RegisterCoalescer: Assume CR_Replace for SubRangeJoin

[arsenm]

Summary

Currently it's possible for following
check on V.WriteLanes (which is not really meaningful
during SubRangeJoin) to pass for one half of the pair,
and then fall through to to one of the impossible
or unresolved states. This then fails as inconsistent
on the other half.

During the main range join, the check between V.WriteLanes
and OtherV.ValidLanes must have passed, meaning this
should be a CR_Replace.

Fixes most of the testcases in bugs 39542 and 39602

[arsenm]

ping

[arsenm]

ping

[arsenm]

ping

[arsenm]

ping

[arsenm]

ping

[qcolombet]

LGTM

[arsenm]

r350678

lib/CodeGen/RegisterCoalescer.cpp

JoinVals::analyzeValue(unsigned ValNo, JoinVals &Other) {
   //   %this  = COPY %ext <-- Erase this copy
   //
   if (DefMI->isFullCopy() && !CP.isPartial() &&
       valuesIdentical(VNI, V.OtherVNI, Other)) {
     V.Identical = true;
     return CR_Erase;
   }
 
+  // The remaining checks apply to the lanes, which aren't tracked here.  This
+  // was already decided to be OK via the following CR_Replace condition.
+  // CR_Replace.
+  if (SubRangeJoin)
+    return CR_Replace;
+
   // If the lanes written by this instruction were all undef in OtherVNI, it is
   // still safe to join the live ranges. This can't be done with a simple value
   // mapping, though - OtherVNI will map to multiple values:
   //
   //   1 %dst:ssub0 = FOO                <-- OtherVNI
   //   2 %src = BAR                      <-- VNI
   //   3 %dst:ssub1 = COPY killed %src    <-- Eliminate this copy.
   //   4 BAZ killed %dst

test/CodeGen/AMDGPU/regcoalesce-cannot-join-failures.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn-amd-amdhsa -verify-coalescing -run-pass=simple-register-coalescing -verify-machineinstrs -o - %s | FileCheck %s
+
+---
+name: couldnt_join_subrange_implicit_def_pred_block
+tracksRegLiveness: true
+body:             |
+  ; CHECK-LABEL: name: couldnt_join_subrange_implicit_def_pred_block
+  ; CHECK: bb.0:
+  ; CHECK:   successors: %bb.1(0x80000000)
+  ; CHECK:   undef %0.sub0:sreg_64_xexec = IMPLICIT_DEF
+  ; CHECK: bb.1:
+  ; CHECK:   successors: %bb.2(0x80000000)
+  ; CHECK:   %0.sub1:sreg_64_xexec = COPY %0.sub0
+  ; CHECK:   S_BRANCH %bb.2
+  ; CHECK: bb.2:
+  ; CHECK:   S_ENDPGM implicit %0
+  bb.0:
+    successors: %bb.1
+
+    undef %0.sub0:sreg_64_xexec = IMPLICIT_DEF
+
+  bb.1:
+    successors: %bb.2
+
+    %1:sreg_64 = COPY %0:sreg_64_xexec
+    %0.sub1:sreg_64_xexec = COPY %0.sub0:sreg_64_xexec
+    S_BRANCH %bb.2
+
+  bb.2:
+    dead %2:sreg_32_xm0 = COPY %0.sub0:sreg_64_xexec
+    S_ENDPGM implicit killed %1
+
+...
+---
+name: couldnt_join_subrange_no_implicit_def_inst
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    ; CHECK-LABEL: name: couldnt_join_subrange_no_implicit_def_inst
+    ; CHECK: undef %0.sub0:sreg_64 = S_MOV_B32 0
+    ; CHECK: %0.sub1:sreg_64 = COPY %0.sub0
+    ; CHECK: S_ENDPGM implicit %0.sub1
+    undef %0.sub0:sreg_64 = S_MOV_B32 0
+    %1:sreg_64 = COPY %0:sreg_64
+    %0.sub1:sreg_64 = COPY %0.sub0:sreg_64
+    S_ENDPGM implicit %1.sub1:sreg_64
+
+...
+---
+name: couldnt_join_subrange0
+tracksRegLiveness: true
+body:             |
+  ; CHECK-LABEL: name: couldnt_join_subrange0
+  ; CHECK: bb.0:
+  ; CHECK:   successors: %bb.1(0x80000000)
+  ; CHECK:   undef %0.sub1:sreg_64 = S_MOV_B32 -1
+  ; CHECK: bb.1:
+  ; CHECK:   %0.sub0:sreg_64 = S_MOV_B32 0
+  ; CHECK:   [[COPY:%[0-9]+]]:sreg_64 = COPY %0
+  ; CHECK:   dead %0.sub1:sreg_64 = COPY %0.sub0
+  ; CHECK:   S_ENDPGM implicit [[COPY]].sub1
+  bb.0:
+    successors: %bb.1
+    undef %0.sub1:sreg_64 = S_MOV_B32 -1
+
+  bb.1:
+    %0.sub0:sreg_64 = S_MOV_B32 0
+    %1:sreg_64 = COPY %0:sreg_64
+    dead %0.sub1:sreg_64 = COPY %0.sub0:sreg_64
+    S_ENDPGM implicit %1.sub1:sreg_64
+
+...
+---
+name: lanes_not_tracked_subreg_join_couldnt_join_subrange
+tracksRegLiveness: true
+body:             |
+  bb.0:
+
+    ; CHECK-LABEL: name: lanes_not_tracked_subreg_join_couldnt_join_subrange
+    ; CHECK: undef %0.sub0:sreg_64_xexec = S_MOV_B32 0
+    ; CHECK: %0.sub1:sreg_64_xexec = S_MOV_B32 0
+    ; CHECK: S_NOP 0, implicit %0.sub1
+    ; CHECK: S_NOP 0, implicit %0
+    ; CHECK: S_ENDPGM
+    undef %0.sub0:sreg_64_xexec = S_MOV_B32 0
+    %1:sreg_64 = COPY %0
+    %0.sub1:sreg_64_xexec = S_MOV_B32 0
+    S_NOP 0, implicit %0.sub1
+    S_NOP 0, implicit %1
+    S_ENDPGM
+
+...
+---
+name: couldnt_join_subrange1
+tracksRegLiveness: true
+body:             |
+  ; CHECK-LABEL: name: couldnt_join_subrange1
+  ; CHECK: bb.0:
+  ; CHECK:   successors: %bb.1(0x80000000)
+  ; CHECK:   undef %0.sub0:sreg_64_xexec = S_MOV_B32 0
+  ; CHECK:   %0.sub1:sreg_64_xexec = COPY %0.sub0
+  ; CHECK: bb.1:
+  ; CHECK:   S_NOP 0, implicit %0.sub1
+  ; CHECK:   S_ENDPGM implicit %0
+  bb.0:
+    successors: %bb.1
+
+    undef %0.sub0:sreg_64_xexec = S_MOV_B32 0
+    %1:sreg_64 = COPY %0
+    %0.sub1:sreg_64_xexec = COPY %0.sub0
+
+  bb.1:
+
+    S_NOP 0, implicit %0.sub1
+    S_ENDPGM implicit %1
+
+...