Skip to content

ValueTracking: Improve computeKnownFPClass fmul handling #173247

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
arsenm wants to merge 1 commit into
base: users/arsenm/valuetracking/add-baseline-tests-computeKnownFPClass-fmul-zero-inf
Choose a base branch
from
Open

ValueTracking: Improve computeKnownFPClass fmul handling #173247

arsenm wants to merge 1 commit into from
+71 −54

Conversation

@arsenm
Copy link
Contributor

@arsenm arsenm commented Dec 22, 2025

Improve known non-nan sign bit tracking. Handle cases with
a known 0 or inf input of indeterminate sign.

The tails of some library functions have sign management
for special cases.

@arsenm Graphite App
Copy link
Contributor Author

arsenm commented Dec 22, 2025 •
edited
Loading

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

This stack of pull requests is managed by Graphite. Learn more about stacking.

@arsenm arsenm mentioned this pull request Dec 22, 2025
Open
@arsenm arsenm added the floating-point Floating-point math label Dec 22, 2025 — with Graphite App
@arsenm arsenm marked this pull request as ready for review December 22, 2025 12:10
@arsenm arsenm requested a review from nikic as a code owner December 22, 2025 12:10
@llvmbot llvmbot added llvm:analysis Includes value tracking, cost tables and constant folding llvm:transforms labels Dec 22, 2025
@llvmbot
Copy link
Member

llvmbot commented Dec 22, 2025 •
edited
Loading

@llvm/pr-subscribers-backend-amdgpu
@llvm/pr-subscribers-llvm-transforms

@llvm/pr-subscribers-llvm-analysis

Author: Matt Arsenault (arsenm)

Changes

Improve known non-nan sign bit tracking. Handle cases with
a known 0 or inf input of indeterminate sign.

The tails of some library functions have sign management
for special cases.

Patch is 23.36 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/173247.diff

3 Files Affected:

  • (modified) llvm/lib/Analysis/ValueTracking.cpp (+33-12)
  • (modified) llvm/test/Transforms/Attributor/nofpclass-fmul.ll (+35-35)
  • (modified) llvm/test/Transforms/Attributor/nofpclass-nan-fmul.ll (+1-1)
diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index 0321c08f55776..c4ad71f668506 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -5722,14 +5722,8 @@ void computeKnownFPClass(const Value *V, const APInt &DemandedElts,
     if (Op->getOperand(0) == Op->getOperand(1))
       Known.knownNot(fcNegative);
 
-    if ((InterestedClasses & fcNan) != fcNan)
-      break;
-
-    // fcSubnormal is only needed in case of DAZ.
-    const FPClassTest NeedForNan = fcNan | fcInf | fcZero | fcSubnormal;
-
     KnownFPClass KnownLHS, KnownRHS;
-    computeKnownFPClass(Op->getOperand(1), DemandedElts, NeedForNan, KnownRHS,
+    computeKnownFPClass(Op->getOperand(1), DemandedElts, fcAllFlags, KnownRHS,
                         Q, Depth + 1);
 
     const APFloat *CRHS;
@@ -5749,12 +5743,39 @@ void computeKnownFPClass(const Value *V, const APInt &DemandedElts,
         Known.knownNot(fcSubnormal);
     }
 
