A high-performance inference engine for AI models
#pragma once
#include <metal_stdlib>
using namespace metal;
struct ThreadContext {
uint simd_lane_id; // [[thread_index_in_simdgroup]] — 0..simdgroup_size
uint simdgroup_index; // [[simdgroup_index_in_threadgroup]] —
// 0..simdgroups_per_threadgroup
uint simdgroup_size; // [[threads_per_simdgroup]] — typically 32
uint simdgroups_per_threadgroup; // [[simdgroups_per_threadgroup]]
uint3 threadgroup_position; // [[threadgroup_position_in_grid]]
uint3 threadgroup_size;
uint3 threadgroup_count;
uint3 grid_size;
};