[llvm] [RegisterCoalescer] Fix subrange update when rematerialization widens a def (PR #151974)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-llvm-regalloc

@llvm/pr-subscribers-backend-systemz

Author: None (KRM7)

<details>
<summary>Changes</summary>

Currently, when an instruction rematerialized by the register coalescer defines more subregs of the destination register
than the original COPY instruction did, we only add dead defs for the newly defined subregs if they were not defined anywhere
else. For example, consider something like this before rematerialization:
```
 %0:reg64 = CONSTANT 1
 %1:reg128.sub_lo64_lo32 = COPY %0.lo32
 %1:reg128.sub_lo64_hi32 = ...
 ...
```
that would look like this after rematerializing `%0`:
```
 %0:reg64 = CONSTANT 2
 %1:reg128.sub_lo64 = CONSTANT 2
 %1:reg128.sub_lo64_hi32 = ...
 ...
```
A dead def would not be added for `%1.sub_lo64_hi32` at the 2nd instruction because it's subrange wasn't empty beforehand.

---
Full diff: https://github.com/llvm/llvm-project/pull/151974.diff


2 Files Affected:

- (modified) llvm/lib/CodeGen/RegisterCoalescer.cpp (+5-5) 
- (modified) llvm/test/CodeGen/SystemZ/regcoal-subranges-update-remat.mir (+25) 


``````````diff
diff --git a/llvm/lib/CodeGen/RegisterCoalescer.cpp b/llvm/lib/CodeGen/RegisterCoalescer.cpp
index 2d7987a2e1988..514f2f02d6425 100644
--- a/llvm/lib/CodeGen/RegisterCoalescer.cpp
+++ b/llvm/lib/CodeGen/RegisterCoalescer.cpp
@@ -1624,11 +1624,11 @@ bool RegisterCoalescer::reMaterializeTrivialDef(const CoalescerPair &CP,
           UpdatedSubRanges = true;
         } else {
           // We know that this lane is defined by this instruction,
-          // but at this point it may be empty because it is not used by
-          // anything. This happens when updateRegDefUses adds the missing
-          // lanes. Assign that lane a dead def so that the interferences
-          // are properly modeled.
-          if (SR.empty())
+          // but at this point it might not be live because it was not defined
+          // by the original instruction. This happens when the
+          // rematerialization widens the defined register. Assign that lane a
+          // dead def so that the interferences are properly modeled.
+          if (!SR.liveAt(DefIndex))
             SR.createDeadDef(DefIndex, Alloc);
         }
       }
diff --git a/llvm/test/CodeGen/SystemZ/regcoal-subranges-update-remat.mir b/llvm/test/CodeGen/SystemZ/regcoal-subranges-update-remat.mir
index e3207df799449..d9fe810e23b3a 100644
--- a/llvm/test/CodeGen/SystemZ/regcoal-subranges-update-remat.mir
+++ b/llvm/test/CodeGen/SystemZ/regcoal-subranges-update-remat.mir
@@ -43,3 +43,28 @@ body:             |
     %3:gr32bit = COPY killed %1
     Return implicit %3
 ...
+
+---
+name:            test_dead_at_remat_later_defined
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    ; CHECK-LABEL: name: test_dead_at_remat_later_defined
+    ; CHECK: undef [[LHI:%[0-9]+]].subreg_l32:gr128bit = LHI 0
+    ; CHECK-NEXT: [[LHI:%[0-9]+]].subreg_l64:gr128bit = LGHI 2
+    ; CHECK-NEXT: [[LHI1:%[0-9]+]]:gr32bit = LHI 1
+    ; CHECK-NEXT: [[LHI:%[0-9]+]].subreg_lh32:gr128bit = COPY [[LHI1]]
+    ; CHECK-NEXT: [[LGHI:%[0-9]+]]:gr64bit = LGHI 2
+    ; CHECK-NEXT: [[LHI:%[0-9]+]].subreg_h32:gr128bit = COPY [[LGHI]].subreg_l32
+    ; CHECK-NEXT: $r0q = COPY [[LHI]]
+    ; CHECK-NEXT: $r4d = COPY [[LGHI]].subreg_h32
+    %0:gr64bit = LGHI 2
+    %1:gr32bit = LHI 0
+    %2:gr32bit = LHI 1
+    undef %3.subreg_ll32:gr128bit = COPY %0.subreg_l32
+    %3.subreg_lh32:gr128bit = COPY %2
+    %3.subreg_l32:gr128bit = COPY %1
+    %3.subreg_h32:gr128bit = COPY %0.subreg_l32
+    $r0q = COPY %3
+    $r4d = COPY %0.subreg_h32
+...

``````````

</details>


https://github.com/llvm/llvm-project/pull/151974