-    if (!KnownRHS.isKnownNeverNaN())
-      break;
-
-    computeKnownFPClass(Op->getOperand(0), DemandedElts, NeedForNan, KnownLHS,
+    computeKnownFPClass(Op->getOperand(0), DemandedElts, fcAllFlags, KnownLHS,
                         Q, Depth + 1);
-    if (!KnownLHS.isKnownNeverNaN())
+
+    // xor sign bit.
+    if ((KnownLHS.isKnownNever(fcNegative) &&
+         KnownRHS.isKnownNever(fcNegative)) ||
+        (KnownLHS.isKnownNever(fcPositive) &&
+         KnownRHS.isKnownNever(fcPositive)))
+      Known.knownNot(fcNegative);
+
+    if ((KnownLHS.isKnownAlways(fcNegative | fcNan) &&
+         KnownRHS.isKnownNever(fcNegative)) ||
+        (KnownLHS.isKnownNever(fcNegative) &&
+         KnownRHS.isKnownAlways(fcNegative | fcNan)))
+      Known.knownNot(fcPositive);
+
+    // inf * anything => inf or nan
+    if (KnownLHS.isKnownAlways(fcInf) || KnownRHS.isKnownAlways(fcInf))
+      Known.knownNot(fcNormal | fcSubnormal | fcZero);
+
+    // 0 * anything => 0 or nan
+    if (KnownRHS.isKnownAlways(fcZero | fcNan) ||
+        KnownLHS.isKnownAlways(fcZero | fcNan))
+      Known.knownNot(fcNormal | fcSubnormal | fcInf);
+
+    // +/-0 * +/-inf = nan
+    if ((KnownLHS.isKnownAlways(fcZero | fcNan) &&
+         KnownRHS.isKnownAlways(fcInf | fcNan)) ||
+        (KnownLHS.isKnownAlways(fcInf | fcNan) &&
+         KnownRHS.isKnownAlways(fcZero | fcNan)))
+      Known.knownNot(~fcNan);
+
+    if (!KnownLHS.isKnownNeverNaN() || !KnownRHS.isKnownNeverNaN())
       break;
 
     if (KnownLHS.SignBit && KnownRHS.SignBit) {
diff --git a/llvm/test/Transforms/Attributor/nofpclass-fmul.ll b/llvm/test/Transforms/Attributor/nofpclass-fmul.ll
index 6367fc933b4b5..306a14a708566 100644
--- a/llvm/test/Transforms/Attributor/nofpclass-fmul.ll
+++ b/llvm/test/Transforms/Attributor/nofpclass-fmul.ll
@@ -257,7 +257,7 @@ define <2 x float> @ret_mul_partially_foldable_exponent_v2f32(<2 x float> %arg0)
 }
 
 define float @ret_mul_f32_0(float %arg0) {
-; CHECK-LABEL: define float @ret_mul_f32_0(
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_mul_f32_0(
 ; CHECK-SAME: float [[ARG0:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[CALL:%.*]] = fmul float [[ARG0]], 0.000000e+00
 ; CHECK-NEXT:    ret float [[CALL]]
@@ -267,7 +267,7 @@ define float @ret_mul_f32_0(float %arg0) {
 }
 
 define float @ret_mul_f32_inf(float %arg0) {
-; CHECK-LABEL: define nofpclass(sub) float @ret_mul_f32_inf(
+; CHECK-LABEL: define nofpclass(zero sub norm) float @ret_mul_f32_inf(
 ; CHECK-SAME: float [[ARG0:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[CALL:%.*]] = fmul float [[ARG0]], 0x7FF0000000000000
 ; CHECK-NEXT:    ret float [[CALL]]
@@ -277,7 +277,7 @@ define float @ret_mul_f32_inf(float %arg0) {
 }
 
 define float @ret_mul_f32_nan(float %arg0) {
-; CHECK-LABEL: define float @ret_mul_f32_nan(
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_mul_f32_nan(
 ; CHECK-SAME: float [[ARG0:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[CALL:%.*]] = fmul float [[ARG0]], 0x7FF8000000000000
 ; CHECK-NEXT:    ret float [[CALL]]
@@ -287,7 +287,7 @@ define float @ret_mul_f32_nan(float %arg0) {
 }
 
 define float @ret_fmul_f32_known_zero_or_nan_lhs(float nofpclass(inf norm sub) %arg0, float %arg1) {
-; CHECK-LABEL: define float @ret_fmul_f32_known_zero_or_nan_lhs(
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_fmul_f32_known_zero_or_nan_lhs(
 ; CHECK-SAME: float nofpclass(inf sub norm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -297,7 +297,7 @@ define float @ret_fmul_f32_known_zero_or_nan_lhs(float nofpclass(inf norm sub) %
 }
 
