[llvm-branch-commits] [llvm] ValueTracking: Handle fdiv by known 0 case in computeKnownFPClass (PR #174855)
Matt Arsenault via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Thu Jan 8 09:13:57 PST 2026
https://github.com/arsenm updated https://github.com/llvm/llvm-project/pull/174855
>From 16afdce72ad2e28456a6220a51cf796942e5022b Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Wed, 7 Jan 2026 19:12:28 +0100
Subject: [PATCH] ValueTracking: Handle fdiv by known 0 case in
computeKnownFPClass
This will help subsume some transforms instsimplify performs
in the middle of SimplifyDemandedFPClass
---
llvm/lib/Analysis/ValueTracking.cpp | 8 +++++
.../Transforms/Attributor/nofpclass-fdiv.ll | 32 +++++++++----------
2 files changed, 24 insertions(+), 16 deletions(-)
diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index 4776020df0bc5..6ee8882850eab 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -5708,6 +5708,14 @@ void computeKnownFPClass(const Value *V, const APInt &DemandedElts,
(KnownLHS.isKnownNever(fcNegative) &&
KnownRHS.isKnownNever(fcPositive)))
Known.knownNot(fcPositive);
+
+ // 0 / x => 0 or nan
+ if (KnownLHS.isKnownAlways(fcZero))
+ Known.knownNot(fcSubnormal | fcNormal | fcInf);
+
+ // x / 0 => nan or inf
+ if (KnownRHS.isKnownAlways(fcZero))
+ Known.knownNot(fcFinite);
} else {
// Inf REM x and x REM 0 produce NaN.
if (KnownLHS.isKnownNeverNaN() && KnownRHS.isKnownNeverNaN() &&
diff --git a/llvm/test/Transforms/Attributor/nofpclass-fdiv.ll b/llvm/test/Transforms/Attributor/nofpclass-fdiv.ll
index 8efed69d9d7c4..e2ebc6ee62aad 100644
--- a/llvm/test/Transforms/Attributor/nofpclass-fdiv.ll
+++ b/llvm/test/Transforms/Attributor/nofpclass-fdiv.ll
@@ -609,7 +609,7 @@ define float @ret_fdiv_f32_known_zero_lhs(float nofpclass(nan inf norm sub) %arg
}
define float @ret_fdiv_f32_known_zero_rhs(float %arg0, float nofpclass(nan inf norm sub) %arg1) {
-; CHECK-LABEL: define float @ret_fdiv_f32_known_zero_rhs
+; CHECK-LABEL: define nofpclass(zero sub norm) float @ret_fdiv_f32_known_zero_rhs
; CHECK-SAME: (float [[ARG0:%.*]], float nofpclass(nan inf sub norm) [[ARG1:%.*]]) #[[ATTR4]] {
; CHECK-NEXT: [[FDIV:%.*]] = fdiv float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FDIV]]
@@ -649,7 +649,7 @@ define float @ret_fdiv_f32_known_pzero_lhs(float nofpclass(nan inf norm sub nzer
}
define float @ret_fdiv_f32_known_pzero_rhs(float %arg0, float nofpclass(nan inf norm sub nzero) %arg1) {
-; CHECK-LABEL: define float @ret_fdiv_f32_known_pzero_rhs
+; CHECK-LABEL: define nofpclass(zero sub norm) float @ret_fdiv_f32_known_pzero_rhs
; CHECK-SAME: (float [[ARG0:%.*]], float nofpclass(nan inf nzero sub norm) [[ARG1:%.*]]) #[[ATTR4]] {
; CHECK-NEXT: [[FDIV:%.*]] = fdiv float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FDIV]]
@@ -689,7 +689,7 @@ define float @ret_fdiv_f32_known_nzero_lhs(float nofpclass(nan inf norm sub pzer
}
define float @ret_fdiv_f32_known_nzero_rhs(float %arg0, float nofpclass(nan inf norm sub pzero) %arg1) {
