aclnnWeightQuantMatmulAllReduce
支持的产品型号
- Atlas A2训练系列产品/Atlas 800I A2推理产品。
说明: 使用该接口时,请确保驱动固件包和CANN包都为配套的8.0.RC2版本或者配套的更高版本,否则将会引发报错,比如BUS ERROR等。
接口原型
每个算子分为两段式接口,必须先调用“aclnnWeightQuantMatmulAllReduceGetWorkspaceSize”接口获取计算所需workspace大小以及包含了算子计算流程的执行器,再调用“aclnnWeightQuantMatmulAllReduce”接口执行计算。
aclnnStatus aclnnWeightQuantMatmulAllReduceGetWorkspaceSize(const aclTensor *x1, const aclTensor *x2, const aclTensor *bias, const aclTensor *antiquantScale, const aclTensor *antiquantOffset, const aclTensor *x3, const char* group, const char *reduceOp, int64_t commTurn, int64_t streamMode, int64_t antiquantGroupSize, const aclTensor *output, uint64_t *workspaceSize, aclOpExecutor **executor);
aclnnStatus aclnnWeightQuantMatmulAllReduce(void *workspace, uint64_t workspaceSize, aclOpExecutor *executor, const aclrtStream stream);
功能描述
- 算子功能:对入参x2进行伪量化计算后,完成mm + all_reduce_base计算。
- 计算公式:
aclnnWeightQuantMatmulAllReduceGetWorkspaceSize
参数说明:
- x1(const aclTensor *, 计算输入):Device侧的aclTensor,mm左矩阵,维度可为2维或者3维,数据类型支持BFLOAT16、FLOAT16,数据格式支持ND。
- x2(const aclTensor *, 计算输入):Device侧的2维aclTensor,mm右矩阵,数据类型支持INT8、INT4,数据格式支持ND、FRACTAL_NZ。
- bias(const aclTensor *, 计算输入):Device侧的aclTensor,对应计算公式中bias偏移。维度为1维,数据类型支持BFLOAT16、FLOAT16,数据格式支持ND,可选,可为空,非空时shape和x2最后一维相等。
- antiquantScale(const aclTensor *, 计算输入):对x2进行伪量化计算的scale参数,数据类型支持BFLOAT16、FLOAT16,数据格式支持ND。per_tensor场景shape为[1];per_channel场景shape为[n]/[1,n],n为x2最后一维的大小;per_group场景shape为(ceil(k,antiquantGroupSize),n)。
- antiquantOffset(const aclTensor *, 计算输入):对x2进行伪量化计算的offset参数,数据类型支持BFLOAT16、FLOAT16,数据格式支持ND,可选,可为空,非空时shape与antiquantScale一致。
- x3(const aclTensor *, 计算输入):可选,matmul计算后的偏移,数据类型支持BFLOAT16、FLOAT16;数据格式支持ND,可选,可为空,非空时shape与matmul计算后的shape相同。
- group(const char *, 计算输入):通信域名称,数据类型支持String,通过Hccl提供的接口获取:extern HcclResult HcclGetCommName(HcclComm comm, char* commName); commName即为group。
- reduceOp(const char *, 计算输入):reduce操作类型,数据类型支持String,目前仅支持"sum"。
- commTurn(int64_t, 计算输入):通信数据切分数,即总数据量/单次通信量,数据类型支持int64_t,当前版本仅支持输入0。
- streamMode(int64_t,计算输入):Host侧的整型,AscendCL流模式的枚举,当前只支持枚举值1,数据类型支持int64_t。
- antiquantGroupSize(int64_t,计算输入):伪量化per_group模式下,对x2进行反量化计算的groupSize输入;当不支持per_group时,传入0,支持时,传入值的范围为[32,min(k-1,INT_MAX)],且为32的倍数;k取值范围与matmul接口保持一致。
- output(aclTensor *, 输出):计算+通信的结果,数据类型支持BFLOAT16、FLOAT16;数据类型与x1保持一致;shape除最后一维与x1除最后一维相等,最后一维和x2最后一维相等。
- workspaceSize(uint64_t *, 出参):返回需要在Device侧申请的workspace大小。
- executor(aclOpExecutor **, 出参):返回op执行器,包含了算子计算流程。
返回值:
返回aclnnStatus状态码,具体参见aclnn返回码。
aclnnWeightQuantMatmulAllReduce
参数说明:
- workspace(void*,入参):在Device侧申请的workspace内存地址。
- workspaceSize(uint64_t,入参):在Device侧申请的workspace大小,由第一段接口aclnnWeightQuantMatmulAllReduceGetWorkspaceSize。
- executor(aclOpExecutor*,入参):op执行器,包含了算子计算流程。
- stream(const aclrtStream,入参):指定执行任务的AscendCL stream流。
返回值:
返回aclnnStatus状态码,具体参见aclnn返回码。
约束与限制
- 增量场景不使能MC2,全量场景使能MC2。
- 输入x1可为2维或者3维,其维度为(b, s, k)或者(m, k)。x2必须是2维。其维度为(k, n),k轴满足matmul算子入参要求,k轴相等,k、n的范围为[1, 65535]。
- 传入的x1、x2、antiquantScale或者output不为空指针。