 define float @ret_fmul_f32_known_zero_or_nan_rhs(float %arg0, float nofpclass(inf norm sub) %arg1) {
-; CHECK-LABEL: define float @ret_fmul_f32_known_zero_or_nan_rhs(
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_fmul_f32_known_zero_or_nan_rhs(
 ; CHECK-SAME: float [[ARG0:%.*]], float nofpclass(inf sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -307,7 +307,7 @@ define float @ret_fmul_f32_known_zero_or_nan_rhs(float %arg0, float nofpclass(in
 }
 
 define float @ret_fmul_f32_known_zero_lhs(float nofpclass(nan inf norm sub) %arg0, float %arg1) {
-; CHECK-LABEL: define float @ret_fmul_f32_known_zero_lhs(
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_fmul_f32_known_zero_lhs(
 ; CHECK-SAME: float nofpclass(nan inf sub norm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -317,7 +317,7 @@ define float @ret_fmul_f32_known_zero_lhs(float nofpclass(nan inf norm sub) %arg
 }
 
 define float @ret_fmul_f32_known_zero_rhs(float %arg0, float nofpclass(nan inf norm sub) %arg1) {
-; CHECK-LABEL: define float @ret_fmul_f32_known_zero_rhs(
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_fmul_f32_known_zero_rhs(
 ; CHECK-SAME: float [[ARG0:%.*]], float nofpclass(nan inf sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -327,7 +327,7 @@ define float @ret_fmul_f32_known_zero_rhs(float %arg0, float nofpclass(nan inf n
 }
 
 define float @ret_fmul_f32_known_pzero_or_nan_lhs(float nofpclass(inf norm sub nzero) %arg0, float %arg1) {
-; CHECK-LABEL: define float @ret_fmul_f32_known_pzero_or_nan_lhs(
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_fmul_f32_known_pzero_or_nan_lhs(
 ; CHECK-SAME: float nofpclass(inf nzero sub norm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -337,7 +337,7 @@ define float @ret_fmul_f32_known_pzero_or_nan_lhs(float nofpclass(inf norm sub n
 }
 
 define float @ret_fmul_f32_known_pzero_or_nan_rhs(float %arg0, float nofpclass(inf norm sub nzero) %arg1) {
-; CHECK-LABEL: define float @ret_fmul_f32_known_pzero_or_nan_rhs(
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_fmul_f32_known_pzero_or_nan_rhs(
 ; CHECK-SAME: float [[ARG0:%.*]], float nofpclass(inf nzero sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -347,7 +347,7 @@ define float @ret_fmul_f32_known_pzero_or_nan_rhs(float %arg0, float nofpclass(i
 }
 
 define float @ret_fmul_f32_known_pzero_lhs(float nofpclass(nan inf norm sub nzero) %arg0, float %arg1) {
-; CHECK-LABEL: define float @ret_fmul_f32_known_pzero_lhs(
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_fmul_f32_known_pzero_lhs(
 ; CHECK-SAME: float nofpclass(nan inf nzero sub norm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -357,7 +357,7 @@ define float @ret_fmul_f32_known_pzero_lhs(float nofpclass(nan inf norm sub nzer
 }
 
 define float @ret_fmul_f32_known_pzero_rhs(float %arg0, float nofpclass(nan inf norm sub nzero) %arg1) {
-; CHECK-LABEL: define float @ret_fmul_f32_known_pzero_rhs(
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_fmul_f32_known_pzero_rhs(
 ; CHECK-SAME: float [[ARG0:%.*]], float nofpclass(nan inf nzero sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -367,7 +367,7 @@ define float @ret_fmul_f32_known_pzero_rhs(float %arg0, float nofpclass(nan inf
 }
 
