31 skills found · Page 1 of 2
ncatlin / RgatAn instruction trace visualisation tool for dynamic program analysis
s7nfo / CirronCirron measures performance counters (instructions executed, etc.) and traces system calls a piece of Python or Ruby code executes.
ARM-software / Tarmac Trace UtilitiesTools for analyzing and browsing Tarmac instruction traces.
CYRUS-STUDIO / Frida Stalker使用 Frida Stalker 反 OLLVM 算法还原（函数调用分析、参数分析、打印调用堆栈、指令&寄存器Trace）。 Using Frida Stalker to Bypass OLLVM and Restore (Function Call Analysis, Parameter Analysis, Stack Trace Printing, Instruction & Registers Trace).
mibho / X64dbg Vmp Traceunorthodox approach to analyze a trace, but this helped me get comfy with x64 instructions overall (excluding sse/avx/etc lol), cleared up A LOT of misconceptions I had regarding VMP, and helped me not be as spooked as before about trying to use complex libs
projectbtle / ArgXtractTool to trace ARM Cortex-M assembly instructions and extract arguments to configuration APIs (supervisor calls or function calls).
scotty-kdw / ARM TracerTrace Log Generation (CLI) on a target device or emulator : Generating context information of every instruction from a specific point (input data) to a crash point
trishume / PerftraceTrace instruction execution using perf breakpoints in Python
mountainstorm / FlowFlow control tracer/debugger for OSX; provides a complete trace of all instructions executed by a process. Perfect if you want to know exactly what a process did/does
pulp-platform / Trace DebuggerCapture retired instructions of a RISC-V Core and compress them to a sequence of packets.
SOYJUN / Implement ODR ProtocolOverview For this assignment you will be developing and implementing : An On-Demand shortest-hop Routing (ODR) protocol for networks of fixed but arbitrary and unknown connectivity, using PF_PACKET sockets. The implementation is based on (a simplified version of) the AODV algorithm. Time client and server applications that send requests and replies to each other across the network using ODR. An API you will implement using Unix domain datagram sockets enables applications to communicate with the ODR mechanism running locally at their nodes. I shall be discussing the assignment in class on Wednesday, October 29, and Monday, November 3. The following should prove useful reference material for the assignment : Sections 15.1, 15.2, 15.4 & 15.6, Chapter 15, on Unix domain datagram sockets. PF_PACKET(7) from the Linux manual pages. You might find these notes made by a past CSE 533 student useful. Also, the following link http://www.pdbuchan.com/rawsock/rawsock.html contains useful code samples that use PF_PACKET sockets (as well as other code samples that use raw IP sockets which you do not need for this assignment, though you will be using these types of sockets for Assignment 4). Charles E. Perkins & Elizabeth M. Royer. “Ad-hoc On-Demand Distance Vector Routing.” Proceedings of the 2nd IEEE Workshop on Mobile Computing Systems and Applications, New Orleans, Louisiana, February 1999, pp. 90 - 100. The VMware environment minix.cs.stonybrook.edu is a Linux box running VMware. A cluster of ten Linux virtual machines, called vm1 through vm10, on which you can gain access as root and run your code have been created on minix. See VMware Environment Hosts for further details. VMware instructions takes you to a page that explains how to use the system. The ten virtual machines have been configured into a small virtual intranet of Ethernet LANs whose topology is (in principle) unknown to you. There is a course account cse533 on node minix, with home directory /users/cse533. In there, you will find a subdirectory Stevens/unpv13e , exactly as you are used to having on the cs system. You should develop your source code and makefiles for handing in accordingly. You will be handing in your source code on the minix node. Note that you do not need to link against the socket library (-lsocket) in Linux. The same is true for -lnsl and -lresolv. For example, take a look at how the LIBS variable is defined for Solaris, in /home/courses/cse533/Stevens/unpv13e_solaris2.10/Make.defines (on compserv1, say) : LIBS = ../libunp.a -lresolv -lsocket -lnsl -lpthread But if you take a look at Make.defines on minix (/users/cse533/Stevens/unpv13e/Make.defines) you will find only: LIBS = ../libunp.a -lpthread The nodes vm1 , . . . . . , vm10 are all multihomed : each has two (or more) interfaces. The interface ‘eth0 ’ should be completely ignored and is not to be used for this assignment (because it shows all ten nodes as if belonging to the same single Ethernet 192.168.1.0/24, rather than to an intranet composed of several Ethernets). Note that vm1 , . . . . . , vm10 are virtual machines, not real ones. One implication of this is that you will