The attached test case contains two functions.  Each one shows one way that optimizing the code can make it slower and larger.

The first one has a small structure with type double members.  The code zeros them out ...
 

   ts1.a = 0;
   ts1.b = 0;
   ts1.c = 0;
   ts1.d = 0;

The second one has a very large structure with type float members.  The float members are in two groups, more than a data page apart in the structure.  The code adds members from different pages together ...

   ts2.pageA_a += ts2.pageB_a;
   ts2.pageA_b += ts2.pageB_b;
   ts2.pageA_c += ts2.pageB_c;

Build it with optimization disabled ...

% cl2000 --src_interlist --float_support=fpu32 --opt_level=off testcase.c

Inspect the assembly code file testcase.asm to see that f1 has 8 instructions and f2 has 79 instructions.

Build it with optimization level 2 ...

% cl2000 --src_interlist --float_support=fpu32 --opt_level=2 testcase.c

Note that f1 has 11 instructions and f2 has 109 instructions.