 define float @ret_fmul_f32_known_nzero_or_nan_lhs(float nofpclass(inf norm sub pzero) %arg0, float %arg1) {
-; CHECK-LABEL: define float @ret_fmul_f32_known_nzero_or_nan_lhs(
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_fmul_f32_known_nzero_or_nan_lhs(
 ; CHECK-SAME: float nofpclass(inf pzero sub norm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -377,7 +377,7 @@ define float @ret_fmul_f32_known_nzero_or_nan_lhs(float nofpclass(inf norm sub p
 }
 
 define float @ret_fmul_f32_known_nzero_or_nan_rhs(float %arg0, float nofpclass(inf norm sub pzero) %arg1) {
-; CHECK-LABEL: define float @ret_fmul_f32_known_nzero_or_nan_rhs(
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_fmul_f32_known_nzero_or_nan_rhs(
 ; CHECK-SAME: float [[ARG0:%.*]], float nofpclass(inf pzero sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -387,7 +387,7 @@ define float @ret_fmul_f32_known_nzero_or_nan_rhs(float %arg0, float nofpclass(i
 }
 
 define float @ret_fmul_f32_known_nzero_lhs(float nofpclass(nan inf norm sub pzero) %arg0, float %arg1) {
-; CHECK-LABEL: define float @ret_fmul_f32_known_nzero_lhs(
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_fmul_f32_known_nzero_lhs(
 ; CHECK-SAME: float nofpclass(nan inf pzero sub norm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -397,7 +397,7 @@ define float @ret_fmul_f32_known_nzero_lhs(float nofpclass(nan inf norm sub pzer
 }
 
 define float @ret_fmul_f32_known_nzero_rhs(float %arg0, float nofpclass(nan inf norm sub pzero) %arg1) {
-; CHECK-LABEL: define float @ret_fmul_f32_known_nzero_rhs(
+; CHECK-LABEL: define nofpclass(inf sub norm) float @ret_fmul_f32_known_nzero_rhs(
 ; CHECK-SAME: float [[ARG0:%.*]], float nofpclass(nan inf pzero sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -427,7 +427,7 @@ define float @ret_fmul_f32_known_inf_or_nan_rhs(float %arg0, float nofpclass(zer
 }
 
 define float @ret_fmul_f32_known_inf_lhs(float nofpclass(nan zero norm sub) %arg0, float %arg1) {
-; CHECK-LABEL: define float @ret_fmul_f32_known_inf_lhs(
+; CHECK-LABEL: define nofpclass(zero sub norm) float @ret_fmul_f32_known_inf_lhs(
 ; CHECK-SAME: float nofpclass(nan zero sub norm) [[ARG0:%.*]], float [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -437,7 +437,7 @@ define float @ret_fmul_f32_known_inf_lhs(float nofpclass(nan zero norm sub) %arg
 }
 
 define float @ret_fmul_f32_known_inf_rhs(float %arg0, float nofpclass(nan zero norm sub) %arg1) {
-; CHECK-LABEL: define float @ret_fmul_f32_known_inf_rhs(
+; CHECK-LABEL: define nofpclass(zero sub norm) float @ret_fmul_f32_known_inf_rhs(
 ; CHECK-SAME: float [[ARG0:%.*]], float nofpclass(nan zero sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -488,7 +488,7 @@ define float @ret_fmul_f32_known_ninf_or_nan_rhs(float %arg0, float nofpclass(pi
 
 ; -> nan
 define float @ret_known_inf_mul_known_zero(float nofpclass(nan norm sub zero) %arg0, float nofpclass(nan inf norm sub) %arg1) {
-; CHECK-LABEL: define float @ret_known_inf_mul_known_zero(
+; CHECK-LABEL: define nofpclass(inf zero sub norm) float @ret_known_inf_mul_known_zero(
 ; CHECK-SAME: float nofpclass(nan zero sub norm) [[ARG0:%.*]], float nofpclass(nan inf sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -499,7 +499,7 @@ define float @ret_known_inf_mul_known_zero(float nofpclass(nan norm sub zero) %a
 
