d0p1/tcc-stupidos
1
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Faster load/store arm64

Browse source 
The load/store code is optimized to make better use of the offsets
present in the load/store instructions.
Also use GOT reloc's instead of ABS64 relocs.

arm64-gen.c/arm64_check_offset:
- New function to split offset used by load/store and by arm64_sym.

arm64-gen.c/arm64_sym:
- Use GOT reloc's instead of ABS64 relocs.

arm64-gen.c/load arm64-gen.c/store:
- Use new arm64_check_offset function.

arm64-gen.c/gen_bounds_prolog arm64-gen.c/gen_bounds_epilog:
- Use GOT reloc's instaed of ABS64 relocs.
This commit is contained in:
herman ten brugge 2020-07-30 09:26:20 +02:00
parent 58be11f701
commit d1ce34448f
1 changed files with 71 additions and 34 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

105
arm64-gen.c
View file

@ -283,6 +283,23 @@ static void arm64_spoff(int reg, uint64_t off)
} }
} }
/* invert 0: return value to use for store/load */
/* invert 1: return value to use for arm64_sym */
static uint64_t arm64_check_offset(int invert, int sz_, uint64_t off)
{
uint32_t sz = sz_;
if (!(off & ~((uint32_t)0xfff << sz)) ||
(off < 256 || -off <= 256))
return invert ? off : 0ul;
else if ((off & ((uint32_t)0xfff << sz)))
return invert ? off & ((uint32_t)0xfff << sz)
: off & ~((uint32_t)0xfff << sz);
else if (off & 0x1ff)
return invert ? off & 0x1ff : off & ~0x1ff;
else
return invert ? 0ul : off;
}
static void arm64_ldrx(int sg, int sz_, int dst, int bas, uint64_t off) static void arm64_ldrx(int sg, int sz_, int dst, int bas, uint64_t off)
{ {
uint32_t sz = sz_; uint32_t sz = sz_;
@ -432,12 +449,28 @@ static void arm64_strv(int sz_, int dst, int bas, uint64_t off)
static void arm64_sym(int r, Sym *sym, unsigned long addend) static void arm64_sym(int r, Sym *sym, unsigned long addend)
{ {
o(0x10000060 | r); // adr xr,pc+12 greloca(cur_text_section, sym, ind, R_AARCH64_ADR_GOT_PAGE, 0);
o(0xf9400000 | r | (r << 5)); // ldr xr,[xr] o(0x90000000 | r); // adrp xr, #sym
o(0x14000003); // b + 8 greloca(cur_text_section, sym, ind, R_AARCH64_LD64_GOT_LO12_NC, 0);
greloca(cur_text_section, sym, ind, R_AARCH64_ABS64, addend); o(0xf9400000 | r | (r << 5)); // ld xr,[xr, #sym]
o(0); if (addend) {
o(0); // add xr, xr, #addend
if (addend & 0xffful)
o(0x91000000 | r | r << 5 | (addend & 0xfff) << 10);
if (addend > 0xffful) {
// add xr, xr, #addend, lsl #12
if (addend & 0xfff000ul)
o(0x91400000 | r | r << 5 | ((addend >> 12) & 0xfff) << 10);
if (addend > 0xfffffful) {
/* very unlikely */
int t = r ? 0 : 1;
o(0xf81f0fe0 | t); /* str xt, [sp, #-16]! */
arm64_movimm(t, addend & ~0xfffffful); // use xt for addent
o(0x91000000 | r | (t << 5)); /* add xr, xt, #0 */
o(0xf84107e0 | t); /* ldr xt, [sp], #16 */
}
}
}
} }
static void arm64_load_cmp(int r, SValue *sv); static void arm64_load_cmp(int r, SValue *sv);
@ -460,12 +493,14 @@ ST_FUNC void load(int r, SValue *sv)
} }
if (svr == (VT_CONST | VT_LVAL)) { if (svr == (VT_CONST | VT_LVAL)) {
arm64_sym(30, sv->sym, sv->c.i); // use x30 for address arm64_sym(30, sv->sym, // use x30 for address
arm64_check_offset(0, arm64_type_size(svtt), sv->c.i));
if (IS_FREG(r)) if (IS_FREG(r))
arm64_ldrv(arm64_type_size(svtt), fltr(r), 30, 0); arm64_ldrv(arm64_type_size(svtt), fltr(r), 30,
arm64_check_offset(1, arm64_type_size(svtt), sv->c.i));
else else
arm64_ldrx(!(svtt & VT_UNSIGNED), arm64_type_size(svtt), arm64_ldrx(!(svtt&VT_UNSIGNED), arm64_type_size(svtt), intr(r), 30,
intr(r), 30, 0); arm64_check_offset(1, arm64_type_size(svtt), sv->c.i));
return; return;
} }
@ -481,12 +516,14 @@ ST_FUNC void load(int r, SValue *sv)
} }
if (svr == (VT_CONST | VT_LVAL | VT_SYM)) { if (svr == (VT_CONST | VT_LVAL | VT_SYM)) {
arm64_sym(30, sv->sym, svcul); // use x30 for address arm64_sym(30, sv->sym, // use x30 for address
arm64_check_offset(0, arm64_type_size(svtt), svcul));
if (IS_FREG(r)) if (IS_FREG(r))
arm64_ldrv(arm64_type_size(svtt), fltr(r), 30, 0); arm64_ldrv(arm64_type_size(svtt), fltr(r), 30,
arm64_check_offset(1, arm64_type_size(svtt), svcul));
else else
arm64_ldrx(!(svtt & VT_UNSIGNED), arm64_type_size(svtt), arm64_ldrx(!(svtt&VT_UNSIGNED), arm64_type_size(svtt), intr(r), 30,
intr(r), 30, 0); arm64_check_offset(1, arm64_type_size(svtt), svcul));
return; return;
} }
@ -571,11 +608,14 @@ ST_FUNC void store(int r, SValue *sv)
} }
if (svr == (VT_CONST | VT_LVAL)) { if (svr == (VT_CONST | VT_LVAL)) {
arm64_sym(30, sv->sym, sv->c.i); // use x30 for address arm64_sym(30, sv->sym, // use x30 for address
arm64_check_offset(0, arm64_type_size(svtt), sv->c.i));
if (IS_FREG(r)) if (IS_FREG(r))
arm64_strv(arm64_type_size(svtt), fltr(r), 30, 0); arm64_strv(arm64_type_size(svtt), fltr(r), 30,
arm64_check_offset(1, arm64_type_size(svtt), sv->c.i));
else else
arm64_strx(arm64_type_size(svtt), intr(r), 30, 0); arm64_strx(arm64_type_size(svtt), intr(r), 30,
arm64_check_offset(1, arm64_type_size(svtt), sv->c.i));
return; return;
} }
@ -588,11 +628,14 @@ ST_FUNC void store(int r, SValue *sv)
} }
if (svr == (VT_CONST | VT_LVAL | VT_SYM)) { if (svr == (VT_CONST | VT_LVAL | VT_SYM)) {
arm64_sym(30, sv->sym, svcul); // use x30 for address arm64_sym(30, sv->sym, // use x30 for address
arm64_check_offset(0, arm64_type_size(svtt), svcul));
if (IS_FREG(r)) if (IS_FREG(r))
arm64_strv(arm64_type_size(svtt), fltr(r), 30, 0); arm64_strv(arm64_type_size(svtt), fltr(r), 30,
arm64_check_offset(1, arm64_type_size(svtt), svcul));
else else
arm64_strx(arm64_type_size(svtt), intr(r), 30, 0); arm64_strx(arm64_type_size(svtt), intr(r), 30,
arm64_check_offset(1, arm64_type_size(svtt), svcul));
return; return;
} }
@ -634,8 +677,6 @@ static void gen_bounds_prolog(void)
o(0xd503201f); /* nop -> mov x0, lbound section pointer */ o(0xd503201f); /* nop -> mov x0, lbound section pointer */
o(0xd503201f); o(0xd503201f);
o(0xd503201f); o(0xd503201f);
o(0xd503201f);
o(0xd503201f);
o(0xd503201f); /* nop -> call __bound_local_new */ o(0xd503201f); /* nop -> call __bound_local_new */
} }
@ -660,12 +701,10 @@ static void gen_bounds_epilog(void)
if (offset_modified) { if (offset_modified) {
saved_ind = ind; saved_ind = ind;
ind = func_bound_ind; ind = func_bound_ind;
o(0x10000060 | 0); // adr x0,pc+12 greloca(cur_text_section, sym_data, ind, R_AARCH64_ADR_GOT_PAGE, 0);
o(0xf9400000 | 0 | (0 << 5)); // ldr x0,[x0] o(0x90000000 | 0); // adrp x0, #sym_data
o(0x14000003); // b + 8 greloca(cur_text_section, sym_data, ind, R_AARCH64_LD64_GOT_LO12_NC, 0);
greloca(cur_text_section, sym_data, ind, R_AARCH64_ABS64, 0); o(0xf9400000 | 0 | (0 << 5)); // ld x0,[x0, #sym_data]
o(0);
o(0);
gen_bounds_call(TOK___bound_local_new); gen_bounds_call(TOK___bound_local_new);
ind = saved_ind; ind = saved_ind;
} }
@ -673,12 +712,10 @@ static void gen_bounds_epilog(void)
/* generate bound check local freeing */ /* generate bound check local freeing */
o(0xf81f0fe0); /* str x0, [sp, #-16]! */ o(0xf81f0fe0); /* str x0, [sp, #-16]! */
o(0x3c9f0fe0); /* str q0, [sp, #-16]! */ o(0x3c9f0fe0); /* str q0, [sp, #-16]! */
o(0x10000060 | 0); // adr x0,pc+12 greloca(cur_text_section, sym_data, ind, R_AARCH64_ADR_GOT_PAGE, 0);
o(0xf9400000 | 0 | (0 << 5)); // ldr x0,[x0] o(0x90000000 | 0); // adrp x0, #sym_data
o(0x14000003); // b + 8 greloca(cur_text_section, sym_data, ind, R_AARCH64_LD64_GOT_LO12_NC, 0);
greloca(cur_text_section, sym_data, ind, R_AARCH64_ABS64, 0); o(0xf9400000 | 0 | (0 << 5)); // ld x0,[x0, #sym_data]
o(0);
o(0);
gen_bounds_call(TOK___bound_local_delete); gen_bounds_call(TOK___bound_local_delete);
o(0x3cc107e0); /* ldr q0, [sp], #16 */ o(0x3cc107e0); /* ldr q0, [sp], #16 */
o(0xf84107e0); /* ldr x0, [sp], #16 */ o(0xf84107e0); /* ldr x0, [sp], #16 */