That's not possible using the instruction with a constant offset. To generate the code, the offset would need to be known at compile time, and not be variable. You have to use a different instruction, an indirect load with base register and offset:
int foo(int64_t offset, double value){ asm volatile ("movsd %0, (%%rsp,%1)" :: "x" (value), "r" (offset) : "memory" );}You could also let the CPU do the multiplication by 8 by using a scaled offset addressing mode:
int foo(int64_t offset, double value){ asm volatile ("movsd %0, (%%rsp,%1,8)" :: "x" (value), "r" (offset) : "memory" );}Or if you want to emulate push, then sub $8, %%rsp / movsd %0, (%%rsp), but you can't mess with the stack pointer from inline asm without breaking compiler-generated code.
