[ValueTracking] Teach isGuaranteedNotToBeUndefOrPoison about splats #163570

c-rhodes · 2025-10-15T14:41:14Z

Splats include two poison values, but only the poison-ness of the splatted value actually matters.

llvmbot · 2025-10-15T14:41:47Z

@llvm/pr-subscribers-clang
@llvm/pr-subscribers-llvm-transforms

@llvm/pr-subscribers-llvm-analysis

Author: Cullen Rhodes (c-rhodes)

Changes

Splats include two poison values, but only the poison-ness of the splatted value actually matters.

Full diff: https://github.com/llvm/llvm-project/pull/163570.diff

2 Files Affected:

(modified) llvm/lib/Analysis/ValueTracking.cpp (+5)
(modified) llvm/unittests/Analysis/ValueTrackingTest.cpp (+10)

diff --git a/llvm/lib/Analysis/ValueTracking.cpp b/llvm/lib/Analysis/ValueTracking.cpp
index 9655c886f4441..b0016c36bf5da 100644
--- a/llvm/lib/Analysis/ValueTracking.cpp
+++ b/llvm/lib/Analysis/ValueTracking.cpp
@@ -7678,6 +7678,7 @@ static bool isGuaranteedNotToBeUndefOrPoison(
         return true;
     }
 
+    Value *Splat;
     if (!::canCreateUndefOrPoison(Opr, Kind,
                                   /*ConsiderFlagsAndMetadata=*/true)) {
       if (const auto *PN = dyn_cast<PHINode>(V)) {
@@ -7695,6 +7696,10 @@ static bool isGuaranteedNotToBeUndefOrPoison(
         }
         if (IsWellDefined)
           return true;
+      } else if (isa<ShuffleVectorInst>(Opr) && (Splat = getSplatValue(Opr))) {
+        // For splats we only need to check the value being splatted.
+        if (OpCheck(Splat))
+          return true;
       } else if (all_of(Opr->operands(), OpCheck))
         return true;
     }
diff --git a/llvm/unittests/Analysis/ValueTrackingTest.cpp b/llvm/unittests/Analysis/ValueTrackingTest.cpp
index 559a0b724f383..bb0280ee69cfd 100644
--- a/llvm/unittests/Analysis/ValueTrackingTest.cpp
+++ b/llvm/unittests/Analysis/ValueTrackingTest.cpp
@@ -1091,6 +1091,16 @@ TEST_F(ValueTrackingTest, isGuaranteedNotToBeUndefOrPoison) {
   }
 }
 
+TEST_F(ValueTrackingTest, isGuaranteedNotToBeUndefOrPoison_splat) {
+  parseAssembly(
+      "define <4 x i32> @test(i32 noundef %x) {\n"
+      "  %ins = insertelement <4 x i32> poison, i32 %x, i32 0\n"
+      "  %A = shufflevector <4 x i32> %ins, <4 x i32> poison, <4 x i32> zeroinitializer\n"
+      "  ret <4 x i32> %A\n"
+      "}");
+  EXPECT_TRUE(isGuaranteedNotToBeUndefOrPoison(A));
+}
+
 TEST_F(ValueTrackingTest, isGuaranteedNotToBeUndefOrPoison_assume) {
   parseAssembly("declare i1 @f_i1()\n"
                 "declare i32 @f_i32()\n"

github-actions · 2025-10-15T14:42:57Z

⚠️ C/C++ code formatter, clang-format found issues in your code. ⚠️

You can test this locally with the following command:

git-clang-format --diff origin/main HEAD --extensions c,cpp -- clang/test/CodeGen/arm-mve-intrinsics/dup.c clang/test/Headers/wasm.c llvm/lib/Analysis/ValueTracking.cpp llvm/unittests/Analysis/ValueTrackingTest.cpp --diff_from_common_commit

⚠️
The reproduction instructions above might return results for more than one PR
in a stack if you are using a stacked PR workflow. You can limit the results by
changing origin/main to the base branch/commit you want to compare against.
⚠️

View the diff from clang-format here.

diff --git a/llvm/unittests/Analysis/ValueTrackingTest.cpp b/llvm/unittests/Analysis/ValueTrackingTest.cpp
index bb0280ee6..2c22a7a99 100644
--- a/llvm/unittests/Analysis/ValueTrackingTest.cpp
+++ b/llvm/unittests/Analysis/ValueTrackingTest.cpp
@@ -1092,12 +1092,12 @@ TEST_F(ValueTrackingTest, isGuaranteedNotToBeUndefOrPoison) {
 }
 
