Alignment of scalar values should be specified

[asb]

The calling convention description is incomplete without specifying the alignment of scalar values. Of course the alignment is usually sizeof(ty), but some 32-bit ABIs allow doubles to be 4-byte aligned.

[palmer-dabbelt]

Thanks for going through the documentation, it's great to make sure the LLVM and GCC stuff match before people start really using our stuff. Looks like we're 64-bit aligned for uint64_t on rv32g. @sorear, can you document it?

palmer gamma02 ~ $ cat doublealign.c 
#include <stdint.h>

struct aligntest {
	uint32_t a;
	uint64_t b;
	uint32_t c;
};

void func(void *);

extern struct aligntest aligntest;

int main() {
	func(&aligntest.a);
	func(&aligntest.b);
	func(&aligntest.c);
	return 0;
};
	
	
palmer gamma02 ~ $ riscv64-unknown-linux-gnu-gcc doublealign.c -march=rv32g -mabi=ilp32 -O3 -S -o -
	.file	"doublealign.c"
	.option nopic
	.section	.text.startup,"ax",@progbits
	.align	2
	.globl	main
	.type	main, @function
main:
	lui	a0,%hi(aligntest)
	add	sp,sp,-16
	add	a0,a0,%lo(aligntest)
	sw	ra,12(sp)
	call	func
	lui	a0,%hi(aligntest+8)
	add	a0,a0,%lo(aligntest+8)
	call	func
	lui	a0,%hi(aligntest+16)
	add	a0,a0,%lo(aligntest+16)
	call	func
	lw	ra,12(sp)
	li	a0,0
	add	sp,sp,16
	jr	ra
	.size	main, .-main
	.ident	"GCC: (GNU) 6.1.0"

[asb]

Thanks Palmer, it was actually double fp I was referring to, though I would guess these are 64-bit aligned on RV32 as well (can't ssh into my machine with riscv-gcc to verify right now).

[palmer-dabbelt]

doubles would definitely want to be 64-bit aligned, since there's a fld instruction even on RV32D.