[llvm] de7bfec - ValueTracking: Handle frem in computeKnownFPClass
Matt Arsenault via llvm-commits
llvm-commits at lists.llvm.org
Wed Apr 26 03:22:42 PDT 2023
Author: Matt Arsenault
Date: 2023-04-26T06:22:30-04:00
New Revision: de7bfec8850ee402493d8386225129dd5ef33b5c
URL: https://github.com/llvm/llvm-project/commit/de7bfec8850ee402493d8386225129dd5ef33b5c
DIFF: https://github.com/llvm/llvm-project/commit/de7bfec8850ee402493d8386225129dd5ef33b5c.diff
LOG: ValueTracking: Handle frem in computeKnownFPClass
I barely understand what this does, but try to
handle the last case required to delete cannotBeOrderedLessThanZeroImpl.
Also improve by following fdiv handling for nans and identical operand case.
Added:
Modified:
llvm/lib/Analysis/ValueTracking.cpp
llvm/test/Transforms/Attributor/nofpclass-frem.ll
Removed:
################################################################################
diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index b47e7af0cbf6..6a02ab67c5a5 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -4525,7 +4525,8 @@ void computeKnownFPClass(const Value *V, const APInt &DemandedElts,
if (Depth == MaxAnalysisRecursionDepth)
return;
- switch (Op->getOpcode()) {
+ const unsigned Opc = Op->getOpcode();
+ switch (Opc) {
case Instruction::FNeg: {
computeKnownFPClass(Op->getOperand(0), DemandedElts, InterestedClasses,
Known, Depth + 1, Q, TLI);
@@ -4823,20 +4824,28 @@ void computeKnownFPClass(const Value *V, const APInt &DemandedElts,
break;
}
- case Instruction::FDiv: {
- // X / X is always exactly 1.0 or a NaN.
+ case Instruction::FDiv:
+ case Instruction::FRem: {
if (Op->getOperand(0) == Op->getOperand(1)) {
// TODO: Could filter out snan if we inspect the operand
- Known.KnownFPClasses = fcNan | fcPosNormal;
+ if (Op->getOpcode() == Instruction::FDiv) {
+ // X / X is always exactly 1.0 or a NaN.
+ Known.KnownFPClasses = fcNan | fcPosNormal;
+ } else {
+ // X % X is always exactly [+-]0.0 or a NaN.
+ Known.KnownFPClasses = fcNan | fcZero;
+ }
+
break;
}
const bool WantNan = (InterestedClasses & fcNan) != fcNone;
const bool WantNegative = (InterestedClasses & fcNegative) != fcNone;
- if (!WantNan && !WantNegative)
+ const bool WantPositive =
+ Opc == Instruction::FRem && (InterestedClasses & fcPositive) != fcNone;
+ if (!WantNan && !WantNegative && !WantPositive)
break;
- // TODO: FRem
KnownFPClass KnownLHS, KnownRHS;
computeKnownFPClass(Op->getOperand(1), DemandedElts,
@@ -4846,26 +4855,46 @@ void computeKnownFPClass(const Value *V, const APInt &DemandedElts,
bool KnowSomethingUseful =
KnownRHS.isKnownNeverNaN() || KnownRHS.isKnownNever(fcNegative);
- if (KnowSomethingUseful) {
+ if (KnowSomethingUseful || WantPositive) {
+ const FPClassTest InterestedLHS =
+ WantPositive ? fcAllFlags
+ : fcNan | fcInf | fcZero | fcSubnormal | fcNegative;
+
computeKnownFPClass(Op->getOperand(0), DemandedElts,
- fcNan | fcInf | fcZero | fcNegative, KnownLHS,
+ InterestedClasses & InterestedLHS, KnownLHS,
Depth + 1, Q, TLI);
}
const Function *F = cast<Instruction>(Op)->getFunction();
- // Only 0/0, Inf/Inf, Inf REM x and x REM 0 produce NaN.
- if (KnownLHS.isKnownNeverNaN() && KnownRHS.isKnownNeverNaN() &&
- (KnownLHS.isKnownNeverInfinity() || KnownRHS.isKnownNeverInfinity()) &&
- (KnownLHS.isKnownNeverLogicalZero(*F, Op->getType()) ||
- KnownRHS.isKnownNeverLogicalZero(*F, Op->getType()))) {
- Known.knownNot(fcNan);
- }
+ if (Op->getOpcode() == Instruction::FDiv) {
+ // Only 0/0, Inf/Inf produce NaN.
