Index: llvm/lib/Analysis/ValueTracking.cpp
===================================================================
--- llvm/lib/Analysis/ValueTracking.cpp
+++ llvm/lib/Analysis/ValueTracking.cpp
@@ -4571,6 +4571,20 @@
 
     break;
   }
+  case Instruction::FPExt: {
+    // Infinity, nan and zero propagate from source.
+    computeKnownFPClass(Op->getOperand(0), DemandedElts, InterestedClasses,
+                        Known, Depth + 1, Q, TLI);
+
+    // All subnormal inputs should be in the normal range in the result type.
+    // TODO: Does this hold for all exotic types
+    Known.knownNot(fcSubnormal);
+
+    // Sign bit of a nan isn't guaranteed.
+    if (!Known.isKnownNeverNaN())
+      Known.SignBit = std::nullopt;
+    break;
+  }
   case Instruction::SIToFP:
   case Instruction::UIToFP: {
     // Cannot produce nan
Index: llvm/test/Transforms/Attributor/nofpclass-fpext.ll
===================================================================
--- llvm/test/Transforms/Attributor/nofpclass-fpext.ll
+++ llvm/test/Transforms/Attributor/nofpclass-fpext.ll
@@ -2,7 +2,7 @@
 ; RUN: opt -aa-pipeline=basic-aa -passes=attributor -attributor-manifest-internal -S < %s | FileCheck %s --check-prefixes=CHECK,TUNIT
 
 define double @ret_fpext(float %arg0) {
-; CHECK-LABEL: define double @ret_fpext
+; CHECK-LABEL: define nofpclass(sub) double @ret_fpext
 ; CHECK-SAME: (float [[ARG0:%.*]]) #[[ATTR0:[0-9]+]] {
 ; CHECK-NEXT:    [[EXT:%.*]] = fpext float [[ARG0]] to double
 ; CHECK-NEXT:    ret double [[EXT]]
@@ -12,7 +12,7 @@
 }
 
 define double @ret_fpext_nonan(float nofpclass(nan) %arg0) {
-; CHECK-LABEL: define double @ret_fpext_nonan
+; CHECK-LABEL: define nofpclass(nan sub) double @ret_fpext_nonan
 ; CHECK-SAME: (float nofpclass(nan) [[ARG0:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[EXT:%.*]] = fpext float [[ARG0]] to double
 ; CHECK-NEXT:    ret double [[EXT]]
@@ -22,7 +22,7 @@
 }
 
 define double @ret_fpext_noqnan(float nofpclass(qnan) %arg0) {
-; CHECK-LABEL: define double @ret_fpext_noqnan
+; CHECK-LABEL: define nofpclass(qnan sub) double @ret_fpext_noqnan
 ; CHECK-SAME: (float nofpclass(qnan) [[ARG0:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[EXT:%.*]] = fpext float [[ARG0]] to double
 ; CHECK-NEXT:    ret double [[EXT]]
@@ -32,7 +32,7 @@
 }
 
 define double @ret_fpext_nosnan(float nofpclass(snan) %arg0) {
-; CHECK-LABEL: define double @ret_fpext_nosnan
+; CHECK-LABEL: define nofpclass(snan sub) double @ret_fpext_nosnan
 ; CHECK-SAME: (float nofpclass(snan) [[ARG0:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[EXT:%.*]] = fpext float [[ARG0]] to double
 ; CHECK-NEXT:    ret double [[EXT]]
@@ -42,7 +42,7 @@
 }
 
 define double @ret_fpext_noinf(float nofpclass(inf) %arg0) {
-; CHECK-LABEL: define double @ret_fpext_noinf
+; CHECK-LABEL: define nofpclass(inf sub) double @ret_fpext_noinf
 ; CHECK-SAME: (float nofpclass(inf) [[ARG0:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[EXT:%.*]] = fpext float [[ARG0]] to double
 ; CHECK-NEXT:    ret double [[EXT]]
@@ -52,7 +52,7 @@
 }
 
