[PATCH] D29423: [AMDGPU] Account workgroup size in LDS occupancy limits

Matt Arsenault via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Wed Feb 1 14:10:24 PST 2017

arsenm added inline comments.

Comment at: lib/Target/AMDGPU/AMDGPUSubtarget.cpp:71-74
+  if (LocalMemorySize == 0) {
+    if (getGeneration() == SOUTHERN_ISLANDS) LocalMemorySize = 32768;
+    else if (getGeneration() >= SEA_ISLANDS) LocalMemorySize = 65536;
+  }
This should never happen. These values are already defined by the subtarget feature which set these values. Are you seeing this happen? Is the feature missing for some reason from some processor definitions?

Comment at: test/CodeGen/AMDGPU/large-work-group-promote-alloca.ll:71
 ; CHECK-LABEL: @occupancy_0(
-; CHECK: alloca [5 x i32]
+; CI: alloca [5 x i32]
 define void @occupancy_0(i32 addrspace(1)* nocapture %out, i32 addrspace(1)* nocapture %in) #3 {
Can you add the check for the other case too



More information about the llvm-commits mailing list