[llvm] 1badf7c - [InstComine] Forego of one-use check in `(X - (X & Y)) --> (X & ~Y)` if Y is a constant
Roman Lebedev via llvm-commits
llvm-commits at lists.llvm.org
Fri Mar 6 10:39:40 PST 2020
Author: Roman Lebedev
Date: 2020-03-06T21:39:07+03:00
New Revision: 1badf7c33a5d01900c77646f750e2ea11ad8bf5a
URL: https://github.com/llvm/llvm-project/commit/1badf7c33a5d01900c77646f750e2ea11ad8bf5a
DIFF: https://github.com/llvm/llvm-project/commit/1badf7c33a5d01900c77646f750e2ea11ad8bf5a.diff
LOG: [InstComine] Forego of one-use check in `(X - (X & Y)) --> (X & ~Y)` if Y is a constant
Summary:
This is potentially more friendly for further optimizations,
analysies, e.g.: https://godbolt.org/z/G24anE
This resolves phase-ordering bug that was introduced
in D75145 for https://godbolt.org/z/2gBwF2
https://godbolt.org/z/XvgSua
Reviewers: spatel, nikic, dmgreen, xbolva00
Reviewed By: nikic, xbolva00
Subscribers: hiraditya, zzheng, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D75757
Added:
Modified:
llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
llvm/test/Transforms/InstCombine/sub.ll
llvm/test/Transforms/LoopUnroll/runtime-loop-multiple-exits.ll
llvm/test/Transforms/LoopUnroll/runtime-unroll-remainder.ll
llvm/test/Transforms/LoopVectorize/X86/float-induction-x86.ll
llvm/test/Transforms/PhaseOrdering/reassociate-after-unroll.ll
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
index 3eaa1b60b215..a78125104836 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp
@@ -1917,6 +1917,12 @@ Instruction *InstCombiner::visitSub(BinaryOperator &I) {
return NewSel;
}
+ // (X - (X & Y)) --> (X & ~Y)
+ if (match(Op1, m_c_And(m_Specific(Op0), m_Value(Y))) &&
+ (Op1->hasOneUse() || isa<Constant>(Y)))
+ return BinaryOperator::CreateAnd(
+ Op0, Builder.CreateNot(Y, Y->getName() + ".not"));
+
if (Op1->hasOneUse()) {
Value *Y = nullptr, *Z = nullptr;
Constant *C = nullptr;
@@ -1926,11 +1932,6 @@ Instruction *InstCombiner::visitSub(BinaryOperator &I) {
return BinaryOperator::CreateAdd(Op0,
Builder.CreateSub(Z, Y, Op1->getName()));
- // (X - (X & Y)) --> (X & ~Y)
- if (match(Op1, m_c_And(m_Value(Y), m_Specific(Op0))))
- return BinaryOperator::CreateAnd(Op0,
- Builder.CreateNot(Y, Y->getName() + ".not"));
-
// Subtracting -1/0 is the same as adding 1/0:
// sub [nsw] Op0, sext(bool Y) -> add [nsw] Op0, zext(bool Y)
// 'nuw' is dropped in favor of the canonical form.
diff --git a/llvm/test/Transforms/InstCombine/sub.ll b/llvm/test/Transforms/InstCombine/sub.ll
index 1b8ef83f7b63..dc1426c21e5f 100644
--- a/llvm/test/Transforms/InstCombine/sub.ll
+++ b/llvm/test/Transforms/InstCombine/sub.ll
@@ -1531,7 +1531,7 @@ define i8 @test75(i8 %x) {
; CHECK-LABEL: @test75(
; CHECK-NEXT: [[T0:%.*]] = and i8 [[X:%.*]], -8
; CHECK-NEXT: call void @use8(i8 [[T0]])
-; CHECK-NEXT: [[T1:%.*]] = sub i8 [[X]], [[T0]]
+; CHECK-NEXT: [[T1:%.*]] = and i8 [[X]], 7
; CHECK-NEXT: ret i8 [[T1]]
;
%t0 = and i8 %x, -8
diff --git a/llvm/test/Transforms/LoopUnroll/runtime-loop-multiple-exits.ll b/llvm/test/Transforms/LoopUnroll/runtime-loop-multiple-exits.ll
