[PATCH] D105709: [AMDGPU][GlobalISel] Insert an and with exec before s_cbranch_vccnz if necessary

[Mirko Brkusanin via Phabricator via llvm-commits]

mbrkusanin added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp:2496
+    const unsigned CondDefOpc = MRI->getUniqueVRegDef(CondReg)->getOpcode();
+    if (CondDefOpc != AMDGPU::G_ICMP && CondDefOpc != AMDGPU::G_FCMP) {
+      const bool Is64 = STI.isWave64();
----------------
arsenm wrote:
> foad wrote:
> > arsenm wrote:
> > > This isn't the precise condition. It should also cover class intrinsics and maybe a few other cases. I thought we had an is known vector bool helper somewhere?
> > > 
> > For SelectionDAG we have isBoolSGPR. I don't know of anything similar for GlobalISel.
> We should have the equivalent
I could not find a global-isel equivalent.


================
Comment at: llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-brcond.mir:231-233
+    %2:sgpr(s64) = G_INTRINSIC intrinsic(@llvm.amdgcn.icmp), %0:vgpr(s64), %1:vgpr(s64), 32
+    %3:sgpr(s1) = G_TRUNC %2
+    %4:vcc(s1) = COPY %3
----------------
Is this correct? As far as I can tell this is the only way to copy sgpr to vcc.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D105709/new/

https://reviews.llvm.org/D105709