ValueTracking: Implement computeKnownFPClass for canonicalize

diff --git a/llvm/include/llvm/ADT/FloatingPointMode.h b/llvm/include/llvm/ADT/FloatingPointMode.h
index dd7719b..a7198ba 100644 (file)
--- a/llvm/include/llvm/ADT/FloatingPointMode.h
+++ b/llvm/include/llvm/ADT/FloatingPointMode.h
@@ -132,6 +132,18 @@ struct DenormalMode {
            Input != DenormalModeKind::Invalid;
   }
 
+  /// Return true if input denormals must be implicitly treated as 0.
+  constexpr bool inputsAreZero() const {
+    return Input == DenormalModeKind::PreserveSign ||
+           Input == DenormalModeKind::PositiveZero;
+  }
+
+  /// Return true if output denormals should be flushed to 0.
+  constexpr bool outputsAreZero() const {
+    return Output == DenormalModeKind::PreserveSign ||
+           Output == DenormalModeKind::PositiveZero;
+  }
+
   inline void print(raw_ostream &OS) const;
 
   inline std::string str() const {

diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index 6f02be1..b4ff594 100644 (file)
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -4448,6 +4448,26 @@ void computeKnownFPClass(const Value *V, const APInt &DemandedElts,
           Known.knownNot(fcNan);
         break;
       }
+      case Intrinsic::canonicalize: {
+        computeKnownFPClass(II->getArgOperand(0), DemandedElts,
+                            InterestedClasses, Known, Depth + 1, Q, TLI);
+        // Canonicalize is guaranteed to quiet signaling nans.
+        Known.knownNot(fcSNan);
+
+        // If the parent function flushes denormals, the canonical output cannot
+        // be a denormal.
+        const fltSemantics &FPType = II->getType()->getFltSemantics();
+        DenormalMode DenormMode = II->getFunction()->getDenormalMode(FPType);
+        if (DenormMode.inputsAreZero() || DenormMode.outputsAreZero())
+          Known.knownNot(fcSubnormal);
+
+        if (DenormMode.Input == DenormalMode::PositiveZero ||
+            (DenormMode.Output == DenormalMode::PositiveZero &&
+             DenormMode.Input == DenormalMode::IEEE))
+          Known.knownNot(fcNegZero);
+
+        break;
+      }
       default:
         break;
       }

diff --git a/llvm/test/Transforms/Attributor/nofpclass-canonicalize.ll b/llvm/test/Transforms/Attributor/nofpclass-canonicalize.ll
index 26022a8..8fdf5ab 100644 (file)
--- a/llvm/test/Transforms/Attributor/nofpclass-canonicalize.ll
+++ b/llvm/test/Transforms/Attributor/nofpclass-canonicalize.ll
@@ -4,9 +4,9 @@
 declare float @llvm.canonicalize.f32(float)
 
 define float @ret_canonicalize(float %arg0) {
-; CHECK-LABEL: define float @ret_canonicalize
+; CHECK-LABEL: define nofpclass(snan) float @ret_canonicalize
 ; CHECK-SAME: (float [[ARG0:%.*]]) #[[ATTR1:[0-9]+]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10:[0-9]+]]
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(snan) float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10:[0-9]+]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.canonicalize.f32(float %arg0)
@@ -14,9 +14,9 @@ define float @ret_canonicalize(float %arg0) {
 }
 
 define float @ret_canonicalize_noinf(float nofpclass(inf) %arg0) {
-; CHECK-LABEL: define float @ret_canonicalize_noinf
+; CHECK-LABEL: define nofpclass(snan inf) float @ret_canonicalize_noinf
 ; CHECK-SAME: (float nofpclass(inf) [[ARG0:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(snan inf) float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.canonicalize.f32(float %arg0)
@@ -24,9 +24,9 @@ define float @ret_canonicalize_noinf(float nofpclass(inf) %arg0) {
 }
 
 define float @ret_canonicalize_dynamic_denormal(float nofpclass(inf) %arg0) "denormal-fp-math"="dynamic,dynamic" {
-; CHECK-LABEL: define float @ret_canonicalize_dynamic_denormal
+; CHECK-LABEL: define nofpclass(snan inf) float @ret_canonicalize_dynamic_denormal
 ; CHECK-SAME: (float nofpclass(inf) [[ARG0:%.*]]) #[[ATTR2:[0-9]+]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(snan inf) float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.canonicalize.f32(float %arg0)
@@ -34,9 +34,9 @@ define float @ret_canonicalize_dynamic_denormal(float nofpclass(inf) %arg0) "den
 }
 