index e91bd2eff9c0..f905355f1a8a 100644
--- a/llvm/test/Transforms/LoopUnroll/runtime-loop-multiple-exits.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-loop-multiple-exits.ll
@@ -19,7 +19,7 @@ define void @test1(i64 %trip, i1 %cond) {
; EPILOG-NEXT: [[TMP1:%.*]] = icmp ult i64 [[TMP0]], 7
; EPILOG-NEXT: br i1 [[TMP1]], label %exit2.loopexit.unr-lcssa, label [[ENTRY_NEW:%.*]]
; EPILOG: entry.new:
-; EPILOG-NEXT: [[UNROLL_ITER:%.*]] = sub i64 [[TRIP]], [[XTRAITER]]
+; EPILOG-NEXT: [[UNROLL_ITER:%.*]] = and i64 [[TRIP]], -8
; EPILOG-NEXT: br label [[LOOP_HEADER:%.*]]
; EPILOG: loop_latch.epil:
; EPILOG-NEXT: %epil.iter.sub = add i64 %epil.iter, -1
@@ -147,7 +147,7 @@ define void @test3(i64 %trip, i64 %add) {
; EPILOG-NEXT: [[TMP1:%.*]] = icmp ult i64 [[TMP0]], 7
; EPILOG-NEXT: br i1 [[TMP1]], label %exit2.loopexit.unr-lcssa, label [[ENTRY_NEW:%.*]]
; EPILOG: entry.new:
-; EPILOG-NEXT: %unroll_iter = sub i64 [[TRIP]], [[XTRAITER]]
+; EPILOG-NEXT: %unroll_iter = and i64 [[TRIP]], -8
; EPILOG-NEXT: br label [[LOOP_HEADER:%.*]]
; EPILOG: loop_header:
; EPILOG-NEXT: %sum = phi i64 [ 0, %entry.new ], [ %sum.next.7, %loop_latch.7 ]
diff --git a/llvm/test/Transforms/LoopUnroll/runtime-unroll-remainder.ll b/llvm/test/Transforms/LoopUnroll/runtime-unroll-remainder.ll
index b85e09bc3bc8..6ea1b5409c74 100644
--- a/llvm/test/Transforms/LoopUnroll/runtime-unroll-remainder.ll
+++ b/llvm/test/Transforms/LoopUnroll/runtime-unroll-remainder.ll
@@ -21,7 +21,7 @@ for.cond.cleanup:
; CHECK: br i1 [[CMP]], label %[[CLEANUP:.*]], label %for.body.lr.ph.new
; CHECK-LABEL: for.body.lr.ph.new:
-; CHECK: %unroll_iter = sub nsw i64 %wide.trip.count, %xtraiter
+; CHECK: %unroll_iter = and i64 %wide.trip.count, 4294967292
; CHECK: br label %for.body
; CHECK: [[CLEANUP]]:
diff --git a/llvm/test/Transforms/LoopVectorize/X86/float-induction-x86.ll b/llvm/test/Transforms/LoopVectorize/X86/float-induction-x86.ll
index 3d579915566b..4ad2db4c5950 100644
--- a/llvm/test/Transforms/LoopVectorize/X86/float-induction-x86.ll
+++ b/llvm/test/Transforms/LoopVectorize/X86/float-induction-x86.ll
@@ -35,7 +35,7 @@ define void @fp_iv_loop1(float* noalias nocapture %A, i32 %N) #0 {
; AUTO_VEC-NEXT: [[TMP4:%.*]] = icmp ult i64 [[TMP1]], 96
; AUTO_VEC-NEXT: br i1 [[TMP4]], label [[MIDDLE_BLOCK_UNR_LCSSA:%.*]], label [[VECTOR_PH_NEW:%.*]]
; AUTO_VEC: vector.ph.new:
-; AUTO_VEC-NEXT: [[UNROLL_ITER:%.*]] = sub nsw i64 [[TMP3]], [[XTRAITER]]
+; AUTO_VEC-NEXT: [[UNROLL_ITER:%.*]] = and i64 [[TMP3]], 1152921504606846972
; AUTO_VEC-NEXT: br label [[VECTOR_BODY:%.*]]
; AUTO_VEC: vector.body:
; AUTO_VEC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[INDEX_NEXT_3:%.*]], [[VECTOR_BODY]] ]
@@ -203,7 +203,7 @@ define void @fp_iv_loop2(float* noalias nocapture %A, i32 %N) #1 {
; AUTO_VEC-NEXT: [[TMP1:%.*]] = icmp ult i64 [[TMP0]], 7
; AUTO_VEC-NEXT: br i1 [[TMP1]], label [[FOR_END_LOOPEXIT_UNR_LCSSA:%.*]], label [[FOR_BODY_PREHEADER_NEW:%.*]]
