[clang codegen] Emit int TBAA metadata on FP math libcall expf #96025

vfdff · 2024-06-19T05:31:41Z

Base on the discussion https://discourse.llvm.org/t/fp-can-we-add-pure-attribute-for-math-library-functions-default/79459, math libcalls set errno, so it should emit "int" TBAA metadata on FP libcalls to solve the alias issue.

Note: Only add support for expf in this PR

Fix #86635

llvmbot · 2024-06-19T05:32:10Z

@llvm/pr-subscribers-clang-codegen

@llvm/pr-subscribers-clang

Author: Allen (vfdff)

Changes

Base on the discussion https://discourse.llvm.org/t/fp-can-we-add-pure-attribute-for-math-library-functions-default/79459, math libcalls set errno, so it should emit "int" TBAA metadata on FP libcalls to solve the alias issue.

Fix #86635

Full diff: https://github.com/llvm/llvm-project/pull/96025.diff

2 Files Affected:

(modified) clang/lib/CodeGen/CGBuiltin.cpp (+32-1)
(added) clang/test/CodeGen/math-libcalls-tbaa.cpp (+22)

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 08a89bd123d03..dc4af109cdbdb 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -707,7 +707,38 @@ static RValue emitLibraryCall(CodeGenFunction &CGF, const FunctionDecl *FD,
                               const CallExpr *E, llvm::Constant *calleeValue) {
   CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E);
   CGCallee callee = CGCallee::forDirect(calleeValue, GlobalDecl(FD));
-  return CGF.EmitCall(E->getCallee()->getType(), callee, E, ReturnValueSlot());
+  RValue Call =
+      CGF.EmitCall(E->getCallee()->getType(), callee, E, ReturnValueSlot());
+
+  // Check the supported intrinsic.
+  if (unsigned BuiltinID = FD->getBuiltinID()) {
+    auto IntrinsicID = [&]() -> unsigned {
+      switch (BuiltinID) {
+      case Builtin::BIexpf:
+      case Builtin::BI__builtin_expf:
+      case Builtin::BI__builtin_expf128:
+        return true;
+      }
+      // TODO: support more FP math libcalls
+      return false;
+    }();
+
+    if (IntrinsicID) {
+      llvm::MDBuilder MDHelper(CGF.getLLVMContext());
+      MDNode *RootMD;
+      if (CGF.getLangOpts().CPlusPlus)
+        RootMD = MDHelper.createTBAARoot("Simple C++ TBAA");
+      else
+        RootMD = MDHelper.createTBAARoot("Simple C/C++ TBAA");
+      // Emit "int" TBAA metadata on FP math libcalls.
+      MDNode *AliasType = MDHelper.createTBAANode("int", RootMD);
+      MDNode *MDInt = MDHelper.createTBAAStructTagNode(AliasType, AliasType, 0);
+
+      Value *Val = Call.getScalarVal();
+      cast<llvm::Instruction>(Val)->setMetadata(LLVMContext::MD_tbaa, MDInt);
+    }
+  }
+  return Call;
 }
 
 /// Emit a call to llvm.{sadd,uadd,ssub,usub,smul,umul}.with.overflow.*
diff --git a/clang/test/CodeGen/math-libcalls-tbaa.cpp b/clang/test/CodeGen/math-libcalls-tbaa.cpp
new file mode 100644
index 0000000000000..b3241c97da71f
--- /dev/null
+++ b/clang/test/CodeGen/math-libcalls-tbaa.cpp
@@ -0,0 +1,22 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 2
+// RUN:  %clang -S -O3 -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefixes=CHECK
+
+#include <math.h>
+
+
+// Emit int TBAA metadata on FP math libcalls, which is useful for alias analysis
+
+// CHECK-LABEL: define dso_local noundef float @_Z3fooPffi
+// CHECK-SAME: (ptr nocapture noundef readonly [[NUM:%.*]], float noundef [[R2INV:%.*]], i32 noundef [[N:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[NUM]], i64 40
+// CHECK-NEXT:    [[TMP0:%.*]] = load float, ptr [[ARRAYIDX]], align 4, !tbaa [[TBAA6:![0-9]+]]
+// CHECK-NEXT:    [[CALL_I:%.*]] = tail call noundef float @expf(float noundef [[TMP0]]) #[[ATTR2:[0-9]+]], !tbaa [[TBAA10:![0-9]+]]
+// CHECK-NEXT:    [[MUL:%.*]] = fmul float [[TMP0]], [[CALL_I]]
+// CHECK-NEXT:    ret float [[MUL]]
+//
+float foo (float num[], float r2inv, int n) {
+   const float expm2 =  std::exp(num[10]);  # Emit TBAA metadata on @expf
+   float tmp = expm2 * num[10];
+   return tmp;
+}

efriedma-quic

Ideally we could identify errno more precisely somehow, but I guess this is better than nothing.

