Skip to content

MC,AMDGPU: Don't pad .text with s_code_end if it would otherwise be empty #147980

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 4 commits into
base: main
Choose a base branch
from

Conversation

trenouf
Copy link
Collaborator

@trenouf trenouf commented Jul 10, 2025

We don't want that padding in a module that only contains data, not code.

Also fix MCSection::hasInstructions() so it works with the asm streamer too.

We don't want that padding in a module that only contains data, not
code.
@llvmbot
Copy link
Member

llvmbot commented Jul 10, 2025

@llvm/pr-subscribers-mc

@llvm/pr-subscribers-backend-amdgpu

Author: Tim Renouf (trenouf)

Changes

We don't want that padding in a module that only contains data, not code.


Full diff: https://github.com/llvm/llvm-project/pull/147980.diff

1 Files Affected:

  • (modified) llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp (+3-1)
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp b/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
index 749b9efc81378..3765c1d2b106c 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
@@ -486,10 +486,12 @@ bool AMDGPUAsmPrinter::doFinalization(Module &M) {
   // Pad with s_code_end to help tools and guard against instruction prefetch
   // causing stale data in caches. Arguably this should be done by the linker,
   // which is why this isn't done for Mesa.
+  // Don't do it if there are no functions.
   const MCSubtargetInfo &STI = *getGlobalSTI();
   if ((AMDGPU::isGFX10Plus(STI) || AMDGPU::isGFX90A(STI)) &&
       (STI.getTargetTriple().getOS() == Triple::AMDHSA ||
-       STI.getTargetTriple().getOS() == Triple::AMDPAL)) {
+       STI.getTargetTriple().getOS() == Triple::AMDPAL) &&
+      !M.empty()) {
     OutStreamer->switchSection(getObjFileLowering().getTextSection());
     getTargetStreamer()->EmitCodeEnd(STI);
   }

Copy link
Contributor

@arsenm arsenm left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Testcase?

const MCSubtargetInfo &STI = *getGlobalSTI();
if ((AMDGPU::isGFX10Plus(STI) || AMDGPU::isGFX90A(STI)) &&
(STI.getTargetTriple().getOS() == Triple::AMDHSA ||
STI.getTargetTriple().getOS() == Triple::AMDPAL)) {
STI.getTargetTriple().getOS() == Triple::AMDPAL) &&
!M.empty()) {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Seems reasonable but I wonder if there's a better way to test "did we emit any code?". E.g. what if the module contains some function declarations but no definitions?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Plus this may misdetect available_externally functions which will be dropped

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I changed to using MCSection::hasInstructions(), which I had to fix to work with the asm streamer.

Also added test. The positive case (when it does add padding) already has a test.

@llvmbot llvmbot added the mc Machine (object) code label Jul 11, 2025
@trenouf trenouf changed the title AMDGPU: Don't pad .text with s_code_end if it would otherwise be empty MC,AMDGPU: Don't pad .text with s_code_end if it would otherwise be empty Jul 11, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
backend:AMDGPU mc Machine (object) code
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants