github-mirrors/vortex
1
0
Fork 0
mirror of https://github.com/vortexgpgpu/vortex.git synced 2025-04-23 21:39:10 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

vx_serial runtime API

Browse source
This commit is contained in:
Blaise Tine 2021-06-28 05:43:36 -04:00
parent 342c07f8d6
commit 47d317f17a
6 changed files with 163 additions and 46 deletions
Download patch file Download diff file Expand all files Collapse all files

4
runtime/Makefile
View file

@ -5,12 +5,12 @@ AR = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-gcc-ar
DP = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objdump
CP = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objcopy
CFLAGS += -O3 -march=rv32imf -mabi=ilp32f -fno-exceptions -fdata-sections -ffunction-sections
CFLAGS += -O3 -march=rv32imf -mabi=ilp32f -Wstack-usage=1024 -fno-exceptions -fdata-sections -ffunction-sections
CFLAGS += -I./include -I../hw
PROJECT = libvortexrt
SRCS = ./src/vx_start.S ./src/vx_print.S ./src/vx_print.c ./src/vx_spawn.c ./src/vx_perf.c
SRCS = ./src/vx_start.S ./src/vx_syscalls.c ./src/vx_print.S ./src/vx_print.c ./src/vx_spawn.c ./src/vx_spawn.S ./src/vx_perf.c
OBJS := $(addsuffix .o, $(notdir $(SRCS)))

8
runtime/include/vx_spawn.h
View file

@ -26,11 +26,13 @@ typedef void (*pfn_workgroup_func) (
uint32_t /* group_z */
);
typedef void (*pfn_callback)(int task_id, const void *arg);
typedef void (*pfn_callback)(int task_id, void *arg);
void vx_spawn_kernel(struct context_t * ctx, pfn_workgroup_func wg_func, const void * args);
void vx_spawn_kernel(struct context_t * ctx, pfn_workgroup_func wg_func, void * arg);
void vx_spawn_tasks(int num_tasks, pfn_callback callback, const void * args);
void vx_spawn_tasks(int num_tasks, pfn_callback callback, void * arg);
void vx_serial(pfn_callback callback, void * arg);
#ifdef __cplusplus
}

50
runtime/src/vx_print.c
View file

@ -1,4 +1,5 @@
#include <vx_print.h>
#include <vx_spawn.h>
#include <vx_intrinsics.h>
#include <stdlib.h>
#include <stdbool.h>
@ -8,45 +9,32 @@
extern "C" {
#endif
int __attribute__((noinline)) __vprintf(int index, int tid, const char* format, va_list va) {
__if (index == tid) {
return vprintf(format, va);
}__endif
return 0;
struct printf_arg_t {
const char* format;
va_list va;
int ret;
};
static void __printf_callback(int task_id, void* arg) {
struct printf_arg_t* p_arg = (struct printf_arg_t*)(arg);
p_arg->ret = vprintf(p_arg->format, p_arg->va);
}
int vx_vprintf(const char* format, va_list va) {
int ret = 0;
// need to execute single-threaded due to potential thread-data dependency
// use manual goto loop to disable compiler optimizations affceting split/join placement
volatile int nt = vx_num_threads();
int tid = vx_thread_id();
for (int i = 0; i < nt; ++i) {
ret |= __vprintf(i, tid, format, va);
}
return ret;
// need to execute 'vprintf' single-threaded due to potential thread-data dependency
struct printf_arg_t arg;
arg.format = format;
arg.va = va;
vx_serial(__printf_callback, &arg);
return arg.ret;
}
int vx_printf(const char * format, ...) {
int ret = 0;
// need to execute single-threaded due to potential thread-data dependency
// use manual goto loop to disable compiler optimizations affceting split/join placement
volatile int nt = vx_num_threads();
int tid = vx_thread_id();
int ret;
va_list va;
va_start(va, format);
for (int i = 0; i < nt; ++i) {
ret |= __vprintf(i, tid, format, va);
}
va_end(va);
ret = vx_vprintf(format, va);
va_end(va);
return ret;
}

36
runtime/src/vx_spawn.S Normal file
View file

@ -0,0 +1,36 @@
.type vx_serial, @function
.global vx_serial
vx_serial:
addi sp, sp, -24
sw ra, 20(sp)
sw s4, 16(sp)
sw s3, 12(sp)
sw s2, 8(sp)
sw s1, 4(sp)
sw s0, 0(sp)
mv s4, a0 # callback
mv s3, a1 # arg
csrr s2, 0xfc0 # NT
csrr s1, 0xcc0 # tid
li s0, 0 # index
label_loop:
sub t0, s0, s1
snez t0, t0
.insn s 0x6b, 2, x0, 0(t0) # split t0
bnez t0, label_join
mv a0, s0 # a0 <- index
mv a1, s3 # a1 <- arg
jalr s4 # callback(index, arg)
label_join:
.insn s 0x6b, 3, x0, 0(x0) # join
addi s0, s0, 1
blt s0, s2, label_loop
lw ra, 20(sp)
lw s4, 16(sp)
lw s3, 12(sp)
lw s2, 8(sp)
lw s1, 4(sp)
lw s0, 0(sp)
addi sp, sp, 24
ret

