[PATCH] D11883: AMDGPU/SI: Better handle s_wait insertion

[Axel Davy via llvm-commits]

axeldavy added inline comments.

================
Comment at: test/CodeGen/AMDGPU/wait.ll:7-8
@@ -6,3 +6,4 @@
 ; CHECK: s_load_dwordx4
-; CHECK: s_waitcnt vmcnt(0) lgkmcnt(0){{$}}
+; CHECK: s_waitcnt vmcnt(0)
+; CHECK: s_waitcnt lgkmcnt(0)
 ; CHECK: s_endpgm
----------------
tstellarAMD wrote:
> I would add a check for the instruction between these two s_waitcnt instructions.  We want  to make sure some future change doesn't regress us, and cause us to emit two s_waitcnt instructions in a row.
OK sure

================
Comment at: test/CodeGen/AMDGPU/wait2.ll:1
@@ +1,2 @@
+; RUN: llc -march=amdgcn --misched=ilpmax -mcpu=SI -verify-machineinstrs < %s | FileCheck -strict-whitespace %s
+; RUN: llc -march=amdgcn --misched=ilpmax -mcpu=tonga -verify-machineinstrs < %s | FileCheck -strict-whitespace %s
----------------
tstellarAMD wrote:
> Is there are reason why this test was added to a new file and not to wait.ll ?
yes, the run command is different (and uses the ilpmax scheduler)


http://reviews.llvm.org/D11883