Can we restrict this to targets where libm actually modifies errno?

How does this interact with StrictFP?

efriedma-quic · 2024-06-19T23:21:22Z

clang/lib/CodeGen/CGBuiltin.cpp

+      llvm::MDBuilder MDHelper(CGF.getLLVMContext());
+      MDNode *RootMD;
+      if (CGF.getLangOpts().CPlusPlus)
+        RootMD = MDHelper.createTBAARoot("Simple C++ TBAA");


Please don't copy-paste code from CodeGenTBAA.

I don't find the way to call CodeGenTBAA::getRoot directly, so add a public interface getTBAARoot, thanks.

vfdff · 2024-06-20T03:50:31Z

Can we restrict this to targets where libm actually modifies errno?

Do you mean it is better to have a function list, which actually modifies errno ? For example, we should restrict to __exp10, but but not a general top call __exp ?
https://github.com/lattera/glibc/blob/895ef79e04a953cac1493863bcae29ad85657ee1/math/w_exp10_template.c#L35

How does this interact with StrictFP?

I don't think it related to StrictFP, because we only care whether the global variable errno is modified? which may bring in alias to other global variable. As now the type of errno is int, so we record that for related function.

vfdff · 2024-06-21T02:41:05Z

hi @efriedma-quic, I surprise to find there are 4 new failed tests with new commit 9990877a2 because timeout, but the change seems tiny, and doesn't see marked change on compile time local with this commit , so I need some time to figure out the reason. If you have any suggestions, it's also very welcome.

efriedma-quic · 2024-06-21T21:17:35Z

Can we restrict this to targets where libm actually modifies errno?

Do you mean it is better to have a function list, which actually modifies errno ? For example, we should restrict to __exp10, but but not a general top call __exp ?

I said targets, not functions. For example, MacOS never set errno for math functions.

efriedma-quic · 2024-06-21T21:15:19Z

clang/lib/CodeGen/CGBuiltin.cpp

+      llvm::MDBuilder MDHelper(CGF.getLLVMContext());
+      MDNode *RootMD = CGF.CGM.getTBAARoot();
+      // Emit "int" TBAA metadata on FP math libcalls.
+      MDNode *AliasType = MDHelper.createTBAANode("int", RootMD);


We should have code somewhere for getting metadata for "int"?

sorry, I don't find the metadata for "int", and the createTBAAScalarTypeNode is only used in CodeGenTBAA::createScalarTypeNode

ps：I will update the createTBAANode into createTBAAScalarTypeNode according switch from old TBAA format to the new struct-path aware TBAA format

Looked it up; should be CGM.getTBAATypeInfo(Context.IntTy)?

Yes, apply your comment, thanks

efriedma-quic · 2024-06-21T21:21:13Z

I can't think of any obvious reason this would cause timeouts. I mean, you're adding metadata to a bunch of functions, and if something handles that metadata inefficiently, things could easily explode. But nothing specific comes to mind.

efriedma-quic · 2024-06-24T17:38:44Z

clang/lib/CodeGen/CGBuiltin.cpp

+    }();
+
+    const llvm::Triple &T = CGF.getTarget().getTriple();
+    // Restrict to Linux because not all targets set errno, such as MacOS.


LangOpts.MathErrno

efriedma-quic · 2024-06-24T17:42:26Z

clang/lib/CodeGen/CGBuiltin.cpp

+      llvm::MDBuilder MDHelper(CGF.getLLVMContext());
+      MDNode *RootMD = CGF.CGM.getTBAARoot();
+      // Emit "int" TBAA metadata on FP math libcalls.
+      MDNode *AliasType = MDHelper.createTBAANode("int", RootMD);


Looked it up; should be CGM.getTBAATypeInfo(Context.IntTy)?

vfdff · 2024-06-25T04:01:06Z

It report error Access tag metadata must have either 4 or 5 operands with CGM.getTBAATypeInfo(Context.IntTy)

p MDHelper.createTBAAScalarTypeNode("int", RootMD)
  <0xaaaabe1f62a8> = !{!"int", <0xaaaabddc68f8>, i64 0}  -- return a Type descriptors
p CGF.CGM.getTBAATypeInfo(CGF.getContext().IntTy)
  <0xaaaabe0c93b8> = !{<0xaaaabddfd208>, i64 4, !"int"}  --  Is it a new path-tbaa?

