cuQRTM: A Cuda repository from kisityan

title	date	author	mathjax
cu$Q$-RTM Manual	2018-06-06	Yufeng Wang	true

Overview of cu$Q$-RTM package

cu-QRTM is a CUDA-based code package that implements $Q$-RTM based on a set of stable and efficient strategies, such as streamed CUFFT, checkpointing-assisted time-reversal reconstruction (CATRC) and adaptive stabilization scheme. This package is provided for accelerating conventional CPU-based $Q$-RTM, and mimicking how a geophysicist writes down a seismic processing modules such as modeling, imaging and inversion in the framework of the CPU-GPU heterogeneous computing platform. We provide two package versions: cuQRTM-Express for quick execution with 4 shots, which can be done within 3 min under single GPU card (GTX 760); cuQRTM-Standard for standard excution with 64 shots, which will take 10 min under 4 GPU cards (Tesla K10).

The architecture of cu$Q$-RTM package

input: accurate velocity and $Q$ model for $Q$-RTM:
- acc_vp.txt: quasi-Marmousi velocity model;
- acc_Qp.txt: quasi-Marmousi $Q$ model;
- ascii2bin.m: converting ASCII files to Binary files.
output: generated results such as seismograms, images of each shot and final stacked images. What we are most interested in is final images, wich includes:
- Final_image_cor_type0.dat: image from acoustic RTM;
- Final_image_cor_type1.dat: image from viscoacoustic RTM with compensation;
- Final_image_cor_type2.dat: image from $Q$-RTM using low-pass filtering;
- Final_image_cor_type3.dat: image from $Q$-RTM using adaptive stabilization;
plot: scripts for plotting figures, which includes:
- /madagascar/SConstruct: plot images, velocity and $Q$ models;
- /matlab/martrace: plot extracted trace from final migrated images for comparison.
Myfunctions.h: header file;
CUDAQRTM.cu: cuda code file;
QRTM.cpp: c++ code file, there are serveal important flags and parameters to control performance of $Q$-RTM, which includes:
- RTMtype: you can change this flag to generate different migrated images.

int RTMtype=0;		// RTMtype=0 for acoustic RTM
			// RTMtype=1 for viscoacoustic RTM without compensation
			// RTMtype=2 for QRTM using low-pass filtering
			// RTMtype=3 for QRTM using adaptive stabilization scheme

- `GPU_N` you can set GPU_N=n, where n denotes the number of GPU cards you will use.

int i,GPU_N;		// GPU_N stands for the total number of GPUs per node
getdevice(&GPU_N);	// Obtain the number of GPUs per node
printf("The available Device number is %d on %s\n",GPU_N,processor_name);

- `kx_cut` and `kz_cut`: these parameters are defined for low-pass filtering.

float kx_cut=3.0*2*PI*f0/vp_max;	// be careful to change this parameter
float kz_cut=3.0*2*PI*f0/vp_max;
float kx_cut_inv=3.0*2*PI*f0/vp_max;
float kz_cut_inv=3.0*2*PI*f0/vp_max;
float taper_ratio=0.2;				// taper ratio for turkey window filter

- `sigma` and `Order`: these two parameters are defined for adaptive stabilization,

float sigma=2.5e-3;	// be careful to change this parameter
int Order=1;		// defult

Makefile: excution script.

Prerequisites

cu-QRTM package is developed under Linux system, which should be equipped with the following environments:

CUDA environment (for example, -I/usr/local/cuda-8.0/include -L/usr/local/cuda-8.0/lib64);
MPI environment (for example, -I/home/wyf/intel/impi/5.0.1.035/intel64/include -L/home/wyf/intel/impi/5.0.1.035/intel64/lib/);
matlab;
madagascar.

How to run this package

If you want to quick test the package, please use fast version cuQRTM-Express; cuQRTM-Standard should be excuted on cluster with multi-GPUs (or you can excute on a sigle GPU card within an hour).

Step 1: Run the matlab file ascii2bin.m in ./input to convert the ASCII data into binary data;
Step 2: Confirm the environment in Makefile, and replace the folder path with your own enviroment path;

#! /bin/sh
# compiler
CC=nvcc
# set CUDA and MPI environment path 
INC=-I/usr/local/cuda-8.0/include -I/home/wyf/intel/impi/5.0.1.035/intel64/include 
LIB=-L/usr/local/cuda-8.0/lib64 -L/home/wyf/intel/impi/5.0.1.035/intel64/lib/ 
# set CUDA and MPI Dynamic link library
LINK= -lcudart -lcufft -lm -lmpich -lpthread -lrt -DMPICH_IGNORE_CXX_SEEK  -DMPICH_SKIP_MPICXX
# CUDA and C++ source codes
SOURCES=CUDAQRTM.cu QRTM.cpp
EXECNAME=QRTM
# Execution
all:
	$(CC) -v -o $(EXECNAME) $(SOURCES) $(INC) $(LIB) $(LINK)
	rm -f *.o 
	nohup mpirun -np 1 ./QRTM &

Step 3: Run the Makefile by the command line: make;
Step 4: View generated files in the folder ./ouput;

make
vi nohup
cd output/
ximage n1=234 < Final_image_cor_type0.dat hbox=300 &
ximage n1=234 < Final_image_cor_type1.dat hbox=300 &
ximage n1=234 < Final_image_cor_type2.dat hbox=300 &
ximage n1=234 < Final_image_cor_type3.dat hbox=300 &

Step 5: Plot figures by run /plot/madagascar/SConstruct and /plot/matlab/martrace.m.

scons view
cd Fig/
vpconvert format=pdf *.vpl

Contact me

I am Yufeng Wang, a PhD candidate from China University of Petroleum, Beijing. If you have any question about this coda package, please feel free to contact me by Email:hellowangyf@163.com.

Copyright

cu-QRTM is a CUDA-based code package that implements $Q$-RTM based on a set of stable and efficient strategies.

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.