Takes param size and align from literal block in OpEnqueueKernel (#3123)

- Implemented the TODO in the visitCallEnqueueKernel function
- Takes param size and align from literal block
- If dynamic allocation is used store instructions are processed to
obtain size and align
- Modified test

Original commit:
KhronosGroup/SPIRV-LLVM-Translator@ff7db6ef1c71860

Author: EbinJose2002

llvm-spirv/lib/SPIRV/OCLToSPIRV.cpp

@@ -1501,14 +1501,54 @@ void OCLToSPIRVBase::visitCallEnqueueKernel(CallInst *CI,
   // Param: Pointer to block literal
   Value *BlockLiteral = CI->getArgOperand(BlockFIdx + 1);
   Args.push_back(BlockLiteral);
+  BlockLiteral = BlockLiteral->stripPointerCasts();
+  if (auto *GV = dyn_cast<GlobalVariable>(BlockLiteral)) {
+    assert(GV->hasInitializer() && "Block literal should have an initializer");
+    Constant *Init = GV->getInitializer();
+    if (auto *Struct = dyn_cast<ConstantStruct>(Init)) {
+      Constant *SizeConst = Struct->getOperand(0);
+      Constant *AlignConst = Struct->getOperand(1);
+      auto *SizeVal = dyn_cast<ConstantInt>(SizeConst);
+      auto *AlignVal = dyn_cast<ConstantInt>(AlignConst);
+      Args.push_back(SizeVal);
+      Args.push_back(AlignVal);
+    }
+  } else {
 
-  // Param Size: Size of block literal structure
-  // Param Aligment: Aligment of block literal structure
-  // TODO: these numbers should be obtained from block literal structure
-  Type *ParamType = getBlockStructType(BlockLiteral);
-  Args.push_back(getInt32(M, DL.getTypeStoreSize(ParamType)));
-  Args.push_back(getInt32(M, DL.getPrefTypeAlign(ParamType).value()));
+    Value *Base = getUnderlyingObject(BlockLiteral);
+    Value *ParamSizeVal = nullptr;
+    Value *ParamAlignVal = nullptr;
+
+    for (User *U : Base->users()) {
+      if (auto *GEP = dyn_cast<GetElementPtrInst>(U)) {
+        if (GEP->getNumIndices() < 2)
+          continue;
+        auto *CI1 = dyn_cast<ConstantInt>(GEP->getOperand(2));
+        if (!CI1)
+          continue;
+        uint64_t FieldIndex = CI1->getZExtValue();
+        for (User *GEPUser : GEP->users()) {
+          if (auto *Store = dyn_cast<StoreInst>(GEPUser)) {
+            if (FieldIndex == 0)
+              ParamSizeVal = dyn_cast<ConstantInt>(Store->getValueOperand());
+            else if (FieldIndex == 1)
+              ParamAlignVal = dyn_cast<ConstantInt>(Store->getValueOperand());
+          }
+        }
+      }
+      if (ParamSizeVal && ParamAlignVal)
+        break;
+    }
+    Type *ParamType = getBlockStructType(BlockLiteral);
+    // Fallback to default if not found
+    if (!ParamSizeVal)
+      ParamSizeVal = getInt32(M, DL.getTypeStoreSize(ParamType));
+    if (!ParamAlignVal)
+      ParamAlignVal = getInt32(M, DL.getPrefTypeAlign(ParamType).value());
 
+    Args.push_back(ParamSizeVal);
+    Args.push_back(ParamAlignVal);
+  }
   // Local sizes arguments: Sizes of block invoke arguments
   // Clang 6.0 and higher generates local size operands as an array,
   // so we need to unpack them

llvm-spirv/test/transcoding/enqueue_kernel.cl

@@ -37,11 +37,11 @@
 // CHECK-SPIRV: TypeInt [[Int32Ty:[0-9]+]] 32
 // CHECK-SPIRV: TypeInt [[Int8Ty:[0-9]+]] 8
 // CHECK-SPIRV: Constant [[Int32Ty]] [[ConstInt12:[0-9]+]] 12
+// CHECK-SPIRV: Constant [[Int32Ty]] [[ConstInt4:[0-9]+]] 4
 // CHECK-SPIRV: Constant [[Int32Ty]] [[ConstInt0:[0-9]+]] 0
-// CHECK-SPIRV: Constant [[Int32Ty]] [[ConstInt17:[0-9]+]] 21
+// CHECK-SPIRV: Constant [[Int32Ty]] [[ConstInt21:[0-9]+]] 21
 // CHECK-SPIRV: Constant [[Int32Ty]] [[ConstInt2:[0-9]+]] 2
-// CHECK-SPIRV: Constant [[Int32Ty]] [[ConstInt8:[0-9]+]] 8
-// CHECK-SPIRV: Constant [[Int32Ty]] [[ConstInt20:[0-9]+]] 24
+// CHECK-SPIRV: Constant [[Int32Ty]] [[ConstInt24:[0-9]+]] 24
 