vfdff · 2024-06-25T06:33:45Z

It report error Access tag metadata must have either 4 or 5 operands with CGM.getTBAATypeInfo(Context.IntTy)

p MDHelper.createTBAAScalarTypeNode("int", RootMD)
  <0xaaaabe1f62a8> = !{!"int", <0xaaaabddc68f8>, i64 0}  -- return a Type descriptors
p CGF.CGM.getTBAATypeInfo(CGF.getContext().IntTy)
  <0xaaaabe0c93b8> = !{<0xaaaabddfd208>, i64 4, !"int"}  --  Is it a new path-tbaa?

Oh, it works fine when I delete the option -Xclang -new-struct-path-tbaa

efriedma-quic · 2024-06-25T16:41:51Z

clang/lib/CodeGen/CGBuiltin.cpp

+      // Emit "int" TBAA metadata on FP math libcalls.
+      clang::QualType IntTy = Context.IntTy;
+      MDNode *AliasType = CGF.CGM.getTBAATypeInfo(IntTy);
+      MDNode *MDInt = MDHelper.createTBAAStructTagNode(AliasType, AliasType, 0);


I think you want CodeGenModule::getTBAAAccessTagInfo, not getTBAATypeInfo? (My mistake.)

Apply your comment, thanks very much

vfdff · 2024-06-29T07:06:42Z

hi, is there any new suggestion ?

vfdff · 2024-07-03T10:47:16Z

ping ?

zahiraam · 2024-07-03T15:13:47Z

clang/test/CodeGen/math-libcalls-tbaa.cpp

+
+// RUN:  %clang -S -O3 -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefixes=CHECK,NoNewStructPathTBAA
+// RUN:  %clang -S -O3 -Xclang -new-struct-path-tbaa -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefixes=CHECK,NewStructPathTBAA
+


Why are you using clang here instead of clang_cc1. I think in general CodeGen tests should use clang_cc1. I would think the same than what the other tbaa LIT tests are written.

Why does the test fail on Windows? I don't see any reason to require Linux for this test.

I add a head file <math.h> in the test, and it could not find the head file with clang_cc1

I also tried %clang_cc1 -triple aarch64 -internal-isystem %S/../Headers/Inputs/include, but it did not run into the function emitLibraryCall, which is changed by this PR

I'll try to relax the restriction, add that for windows because it run timeout with my first commit.

hi @zahiraam, It still fail wtih run timeout for window, so would it be better to add this restriction?

Can you rewrite the test so you don't need to include math.h? You don't need the whole header; you just need extern "C" float expf(float);.

We try to avoid including system headers in clang regression tests so the tests work consistently across platforms. If we actually need to run something natively for some reason, we it in llvm-test-suite.

Thanks, apply your idea. But the test is still fail with timeout on the windows target.

@vfdff I have uploaded your patch and ran the LIT test on windows. It's failing mostly because the //CHECK lines you have in your LIT test are Linux's output. What you are interested by is the tbaa metada at the call of expf right? So it's enough to just put that check line.
This test is passing for me on Windows (I didn't test it on Linux. Can you do that?):
// RUN: %clang_cc1 -fmath-errno -O3 -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefixes=CHECK,NoNewStructPathTBAA
// RUN: %clang_cc1 -fmath-errno -O3 -new-struct-path-tbaa -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefixes=CHECK,NewStructPathTBAA

extern "C" float expf(float);

// Emit int TBAA metadata on FP math libcalls, which is useful for alias analysis

// CHECK-LABEL: define dso_local noundef float {{.*}}foo
// CHECK-SAME: ptr nocapture noundef readonly [[NUM:%.*]], float noundef [[R2INV:%.*]], i32 noundef [[N:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
// CHECK-NEXT: [[ENTRY:.*:]]
// CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[NUM]], i64 40
// CHECK-NEXT: [[TMP0:%.*]] = load float, ptr [[ARRAYIDX]], align 4, !tbaa [[TBAA2:![0-9]+]]
// CHECK-NEXT: [[CALL:%.*]] = tail call float @expf(float noundef [[TMP0]]) #[[ATTR2:[0-9]+]]