-; CHECK-LABEL: define float @ret_fdiv_f32_known_nzero_rhs
+; CHECK-LABEL: define nofpclass(zero sub norm) float @ret_fdiv_f32_known_nzero_rhs
; CHECK-SAME: (float [[ARG0:%.*]], float nofpclass(nan inf pzero sub norm) [[ARG1:%.*]]) #[[ATTR4]] {
; CHECK-NEXT: [[FDIV:%.*]] = fdiv float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FDIV]]
@@ -780,7 +780,7 @@ define float @ret_fdiv_f32_known_ninf_or_nan_rhs(float %arg0, float nofpclass(pi
; -> nan
define float @ret_known_inf_fdiv_known_zero(float nofpclass(nan norm sub zero) %arg0, float nofpclass(nan inf norm sub) %arg1) {
-; CHECK-LABEL: define nofpclass(nan) float @ret_known_inf_fdiv_known_zero
+; CHECK-LABEL: define nofpclass(nan zero sub norm) float @ret_known_inf_fdiv_known_zero
; CHECK-SAME: (float nofpclass(nan zero sub norm) [[ARG0:%.*]], float nofpclass(nan inf sub norm) [[ARG1:%.*]]) #[[ATTR4]] {
; CHECK-NEXT: [[FDIV:%.*]] = fdiv float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FDIV]]
@@ -789,9 +789,9 @@ define float @ret_known_inf_fdiv_known_zero(float nofpclass(nan norm sub zero) %
ret float %fdiv
}
-; -> nan
+; -> inf or nan
define float @ret_known_inf_or_nan_fdiv_known_zero(float nofpclass(norm sub zero) %arg0, float nofpclass(nan inf norm sub) %arg1) {
-; CHECK-LABEL: define float @ret_known_inf_or_nan_fdiv_known_zero
+; CHECK-LABEL: define nofpclass(zero sub norm) float @ret_known_inf_or_nan_fdiv_known_zero
; CHECK-SAME: (float nofpclass(zero sub norm) [[ARG0:%.*]], float nofpclass(nan inf sub norm) [[ARG1:%.*]]) #[[ATTR4]] {
; CHECK-NEXT: [[FDIV:%.*]] = fdiv float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FDIV]]
@@ -811,9 +811,9 @@ define float @ret_known_inf_fdiv_known_zero_or_nan(float nofpclass(nan norm sub
ret float %fdiv
}
-; -> nan
+; -> inf
define float @ret_known_ninf_fdiv_known_zero(float nofpclass(nan pinf norm sub zero) %arg0, float nofpclass(nan inf norm sub) %arg1) {
-; CHECK-LABEL: define nofpclass(nan) float @ret_known_ninf_fdiv_known_zero
+; CHECK-LABEL: define nofpclass(nan zero sub norm) float @ret_known_ninf_fdiv_known_zero
; CHECK-SAME: (float nofpclass(nan pinf zero sub norm) [[ARG0:%.*]], float nofpclass(nan inf sub norm) [[ARG1:%.*]]) #[[ATTR4]] {
; CHECK-NEXT: [[FDIV:%.*]] = fdiv float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FDIV]]
@@ -822,9 +822,9 @@ define float @ret_known_ninf_fdiv_known_zero(float nofpclass(nan pinf norm sub z
ret float %fdiv
}
-; -> nan
+; -> pinf
define float @ret_known_pinf_fdiv_known_zero(float nofpclass(nan ninf norm sub zero) %arg0, float nofpclass(nan inf norm sub) %arg1) {
-; CHECK-LABEL: define nofpclass(nan) float @ret_known_pinf_fdiv_known_zero
+; CHECK-LABEL: define nofpclass(nan zero sub norm) float @ret_known_pinf_fdiv_known_zero
; CHECK-SAME: (float nofpclass(nan ninf zero sub norm) [[ARG0:%.*]], float nofpclass(nan inf sub norm) [[ARG1:%.*]]) #[[ATTR4]] {
; CHECK-NEXT: [[FDIV:%.*]] = fdiv float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FDIV]]
