mirror of
https://github.com/luxfi/zap.git
synced 2026-10-30 15:04:16 +00:00
zeekay
d14b104790
Three orthogonal additions, one wire format. v2 package — generic ad-hoc schema dispatch (~/work/lux/zap/v2): - README + IMPOSSIBILITY + BENCH_RESULTS document the v1↔v2 split. - builder/field/iter/list/pool/registry/schema/view + codegen package emit v1-equivalent fast paths from declarative .zap sources. - _compile_fail_test/ + examples/ cover negative and positive cases. - v2_test.go exercises the generic dispatch path. - doc.go in v1 marks v1 deprecated for *new* schema authoring; existing v1 buffers stay parseable by both v1 and v2. bufpool — pooled read slabs (bufpool.go + bufpool_test.go): - Message gains optional *bufRef; Release() returns the slab to its pool when sourced from the pooled read path, GC-managed otherwise. - ParseHeader returns (data, rootOffset) without allocating *Message, used by zapv2 to avoid two allocations per parse. - bench_read_test.go + node_quic_stream_test.go cover both paths. Transport allocator + UMA/RDMA cgo (transport/*): - Transport.Caps()/Buffer interface: capability + slab-typed buffer contract. Bytes() vs DevicePtr() decoupled so UMA and GPUDirect can share the same call surface. - DPDK PMD + GPUDirect RDMA + CUDA UMA real cgo backends gated behind build tags; stub paths remain default on darwin / non-Linux. - transport_test.go covers Caps reporting + Buffer lifecycle on the default transport. LLM.md + go.mod follow the additions.
69 lines
1.6 KiB
C
69 lines
1.6 KiB
C
// Copyright (C) 2025-2026, Lux Industries Inc. All rights reserved.
|
|
// See the file LICENSE for licensing terms.
|
|
//
|
|
//go:build cgo && linux && cuda
|
|
//
|
|
// cudaMallocManaged-backed UMA pool, host-side glue. Compiled into the
|
|
// luxfi/zap binary only when the cgo + linux + cuda build tags are set;
|
|
// see uma_linux_cuda.go for the Go cgo bridge.
|
|
//
|
|
// We keep this file pure C (not .cu) so the host's plain `gcc`/`cc` can
|
|
// build it; the CUDA runtime is linked via -lcudart and headers are
|
|
// pulled from cuda_runtime_api.h (the C-only subset of the CUDA SDK).
|
|
|
|
#include "uma_cuda.h"
|
|
|
|
#include <cuda_runtime_api.h>
|
|
#include <string.h>
|
|
|
|
int luxzap_cuda_probe(void) {
|
|
int count = 0;
|
|
cudaError_t e = cudaGetDeviceCount(&count);
|
|
if (e != cudaSuccess) {
|
|
return (int)e;
|
|
}
|
|
if (count == 0) {
|
|
return -1; // no device — distinct from a cuda error
|
|
}
|
|
return 0;
|
|
}
|
|
|
|
int luxzap_cuda_alloc_managed(size_t size, void** out_ptr) {
|
|
if (!out_ptr) {
|
|
return (int)cudaErrorInvalidValue;
|
|
}
|
|
*out_ptr = NULL;
|
|
void* p = NULL;
|
|
cudaError_t e = cudaMallocManaged(&p, size, cudaMemAttachGlobal);
|
|
if (e != cudaSuccess) {
|
|
return (int)e;
|
|
}
|
|
*out_ptr = p;
|
|
return 0;
|
|
}
|
|
|
|
int luxzap_cuda_free(void* ptr) {
|
|
if (!ptr) {
|
|
return 0;
|
|
}
|
|
return (int)cudaFree(ptr);
|
|
}
|
|
|
|
size_t luxzap_cuda_device_name(char* buf, size_t buflen) {
|
|
if (!buf || buflen == 0) {
|
|
return 0;
|
|
}
|
|
buf[0] = '\0';
|
|
struct cudaDeviceProp prop;
|
|
if (cudaGetDeviceProperties(&prop, 0) != cudaSuccess) {
|
|
return 0;
|
|
}
|
|
size_t n = strnlen(prop.name, sizeof(prop.name));
|
|
if (n >= buflen) {
|
|
n = buflen - 1;
|
|
}
|
|
memcpy(buf, prop.name, n);
|
|
buf[n] = '\0';
|
|
return n;
|
|
}
|