 TEST_F(ValueTrackingTest, isGuaranteedNotToBeUndefOrPoison_splat) {
-  parseAssembly(
-      "define <4 x i32> @test(i32 noundef %x) {\n"
-      "  %ins = insertelement <4 x i32> poison, i32 %x, i32 0\n"
-      "  %A = shufflevector <4 x i32> %ins, <4 x i32> poison, <4 x i32> zeroinitializer\n"
-      "  ret <4 x i32> %A\n"
-      "}");
+  parseAssembly("define <4 x i32> @test(i32 noundef %x) {\n"
+                "  %ins = insertelement <4 x i32> poison, i32 %x, i32 0\n"
+                "  %A = shufflevector <4 x i32> %ins, <4 x i32> poison, <4 x "
+                "i32> zeroinitializer\n"
+                "  ret <4 x i32> %A\n"
+                "}");
   EXPECT_TRUE(isGuaranteedNotToBeUndefOrPoison(A));
 }

david-arm

Thanks for this! Seems like a sensible change.

llvm/unittests/Analysis/ValueTrackingTest.cpp

llvm/lib/Analysis/ValueTracking.cpp

dtcxzyw

LGTM. Perhaps we can also teach this function with single-source shufflevectors.

llvm/lib/Analysis/ValueTracking.cpp

nikic

Looks like there are two clang tests that need to be updated as well.

llvm/lib/Analysis/ValueTracking.cpp

david-arm · 2025-10-20T08:08:49Z

llvm/lib/Analysis/ValueTracking.cpp

        }
        if (IsWellDefined)
          return true;
+      } else if (auto *Splat = isa<ShuffleVectorInst>(Opr) ? getSplatValue(Opr)


Just curious about how auto actually works here when selecting between two possibilities? It could choose nullptr, in which case it might be void * I presume? I can only assume it creates the type based on the first option, i.e. getSplatValue. I just wasn't sure whether the code compiles because of luck or an actual C++ specification based on ordering.

The precise rules for determining the result type of a ternary operator are exceedingly complicated, but the tl;dr is that nullptr is of type nullptr_t, which implicitly converts to all pointer types, and that's what makes this work.

llvm-ci · 2025-10-20T10:31:01Z

LLVM Buildbot has detected a new failure on builder llvm-clang-x86_64-sie-ubuntu-fast running on sie-linux-worker while building clang,llvm at step 6 "test-build-unified-tree-check-all".

Full details are available at: https://lab.llvm.org/buildbot/#/builders/144/builds/38197

Here is the relevant piece of the build log for the reference

Step 6 (test-build-unified-tree-check-all) failure: test (failure)
******************** TEST 'lld :: ELF/gc-sections-linker-defined-symbol.s' FAILED ********************
Exit Code: 250

Command Output (stdout):
--
# RUN: at line 3
/home/buildbot/buildbot-root/llvm-clang-x86_64-sie-ubuntu-fast/build/bin/llvm-mc -filetype=obj -triple=x86_64-pc-linux /home/buildbot/buildbot-root/llvm-clang-x86_64-sie-ubuntu-fast/llvm-project/lld/test/ELF/gc-sections-linker-defined-symbol.s -o /home/buildbot/buildbot-root/llvm-clang-x86_64-sie-ubuntu-fast/build/tools/lld/test/ELF/Output/gc-sections-linker-defined-symbol.s.tmp.o
# executed command: /home/buildbot/buildbot-root/llvm-clang-x86_64-sie-ubuntu-fast/build/bin/llvm-mc -filetype=obj -triple=x86_64-pc-linux /home/buildbot/buildbot-root/llvm-clang-x86_64-sie-ubuntu-fast/llvm-project/lld/test/ELF/gc-sections-linker-defined-symbol.s -o /home/buildbot/buildbot-root/llvm-clang-x86_64-sie-ubuntu-fast/build/tools/lld/test/ELF/Output/gc-sections-linker-defined-symbol.s.tmp.o
# note: command had no output on stdout or stderr
# RUN: at line 4
/home/buildbot/buildbot-root/llvm-clang-x86_64-sie-ubuntu-fast/build/bin/ld.lld /home/buildbot/buildbot-root/llvm-clang-x86_64-sie-ubuntu-fast/build/tools/lld/test/ELF/Output/gc-sections-linker-defined-symbol.s.tmp.o -o /home/buildbot/buildbot-root/llvm-clang-x86_64-sie-ubuntu-fast/build/tools/lld/test/ELF/Output/gc-sections-linker-defined-symbol.s.tmp.so --gc-sections -shared
# executed command: /home/buildbot/buildbot-root/llvm-clang-x86_64-sie-ubuntu-fast/build/bin/ld.lld /home/buildbot/buildbot-root/llvm-clang-x86_64-sie-ubuntu-fast/build/tools/lld/test/ELF/Output/gc-sections-linker-defined-symbol.s.tmp.o -o /home/buildbot/buildbot-root/llvm-clang-x86_64-sie-ubuntu-fast/build/tools/lld/test/ELF/Output/gc-sections-linker-defined-symbol.s.tmp.so --gc-sections -shared
# .---command stderr------------
# | terminate called after throwing an instance of 'std::system_error'
# |   what():  Resource temporarily unavailable
# `-----------------------------
# error: command failed with exit status: 250

--

********************

…lvm#163570) Splats include two poison values, but only the poison-ness of the splatted value actually matters.

