maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[InstCombine] matchBSwapOrBitReversem - recognise or(fshl(),fshl()) bswap patterns. 

I'm not certain InstCombinerImpl::matchBSwapOrBitReverse needs to filter the or(op0(),op1()) ops - there are just too many cases that recognizeBSwapOrBitReverseIdiom/collectBitParts handle now (and quickly).
This commit is contained in:
Simon Pilgrim 2020-10-25 10:17:45 +00:00
parent 5e9f172295
commit 3052e474ec
2 changed files with 11 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp
View File

@ -2010,7 +2010,16 @@ Instruction *InstCombinerImpl::matchBSwapOrBitReverse(BinaryOperator &Or,
bool OrWithAnds = match(Op0, m_And(m_Value(), m_Value())) ||
match(Op1, m_And(m_Value(), m_Value()));
if (!OrWithOrs && !OrWithShifts && !OrWithAnds)
// fshl(A,B,C) | D and A | fshl(B,C,D) -> bswap if possible.
// fshr(A,B,C) | D and A | fshr(B,C,D) -> bswap if possible.
bool OrWithFunnels = match(Op0, m_FShl(m_Value(), m_Value(), m_Value())) ||
match(Op0, m_FShr(m_Value(), m_Value(), m_Value())) ||
match(Op0, m_FShl(m_Value(), m_Value(), m_Value())) ||
match(Op0, m_FShr(m_Value(), m_Value(), m_Value()));
// TODO: Do we need all these filtering checks or should we just rely on
// recognizeBSwapOrBitReverseIdiom + collectBitParts to reject them quickly?
if (!OrWithOrs && !OrWithShifts && !OrWithAnds && !OrWithFunnels)
return nullptr;
SmallVector<Instruction *, 4> Insts;

6
llvm/test/Transforms/InstCombine/bswap.ll
View File

@ -687,11 +687,7 @@ define i32 @funnel_binary(i32 %abcd) {
define i32 @funnel_and(i32 %abcd) {
; CHECK-LABEL: @funnel_and(
; CHECK-NEXT: [[ZZCZ:%.*]] = and i32 [[ABCD:%.*]], 65280
; CHECK-NEXT: [[ZCZA:%.*]] = call i32 @llvm.fshl.i32(i32 [[ZZCZ]], i32 [[ABCD]], i32 8)
; CHECK-NEXT: [[ZBZZ:%.*]] = and i32 [[ABCD]], 16711680
; CHECK-NEXT: [[DZBZ:%.*]] = call i32 @llvm.fshl.i32(i32 [[ABCD]], i32 [[ZBZZ]], i32 24)
; CHECK-NEXT: [[DCBA:%.*]] = or i32 [[ZCZA]], [[DZBZ]]
; CHECK-NEXT: [[DCBA:%.*]] = call i32 @llvm.bswap.i32(i32 [[ABCD:%.*]])
; CHECK-NEXT: ret i32 [[DCBA]]
;
%zzcz = and i32 %abcd, 65280