A high-performance inference engine for AI models
const std = @import("std");
const bench_main = @import("bench_main.zig");
pub const mlx = bench_main.mlx;
pub const log = bench_main.log;
pub const io_util = bench_main.io_util;
pub const env = bench_main.env;
export fn benchmarks_mlxserve_main(argc: c_int, argv: [*][*:0]u8) c_int {
var count: usize = 0;
while (std.c.environ[count] != null) : (count += 1) {}
const environ: std.process.Environ.Block = .{ .slice = std.c.environ[0..count :null] };
const args = argv[0..@intCast(argc)];
const gpa = std.heap.c_allocator;
var arena = std.heap.ArenaAllocator.init(std.heap.page_allocator);
defer arena.deinit();
var threaded: std.Io.Threaded = .init(gpa, .{ .argv0 = .init(.{ .vector = args }), .environ = .{ .block = environ } });
defer threaded.deinit();
var environ_map = std.process.Environ.createMap(.{ .block = environ }, gpa) catch return 1;
defer environ_map.deinit();
const preopens = std.process.Preopens.init(arena.allocator()) catch return 1;
bench_main.main(.{
.minimal = .{ .args = .{ .vector = args }, .environ = .{ .block = environ } },
.arena = &arena,
.gpa = gpa,
.io = threaded.io(),
.environ_map = &environ_map,
.preopens = preopens,
}) catch |err| {
log.err("{s}\n", .{@errorName(err)});
return 1;
};
return 0;
}