[llvm] [AMDGPU] Speedup SIFormMemoryClauses live-in register set calculation. (PR #89919)

[Valery Pykhtin via llvm-commits]

https://github.com/vpykhtin created https://github.com/llvm/llvm-project/pull/89919

GCNDownwardRPTracker requires live-in set for a start of every memory clause. The live-in set calculation is expensive calculation and instead of calculating a set for one clause at time the patch calculates all sets for all clauses at once. This is faster because start slot indexes for clauses are ordered and LIS live segments for a register are also ordered so the getLiveRegMap finds a range of start slot indexes that fall into a given live segment of a register.

This patch uses the same approach used in GCNScheduleDAGMILive::getBBLiveInMap

>From 5129cbcc1982cb448992591a7be42c33395d64ad Mon Sep 17 00:00:00 2001
From: Valery Pykhtin <valery.pykhtin at gmail.com>
Date: Tue, 23 Apr 2024 17:24:50 +0200
Subject: [PATCH] [AMDGPU] Speedup SIFormMemoryClauses live-in register set
 calculation

This patch uses the same approach used in GCNScheduleDAGMILive::getBBLiveInMap
---
 .../lib/Target/AMDGPU/SIFormMemoryClauses.cpp | 34 ++++++++++++++-----
 1 file changed, 26 insertions(+), 8 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/SIFormMemoryClauses.cpp b/llvm/lib/Target/AMDGPU/SIFormMemoryClauses.cpp
index edcfd994033e79..b496c8d449edf2 100644
--- a/llvm/lib/Target/AMDGPU/SIFormMemoryClauses.cpp
+++ b/llvm/lib/Target/AMDGPU/SIFormMemoryClauses.cpp
@@ -275,10 +275,29 @@ bool SIFormMemoryClauses::runOnMachineFunction(MachineFunction &MF) {
   unsigned FuncMaxClause = MF.getFunction().getFnAttributeAsParsedInteger(
       "amdgpu-max-memory-clause", MaxClause);
 
-  for (MachineBasicBlock &MBB : MF) {
-    GCNDownwardRPTracker RPT(*LIS);
+  SmallVector<MachineInstr *, 16> FirstBBClauseMI;
+  for (auto &MBB : MF) {
+    for (auto &MI : MBB) {
+      if (!MI.isMetaInstruction() &&
+          isValidClauseInst(MI, isVMEMClauseInst(MI))) {
+        FirstBBClauseMI.push_back(&MI);
+        break;
+      }
+    }
+  }
+  if (FirstBBClauseMI.empty())
+    return false;
+
+  auto LRM = getLiveRegMap(FirstBBClauseMI, false /*After*/, *LIS);
+
+  GCNDownwardRPTracker RPT(*LIS);
+  for (auto *FirstMI : FirstBBClauseMI) {
+    auto &MBB = *FirstMI->getParent();
+    RPT.reset(*FirstMI, &LRM[FirstMI]);
     MachineBasicBlock::instr_iterator Next;
-    for (auto I = MBB.instr_begin(), E = MBB.instr_end(); I != E; I = Next) {
+    for (auto I = MachineBasicBlock::instr_iterator(FirstMI),
+              E = MBB.instr_end();
+         I != E; I = Next) {
       MachineInstr &MI = *I;
       Next = std::next(I);
 
@@ -287,12 +306,11 @@ bool SIFormMemoryClauses::runOnMachineFunction(MachineFunction &MF) {
 
       bool IsVMEM = isVMEMClauseInst(MI);
 
-      if (!isValidClauseInst(MI, IsVMEM))
-        continue;
+      if (&MI != FirstMI) {
+        if (!isValidClauseInst(MI, IsVMEM))
+          continue;
 
-      if (!RPT.getNext().isValid())
-        RPT.reset(MI);
-      else { // Advance the state to the current MI.
+        // Advance the state to the current MI.
         RPT.advance(MachineBasicBlock::const_iterator(MI));
         RPT.advanceBeforeNext();
       }