lzhengning
/
JittorMirror
mirror of https://github.com/Jittor/Jittor
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #646 from Jittor/fixHW 

adjust aclnn.h reference
This commit is contained in:
Yuxuan Han 2025-06-17 13:34:28 +08:00 committed by GitHub
parent 330dec69d2 3cf5d7f2a4
commit daf04e9fb5
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
9 changed files with 32 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
python/jittor/compiler.py
View File

@ -1191,6 +1191,8 @@ ascend_toolkit_home = os.getenv('ASCEND_TOOLKIT_HOME')
# build cache_compile
cc_flags += f" -I\"{os.path.join(jittor_path, 'src')}\" "
if ascend_toolkit_home:
cc_flags += f" -I\"{os.path.join(jittor_path, 'extern')}\" "
cc_flags += f" -I\"{os.path.join(ascend_toolkit_home, 'include')}\" "
cc_flags += f" -I\"{os.path.join(ascend_toolkit_home, 'include/acl')}\" "

1
python/jittor/extern/acl/acl_jittor.h vendored
View File

@ -7,6 +7,7 @@
#pragma once
#include "common.h"
#include <acl/acl.h>
#include "aclnn.h"
std::string acl_error_to_string(aclError error);

1
python/jittor/extern/acl/aclops/utils.cc vendored
View File

@ -5,6 +5,7 @@
#include <Python.h>
#include <pystate.h>
#include "utils.h"
#include "aclnn.h"
namespace jittor
{

1
python/jittor/extern/acl/aclops/utils.h vendored
View File

@ -6,6 +6,7 @@
#include <Python.h>
#include <pystate.h>
#include "misc/nano_string.h"
#include "aclnn.h"
namespace jittor
{

3
python/jittor/extern/mpi/ops/mpi_reduce_op.cc vendored
View File

@ -49,10 +49,13 @@ MpiReduceOp::MpiReduceOp(Var* x, NanoString op, int root) : x(x), op(op), root(r
forward(var);
return;
} else if (hccl_reduce) {
auto var = hccl_reduce(x, "sum", root);
//exe.run_sync({var}, true);
forward(var);
return;
}
}
#endif
y = create_output(nullptr, x->dtype());
}

2
python/jittor/src/common.h
View File

@ -8,7 +8,7 @@
#include <memory>
#include <functional>
#include "utils/log.h"
#include "../extern/acl/aclnn/aclnn.h"
// #include "../extern/acl/aclnn/aclnn.h"
#define JIT_TEST(name) extern void jit_test_ ## name ()
void expect_error(std::function<void()> func);

6
python/jittor/src/ops/array_op.cc
View File

@ -31,9 +31,9 @@ cudaEvent_t event;
struct Init {
Init() {
if (!get_device_count()) return;
//checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
//checkCudaErrors(cudaEventCreate(&event, cudaEventDisableTiming));
stream = aclstream;
checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
checkCudaErrors(cudaEventCreate(&event, cudaEventDisableTiming));
// stream = aclstream;
}
~Init() {
if (!get_device_count()) return;

2
python/jittor/src/ops/copy_op.cc
View File

@ -17,7 +17,7 @@
namespace jittor {
EXTERN_LIB aclrtStream aclstream;
// EXTERN_LIB aclrtStream aclstream;
CopyOp::CopyOp(Var* x) {
flags.set(NodeFlags::_cpu);

20
python/jittor/src/ops/fetch_op.cc
View File

@ -47,7 +47,7 @@ Init() {
if (!get_device_count()) return;
checkCudaErrors(cudaStreamCreateWithFlags(&stream, cudaStreamNonBlocking));
checkCudaErrors(cudaEventCreate(&event, cudaEventDisableTiming));
stream = aclstream;
// stream = aclstream;
}
~Init() {
if (!get_device_count()) return;
@ -123,11 +123,11 @@ void FetchOp::run() {
new (&allocation) Allocation(&cuda_dual_allocator, v->size);
// mostly device to device
#if IS_CUDA
// checkCudaErrors(cudaMemcpyAsync(
// allocation.ptr, v->mem_ptr, v->size, cudaMemcpyDefault, stream));
checkCudaErrors(cudaMemcpyAsync(
allocation.ptr, v->size, v->mem_ptr, v->size, cudaMemcpyDefault, aclstream));
checkCudaErrors(aclrtSynchronizeStream(aclstream));
allocation.ptr, v->mem_ptr, v->size, cudaMemcpyDefault, stream));
// checkCudaErrors(cudaMemcpyAsync(
// allocation.ptr, v->size, v->mem_ptr, v->size, cudaMemcpyDefault, aclstream));
// checkCudaErrors(aclrtSynchronizeStream(aclstream));
#else
checkCudaErrors(cudaMemcpyAsync(
allocation.ptr, v->mem_ptr, v->size, cudaMemcpyDeviceToDevice, stream));
@ -135,11 +135,11 @@ void FetchOp::run() {
auto host_ptr = cuda_dual_allocator.get_dual_allocation(
allocation.allocation).host_ptr;
// device to host
// checkCudaErrors(cudaMemcpyAsync(
// host_ptr, allocation.ptr, v->size, cudaMemcpyDeviceToHost, stream));
checkCudaErrors(aclrtMemcpyAsync(
host_ptr, v->size, allocation.ptr, v->size, cudaMemcpyDeviceToHost, aclstream));
checkCudaErrors(aclrtSynchronizeStream(aclstream));
checkCudaErrors(cudaMemcpyAsync(
host_ptr, allocation.ptr, v->size, cudaMemcpyDeviceToHost, stream));
// checkCudaErrors(aclrtMemcpyAsync(
// host_ptr, v->size, allocation.ptr, v->size, cudaMemcpyDeviceToHost, aclstream));
// checkCudaErrors(aclrtSynchronizeStream(aclstream));
allocation.ptr = host_ptr;
has_cuda_memcpy = true;
} else