[BPF] generate proper code for __builtin_bpf_load_u32_to_ptr()

[yonghong-song]

Summary

This is the BPF backend support for __builtin_bpf_load_u32_to_ptr().
A new codegen only opcode LOAD_U32_TO_PTR is introduced
for IR builtin int_bpf_load_u32_to_ptr(). Finally, a
single LDW instruction is generated for the builtin.

[yonghong-song]

The corresponding clang change: https://reviews.llvm.org/D81479

[yonghong-song]

generate code for LOAD_U32_TO_PTR through BPFInstrInfo.td specificaiton instead of rewrite at AsmPrint time.

[yonghong-song]

user inline asm can do the work so abandon this patch.

llvm/lib/Target/BPF/BPFAsmPrinter.cpp

else
     O << "(" << BPFInstPrinter::getRegisterName(BaseMO.getReg()) << " + " << Offset << ")";
 
   return false;
 }
 
 void BPFAsmPrinter::emitInstruction(const MachineInstr *MI) {
   MCInst TmpInst;
 
-  if (!BTF || !BTF->InstLower(MI, TmpInst)) {
+  if (MI->getOpcode() == BPF::LOAD_U32_TO_PTR) {
+    TmpInst.setOpcode(BPF::LDW);
+    TmpInst.addOperand(MCOperand::createReg(MI->getOperand(0).getReg()));
+    TmpInst.addOperand(MCOperand::createReg(MI->getOperand(1).getReg()));
+    TmpInst.addOperand(MCOperand::createImm(MI->getOperand(2).getImm()));
+  } else if (!BTF || !BTF->InstLower(MI, TmpInst)) {
     BPFMCInstLower MCInstLowering(OutContext, *this);
     MCInstLowering.Lower(MI, TmpInst);
   }
   EmitToStreamer(*OutStreamer, TmpInst);
 }
 
 // Force static initialization.
 extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeBPFAsmPrinter() {
   RegisterAsmPrinter<BPFAsmPrinter> X(getTheBPFleTarget());
   RegisterAsmPrinter<BPFAsmPrinter> Y(getTheBPFbeTarget());
   RegisterAsmPrinter<BPFAsmPrinter> Z(getTheBPFTarget());
 }

llvm/lib/Target/BPF/BPFInstrInfo.td

def LD_pseudo
 
   let Inst{51-48} = dst;
   let Inst{55-52} = pseudo;
   let Inst{47-32} = 0;
   let Inst{31-0} = imm{31-0};
   let BPFClass = BPF_LD;
 }
 
+let isCodeGenOnly = 1 in {
+  def LOAD_U32_TO_PTR : TYPE_LD_ST<BPF_MEM.Value, BPF_W.Value,
+                                   (outs GPR:$dst),
+                                   (ins GPR:$src, u64imm:$offset),
+                                   "$dst = (void *)(long)*(u32 *)($src + $offset)",
+                                   [(set GPR:$dst, (int_bpf_load_u32_to_ptr GPR:$src, imm:$offset))]>;
+}
+
 // STORE instructions
 class STORE<BPFWidthModifer SizeOp, string OpcodeStr, list<dag> Pattern>
     : TYPE_LD_ST<BPF_MEM.Value, SizeOp.Value,
                  (outs),
                  (ins GPR:$src, MEMri:$addr),
                  "*("#OpcodeStr#" *)($addr) = $src",
                  Pattern> {
   bits<4> src;

llvm/test/CodeGen/BPF/load_u32_to_ptr.ll

+; RUN: llc < %s -march=bpfel | FileCheck %s
+; RUN: llc < %s -march=bpfel -mattr=+alu32 | FileCheck %s
+
+%struct.t = type { i32, i32 }
+
+; Function Attrs: nounwind readonly
+define dso_local i8* @test(%struct.t* readonly %arg) local_unnamed_addr {
+entry:
+  %0 = bitcast %struct.t* %arg to i8*
+  %1 = tail call i8* @llvm.bpf.load.u32.to.ptr(i8* %0, i64 4)
+  ret i8* %1
+}
+
+; CHECK:        r0 = *(u32 *)(r{{[0-9]+}} + 4)
+; CHECK-NEXT:   exit
+
+; Function Attrs: nounwind readonly
+declare i8* @llvm.bpf.load.u32.to.ptr(i8*, i64) #1