diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index 1f974c148af4..44a1240f5635 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -5727,6 +5727,45 @@ void computeKnownFPClass(const Value *V, const APInt &DemandedElts,
       Known.knownNot(fcNan | fcInf | fcNegative);
       break;
     }
+    case Intrinsic::convert_from_arbitrary_fp: {
+      auto *MD = cast<MetadataAsValue>(II->getArgOperand(1))->getMetadata();
+      StringRef FormatStr = cast<MDString>(MD)->getString();
+
+      const fltSemantics *SrcSemantics =
+          APFloat::getArbitraryFPSemantics(FormatStr);
+      if (!SrcSemantics)
+        break;
+
+      const fltSemantics DstSemantics =
+          II->getType()->getScalarType()->getFltSemantics();
+
+      if (!APFloat::semanticsHasNaN(*SrcSemantics))
+        Known.knownNot(fcNan);
+
+      // fcInf can only be cleared if the source format has no Inf encoding
+      // and the dst max exp can accommodate src max exp.
+      if (!APFloat::semanticsHasInf(*SrcSemantics) &&
+          APFloat::semanticsMaxExponent(*SrcSemantics) <=
+              APFloat::semanticsMaxExponent(DstSemantics))
+        Known.knownNot(fcInf);
+
+      // Check and clear all neg flags for formats that do not have signed
+      // representation.
+      if (!APFloat::semanticsHasSignedRepr(*SrcSemantics))
+        Known.knownNot(fcNegative);
+
+      // Check if format has no zero at all (Float8E8M0FNU), or no negative
+      // zero.
+      if (!APFloat::semanticsHasZero(*SrcSemantics))
+        Known.knownNot(fcZero);
+      else if (SrcSemantics->nanEncoding == fltNanEncoding::NegativeZero)
+        Known.knownNot(fcNegZero);
+
+      // If src lands normally in dest, the result can never be subnormal.
+      if (APFloat::isRepresentableAsNormalIn(*SrcSemantics, DstSemantics))
+        Known.knownNot(fcSubnormal);
+      break;
+    }
     default:
       break;
     }
diff --git a/llvm/test/Transforms/Attributor/nofpclass-arbitrary-fp.ll b/llvm/test/Transforms/Attributor/nofpclass-arbitrary-fp.ll
new file mode 100644
index 000000000000..c87a983882f6
--- /dev/null
+++ b/llvm/test/Transforms/Attributor/nofpclass-arbitrary-fp.ll
@@ -0,0 +1,192 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt -aa-pipeline=basic-aa -passes=attributor -attributor-manifest-internal -S < %s | FileCheck %s
+
+define float @test_convert_float8e5m2_to_f32(i8 %bits) {
+; CHECK-LABEL: define nofpclass(sub) float @test_convert_float8e5m2_to_f32(
+; CHECK-SAME: i8 [[BITS:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:    [[FP:%.*]] = call nofpclass(sub) float @llvm.convert.from.arbitrary.fp.f32.i8(i8 [[BITS]], metadata !"Float8E5M2") #[[ATTR2:[0-9]+]]
+; CHECK-NEXT:    ret float [[FP]]
+;
+  %fp = call float @llvm.convert.from.arbitrary.fp.f32.i8(i8 %bits, metadata !"Float8E5M2")
+  ret float %fp
+}
+
+define float @test_convert_float8e5m2fnuz_to_f32(i8 %bits) {
+; CHECK-LABEL: define float @test_convert_float8e5m2fnuz_to_f32(
