Best to describe it using an example. Here is a copy of the post I've put in D forum:

I made the following example to see how far can inline asm blocks use template args: (Note: This bugzilla is about statement #1 only. For #4, I'll open another one)
         ========================================
         void asm_template(T)(T Tval,T* pDest)
         {
                 asm{
                         fld Tval;
                         mov EAX,pDest;
                         fstp T ptr[EAX];        //#1
                         fstp float ptr[EAX];    //#2
                         mov EDX,T.sizeof;       //#3
                         mov ECX,float.sizeof;   //#4
                 }//asm
         }//asm_template
                 
         void main()
         {
           float f1=4.0;
           float f2=5.0;
         
           asm_template!(float)(f1,&f2);
         }
         ========================================
 Here is the compilation result of the 4 statements marked above:
 #1: ERROR: "cannot use type float as an operand"
 #2: OK
 #3: OK!!!
 #4: ERROR: "ptr expected"

For me, if #1 would be illegal, then the error should not have mentioned "type float". In other words, why would it fail if T is correctly interpreted to float? In conclusion, it would make sense if such statement would be accepted, thus enabling writing general ASM code.