[llvm] [AMDGPU] Do not use 64-bit literals with VOP3* encodings (PR #156602)

[Jay Foad via llvm-commits]

================
@@ -398,9 +398,9 @@ v_lshlrev_b64 v[4:5], 0xaf123456, v[2:3]
 // GFX1250: v_lshlrev_b64_e32 v[4:5], 0xaf123456, v[2:3] ; encoding: [0xff,0x04,0x08,0x3e,0x56,0x34,0x12,0xaf]
 // GFX9-ERR: :[[@LINE-3]]:23: error: literal operands are not supported
 
-v_lshlrev_b64 v[6:7], v1, 0x3f717273
-// GFX10: v_lshlrev_b64 v[6:7], v1, 0x3f717273    ; encoding: [0x06,0x00,0xff,0xd6,0x01,0xff,0x01,0x00,0x73,0x72,0x71,0x3f]
-// GFX1250: v_lshlrev_b64_e64 v[6:7], v1, 0x3f717273 ; encoding: [0x06,0x00,0x1f,0xd5,0x01,0xff,0x01,0x00,0x73,0x72,0x71,0x3f]
+v_lshlrev_b64 v[6:7], v1, 0xbf717273
----------------
jayfoad wrote:

You can see in the first commit in the PR that it introduces a bogus lit64, which the second commit fixes.

https://github.com/llvm/llvm-project/pull/156602