20
runtime/src/vx_spawn.c
View file

@ -12,7 +12,7 @@ extern "C" {
typedef struct {
pfn_callback callback;
const void * args;
void * arg;
int offset;
int N;
int R;
@ -21,7 +21,7 @@ typedef struct {
typedef struct {
struct context_t * ctx;
pfn_workgroup_func wg_func;
const void * args;
void * arg;
int offset;
int N;
int R;
@ -57,7 +57,7 @@ static void spawn_tasks_callback() {
int offset = p_wspawn_args->offset + (wK * NT) + (tid * tK);
for (int task_id = offset, N = task_id + tK; task_id < N; ++task_id) {
(p_wspawn_args->callback)(task_id, p_wspawn_args->args);
(p_wspawn_args->callback)(task_id, p_wspawn_args->arg);
}
vx_tmc(0 == wid);
@ -72,12 +72,12 @@ void spawn_remaining_tasks_callback(int nthreads) {
wspawn_tasks_args_t* p_wspawn_args = (wspawn_tasks_args_t*)g_wspawn_args[core_id];
int task_id = p_wspawn_args->offset + tid;
(p_wspawn_args->callback)(task_id, p_wspawn_args->args);
(p_wspawn_args->callback)(task_id, p_wspawn_args->arg);
vx_tmc(1);
}
void vx_spawn_tasks(int num_tasks, pfn_callback callback , const void * args) {
void vx_spawn_tasks(int num_tasks, pfn_callback callback , void * arg) {
// device specs
int NC = vx_num_cores();
int NW = vx_num_warps();
@ -112,7 +112,7 @@ void vx_spawn_tasks(int num_tasks, pfn_callback callback , const void * args) {
fW = 1;
//--
wspawn_tasks_args_t wspawn_args = { callback, args, core_id * tasks_per_core, fW, rW };
wspawn_tasks_args_t wspawn_args = { callback, arg, core_id * tasks_per_core, fW, rW };
g_wspawn_args[core_id] = &wspawn_args;
//--
@ -159,7 +159,7 @@ static void spawn_kernel_callback() {
int gid1 = p_wspawn_args->ctx->global_offset[1] + j;
int gid2 = p_wspawn_args->ctx->global_offset[2] + k;
(p_wspawn_args->wg_func)(p_wspawn_args->args, p_wspawn_args->ctx, gid0, gid1, gid2);
(p_wspawn_args->wg_func)(p_wspawn_args->arg, p_wspawn_args->ctx, gid0, gid1, gid2);
}
vx_tmc(0 == wid);
@ -188,12 +188,12 @@ static void spawn_kernel_remaining_callback(int nthreads) {
int gid1 = p_wspawn_args->ctx->global_offset[1] + j;
int gid2 = p_wspawn_args->ctx->global_offset[2] + k;
(p_wspawn_args->wg_func)(p_wspawn_args->args, p_wspawn_args->ctx, gid0, gid1, gid2);
(p_wspawn_args->wg_func)(p_wspawn_args->arg, p_wspawn_args->ctx, gid0, gid1, gid2);
vx_tmc(1);
}
void vx_spawn_kernel(struct context_t * ctx, pfn_workgroup_func wg_func, const void * args) {
void vx_spawn_kernel(struct context_t * ctx, pfn_workgroup_func wg_func, void * arg) {
// total number of WGs
int X = ctx->num_groups[0];
int Y = ctx->num_groups[1];
@ -241,7 +241,7 @@ void vx_spawn_kernel(struct context_t * ctx, pfn_workgroup_func wg_func, const v
char log2X = fast_log2(X);
//--
wspawn_kernel_args_t wspawn_args = { ctx, wg_func, args, core_id * wgs_per_core, fW, rW, isXYpow2, isXpow2, log2XY, log2X };
wspawn_kernel_args_t wspawn_args = { ctx, wg_func, arg, core_id * wgs_per_core, fW, rW, isXYpow2, isXpow2, log2XY, log2X };
g_wspawn_args[core_id] = &wspawn_args;
//--

91
runtime/src/vx_syscalls.c Normal file
View file

@ -0,0 +1,91 @@
#include <sys/stat.h>
#include <newlib.h>
#include <unistd.h>
#include <vx_intrinsics.h>
#include <vx_print.h>
int _close(int file) { return -1; }
int _fstat(int file, struct stat *st) { return -1; }
int _isatty(int file) { return 0; }
int _lseek(int file, int ptr, int dir) { return 0; }
int _open(const char *name, int flags, int mode) { return -1; }
int _read(int file, char *ptr, int len) { return -1; }
caddr_t _sbrk(int incr) { return 0; }
int _write(int file, char *ptr, int len) {
int i;
for (i = 0; i < len; ++i) {
vx_putchar(*ptr++);
}
return len;
}
int _kill(int pid, int sig) { return -1; }
int _getpid() {
return vx_warp_gid();
}
#ifdef HAVE_INITFINI_ARRAY
/* These magic symbols are provided by the linker. */
extern void (*__preinit_array_start []) (void) __attribute__((weak));
extern void (*__preinit_array_end []) (void) __attribute__((weak));
extern void (*__init_array_start []) (void) __attribute__((weak));
extern void (*__init_array_end []) (void) __attribute__((weak));
#ifdef HAVE_INIT_FINI
extern void _init (void);
#endif
/* Iterate over all the init routines. */
void
__libc_init_array (void)
{
size_t count;
size_t i;
count = __preinit_array_end - __preinit_array_start;
for (i = 0; i < count; i++)
__preinit_array_start[i] ();
#ifdef HAVE_INIT_FINI
_init ();
#endif
count = __init_array_end - __init_array_start;
for (i = 0; i < count; i++)
__init_array_start[i] ();
}
#endif
#ifdef HAVE_INITFINI_ARRAY
extern void (*__fini_array_start []) (void) __attribute__((weak));
extern void (*__fini_array_end []) (void) __attribute__((weak));
#ifdef HAVE_INIT_FINI
extern void _fini (void);
#endif
/* Run all the cleanup routines. */
void
__libc_fini_array (void)
{
size_t count;
size_t i;
count = __fini_array_end - __fini_array_start;
for (i = count; i > 0; i--)
__fini_array_start[i-1] ();
#ifdef HAVE_INIT_FINI
_fini ();
#endif
}
#endif