+ if (KnownLHS.isKnownNeverNaN() && KnownRHS.isKnownNeverNaN() &&
+ (KnownLHS.isKnownNeverInfinity() ||
+ KnownRHS.isKnownNeverInfinity()) &&
+ (KnownLHS.isKnownNeverLogicalZero(*F, Op->getType()) ||
+ KnownRHS.isKnownNeverLogicalZero(*F, Op->getType()))) {
+ Known.knownNot(fcNan);
+ }
- // X / -0.0 is -Inf (or NaN).
- // +X / +X is +X
- if (KnownLHS.isKnownNever(fcNegative) && KnownRHS.isKnownNever(fcNegative))
- Known.knownNot(fcNegative);
+ // X / -0.0 is -Inf (or NaN).
+ // +X / +X is +X
+ if (KnownLHS.isKnownNever(fcNegative) && KnownRHS.isKnownNever(fcNegative))
+ Known.knownNot(fcNegative);
+ } else {
+ // Inf REM x and x REM 0 produce NaN.
+ if (KnownLHS.isKnownNeverNaN() && KnownRHS.isKnownNeverNaN() &&
+ KnownLHS.isKnownNeverInfinity() &&
+ KnownRHS.isKnownNeverLogicalZero(*F, Op->getType())) {
+ Known.knownNot(fcNan);
+ }
+
+ // The sign for frem is the same as the first operand.
+ if (KnownLHS.isKnownNever(fcNegative))
+ Known.knownNot(fcNegative);
+ if (KnownLHS.isKnownNever(fcPositive))
+ Known.knownNot(fcPositive);
+ }
break;
}
diff --git a/llvm/test/Transforms/Attributor/nofpclass-frem.ll b/llvm/test/Transforms/Attributor/nofpclass-frem.ll
index d7ace0e88b96..930b1a77da57 100644
--- a/llvm/test/Transforms/Attributor/nofpclass-frem.ll
+++ b/llvm/test/Transforms/Attributor/nofpclass-frem.ll
@@ -82,7 +82,7 @@ define float @ret_frem_ieee_nonan_nozero__nonan_noinf(float nofpclass(nan zero)
}
define float @ret_frem_ieee_nonan_noinf__nonan_nozero(float nofpclass(nan inf) %arg0, float nofpclass(nan zero) %arg1) #0 {
-; CHECK-LABEL: define float @ret_frem_ieee_nonan_noinf__nonan_nozero
+; CHECK-LABEL: define nofpclass(nan) float @ret_frem_ieee_nonan_noinf__nonan_nozero
; CHECK-SAME: (float nofpclass(nan inf) [[ARG0:%.*]], float nofpclass(nan zero) [[ARG1:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FREM]]
@@ -122,7 +122,7 @@ define float @ret_frem_daz_nonan_nozero_nosub__nonan_noinf(float nofpclass(nan z
}
define float @ret_frem_daz_nonan_noinf__nonan_nozero_nosub(float nofpclass(nan inf) %arg0, float nofpclass(nan zero sub) %arg1) #1 {
-; CHECK-LABEL: define float @ret_frem_daz_nonan_noinf__nonan_nozero_nosub
+; CHECK-LABEL: define nofpclass(nan) float @ret_frem_daz_nonan_noinf__nonan_nozero_nosub
; CHECK-SAME: (float nofpclass(nan inf) [[ARG0:%.*]], float nofpclass(nan zero sub) [[ARG1:%.*]]) #[[ATTR1]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FREM]]
@@ -142,7 +142,7 @@ define float @ret_frem_ieee_nonan_noinf__nonan_noinf(float nofpclass(nan) %arg0,
}
define float @ret_frem_ieee_nonan_noinf_nozero__nonan_noinf_nozero(float nofpclass(nan inf zero) %arg0, float nofpclass(nan inf zero) %arg1) #0 {
-; CHECK-LABEL: define float @ret_frem_ieee_nonan_noinf_nozero__nonan_noinf_nozero
+; CHECK-LABEL: define nofpclass(nan) float @ret_frem_ieee_nonan_noinf_nozero__nonan_noinf_nozero
; CHECK-SAME: (float nofpclass(nan inf zero) [[ARG0:%.*]], float nofpclass(nan inf zero) [[ARG1:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FREM]]
@@ -202,7 +202,7 @@ define float @ret_frem_ieee_nonan_noinf_nozero__nonan_noinf_noinf(float nofpclas
}
define float @ret_frem_ieee_nonan_noinf_noinf__nonan_noinf_nozero(float nofpclass(nan inf) %arg0, float nofpclass(nan inf zero) %arg1) #0 {
-; CHECK-LABEL: define float @ret_frem_ieee_nonan_noinf_noinf__nonan_noinf_nozero
+; CHECK-LABEL: define nofpclass(nan) float @ret_frem_ieee_nonan_noinf_noinf__nonan_noinf_nozero
