[InstCombine] Check isGuaranteedNotToBeUndef in haveNoCommonBitsSetSp… #74390
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-llvm-analysis @llvm/pr-subscribers-llvm-transforms Author: Craig Topper (topperc) Changes…ecialCases. It's not safe for InstCombine to add disjoint metadata when converting add to or otherwise. I've added noundef attribute to preserve existing test behavior. A few questions. Do we only want to do this when the caller is going to set the disjoint flag? Alternatively, we could freeze the inputs? Patch is 29.43 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/74390.diff 9 Files Affected:
diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index 1f09d912f7339..ad1806ee231cf 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -187,36 +187,44 @@ KnownBits llvm::computeKnownBits(const Value *V, const APInt &DemandedElts,
}
static bool haveNoCommonBitsSetSpecialCases(const Value *LHS,
- const Value *RHS) {
+ const Value *RHS,
+ const SimplifyQuery &SQ) {
// Look for an inverted mask: (X & ~M) op (Y & M).
{
Value *M;
if (match(LHS, m_c_And(m_Not(m_Value(M)), m_Value())) &&
- match(RHS, m_c_And(m_Specific(M), m_Value())))
+ match(RHS, m_c_And(m_Specific(M), m_Value())) &&
+ isGuaranteedNotToBeUndef(M, SQ.AC, SQ.CxtI, SQ.DT))
return true;
}
// X op (Y & ~X)
- if (match(RHS, m_c_And(m_Not(m_Specific(LHS)), m_Value())))
+ if (match(RHS, m_c_And(m_Not(m_Specific(LHS)), m_Value())) &&
+ isGuaranteedNotToBeUndef(LHS, SQ.AC, SQ.CxtI, SQ.DT))
return true;
// X op ((X & Y) ^ Y) -- this is the canonical form of the previous pattern
// for constant Y.
Value *Y;
- if (match(RHS, m_c_Xor(m_c_And(m_Specific(LHS), m_Value(Y)), m_Deferred(Y))))
+ if (match(RHS, m_c_Xor(m_c_And(m_Specific(LHS), m_Value(Y)), m_Deferred(Y))) &&
+ isGuaranteedNotToBeUndef(LHS, SQ.AC, SQ.CxtI, SQ.DT) &&
+ isGuaranteedNotToBeUndef(Y, SQ.AC, SQ.CxtI, SQ.DT))
return true;
// Peek through extends to find a 'not' of the other side:
// (ext Y) op ext(~Y)
if (match(LHS, m_ZExtOrSExt(m_Value(Y))) &&
- match(RHS, m_ZExtOrSExt(m_Not(m_Specific(Y)))))
+ match(RHS, m_ZExtOrSExt(m_Not(m_Specific(Y)))) &&
+ isGuaranteedNotToBeUndef(Y, SQ.AC, SQ.CxtI, SQ.DT))
return true;
// Look for: (A & B) op ~(A | B)
{
Value *A, *B;
if (match(LHS, m_And(m_Value(A), m_Value(B))) &&
- match(RHS, m_Not(m_c_Or(m_Specific(A), m_Specific(B)))))
+ match(RHS, m_Not(m_c_Or(m_Specific(A), m_Specific(B)))) &&
+ isGuaranteedNotToBeUndef(A, SQ.AC, SQ.CxtI, SQ.DT) &&
+ isGuaranteedNotToBeUndef(B, SQ.AC, SQ.CxtI, SQ.DT))
return true;
}
