[AMDGPU][GlobalISel] Legalize memcpy family of intrinsics

Legalize G_MEMCPY, G_MEMMOVE, G_MEMSET and G_MEMCPY_INLINE.

Corresponding intrinsics are replaced by a loop that uses loads/stores in
AMDGPULowerIntrinsics pass unless their length is a constant lower then
MemIntrinsicExpandSizeThresholdOpt (default 1024). Any G_MEM* instruction that
reaches legalizer should have a const length argument and should be expanded
into appropriate number of loads + stores.

Differential Revision: https://reviews.llvm.org/D108357

llvm/include/llvm/CodeGen/GlobalISel/CombinerHelper.h

@@ -594,19 +594,6 @@ class CombinerHelper {
   bool tryEmitMemcpyInline(MachineInstr &MI);
 
 private:
-  // Memcpy family optimization helpers.
-  bool tryEmitMemcpyInline(MachineInstr &MI, Register Dst, Register Src,
-                           uint64_t KnownLen, Align DstAlign, Align SrcAlign,
-                           bool IsVolatile);
-  bool optimizeMemcpy(MachineInstr &MI, Register Dst, Register Src,
-                      uint64_t KnownLen, uint64_t Limit, Align DstAlign,
-                      Align SrcAlign, bool IsVolatile);
-  bool optimizeMemmove(MachineInstr &MI, Register Dst, Register Src,
-                       uint64_t KnownLen, Align DstAlign, Align SrcAlign,
-                       bool IsVolatile);
-  bool optimizeMemset(MachineInstr &MI, Register Dst, Register Val,
-                      uint64_t KnownLen, Align DstAlign, bool IsVolatile);
-
   /// Given a non-indexed load or store instruction \p MI, find an offset that
   /// can be usefully and legally folded into it as a post-indexing operation.
   ///

llvm/include/llvm/CodeGen/GlobalISel/LegalizerHelper.h

@@ -256,6 +256,20 @@ class LegalizerHelper {
                                         LLT SrcTy, LLT NarrowTy,
                                         unsigned ScalarOpc);
 
+  // Memcpy family legalization helpers.
+  LegalizeResult lowerMemset(MachineInstr &MI, Register Dst, Register Val,
+                             uint64_t KnownLen, Align Alignment,
+                             bool IsVolatile);
+  LegalizeResult lowerMemcpyInline(MachineInstr &MI, Register Dst, Register Src,
+                                   uint64_t KnownLen, Align DstAlign,
+                                   Align SrcAlign, bool IsVolatile);
+  LegalizeResult lowerMemcpy(MachineInstr &MI, Register Dst, Register Src,
+                             uint64_t KnownLen, uint64_t Limit, Align DstAlign,
+                             Align SrcAlign, bool IsVolatile);
+  LegalizeResult lowerMemmove(MachineInstr &MI, Register Dst, Register Src,
+                              uint64_t KnownLen, Align DstAlign, Align SrcAlign,
+                              bool IsVolatile);
+
 public:
   /// Return the alignment to use for a stack temporary object with the given
   /// type.
@@ -403,6 +417,8 @@ class LegalizerHelper {
   LegalizeResult lowerAbsToAddXor(MachineInstr &MI);
   LegalizeResult lowerAbsToMaxNeg(MachineInstr &MI);
   LegalizeResult lowerVectorReduction(MachineInstr &MI);
+  LegalizeResult lowerMemcpyInline(MachineInstr &MI);
+  LegalizeResult lowerMemCpyFamily(MachineInstr &MI, unsigned MaxLen = 0);
 };
 
 /// Helper function that creates a libcall to the given \p Name using the given