[LoongArch] Implement isZextFree

This returns true for 8-bit and 16-bit loads, allowing ld.bu/ld.hu to be selected and avoiding unnecessary masks.

Signed-off-by: WANG Rui <[email protected]>

Reviewed By: SixWeining, xen0n

Differential Revision: https://reviews.llvm.org/D154819

Author: heiher

llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp

@@ -3255,6 +3255,21 @@ bool LoongArchTargetLowering::isLegalAddImmediate(int64_t Imm) const {
   return isInt<12>(Imm);
 }
 
+bool LoongArchTargetLowering::isZExtFree(SDValue Val, EVT VT2) const {
+  // Zexts are free if they can be combined with a load.
+  // Don't advertise i32->i64 zextload as being free for LA64. It interacts
+  // poorly with type legalization of compares preferring sext.
+  if (auto *LD = dyn_cast<LoadSDNode>(Val)) {
+    EVT MemVT = LD->getMemoryVT();
+    if ((MemVT == MVT::i8 || MemVT == MVT::i16) &&
+        (LD->getExtensionType() == ISD::NON_EXTLOAD ||
+         LD->getExtensionType() == ISD::ZEXTLOAD))
+      return true;
+  }
+
+  return TargetLowering::isZExtFree(Val, VT2);
+}
+
 bool LoongArchTargetLowering::hasAndNotCompare(SDValue Y) const {
   // TODO: Support vectors.
   if (Y.getValueType().isVector())

llvm/lib/Target/LoongArch/LoongArchISelLowering.h

@@ -204,6 +204,7 @@ class LoongArchTargetLowering : public TargetLowering {
 
   bool isLegalICmpImmediate(int64_t Imm) const override;
   bool isLegalAddImmediate(int64_t Imm) const override;
+  bool isZExtFree(SDValue Val, EVT VT2) const override;
 
   bool hasAndNotCompare(SDValue Y) const override;
 

llvm/test/CodeGen/LoongArch/zext-with-load-is-free.ll

@@ -7,14 +7,12 @@
 define zeroext i8 @test_zext_i8(ptr %p) nounwind {
 ; LA32-LABEL: test_zext_i8:
 ; LA32:       # %bb.0:
-; LA32-NEXT:    ld.b $a0, $a0, 0
-; LA32-NEXT:    andi $a0, $a0, 255
+; LA32-NEXT:    ld.bu $a0, $a0, 0
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: test_zext_i8:
 ; LA64:       # %bb.0:
-; LA64-NEXT:    ld.b $a0, $a0, 0
-; LA64-NEXT:    andi $a0, $a0, 255
+; LA64-NEXT:    ld.bu $a0, $a0, 0
 ; LA64-NEXT:    ret
   %a = load i8, ptr %p, align 1
   br label %exit
@@ -26,16 +24,14 @@ define zeroext i16 @test_zext_i16(ptr %p) nounwind {
 ; LA32-LABEL: test_zext_i16:
 ; LA32:       # %bb.0:
 ; LA32-NEXT:    ld.bu $a1, $a0, 0
-; LA32-NEXT:    ld.b $a0, $a0, 1
+; LA32-NEXT:    ld.bu $a0, $a0, 1
 ; LA32-NEXT:    slli.w $a0, $a0, 8
 ; LA32-NEXT:    or $a0, $a0, $a1
-; LA32-NEXT:    bstrpick.w $a0, $a0, 15, 0
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: test_zext_i16:
 ; LA64:       # %bb.0:
-; LA64-NEXT:    ld.h $a0, $a0, 0
-; LA64-NEXT:    bstrpick.d $a0, $a0, 15, 0
+; LA64-NEXT:    ld.hu $a0, $a0, 0
 ; LA64-NEXT:    ret
   %a = load i16, ptr %p, align 1
   br label %exit