@@ -234,8 +242,8 @@ bool llvm::haveNoCommonBitsSet(const WithCache<const Value *> &LHSCache,
assert(LHS->getType()->isIntOrIntVectorTy() &&
"LHS and RHS should be integers");
- if (haveNoCommonBitsSetSpecialCases(LHS, RHS) ||
- haveNoCommonBitsSetSpecialCases(RHS, LHS))
+ if (haveNoCommonBitsSetSpecialCases(LHS, RHS, SQ) ||
+ haveNoCommonBitsSetSpecialCases(RHS, LHS, SQ))
return true;
return KnownBits::haveNoCommonBitsSet(LHSCache.getKnownBits(SQ),
diff --git a/llvm/test/Transforms/InstCombine/add.ll b/llvm/test/Transforms/InstCombine/add.ll
index db9eafe998ebb..b9280016b1e62 100644
--- a/llvm/test/Transforms/InstCombine/add.ll
+++ b/llvm/test/Transforms/InstCombine/add.ll
@@ -1511,7 +1511,7 @@ define i8 @add_like_or_t2_extrause(i8 %x) {
ret i8 %r
}
-define i8 @add_and_xor(i8 %x, i8 %y) {
+define i8 @add_and_xor(i8 noundef %x, i8 %y) {
; CHECK-LABEL: @add_and_xor(
; CHECK-NEXT: [[ADD:%.*]] = or i8 [[Y:%.*]], [[X:%.*]]
; CHECK-NEXT: ret i8 [[ADD]]
@@ -1548,7 +1548,7 @@ define i8 @add_and_xor_wrong_op(i8 %x, i8 %y, i8 %z) {
ret i8 %add
}
-define i8 @add_and_xor_commuted1(i8 %x, i8 %_y) {
+define i8 @add_and_xor_commuted1(i8 noundef %x, i8 %_y) {
; CHECK-LABEL: @add_and_xor_commuted1(
; CHECK-NEXT: [[Y:%.*]] = udiv i8 42, [[_Y:%.*]]
; CHECK-NEXT: [[ADD:%.*]] = or i8 [[Y]], [[X:%.*]]
@@ -1561,7 +1561,7 @@ define i8 @add_and_xor_commuted1(i8 %x, i8 %_y) {
ret i8 %add
}
-define i8 @add_and_xor_commuted2(i8 %_x, i8 %y) {
+define i8 @add_and_xor_commuted2(i8 noundef %_x, i8 %y) {
; CHECK-LABEL: @add_and_xor_commuted2(
; CHECK-NEXT: [[X:%.*]] = udiv i8 42, [[_X:%.*]]
; CHECK-NEXT: [[ADD:%.*]] = or i8 [[X]], [[Y:%.*]]
@@ -1574,7 +1574,7 @@ define i8 @add_and_xor_commuted2(i8 %_x, i8 %y) {
ret i8 %add
}
-define i8 @add_and_xor_commuted3(i8 %_x, i8 %_y) {
+define i8 @add_and_xor_commuted3(i8 noundef %_x, i8 %_y) {
; CHECK-LABEL: @add_and_xor_commuted3(
; CHECK-NEXT: [[X:%.*]] = udiv i8 42, [[_X:%.*]]
; CHECK-NEXT: [[Y:%.*]] = udiv i8 42, [[_Y:%.*]]
@@ -1589,7 +1589,7 @@ define i8 @add_and_xor_commuted3(i8 %_x, i8 %_y) {
ret i8 %add
}
-define i8 @add_and_xor_extra_use(i8 %x, i8 %y) {
+define i8 @add_and_xor_extra_use(i8 noundef %x, i8 %y) {
; CHECK-LABEL: @add_and_xor_extra_use(
; CHECK-NEXT: [[XOR:%.*]] = xor i8 [[X:%.*]], -1
; CHECK-NEXT: call void @use(i8 [[XOR]])
@@ -1606,7 +1606,7 @@ define i8 @add_and_xor_extra_use(i8 %x, i8 %y) {
ret i8 %add
}
-define i8 @add_xor_and_const(i8 %x) {
+define i8 @add_xor_and_const(i8 noundef %x) {
; CHECK-LABEL: @add_xor_and_const(
; CHECK-NEXT: [[ADD:%.*]] = or i8 [[X:%.*]], 42
; CHECK-NEXT: ret i8 [[ADD]]