not be able to find out what their (virtual) IP addresses are by using nslookup and such. To find out these IP addresses, you need to look at the file /etc/hosts on minix. More to the point, invoking gethostbyname for a given vm will return to you only the (primary) IP address associated with the interface eth0 of that vm (which is the interface you will not be using). It will not return to you any other IP address for the node. Similarly, gethostbyaddr will return the vm node name only if you give it the (primary) IP address associated with the interface eth0 for the node. It will return nothing if you give it any other IP address for the node, even though the address is perfectly valid. Because of this, and because it will ease your task to be able to use gethostbyname and gethostbyaddr in a straightforward way, we shall adopt the (primary) IP addresses associated with interfaces eth0 as the ‘canonical’ IP addresses for the nodes (more on this below). Time client and server A time server runs on each of the ten vm machines. The client code should also be available on each vm so that it can be evoked at any of them. Normally, time clients/servers exchange request/reply messages using the TCP/UDP socket API that, effectively, enables them to receive service (indirectly, via the transport layer) from the local IP mechanism running at their nodes. You are to implement an API using Unix domain sockets to access the local ODR service directly (somewhat similar, in effect, to the way that raw sockets permit an application to access IP directly). Use Unix domain SOCK_DGRAM, rather than SOCK_STREAM, sockets (see Figures 15.5 & 15.6, pp. 418 - 419). API You need to implement a msg_send function that will be called by clients/servers to send requests/replies. The parameters of the function consist of : int giving the socket descriptor for write char* giving the ‘canonical’ IP address for the destination node, in presentation format int giving the destination ‘port’ number char* giving message to be sent int flag if set, force a route rediscovery to the destination node even if a non-‘stale’ route already exists (see below) msg_send will format these parameters into a single char sequence which is written to the Unix domain socket that a client/server process creates. The sequence will be read by the local ODR from a Unix domain socket that the ODR process creates for itself. Recall that the ‘canonical’ IP address for a vm node is the (primary) IP address associated with the eth0 interface for the node. It is what will be returned to you by a call to gethostbyname. Similarly, we need a msg_recv function which will do a (blocking) read on the application domain socket and return with : int giving socket descriptor for read char* giving message received char* giving ‘canonical’ IP address for the source node of message, in presentation format int* giving source ‘port’ number This information is written as a single char sequence by the ODR process to the domain socket that it creates for itself. It is read by msg_recv from the domain socket the client/server process creates, decomposed into the three components above, and returned to the caller of msg_recv. Also see the section below entitled ODR and the API. Client When a client is evoked at a node, it creates a domain datagram socket. The client should bind its socket to a ‘temporary’ (i.e., not ‘well-known’) sun_path name obtained from a call to tmpnam() (cf. line 10, Figure 15.6, p. 419) so that multiple clients may run at the same node. Note that tmpnam() is actually highly deprecated. You should use the mkstemp() function instead - look up the online man pages on minix (‘man mkstemp’) for details. As you run client code again and again during the development stage, the temporary files created by the calls to tmpnam / mkstemp start to proliferate since these files are not automatically removed when the client code terminates. You need to explicitly remove the file created by the client evocation by issuing a call to unlink() or to remove() in your client code just before the client code exits. See the online man pages on minix (‘man unlink’, ‘man remove’) for details. The client then enters an infinite loop repeating the steps below. The client prompts the user to choose one of vm1 , . . . . . , vm10 as a server node. Client msg_sends a 1 or 2 byte message to server and prints out on stdout the message client at node vm i1 sending request to server at vm i2 (In general, throughout this assignment, “trace” messages such as the one above should give the vm names and not IP addresses of the nodes.) Client then blocks in msg_recv awaiting response. This attempt to read from the domain socket should be backed up by a timeout in case no response ever comes. I leave it up to you