[llvm-branch-commits] [llvm] [LTO] Prevent crashes if a diagnostic is emitted while compiling with `--lto-partitions` (PR #238259)

Pierre van Houtryve via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Tue Nov 3 01:25:00 PST 2026


https://github.com/Pierre-vh created https://github.com/llvm/llvm-project/pull/238259

Claude found this bug for me while working on the earlier path.
I checked it and I think this fix makes sense. Without the fix in Config.h, the test crashes with `bad_function_call`.

There were other ways to fix it but I think this makes the most sense as it fixes all users.
I only check the full LTO path as all thinLTO paths set a diagnostic handler, and I don't think adding a unit test just to check
the thinLTO path without a diag handler was worth it.

Assisted-By: Claude Sonnet 5.5

>From 781f92433a8b8e88d347ed9251481348960b03f4 Mon Sep 17 00:00:00 2001
From: pvanhout <pierre.vanhoutryve at amd.com>
Date: Tue, 3 Nov 2026 10:22:06 +0100
Subject: [PATCH] [LTO] Prevent crashes if a diagnostic is emitted while
 compiling with `--lto-partitions`

Claude found this bug for me while working on the earlier path.
I checked it and I think this fix makes sense. Without the fix in Config.h, the test crashes with `bad_function_call`.

There were other ways to fix it but I think this makes the most sense as it fixes all users.
I only check the full LTO path as all thinLTO paths set a diagnostic handler, and I don't think adding a unit test just to check
the thinLTO path without a diag handler was worth it.

Assisted-By: Claude Sonnet 5.5
---
 llvm/include/llvm/LTO/Config.h               |  4 +++
 llvm/test/LTO/AMDGPU/parallel-diagnostics.ll | 28 ++++++++++++++++++++
 2 files changed, 32 insertions(+)
 create mode 100644 llvm/test/LTO/AMDGPU/parallel-diagnostics.ll

diff --git a/llvm/include/llvm/LTO/Config.h b/llvm/include/llvm/LTO/Config.h
index 60cb2a08c0381b..6c6fa9827bc28f 100644
--- a/llvm/include/llvm/LTO/Config.h
+++ b/llvm/include/llvm/LTO/Config.h
@@ -322,6 +322,10 @@ struct LTOLLVMDiagnosticHandler : public DiagnosticHandler {
   LTOLLVMDiagnosticHandler(DiagnosticHandlerFunction *DiagHandlerFn)
       : Fn(DiagHandlerFn) {}
   bool handleDiagnostics(const DiagnosticInfo &DI) override {
+    // Let the context apply its default handling if the client did not
+    // provide a callback.
+    if (!*Fn)
+      return false;
     (*Fn)(DI);
     return true;
   }
diff --git a/llvm/test/LTO/AMDGPU/parallel-diagnostics.ll b/llvm/test/LTO/AMDGPU/parallel-diagnostics.ll
new file mode 100644
index 00000000000000..190143bb2539e9
--- /dev/null
+++ b/llvm/test/LTO/AMDGPU/parallel-diagnostics.ll
@@ -0,0 +1,28 @@
+; REQUIRES: amdgpu-registered-target
+
+; Check that diagnostics emitted while code-generating the parallel partitions
+; (-j > 1) are reported instead of crashing. The kernel resource usage remarks
+; are emitted by the AMDGPU backend during codegen. There is a single kernel so
+; that only one partition emits remarks and the output is not interleaved.
+
+; RUN: rm -rf %t && mkdir %t
+; RUN: llvm-as -o %t.bc %s
+
+; Without a cache.
+; RUN: llvm-lto -exported-symbol=A -j2 -pass-remarks-analysis=kernel-resource-usage -o %t/nocache.o %t.bc 2>&1 | FileCheck %s
+; RUN: llvm-nm %t/nocache.o.0 | FileCheck %s --check-prefix=SYMS
+
+; With a cache, on a miss.
+; RUN: llvm-lto -exported-symbol=A -j2 -pass-remarks-analysis=kernel-resource-usage -thinlto-cache-dir=%t/cache -o %t/miss.o %t.bc 2>&1 | FileCheck %s
+; RUN: llvm-nm %t/miss.o.0 | FileCheck %s --check-prefix=SYMS
+
+; CHECK: remark: <unknown>:0:0: Function Name: A
+
+; SYMS: T A
+
+target datalayout = "e-m:e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128:128:48-p9:192:256:256:32-p10:32:32-p11:32:32-p12:32:32-p13:32:32-p14:32:32-p15:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1-ni:7:8:9"
+target triple = "amdgpu9.00-amd-amdhsa"
+
+define amdgpu_kernel void @A() {
+  ret void
+}



More information about the llvm-branch-commits mailing list