 ; -> nan
 define float @ret_known_inf_or_nan_mul_known_zero(float nofpclass(norm sub zero) %arg0, float nofpclass(nan inf norm sub) %arg1) {
-; CHECK-LABEL: define float @ret_known_inf_or_nan_mul_known_zero(
+; CHECK-LABEL: define nofpclass(inf zero sub norm) float @ret_known_inf_or_nan_mul_known_zero(
 ; CHECK-SAME: float nofpclass(zero sub norm) [[ARG0:%.*]], float nofpclass(nan inf sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -510,7 +510,7 @@ define float @ret_known_inf_or_nan_mul_known_zero(float nofpclass(norm sub zero)
 
 ; -> nan
 define float @ret_known_inf_mul_known_zero_or_nan(float nofpclass(nan norm sub zero) %arg0, float nofpclass(inf norm sub) %arg1) {
-; CHECK-LABEL: define float @ret_known_inf_mul_known_zero_or_nan(
+; CHECK-LABEL: define nofpclass(inf zero sub norm) float @ret_known_inf_mul_known_zero_or_nan(
 ; CHECK-SAME: float nofpclass(nan zero sub norm) [[ARG0:%.*]], float nofpclass(inf sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -521,7 +521,7 @@ define float @ret_known_inf_mul_known_zero_or_nan(float nofpclass(nan norm sub z
 
 ; -> nan
 define float @ret_known_ninf_mul_known_zero(float nofpclass(nan pinf norm sub zero) %arg0, float nofpclass(nan inf norm sub) %arg1) {
-; CHECK-LABEL: define float @ret_known_ninf_mul_known_zero(
+; CHECK-LABEL: define nofpclass(inf zero sub norm) float @ret_known_ninf_mul_known_zero(
 ; CHECK-SAME: float nofpclass(nan pinf zero sub norm) [[ARG0:%.*]], float nofpclass(nan inf sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -532,7 +532,7 @@ define float @ret_known_ninf_mul_known_zero(float nofpclass(nan pinf norm sub ze
 
 ; -> nan
 define float @ret_known_pinf_mul_known_zero(float nofpclass(nan ninf norm sub zero) %arg0, float nofpclass(nan inf norm sub) %arg1) {
-; CHECK-LABEL: define float @ret_known_pinf_mul_known_zero(
+; CHECK-LABEL: define nofpclass(inf zero sub norm) float @ret_known_pinf_mul_known_zero(
 ; CHECK-SAME: float nofpclass(nan ninf zero sub norm) [[ARG0:%.*]], float nofpclass(nan inf sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -543,7 +543,7 @@ define float @ret_known_pinf_mul_known_zero(float nofpclass(nan ninf norm sub ze
 
 ; -> nan
 define float @ret_known_zero_mul_known_inf(float nofpclass(nan inf norm sub) %arg0, float nofpclass(nan norm sub zero) %arg1) {
-; CHECK-LABEL: define float @ret_known_zero_mul_known_inf(
+; CHECK-LABEL: define nofpclass(inf zero sub norm) float @ret_known_zero_mul_known_inf(
 ; CHECK-SAME: float nofpclass(nan inf sub norm) [[ARG0:%.*]], float nofpclass(nan zero sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -554,7 +554,7 @@ define float @ret_known_zero_mul_known_inf(float nofpclass(nan inf norm sub) %ar
 
 ; -> nan
 define float @ret_known_zero_mul_known_inf_or_nan(float nofpclass(nan inf norm sub) %arg0, float nofpclass(norm sub zero) %arg1) {
-; CHECK-LABEL: define float @ret_known_zero_mul_known_inf_or_nan(
+; CHECK-LABEL: define nofpclass(inf zero sub norm) float @ret_known_zero_mul_known_inf_or_nan(
 ; CHECK-SAME: float nofpclass(nan inf sub norm) [[ARG0:%.*]], float nofpclass(zero sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -565,7 +565,7 @@ define float @ret_known_zero_mul_known_inf_or_nan(float nofpclass(nan inf norm s
 