whether you ‘wrap’ the call to msg_recv in a timeout, or you implement the timeout inside msg_recv itself. When the client receives a response it prints out on stdout the message client at node vm i1 : received from vm i2 <timestamp> If, on the other hand, the client times out, it should print out the message client at node vm i1 : timeout on response from vm i2 The client then retransmits the message out, setting the flag parameter in msg_send to force a route rediscovery, and prints out an appropriate message on stdout. This is done only once, when a timeout for a given message to the server occurs for the first time. Client repeats steps 1. - 3. Server The server creates a domain datagram socket. The server socket is assumed to have a (node-local) ‘well-known’ sun_path name which it binds to. This ‘well-known’ sun_path name is designated by a (network-wide) ‘well-known’ ‘port’ value. The time client uses this ‘port’ value to communicate with the server. The server enters an infinite sequence of calls to msg_recv followed by msg_send, awaiting client requests and responding to them. When it responds to a client request, it prints out on stdout the message server at node vm i1 responding to request from vm i2 ODR The ODR process runs on each of the ten vm machines. It is evoked with a single command line argument which gives a “staleness” time parameter, in seconds. It uses get_hw_addrs (available to you on minix in ~cse533/Asgn3_code) to obtain the index, and associated (unicast) IP and Ethernet addresses for each of the node’s interfaces, except for the eth0 and lo (loopback) interfaces, which should be ignored. In the subdirectory ~cse533/Asgn3_code (/users/cse533/Asgn3_code) on minix I am providing you with two functions, get_hw_addrs and prhwaddrs. These are analogous to the get_ifi_info_plus and prifinfo_plus of Assignment 2. Like get_ifi_info_plus, get_hw_addrs uses ioctl. get_hw_addrs gets the (primary) IP address, alias IP addresses (if any), HW address, and interface name and index value for each of the node's interfaces (including the loopback interface lo). prhwaddrs prints that information out. You should modify and use these functions as needed. Note that if an interface has no HW address associated with it (this is, typically, the case for the loopback interface lo for example), then ioctl returns get_hw_addrs a HW address which is the equivalent of 00:00:00:00:00:00 . get_hw_addrs stores this in the appropriate field of its data structures as it would with any HW address returned by ioctl, but when prhwaddrs comes across such an address, it prints a blank line instead of its usual ‘HWaddr = xx:xx:xx:xx:xx:xx’. The ODR process creates one or more PF_PACKET sockets. You will need to try out PF_PACKET sockets for yourselves and familiarize yourselves with how they behave. If, when you read from the socket and provide a sockaddr_ll structure, the kernel returns to you the index of the interface on which the incoming frame was received, then one socket will be enough. Otherwise, somewhat in the manner of Assignment 2, you shall have to create a PF_PACKET socket for every interface of interest (which are all the interfaces of the node, excluding interfaces lo and eth0 ), and bind a socket to each interface. Furthermore, if the kernel also returns to you the source Ethernet address of the frame in the sockaddr_ll structure, then you can make do with SOCK_DGRAM type PF_PACKET sockets; otherwise you shall have to use SOCK_RAW type sockets (although I would prefer you to use SOCK_RAW type sockets anyway, even if it turns out you can make do with SOCK_DGRAM type). The socket(s) should have a protocol value (no larger than 0xffff so that it fits in two bytes; this value is given as a network-byte-order parameter in the call(s) to function socket) that identifies your ODR protocol. The <linux/if_ether.h> include file (i.e., the file /usr/include/linux/if_ether.h) contains protocol values defined for the standard protocols typically found on an Ethernet LAN, as well as other values such as ETH_P_ALL. You should set protocol to a value of your choice which is not a <linux/if_ether.h> value, but which is, hopefully, unique to yourself. Remember that you will all be running your code using the same root account on the vm1 , . . . . . , vm10 nodes. So if two of you happen to choose the same protocol value and happen to be running on the same vm node at the same time, your applications will receive each other’s frames. For that reason, try to choose a protocol value for the socket(s) that is likely to be unique to yourself (something based on your Stony Brook student ID number, for example). This value effectively becomes the protocol value for your implementation of ODR, as opposed to some other cse 533 student's