+; CHECK-SAME: i8 [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call float @llvm.convert.from.arbitrary.fp.f32.i8(i8 [[BITS]], metadata !"Float8E5M2FNUZ") #[[ATTR2]]
+; CHECK-NEXT:    ret float [[FP]]
+;
+  %fp = call float @llvm.convert.from.arbitrary.fp.f32.i8(i8 %bits, metadata !"Float8E5M2FNUZ")
+  ret float %fp
+}
+
+define float @test_convert_float8e4m3_to_f32(i8 %bits) {
+; CHECK-LABEL: define float @test_convert_float8e4m3_to_f32(
+; CHECK-SAME: i8 [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call float @llvm.convert.from.arbitrary.fp.f32.i8(i8 [[BITS]], metadata !"Float8E4M3") #[[ATTR2]]
+; CHECK-NEXT:    ret float [[FP]]
+;
+  %fp = call float @llvm.convert.from.arbitrary.fp.f32.i8(i8 %bits, metadata !"Float8E4M3")
+  ret float %fp
+}
+
+define float @test_convert_float8e4m3fn_to_f32(i8 %bits) {
+; CHECK-LABEL: define nofpclass(inf sub) float @test_convert_float8e4m3fn_to_f32(
+; CHECK-SAME: i8 [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call nofpclass(inf sub) float @llvm.convert.from.arbitrary.fp.f32.i8(i8 [[BITS]], metadata !"Float8E4M3FN") #[[ATTR2]]
+; CHECK-NEXT:    ret float [[FP]]
+;
+  %fp = call float @llvm.convert.from.arbitrary.fp.f32.i8(i8 %bits, metadata !"Float8E4M3FN")
+  ret float %fp
+}
+
+define float @test_convert_float8e4m3fnuz_to_f32(i8 %bits) {
+; CHECK-LABEL: define float @test_convert_float8e4m3fnuz_to_f32(
+; CHECK-SAME: i8 [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call float @llvm.convert.from.arbitrary.fp.f32.i8(i8 [[BITS]], metadata !"Float8E4M3FNUZ") #[[ATTR2]]
+; CHECK-NEXT:    ret float [[FP]]
+;
+  %fp = call float @llvm.convert.from.arbitrary.fp.f32.i8(i8 %bits, metadata !"Float8E4M3FNUZ")
+  ret float %fp
+}
+
+define float @test_convert_float8e4m3b11fnuz_to_f32(i8 %bits) {
+; CHECK-LABEL: define float @test_convert_float8e4m3b11fnuz_to_f32(
+; CHECK-SAME: i8 [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call float @llvm.convert.from.arbitrary.fp.f32.i8(i8 [[BITS]], metadata !"Float8E4M3B11FNUZ") #[[ATTR2]]
+; CHECK-NEXT:    ret float [[FP]]
+;
+  %fp = call float @llvm.convert.from.arbitrary.fp.f32.i8(i8 %bits, metadata !"Float8E4M3B11FNUZ")
+  ret float %fp
+}
+
+define float @test_convert_float8e3m4_to_f32(i8 %bits) {
+; CHECK-LABEL: define float @test_convert_float8e3m4_to_f32(
+; CHECK-SAME: i8 [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call float @llvm.convert.from.arbitrary.fp.f32.i8(i8 [[BITS]], metadata !"Float8E3M4") #[[ATTR2]]
+; CHECK-NEXT:    ret float [[FP]]
+;
+  %fp = call float @llvm.convert.from.arbitrary.fp.f32.i8(i8 %bits, metadata !"Float8E3M4")
+  ret float %fp
+}
+
+define float @test_convert_float8e8m0fnu_to_f32(i8 %bits) {
+; CHECK-LABEL: define float @test_convert_float8e8m0fnu_to_f32(
+; CHECK-SAME: i8 [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call float @llvm.convert.from.arbitrary.fp.f32.i8(i8 [[BITS]], metadata !"Float8E8M0FNU") #[[ATTR2]]