- x3(非空场景)以及输出output除最后一维皆与输入x1除最后一维相等,x3(非空场景)以及输出output的最后1维与输入x2的最后1维相等。bias若非空,shape大小与output最后一维相等。antiquantScale满足per-tensor场景shape为[1],per-channel场景shape为[1,n]/[n],n如果为1时,per-tensor和per-channel都只有1个数值,此时统一当做是per-tensor场景,per-group场景shape为[ceil(k,antiquantGroupSize),n]。antiquantOffset若非空,shape与antiquantScale一致。
- x1和x2,x3(非空场景)、antiquantScale、antiquantOffset(非空场景)、output、bias(非空场景)的数据类型和数据格式需要在支持的范围之内。
- x1,antiquantScale,antiquantOffset(非空场景),x3(非空场景)、bias(非空场景)output的数据类型相同 。
- antiquantGroupSize取值满足取值范围且为32倍数。
- 只支持x2矩阵转置/不转置,x1矩阵不支持转置场景。
- 在长序列场景,随着b/s或者m的增大,可能出现OOM或者计算超时。
- x2的Format为FRACTAL_NZ时,维度仅支持2维,配合aclnnCalculateMatmulWeightSizeV2/aclnnTransMatmulWeightGetWorkspaceSize/aclnnTransMatmulWeight完成输入ND到NZ的转换。
- Atlas A2训练系列产品/Atlas 800I A2推理产品支持1、2、4、8卡,并且仅支持hccs链路all mesh组网。
调用示例
示例代码如下,仅供参考,具体编译和执行过程请参考编译与运行样例。
#include <iostream>
#include <vector>
#include <thread>
#include <string.h>
#include "aclnnop/aclnn_weight_quant_matmul_all_reduce.h"
int ndev = 8;
#define CHECK_RET(cond, return_expr) \
do { \
if (!(cond)) { \
return_expr; \
} \
} while (0)
#define LOG_PRINT(message, ...) \
do { \
printf(message, ##__VA_ARGS__); \
} while (0)
int64_t GetShapeSize(const std::vector<int64_t> &shape) {
int64_t shapeSize = 1;
for (auto i: shape) {
shapeSize *= i;
}
return shapeSize;
}
template<typename T>
int CreateAclTensor(const std::vector<T> &hostData, const std::vector<int64_t> &shape, void **deviceAddr,
aclDataType dataType, aclTensor **tensor) {
auto size = GetShapeSize(shape) * sizeof(T);
// 调用aclrtMalloc申请device侧内存
auto ret = aclrtMalloc(deviceAddr, size, ACL_MEM_MALLOC_HUGE_FIRST);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtMalloc failed. ERROR: %d\n", ret); return ret);
// 调用aclrtMemcpy将host侧数据拷贝到device侧内存上
ret = aclrtMemcpy(*deviceAddr, size, hostData.data(), size, ACL_MEMCPY_HOST_TO_DEVICE);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtMemcpy failed. ERROR: %d\n", ret); return ret);
// 计算连续tensor的strides
std::vector<int64_t> strides(shape.size(), 1);
for (int64_t i = shape.size() - 2; i >= 0; i--) {
strides[i] = shape[i + 1] * strides[i + 1];
}
// 调用aclCreateTensor接口创建aclTensor
*tensor = aclCreateTensor(shape.data(), shape.size(), dataType, strides.data(), 0, aclFormat::ACL_FORMAT_ND,
shape.data(), shape.size(), *deviceAddr);
return 0;
}
struct Args {
uint32_t rankId;
HcclComm hcclComm;
aclrtStream stream;
aclrtContext context;
};
int launchOneThreadweightQuantmatmulAllReduce(Args &args) {
int ret;
ret = aclrtSetCurrentContext(args.context);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSetCurrentContext failed. ERROR: %d\n", ret); return ret);
char hcom_name[128];
ret = HcclGetCommName(args.hcclComm, hcom_name);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("[ERROR] HcclGetCommName failed. ret = %d \n", ret); return -1);
LOG_PRINT("[INFO] rank %d hcom: %s stream: %p, context : %p\n", args.rankId, hcom_name, args.stream,
args.context);
std::vector<int64_t> x1Shape = {32, 64};
std::vector<int64_t> x2Shape = {64, 128};
std::vector<int64_t> biasShape = {128};
std::vector<int64_t> antiquantScaleShape = {128};
std::vector<int64_t> antiquantOffsetShape = {128};
std::vector<int64_t> x3Shape = {32, 128};
std::vector<int64_t> outShape = {32, 128};