implementation. Because your value of protocol is to be carried in the frame type field of the Ethernet frame header, the value chosen should be not less than 1536 (0x600) so that it is not misinterpreted as the length of an Ethernet 802.3 frame. Note from the man pages for packet(7) that frames are passed to and from the socket without any processing in the frame content by the device driver on the other side of the socket, except for calculating and tagging on the 4-byte CRC trailer for outgoing frames, and stripping that trailer before delivering incoming frames to the socket. Nevertheless, if you write a frame that is less than 60 bytes, the necessary padding is automatically added by the device driver so that the frame that is actually transmitted out is the minimum Ethernet size of 64 bytes. When reading from the socket, however, any such padding that was introduced into a short frame at the sending node to bring it up to the minimum frame size is not stripped off - it is included in what you receive from the socket (thus, the minimum number of bytes you receive should never be less than 60). Also, you will have to build the frame header for outgoing frames yourselves (assuming you use SOCK_RAW type sockets). Bear in mind that the field values in that header have to be in network order. The ODR process also creates a domain datagram socket for communication with application processes at the node, and binds the socket to a ‘well known’ sun_path name for the ODR service. Because it is dealing with fixed topologies, ODR is, by and large, considerably simpler than AODV. In particular, discovered routes are relatively stable and there is no need for all the paraphernalia that goes with the possibility of routes changing (such as maintenance of active nodes in the routing tables and timeout mechanisms; timeouts on reverse links; lifetime field in the RREP messages; etc.) Nor will we be implementing source_sequence_#s (in the RREQ messages), and dest_sequence_# (in RREQ and RREP messages). In reality, we should (though we will not, for the sake of simplicity, be doing so) implement some sort of sequence number mechanism, or some alternative mechanism such as split-horizon for example, if we are to avoid possible scenarios of routing loops in a “count to infinity” context (I shall explain this point in class). However, we want ODR to discover shortest-hop paths, and we want it to do so in a reasonably efficient manner. This necessitates having one or two aspects of its operations work in a different, possibly slightly more complicated, way than AODV does. ODR has several basic responsibilities : Build and maintain a routing table. For each destination in the table, the routing table structure should include, at a minimum, the next-hop node (in the form of the Ethernet address for that node) and outgoing interface index, the number of hops to the destination, and a timestamp of when the the routing table entry was made or last “reconfirmed” / updated. Note that a destination node in the table is to be identified only by its ‘canonical’ IP address, and not by any other IP addresses the node has. Generate a RREQ in response to a time client calling msg_send for a destination for which ODR has no route (or for which a route exists, but msg_send has the flag parameter set or the route has gone ‘stale’ – see below), and ‘flood’ the RREQ out on all the node’s interfaces (except for the interface it came in on and, of course, the interfaces eth0 and lo). Flooding is done using an Ethernet broadcast destination address (0xff:ff:ff:ff:ff:ff) in the outgoing frame header. Note that a copy of the broadcast packet is supposed to / might be looped back to the node that sends it (see p. 535 in the Stevens textbook). ODR will have to take care not to treat these copies as new incoming RREQs. Also note that ODR at the client node increments the broadcast_id every time it issues a new RREQ for any destination node. When a RREQ is received, ODR has to generate a RREP if it is at the destination node, or if it is at an intermediate node that happens to have a route (which is not ‘stale’ – see below) to the destination. Otherwise, it must propagate the RREQ by flooding it out on all the node’s interfaces (except the interface the RREQ arrived on). Note that as it processes received RREQs, ODR should enter the ‘reverse’ route back to the source node into its routing table, or update an existing entry back to the source node if the RREQ received shows a shorter-hop route, or a route with the same number of hops but going through a different neighbour. The timestamp associated with the table entry should be updated whenever an existing route is either “reconfirmed” or updated. Obviously, if the node is going to generate a RREP, updating