[Hexagon] Implement shouldConvertConstantLoadToIntImm (#146452)

This will convert loads of constant strings to immediate values. Put
this behind a flag that is enabled by default so that we can toggle it
if need be.

Author: svs-quic

llvm/lib/Target/Hexagon/HexagonISelLowering.cpp

@@ -102,6 +102,10 @@ static cl::opt<int>
     MaxStoresPerMemsetOptSizeCL("max-store-memset-Os", cl::Hidden, cl::init(4),
                                 cl::desc("Max #stores to inline memset"));
 
+static cl::opt<bool>
+    ConstantLoadsToImm("constant-loads-to-imm", cl::Hidden, cl::init(true),
+                       cl::desc("Convert constant loads to immediate values."));
+
 static cl::opt<bool> AlignLoads("hexagon-align-loads",
   cl::Hidden, cl::init(false),
   cl::desc("Rewrite unaligned loads as a pair of aligned loads"));
@@ -3607,6 +3611,18 @@ bool HexagonTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT,
   return true;
 }
 
+/// Returns true if it is beneficial to convert a load of a constant
+/// to just the constant itself.
+bool HexagonTargetLowering::shouldConvertConstantLoadToIntImm(const APInt &Imm,
+                                                              Type *Ty) const {
+  if (!ConstantLoadsToImm)
+    return false;
+
+  assert(Ty->isIntegerTy());
+  unsigned BitSize = Ty->getPrimitiveSizeInBits();
+  return (BitSize > 0 && BitSize <= 64);
+}
+
 /// isLegalAddressingMode - Return true if the addressing mode represented by
 /// AM is legal for this target, for a load/store of the specified type.
 bool HexagonTargetLowering::isLegalAddressingMode(const DataLayout &DL,

llvm/lib/Target/Hexagon/HexagonISelLowering.h

@@ -342,6 +342,11 @@ class HexagonTargetLowering : public TargetLowering {
   SDValue getPICJumpTableRelocBase(SDValue Table, SelectionDAG &DAG)
                                    const override;
 
+  /// Returns true if it is beneficial to convert a load of a constant
+  /// to just the constant itself.
+  bool shouldConvertConstantLoadToIntImm(const APInt &Imm,
+                                         Type *Ty) const override;
+
   bool shouldReduceLoadWidth(SDNode *Load, ISD::LoadExtType ExtTy, EVT NewVT,
                              std::optional<unsigned> ByteOffset) const override;
 

llvm/test/CodeGen/Hexagon/hexagon-strcpy.ll

@@ -0,0 +1,57 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -march=hexagon -verify-machineinstrs  < %s | FileCheck %s
+
+@.str = private unnamed_addr constant [31 x i8] c"DHRYSTONE PROGRAM, 3'RD STRING\00", align 1
+@.str1 = private unnamed_addr constant [3 x i8] c"%s\00", align 1
+
+; Function Attrs: nounwind
+declare i32 @printf(i8* nocapture readonly, ...)
+
+; Function Attrs: nounwind
+define i32 @main() {
+; CHECK-LABEL: main:
+; CHECK:         .cfi_startproc
+; CHECK-NEXT:  // %bb.0: // %entry
+; CHECK-NEXT:    .cfi_def_cfa r30, 8
+; CHECK-NEXT:    .cfi_offset r31, -4
+; CHECK-NEXT:    .cfi_offset r30, -8
+; CHECK-NEXT:    {
+; CHECK-NEXT:     r0 = ##.L.str1
+; CHECK-NEXT:     r3:2 = CONST64(#2325073635944967245)
+; CHECK-NEXT:     allocframe(r29,#40):raw
+; CHECK-NEXT:    }
+; CHECK-NEXT:    {
+; CHECK-NEXT:     r1 = add(r29,#8)
+; CHECK-NEXT:     r7:6 = CONST64(#4706902966564560965)
+; CHECK-NEXT:     r5:4 = CONST64(#5642821575076104260)
+; CHECK-NEXT:    }
+; CHECK-NEXT:    {
+; CHECK-NEXT:     memb(r29+#38) = #0
+; CHECK-NEXT:     memw(r29+#0) = r1
+; CHECK-NEXT:    }
+; CHECK-NEXT:    {
+; CHECK-NEXT:     memd(r29+#24) = r3:2
+; CHECK-NEXT:     memd(r29+#16) = r7:6
+; CHECK-NEXT:    }
+; CHECK-NEXT:    {
+; CHECK-NEXT:     memd(r29+#8) = r5:4
+; CHECK-NEXT:     memh(r29+#36) = ##18254
+; CHECK-NEXT:    }
+; CHECK-NEXT:    {
+; CHECK-NEXT:     call printf
+; CHECK-NEXT:     memw(r29+#32) = ##1230132307
+; CHECK-NEXT:    }
+; CHECK-NEXT:    {
+; CHECK-NEXT:     r0 = #0
+; CHECK-NEXT:     dealloc_return
+; CHECK-NEXT:    }
+entry:
+  %blah = alloca [30 x i8], align 8
+  %arraydecay = getelementptr inbounds [30 x i8], [30 x i8]* %blah, i32 0, i32 0
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %arraydecay, i8* getelementptr inbounds ([31 x i8], [31 x i8]* @.str, i32 0, i32 0), i32 31, i32 1, i1 false)
+  %call2 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([3 x i8], [3 x i8]* @.str1, i32 0, i32 0), i8* %arraydecay)
+  ret i32 0
+}
+
+; Function Attrs: nounwind
+declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture readonly, i32, i32, i1)