[llvm] [AMDGPU] Update minmax3-tree-reduction.ll for true16. NFC (PR #191879)

[Stanislav Mekhanoshin via llvm-commits]

================
@@ -507,22 +518,49 @@ define bfloat @v_no_max3_maxnum_tree4_bf16(bfloat %a, bfloat %b, bfloat %c, bflo
 ; GFX9-NEXT:    v_lshrrev_b32_e32 v0, 16, v0
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
-; GFX1250-LABEL: v_no_max3_maxnum_tree4_bf16:
-; GFX1250:       ; %bb.0:
-; GFX1250-NEXT:    s_wait_loadcnt_dscnt 0x0
-; GFX1250-NEXT:    s_wait_kmcnt 0x0
-; GFX1250-NEXT:    v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v3, 16, v3
-; GFX1250-NEXT:    v_dual_lshlrev_b32 v2, 16, v2 :: v_dual_lshlrev_b32 v0, 16, v0
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-NEXT:    v_dual_max_num_f32 v2, v2, v3 :: v_dual_max_num_f32 v0, v0, v1
-; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v1, v2, s0
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
-; GFX1250-NEXT:    v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0
-; GFX1250-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX1250-NEXT:    v_max_num_f32_e32 v0, v0, v1
-; GFX1250-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
-; GFX1250-NEXT:    s_set_pc_i64 s[30:31]
+; GFX1250-FAKE16-LABEL: v_no_max3_maxnum_tree4_bf16:
+; GFX1250-FAKE16:       ; %bb.0:
+; GFX1250-FAKE16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-FAKE16-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-FAKE16-NEXT:    v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v3, 16, v3
+; GFX1250-FAKE16-NEXT:    v_dual_lshlrev_b32 v2, 16, v2 :: v_dual_lshlrev_b32 v0, 16, v0
+; GFX1250-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT:    v_dual_max_num_f32 v2, v2, v3 :: v_dual_max_num_f32 v0, v0, v1
+; GFX1250-FAKE16-NEXT:    v_cvt_pk_bf16_f32 v1, v2, s0
+; GFX1250-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
+; GFX1250-FAKE16-NEXT:    v_dual_lshlrev_b32 v1, 16, v1 :: v_dual_lshlrev_b32 v0, 16, v0
+; GFX1250-FAKE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-FAKE16-NEXT:    v_max_num_f32_e32 v0, v0, v1
+; GFX1250-FAKE16-NEXT:    v_cvt_pk_bf16_f32 v0, v0, s0
+; GFX1250-FAKE16-NEXT:    s_set_pc_i64 s[30:31]
+;
+; GFX1250-REAL16-LABEL: v_no_max3_maxnum_tree4_bf16:
+; GFX1250-REAL16:       ; %bb.0:
+; GFX1250-REAL16-NEXT:    s_wait_loadcnt_dscnt 0x0
+; GFX1250-REAL16-NEXT:    s_wait_kmcnt 0x0
+; GFX1250-REAL16-NEXT:    v_mov_b16_e32 v4.l, 0
+; GFX1250-REAL16-NEXT:    v_mov_b16_e32 v4.h, v1.l
----------------
rampitec wrote:

We may need to find a way to use `lshl` here if we are updating both halves, or maybe some other form of pack instruction. `lshl` is good though because it can become dual.

https://github.com/llvm/llvm-project/pull/191879