+; CHECK-NEXT:    ret float [[FP]]
+;
+  %fp = call float @llvm.convert.from.arbitrary.fp.f32.i8(i8 %bits, metadata !"Float8E8M0FNU")
+  ret float %fp
+}
+
+define float @test_convert_float6e3m2fn_to_f32(i6 %bits) {
+; CHECK-LABEL: define nofpclass(nan inf sub) float @test_convert_float6e3m2fn_to_f32(
+; CHECK-SAME: i6 [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call nofpclass(nan inf sub) float @llvm.convert.from.arbitrary.fp.f32.i6(i6 [[BITS]], metadata !"Float6E3M2FN") #[[ATTR2]]
+; CHECK-NEXT:    ret float [[FP]]
+;
+  %fp = call float @llvm.convert.from.arbitrary.fp.f32.i6(i6 %bits, metadata !"Float6E3M2FN")
+  ret float %fp
+}
+
+define float @test_convert_float6e2m3fn_to_f32(i6 %bits) {
+; CHECK-LABEL: define nofpclass(nan inf sub) float @test_convert_float6e2m3fn_to_f32(
+; CHECK-SAME: i6 [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call nofpclass(nan inf sub) float @llvm.convert.from.arbitrary.fp.f32.i6(i6 [[BITS]], metadata !"Float6E2M3FN") #[[ATTR2]]
+; CHECK-NEXT:    ret float [[FP]]
+;
+  %fp = call float @llvm.convert.from.arbitrary.fp.f32.i6(i6 %bits, metadata !"Float6E2M3FN")
+  ret float %fp
+}
+
+define float @test_convert_float4e2m1fn_to_f32(i4 %bits) {
+; CHECK-LABEL: define nofpclass(nan inf sub) float @test_convert_float4e2m1fn_to_f32(
+; CHECK-SAME: i4 [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call nofpclass(nan inf sub) float @llvm.convert.from.arbitrary.fp.f32.i4(i4 [[BITS]], metadata !"Float4E2M1FN") #[[ATTR2]]
+; CHECK-NEXT:    ret float [[FP]]
+;
+  %fp = call float @llvm.convert.from.arbitrary.fp.f32.i4(i4 %bits, metadata !"Float4E2M1FN")
+  ret float %fp
+}
+
+define double @test_convert_float4e2m1fn_to_f64(i4 %bits) {
+; CHECK-LABEL: define nofpclass(nan inf sub) double @test_convert_float4e2m1fn_to_f64(
+; CHECK-SAME: i4 [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call nofpclass(nan inf sub) double @llvm.convert.from.arbitrary.fp.f64.i4(i4 [[BITS]], metadata !"Float4E2M1FN") #[[ATTR2]]
+; CHECK-NEXT:    ret double [[FP]]
+;
+  %fp = call double @llvm.convert.from.arbitrary.fp.f64.i4(i4 %bits, metadata !"Float4E2M1FN")
+  ret double %fp
+}
+
+define half @test_convert_float8e4m3fn_to_f16(i8 %bits) {
+; CHECK-LABEL: define nofpclass(inf sub) half @test_convert_float8e4m3fn_to_f16(
+; CHECK-SAME: i8 [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call nofpclass(inf sub) half @llvm.convert.from.arbitrary.fp.f16.i8(i8 [[BITS]], metadata !"Float8E4M3FN") #[[ATTR2]]
+; CHECK-NEXT:    ret half [[FP]]
+;
+  %fp = call half @llvm.convert.from.arbitrary.fp.f16.i8(i8 %bits, metadata !"Float8E4M3FN")
+  ret half %fp
+}
+
+define half @test_convert_float6e2m3fn_to_f16(i6 %bits) {
+; CHECK-LABEL: define nofpclass(nan inf sub) half @test_convert_float6e2m3fn_to_f16(
+; CHECK-SAME: i6 [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call nofpclass(nan inf sub) half @llvm.convert.from.arbitrary.fp.f16.i6(i6 [[BITS]], metadata !"Float6E2M3FN") #[[ATTR2]]