@@ -833,9 +833,9 @@ define float @ret_known_pinf_fdiv_known_zero(float nofpclass(nan ninf norm sub z
ret float %fdiv
}
-; -> nan
+; -> zero
define float @ret_known_zero_fdiv_known_inf(float nofpclass(nan inf norm sub) %arg0, float nofpclass(nan norm sub zero) %arg1) {
-; CHECK-LABEL: define nofpclass(nan) float @ret_known_zero_fdiv_known_inf
+; CHECK-LABEL: define nofpclass(nan inf sub norm) float @ret_known_zero_fdiv_known_inf
; CHECK-SAME: (float nofpclass(nan inf sub norm) [[ARG0:%.*]], float nofpclass(nan zero sub norm) [[ARG1:%.*]]) #[[ATTR4]] {
; CHECK-NEXT: [[FDIV:%.*]] = fdiv float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FDIV]]
@@ -844,7 +844,7 @@ define float @ret_known_zero_fdiv_known_inf(float nofpclass(nan inf norm sub) %a
ret float %fdiv
}
-; -> nan
+; -> zero
define float @ret_known_zero_fdiv_known_inf_or_nan(float nofpclass(nan inf norm sub) %arg0, float nofpclass(norm sub zero) %arg1) {
; CHECK-LABEL: define float @ret_known_zero_fdiv_known_inf_or_nan
; CHECK-SAME: (float nofpclass(nan inf sub norm) [[ARG0:%.*]], float nofpclass(zero sub norm) [[ARG1:%.*]]) #[[ATTR4]] {
@@ -855,7 +855,7 @@ define float @ret_known_zero_fdiv_known_inf_or_nan(float nofpclass(nan inf norm
ret float %fdiv
}
-; -> nan
+; -> zero or nan
define float @ret_known_zero_or_nan_fdiv_known_inf(float nofpclass(inf norm sub) %arg0, float nofpclass(nan norm sub zero) %arg1) {
; CHECK-LABEL: define float @ret_known_zero_or_nan_fdiv_known_inf
; CHECK-SAME: (float nofpclass(inf sub norm) [[ARG0:%.*]], float nofpclass(nan zero sub norm) [[ARG1:%.*]]) #[[ATTR4]] {
@@ -995,7 +995,7 @@ define float @ret_known_inf_or_nan_fdiv_known_inf(float nofpclass(norm sub zero)
; Could be zero of either sign
define float @ret_known_zero_fdiv_known_zero(float nofpclass(inf norm sub nan) %arg0, float nofpclass(inf norm sub nan) %arg1) {
-; CHECK-LABEL: define float @ret_known_zero_fdiv_known_zero
+; CHECK-LABEL: define nofpclass(inf zero sub norm) float @ret_known_zero_fdiv_known_zero
; CHECK-SAME: (float nofpclass(nan inf sub norm) [[ARG0:%.*]], float nofpclass(nan inf sub norm) [[ARG1:%.*]]) #[[ATTR4]] {
; CHECK-NEXT: [[FDIV:%.*]] = fdiv float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FDIV]]
@@ -1017,7 +1017,7 @@ define float @ret_known_zero_fdiv_known_zero_or_nan(float nofpclass(inf norm sub
; Could be zero of either sign, or nan
define float @ret_known_zero_or_nan_fdiv_known_zero(float nofpclass(inf norm sub) %arg0, float nofpclass(inf norm sub nan) %arg1) {
-; CHECK-LABEL: define float @ret_known_zero_or_nan_fdiv_known_zero
+; CHECK-LABEL: define nofpclass(zero sub norm) float @ret_known_zero_or_nan_fdiv_known_zero
; CHECK-SAME: (float nofpclass(inf sub norm) [[ARG0:%.*]], float nofpclass(nan inf sub norm) [[ARG1:%.*]]) #[[ATTR4]] {
; CHECK-NEXT: [[FDIV:%.*]] = fdiv float [[ARG0]], [[ARG1]]
; CHECK-NEXT: ret float [[FDIV]]
More information about the llvm-branch-commits
mailing list