Examine the code generated by GCC for functions that have st

Examine the code generated by GCC for functions that have structures as arguments and return values, and from this see how these language features are typically implemented. The following C code has a function word_sum having structures as argument and return values, and a function prod that calls process: typedef struct {long a[2]; long *p;} strA; typedef struct {long u[2]; long q;} strB; strB process(strA s) {strB r.u[0] = s.a[1]; r.u[1] = s.a[0]; r.q = *s.p; return r; long eval (long x, long y, long z) {strA s; s.a[0] = x; s.a[1] = y; s.p = &z; strB r = process(s); return r.u[0] + r.u[1] + r.q;} GCCgenerates the following code for these two functions: strB process(strA s) process: movq %rdi, %rax movq 24(%rsp), %rdx movq (%rdx), %rdx movq 16(%rsp), %rcx movq %rcx, (%rdi) movq 8(%rsp), %rcx movq %rcx, 8(%rdi) movq %rdx, 16(%rdi) ret long eval (long x, long y, long z) x in %rdi, y in %rsi, z in %rdx eval: subq $104, %rsp movq %rdx, 24(%rsp) leaq 24(%rsp), %rax movq %rdi, (%rsp) movq %rsi, 8(%rsp) movq %rax, 16(%rsp) leaq 64(%rsp), %rdi call process movq 72(%rsp), %rax addq 64(%rsp), %rax addq 80(%rsp), %rax addq $104, %rsp ret We can see on line 2 of function eval that it allocates 104 bytes on the stack. Diagram the stack frame for eval, showing the values that it stores on the stack prior to calling process. What value does eval pass in its call to process? How does the code for process access the elements of structure argument s? How does the code for process set the fields of result structure r? Complete your diagram of the stack frame for eval, showing how eval accesses the elements of structure r following the return from process. What general principles can you discern about how structure values are passed as function arguments and how they are returned as function results?

Solution

eval creates a new variable of type strA using its input variables x,y and z.

It then passes this newly created structure to function process

How it actually happens when converted to assembly code ? Well eval knows size of strA ( as compiler knows that strA has two variables of long type, a[0],a[1] and one pointer p, each of 8 bytes ) is 24bytes. Thus it changes stack pointer by 24bytes before calling process, as seen in instruction 4 of eval\'s assembly code. Then it sets values at offset of 8 bytes ( i.e. first at 0, then at 8 and finally at 16 bytes from stack pointer ), and these values are what structure elements should be. Finally it calls process, and process navigates using its stack pointer, gets the value and does the job.