From: Nicolai Haehnle <nhaehnle@gmail.com>
Date: Tue, 28 Nov 2017 08:42:46 +0000 (+0000)
Subject: AMDGPU: Re-organize the outer loop of SILoadStoreOptimizer
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=f62da810dff0ea2bb0bb2d478650a0e719371cbc;p=llvm

AMDGPU: Re-organize the outer loop of SILoadStoreOptimizer

Summary:
The entire algorithm operates per basic-block, so for cache locality
it should be better to re-optimize a basic-block immediately rather than
in a separate loop.

I don't have performance measurements.

Change-Id: I85106570bd623c4ff277faaa50ee43258e1ddcc5

Reviewers: arsenm, rampitec

Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye

Differential Revision: https://reviews.llvm.org/D40344

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@319156 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp b/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
index 14c9c8ff728..48bfc2dac2d 100644
--- a/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
+++ b/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp
@@ -14,7 +14,7 @@
 // ==>
 //   ds_read2_b32 v[0:1], v2, offset0:4 offset1:8
 //
-// The same is done for certain SMEM opcodes, e.g.:
+// The same is done for certain SMEM and VMEM opcodes, e.g.:
 //  s_buffer_load_dword s4, s[0:3], 4
 //  s_buffer_load_dword s5, s[0:3], 8
 // ==>
@@ -892,14 +892,13 @@ bool SILoadStoreOptimizer::runOnMachineFunction(MachineFunction &MF) {
   DEBUG(dbgs() << "Running SILoadStoreOptimizer\n");
 
   bool Modified = false;
-  CreatedX2 = 0;
 
-  for (MachineBasicBlock &MBB : MF)
+  for (MachineBasicBlock &MBB : MF) {
+    CreatedX2 = 0;
     Modified |= optimizeBlock(MBB);
 
-  // Run again to convert x2 to x4.
-  if (CreatedX2 >= 1) {
-    for (MachineBasicBlock &MBB : MF)
+    // Run again to convert x2 to x4.
+    if (CreatedX2 >= 1)
       Modified |= optimizeBlock(MBB);
   }