[llvm] Add mad support for v_pk_* 16 bit integer (PR #95104)

[Jay Foad via llvm-commits]

================
@@ -2999,16 +2993,15 @@ define <2 x i16> @v_mul_284_add_82_v2i16(<2 x i16> %arg) {
 ; GFX9:       ; %bb.0:
 ; GFX9-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GFX9-NEXT:    s_movk_i32 s4, 0x11c
-; GFX9-NEXT:    v_pk_mul_lo_u16 v0, v0, s4 op_sel_hi:[1,0]
-; GFX9-NEXT:    s_movk_i32 s4, 0x52
-; GFX9-NEXT:    v_pk_add_u16 v0, v0, s4 op_sel_hi:[1,0]
+; GFX9-NEXT:    v_mov_b32_e32 v1, 0x52
+; GFX9-NEXT:    v_pk_mad_u16 v0, v0, s4, v1 op_sel_hi:[1,0,0]
 ; GFX9-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GFX10-LABEL: v_mul_284_add_82_v2i16:
 ; GFX10:       ; %bb.0:
 ; GFX10-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX10-NEXT:    v_pk_mul_lo_u16 v0, 0x11c, v0 op_sel_hi:[0,1]
-; GFX10-NEXT:    v_pk_add_u16 v0, 0x52, v0 op_sel_hi:[0,1]
+; GFX10-NEXT:    s_movk_i32 s4, 0x11c
+; GFX10-NEXT:    v_pk_mad_u16 v0, v0, s4, 0x52 op_sel_hi:[1,0,0]
----------------
jayfoad wrote:

You can't encode two different literals in one instruction. 0x52 and 0x11c are both too big to be inline constants.

https://github.com/llvm/llvm-project/pull/95104