blob: bdbf6e9572217a169d9d9487917c582aff6fb8c5 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
|
Fix hiprtc -fgpu-rdc with llvm-23
LLVM 23 does not assign GUIDs to functions linked with -mlink-builtin-bitcode-postopt,
so LTO summary emission aborts with "expected function definition ... to have an
associated value info". Link device libraries pre-opt for HIP, like the clang driver does.
Bug: https://github.com/llvm/llvm-project/pull/211155
--- a/src/comgr-compiler.cpp
+++ b/src/comgr-compiler.cpp
@@ -38,6 +38,7 @@
#include "clang/Frontend/TextDiagnosticPrinter.h"
#include "clang/FrontendTool/Utils.h"
#include "clang/Options/Options.h"
+#include "llvm/Config/llvm-config.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/StringExtras.h"
@@ -1712,8 +1713,15 @@ amd_comgr_status_t AMDGPUCompiler::compileToBitcode(bool WithDeviceLibs) {
// necessary for HIP (for example if HIP adopts the same AMDGPUSimplifyLibs
// strategy that potentially introduces undefined device-library symbols),
// we will need also apply this option in compileToRelocatable().
- Args.push_back("-Xclang");
- Args.push_back("-mlink-builtin-bitcode-postopt");
+#if LLVM_VERSION_MAJOR < 24
+ // LLVM 23 does not assign GUIDs to post-opt linked functions, which
+ // breaks LTO summary emission for -fgpu-rdc; HIP does not need postopt
+ if (getLanguage() != AMD_COMGR_LANGUAGE_HIP)
+#endif
+ {
+ Args.push_back("-Xclang");
+ Args.push_back("-mlink-builtin-bitcode-postopt");
+ }
}
return processFiles(AMD_COMGR_DATA_KIND_BC, ".bc");
|
