[PATCH] D128527: [AMDGPU] Use GFX11 S_PACK_HL instruction in more cases

[Joe Nash via Phabricator via llvm-commits]

Joe_Nash added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/SIInstrInfo.cpp:7090
   }
+  case AMDGPU::S_PACK_HL_B32_B16: {
+    Register TmpReg = MRI.createVirtualRegister(&AMDGPU::VGPR_32RegClass);
----------------
I don't think there is any test coverage for this block.


================
Comment at: llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-build-vector-trunc.v2s16.mir:3
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefixes=GFX9PLUS,GFX9 %s
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx1100 -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck -check-prefixes=GFX9PLUS,GFX11 %s
 
----------------
foad wrote:
> @Joe_Nash mentioned offline that it might be better to pre-commit all the added RUN lines.
I think the codegen changes are reviewable by comparing with the gfx10 checks. The main reason to pre-commit tests would be separating NFC from this this change. Up to you.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D128527/new/

https://reviews.llvm.org/D128527