an existing entry back to the source node with a more efficient route, or a same-hops route using a different neighbour, should be done before the RREP is generated. Unlike AODV, when an intermediate node receives a RREQ for which it generates a RREP, it should nevertheless continue to flood the RREQ it received if the RREQ pertains to a source node whose existence it has heretofore been unaware of, or the RREQ gives it a more efficient route than it knew of back to the source node (the reason for continuing to flood the RREQ is so that other nodes in the intranet also become aware of the existence of the source node or of the potentially more optimal reverse route to it, and update their tables accordingly). However, since an RREP for this RREQ is being sent by our node, we do not want other nodes who receive the RREQ propagated by our node, and who might be in a position to do so, to also send RREPs. So we need to introduce a field in the RREQ message, not present in the AODV specifications, which acts like a “RREP already sent” field. Our node sets this field before further propagating the RREQ and nodes receiving an RREQ with this field set do not send RREPs in response, even if they are in a position to do so. ODR may, of course, receive multiple, distinct instances of the same RREQ (the combination of source_addr and broadcast_id uniquely identifies the RREQ). Such RREQs should not be flooded out unless they have a lower hop count than instances of that RREQ that had previously been received. By the same token, if ODR is in a position to send out a RREP, and has already done so for this, now repeating, RREQ , it should not send out another RREP unless the RREQ shows a more efficient, previously unknown, reverse route back to the source node. In other words, ODR should not generate essentially duplicative RREPs, nor generate RREPs to instances of RREQs that reflect reverse routes to the source that are not more efficient than what we already have. Relay RREPs received back to the source node (this is done using the ‘reverse’ route entered into the routing table when the corresponding RREQ was processed). At the same time, a ‘forward’ path to the destination is entered into the routing table. ODR could receive multiple, distinct RREPs for the same RREQ. The ‘forward’ route entered in the routing table should be updated to reflect the shortest-hop route to the destination, and RREPs reflecting suboptimal routes should not be relayed back to the source. In general, maintaining a route and its associated timestamp in the table in response to RREPs received is done in the same manner described above for RREQs. Forward time client/server messages along the next hop. (The following is important – you will lose points if you do not implement it.) Note that such application payload messages (especially if they are the initial request from the client to the server, rather than the server response back to the client) can be like “free” RREPs, enabling nodes along the path from source (client) to destination (server) node to build a reverse path back to the client node whose existence they were heretofore unaware of (or, possibly, to update an existing route with a more optimal one). Before it forwards an application payload message along the next hop, ODR at an intermediate node (and also at the final destination node) should use the message to update its routing table in this way. Thus, calls to msg_send by time servers should never cause ODR at the server node to initiate RREQs, since the receipt of a time client request implies that a route back to the client node should now exist in the routing table. The only exception to this is if the server node has a staleness parameter of zero (see below). A routing table entry has associated with it a timestamp that gives the time the entry was made into the table. When a client at a node calls msg_send, and if an entry for the destination node already exists in the routing table, ODR first checks that the routing information is not ‘stale’. A stale routing table entry is one that is older than the value defined by the staleness parameter given as a command line argument to the ODR process when it is executed. ODR deletes stale entries (as well as non-stale entries when the flag parameter in msg_send is set) and initiates a route rediscovery by issuing a RREQ for the destination node. This will force periodic updating of the routing tables to take care of failed nodes along the current path, Ethernet addresses that might have changed, and so on. Similarly, as RREQs propagate through the intranet, existing stale table entries at intermediate nodes are deleted and new route discoveries propagated. As noted above when discussing the processing of RREQs and RREPs, the associated