[llvm] [AMDGPU][True16] remove pack32 pattern from true16 mode (PR #171756)

[Shilei Tian via llvm-commits]

================
@@ -1848,10 +1848,9 @@ define <2 x half> @fmul_select_v2f16_test3(<2 x half> %x, <2 x i32> %bool.arg1,
 ; GFX11-SDAG-TRUE16-NEXT:    v_mov_b16_e32 v2.l, 0x3c00
 ; GFX11-SDAG-TRUE16-NEXT:    v_cmp_eq_u32_e64 s0, v1, v3
 ; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
-; GFX11-SDAG-TRUE16-NEXT:    v_cndmask_b16 v1.l, v2.l, 0x4000, vcc_lo
-; GFX11-SDAG-TRUE16-NEXT:    v_cndmask_b16 v1.h, v2.l, 0x4000, s0
-; GFX11-SDAG-TRUE16-NEXT:    s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-SDAG-TRUE16-NEXT:    v_pack_b32_f16 v1, v1.h, v1.l
----------------
shiltian wrote:

I vaguely remember someone mentioning this unnecessary packing a couple of months ago. Glad to see it's gone.

https://github.com/llvm/llvm-project/pull/171756