; CHECK-SAME: (float nofpclass(nan inf) [[ARG0:%.*]], float nofpclass(nan inf zero) [[ARG1:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FREM]]
@@ -232,7 +232,7 @@ define float @ret_frem_ieee_nonan_noinf_noposzero__nonan_noinf_noinf(float nofpc
}
define float @ret_frem_ieee_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosub(float nofpclass(nan inf zero sub) %arg0, float nofpclass(nan inf zero sub) %arg1) #0 {
-; CHECK-LABEL: define float @ret_frem_ieee_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosub
+; CHECK-LABEL: define nofpclass(nan) float @ret_frem_ieee_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosub
; CHECK-SAME: (float nofpclass(nan inf zero sub) [[ARG0:%.*]], float nofpclass(nan inf zero sub) [[ARG1:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FREM]]
@@ -243,7 +243,7 @@ define float @ret_frem_ieee_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosub(f
; Denormal mode doesn't matter because sources are nofpclass(sub)
define float @ret_frem_daz_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosub(float nofpclass(nan inf zero sub) %arg0, float nofpclass(nan inf zero sub) %arg1) #1 {
-; CHECK-LABEL: define float @ret_frem_daz_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosub
+; CHECK-LABEL: define nofpclass(nan) float @ret_frem_daz_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosub
; CHECK-SAME: (float nofpclass(nan inf zero sub) [[ARG0:%.*]], float nofpclass(nan inf zero sub) [[ARG1:%.*]]) #[[ATTR1]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FREM]]
@@ -253,7 +253,7 @@ define float @ret_frem_daz_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosub(fl
}
define float @ret_frem_dapz_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosub(float nofpclass(nan inf zero sub) %arg0, float nofpclass(nan inf zero sub) %arg1) #2 {
-; CHECK-LABEL: define float @ret_frem_dapz_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosub
+; CHECK-LABEL: define nofpclass(nan) float @ret_frem_dapz_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosub
; CHECK-SAME: (float nofpclass(nan inf zero sub) [[ARG0:%.*]], float nofpclass(nan inf zero sub) [[ARG1:%.*]]) #[[ATTR2]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FREM]]
@@ -263,7 +263,7 @@ define float @ret_frem_dapz_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosub(f
}
define float @ret_frem_dynamic_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosub(float nofpclass(nan inf zero sub) %arg0, float nofpclass(nan inf zero sub) %arg1) #3 {
-; CHECK-LABEL: define float @ret_frem_dynamic_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosub
+; CHECK-LABEL: define nofpclass(nan) float @ret_frem_dynamic_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosub
; CHECK-SAME: (float nofpclass(nan inf zero sub) [[ARG0:%.*]], float nofpclass(nan inf zero sub) [[ARG1:%.*]]) #[[ATTR3]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FREM]]
@@ -274,7 +274,7 @@ define float @ret_frem_dynamic_nonan_noinf_nozero_nosub__nonan_noinf_nozero_nosu
; Missing no-subnormal on lhs
define float @ret_frem_daz_nonan_noinf_nozero__nonan_noinf_nozero_nosub(float nofpclass(nan inf zero) %arg0, float nofpclass(nan inf zero sub) %arg1) #1 {
-; CHECK-LABEL: define float @ret_frem_daz_nonan_noinf_nozero__nonan_noinf_nozero_nosub
+; CHECK-LABEL: define nofpclass(nan) float @ret_frem_daz_nonan_noinf_nozero__nonan_noinf_nozero_nosub
; CHECK-SAME: (float nofpclass(nan inf zero) [[ARG0:%.*]], float nofpclass(nan inf zero sub) [[ARG1:%.*]]) #[[ATTR1]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FREM]]