@@ -1630,7 +1630,7 @@ define i8 @add_xor_and_const_wrong_const(i8 %x) {
ret i8 %add
}
-define i8 @add_xor_and_var(i8 %x, i8 %y) {
+define i8 @add_xor_and_var(i8 noundef %x, i8 noundef %y) {
; CHECK-LABEL: @add_xor_and_var(
; CHECK-NEXT: [[AND:%.*]] = and i8 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: call void @use(i8 [[AND]])
@@ -1674,7 +1674,7 @@ define i8 @add_xor_and_var_wrong_op2(i8 %x, i8 %y, i8 %z) {
ret i8 %add
}
-define i8 @add_xor_and_var_commuted1(i8 %x, i8 %y) {
+define i8 @add_xor_and_var_commuted1(i8 noundef %x, i8 noundef %y) {
; CHECK-LABEL: @add_xor_and_var_commuted1(
; CHECK-NEXT: [[AND:%.*]] = and i8 [[Y:%.*]], [[X:%.*]]
; CHECK-NEXT: call void @use(i8 [[AND]])
@@ -1688,7 +1688,7 @@ define i8 @add_xor_and_var_commuted1(i8 %x, i8 %y) {
ret i8 %add
}
-define i8 @add_xor_and_var_commuted2(i8 %_x, i8 %_y) {
+define i8 @add_xor_and_var_commuted2(i8 noundef %_x, i8 noundef %_y) {
; CHECK-LABEL: @add_xor_and_var_commuted2(
; CHECK-NEXT: [[X:%.*]] = udiv i8 42, [[_X:%.*]]
; CHECK-NEXT: [[Y:%.*]] = udiv i8 42, [[_Y:%.*]]
@@ -1706,7 +1706,7 @@ define i8 @add_xor_and_var_commuted2(i8 %_x, i8 %_y) {
ret i8 %add
}
-define i8 @add_xor_and_var_commuted3(i8 %x, i8 %_y) {
+define i8 @add_xor_and_var_commuted3(i8 noundef %x, i8 noundef %_y) {
; CHECK-LABEL: @add_xor_and_var_commuted3(
; CHECK-NEXT: [[Y:%.*]] = udiv i8 42, [[_Y:%.*]]
; CHECK-NEXT: [[AND:%.*]] = and i8 [[Y]], [[X:%.*]]
@@ -1722,7 +1722,7 @@ define i8 @add_xor_and_var_commuted3(i8 %x, i8 %_y) {
ret i8 %add
}
-define i8 @add_xor_and_var_commuted4(i8 %_x, i8 %y) {
+define i8 @add_xor_and_var_commuted4(i8 noundef %_x, i8 noundef %y) {
; CHECK-LABEL: @add_xor_and_var_commuted4(
; CHECK-NEXT: [[X:%.*]] = udiv i8 42, [[_X:%.*]]
; CHECK-NEXT: [[AND:%.*]] = and i8 [[X]], [[Y:%.*]]
@@ -1738,7 +1738,7 @@ define i8 @add_xor_and_var_commuted4(i8 %_x, i8 %y) {
ret i8 %add
}
-define i8 @add_xor_and_var_commuted5(i8 %_x, i8 %_y) {
+define i8 @add_xor_and_var_commuted5(i8 noundef %_x, i8 noundef %_y) {
; CHECK-LABEL: @add_xor_and_var_commuted5(
; CHECK-NEXT: [[X:%.*]] = udiv i8 42, [[_X:%.*]]
; CHECK-NEXT: [[Y:%.*]] = udiv i8 42, [[_Y:%.*]]
@@ -1756,7 +1756,7 @@ define i8 @add_xor_and_var_commuted5(i8 %_x, i8 %_y) {
ret i8 %add
}
-define i8 @add_xor_and_var_commuted6(i8 %_x, i8 %_y) {
+define i8 @add_xor_and_var_commuted6(i8 noundef %_x, i8 noundef %_y) {
; CHECK-LABEL: @add_xor_and_var_commuted6(
; CHECK-NEXT: [[X:%.*]] = udiv i8 42, [[_X:%.*]]
; CHECK-NEXT: [[Y:%.*]] = udiv i8 42, [[_Y:%.*]]
