github-mirrors/vortex
1
0
Fork 0
mirror of https://github.com/vortexgpgpu/vortex.git synced 2025-04-24 13:57:17 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

vx_spawn_threads implementation

Browse source
This commit is contained in:
Blaise Tine 2024-06-07 07:52:15 -07:00
parent 8c5a783477
commit 96cb381885
5 changed files with 208 additions and 152 deletions
Download patch file Download diff file Expand all files Collapse all files

46
kernel/include/vx_spawn.h
View file

@ -14,28 +14,54 @@
#ifndef __VX_SPAWN_H__
#define __VX_SPAWN_H__
#include <VX_types.h>
#include <vx_intrinsics.h>
#include <stdint.h>
#ifdef __cplusplus
extern "C" {
#endif
typedef void (*vx_spawn_tasks_cb)(int task_id, const void *arg);
typedef void (*vx_spawn_task_groups_cb)(int local_task_id, int group_id, int local_group_id, int warps_per_group, const void *arg);
typedef void (*vx_spawn_tasks_cb)(uint32_t task_id, const void *arg);
typedef void (*vx_serial_cb)(const void *arg);
void vx_spawn_tasks(int num_tasks, vx_spawn_tasks_cb callback, const void * arg);
void vx_spawn_tasks(uint32_t num_tasks, vx_spawn_tasks_cb callback, const void * arg);
void vx_spawn_task_groups(int num_groups, int group_size, vx_spawn_task_groups_cb callback, const void * arg);
void vx_serial(vx_serial_cb callback, void * arg);
inline void* vx_local_malloc(int local_group_id, int size) {
return (int8_t*)csr_read(VX_CSR_LOCAL_MEM_BASE) + local_group_id * size;
}
///////////////////////////////////////////////////////////////////////////////
void vx_serial(vx_serial_cb callback, const void * arg);
typedef union {
struct {
uint32_t x;
uint32_t y;
uint32_t z;
};
uint32_t m[3];
} dim3_t;
extern __thread dim3_t blockIdx;
extern __thread dim3_t threadIdx;
extern dim3_t gridDim;
extern dim3_t blockDim;
extern __thread uint32_t __local_group_id;
extern uint32_t __groups_per_core;
extern uint32_t __warps_per_group;
typedef void (*vx_kernel_func_cb)(const void *arg);
#define __local_mem(size) \
(void*)((int8_t*)csr_read(VX_CSR_LOCAL_MEM_BASE) + __local_group_id * size)
#define __syncthreads() \
vx_barrier(__COUNTER__ * __groups_per_core + __local_group_id, __warps_per_group)
int vx_spawn_threads(uint32_t dimension,
const uint32_t* grid_dim,
const uint32_t* block_dim,
vx_kernel_func_cb kernel_func,
const void* arg);
#ifdef __cplusplus
}