timestamp for an existing table entry is updated in response to having the route either “reconfirmed” or updated (this applies to both reverse routes, by virtue of RREQs received, and to forward routes, by virtue of RREPs). Finally, note that a staleness parameter of 0 essentially indicates that the discovered route will be used only once, when first discovered, and then discarded. Effectively, an ODR with staleness parameter 0 maintains no real routing table at all ; instead, it forces route discoveries at every step of its operation. As a practical matter, ODR should be run with staleness parameter values that are considerably larger than the longest RTT on the intranet, otherwise performance will degrade considerably (and collapse entirely as the parameter values approach 0). Nevertheless, for robustness, we need to implement a mechanism by which an intermediate node that receives a RREP or application payload message for forwarding and finds that its relevant routing table entry has since gone stale, can intiate a RREQ to rediscover the route it needs. RREQ, RREP, and time client/server request/response messages will all have to be carried as encapsulated ODR protocol messages that form the data payload of Ethernet frames. So we need to design the structure of ODR protocol messages. The format should contain a type field (0 for RREQ, 1 for RREP, 2 for application payload ). The remaining fields in an ODR message will depend on what type it is. The fields needed for (our simplified versions of AODV’s) RREQ and RREP should be fairly clear to you, but keep in mind that you need to introduce two extra fields: The “RREP already sent” bit or field in RREQ messages, as mentioned above. A “forced discovery” bit or field in both RREQ and RREP messages: When a client application forces route rediscovery, this bit should be set in the RREQ issued by the client node ODR. Intermediate nodes that are not the destination node but which do have a route to the destination node should not respond with RREPs to an RREQ which has the forced discovery field set. Instead, they should continue to flood the RREQ so that it eventually reaches the destination node which will then respond with an RREP. The intermediate nodes relaying such an RREQ must update their ‘reverse’ route back to the source node accordingly, even if the new route is less efficient (i.e., has more hops) than the one they currently have in their routing table. The destination node responds to the RREQ with an RREP in which this field is also set. Intermediate nodes that receive such a forced discovery RREP must update their ‘forward’ route to the destination node accordingly, even if the new route is less efficient (i.e., has more hops) than the one they currently have in their routing table. This behaviour will cause a forced discovery RREQ to be responded to only by the destination node itself and not any other node, and will cause intermediate nodes to update their routing tables to both source and destination nodes in accordance with the latest routing information received, to cover the possibility that older routes are no longer valid because nodes and/or links along their paths have gone down. A type 2, application payload, message needs to contain the following type of information : type = 2 ‘canonical’ IP address of source node ‘port’ number of source application process (This, of course, is not a real port number in the TCP/UDP sense, but simply a value that ODR at the source node uses to designate the sun_path name for the source application’s domain socket.) ‘canonical’ IP address of destination node ‘port’ number of destination application process (This is passed to ODR by the application process at the source node when it calls msg_send. Its designates the sun_path name for an application’s domain socket at the destination node.) hop count (This starts at 0 and is incremented by 1 at each hop so that ODR can make use of the message to update its routing table, as discussed above.) number of bytes in application message The fields above essentially constitute a ‘header’ for the ODR message. Note that fields which you choose to have carry numeric values (rather than ascii characters, for example) must be in network byte order. ODR-defined numeric-valued fields in type 0, RREQ, and type 1, RREP, messages must, of course, also be in network byte order. Also note that only the ‘canonical’ IP addresses are used for the source and destination nodes in the ODR header. The same has to be true in the headers for type 0, RREQ, and type 1, RREP, messages. The general rule is that ODR messages only carry ‘canonical’ IP node addresses. The last field in the type 2 ODR message is essentially the data payload of the message. application message given in the call