[PATCH] D29505: [AMDGPU] Bump -amdgpu-unroll-threshold-private to 2000

Stanislav Mekhanoshin via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Fri Feb 3 11:52:16 PST 2017

rampitec created this revision.
Herald added a reviewer: tstellarAMD.
Herald added subscribers: tpr, tony-tye, yaxunl, nhaehnle, wdng, kzhuravl.

This has quite positive performance impact according to measurements.
Before previous fixes to limit the optimization that was too high
and blowed compile time and scratch usage, but now this is gone and
we can bump the threshold.




Index: lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
--- lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
+++ lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
@@ -32,7 +32,7 @@
 static cl::opt<unsigned> UnrollThresholdPrivate(
   cl::desc("Unroll threshold for AMDGPU if private memory used in a loop"),
-  cl::init(800), cl::Hidden);
+  cl::init(2000), cl::Hidden);
 void AMDGPUTTIImpl::getUnrollingPreferences(Loop *L,
                                             TTI::UnrollingPreferences &UP) {

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D29505.86998.patch
Type: text/x-patch
Size: 607 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170203/c0e1a5cc/attachment.bin>

More information about the llvm-commits mailing list