; AUTO_VEC: for.body.preheader.new:
-; AUTO_VEC-NEXT: [[UNROLL_ITER:%.*]] = sub nsw i64 [[ZEXT]], [[XTRAITER]]
+; AUTO_VEC-NEXT: [[UNROLL_ITER:%.*]] = and i64 [[ZEXT]], 4294967288
; AUTO_VEC-NEXT: br label [[FOR_BODY:%.*]]
; AUTO_VEC: for.body:
; AUTO_VEC-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[INDVARS_IV_NEXT_7:%.*]], [[FOR_BODY]] ]
@@ -306,7 +306,7 @@ define double @external_use_with_fast_math(double* %a, i64 %n) {
; AUTO_VEC-NEXT: [[TMP5:%.*]] = icmp ult i64 [[TMP2]], 48
; AUTO_VEC-NEXT: br i1 [[TMP5]], label [[MIDDLE_BLOCK_UNR_LCSSA:%.*]], label [[VECTOR_PH_NEW:%.*]]
; AUTO_VEC: vector.ph.new:
-; AUTO_VEC-NEXT: [[UNROLL_ITER:%.*]] = sub nsw i64 [[TMP4]], [[XTRAITER]]
+; AUTO_VEC-NEXT: [[UNROLL_ITER:%.*]] = and i64 [[TMP4]], 2305843009213693948
; AUTO_VEC-NEXT: br label [[VECTOR_BODY:%.*]]
; AUTO_VEC: vector.body:
; AUTO_VEC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH_NEW]] ], [ [[INDEX_NEXT_3:%.*]], [[VECTOR_BODY]] ]
@@ -459,7 +459,7 @@ define double @external_use_without_fast_math(double* %a, i64 %n) {
; AUTO_VEC-NEXT: [[TMP2:%.*]] = icmp ult i64 [[TMP1]], 7
; AUTO_VEC-NEXT: br i1 [[TMP2]], label [[FOR_END_UNR_LCSSA:%.*]], label [[ENTRY_NEW:%.*]]
; AUTO_VEC: entry.new:
-; AUTO_VEC-NEXT: [[UNROLL_ITER:%.*]] = sub nsw i64 [[SMAX]], [[XTRAITER]]
+; AUTO_VEC-NEXT: [[UNROLL_ITER:%.*]] = and i64 [[SMAX]], 9223372036854775800
; AUTO_VEC-NEXT: br label [[FOR_BODY:%.*]]
; AUTO_VEC: for.body:
; AUTO_VEC-NEXT: [[I:%.*]] = phi i64 [ 0, [[ENTRY_NEW]] ], [ [[I_NEXT_7:%.*]], [[FOR_BODY]] ]
diff --git a/llvm/test/Transforms/PhaseOrdering/reassociate-after-unroll.ll b/llvm/test/Transforms/PhaseOrdering/reassociate-after-unroll.ll
index 0441e8a93659..aff7dccb0b1e 100644
--- a/llvm/test/Transforms/PhaseOrdering/reassociate-after-unroll.ll
+++ b/llvm/test/Transforms/PhaseOrdering/reassociate-after-unroll.ll
@@ -19,7 +19,7 @@ define dso_local i64 @func(i64 %blah, i64 %limit) #0 {
; CHECK-NEXT: [[TMP1:%.*]] = icmp ult i64 [[TMP0]], 7
; CHECK-NEXT: br i1 [[TMP1]], label [[FOR_COND_CLEANUP_LOOPEXIT_UNR_LCSSA:%.*]], label [[FOR_BODY_LR_PH_NEW:%.*]]
; CHECK: for.body.lr.ph.new:
-; CHECK-NEXT: [[UNROLL_ITER:%.*]] = sub i64 [[LIMIT]], [[XTRAITER]]
+; CHECK-NEXT: [[UNROLL_ITER:%.*]] = and i64 [[LIMIT]], -8
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
; CHECK: for.cond.cleanup.loopexit.unr-lcssa:
; CHECK-NEXT: [[ADD_LCSSA_PH:%.*]] = phi i64 [ undef, [[FOR_BODY_LR_PH]] ], [ [[ADD_7:%.*]], [[FOR_BODY]] ]
@@ -66,7 +66,7 @@ define dso_local i64 @func(i64 %blah, i64 %limit) #0 {
; NPM-NEXT: [[TMP1:%.*]] = icmp ult i64 [[TMP0]], 7
; NPM-NEXT: br i1 [[TMP1]], label [[FOR_COND_CLEANUP_LOOPEXIT_UNR_LCSSA:%.*]], label [[FOR_BODY_LR_PH_NEW:%.*]]
; NPM: for.body.lr.ph.new:
-; NPM-NEXT: [[UNROLL_ITER:%.*]] = sub i64 [[LIMIT]], [[XTRAITER]]
+; NPM-NEXT: [[UNROLL_ITER:%.*]] = and i64 [[LIMIT]], -8
; NPM-NEXT: [[AND_0:%.*]] = and i64 [[CONV]], 1
; NPM-NEXT: br label [[FOR_BODY:%.*]]
; NPM: for.cond.cleanup.loopexit.unr-lcssa:
More information about the llvm-commits
mailing list