@@ -1774,7 +1774,7 @@ define i8 @add_xor_and_var_commuted6(i8 %_x, i8 %_y) {
ret i8 %add
}
-define i8 @add_xor_and_var_commuted7(i8 %_x, i8 %_y) {
+define i8 @add_xor_and_var_commuted7(i8 noundef %_x, i8 noundef %_y) {
; CHECK-LABEL: @add_xor_and_var_commuted7(
; CHECK-NEXT: [[X:%.*]] = udiv i8 42, [[_X:%.*]]
; CHECK-NEXT: [[Y:%.*]] = udiv i8 42, [[_Y:%.*]]
@@ -1792,7 +1792,7 @@ define i8 @add_xor_and_var_commuted7(i8 %_x, i8 %_y) {
ret i8 %add
}
-define i8 @add_xor_and_var_extra_use(i8 %x, i8 %y) {
+define i8 @add_xor_and_var_extra_use(i8 noundef %x, i8 noundef %y) {
; CHECK-LABEL: @add_xor_and_var_extra_use(
; CHECK-NEXT: [[AND:%.*]] = and i8 [[X:%.*]], [[Y:%.*]]
; CHECK-NEXT: call void @use(i8 [[AND]])
@@ -2573,7 +2573,7 @@ define <vscale x 1 x i32> @add_to_or_scalable(<vscale x 1 x i32> %in) {
ret <vscale x 1 x i32> %add
}
-define i5 @zext_zext_not(i3 %x) {
+define i5 @zext_zext_not(i3 noundef %x) {
; CHECK-LABEL: @zext_zext_not(
; CHECK-NEXT: ret i5 7
;
@@ -2584,7 +2584,7 @@ define i5 @zext_zext_not(i3 %x) {
ret i5 %r
}
-define <2 x i5> @zext_zext_not_commute(<2 x i3> %x) {
+define <2 x i5> @zext_zext_not_commute(<2 x i3> noundef %x) {
; CHECK-LABEL: @zext_zext_not_commute(
; CHECK-NEXT: ret <2 x i5> <i5 7, i5 7>
;
@@ -2595,7 +2595,7 @@ define <2 x i5> @zext_zext_not_commute(<2 x i3> %x) {
ret <2 x i5> %r
}
-define i9 @sext_sext_not(i3 %x) {
+define i9 @sext_sext_not(i3 noundef %x) {
; CHECK-LABEL: @sext_sext_not(
; CHECK-NEXT: ret i9 -1
;
@@ -2606,7 +2606,7 @@ define i9 @sext_sext_not(i3 %x) {
ret i9 %r
}
-define i8 @sext_sext_not_commute(i3 %x) {
+define i8 @sext_sext_not_commute(i3 noundef %x) {
; CHECK-LABEL: @sext_sext_not_commute(
; CHECK-NEXT: [[SX:%.*]] = sext i3 [[X:%.*]] to i8
; CHECK-NEXT: call void @use(i8 [[SX]])
@@ -2621,7 +2621,7 @@ define i8 @sext_sext_not_commute(i3 %x) {
ret i8 %r
}
-define i5 @zext_sext_not(i4 %x) {
+define i5 @zext_sext_not(i4 noundef %x) {
; CHECK-LABEL: @zext_sext_not(
; CHECK-NEXT: [[ZX:%.*]] = zext i4 [[X:%.*]] to i5
; CHECK-NEXT: [[NOTX:%.*]] = xor i4 [[X]], -1
@@ -2636,7 +2636,7 @@ define i5 @zext_sext_not(i4 %x) {
ret i5 %r
}
-define i8 @zext_sext_not_commute(i4 %x) {
+define i8 @zext_sext_not_commute(i4 noundef %x) {
; CHECK-LABEL: @zext_sext_not_commute(
; CHECK-NEXT: [[ZX:%.*]] = zext i4 [[X:%.*]] to i8
; CHECK-NEXT: call void @use(i8 [[ZX]])
@@ -2655,7 +2655,7 @@ define i8 @zext_sext_not_commute(i4 %x) {
ret i8 %r
}
-define i9 @sext_zext_not(i4 %x) {
+define i9 @sext_zext_not(i4 noundef %x) {
; CHECK-LABEL: @sext_zext_not(
; CHECK-NEXT: [[SX:%.*]] = sext i4 [[X:%.*]] to i9