 define double @ret_fpext_nopinf(float nofpclass(pinf) %arg0) {
-; CHECK-LABEL: define double @ret_fpext_nopinf
+; CHECK-LABEL: define nofpclass(pinf sub) double @ret_fpext_nopinf
 ; CHECK-SAME: (float nofpclass(pinf) [[ARG0:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[EXT:%.*]] = fpext float [[ARG0]] to double
 ; CHECK-NEXT:    ret double [[EXT]]
@@ -62,7 +62,7 @@
 }
 
 define double @ret_fpext_noninf(float nofpclass(ninf) %arg0) {
-; CHECK-LABEL: define double @ret_fpext_noninf
+; CHECK-LABEL: define nofpclass(ninf sub) double @ret_fpext_noninf
 ; CHECK-SAME: (float nofpclass(ninf) [[ARG0:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[EXT:%.*]] = fpext float [[ARG0]] to double
 ; CHECK-NEXT:    ret double [[EXT]]
@@ -72,7 +72,7 @@
 }
 
 define double @ret_fpext_nozero(float nofpclass(zero) %arg0) {
-; CHECK-LABEL: define double @ret_fpext_nozero
+; CHECK-LABEL: define nofpclass(zero sub) double @ret_fpext_nozero
 ; CHECK-SAME: (float nofpclass(zero) [[ARG0:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[EXT:%.*]] = fpext float [[ARG0]] to double
 ; CHECK-NEXT:    ret double [[EXT]]
@@ -82,7 +82,7 @@
 }
 
 define double @ret_fpext_nopzero(float nofpclass(pzero) %arg0) {
-; CHECK-LABEL: define double @ret_fpext_nopzero
+; CHECK-LABEL: define nofpclass(pzero sub) double @ret_fpext_nopzero
 ; CHECK-SAME: (float nofpclass(pzero) [[ARG0:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[EXT:%.*]] = fpext float [[ARG0]] to double
 ; CHECK-NEXT:    ret double [[EXT]]
@@ -92,7 +92,7 @@
 }
 
 define double @ret_fpext_nonzero(float nofpclass(nzero) %arg0) {
-; CHECK-LABEL: define double @ret_fpext_nonzero
+; CHECK-LABEL: define nofpclass(nzero sub) double @ret_fpext_nonzero
 ; CHECK-SAME: (float nofpclass(nzero) [[ARG0:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[EXT:%.*]] = fpext float [[ARG0]] to double
 ; CHECK-NEXT:    ret double [[EXT]]
@@ -102,7 +102,7 @@
 }
 
 define double @ret_fpext_nonan_noinf(float nofpclass(nan inf) %arg0) {
-; CHECK-LABEL: define double @ret_fpext_nonan_noinf
+; CHECK-LABEL: define nofpclass(nan inf sub) double @ret_fpext_nonan_noinf
 ; CHECK-SAME: (float nofpclass(nan inf) [[ARG0:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[EXT:%.*]] = fpext float [[ARG0]] to double
 ; CHECK-NEXT:    ret double [[EXT]]
@@ -112,7 +112,7 @@
 }
 
 define double @ret_fpext_nosub(float nofpclass(sub) %arg0) {
-; CHECK-LABEL: define double @ret_fpext_nosub
+; CHECK-LABEL: define nofpclass(sub) double @ret_fpext_nosub
 ; CHECK-SAME: (float nofpclass(sub) [[ARG0:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[EXT:%.*]] = fpext float [[ARG0]] to double
 ; CHECK-NEXT:    ret double [[EXT]]
@@ -122,7 +122,7 @@
 }
 
 define double @ret_fpext_nonorm(float nofpclass(norm) %arg0) {
-; CHECK-LABEL: define double @ret_fpext_nonorm
+; CHECK-LABEL: define nofpclass(sub norm) double @ret_fpext_nonorm
 ; CHECK-SAME: (float nofpclass(norm) [[ARG0:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[EXT:%.*]] = fpext float [[ARG0]] to double
 ; CHECK-NEXT:    ret double [[EXT]]