Commit b7096f76 authored by Tom Stellard's avatar Tom Stellard
Browse files

Merging r348462: 

------------------------------------------------------------------------
r348462 | lebedevri | 2018-12-06 00:14:24 -0800 (Thu, 06 Dec 2018) | 13 lines

[InstCombine] foldICmpWithLowBitMaskedVal(): don't miscompile -1 vector elts

I was finally able to quantify what i thought was missing in the fix,
it was vector constants. If we have a scalar (and %x, -1),
it will be instsimplified before we reach this code,
but if it is a vector, we may still have a -1 element.

Thus, we want to avoid the fold if *at least one* element is -1.
Or in other words, ignoring the undef elements, no sign bits
should be set. Thus, m_NonNegative().

A follow-up for rL348181
https://bugs.llvm.org/show_bug.cgi?id=39861
------------------------------------------------------------------------

llvm-svn: 348538
parent 9e856fab

llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp

+4 −0
Original line number Diff line number Diff line
@@ -2926,6 +2926,8 @@ static Value *foldICmpWithLowBitMaskedVal(ICmpInst &I, 
      return nullptr;         // Ignore the other case.

    if (!match(M, m_Constant())) // Can not do this fold with non-constant.

      return nullptr;

    if (!match(M, m_NonNegative())) // Must not have any -1 vector elements.

      return nullptr;

    DstPred = ICmpInst::Predicate::ICMP_SLE;

    break;

  case ICmpInst::Predicate::ICMP_SLT:
@@ -2934,6 +2936,8 @@ static Value *foldICmpWithLowBitMaskedVal(ICmpInst &I, 
      return nullptr;         // Ignore the other case.

    if (!match(M, m_Constant())) // Can not do this fold with non-constant.

      return nullptr;

    if (!match(M, m_NonNegative())) // Must not have any -1 vector elements.

      return nullptr;

    DstPred = ICmpInst::Predicate::ICMP_SGT;

    break;

  case ICmpInst::Predicate::ICMP_SLE:

llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-sge-to-icmp-sle.ll

+6 −4
Original line number Diff line number Diff line
@@ -205,8 +205,9 @@ define i1 @nv(i8 %x, i8 %y) { 


define <2 x i1> @n3_vec(<2 x i8> %x) {

; CHECK-LABEL: @n3_vec(

; CHECK-NEXT:    [[TMP1:%.*]] = icmp slt <2 x i8> [[X:%.*]], <i8 4, i8 0>

; CHECK-NEXT:    ret <2 x i1> [[TMP1]]

; CHECK-NEXT:    [[TMP0:%.*]] = and <2 x i8> [[X:%.*]], <i8 3, i8 -1>

; CHECK-NEXT:    [[RET:%.*]] = icmp sge <2 x i8> [[TMP0]], [[X]]

; CHECK-NEXT:    ret <2 x i1> [[RET]]

;

  %tmp0 = and <2 x i8> %x, <i8 3, i8 -1>

  %ret = icmp sge <2 x i8> %tmp0, %x
@@ -215,8 +216,9 @@ define <2 x i1> @n3_vec(<2 x i8> %x) { 


define <3 x i1> @n4_vec(<3 x i8> %x) {

; CHECK-LABEL: @n4_vec(

; CHECK-NEXT:    [[TMP1:%.*]] = icmp slt <3 x i8> [[X:%.*]], <i8 4, i8 undef, i8 0>

; CHECK-NEXT:    ret <3 x i1> [[TMP1]]

; CHECK-NEXT:    [[TMP0:%.*]] = and <3 x i8> [[X:%.*]], <i8 3, i8 undef, i8 -1>

; CHECK-NEXT:    [[RET:%.*]] = icmp sge <3 x i8> [[TMP0]], [[X]]

; CHECK-NEXT:    ret <3 x i1> [[RET]]

;

  %tmp0 = and <3 x i8> %x, <i8 3, i8 undef, i8 -1>

  %ret = icmp sge <3 x i8> %tmp0, %x

llvm/test/Transforms/InstCombine/canonicalize-constant-low-bit-mask-and-icmp-slt-to-icmp-sgt.ll

+6 −4
Original line number Diff line number Diff line
@@ -205,8 +205,9 @@ define i1 @nv(i8 %x, i8 %y) { 


define <2 x i1> @n3(<2 x i8> %x) {

; CHECK-LABEL: @n3(

; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt <2 x i8> [[X:%.*]], <i8 3, i8 -1>

; CHECK-NEXT:    ret <2 x i1> [[TMP1]]

; CHECK-NEXT:    [[TMP0:%.*]] = and <2 x i8> [[X:%.*]], <i8 3, i8 -1>

; CHECK-NEXT:    [[RET:%.*]] = icmp slt <2 x i8> [[TMP0]], [[X]]

; CHECK-NEXT:    ret <2 x i1> [[RET]]

;

  %tmp0 = and <2 x i8> %x, <i8 3, i8 -1>

  %ret = icmp slt <2 x i8> %tmp0, %x
@@ -215,8 +216,9 @@ define <2 x i1> @n3(<2 x i8> %x) { 


define <3 x i1> @n4(<3 x i8> %x) {

; CHECK-LABEL: @n4(

; CHECK-NEXT:    [[TMP1:%.*]] = icmp sgt <3 x i8> [[X:%.*]], <i8 3, i8 undef, i8 -1>

; CHECK-NEXT:    ret <3 x i1> [[TMP1]]

; CHECK-NEXT:    [[TMP0:%.*]] = and <3 x i8> [[X:%.*]], <i8 3, i8 undef, i8 -1>

; CHECK-NEXT:    [[RET:%.*]] = icmp slt <3 x i8> [[TMP0]], [[X]]

; CHECK-NEXT:    ret <3 x i1> [[RET]]

;

  %tmp0 = and <3 x i8> %x, <i8 3, i8 undef, i8 -1>

  %ret = icmp slt <3 x i8> %tmp0, %x