; CHECK-NEXT: [[NOTX:%.*]] = xor i4 [[X]], -1
@@ -2670,7 +2670,7 @@ define i9 @sext_zext_not(i4 %x) {
ret i9 %r
}
-define i9 @sext_zext_not_commute(i4 %x) {
+define i9 @sext_zext_not_commute(i4 noundef %x) {
; CHECK-LABEL: @sext_zext_not_commute(
; CHECK-NEXT: [[SX:%.*]] = sext i4 [[X:%.*]] to i9
; CHECK-NEXT: [[NOTX:%.*]] = xor i4 [[X]], -1
diff --git a/llvm/test/Transforms/InstCombine/and-or-not.ll b/llvm/test/Transforms/InstCombine/and-or-not.ll
index c896c8f100380..ca093eba1b568 100644
--- a/llvm/test/Transforms/InstCombine/and-or-not.ll
+++ b/llvm/test/Transforms/InstCombine/and-or-not.ll
@@ -548,7 +548,7 @@ define i32 @and_to_nxor_multiuse(float %fa, float %fb) {
; (a & b) | ~(a | b) --> ~(a ^ b)
; TODO: this increases instruction count if the pieces have additional users
-define i32 @or_to_nxor_multiuse(i32 %a, i32 %b) {
+define i32 @or_to_nxor_multiuse(i32 noundef %a, i32 noundef %b) {
; CHECK-LABEL: @or_to_nxor_multiuse(
; CHECK-NEXT: [[AND:%.*]] = and i32 [[A:%.*]], [[B:%.*]]
; CHECK-NEXT: [[OR:%.*]] = or i32 [[A]], [[B]]
diff --git a/llvm/test/Transforms/InstCombine/logical-select.ll b/llvm/test/Transforms/InstCombine/logical-select.ll
index 31848bc911892..c3eec5eebf2eb 100644
--- a/llvm/test/Transforms/InstCombine/logical-select.ll
+++ b/llvm/test/Transforms/InstCombine/logical-select.ll
@@ -762,7 +762,7 @@ define <vscale x 2 x i64> @bitcast_vec_cond_scalable(<vscale x 16 x i1> %cond, <
; Negative test - bitcast of condition from wide source element type cannot be converted to select.
-define <8 x i3> @bitcast_vec_cond_commute1(<3 x i1> %cond, <8 x i3> %pc, <8 x i3> %d) {
+define <8 x i3> @bitcast_vec_cond_commute1(<3 x i1> noundef %cond, <8 x i3> %pc, <8 x i3> %d) {
; CHECK-LABEL: @bitcast_vec_cond_commute1(
; CHECK-NEXT: [[C:%.*]] = mul <8 x i3> [[PC:%.*]], [[PC]]
; CHECK-NEXT: [[S:%.*]] = sext <3 x i1> [[COND:%.*]] to <3 x i8>
@@ -830,7 +830,7 @@ define <2 x i16> @bitcast_vec_cond_commute3(<4 x i8> %cond, <2 x i16> %pc, <2 x
; Don't crash on invalid type for compute signbits.
-define <2 x i64> @bitcast_fp_vec_cond(<2 x double> %s, <2 x i64> %c, <2 x i64> %d) {
+define <2 x i64> @bitcast_fp_vec_cond(<2 x double> noundef %s, <2 x i64> %c, <2 x i64> %d) {
; CHECK-LABEL: @bitcast_fp_vec_cond(
; CHECK-NEXT: [[T9:%.*]] = bitcast <2 x double> [[S:%.*]] to <2 x i64>
; CHECK-NEXT: [[NOTT9:%.*]] = xor <2 x i64> [[T9]], <i64 -1, i64 -1>
@@ -849,7 +849,7 @@ define <2 x i64> @bitcast_fp_vec_cond(<2 x double> %s, <2 x i64> %c, <2 x i64> %
; Wider source type would be ok except poison could leak across elements.
