_GroupIndex) {}
 // INLINE-NEXT:   alloca
 // INLINE-NEXT:   store i32 12
 // INLINE-NEXT:   store i32 13
+// INLINE-NEXT:   %[[HANDLE:.*]] = call target("dx.CBuffer", target("dx.Layout", %"__cblayout_$Globals", 4, 0))
+// INLINE-NEXT-SAME: @"llvm.dx.resource.handlefromimplicitbinding.tdx.CBuffer_tdx.Layout_s___cblayout_$Globalss_4_0tt"(i32 0, i32 0, i32 1, i32 0, i1 false)
+// INLINE-NEXT:   store target("dx.CBuffer", target("dx.Layout", %"__cblayout_$Globals", 4, 0)) %[[HANDLE]], ptr @"$Globals.cb", align 4
 // INLINE-NEXT:   %0 = call i32 @llvm.dx.flattened.thread.id.in.group()
 // INLINE-NEXT:   store i32 %
 // INLINE-NEXT:   store i32 0

clang/test/CodeGenHLSL/cbuffer.hlsl

@@ -102,7 +102,7 @@ typedef uint32_t4 uint32_t8[2];
 typedef uint4 T1;
 typedef T1 T2[2]; // check a double typedef
 
-cbuffer CBTypedefArray {
+cbuffer CBTypedefArray : register(space2) {
   uint32_t8 t1[2];
   T2 t2[2];
 }
@@ -268,16 +268,64 @@ cbuffer CB_C {
 
 // CHECK: define internal void @_init_buffer_CBScalars.cb()
 // CHECK-NEXT: entry:
-// CHECK-NEXT: %[[HANDLE1:.*]] = call target("dx.CBuffer", target("dx.Layout", %__cblayout_CBScalars, 56, 0, 8, 16, 24, 32, 36, 40, 48))
+// CHECK-NEXT: %CBScalars.cb_h = call target("dx.CBuffer", target("dx.Layout", %__cblayout_CBScalars, 56, 0, 8, 16, 24, 32, 36, 40, 48))
 // CHECK-SAME: @llvm.dx.resource.handlefrombinding.tdx.CBuffer_tdx.Layout_s___cblayout_CBScalarss_56_0_8_16_24_32_36_40_48tt(i32 5, i32 1, i32 1, i32 0, i1 false)
 // CHECK-NEXT: store target("dx.CBuffer", target("dx.Layout", %__cblayout_CBScalars, 56, 0, 8, 16, 24, 32, 36, 40, 48)) %CBScalars.cb_h, ptr @CBScalars.cb, align 4
 
+// CHECK: define internal void @_init_buffer_CBVectors.cb()
+// CHECK-NEXT: entry:
+// CHECK-NEXT: %CBVectors.cb_h = call target("dx.CBuffer", target("dx.Layout", %__cblayout_CBVectors, 136, 0, 16, 40, 48, 80, 96, 112))
+// CHECK-SAME: @llvm.dx.resource.handlefromimplicitbinding.tdx.CBuffer_tdx.Layout_s___cblayout_CBVectorss_136_0_16_40_48_80_96_112tt(i32 0, i32 0, i32 1, i32 0, i1 false)
+// CHECK-NEXT: store target("dx.CBuffer", target("dx.Layout", %__cblayout_CBVectors, 136, 0, 16, 40, 48, 80, 96, 112)) %CBVectors.cb_h, ptr @CBVectors.cb, align 4
+
 // CHECK: define internal void @_init_buffer_CBArrays.cb()
 // CHECK-NEXT: entry:
-// CHECK-NEXT: %[[HANDLE2:.*]] = call target("dx.CBuffer", target("dx.Layout", %__cblayout_CBArrays, 708, 0, 48, 112, 176, 224, 608, 624, 656))
+// CHECK-NEXT: %CBArrays.cb_h = call target("dx.CBuffer", target("dx.Layout", %__cblayout_CBArrays, 708, 0, 48, 112, 176, 224, 608, 624, 656))
 // CHECK-SAME: @llvm.dx.resource.handlefrombinding.tdx.CBuffer_tdx.Layout_s___cblayout_CBArrayss_708_0_48_112_176_224_608_624_656tt(i32 0, i32 2, i32 1, i32 0, i1 false)
 // CHECK-NEXT: store target("dx.CBuffer", target("dx.Layout", %__cblayout_CBArrays, 708, 0, 48, 112, 176, 224, 608, 624, 656)) %CBArrays.cb_h, ptr @CBArrays.cb, align 4
 