to msg_send An ODR protocol message is encapsulated as the data payload of an Ethernet frame whose header it fills in as follows : source address = Ethernet address of outgoing interface of the current node where ODR is processing the message. destination address = Ethernet broadcast address for type 0 messages; Ethernet address of next hop node for type 1 & 2 messages. protocol field = protocol value for the ODR PF_PACKET socket(s). Last but not least, whenever ODR writes an Ethernet frame out through its socket, it prints out on stdout the message ODR at node vm i1 : sending frame hdr src vm i1 dest addr ODR msg type n src vm i2 dest vm i3 where addr is in presentation format (i.e., hexadecimal xx:xx:xx:xx:xx:xx) and gives the destination Ethernet address in the outgoing frame header. Other nodes in the message should be identified by their vm name. A message should be printed out for each packet sent out on a distinct interface. ODR and the API When the ODR process first starts, it must construct a table in which it enters all well-known ‘port’ numbers and their corresponding sun_path names. These will constitute permanent entries in the table. Thereafter, whenever it reads a message off its domain socket, it must obtain the sun_path name for the peer process socket and check whether that name is entered in the table. If not, it must select an ‘ephemeral’ ‘port’ value by which to designate the peer sun_path name and enter the pair < port value , sun_path name > into the table. Such entries cannot be permanent otherwise the table will grow unboundedly in time, with entries surviving for ever, beyond the peer processes’ demise. We must associate a time_to_live field with a non-permanent table entry, and purge the entry if nothing is heard from the peer for that amount of time. Every time a peer process for which a non-permanent table entry exists communicates with ODR, its time_to_live value should be reinitialized. Note that when ODR writes to a peer, it is possible for the write to fail because the peer does not exist : it could be a ‘well-known’ service that is not running, or we could be in the interval between a process with a non-permanent table entry terminating and the expiration of its time_to_live value. Notes A proper implementation of ODR would probably require that RREQ and RREP messages be backed up by some kind of timeout and retransmission mechanism since the network transmission environment is not reliable. This would considerably complicate the implementation (because at any given moment, a node could have multiple RREQs that it has flooded out, but for which it has still not received RREPs; the situation is further complicated by the fact that not all intermediate nodes receiving and relaying RREQs necessarily lie on a path to the destination, and therefore should expect to receive RREPs), and, learning-wise, would not add much to the experience you should have gained from Assignment 2.
pulp-platform / TrdbRISC-V Processor Tracing tools and library
Aryia-Behroziuan / NumpyQuickstart tutorial Prerequisites Before reading this tutorial you should know a bit of Python. If you would like to refresh your memory, take a look at the Python tutorial. If you wish to work the examples in this tutorial, you must also have some software installed on your computer. Please see https://scipy.org/install.html for instructions. Learner profile This tutorial is intended as a quick overview of algebra and arrays in NumPy and want to understand how n-dimensional (n>=2) arrays are represented and can be manipulated. In particular, if you don’t know how to apply common functions to n-dimensional arrays (without using for-loops), or if you want to understand axis and shape properties for n-dimensional arrays, this tutorial might be of help. Learning Objectives After this tutorial, you should be able to: Understand the difference between one-, two- and n-dimensional arrays in NumPy; Understand how to apply some linear algebra operations to n-dimensional arrays without using for-loops; Understand axis and shape properties for n-dimensional arrays. The Basics NumPy’s main object is the homogeneous multidimensional array. It is a table of elements (usually numbers), all of the same type, indexed by a tuple of non-negative integers. In NumPy dimensions are called axes. For example, the coordinates of a point in 3D space [1, 2, 1] has one axis. That axis has 3 elements in it, so we say it has a length of 3. In the example pictured below, the array has 2 axes. The first axis has a length of 2, the second axis has a length of 3. [[ 1., 0., 0.], [ 0., 1., 2.]] NumPy’s array class is called ndarray. It is also known by the alias array. Note that numpy.array is not the same as the Standard Python Library class array.array, which