[llvm] [AMDGPU][MC] Add dpp for V_PK_FMAC_F16 for GFX10 (PR #79598)

[Joe Nash via llvm-commits]

================
@@ -291,4 +291,4 @@ v_pk_add_u16 v5, v1, 123456.0
 // FIXME: v_pk_fmac_f16 cannot be promoted to VOP3 so '_e32' suffix is not valid
 v_pk_fmac_f16 v5, 0x12345678, v2
 // NOGFX9: :[[@LINE-1]]:{{[0-9]+}}: error: instruction not supported on this GPU
-// GFX10: v_pk_fmac_f16 v5, 0x12345678, v2 ; encoding: [0xff,0x04,0x0a,0x78,0x78,0x56,0x34,0x12]
+// GFX10: v_pk_fmac_f16_e32 v5, 0x12345678, v2 ; encoding: [0xff,0x04,0x0a,0x78,0x78,0x56,0x34,0x12]
----------------
Sisyph wrote:

If I have the convention right, as stated in this patch 0f5ebbcc7fc38f587ffd0b84da4693a8625c1ccb, instructions without a VOP3 form should not have _e32.  It looks like removing IsSingle caused _e32 to be added to the mnemonic. Please change it back.

https://github.com/llvm/llvm-project/pull/79598