[ValueTracking] Teach isGuaranteedNotToBeUndefOrPoison about splats

c8cdc8c

Splats include two poison values, but only the poison-ness of the splatted value actually matters.

c-rhodes requested review from david-arm, dtcxzyw and nikic October 15, 2025 14:41

llvmbot added the llvm:analysis Includes value tracking, cost tables and constant folding label Oct 15, 2025

c-rhodes mentioned this pull request Oct 15, 2025

[InstCombine] Push freeze through non-recurrence PHIs #157678

Open

david-arm reviewed Oct 15, 2025

View reviewed changes

llvm/unittests/Analysis/ValueTrackingTest.cpp Show resolved Hide resolved

llvm/lib/Analysis/ValueTracking.cpp Outdated Show resolved Hide resolved

address comments

ec52263

llvmbot added llvm:instcombine Covers the InstCombine, InstSimplify and AggressiveInstCombine passes llvm:transforms labels Oct 15, 2025

dtcxzyw approved these changes Oct 15, 2025

View reviewed changes

llvm/lib/Analysis/ValueTracking.cpp Outdated Show resolved Hide resolved

nikic reviewed Oct 15, 2025

View reviewed changes

llvm/lib/Analysis/ValueTracking.cpp Outdated Show resolved Hide resolved

antoniofrighetto reviewed Oct 15, 2025

View reviewed changes

llvm/lib/Analysis/ValueTracking.cpp Outdated Show resolved Hide resolved

address comments

6cfa5ae

llvmbot added the clang Clang issues not falling into any other category label Oct 16, 2025

address comments

7bf81c7

nikic approved these changes Oct 18, 2025

View reviewed changes

david-arm reviewed Oct 20, 2025

View reviewed changes

c-rhodes merged commit 5b5eacc into llvm:main Oct 20, 2025
9 of 10 checks passed

Lukacma pushed a commit to Lukacma/llvm-project that referenced this pull request Oct 29, 2025

[ValueTracking] Teach isGuaranteedNotToBeUndefOrPoison about splats (l…

fb7ecdd

…lvm#163570) Splats include two poison values, but only the poison-ness of the splatted value actually matters.

aokblast pushed a commit to aokblast/llvm-project that referenced this pull request Oct 30, 2025

[ValueTracking] Teach isGuaranteedNotToBeUndefOrPoison about splats (l…

e3ea0b7

…lvm#163570) Splats include two poison values, but only the poison-ness of the splatted value actually matters.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[ValueTracking] Teach isGuaranteedNotToBeUndefOrPoison about splats #163570

[ValueTracking] Teach isGuaranteedNotToBeUndefOrPoison about splats #163570

Uh oh!

c-rhodes commented Oct 15, 2025

Uh oh!

llvmbot commented Oct 15, 2025 •

edited

Loading

Uh oh!

github-actions bot commented Oct 15, 2025 •

edited

Loading

Uh oh!

david-arm left a comment

Uh oh!

Uh oh!

Uh oh!

dtcxzyw left a comment

Uh oh!

Uh oh!

nikic left a comment

Uh oh!

Uh oh!

Uh oh!

david-arm Oct 20, 2025

Uh oh!

nikic Oct 20, 2025

Uh oh!

Uh oh!

llvm-ci commented Oct 20, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

[ValueTracking] Teach isGuaranteedNotToBeUndefOrPoison about splats #163570

[ValueTracking] Teach isGuaranteedNotToBeUndefOrPoison about splats #163570

Uh oh!

Conversation

c-rhodes commented Oct 15, 2025

Uh oh!

llvmbot commented Oct 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Oct 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

david-arm left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

dtcxzyw left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

nikic left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

david-arm Oct 20, 2025

Choose a reason for hiding this comment

Uh oh!

nikic Oct 20, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

llvm-ci commented Oct 20, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

llvmbot commented Oct 15, 2025 •

edited

Loading

github-actions bot commented Oct 15, 2025 •

edited

Loading