 ; -> nan
 define float @ret_known_zero_or_nan_mul_known_inf(float nofpclass(inf norm sub) %arg0, float nofpclass(nan norm sub zero) %arg1) {
-; CHECK-LABEL: define float @ret_known_zero_or_nan_mul_known_inf(
+; CHECK-LABEL: define nofpclass(inf zero sub norm) float @ret_known_zero_or_nan_mul_known_inf(
 ; CHECK-SAME: float nofpclass(inf sub norm) [[ARG0:%.*]], float nofpclass(nan zero sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -602,7 +602,7 @@ define float @ret_fmul_rhs_known_positive_or_nan(float %lhs, float %rhs) {
 
 ; Result must be positive or nan
 define float @ret_fmul_both_signs_positive_or_nan(float %lhs, float %rhs) {
-; CHECK-LABEL: define float @ret_fmul_both_signs_positive_or_nan(
+; CHECK-LABEL: define nofpclass(ninf nzero nsub nnorm) float @ret_fmul_both_signs_positive_or_nan(
 ; CHECK-SAME: float [[LHS:%.*]], float [[RHS:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[LHS_FABS:%.*]] = call float @llvm.fabs.f32(float [[LHS]]) #[[ATTR2]]
 ; CHECK-NEXT:    [[RHS_FABS:%.*]] = call float @llvm.fabs.f32(float [[RHS]]) #[[ATTR2]]
@@ -617,7 +617,7 @@ define float @ret_fmul_both_signs_positive_or_nan(float %lhs, float %rhs) {
 
 ; Result must be positive or nan
 define float @ret_fmul_both_signs_negative_or_nan(float %lhs, float %rhs) {
-; CHECK-LABEL: define float @ret_fmul_both_signs_negative_or_nan(
+; CHECK-LABEL: define nofpclass(ninf nzero nsub nnorm) float @ret_fmul_both_signs_negative_or_nan(
 ; CHECK-SAME: float [[LHS:%.*]], float [[RHS:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[LHS_FABS:%.*]] = call float @llvm.fabs.f32(float [[LHS]]) #[[ATTR2]]
 ; CHECK-NEXT:    [[RHS_FABS:%.*]] = call float @llvm.fabs.f32(float [[RHS]]) #[[ATTR2]]
@@ -636,7 +636,7 @@ define float @ret_fmul_both_signs_negative_or_nan(float %lhs, float %rhs) {
 
 ; Result must be negative or nan
 define float @ret_fmul_lhs_negative_rhs_positive(float %lhs, float %rhs) {
-; CHECK-LABEL: define float @ret_fmul_lhs_negative_rhs_positive(
+; CHECK-LABEL: define nofpclass(pinf pzero psub pnorm) float @ret_fmul_lhs_negative_rhs_positive(
 ; CHECK-SAME: float [[LHS:%.*]], float [[RHS:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[LHS_FABS:%.*]] = call float @llvm.fabs.f32(float [[LHS]]) #[[ATTR2]]
 ; CHECK-NEXT:    [[RHS_FABS:%.*]] = call float @llvm.fabs.f32(float [[RHS]]) #[[ATTR2]]
@@ -653,7 +653,7 @@ define float @ret_fmul_lhs_negative_rhs_positive(float %lhs, float %rhs) {
 
 ; Result must be negative or nan
 define float @ret_fmul_rhs_negative_lhs_positive(float %lhs, float %rhs) {
-; CHECK-LABEL: define float @ret_fmul_rhs_negative_lhs_positive(
+; CHECK-LABEL: define nofpclass(pinf pzero psub pnorm) float @ret_fmul_rhs_negative_lhs_positive(
 ; CHECK-SAME: float [[LHS:%.*]], float [[RHS:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[LHS_FABS:%.*]] = call float @llvm.fabs.f32(float [[LHS]]) #[[ATTR2]]
 ; CHECK-NEXT:    [[RHS_FABS:%.*]] = call float @llvm.fabs.f32(float [[RHS]]) #[[ATTR2]]
@@ -670,7 +670,7 @@ define float @ret_fmul_rhs_negative_lhs_positive(float %lhs, float %rhs) {
 
