[PATCH] D125261: [AMDGPU] gfx11 subtarget features & early tests

[Petar Avramovic via Phabricator via llvm-commits]

Petar.Avramovic added a comment.

Can you also add comment that test changes for some memory instructions are caused by subtarget feature that allows offset folding (to be precise not having feature that forbids offset folding). Maybe separate these tests to another patch.



================
Comment at: llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-load-flat.mir:935-956
     ; GFX10-LABEL: name: load_flat_s32_from_1_gep_2047
     ; GFX10: liveins: $vgpr0_vgpr1
     ; GFX10-NEXT: {{  $}}
     ; GFX10-NEXT: [[COPY:%[0-9]+]]:vreg_64 = COPY $vgpr0_vgpr1
     ; GFX10-NEXT: [[V_MOV_B32_e32_:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 2047, implicit $exec
     ; GFX10-NEXT: [[V_MOV_B32_e32_1:%[0-9]+]]:vgpr_32 = V_MOV_B32_e32 0, implicit $exec
     ; GFX10-NEXT: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[V_MOV_B32_e32_]], %subreg.sub0, [[V_MOV_B32_e32_1]], %subreg.sub1
----------------
Petar.Avramovic wrote:
> This is interesting test change, which patch allows offset fold for this?
GFX11 does not have subtarget feature that stops folding of offsets for some address spaces (here hasFlatSegmentOffsetBug()).


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D125261/new/

https://reviews.llvm.org/D125261