 define float @ret_canonicalize_daz_denormal(float nofpclass(inf) %arg0) "denormal-fp-math"="dynamic,preserve-sign" {
-; CHECK-LABEL: define float @ret_canonicalize_daz_denormal
+; CHECK-LABEL: define nofpclass(snan inf sub) float @ret_canonicalize_daz_denormal
 ; CHECK-SAME: (float nofpclass(inf) [[ARG0:%.*]]) #[[ATTR3:[0-9]+]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(snan inf sub) float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.canonicalize.f32(float %arg0)
@@ -44,9 +44,9 @@ define float @ret_canonicalize_daz_denormal(float nofpclass(inf) %arg0) "denorma
 }
 
 define float @ret_canonicalize_dapz_denormal(float nofpclass(inf) %arg0) "denormal-fp-math"="dynamic,positive-zero" {
-; CHECK-LABEL: define float @ret_canonicalize_dapz_denormal
+; CHECK-LABEL: define nofpclass(snan inf nzero sub) float @ret_canonicalize_dapz_denormal
 ; CHECK-SAME: (float nofpclass(inf) [[ARG0:%.*]]) #[[ATTR4:[0-9]+]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(snan inf nzero sub) float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.canonicalize.f32(float %arg0)
@@ -54,9 +54,9 @@ define float @ret_canonicalize_dapz_denormal(float nofpclass(inf) %arg0) "denorm
 }
 
 define float @ret_canonicalize_ftpz_dapz_denormal(float nofpclass(inf) %arg0) "denormal-fp-math"="positive-zero,preserve-sign" {
-; CHECK-LABEL: define float @ret_canonicalize_ftpz_dapz_denormal
+; CHECK-LABEL: define nofpclass(snan inf sub) float @ret_canonicalize_ftpz_dapz_denormal
 ; CHECK-SAME: (float nofpclass(inf) [[ARG0:%.*]]) #[[ATTR5:[0-9]+]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(snan inf sub) float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.canonicalize.f32(float %arg0)
@@ -64,9 +64,9 @@ define float @ret_canonicalize_ftpz_dapz_denormal(float nofpclass(inf) %arg0) "d
 }
 
 define float @ret_canonicalize_ftpz_ieee_denormal(float nofpclass(inf) %arg0) "denormal-fp-math"="positive-zero,ieee" {
-; CHECK-LABEL: define float @ret_canonicalize_ftpz_ieee_denormal
+; CHECK-LABEL: define nofpclass(snan inf nzero sub) float @ret_canonicalize_ftpz_ieee_denormal
 ; CHECK-SAME: (float nofpclass(inf) [[ARG0:%.*]]) #[[ATTR6:[0-9]+]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(snan inf nzero sub) float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.canonicalize.f32(float %arg0)
@@ -74,9 +74,9 @@ define float @ret_canonicalize_ftpz_ieee_denormal(float nofpclass(inf) %arg0) "d
 }
 
 define float @ret_canonicalize_ftpz_dynamic_denormal(float nofpclass(inf) %arg0) "denormal-fp-math"="positive-zero,dynamic" {
-; CHECK-LABEL: define float @ret_canonicalize_ftpz_dynamic_denormal
+; CHECK-LABEL: define nofpclass(snan inf sub) float @ret_canonicalize_ftpz_dynamic_denormal
 ; CHECK-SAME: (float nofpclass(inf) [[ARG0:%.*]]) #[[ATTR7:[0-9]+]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(snan inf sub) float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.canonicalize.f32(float %arg0)
@@ -84,9 +84,9 @@ define float @ret_canonicalize_ftpz_dynamic_denormal(float nofpclass(inf) %arg0)
 }
 
 define float @ret_canonicalize_ftz_denormal(float nofpclass(inf) %arg0) "denormal-fp-math"="preserve-sign,dynamic" {
-; CHECK-LABEL: define float @ret_canonicalize_ftz_denormal
+; CHECK-LABEL: define nofpclass(snan inf sub) float @ret_canonicalize_ftz_denormal
 ; CHECK-SAME: (float nofpclass(inf) [[ARG0:%.*]]) #[[ATTR8:[0-9]+]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(snan inf sub) float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.canonicalize.f32(float %arg0)
@@ -94,9 +94,9 @@ define float @ret_canonicalize_ftz_denormal(float nofpclass(inf) %arg0) "denorma
 }
 
 define float @ret_canonicalize_ieee_denormal(float nofpclass(inf) %arg0) "denormal-fp-math"="ieee,ieee" {
-; CHECK-LABEL: define float @ret_canonicalize_ieee_denormal
+; CHECK-LABEL: define nofpclass(snan inf) float @ret_canonicalize_ieee_denormal
 ; CHECK-SAME: (float nofpclass(inf) [[ARG0:%.*]]) #[[ATTR9:[0-9]+]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
+; CHECK-NEXT:    [[CALL:%.*]] = call nofpclass(snan inf) float @llvm.canonicalize.f32(float [[ARG0]]) #[[ATTR10]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
   %call = call float @llvm.canonicalize.f32(float %arg0)