[llvm] 4e34915 - [GuardWidening] Make sure widened condition operands are available at insertion point

Dmitry Makogon via llvm-commits llvm-commits at lists.llvm.org
Tue Feb 28 01:30:59 PST 2023


Author: Dmitry Makogon
Date: 2023-02-28T16:30:09+07:00
New Revision: 4e34915b6d27d950b18dea010303de05a0a18aef

URL: https://github.com/llvm/llvm-project/commit/4e34915b6d27d950b18dea010303de05a0a18aef
DIFF: https://github.com/llvm/llvm-project/commit/4e34915b6d27d950b18dea010303de05a0a18aef.diff

LOG: [GuardWidening] Make sure widened condition operands are available at insertion point

This fixes a possible issue when we could hoist an instruction up to a widenable
condition intrinsic call without making sure its operands are available at
hoisiting point.

Recently insertion point finding algorithm changed a bit, so this availability
check became necessary.

Verifier would crash after we handled the following special case:
  L >u C0 && L >u C1  ->  L >u max(C0, C1),
Previously we would insert the new condition right before the widenable condition
branch where all L operands were available.
Now we may choose the widenable condition intrinsic call as insertion point and it may
happen so that the L operands are computed after the call, so we have to make sure that
L operands are available at the point we want to insert it.

Differential Revision: https://reviews.llvm.org/D144944

Added: 
    

Modified: 
    llvm/lib/Transforms/Scalar/GuardWidening.cpp
    llvm/test/Transforms/GuardWidening/widen-cond-with-operands.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/Scalar/GuardWidening.cpp b/llvm/lib/Transforms/Scalar/GuardWidening.cpp
index 199b3a7bbebca..3ded658fd22f1 100644
--- a/llvm/lib/Transforms/Scalar/GuardWidening.cpp
+++ b/llvm/lib/Transforms/Scalar/GuardWidening.cpp
@@ -552,6 +552,8 @@ bool GuardWideningImpl::widenCondCommon(Value *Cond0, Value *Cond1,
           if (InsertPt) {
             ConstantInt *NewRHS =
                 ConstantInt::get(Cond0->getContext(), NewRHSAP);
+            assert(isAvailableAt(LHS, InsertPt) && "must be");
+            makeAvailableAt(LHS, InsertPt);
             Result = new ICmpInst(InsertPt, Pred, LHS, NewRHS, "wide.chk");
           }
           return true;

diff  --git a/llvm/test/Transforms/GuardWidening/widen-cond-with-operands.ll b/llvm/test/Transforms/GuardWidening/widen-cond-with-operands.ll
index 398956e6c9350..dd3c15a44912c 100644
--- a/llvm/test/Transforms/GuardWidening/widen-cond-with-operands.ll
+++ b/llvm/test/Transforms/GuardWidening/widen-cond-with-operands.ll
@@ -1,13 +1,31 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -S -passes=guard-widening,verify < %s | FileCheck %s
 target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128-ni:1-p2:32:8:8:32-ni:2"
 target triple = "x86_64-unknown-linux-gnu"
 
-; REQUIRES: asserts
-; XFAIL: *
-
-; GuardWidening moves 'check' instruction before 'wc1' without its operand ('zero').
-; This is incorrect as 'zero' is not available at that point.
+; Make sure GuardWidening moves 'check' instruction before 'wc1' together with its operand - 'zero'.
 define void @foo() {
+; CHECK-LABEL: @foo(
+; CHECK-NEXT:  bb:
+; CHECK-NEXT:    [[ZERO:%.*]] = add i32 0, 0
+; CHECK-NEXT:    [[WIDE_CHK:%.*]] = icmp ult i32 [[ZERO]], 0
+; CHECK-NEXT:    [[WC1:%.*]] = call i1 @llvm.experimental.widenable.condition()
+; CHECK-NEXT:    [[WC2:%.*]] = call i1 @llvm.experimental.widenable.condition()
+; CHECK-NEXT:    [[CHECK:%.*]] = icmp ult i32 [[ZERO]], 0
+; CHECK-NEXT:    [[C2:%.*]] = and i1 [[CHECK]], [[WC2]]
+; CHECK-NEXT:    [[C1:%.*]] = and i1 [[WIDE_CHK]], [[WC1]]
+; CHECK-NEXT:    br i1 [[C1]], label [[BB6:%.*]], label [[BB9:%.*]]
+; CHECK:       bb6:
+; CHECK-NEXT:    br i1 true, label [[BB7:%.*]], label [[BB8:%.*]]
+; CHECK:       bb7:
+; CHECK-NEXT:    ret void
+; CHECK:       bb8:
+; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid(i32 0) [ "deopt"() ]
+; CHECK-NEXT:    ret void
+; CHECK:       bb9:
+; CHECK-NEXT:    call void (...) @llvm.experimental.deoptimize.isVoid(i32 0) [ "deopt"() ]
+; CHECK-NEXT:    ret void
+;
 bb:
   %wc1 = call i1 @llvm.experimental.widenable.condition()
   %wc2 = call i1 @llvm.experimental.widenable.condition()


        


More information about the llvm-commits mailing list