+// CHECK: define internal void @_init_buffer_CBTypedefArray.cb()
+// CHECK-NEXT: entry:
+// CHECK-NEXT: %CBTypedefArray.cb_h = call target("dx.CBuffer", target("dx.Layout", %__cblayout_CBTypedefArray, 128, 0, 64))
+// CHECK-SAME: @llvm.dx.resource.handlefromimplicitbinding.tdx.CBuffer_tdx.Layout_s___cblayout_CBTypedefArrays_128_0_64tt(i32 1, i32 2, i32 1, i32 0, i1 false)
+// CHECK-NEXT: store target("dx.CBuffer", target("dx.Layout", %__cblayout_CBTypedefArray, 128, 0, 64)) %CBTypedefArray.cb_h, ptr @CBTypedefArray.cb, align 4
+
+// CHECK: define internal void @_init_buffer_CBStructs.cb()
+// CHECK-NEXT: entry:
+// CHECK-NEXT:   %CBStructs.cb_h = call target("dx.CBuffer", target("dx.Layout", %__cblayout_CBStructs, 246, 0, 16, 32, 64, 144, 238, 240))
+// CHECK-SAME: @llvm.dx.resource.handlefromimplicitbinding.tdx.CBuffer_tdx.Layout_s___cblayout_CBStructss_246_0_16_32_64_144_238_240tt(i32 2, i32 0, i32 1, i32 0, i1 false)
+// CHECK-NEXT:   store target("dx.CBuffer", target("dx.Layout", %__cblayout_CBStructs, 246, 0, 16, 32, 64, 144, 238, 240)) %CBStructs.cb_h, ptr @CBStructs.cb, align 4
+
+// CHECK: define internal void @_init_buffer_CBClasses.cb()
+// CHECK-NEXT: entry:
+// CHECK-NEXT: %CBClasses.cb_h = call target("dx.CBuffer", target("dx.Layout", %__cblayout_CBClasses, 260, 0, 16, 32, 112))
+// CHECK-SAME: @llvm.dx.resource.handlefromimplicitbinding.tdx.CBuffer_tdx.Layout_s___cblayout_CBClassess_260_0_16_32_112tt(i32 3, i32 0, i32 1, i32 0, i1 false)
+// CHECK-NEXT: store target("dx.CBuffer", target("dx.Layout", %__cblayout_CBClasses, 260, 0, 16, 32, 112)) %CBClasses.cb_h, ptr @CBClasses.cb, align 4
+
+// CHECK: define internal void @_init_buffer_CBMix.cb()
+// CHECK-NEXT: entry:
+// CHECK-NEXT: %CBMix.cb_h = call target("dx.CBuffer", target("dx.Layout", %__cblayout_CBMix, 170, 0, 24, 32, 120, 128, 136, 144, 152, 160, 168))
+// CHECK-SAME: @llvm.dx.resource.handlefromimplicitbinding.tdx.CBuffer_tdx.Layout_s___cblayout_CBMixs_170_0_24_32_120_128_136_144_152_160_168tt(i32 4, i32 0, i32 1, i32 0, i1 false)
+// CHECK-NEXT: store target("dx.CBuffer", target("dx.Layout", %__cblayout_CBMix, 170, 0, 24, 32, 120, 128, 136, 144, 152, 160, 168)) %CBMix.cb_h, ptr @CBMix.cb, align 4
+
+// CHECK: define internal void @_init_buffer_CB_A.cb()
+// CHECK-NEXT: entry:
+// CHECK-NEXT: %CB_A.cb_h = call target("dx.CBuffer", target("dx.Layout", %__cblayout_CB_A, 188, 0, 32, 76, 80, 120, 128, 144, 160, 182))
+// CHECK-SAME: @llvm.dx.resource.handlefromimplicitbinding.tdx.CBuffer_tdx.Layout_s___cblayout_CB_As_188_0_32_76_80_120_128_144_160_182tt(i32 5, i32 0, i32 1, i32 0, i1 false)
+// CHECK-NEXT: store target("dx.CBuffer", target("dx.Layout", %__cblayout_CB_A, 188, 0, 32, 76, 80, 120, 128, 144, 160, 182)) %CB_A.cb_h, ptr @CB_A.cb, align 4
+
+// CHECK: define internal void @_init_buffer_CB_B.cb()
+// CHECK-NEXT: entry:
+// CHECK-NEXT: %CB_B.cb_h = call target("dx.CBuffer", target("dx.Layout", %__cblayout_CB_B, 94, 0, 88))
+// CHECK-SAME: @llvm.dx.resource.handlefromimplicitbinding.tdx.CBuffer_tdx.Layout_s___cblayout_CB_Bs_94_0_88tt(i32 6, i32 0, i32 1, i32 0, i1 false)
+// CHECK-NEXT: store target("dx.CBuffer", target("dx.Layout", %__cblayout_CB_B, 94, 0, 88)) %CB_B.cb_h, ptr @CB_B.cb, align 4
+
+// CHECK: define internal void @_init_buffer_CB_C.cb()
+// CHECK-NEXT: entry:
+// CHECK-NEXT: %CB_C.cb_h = call target("dx.CBuffer", target("dx.Layout", %__cblayout_CB_C, 400, 0, 16, 112, 128, 392))
+// CHECK-SAME: @llvm.dx.resource.handlefromimplicitbinding.tdx.CBuffer_tdx.Layout_s___cblayout_CB_Cs_400_0_16_112_128_392tt(i32 7, i32 0, i32 1, i32 0, i1 false)
+// CHECK-NEXT: store target("dx.CBuffer", target("dx.Layout", %__cblayout_CB_C, 400, 0, 16, 112, 128, 392)) %CB_C.cb_h, ptr @CB_C.cb, align 4
+
 RWBuffer<float> Buf;
 