 // CHECK-SPIRV-TYPED-PTR: TypePointer [[Int8PtrGenTy:[0-9]+]] 8 [[Int8Ty]]
 // CHECK-SPIRV-UNTYPED-PTR: TypeUntypedPointerKHR [[Int8PtrGenTy:[0-9]+]] 8
@@ -78,7 +78,7 @@ kernel void device_side_enqueue(global int *a, global int *b, int i, char c0) {
   // CHECK-SPIRV: Bitcast [[Int8PtrGenTy]] [[BlockLit1:[0-9]+]]
   // CHECK-SPIRV: EnqueueKernel [[Int32Ty]] [[#]] [[#]] [[#]] [[#]]
   // CHECK-SPIRV-SAME: [[ConstInt0]] [[EventNull]] [[#]]
-  // CHECK-SPIRV-SAME: [[#InvokeFunc1]] [[BlockLit1]] [[ConstInt17]] [[ConstInt8]]
+  // CHECK-SPIRV-SAME: [[#InvokeFunc1]] [[BlockLit1]] [[ConstInt21]] [[ConstInt4]]
 
   // CHECK-LLVM: [[Block2:%[0-9]+]] = addrspacecast ptr %block to ptr addrspace(4)
   // CHECK-LLVM: [[Block2Ptr:%[0-9]+]] = bitcast ptr addrspace(4) [[Block2]] to ptr addrspace(4)
@@ -98,7 +98,7 @@ kernel void device_side_enqueue(global int *a, global int *b, int i, char c0) {
   // CHECK-SPIRV: Bitcast [[Int8PtrGenTy]] [[BlockLit2:[0-9]+]]
   // CHECK-SPIRV: EnqueueKernel [[Int32Ty]] [[#]] [[#]] [[#]] [[#]]
   // CHECK-SPIRV-SAME: [[ConstInt2]] [[Event1]] [[Event2]]
-  // CHECK-SPIRV-SAME: [[#InvokeFunc2]] [[BlockLit2]] [[ConstInt20]] [[ConstInt8]]
+  // CHECK-SPIRV-SAME: [[#InvokeFunc2]] [[BlockLit2]] [[ConstInt24]] [[ConstInt4]]
 
   // CHECK-LLVM: [[Block3:%[0-9]+]] = addrspacecast ptr %block4 to ptr addrspace(4)
   // CHECK-LLVM: [[Block3Ptr:%[0-9]+]] = bitcast ptr addrspace(4) [[Block3]] to ptr addrspace(4)
@@ -120,7 +120,7 @@ kernel void device_side_enqueue(global int *a, global int *b, int i, char c0) {
   // CHECK-SPIRV: Bitcast [[Int8PtrGenTy]] [[BlockLit3:[0-9]+]] [[BlockLit3Tmp]]
   // CHECK-SPIRV: EnqueueKernel [[Int32Ty]] [[#]] [[#]] [[#]] [[#]]
   // CHECK-SPIRV-SAME: [[ConstInt2]] [[Event1]] [[Event2]]
-  // CHECK-SPIRV-SAME: [[#InvokeFunc3]] [[BlockLit3]] [[ConstInt12]] [[ConstInt8]]
+  // CHECK-SPIRV-SAME: [[#InvokeFunc3]] [[BlockLit3]] [[ConstInt12]] [[ConstInt4]]
   // CHECK-SPIRV-SAME: [[LocalBuf31]]
 
   // CHECK-LLVM: [[Block0Tmp:%[0-9]+]] = addrspacecast ptr addrspace(1) @__block_literal_global to ptr addrspace(4)
@@ -146,7 +146,7 @@ kernel void device_side_enqueue(global int *a, global int *b, int i, char c0) {
   // CHECK-SPIRV: Bitcast [[Int8PtrGenTy]] [[BlockLit4:[0-9]+]] [[BlockLit4Tmp]]
   // CHECK-SPIRV: EnqueueKernel [[Int32Ty]] [[#]] [[#]] [[#]] [[#]]
   // CHECK-SPIRV-SAME: [[ConstInt0]] [[#]] [[#]]
-  // CHECK-SPIRV-SAME: [[#InvokeFunc4]] [[BlockLit4]] [[ConstInt12]] [[ConstInt8]]
+  // CHECK-SPIRV-SAME: [[#InvokeFunc4]] [[BlockLit4]] [[ConstInt12]] [[ConstInt4]]
   // CHECK-SPIRV-SAME: [[LocalBuf41]] [[LocalBuf42]] [[LocalBuf43]]
 
   // CHECK-LLVM: [[Block1Tmp:%[0-9]+]] = addrspacecast ptr addrspace(1) @__block_literal_global.1 to ptr addrspace(4)
@@ -167,7 +167,7 @@ kernel void device_side_enqueue(global int *a, global int *b, int i, char c0) {
   // CHECK-SPIRV: Bitcast [[Int8PtrGenTy]] [[BlockLit5:[0-9]+]]
   // CHECK-SPIRV: EnqueueKernel [[Int32Ty]] [[#]] [[#]] [[#]] [[#]]
   // CHECK-SPIRV-SAME: [[ConstInt0]] [[#]] [[Event1]]
-  // CHECK-SPIRV-SAME: [[#InvokeFunc5]] [[BlockLit5]] [[ConstInt20]] [[ConstInt8]]
+  // CHECK-SPIRV-SAME: [[#InvokeFunc5]] [[BlockLit5]] [[ConstInt24]] [[ConstInt4]]
 
   // CHECK-LLVM: [[Block5:%[0-9]+]] = addrspacecast ptr %block15 to ptr addrspace(4)
   // CHECK-LLVM: [[Block5Ptr:%[0-9]+]] = bitcast ptr addrspace(4) [[Block5]] to ptr addrspace(4)