[llvm] AMDGPU: Allow only VGPR wide sources in fp6/4/8 conversions (PR #127464)

[Stanislav Mekhanoshin via llvm-commits]

================
@@ -1052,7 +1052,11 @@ class VOP3_CVT_SCALE_SR_PK_F4_F16BF16_TiedInput_Profile<ValueType Src0Ty> :
   let HasFP4DstByteSel = 1;
 }
 
-def VOP3_CVT_SCALE_SR_PK_F4_F32_TiedInput_Profile : VOP3_Profile<VOPProfile<[i32, v2f32, i32, f32]>, VOP3_OPSEL> {
+class VOP3_CVT_SCALE_SR_PK_F4_F32_TiedInput_Profile<VOPProfile P>
+    : VOP3_Profile<P, VOP3_OPSEL> {
+
+  let Src0RC64 = !if(!gt(P.Src0VT.Size, 32), getVOP3VSrcReg9ForVT<P.Src0VT>.ret,
----------------
rampitec wrote:

Description says 'all convert opcodes', so `!if` is not needed.

https://github.com/llvm/llvm-project/pull/127464