[PATCH] D101292: [AMDGPU] Restrict immediate scratch offsets

[Jay Foad via Phabricator via llvm-commits]

foad accepted this revision.
foad added a comment.
This revision is now accepted and ready to land.

LGTM, just nits inline.



================
Comment at: llvm/lib/Target/AMDGPU/AMDGPU.td:175
 
+def FeatureNegativeScratchOffsetBug : SubtargetFeature<"negative-scratch-offset-bug",
+  "NegativeScratchOffsetBug",
----------------
Nit: maybe define these next to FlatSegmentOffsetBug since they're closely related?


================
Comment at: llvm/lib/Target/AMDGPU/SIInstrInfo.cpp:7373-7374
+// are allowed and some are not.
+// In general, flat instructions can only be unsigned, global and scratch
+// instructions can also be negative.
+//
----------------
Nit: "flat instruction offsets can only be non-negative, global and scratch instruction offsets can ..."


================
Comment at: llvm/test/CodeGen/AMDGPU/local-stack-alloc-block-sp-reference.ll:305
 ; FLATSCR-NEXT:    s_waitcnt vmcnt(0)
-; FLATSCR-NEXT:    scratch_load_dwordx4 v[4:7], off, s2 offset:-16 glc
+; FLATSCR-NEXT:    s_movk_i32 s2, 0x2000
+; FLATSCR-NEXT:    scratch_load_dwordx4 v[4:7], off, s2 glc
----------------
Seems a shame that this wasn't CSE'd away.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D101292/new/

https://reviews.llvm.org/D101292