p95768021
/
InfiniTensor
forked from jiuyuan/InfiniTensor
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

fix format

This commit is contained in:
wanghailu 2023-10-23 10:48:35 +08:00
parent b1bdbbf478
commit 412f301323
2 changed files with 73 additions and 63 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
include/ascend/ascend_runtime.h
View File

@ -1,7 +1,6 @@
#pragma once
#include "core/runtime.h"
#include "ascend/ascend_common.h"
#include "core/runtime.h"
#define CHECK_RET(cond, return_expr) \
do { \
@ -25,18 +24,22 @@ class ASCENDRuntimeObj : public RuntimeObj {
size_t workspaceSize;
public:
ASCENDRuntimeObj(int deviceId = 0)
: RuntimeObj(Device::ASCEND, deviceId) {
ASCENDRuntimeObj(int deviceId = 0) : RuntimeObj(Device::ASCEND, deviceId) {
auto ret = aclrtSetDevice(deviceId);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSetDevice failed. ERROR: %d\n", ret));
CHECK_RET(ret == ACL_SUCCESS,
LOG_PRINT("aclrtSetDevice failed. ERROR: %d\n", ret));
ret = aclrtCreateContext(&aclnn, deviceId);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtCreateContext failed. ERROR: %d\n", ret));
CHECK_RET(ret == ACL_SUCCESS,
LOG_PRINT("aclrtCreateContext failed. ERROR: %d\n", ret));
ret = aclrtSetCurrentContext(aclnn);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtSetCurrentContext failed. ERROR: %d\n", ret));
CHECK_RET(ret == ACL_SUCCESS,
LOG_PRINT("aclrtSetCurrentContext failed. ERROR: %d\n", ret));
ret = aclrtCreateStream(&stream);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclrtCreateStream failed. ERROR: %d\n", ret));
CHECK_RET(ret == ACL_SUCCESS,
LOG_PRINT("aclrtCreateStream failed. ERROR: %d\n", ret));
ret = aclInit(nullptr);
CHECK_RET(ret == ACL_SUCCESS, LOG_PRINT("aclInit failed. ERROR: %d\n", ret));
CHECK_RET(ret == ACL_SUCCESS,
LOG_PRINT("aclInit failed. ERROR: %d\n", ret));
// 10GB for Longformer
// size_t longformerNum = 3lu * (1 << 30);
workspaceSize = 3ll << 30; // 3 GB
@ -73,19 +76,19 @@ class ASCENDRuntimeObj : public RuntimeObj {
void copyBlobFromCPU(void *dst, const void *src,
size_t bytes) const override {
aclrtMemcpy(dst, 1024*1024*1024, const_cast<void *>(src), bytes,
aclrtMemcpy(dst, 1024 * 1024 * 1024, const_cast<void *>(src), bytes,
ACL_MEMCPY_HOST_TO_DEVICE);
}
void copyBlobToCPU(void *dst, const void *src,
size_t bytes) const override {
aclrtMemcpy(dst, 1024*1024*1024, const_cast<void *>(src), bytes,
aclrtMemcpy(dst, 1024 * 1024 * 1024, const_cast<void *>(src), bytes,
ACL_MEMCPY_DEVICE_TO_HOST);
}
void copyBlobInsideRuntime(void *dst, const void *src,
size_t bytes) const override {
aclrtMemcpy(dst, 1024*1024*1024, const_cast<void *>(src), bytes,
aclrtMemcpy(dst, 1024 * 1024 * 1024, const_cast<void *>(src), bytes,
ACL_MEMCPY_DEVICE_TO_DEVICE);
}

33
src/kernels/ascend/unary.cc
View File

@ -1,7 +1,7 @@
#include "operators/unary.h"
#include "aclnnop/level2/aclnn_relu.h"
#include "ascend/ascend_kernel_without_config.h"
#include "ascend/ascend_runtime.h"
#include "aclnnop/level2/aclnn_relu.h"
namespace infini {
class ReluAclnn : public ASCENDKernelWithoutConfig {
@ -15,38 +15,45 @@ class ReluAclnn : public ASCENDKernelWithoutConfig {
auto a = op->getInputs(0)->getDims();
std::vector<int64_t> aDim(a.size(), 1);
for(size_t i = 0; i < a.size(); ++i) {
for (size_t i = 0; i < a.size(); ++i) {
aDim[i] = int64_t(a[i]);
}
auto aS = op->getInputs(0)->getStride();
std::vector<int64_t> aStride(aS.size(), 1);
for(size_t i = 0; i < aS.size(); ++i) {
for (size_t i = 0; i < aS.size(); ++i) {
aStride[i] = int64_t(aS[i]);
}
auto c = op->getInputs(0)->getDims();
std::vector<int64_t> cDim(c.size(), 1);
for(size_t i = 0; i < c.size(); ++i) {
for (size_t i = 0; i < c.size(); ++i) {
cDim[i] = int64_t(c[i]);
}
auto cS = op->getInputs(0)->getStride();
std::vector<int64_t> cStride(cS.size(), 1);
for(size_t i = 0; i < cS.size(); ++i) {
for (size_t i = 0; i < cS.size(); ++i) {
cStride[i] = int64_t(cS[i]);
}
auto input = aclCreateTensor(aDim.data(), aDim.size(), ACL_FLOAT, aStride.data(), 0, aclFormat::ACL_FORMAT_ND, aDim.data(), aDim.size(), aData);
auto output = aclCreateTensor(cDim.data(), cDim.size(), ACL_FLOAT, cStride.data(), 0, aclFormat::ACL_FORMAT_ND, cDim.data(), cDim.size(), cData);
auto input = aclCreateTensor(
aDim.data(), aDim.size(), ACL_FLOAT, aStride.data(), 0,
aclFormat::ACL_FORMAT_ND, aDim.data(), aDim.size(), aData);
auto output = aclCreateTensor(
cDim.data(), cDim.size(), ACL_FLOAT, cStride.data(), 0,
aclFormat::ACL_FORMAT_ND, cDim.data(), cDim.size(), cData);
uint64_t workspaceSize = 0;
aclOpExecutor* executor;
aclOpExecutor *executor;
auto ret = aclnnReluGetWorkspaceSize(input, output, &workspaceSize, &executor);
void* workspaceAddr = nullptr;
auto ret =
aclnnReluGetWorkspaceSize(input, output, &workspaceSize, &executor);
void *workspaceAddr = nullptr;
if (workspaceSize > 0) {
ret = aclrtMalloc(&workspaceAddr, workspaceSize, ACL_MEM_MALLOC_HUGE_FIRST);
ret = aclrtMalloc(&workspaceAddr, workspaceSize,
ACL_MEM_MALLOC_HUGE_FIRST);
}
assert(ret == ACL_SUCCESS);
ret = aclnnRelu(workspaceAddr, workspaceSize, executor, context->ASCENDHandle());
ret = aclnnRelu(workspaceAddr, workspaceSize, executor,
context->ASCENDHandle());
assert(ret == ACL_SUCCESS);
ret = aclrtSynchronizeStream(context->ASCENDHandle());
assert(ret == ACL_SUCCESS);
@ -57,4 +64,4 @@ class ReluAclnn : public ASCENDKernelWithoutConfig {
REGISTER_KERNEL(Device::ASCEND, OpType::Relu, DataType::Float32, ReluAclnn,
"relu_ASCEND_float");
};
}; // namespace infini