[llvm] [AMDGPU][WIP] Optimize SGPR spills (PR #93668)

Christudasan Devadasan via llvm-commits llvm-commits at lists.llvm.org
Mon Jun 3 01:44:07 PDT 2024


================
@@ -1406,6 +1406,12 @@ bool GCNPassConfig::addRegAssignAndRewriteOptimized() {
   // since FastRegAlloc does the replacements itself.
   addPass(createVirtRegRewriter(false));
 
+  // As by this point SGPR's RA is done introducing SGPR spills to stack frame
----------------
cdevadas wrote:

```suggestion
  // As by this point SGPR's RA is done introducing SGPR spills to stack frame
  // through SGPRAllocPass. So, invoking StackSlotColoring here, may allow these
  // SGPR spills to re-use stack slots, before these spills is further lowered
  // down via SILowerSGPRSpills(i.e. equivalent of PEI for SGPRs).
```
 // At this point, the sgpr-regalloc has been done and it is good to have the stack
  // slot coloring to try to optimize the SGPR spill stack indices before attempting
  // the custom SGPR spill lowering.

https://github.com/llvm/llvm-project/pull/93668


More information about the llvm-commits mailing list