@@ -356,7 +356,7 @@ define float @ret_frem_daz_noinf__nozero_nosub(float nofpclass(inf) %arg0, float
}
define float @ret_frem_same_operands(float %arg) #0 {
-; CHECK-LABEL: define float @ret_frem_same_operands
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_frem_same_operands
; CHECK-SAME: (float [[ARG:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG]], [[ARG]]
; CHECK-NEXT: ret float [[FREM]]
@@ -366,7 +366,7 @@ define float @ret_frem_same_operands(float %arg) #0 {
}
define float @ret_frem_same_operands_nosnan(float nofpclass(snan) %arg) #0 {
-; CHECK-LABEL: define float @ret_frem_same_operands_nosnan
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_frem_same_operands_nosnan
; CHECK-SAME: (float nofpclass(snan) [[ARG:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG]], [[ARG]]
; CHECK-NEXT: ret float [[FREM]]
@@ -376,7 +376,7 @@ define float @ret_frem_same_operands_nosnan(float nofpclass(snan) %arg) #0 {
}
define float @ret_frem_same_operands_noqnan(float nofpclass(qnan) %arg) #0 {
-; CHECK-LABEL: define float @ret_frem_same_operands_noqnan
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_frem_same_operands_noqnan
; CHECK-SAME: (float nofpclass(qnan) [[ARG:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG]], [[ARG]]
; CHECK-NEXT: ret float [[FREM]]
@@ -386,7 +386,7 @@ define float @ret_frem_same_operands_noqnan(float nofpclass(qnan) %arg) #0 {
}
define float @ret_frem_same_operands_nonan(float nofpclass(nan) %arg) #0 {
-; CHECK-LABEL: define float @ret_frem_same_operands_nonan
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_frem_same_operands_nonan
; CHECK-SAME: (float nofpclass(nan) [[ARG:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG]], [[ARG]]
; CHECK-NEXT: ret float [[FREM]]
@@ -436,7 +436,7 @@ define float @ret_frem_no_neg(float nofpclass(ninf nsub nnorm) %arg0, float nofp
}
define float @ret_frem_no_neg_nzero(float nofpclass(ninf nsub nnorm nzero) %arg0, float nofpclass(ninf nsub nnorm nzero) %arg1) #0 {
-; CHECK-LABEL: define float @ret_frem_no_neg_nzero
+; CHECK-LABEL: define nofpclass(ninf nzero nsub nnorm) float @ret_frem_no_neg_nzero
; CHECK-SAME: (float nofpclass(ninf nzero nsub nnorm) [[ARG0:%.*]], float nofpclass(ninf nzero nsub nnorm) [[ARG1:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FREM]]
@@ -456,7 +456,7 @@ define float @ret_frem_no_neg_rhs_no_nzero(float nofpclass(ninf nsub nnorm) %arg
}
define float @ret_frem_no_neg_no_zero_rhs(float nofpclass(ninf nsub nnorm nzero) %arg0, float nofpclass(ninf nsub nnorm zero) %arg1) #0 {
-; CHECK-LABEL: define float @ret_frem_no_neg_no_zero_rhs
+; CHECK-LABEL: define nofpclass(ninf nzero nsub nnorm) float @ret_frem_no_neg_no_zero_rhs
; CHECK-SAME: (float nofpclass(ninf nzero nsub nnorm) [[ARG0:%.*]], float nofpclass(ninf zero nsub nnorm) [[ARG1:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FREM]]
@@ -466,7 +466,7 @@ define float @ret_frem_no_neg_no_zero_rhs(float nofpclass(ninf nsub nnorm nzero)
}
define float @ret_frem_no_pos_lhs(float nofpclass(pinf psub pnorm pzero) %arg0, float %arg1) #0 {
-; CHECK-LABEL: define float @ret_frem_no_pos_lhs
+; CHECK-LABEL: define nofpclass(pinf pzero psub pnorm) float @ret_frem_no_pos_lhs
; CHECK-SAME: (float nofpclass(pinf pzero psub pnorm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR0]] {
; CHECK-NEXT: [[FREM:%.*]] = frem float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FREM]]
More information about the llvm-commits
mailing list