float foo (float num[], float r2inv, int n) {
const float expm2 = expf(num[10]); // Emit TBAA metadata on @expf
float tmp = expm2 * num[10];
return tmp;
}
//.
// NoNewStructPathTBAA: [[TBAA2]] = !{[[META3:![0-9]+]], [[META3]], i64 0}
// NoNewStructPathTBAA: [[META3]] = !{!"float", [[META4:![0-9]+]], i64 0}
// NoNewStructPathTBAA: [[META4]] = !{!"omnipotent char", [[META5:![0-9]+]], i64 0}
// NoNewStructPathTBAA: [[META5]] = !{!"Simple C++ TBAA"}
//.
// NewStructPathTBAA: [[TBAA2]] = !{[[META3:![0-9]+]], [[META3]], i64 0, i64 4}
// NewStructPathTBAA: [[META3]] = !{[[META4:![0-9]+]], i64 4, !"float"}
// NewStructPathTBAA: [[META4]] = !{[[META5:![0-9]+]], i64 1, !"omnipotent char"}
// NewStructPathTBAA: [[META5]] = !{!"Simple C++ TBAA"}
//.
//// NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
// NewStructPathTBAA: {{.*}}
// NoNewStructPathTBAA: {{.*}}

Thanks @zahiraam very much.
Yes, I expected tbaa metada at the call of expf with the PR. so this checking line don't include the tbaa metada because you have not rebuild the clang with the PR ?

CHECK-NEXT: [[CALL:%.*]] = tail call float @expf(float noundef [[TMP0]]) #[[ATTR2:[0-9]+]]

As there is difference function naming rules between Linux and Windows, I try to add extern "C" for the function, then the case pass on windows now, see #98704. I'll rebase this PR after PR98704 is accepted.

Yes, you are right I didn't notice that. Sorry. There is no TBAA on the expf call. But there should be since I uploaded your entire patch! I don't understand why the extern C would make any difference. Furthermore, I ran the LIT test with it.

Thanks for checking.
I think extern C will make the output function names in the assembly named after the standard C language, i.e. no extra prefix and suffix.

clang/lib/CodeGen/CGBuiltin.cpp

efriedma-quic

How does this interact with StrictFP?

I don't think it related to StrictFP, because we only care whether the global variable errno is modified? which may bring in alias to other global variable. As now the type of errno is int, so we record that for related function.

In strictfp mode, expf() can also modify the floating-point status bits... which count as "memory". So we need to make sure we treat reads and writes to the status bits as aliasing.

clang/lib/CodeGen/CGBuiltin.cpp

efriedma-quic · 2024-07-12T19:58:12Z

clang/test/CodeGen/math-libcalls-tbaa.cpp

+
+// RUN:  %clang -S -O3 -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefixes=CHECK,NoNewStructPathTBAA
+// RUN:  %clang -S -O3 -Xclang -new-struct-path-tbaa -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefixes=CHECK,NewStructPathTBAA
+


There isn't any obvious reason for it to time out on Windows specifically... does specifying a target triple help? Otherwise, not sure. We definitely need to sort out whatever is happening if the compiler breaks on Windows.

vfdff · 2024-07-13T02:46:02Z

hi @efriedma-quic I find the case will timeout even without this changes ,so would it reasonable to add require Linux for this test now on PR98704 ?

detail link: https://buildkite.com/llvm-project/github-pull-requests/builds/80879#0190a9c3-b007-4daa-9656-0c8f00601ee9

Base on the discussion https://discourse.llvm.org/t/fp-can-we-add-pure-attribute-for-math-library-functions-default/79459, math libcalls set errno, so it should emit "int" TBAA metadata on FP libcalls to solve the alias issue. Fix llvm#86635

vfdff · 2024-07-18T01:23:11Z

hi @efriedma-quic @zahiraam I have rebase the PR base on the top main branch, and also addresss all comments, please help to check any new sugguestion? thanks.

zahiraam · 2024-07-18T13:23:33Z

LGTM. Thanks.

efriedma-quic

LGTM with one minor comment

efriedma-quic · 2024-07-18T18:52:23Z

clang/lib/CodeGen/CGBuiltin.cpp

+
+  // Check the supported intrinsic.
+  if (unsigned BuiltinID = FD->getBuiltinID()) {
+    auto IntrinsicID = [&]() -> unsigned {


"IntrinsicID" is a boolean; please rename (maybe something like IsErrnoIntrinsic).

apply your comment, thanks

llvm-ci · 2024-07-19T03:48:25Z

LLVM Buildbot has detected a new failure on builder llvm-clang-x86_64-darwin running on doug-worker-3 while building clang at step 6 "test-build-unified-tree-check-all".

