csullivan

Source: https://www.retrospect.com/au/support/kb/repair_efi_bootloader

This guide details how to rebuild the BCD boot store for windows.


Use diskpart to make System partition available.
(This will allow us to use bcdedit later in this guide to repair the bootloader.)

Boot from the Windows installation media

csullivan

Ubuntu 18.04 VFIO GPU passthrough with a single GPU (onboard automatically disables itself)

On an Ubuntu 18.04 system with two NVIDIA GPUs only the change to Grub below was necessary, I did not need to mess with modprobe or initramfs. In addition, I needed to set the PCI bus for the correct nvidia xconfig to the non-blacklisted GPU following these steps

sudo nvidia-xconfig

This will create a new xorg config at /etc/X11/xorg.conf.

Then add the BusID line in the following part of the file /etc/X11/xorg.conf:

csullivan

#!/bin/bash
function run() {
    device_id=$1
        echo "Starting rpc server on adb device: $device_id"
    preamble='
        spawn adb -s'
        body='shell
        expect "#"
    set cmd "cd /data/local/tmp/; LD_LIBRARY_PATH=/data/local/tmp /data/local/tmp/tvm_rpc server --tracker=IP_ADDRESS:PORT --key=\"your_key_here\""
        send $cmd

csullivan

import tvm
from tvm import te


def intrin_vadd(xo, m, n):
    x = te.placeholder((n,), name="vx")
    y = te.placeholder((n,), name="vy")
    if m % n == 0:
        body = lambda i: x[i] + y[i]
    else:

csullivan

1 OpenCL Platforms found

 CL_PLATFORM_NAME:      QUALCOMM Snapdragon(TM)
 CL_PLATFORM_VERSION:   OpenCL 2.0 QUALCOMM build: commit #ffa098c changeid #I3cb6626e76 Date: 01/16/20 Thu Local Branch:  Remote Branch:
OpenCL Device Info:

 1 devices found supporting OpenCL on: QUALCOMM Snapdragon(TM)

 ----------------------------------

csullivan

csullivan

import argparse
import os

def get_files(directory):
    files = os.listdir(path=directory)
    pbtxts = []
    for f in files:
        if "pbtxt" in f:
            pbtxts.append(f)
    return pbtxts

csullivan

diff --git a/src/ngraph/runtime/gpu/gpu_backend.cpp b/src/ngraph/runtime/gpu/gpu_backend.cpp
index 2f1472e4..f54d4ce0 100644
--- a/src/ngraph/runtime/gpu/gpu_backend.cpp
+++ b/src/ngraph/runtime/gpu/gpu_backend.cpp
@@ -162,6 +162,7 @@ bool runtime::gpu::GPU_Backend::call(shared_ptr<Function> func,
                                      const vector<shared_ptr<runtime::Tensor>>& outputs,
                                      const vector<shared_ptr<runtime::Tensor>>& inputs)
 {
+    std::cout << "call " << func->get_name() << std::endl;
     bool rc = true;

csullivan

diff --git a/src/ngraph/runtime/gpu/gpu_external_function.cpp b/src/ngraph/runtime/gpu/gpu_external_function.cpp
index 71cdd614..e836f16b 100644
--- a/src/ngraph/runtime/gpu/gpu_external_function.cpp
+++ b/src/ngraph/runtime/gpu/gpu_external_function.cpp
@@ -561,7 +561,7 @@ void runtime::gpu::GPU_ExternalFunction::compile()
         m_shared_context->m_primitive_emitter->get_memory_allocator());
 
     ngraph::pass::Manager pass_manager;
-#if CUDNN_VERSION >= 7200
+#if CUDNN_VERSION >= 9200

csullivan

diff --git a/src/ngraph/runtime/gpu/cuda_emitter.cpp b/src/ngraph/runtime/gpu/cuda_emitter.cpp
index a9ef0e00..09fe458b 100644
--- a/src/ngraph/runtime/gpu/cuda_emitter.cpp
+++ b/src/ngraph/runtime/gpu/cuda_emitter.cpp
@@ -3096,11 +3096,11 @@ void* runtime::gpu::CUDAEmitter::get_init_reduce_val(std::string reduce_op, std:
 {
     if (reduce_op == "max")
     {
-        return m_host_parameters->min_by_datatype(data_type);
+        return TypeInfo::Get(data_type)->max_ptr();