[RegAlloc] Fix the terminal rule check for interfere with DstReg

[hstk30-hw]

This maybe a bug which is introduced by commit 6749ae3, and has been present ever since.
In this case, OtherReg always overlaps with DstReg cause they from the Copy all.

[llvmbot]

@llvm/pr-subscribers-backend-x86
@llvm/pr-subscribers-backend-webassembly
@llvm/pr-subscribers-backend-hexagon
@llvm/pr-subscribers-backend-aarch64

@llvm/pr-subscribers-llvm-regalloc

Author: None (hstk30-hw)

Changes

This maybe a bug which is introduced by commit 6749ae3, and has been present ever since.
In this case, OtherReg always overlaps with DstReg cause they from the Copy all.

---

Full diff: https://github.com/llvm/llvm-project/pull/168661.diff

1 Files Affected:

• (modified) llvm/lib/CodeGen/RegisterCoalescer.cpp (+1-1)

diff --git a/llvm/lib/CodeGen/RegisterCoalescer.cpp b/llvm/lib/CodeGen/RegisterCoalescer.cpp
index 25c4375a73ce0..e624088a0964e 100644
--- a/llvm/lib/CodeGen/RegisterCoalescer.cpp
+++ b/llvm/lib/CodeGen/RegisterCoalescer.cpp
@@ -4150,7 +4150,7 @@ bool RegisterCoalescer::applyTerminalRule(const MachineInstr &Copy) const {
       continue;
     Register OtherSrcReg, OtherReg;
     unsigned OtherSrcSubReg = 0, OtherSubReg = 0;
-    if (!isMoveInstr(*TRI, &Copy, OtherSrcReg, OtherReg, OtherSrcSubReg,
+    if (!isMoveInstr(*TRI, &MI, OtherSrcReg, OtherReg, OtherSrcSubReg,
                      OtherSubReg))
       return false;
     if (OtherReg == SrcReg)

[hstk30-hw]

There are many testcases fail, so if this is right, then I will fix and update it.

[s-barannikov]

The fix looks intuitively correct to me

[qcolombet]

Great catch @hstk30-hw!
Do you have a test case that exercises this problem.

[hstk30-hw]

Fix all failed testcase.
I also push a commit hstk30-hw@d86334e which compare the old testcases that not turn on terminal rule with default turn on.

[github-actions]

🐧 Linux x64 Test Results

• 186428 tests passed
• 4863 tests skipped

[hstk30-hw]

Do you have a test case that exercises this problem.

Not for now :)

I see luke get a nice performance improvement on MiBench with a 17% reduction in dynamic instruction count.
Can you test again with this patch? @lukel97
#165961 (comment)

[arsenm]

Do you have a test case that exercises this problem.

Not for now :)

I think @qcolombet meant a mir lit test which shows the change, not benchmark results