[PATCH] D41353: [InstCombine] Adjusting bswap pattern to the new masked shl canonization

[Omer Paparo Bivas via Phabricator via llvm-commits]

opaparo updated this revision to Diff 127501.
opaparo added a comment.

Added a comment explaining the change of pattern in the bswap idiom detection.


Repository:
  rL LLVM

https://reviews.llvm.org/D41353

Files:
  lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
  test/Transforms/InstCombine/bswap.ll


Index: test/Transforms/InstCombine/bswap.ll
===================================================================
--- test/Transforms/InstCombine/bswap.ll
+++ test/Transforms/InstCombine/bswap.ll
@@ -98,14 +98,7 @@
 ; PR23863
 define i32 @test7(i32 %x) {
 ; CHECK-LABEL: @test7(
-; CHECK-NEXT:    [[SHL:%.*]] = shl i32 %x, 16
-; CHECK-NEXT:    [[SHR:%.*]] = lshr i32 %x, 16
-; CHECK-NEXT:    [[OR:%.*]] = or i32 [[SHL]], [[SHR]]
-; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[OR]], 16711935
-; CHECK-NEXT:    [[SHL3:%.*]] = shl nuw i32 [[TMP1]], 8
-; CHECK-NEXT:    [[AND4:%.*]] = lshr i32 [[OR]], 8
-; CHECK-NEXT:    [[SHR5:%.*]] = and i32 [[AND4]], 16711935
-; CHECK-NEXT:    [[OR6:%.*]] = or i32 [[SHL3]], [[SHR5]]
+; CHECK-NEXT:    [[OR6:%.*]] = call i32 @llvm.bswap.i32(i32 %x)
 ; CHECK-NEXT:    ret i32 [[OR6]]
 ;
   %shl = shl i32 %x, 16
@@ -148,13 +141,9 @@
 
 define i16 @test10(i32 %a) {
 ; CHECK-LABEL: @test10(
-; CHECK-NEXT:    [[SHR1:%.*]] = lshr i32 %a, 8
-; CHECK-NEXT:    [[AND1:%.*]] = and i32 [[SHR1]], 255
-; CHECK-NEXT:    [[TMP1:%.*]] = and i32 %a, 255
-; CHECK-NEXT:    [[SHL1:%.*]] = shl nuw nsw i32 [[TMP1]], 8
-; CHECK-NEXT:    [[OR:%.*]] = or i32 [[AND1]], [[SHL1]]
-; CHECK-NEXT:    [[CONV:%.*]] = trunc i32 [[OR]] to i16
-; CHECK-NEXT:    ret i16 [[CONV]]
+; CHECK-NEXT:    [[TRUNC:%.*]] = trunc i32 %a to i16
+; CHECK-NEXT:    [[REV:%.*]] = call i16 @llvm.bswap.i16(i16 [[TRUNC]])
+; CHECK-NEXT:    ret i16 [[REV]]
 ;
   %shr1 = lshr i32 %a, 8
   %and1 = and i32 %shr1, 255
Index: lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
===================================================================
--- lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
+++ lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
@@ -1429,11 +1429,19 @@
   bool OrOfShifts = match(Op0, m_LogicalShift(m_Value(), m_Value())) &&
                     match(Op1, m_LogicalShift(m_Value(), m_Value()));
 
-  // (A & B) | (C & D)                              -> bswap if possible.
-  bool OrOfAnds = match(Op0, m_And(m_Value(), m_Value())) &&
-                  match(Op1, m_And(m_Value(), m_Value()));
-
-  if (!OrOfOrs && !OrOfShifts && !OrOfAnds)
+  // (A << B) | (C & D)                              -> bswap if possible.
+  // The bigger pattern here is ((A & C1) << C2) | ((B >> C2) & C1), which is a
+  // part of the bswap idiom for specific values of C1, C2 (e.g. C1 = 16711935,
+  // C2 = 8 for i32).
+  // Note the different canonic forms for 'shl's and 'shr's. For each of them
+  // the canonic form is set such that the 'and' constants are minimal (the
+  // 'and' is before the shift for 'shl' and after the shift for 'shr').
+  bool OrOfShiftAndAnd = (match(Op0, m_And(m_Value(), m_Value())) &&
+                          match(Op1, m_LogicalShift(m_Value(), m_Value()))) ||
+                         (match(Op0, m_LogicalShift(m_Value(), m_Value())) &&
+                          match(Op1, m_And(m_Value(), m_Value())));
+
+  if (!OrOfOrs && !OrOfShifts && !OrOfShiftAndAnd)
     return nullptr;
 
   SmallVector<Instruction*, 4> Insts;


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D41353.127501.patch
Type: text/x-patch
Size: 3053 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20171219/cbc19401/attachment.bin>