[PATCH] D143787: [X86] Add new pass `X86FixupISel` for fixing up machine-instruction selection.

[Phoebe Wang via Phabricator via llvm-commits]

pengfei added inline comments.


================
Comment at: llvm/lib/Target/X86/X86FixupISel.cpp:27
+
+#define DEBUG_TYPE "x86-fixup-ISel"
+
----------------
We often use all lower case latters.


================
Comment at: llvm/lib/Target/X86/X86FixupISel.cpp:69
+
+  auto processVPERMILPSm = [&](unsigned NewOpc) {
+    if (!ST->hasNoDomainDelayShuffle())
----------------
Lambda function name should starts with upper latter.


================
Comment at: llvm/lib/Target/X86/X86FixupISel.cpp:75
+  };
+  auto processVPERMILPSr = [&](unsigned NewOpc) {
+    unsigned MaskImm = MI.getOperand(NumOperands - 1).getImm();
----------------
ditto.


================
Comment at: llvm/lib/Target/X86/X86FixupISel.cpp:86
+  case X86::VPERMILPSri:
+    return processVPERMILPSr(X86::VSHUFPSrri);
+  case X86::VPERMILPSYri:
----------------
Better to use processVPERMILPSri/processVPERMILPSmi


================
Comment at: llvm/test/CodeGen/X86/opt-pipeline.ll:204
 ; CHECK-NEXT:       X86 LEA Fixup
+; CHECK-NEXT:       X86 Fixup ISel
 ; CHECK-NEXT:       Compressing EVEX instrs to VEX encoding when possible
----------------
Is the pass too far away from `ISel`?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D143787/new/

https://reviews.llvm.org/D143787