 ; Could be inf of either sign
 define float @ret_known_inf_fmul_known_inf(float nofpclass(norm sub zero nan) %arg0, float nofpclass(norm sub zero nan) %arg1) {
-; CHECK-LABEL: define nofpclass(nan) float @ret_known_inf_fmul_known_inf(
+; CHECK-LABEL: define nofpclass(nan zero sub norm) float @ret_known_inf_fmul_known_inf(
 ; CHECK-SAME: float nofpclass(nan zero sub norm) [[ARG0:%.*]], float nofpclass(nan zero sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL:%.*]] = fmul float [[ARG0]], [[ARG1]]
 ; CHECK-NEXT:    ret float [[FMUL]]
@@ -681,7 +681,7 @@ define float @ret_known_inf_fmul_known_inf(float nofpclass(norm sub zero nan) %a
 
 ; Could be inf of either sign, or nan
 define float @ret_known_inf_fmul_known_inf_or_nan(float nofpclass(norm sub zero nan) %arg0, float nofpclass(norm sub zero) %arg1) {
-; CHECK-LABEL: define float @ret_known_inf_fmul_known_inf_or_nan(
+; CHECK-LABEL: define nofpclass(zero sub norm) float @ret_known_inf_fmul_known_inf_or_nan(
 ; CHECK-SAME: float nofpclass(nan zero sub norm) [[ARG0:%.*]], float nofpclass(zero sub norm) [[ARG1:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:    [[FMUL...
[truncated]

@arsenm arsenm mentioned this pull request Dec 22, 2025
Open
@github-actions
Copy link

github-actions bot commented Dec 22, 2025 •
edited
Loading

🪟 Windows x64 Test Results

  • 128866 tests passed
  • 2839 tests skipped

✅ The build succeeded and all tests passed.

@github-actions
Copy link

github-actions bot commented Dec 22, 2025 •
edited
Loading

🐧 Linux x64 Test Results

  • 187777 tests passed
  • 4986 tests skipped

✅ The build succeeded and all tests passed.

@arsenm
ValueTracking: Improve computeKnownFPClass fmul handling
ab51313 
Improve known non-nan sign bit tracking. Handle cases with
a known 0 or inf input of indeterminate sign.

The tails of some library functions have sign management
for special cases.
@arsenm arsenm force-pushed the users/arsenm/valuetracking/add-baseline-tests-computeKnownFPClass-fmul-zero-inf branch from 473fb9d to 894e5f6 Compare December 22, 2025 12:56
@arsenm arsenm force-pushed the users/arsenm/valuetracking/improve-fmul-0-handling branch from eb5db7e to ab51313 Compare December 22, 2025 12:56
@llvmbot llvmbot added backend:AMDGPU llvm:instcombine Covers the InstCombine, InstSimplify and AggressiveInstCombine passes labels Dec 22, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@andykaylor andykaylor Awaiting requested review from andykaylor

@dtcxzyw dtcxzyw Awaiting requested review from dtcxzyw

@jayfoad jayfoad Awaiting requested review from jayfoad

@jcranmer-intel jcranmer-intel Awaiting requested review from jcranmer-intel

@nikic nikic Awaiting requested review from nikic nikic is a code owner

Assignees

No one assigned

Labels

backend:AMDGPU floating-point Floating-point math llvm:analysis Includes value tracking, cost tables and constant folding llvm:instcombine Covers the InstCombine, InstSimplify and AggressiveInstCombine passes llvm:transforms

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@arsenm @llvmbot