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

[Stanislav Mekhanoshin via llvm-commits]

================
@@ -1833,6 +1839,12 @@
 # GFX11: v_dot2_f16_f16 v255, -|0xfe0b|, -|vcc_hi|, null op_sel:[0,0,0,1] ; encoding: [0xff,0x43,0x66,0xd6,0xff,0xd6,0xf0,0x61,0x0b,0xfe,0x00,0x00]
 0xff,0x43,0x66,0xd6,0xff,0xd6,0xf0,0x61,0x0b,0xfe,0x00,0x00
 
+# GFX11: v_dot2_f16_f16 v2, v0, 0x20004000, v2 ; encoding: [0x02,0x00,0x66,0xd6,0x00,0xff,0x09,0x04,0x00,0x40,0x00,0x20]
+0x02,0x00,0x66,0xd6,0x00,0xff,0x09,0x04,0x00,0x40,0x00,0x20
+
+# GFX11: v_dot2_f16_f16 v2, 0x20004000, v0, v2 ; encoding: [0x02,0x00,0x66,0xd6,0xff,0x00,0x0a,0x04,0x00,0x40,0x00,0x20]
+0x02,0x00,0x66,0xd6,0xff,0x00,0x0a,0x04,0x00,0x40,0x00,0x20
----------------
rampitec wrote:

I.e. there is no difference if that is 0x3f803f80 or 0x20004000. This is not an inline constant that way or another.

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