void *x1DeviceAddr = nullptr;
void *x2DeviceAddr = nullptr;
void *biasDeviceAddr = nullptr;
void *antiquantScaleDeviceAddr = nullptr;
void *antiquantOffsetDeviceAddr = nullptr;
void *x3DeviceAddr = nullptr;
void *outDeviceAddr = nullptr;
aclTensor *x1 = nullptr;
aclTensor *x2 = nullptr;
aclTensor *bias = nullptr;
aclTensor *antiquantScale = nullptr;
aclTensor *antiquantOffset = nullptr;
aclTensor *x3 = nullptr;
aclTensor *out = nullptr;
int64_t commTurn = 0;
int64_t streamMode = 1;
int64_t antiquantGroupSize = 0;
uint64_t workspaceSize = 0;
aclOpExecutor *executor;
void *workspaceAddr = nullptr;
long long x1ShapeSize = GetShapeSize(x1Shape);
long long x2ShapeSize = GetShapeSize(x2Shape);
long long biasShapeSize = GetShapeSize(biasShape);
long long antiquantScaleShapeSize = GetShapeSize(antiquantScaleShape);
long long antiquantOffsetShapeSize = GetShapeSize(antiquantOffsetShape);
long long x3ShapeSize = GetShapeSize(x3Shape);
long long outShapeSize = GetShapeSize(outShape);
std::vector<int16_t> x1HostData(x1ShapeSize, 0);
std::vector<int8_t> x2HostData(x2ShapeSize, 0);
std::vector<int16_t> biasHostData(biasShapeSize, 0);
std::vector<int16_t> antiquantScaleHostData(antiquantScaleShapeSize, 0);
std::vector<int16_t> antiquantOffsetHostData(antiquantOffsetShapeSize, 0);
std::vector<int16_t> x3HostData(x3ShapeSize, 0);
std::vector<int16_t> outHostData(outShapeSize, 0);
// 创建 tensor
ret = CreateAclTensor(x1HostData, x1Shape, &x1DeviceAddr, aclDataType::ACL_FLOAT16, &x1);
CHECK_RET(ret == ACL_SUCCESS, return ret);
ret = CreateAclTensor(x2HostData, x2Shape, &x2DeviceAddr, aclDataType::ACL_INT8, &x2);
CHECK_RET(ret == ACL_SUCCESS, return ret);
ret = CreateAclTensor(biasHostData, biasShape, &biasDeviceAddr, aclDataType::ACL_FLOAT16, &bias);
CHECK_RET(ret == ACL_SUCCESS, return ret);
ret = CreateAclTensor(antiquantScaleHostData, antiquantScaleShape, &antiquantScaleDeviceAddr,
aclDataType::ACL_FLOAT16, &antiquantScale);
CHECK_RET(ret == ACL_SUCCESS, return ret);
ret = CreateAclTensor(antiquantOffsetHostData, antiquantOffsetShape, &antiquantOffsetDeviceAddr,
aclDataType::ACL_FLOAT16, &antiquantOffset);
CHECK_RET(ret == ACL_SUCCESS, return ret);
ret = CreateAclTensor(x3HostData, x3Shape, &x3DeviceAddr, aclDataType::ACL_FLOAT16, &x3);
CHECK_RET(ret == ACL_SUCCESS, return ret);
ret = CreateAclTensor(outHostData, outShape, &outDeviceAddr, aclDataType::ACL_FLOAT16, &out);
CHECK_RET(ret == ACL_SUCCESS, return ret);
// 调用第一段接口
ret = aclnnWeightQuantMatmulAllReduceGetWorkspaceSize(x1, x2, bias, antiquantScale, antiquantOffset, x3, hcom_name,
"sum", commTurn, streamMode, antiquantGroupSize, out,
&workspaceSize, &executor);
CHECK_RET(ret == ACL_SUCCESS,
LOG_PRINT("aclnnWeightQuantMatmulAllReduceGetWorkspaceSize failed. ERROR: %d\n", ret); return ret);
// 根据第一段接口计算出的workspaceSize申请device内存
if (workspaceSize > 0) {
ret = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("allocate workspace failed. ERROR: %d\n", ret); return ret);
}
// 调用第二段接口
ret = aclnnWeightQuantMatmulAllReduce(workspaceAddr, workspaceSize, executor, args.stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclnnWeightQuantMatmulAllReduce failed. ERROR: %d\n", ret); return ret);