 [numthreads(4,1,1)]
@@ -288,7 +336,13 @@ void main() {
 // CHECK: define internal void @_GLOBAL__sub_I_cbuffer.hlsl()
 // CHECK-NEXT: entry:
 // CHECK-NEXT: call void @_init_buffer_CBScalars.cb()
+// CHECK-NEXT: call void @_init_buffer_CBVectors.cb()
 // CHECK-NEXT: call void @_init_buffer_CBArrays.cb()
+// CHECK-NEXT: call void @_init_buffer_CBTypedefArray.cb()
+// CHECK-NEXT: call void @_init_buffer_CBStructs.cb()
+// CHECK-NEXT: call void @_init_buffer_CBClasses.cb()
+// CHECK-NEXT: call void @_init_buffer_CBMix.cb()
+// CHECK-NEXT: call void @_init_buffer_CB_A.cb()
 
 // CHECK: !hlsl.cbs = !{![[CBSCALARS:[0-9]+]], ![[CBVECTORS:[0-9]+]], ![[CBARRAYS:[0-9]+]], ![[CBTYPEDEFARRAY:[0-9]+]], ![[CBSTRUCTS:[0-9]+]], ![[CBCLASSES:[0-9]+]],
 // CHECK-SAME: ![[CBMIX:[0-9]+]], ![[CB_A:[0-9]+]], ![[CB_B:[0-9]+]], ![[CB_C:[0-9]+]]}