[PATCH] D80338: [AMDGPU] GFX10 PAL: Only pad with s_code_end in pipeline compilation

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp:323-347
+  if (AMDGPU::isGFX10(STI)) {
+    bool AddGuard = STI.getTargetTriple().getOS() == Triple::AMDHSA;
+    if (!AddGuard && STI.getTargetTriple().getOS() == Triple::AMDPAL) {
+      // On PAL, we add the guard only if doing a full pipeline compile, as
+      // there is no later link step. We can spot a full pipeline compile by
+      // there being a PS and at least one of GS or VS, or instead a CS.
+      bool GotPs = false, GotGsOrVs = false, GotCs = false;
----------------
Should move to predicate function


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp:334
+          case CallingConv::AMDGPU_CS:
+            GotCs = true;
+            break;
----------------
Could break out of the whole loop?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D80338/new/

https://reviews.llvm.org/D80338