anonymous_namespace{GpuSharedMemoryUtils.cpp} Namespace Reference

Functions
llvm::Value *	codegen_smem_dest_slot_ptr (llvm::LLVMContext &context, const QueryMemoryDescriptor &query_mem_desc, llvm::IRBuilder<> &ir_builder, const size_t slot_idx, const TargetInfo &target_info, llvm::Value *dest_byte_stream, llvm::Value *byte_offset)
void	replace_called_function_with (llvm::Function *main_func, const std::string &target_func_name, llvm::Function *replace_func)

Function Documentation

llvm::Value* anonymous_namespace{GpuSharedMemoryUtils.cpp}::codegen_smem_dest_slot_ptr	(	llvm::LLVMContext &	context,
		const QueryMemoryDescriptor &	query_mem_desc,
		llvm::IRBuilder<> &	ir_builder,
		const size_t	slot_idx,
		const TargetInfo &	target_info,
		llvm::Value *	dest_byte_stream,
		llvm::Value *	byte_offset
	)

Definition at line 216 of file GpuSharedMemoryUtils.cpp.

References CHECK, get_compact_type(), QueryMemoryDescriptor::getPaddedSlotWidthBytes(), to_string(), and UNREACHABLE.

Referenced by GpuSharedMemCodeBuilder::codegenInitialization().

                                                                {
  const auto sql_type = get_compact_type(target_info);
  const auto slot_bytes = query_mem_desc.getPaddedSlotWidthBytes(slot_idx);
  auto ptr_type = [&context](const size_t slot_bytes, const SQLTypeInfo& sql_type) {
    if (slot_bytes == sizeof(int32_t)) {
      return llvm::Type::getInt32PtrTy(context, /*address_space=*/3);
    } else {
      CHECK(slot_bytes == sizeof(int64_t));
      return llvm::Type::getInt64PtrTy(context, /*address_space=*/3);
    }
    UNREACHABLE() << "Invalid slot size encountered: " << std::to_string(slot_bytes);
    return llvm::Type::getInt32PtrTy(context, /*address_space=*/3);
  };

  const auto casted_dest_slot_address = ir_builder.CreatePointerCast(
      ir_builder.CreateGEP(
          dest_byte_stream->getType()->getScalarType()->getPointerElementType(),
          dest_byte_stream,
          byte_offset),
      ptr_type(slot_bytes, sql_type),
      "dest_slot_adr_" + std::to_string(slot_idx));
  return casted_dest_slot_address;
}

Here is the call graph for this function:

Here is the caller graph for this function:

void anonymous_namespace{GpuSharedMemoryUtils.cpp}::replace_called_function_with	(	llvm::Function *	main_func,
		const std::string &	target_func_name,
		llvm::Function *	replace_func
	)

searches through the main function for the first appearance of called function "target_func_name", and if found it replaces it with replace_func while keeping the same arguments

Definition at line 375 of file GpuSharedMemoryUtils.cpp.

References run_benchmark_import::args, and UNREACHABLE.

Referenced by GpuSharedMemCodeBuilder::injectFunctionsInto().

                                                              {
  for (auto it = llvm::inst_begin(main_func), e = llvm::inst_end(main_func); it != e;
       ++it) {
    if (!llvm::isa<llvm::CallInst>(*it)) {
      continue;
    }
    auto& instruction = llvm::cast<llvm::CallInst>(*it);
    if (std::string(instruction.getCalledFunction()->getName()) == target_func_name) {
      std::vector<llvm::Value*> args;
      for (size_t i = 0; i < instruction.getNumOperands() - 1; ++i) {
        args.push_back(instruction.getArgOperand(i));
      }
      llvm::ReplaceInstWithInst(&instruction,
                                llvm::CallInst::Create(replace_func, args, ""));
      return;
    }
  }
  UNREACHABLE() << "Target function " << target_func_name << " was not found in "
                << replace_func->getName().str();
}

Here is the caller graph for this function: