[llvm] r269226 - [AArch64] Improve getUsefulBitsForUse for narrow stores.

[Chad Rosier via llvm-commits]

Author: mcrosier
Date: Wed May 11 15:19:54 2016
New Revision: 269226

URL: http://llvm.org/viewvc/llvm-project?rev=269226&view=rev
Log:
[AArch64] Improve getUsefulBitsForUse for narrow stores.

For narrow stores (e.g., strb, srth) we know the upper bits of the register are
unused/not useful. In some cases we can use this information to eliminate
unnecessary instructions.

For example, without this patch we generate (from the 2nd test case):

 ldr w8, [x0]
 and w8, w8, #0xfff0
 bfxil w8, w2, #16, #4
 strh w8, [x1]

and after the patch the 'and' is removed:

 ldr w8, [x0]
 bfxil w8, w2, #16, #4
 strh w8, [x1]
 ret

During the lowering of the bitfield insert instruction the 'and' is eliminated
because we know the upper 16-bits that are masked off are unused and the lower
4-bits that are masked off are overwritten by the insert itself. Therefore, the
'and' is unnecessary.

Differential Revision: http://reviews.llvm.org/D20175

Modified:
    llvm/trunk/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
    llvm/trunk/test/CodeGen/AArch64/bitfield-insert.ll

Modified: llvm/trunk/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp?rev=269226&r1=269225&r2=269226&view=diff
==============================================================================
--- llvm/trunk/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp (original)
+++ llvm/trunk/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp Wed May 11 15:19:54 2016
@@ -1849,6 +1849,20 @@ static void getUsefulBitsForUse(SDNode *
   case AArch64::BFMWri:
   case AArch64::BFMXri:
     return getUsefulBitsFromBFM(SDValue(UserNode, 0), Orig, UsefulBits, Depth);
+
+  case AArch64::STRBui:
+  case AArch64::STRBBui:
+    if (UserNode->getOperand(0) != Orig)
+      return;
+    UsefulBits &= APInt(UsefulBits.getBitWidth(), 0xff);
+    return;
+
+  case AArch64::STRHui:
+  case AArch64::STRHHui:
+    if (UserNode->getOperand(0) != Orig)
+      return;
+    UsefulBits &= APInt(UsefulBits.getBitWidth(), 0xffff);
+    return;
   }
 }
 

Modified: llvm/trunk/test/CodeGen/AArch64/bitfield-insert.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/bitfield-insert.ll?rev=269226&r1=269225&r2=269226&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/AArch64/bitfield-insert.ll (original)
+++ llvm/trunk/test/CodeGen/AArch64/bitfield-insert.ll Wed May 11 15:19:54 2016
@@ -237,3 +237,39 @@ define i32 @test_nouseful_bits(i8 %a, i3
   %shl.4 = shl i32 %or.3, 8     ;   A A A 0
   ret i32 %shl.4
 }
+
+define void @test_nouseful_strb(i32* %ptr32, i8* %ptr8, i32 %x)  {
+entry:
+; CHECK-LABEL: @test_nouseful_strb
+; CHECK: ldr [[REG1:w[0-9]+]],
+; CHECK-NOT:  and {{w[0-9]+}}, {{w[0-9]+}}, #0xf8
+; CHECK-NEXT: bfxil [[REG1]], w2, #16, #3
+; CHECK-NEXT: strb [[REG1]],
+; CHECK-NEXT: ret
+  %0 = load i32, i32* %ptr32, align 8
+  %and = and i32 %0, -8
+  %shr = lshr i32 %x, 16
+  %and1 = and i32 %shr, 7
+  %or = or i32 %and, %and1
+  %trunc = trunc i32 %or to i8
+  store i8 %trunc, i8* %ptr8
+  ret void
+}
+
+define void @test_nouseful_strh(i32* %ptr32, i16* %ptr16, i32 %x)  {
+entry:
+; CHECK-LABEL: @test_nouseful_strh
+; CHECK: ldr [[REG1:w[0-9]+]],
+; CHECK-NOT:  and {{w[0-9]+}}, {{w[0-9]+}}, #0xfff0
+; CHECK-NEXT: bfxil [[REG1]], w2, #16, #4
+; CHECK-NEXT: strh [[REG1]],
+; CHECK-NEXT: ret
+  %0 = load i32, i32* %ptr32, align 8
+  %and = and i32 %0, -16
+  %shr = lshr i32 %x, 16
+  %and1 = and i32 %shr, 15
+  %or = or i32 %and, %and1
+  %trunc = trunc i32 %or to i16
+  store i16 %trunc, i16* %ptr16
+  ret void
+}