+; CHECK-NEXT:    ret half [[FP]]
+;
+  %fp = call half @llvm.convert.from.arbitrary.fp.f16.i6(i6 %bits, metadata !"Float6E2M3FN")
+  ret half %fp
+}
+
+define <4 x half> @test_convert_float8e5m2_to_v4f16(<4 x i8> %bits) {
+; CHECK-LABEL: define <4 x half> @test_convert_float8e5m2_to_v4f16(
+; CHECK-SAME: <4 x i8> [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call <4 x half> @llvm.convert.from.arbitrary.fp.v4f16.v4i8(<4 x i8> [[BITS]], metadata !"Float8E5M2") #[[ATTR2]]
+; CHECK-NEXT:    ret <4 x half> [[FP]]
+;
+  %fp = call <4 x half> @llvm.convert.from.arbitrary.fp.v4f16.v4i8(<4 x i8> %bits, metadata !"Float8E5M2")
+  ret <4 x half> %fp
+}
+
+define <4 x double> @test_convert_float4e2m1fn_to_v4f64(<4 x i4> %bits) {
+; CHECK-LABEL: define nofpclass(nan inf sub) <4 x double> @test_convert_float4e2m1fn_to_v4f64(
+; CHECK-SAME: <4 x i4> [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call nofpclass(nan inf sub) <4 x double> @llvm.convert.from.arbitrary.fp.v4f64.v4i4(<4 x i4> [[BITS]], metadata !"Float4E2M1FN") #[[ATTR2]]
+; CHECK-NEXT:    ret <4 x double> [[FP]]
+;
+  %fp = call <4 x double> @llvm.convert.from.arbitrary.fp.v4f64.v4i4(<4 x i4> %bits, metadata !"Float4E2M1FN")
+  ret <4 x double> %fp
+}
+
+define <4 x double> @test_convert_float6e3m2fn_to_v4f64(<4 x i6> %bits) {
+; CHECK-LABEL: define nofpclass(nan inf sub) <4 x double> @test_convert_float6e3m2fn_to_v4f64(
+; CHECK-SAME: <4 x i6> [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call nofpclass(nan inf sub) <4 x double> @llvm.convert.from.arbitrary.fp.v4f64.v4i6(<4 x i6> [[BITS]], metadata !"Float6E3M2FN") #[[ATTR2]]
+; CHECK-NEXT:    ret <4 x double> [[FP]]
+;
+  %fp = call <4 x double> @llvm.convert.from.arbitrary.fp.v4f64.v4i6(<4 x i6> %bits, metadata !"Float6E3M2FN")
+  ret <4 x double> %fp
+}
+
+define half @test_convert_float8e8m0fnu_to_f16(i8 %bits) {
+; CHECK-LABEL: define half @test_convert_float8e8m0fnu_to_f16(
+; CHECK-SAME: i8 [[BITS:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[FP:%.*]] = call half @llvm.convert.from.arbitrary.fp.f16.i8(i8 [[BITS]], metadata !"Float8E8M0FNU") #[[ATTR2]]
+; CHECK-NEXT:    ret half [[FP]]
+;
+  %fp = call half @llvm.convert.from.arbitrary.fp.f16.i8(i8 %bits, metadata !"Float8E8M0FNU")
+  ret half %fp
+}
+
+declare float @llvm.convert.from.arbitrary.fp.f32.i4(i4, metadata)
+declare float @llvm.convert.from.arbitrary.fp.f32.i6(i6, metadata)
+declare float @llvm.convert.from.arbitrary.fp.f32.i8(i8, metadata)
+declare double @llvm.convert.from.arbitrary.fp.f64.i4(i4, metadata)
+declare half @llvm.convert.from.arbitrary.fp.f16.i8(i8, metadata)
+declare half @llvm.convert.from.arbitrary.fp.f16.i6(i6, metadata)
+declare <4 x double> @llvm.convert.from.arbitrary.fp.v4f64.v4i4(<4 x i4>, metadata)
+declare <4 x half> @llvm.convert.from.arbitrary.fp.v4f16.v4i8(<4 x i8>, metadata)
+declare <4 x double> @llvm.convert.from.arbitrary.fp.v4f64.v4i6(<4 x i6>, metadata)