[PATCH] D63851: [AMDGPU] Packed thread ids in function call ABI

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

rampitec marked an inline comment as done.
rampitec added inline comments.


================
Comment at: lib/Target/AMDGPU/SIISelLowering.cpp:1643
                                       SIMachineFunctionInfo &Info) {
-  if (Info.hasWorkItemIDX())
-    Info.setWorkItemIDX(allocateVGPR32Input(CCInfo));
+  const unsigned Mask = 0x3ff;
+  ArgDescriptor Arg;
----------------
rampitec wrote:
> arsenm wrote:
> > This should be derived from the subtarget maximum workgroup size
> It is derived from the theoretical maximum you can pack into 32 bit register (and agrees with maximum whg size as well of course, which makes it possible).
For the record: there is conformation that any given WG dimension can never be more that 1024. This enables this code and allows to add few more AssertZExt I suppose. Probably there is something in clang as well.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D63851/new/

https://reviews.llvm.org/D63851