[PATCH] D134433: [AMDGPU][GISel] Enable Matching of V2S16 G_BUILD_VECTOR

[Jay Foad via Phabricator via llvm-commits]

foad added inline comments.


================
Comment at: llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement.i16.ll:9
 define amdgpu_ps i16 @extractelement_sgpr_v4i16_sgpr_idx(<4 x i16> addrspace(4)* inreg %ptr, i32 inreg %idx) {
-; GCN-LABEL: extractelement_sgpr_v4i16_sgpr_idx:
-; GCN:       ; %bb.0:
-; GCN-NEXT:    s_load_dwordx2 s[0:1], s[2:3], 0x0
-; GCN-NEXT:    s_lshr_b32 s2, s4, 1
-; GCN-NEXT:    s_cmp_eq_u32 s2, 1
-; GCN-NEXT:    s_waitcnt lgkmcnt(0)
-; GCN-NEXT:    s_cselect_b32 s0, s1, s0
-; GCN-NEXT:    s_and_b32 s1, s4, 1
-; GCN-NEXT:    s_lshl_b32 s1, s1, 4
-; GCN-NEXT:    s_lshr_b32 s0, s0, s1
-; GCN-NEXT:    ; return to shader part epilog
+; GFX9-LABEL: extractelement_sgpr_v4i16_sgpr_idx:
+; GFX9:       ; %bb.0:
----------------
Using buffer_store/load instead of shifts for non-constant indices is a big regression.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D134433/new/

https://reviews.llvm.org/D134433