[PATCH] D100904: [AMDGPU] Lower regbanks reassign threshold to 15000

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

rampitec created this revision.
rampitec added reviewers: msearles, foad.
Herald added subscribers: kerbowa, hiraditya, t-tye, tpr, dstuttard, yaxunl, nhaehnle, jvesely, kzhuravl, arsenm.
rampitec requested review of this revision.
Herald added a subscriber: wdng.
Herald added a project: LLVM.

Let it work on a very small kernels only. Measurements showed
the performance benefit is not worth the compile time.


https://reviews.llvm.org/D100904

Files:
  llvm/lib/Target/AMDGPU/GCNRegBankReassign.cpp


Index: llvm/lib/Target/AMDGPU/GCNRegBankReassign.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/GCNRegBankReassign.cpp
+++ llvm/lib/Target/AMDGPU/GCNRegBankReassign.cpp
@@ -52,7 +52,7 @@
 // Threshold to keep compile time reasonable.
 static cl::opt<unsigned> VRegThresh("amdgpu-regbanks-reassign-threshold",
   cl::desc("Max number of vregs to run the regbanks reassign pass"),
-  cl::init(100000), cl::Hidden);
+  cl::init(15000), cl::Hidden);
 
 #define DEBUG_TYPE "amdgpu-regbanks-reassign"
 


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D100904.339023.patch
Type: text/x-patch
Size: 549 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210420/1dfd1621/attachment.bin>