As I read, Think C doesn't has a full assembler compiler. It is only supports an inline "asm" inserts in C code.
I mean, as
@Crutch says, you can write whole functions in an ASM block, at which point the only "C" being compiled will result in the function prologues and epilogues. If you want to call your ASM from the C, you will have to write this code yourself
anyway so you may as well get the compiler to do it. Even on routine boundaries in the assembler, you'll either need to use prologues/epilogues a la C functions or you'll need to come up with your own calling convention. Do not fetishise assembler; there is nothing inherently good about it, especially for the boring code at the beginnings and ends of functions.
As someone who is coming to the end of a hybrid C/68k ASM project where I couldn't mingle C and ASM (because of the limitations of the compiler suite; see the GEMDOS thread in the Lisa subforum), I can assure you that being able to delegate that kind of thing to the C compiler is a
good thing. Having whole files of assembly routines looks cool but is an invitation to bugs.
Also: this gives you the ability to draft out your algorithms in C, and when they are correct but too slow, you can then translate them into ASM and optimise them by hand.