Full details are available at: https://lab.llvm.org/buildbot/#/builders/23/builds/1154

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

Step 6 (test-build-unified-tree-check-all) failure: test (failure)
******************** TEST 'LLVM :: Analysis/StructuralHash/structural-hash-printer.ll' FAILED ********************
Exit Code: 1

Command Output (stderr):
--
RUN: at line 1: /Volumes/RAMDisk/buildbot-root/x86_64-darwin/build/bin/opt -passes='print<structural-hash>' -disable-output /Volumes/RAMDisk/buildbot-root/x86_64-darwin/llvm-project/llvm/test/Analysis/StructuralHash/structural-hash-printer.ll 2>&1 | /Volumes/RAMDisk/buildbot-root/x86_64-darwin/build/bin/FileCheck /Volumes/RAMDisk/buildbot-root/x86_64-darwin/llvm-project/llvm/test/Analysis/StructuralHash/structural-hash-printer.ll
+ /Volumes/RAMDisk/buildbot-root/x86_64-darwin/build/bin/opt '-passes=print<structural-hash>' -disable-output /Volumes/RAMDisk/buildbot-root/x86_64-darwin/llvm-project/llvm/test/Analysis/StructuralHash/structural-hash-printer.ll
+ /Volumes/RAMDisk/buildbot-root/x86_64-darwin/build/bin/FileCheck /Volumes/RAMDisk/buildbot-root/x86_64-darwin/llvm-project/llvm/test/Analysis/StructuralHash/structural-hash-printer.ll
RUN: at line 2: /Volumes/RAMDisk/buildbot-root/x86_64-darwin/build/bin/opt -passes='print<structural-hash><detailed>' -disable-output /Volumes/RAMDisk/buildbot-root/x86_64-darwin/llvm-project/llvm/test/Analysis/StructuralHash/structural-hash-printer.ll 2>&1 | /Volumes/RAMDisk/buildbot-root/x86_64-darwin/build/bin/FileCheck /Volumes/RAMDisk/buildbot-root/x86_64-darwin/llvm-project/llvm/test/Analysis/StructuralHash/structural-hash-printer.ll -check-prefix=DETAILED-HASH
+ /Volumes/RAMDisk/buildbot-root/x86_64-darwin/build/bin/opt '-passes=print<structural-hash><detailed>' -disable-output /Volumes/RAMDisk/buildbot-root/x86_64-darwin/llvm-project/llvm/test/Analysis/StructuralHash/structural-hash-printer.ll
+ /Volumes/RAMDisk/buildbot-root/x86_64-darwin/build/bin/FileCheck /Volumes/RAMDisk/buildbot-root/x86_64-darwin/llvm-project/llvm/test/Analysis/StructuralHash/structural-hash-printer.ll -check-prefix=DETAILED-HASH
/Volumes/RAMDisk/buildbot-root/x86_64-darwin/llvm-project/llvm/test/Analysis/StructuralHash/structural-hash-printer.ll:21:18: error: DETAILED-HASH: expected string not found in input
; DETAILED-HASH: Module Hash: {{([a-z0-9]{14,})}}
                 ^
<stdin>:1:1: note: scanning from here
Module Hash: f76c8321ace7a
^

Input file: <stdin>
Check file: /Volumes/RAMDisk/buildbot-root/x86_64-darwin/llvm-project/llvm/test/Analysis/StructuralHash/structural-hash-printer.ll

-dump-input=help explains the following input dump.

Input was:
<<<<<<
          1: Module Hash: f76c8321ace7a 
check:21     X~~~~~~~~~~~~~~~~~~~~~~~~~~ error: no match found
          2: Function f1 Hash: 4365ad7e28a90d0f 
check:21     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
          3: Function f2 Hash: 9763f9a6861973af 
check:21     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>>>>>

--

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

Summary: Base on the discussion https://discourse.llvm.org/t/fp-can-we-add-pure-attribute-for-math-library-functions-default/79459, math libcalls set errno, so it should emit "int" TBAA metadata on FP libcalls to solve the alias issue. Note: Only add support for expf in this PR Fix #86635 Test Plan: Reviewers: Subscribers: Tasks: Tags: Differential Revision: https://phabricator.intern.facebook.com/D60251692

