[PATCH] D142262: [AMDGPU] Use std::clamp to avoid overflow in the scheduler

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

rampitec created this revision.
rampitec added a reviewer: kerbowa.
Herald added subscribers: kosarev, foad, hiraditya, tpr, dstuttard, yaxunl, jvesely, kzhuravl, arsenm.
Herald added a project: All.
rampitec requested review of this revision.
Herald added a subscriber: wdng.
Herald added a project: LLVM.

Using std::clamp() from c++17 is a more consistent way to avoid
overflow when calculating SGPR and VGPR pressure limits.


https://reviews.llvm.org/D142262

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


Index: llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp
+++ llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp
@@ -90,17 +90,17 @@
 
   // Subtract error margin and bias from register limits and avoid overflow.
   SGPRCriticalLimit =
-      std::min(SGPRCriticalLimit - SGPRLimitBias, SGPRCriticalLimit);
-  SGPRCriticalLimit =
-      std::min(SGPRCriticalLimit - ErrorMargin, SGPRCriticalLimit);
-  VGPRCriticalLimit =
-      std::min(VGPRCriticalLimit - VGPRLimitBias, VGPRCriticalLimit);
+      SGPRCriticalLimit -
+      std::clamp(SGPRLimitBias + ErrorMargin, 0u, SGPRCriticalLimit);
   VGPRCriticalLimit =
-      std::min(VGPRCriticalLimit - ErrorMargin, VGPRCriticalLimit);
-  SGPRExcessLimit = std::min(SGPRExcessLimit - SGPRLimitBias, SGPRExcessLimit);
-  SGPRExcessLimit = std::min(SGPRExcessLimit - ErrorMargin, SGPRExcessLimit);
-  VGPRExcessLimit = std::min(VGPRExcessLimit - VGPRLimitBias, VGPRExcessLimit);
-  VGPRExcessLimit = std::min(VGPRExcessLimit - ErrorMargin, VGPRExcessLimit);
+      VGPRCriticalLimit -
+      std::clamp(VGPRLimitBias + ErrorMargin, 0u, VGPRCriticalLimit);
+  SGPRExcessLimit =
+      SGPRExcessLimit -
+      std::clamp(SGPRLimitBias + ErrorMargin, 0u, SGPRExcessLimit);
+  VGPRExcessLimit =
+      VGPRExcessLimit -
+      std::clamp(VGPRLimitBias + ErrorMargin, 0u, VGPRExcessLimit);
 
   LLVM_DEBUG(dbgs() << "VGPRCriticalLimit = " << VGPRCriticalLimit
                     << ", VGPRExcessLimit = " << VGPRExcessLimit


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D142262.490988.patch
Type: text/x-patch
Size: 1569 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230120/7190d8a0/attachment.bin>