-define <2 x i64> @bitcast_int_vec_cond(i1 %b, <2 x i64> %c, <2 x i64> %d) {
+define <2 x i64> @bitcast_int_vec_cond(i1 noundef %b, <2 x i64> %c, <2 x i64> %d) {
; CHECK-LABEL: @bitcast_int_vec_cond(
; CHECK-NEXT: [[S:%.*]] = sext i1 [[B:%.*]] to i128
; CHECK-NEXT: [[T9:%.*]] = bitcast i128 [[S]] to <2 x i64>
diff --git a/llvm/test/Transforms/InstCombine/masked-merge-add.ll b/llvm/test/Transforms/InstCombine/masked-merge-add.ll
index c4265eb896b20..f655153108a43 100644
--- a/llvm/test/Transforms/InstCombine/masked-merge-add.ll
+++ b/llvm/test/Transforms/InstCombine/masked-merge-add.ll
@@ -16,7 +16,7 @@
; Most basic positive tests
; ============================================================================ ;
-define i32 @p(i32 %x, i32 %y, i32 %m) {
+define i32 @p(i32 %x, i32 %y, i32 noundef %m) {
; CHECK-LABEL: @p(
; CHECK-NEXT: [[AND:%.*]] = and i32 [[X:%.*]], [[M:%.*]]
; CHECK-NEXT: [[NEG:%.*]] = xor i32 [[M]], -1
@@ -31,7 +31,7 @@ define i32 @p(i32 %x, i32 %y, i32 %m) {
ret i32 %ret
}
-define <2 x i32> @p_splatvec(<2 x i32> %x, <2 x i32> %y, <2 x i32> %m) {
+define <2 x i32> @p_splatvec(<2 x i32> %x, <2 x i32> %y, <2 x i32> noundef %m) {
; CHECK-LABEL: @p_splatvec(
; CHECK-NEXT: [[AND:%.*]] = and <2 x i32> [[X:%.*]], [[M:%.*]]
; CHECK-NEXT: [[NEG:%.*]] = xor <2 x i32> [[M]], <i32 -1, i32 -1>
@@ -46,7 +46,7 @@ define <2 x i32> @p_splatvec(<2 x i32> %x, <2 x i32> %y, <2 x i32> %m) {
ret <2 x i32> %ret
}
-define <3 x i32> @p_vec_undef(<3 x i32> %x, <3 x i32> %y, <3 x i32> %m) {
+define <3 x i32> @p_vec_undef(<3 x i32> %x, <3 x i32> %y, <3 x i32> noundef %m) {
; CHECK-LABEL: @p_vec_undef(
; CHECK-NEXT: [[AND:%.*]] = and <3 x i32> [[X:%.*]], [[M:%.*]]
; CHECK-NEXT: [[NEG:%.*]] = xor <3 x i32> [[M]], <i32 -1, i32 undef, i32 -1>
@@ -180,7 +180,7 @@ define <3 x i32> @p_constmask2_vec_undef(<3 x i32> %x, <3 x i32> %y) {
; Used to make sure that the IR complexity sorting does not interfere.
declare i32 @gen32()
-define i32 @p_commutative0(i32 %x, i32 %y, i32 %m) {
+define i32 @p_commutative0(i32 %x, i32 %y, i32 noundef %m) {
; CHECK-LABEL: @p_commutative0(
; CHECK-NEXT: [[AND:%.*]] = and i32 [[M:%.*]], [[X:%.*]]
; CHECK-NEXT: [[NEG:%.*]] = xor i32 [[M]], -1
@@ -195,7 +195,7 @@ define i32 @p_commutative0(i32 %x, i32 %y, i32 %m) {
ret i32 %ret
}
-define i32 @p_commutative1(i32 %x, i32 %m) {
+define i32 @p_commutative1(i32 %x, i32 noundef %m) {
; CHECK-LABEL: @p_commutative1(
; CHECK-NEXT: [[Y:%.*]] = call i32 @gen32()
; CHECK-NEXT: [[AND:%.*]] = and i32 [[X:%.*]], [[M:%.*]]
@@ -212,7 +212,7 @@ define i32 @p_commutative1(i32 %x, i32 %m) {
ret i32 %ret
}
-define i32 @p_commutative2(i32 %x, i32 %y, i32 %m) {
+define i32 @p_commutative2(i32 %x, i32 %y, i32 noundef %m) {
; CHECK-LABEL: @p_commutative2(
; CHECK-NEXT: [[AND:%.*]] = and i32 [[X:%.*]], [[M:%.*]]
; CHECK-NEXT: [[NEG:%.*]] = xor i32 [[M]], -1
@@ -227,7 +227,7 @@ define i32 @p_commutative2(i32 %x, i32 %y, i32 %m) {
ret i32 %ret
}
-define i32 @p_commutative3(i32 %x, i32 %m) {
+define i32 @p_commutative3(i32 %x, i32 noundef %m) {