) Follow #96025, except expf, more FP math libcalls in libm should also be supported. Fix #86635

vfdff requested review from nikic, jdoerfert, andykaylor, karka228 and zahiraam June 19, 2024 05:31

llvmbot added clang Clang issues not falling into any other category clang:codegen IR generation bugs: mangling, exceptions, etc. labels Jun 19, 2024

vfdff force-pushed the PR86635 branch from 741904a to ed6292f Compare June 19, 2024 05:36

nikic removed their request for review June 19, 2024 08:15

efriedma-quic reviewed Jun 19, 2024

View reviewed changes

efriedma-quic reviewed Jun 21, 2024

View reviewed changes

vfdff requested review from manman-ren and efriedma-quic June 22, 2024 09:05

efriedma-quic reviewed Jun 24, 2024

View reviewed changes

efriedma-quic reviewed Jun 25, 2024

View reviewed changes

vfdff requested a review from efriedma-quic June 27, 2024 00:56

zahiraam reviewed Jul 3, 2024

View reviewed changes

clang/lib/CodeGen/CGBuiltin.cpp Show resolved Hide resolved

vfdff changed the title ~~[TBAA] Emit int TBAA metadata on FP math libcalls~~ [TBAA] Emit int TBAA metadata on FP math libcall expf Jul 4, 2024

vfdff requested a review from zahiraam July 10, 2024 02:14

vfdff force-pushed the PR86635 branch from 5b762d0 to 313fb61 Compare July 10, 2024 02:52

efriedma-quic reviewed Jul 12, 2024

View reviewed changes

vfdff requested a review from efriedma-quic July 15, 2024 01:32

vfdff changed the title ~~[TBAA] Emit int TBAA metadata on FP math libcall expf~~ [clang codegen] Emit int TBAA metadata on FP math libcall expf Jul 17, 2024

vfdff force-pushed the PR86635 branch from d9014f6 to ece0662 Compare July 17, 2024 10:28

zahiraam approved these changes Jul 18, 2024

View reviewed changes

efriedma-quic approved these changes Jul 18, 2024

View reviewed changes

rename IntrinsicID to IsErrnoIntrinsic

f4a88a6

vfdff merged commit 1df2e0c into llvm:main Jul 19, 2024
7 checks passed

arsenm mentioned this pull request Jul 25, 2024

[clang codegen] Emit int TBAA metadata on more FP math libcalls #100302

Merged

vfdff added a commit that referenced this pull request Jul 31, 2024

[clang codegen] Emit int TBAA metadata on more FP math libcalls (#100302

9589c12

) Follow #96025, except expf, more FP math libcalls in libm should also be supported. Fix #86635

MacDue mentioned this pull request Sep 17, 2024

[clang][codegen] Don't mark "int" TBAA on FP libcalls with indirect args #108853

Merged


		// RUN: %clang -S -O3 -emit-llvm -o - -x c++ %s \| FileCheck %s -check-prefixes=CHECK,NoNewStructPathTBAA
		// RUN: %clang -S -O3 -Xclang -new-struct-path-tbaa -emit-llvm -o - -x c++ %s \| FileCheck %s -check-prefixes=CHECK,NewStructPathTBAA

[clang codegen] Emit int TBAA metadata on FP math libcall expf #96025

[clang codegen] Emit int TBAA metadata on FP math libcall expf #96025

Uh oh!

Conversation

vfdff commented Jun 19, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Jun 19, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

efriedma-quic left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vfdff Jun 20, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vfdff commented Jun 20, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

vfdff commented Jun 21, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

efriedma-quic commented Jun 21, 2024

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vfdff Jun 22, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

efriedma-quic commented Jun 21, 2024

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vfdff commented Jun 25, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

vfdff commented Jun 25, 2024

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vfdff commented Jun 29, 2024

Uh oh!

vfdff commented Jul 3, 2024

Uh oh!

Choose a reason for hiding this comment

Uh oh!

vfdff Jul 4, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

efriedma-quic left a comment

vfdff commented Jun 19, 2024 •

edited

Loading

llvmbot commented Jun 19, 2024 •

edited

Loading

vfdff Jun 20, 2024 •

edited

Loading

vfdff commented Jun 20, 2024 •

edited

Loading

vfdff commented Jun 21, 2024 •

edited

Loading

vfdff Jun 22, 2024 •

edited

Loading

vfdff commented Jun 25, 2024 •

edited

Loading

vfdff Jul 4, 2024 •

edited

Loading

vfdff commented Jul 13, 2024 •

edited

Loading