[llvm] [AMDGPU] Fix v_dot2_f16_f16/v_dot2_bf16_bf16 operands (PR #82423)

[Matt Arsenault via llvm-commits]

================
@@ -2116,6 +2116,12 @@ v_dot2_bf16_bf16 v5, -src_scc, |vcc_lo|, -1 op_sel:[0,0,1,0]
 v_dot2_bf16_bf16 v255, -|0xfe0b|, -|vcc_hi|, null op_sel:[0,0,0,1]
 // GFX11: encoding: [0xff,0x43,0x67,0xd6,0xff,0xd6,0xf0,0x61,0x0b,0xfe,0x00,0x00]
 
+v_dot2_bf16_bf16 v2, v0, 0x20004000, v2
+// GFX11: v_dot2_bf16_bf16 v2, v0, 0x20004000, v2 ; encoding: [0x02,0x00,0x67,0xd6,0x00,0xff,0x09,0x04,0x00,0x40,0x00,0x20]
+
+v_dot2_bf16_bf16 v2, 0x20004000, v0, v2
----------------
arsenm wrote:

do we need tests with op_sel + constant? 

https://github.com/llvm/llvm-project/pull/82423