; CHECK-LABEL: @p_commutative3(
; CHECK-NEXT: [[Y:%.*]] = call i32 @gen32()
; CHECK-NEXT: [[AND:%.*]] = and i32 [[M:%.*]], [[X:%.*]]
@@ -244,7 +244,7 @@ define i32 @p_commutative3(i32 %x, i32 %m) {
ret i32 %ret
}
-define i32 @p_commutative4(i32 %x, i32 %y, i32 %m) {
+define i32 @p_commutative4(i32 %x, i32 %y, i32 noundef %m) {
; CHECK-LABEL: @p_commutative4(
; CHECK-NEXT: [[AND:%.*]] = and i32 [[M:%.*]], [[X:%.*]]
; CHECK-NEXT: [[NEG:%.*]] = xor i32 [[M]], -1
@@ -259,7 +259,7 @@ define i32 @p_commutative4(i32 %x, i32 %y, i32 %m) {
ret i32 %ret
}
-define i32 @p_commutative5(i32 %x, i32 %m) {
+define i32 @p_commutative5(i32 %x, i32 noundef %m) {
; CHECK-LABEL: @p_commutative5(
; CHECK-NEXT: [[Y:%.*]] = call i32 @gen32()
; CHECK-NEXT: [[AND:%.*]] = and i32 [[X:%.*]], [[M:%.*]]
@@ -276,7 +276,7 @@ define i32 @p_commutative5(i32 %x, i32 %m) {
ret i32 %ret
}
-define i32 @p_commutative6(i32 %x, i32 %m) {
+define i32 @p_commutative6(i32 %x, i32 noundef %m) {
; CHECK-LABEL: @p_commutative6(
; CHECK-NEXT: [[Y:%.*]] = call i32 @gen32()
; CHECK-NEXT: [[AND:%.*]] = and i32 [[M:%.*]], [[X:%.*]]
@@ -314,7 +314,7 @@ define i32 @p_constmask_commutative(i32 %x, i32 %y) {
declare void @use32(i32)
-define i32 @n0_oneuse(i32 %x, i32 %y, i32 %m) {
+define i32 @n0_oneuse(i32 %x, i32 %y, i32 noundef %m) {
; CHECK-LABEL: @n0_oneuse(
; CHECK-NEXT: [[AND:%.*]] = and i32 [[X:%.*]], [[M:%.*]]
; CHECK-NEXT: [[NEG:%.*]] = xor i32 [[M]], -1
diff --git a/llvm/test/Transforms/InstCombine/masked-merge-or.ll b/llvm/test/Transforms/InstCombine/masked-merge-or.ll
index 7d67b2d495412..b49ec07706e28 100644
--- a/llvm/test/Transforms/InstCombine/masked-merge-or.ll
+++ b/llvm/test/Transforms/InstCombine/masked-merge-or.ll
@@ -16,7 +16,7 @@
; Most basic positive tests
; ============================================================================ ;
-define i32 @p(i32 %x, i32 %y, i32 %m) {
+define i32 @p(i32 %x, i32 %y, i32 noundef %m) {
; CHECK-LABEL: @p(
; CHECK-NEXT: [[AND:%.*]] = and i32 [[X:%.*]], [[M:%.*]]
; CHECK-NEXT: [[NEG:%.*]] = xor i32 [[M]], -1
@@ -31,7 +31,7 @@ define i32 @p(i32 %x, i32 %y, i32 %m) {
ret i32 %ret
}
-define <2 x i32> @p_splatvec(<2 x i32> %x, <2 x i32> %y, <2 x i32> %m) {
+define <2 x i32> @p_splatvec(<2 x i32> %x, <2 x i32> %y, <2 x i32> noundef %m) {
; CHECK-LABEL: @p_splatvec(
; CHECK-NEXT: [[AND:%.*]] = and <2 x i32> [[X:%.*]], [[M:%.*]]
; CHECK-NEXT: [[NEG:%.*]] = xor <2 x i32> [[M]], <i32 -1, i32 -1>
@@ -46,7 +46,7 @@ define <2 x i32> @p_splatvec(<2 x i32> %x, <2 x i32> %y, <2 x i32> %m) {
ret <2 x i32> %ret
}
-define <3 x i32> @p_vec_undef(<3 x i32> %x, <3 x i32> %y, <3 x i32> %m) {
+define <3 x i32> @p_vec_undef(<3 x i32> %x, <3 x i32> %y, <3 x i32> noundef %m) {
; CHECK-LABEL: @p_vec_undef(
; CHECK-NEXT: [[AND:%.*]] = and <3 x i32> [[X:%.*]], [[M:%.*]]
; CHECK-NEXT: [[NEG:%.*]] = xor <3 x i32> [[M]], <i32 -1, i32 undef, i32 -1>
@@ -180,7 +180,7 @@ define <3 x i32> @p_constmask2_vec_undef(<3 x i32> %x, <3 x i32> %y) {
; Used to make sure that the IR complexity sorting does not interfere.