//(固定写法)同步等待任务执行结束
ret = aclrtSynchronizeStreamWithTimeout(args.stream, 10000);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSynchronizeStream failed. ERROR: %d\n", ret); return ret);
LOG_PRINT("device%d aclnnWeightQuantMatmulAllReduce execute success \n", args.rankId);
// 释放device资源,需要根据具体API的接口定义修改
if (x1 != nullptr) {
aclDestroyTensor(x1);
}
if (x2 != nullptr) {
aclDestroyTensor(x2);
}
if (bias != nullptr) {
aclDestroyTensor(bias);
}
if (antiquantScale != nullptr) {
aclDestroyTensor(antiquantScale);
}
if (antiquantOffset != nullptr) {
aclDestroyTensor(antiquantOffset);
}
if (x3 != nullptr) {
aclDestroyTensor(x3);
}
if (out != nullptr) {
aclDestroyTensor(out);
}
if (x1DeviceAddr != nullptr) {
aclrtFree(x1DeviceAddr);
}
if (x2DeviceAddr != nullptr) {
aclrtFree(x2DeviceAddr);
}
if (biasDeviceAddr != nullptr) {
aclrtFree(biasDeviceAddr);
}
if (antiquantScaleDeviceAddr != nullptr) {
aclrtFree(antiquantScaleDeviceAddr);
}
if (antiquantOffsetDeviceAddr != nullptr) {
aclrtFree(antiquantOffsetDeviceAddr);
}
if (x3DeviceAddr != nullptr) {
aclrtFree(x3DeviceAddr);
}
if (outDeviceAddr != nullptr) {
aclrtFree(outDeviceAddr);
}
if (workspaceSize > 0) {
aclrtFree(workspaceAddr);
}
aclrtDestroyStream(args.stream);
HcclCommDestroy(args.hcclComm);
aclrtDestroyContext(args.context);
aclrtResetDevice(args.rankId);
return 0;
}
int main(int argc, char *argv[]) {
int ret;
int32_t devices[ndev];
for (int i = 0; i < ndev; i++) {
devices[i] = i;
}
HcclComm comms[128];
ret = aclInit(nullptr);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclInit failed. ERROR: %d\n", ret); return ret);
// 初始化集合通信域
for (int i = 0; i < ndev; i++) {
ret = aclrtSetDevice(devices[i]);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSetDevice failed. ERROR: %d\n", ret); return ret);
}
ret = HcclCommInitAll(ndev, devices, comms);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("HcclCommInitAll failed. ERROR: %d\n", ret); return ret);
Args args[ndev];
aclrtStream stream[ndev];
aclrtContext context[ndev];
for (uint32_t rankId = 0; rankId < ndev; rankId++) {
ret = aclrtSetDevice(rankId);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSetDevice failed. ERROR: %d\n", ret); return ret);
ret = aclrtCreateContext(&context[rankId], rankId);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtCreateContext failed. ERROR: %d\n", ret); return ret);
ret = aclrtCreateStream(&stream[rankId]);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtCreateStream failed. ERROR: %d\n", ret); return ret);
}
// 启动多线程
std::vector<std::unique_ptr<std::thread>> threads(ndev);
for (uint32_t rankId = 0; rankId < ndev; rankId++) {
args[rankId].rankId = rankId;
args[rankId].hcclComm = comms[rankId];
args[rankId].stream = stream[rankId];
args[rankId].context = context[rankId];
threads[rankId].reset(
new(std::nothrow) std::thread(&launchOneThreadweightQuantmatmulAllReduce, std::ref(args[rankId])));
}
for (uint32_t rankId = 0; rankId < ndev; rankId++) {
threads[rankId]->join();
}
aclFinalize();
return 0;
}