[PATCH] D106768: [AArch4][GlobalISel] Post-legalize combine s64 = G_MERGE s32, 0 -> G_ZEXT.

[Amara Emerson via Phabricator via llvm-commits]

This revision was landed with ongoing or failed builds.
This revision was automatically updated to reflect the committed changes.
Closed by commit rG6af8d360546e: [AArch4][GlobalISel] Post-legalize combine s64 = G_MERGE s32, 0 -> G_ZEXT. (authored by aemerson).

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D106768/new/

https://reviews.llvm.org/D106768

Files:
  llvm/lib/Target/AArch64/AArch64Combine.td
  llvm/lib/Target/AArch64/GISel/AArch64PostLegalizerCombiner.cpp
  llvm/test/CodeGen/AArch64/GlobalISel/postlegalizer-combiner-merge.mir


Index: llvm/test/CodeGen/AArch64/GlobalISel/postlegalizer-combiner-merge.mir
===================================================================
--- llvm/test/CodeGen/AArch64/GlobalISel/postlegalizer-combiner-merge.mir
+++ llvm/test/CodeGen/AArch64/GlobalISel/postlegalizer-combiner-merge.mir
@@ -68,3 +68,25 @@
     RET_ReallyLR implicit $x0
 
 ...
+---
+name:            merge_to_zext
+alignment:       4
+legalized:       true
+liveins:
+  - { reg: '$w0' }
+body:             |
+  bb.1.entry:
+    liveins: $w0
+
+    ; CHECK-LABEL: name: merge_to_zext
+    ; CHECK: %v:_(s32) = COPY $w0
+    ; CHECK: %merge:_(s64) = G_ZEXT %v(s32)
+    ; CHECK: $x0 = COPY %merge(s64)
+    ; CHECK: RET_ReallyLR implicit $x0
+    %v:_(s32) = COPY $w0
+    %zero:_(s32) = G_CONSTANT i32 0
+    %merge:_(s64) = G_MERGE_VALUES %v, %zero
+    $x0 = COPY %merge(s64)
+    RET_ReallyLR implicit $x0
+
+...
Index: llvm/lib/Target/AArch64/GISel/AArch64PostLegalizerCombiner.cpp
===================================================================
--- llvm/lib/Target/AArch64/GISel/AArch64PostLegalizerCombiner.cpp
+++ llvm/lib/Target/AArch64/GISel/AArch64PostLegalizerCombiner.cpp
@@ -23,6 +23,7 @@
 #include "llvm/CodeGen/GlobalISel/Combiner.h"
 #include "llvm/CodeGen/GlobalISel/CombinerHelper.h"
 #include "llvm/CodeGen/GlobalISel/CombinerInfo.h"
+#include "llvm/CodeGen/GlobalISel/GISelChangeObserver.h"
 #include "llvm/CodeGen/GlobalISel/GISelKnownBits.h"
 #include "llvm/CodeGen/GlobalISel/MIPatternMatch.h"
 #include "llvm/CodeGen/GlobalISel/MachineIRBuilder.h"
@@ -240,6 +241,27 @@
   return true;
 }
 
+/// Try to fold a G_MERGE_VALUES of 2 s32 sources, where the second source
+/// is a zero, into a G_ZEXT of the first.
+bool matchFoldMergeToZext(MachineInstr &MI, MachineRegisterInfo &MRI) {
+  auto &Merge = cast<GMerge>(MI);
+  LLT SrcTy = MRI.getType(Merge.getSourceReg(0));
+  if (SrcTy != LLT::scalar(32) || Merge.getNumSources() != 2)
+    return false;
+  return mi_match(Merge.getSourceReg(1), MRI, m_SpecificICst(0));
+}
+
+void applyFoldMergeToZext(MachineInstr &MI, MachineRegisterInfo &MRI,
+                          MachineIRBuilder &B, GISelChangeObserver &Observer) {
+  // Mutate %d(s64) = G_MERGE_VALUES %a(s32), 0(s32)
+  //  ->
+  // %d(s64) = G_ZEXT %a(s32)
+  Observer.changingInstr(MI);
+  MI.setDesc(B.getTII().get(TargetOpcode::G_ZEXT));
+  MI.RemoveOperand(2);
+  Observer.changedInstr(MI);
+}
+
 #define AARCH64POSTLEGALIZERCOMBINERHELPER_GENCOMBINERHELPER_DEPS
 #include "AArch64GenPostLegalizeGICombiner.inc"
 #undef AARCH64POSTLEGALIZERCOMBINERHELPER_GENCOMBINERHELPER_DEPS
Index: llvm/lib/Target/AArch64/AArch64Combine.td
===================================================================
--- llvm/lib/Target/AArch64/AArch64Combine.td
+++ llvm/lib/Target/AArch64/AArch64Combine.td
@@ -182,6 +182,13 @@
   (apply [{ applyFormTruncstore(*${root}, MRI, B, Observer, ${matchinfo}); }])
 >;
 
+def fold_merge_to_zext : GICombineRule<
+  (defs root:$d),
+  (match (wip_match_opcode G_MERGE_VALUES):$d,
+          [{ return matchFoldMergeToZext(*${d}, MRI); }]),
+  (apply [{ applyFoldMergeToZext(*${d}, MRI, B, Observer); }])
+>;
+
 // Post-legalization combines which should happen at all optimization levels.
 // (E.g. ones that facilitate matching for the selector) For example, matching
 // pseudos.
@@ -204,6 +211,6 @@
                         mul_const, redundant_sext_inreg,
                         form_bitfield_extract, rotate_out_of_range,
                         icmp_to_true_false_known_bits, merge_unmerge,
-                        select_combines]> {
+                        select_combines, fold_merge_to_zext]> {
   let DisableRuleOption = "aarch64postlegalizercombiner-disable-rule";
 }


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D106768.361726.patch
Type: text/x-patch
Size: 3724 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210726/7c436e3a/attachment.bin>