declare i32 @gen32()
-define i32 @p_commutative0(i32 %x, i32 %y, i32 %m) {
+define i32 @p_commutative0(i32 %x, i32 %y, i32 noundef %m) {
; CHECK-LABEL: @p_commutative0(
; CHECK-NEXT: [[AND:%.*]] = and i32 [[M:%.*]], [[X:%.*]]
; CHECK-NEXT: [[NEG:%.*]] = xor i32 [[M]], -1
@@ -195,7 +195,7 @@ define i32 @p_commutative0(i32 %x, i32 %y, i32 %m) {
ret i32 %ret
}
-define i32 @p_commutative1(i32 %x, i32 %m) {
+define i32 @p_commutative1(i32 %x, i32 noundef %m) {
; CHECK-LABEL: @p_commutative1(
; CHECK-NEXT: [[Y:%.*]] = call i32 @gen32()
; CHECK-NEXT: [[AND:%.*]] = and i32 [[X:%.*]], [[M:%.*]]
@@ -212,7 +212,7 @@ define i32 @p_commutative1(i32 %x, i32 %m) {
ret i32 %ret
}
-define i32 @p_commutative2(i32 %x, i32 %y, i32 %m) {
+define i32 @p_commutative2(i32 %x, i32 %y, i32 noundef %m) {
; CHECK-LABEL: @p_commutative2(
; CHECK-NEXT: [[AND:%.*]] = and i32 [[X:%.*]], [[M:%.*]]
; CHECK-NEXT: [[NEG:%.*]] = xor i32 [[M]], -1
@@ -227,7 +227,7 @@ define i32 @p_commutative2(i32 %x, i32 %y, i32 %m) {
ret i32 %ret
}
-define i32 @p_commutative3(i32 %x, i32 %m) {
+define i32 @p_commutative3(i32 %x, i32 noundef %m) {
; CHECK-LABEL: @p_commutative3(
; CHECK-NEXT: [[Y:%.*]] = call i32 @gen32()
; CHECK-NEXT: [[AND:%.*]] = and i32 [[M:%.*]], [[X:%.*]]
@@ -2...
[truncated]
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
nikic
approved these changes
Dec 5, 2023
I'm pretty sure that some other uses that don't set the disjoint flag are also affected, just in ways that we don't notice. E.g. SCEV uses haveNoCommonBitsSet to create an |
…ecialCases. It's not safe for InstCombine to add disjoint metadata when converting add to or otherwise. I've added noundef attribute to preserve existing test behavior. A few questions. Do we only want to do this when the caller is going to set the disjoint flag? Alternatively, we could freeze the inputs?
topperc
force-pushed
the
pr/disjoint-undef
branch
from
ed7af7c
to
ba5f9eb
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
…ecialCases.
It's not safe for InstCombine to add disjoint metadata when converting add to or otherwise.
I've added noundef attribute to preserve existing test behavior.
A few questions.
Do we only want to do this when the caller is going to set the disjoint flag? Alternatively, we could freeze the inputs?