only handles one-dimensional arrays and offers less functionality. The more important attributes of an ndarray object are: ndarray.ndim the number of axes (dimensions) of the array. ndarray.shape the dimensions of the array. This is a tuple of integers indicating the size of the array in each dimension. For a matrix with n rows and m columns, shape will be (n,m). The length of the shape tuple is therefore the number of axes, ndim. ndarray.size the total number of elements of the array. This is equal to the product of the elements of shape. ndarray.dtype an object describing the type of the elements in the array. One can create or specify dtype’s using standard Python types. Additionally NumPy provides types of its own. numpy.int32, numpy.int16, and numpy.float64 are some examples. ndarray.itemsize the size in bytes of each element of the array. For example, an array of elements of type float64 has itemsize 8 (=64/8), while one of type complex32 has itemsize 4 (=32/8). It is equivalent to ndarray.dtype.itemsize. ndarray.data the buffer containing the actual elements of the array. Normally, we won’t need to use this attribute because we will access the elements in an array using indexing facilities. An example >>> import numpy as np a = np.arange(15).reshape(3, 5) a array([[ 0, 1, 2, 3, 4], [ 5, 6, 7, 8, 9], [10, 11, 12, 13, 14]]) a.shape (3, 5) a.ndim 2 a.dtype.name 'int64' a.itemsize 8 a.size 15 type(a) <class 'numpy.ndarray'> b = np.array([6, 7, 8]) b array([6, 7, 8]) type(b) <class 'numpy.ndarray'> Array Creation There are several ways to create arrays. For example, you can create an array from a regular Python list or tuple using the array function. The type of the resulting array is deduced from the type of the elements in the sequences. >>> >>> import numpy as np >>> a = np.array([2,3,4]) >>> a array([2, 3, 4]) >>> a.dtype dtype('int64') >>> b = np.array([1.2, 3.5, 5.1]) >>> b.dtype dtype('float64') A frequent error consists in calling array with multiple arguments, rather than providing a single sequence as an argument. >>> >>> a = np.array(1,2,3,4) # WRONG Traceback (most recent call last): ... TypeError: array() takes from 1 to 2 positional arguments but 4 were given >>> a = np.array([1,2,3,4]) # RIGHT array transforms sequences of sequences into two-dimensional arrays, sequences of sequences of sequences into three-dimensional arrays, and so on. >>> >>> b = np.array([(1.5,2,3), (4,5,6)]) >>> b array([[1.5, 2. , 3. ], [4. , 5. , 6. ]]) The type of the array can also be explicitly specified at creation time: >>> >>> c = np.array( [ [1,2], [3,4] ], dtype=complex ) >>> c array([[1.+0.j, 2.+0.j], [3.+0.j, 4.+0.j]]) Often, the elements of an array are originally unknown, but its size is known. Hence, NumPy offers several functions to create arrays with initial placeholder content. These minimize the necessity of growing arrays, an expensive operation. The function zeros creates an array full of zeros, the function ones creates an array full of ones, and the function empty creates an array whose initial content is random and depends on the state of the memory. By default, the dtype of the created array is float64. >>> >>> np.zeros((3, 4)) array([[0., 0., 0., 0.], [0., 0., 0., 0.], [0., 0., 0., 0.]]) >>> np.ones( (2,3,4), dtype=np.int16 ) # dtype can also be specified array([[[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]], [[1, 1, 1, 1], [1, 1, 1, 1], [1, 1, 1, 1]]], dtype=int16) >>> np.empty( (2,3) ) # uninitialized array([[ 3.73603959e-262, 6.02658058e-154, 6.55490914e-260], # may vary [ 5.30498948e-313, 3.14673309e-307, 1.00000000e+000]]) To create sequences of numbers, NumPy provides the arange function which is analogous to the Python built-in range, but returns an array. >>> >>> np.arange( 10, 30, 5 ) array([10, 15, 20, 25]) >>> np.arange( 0, 2, 0.3 ) # it accepts float arguments array([0. , 0.3, 0.6, 0.9, 1.2, 1.5, 1.8]) When arange is used with floating point arguments, it is generally not possible to predict the number of elements obtained, due to the finite floating point precision. For this reason, it is usually better to use the function linspace that receives as an argument the number of elements that we want, instead of the step: >>> >>> from numpy import pi >>> np.linspace( 0, 2, 9 ) # 9 numbers from 0 to 2 array([0. , 0.25, 0.5 , 0.75, 1. , 1.25, 1.5 , 1.75, 2. ]) >>> x = np.linspace( 0, 2*pi, 100 ) # useful to evaluate function at lots of points >>> f = np.sin(x) See also array, zeros, zeros_like, ones, ones_like, empty, empty_like, arange, linspace, numpy.random.Generator.rand, numpy.random.Generator.randn, fromfunction, fromfile Printing Arrays When you print an array, NumPy displays it in a similar way to nested lists, but with the following layout: the last axis is printed from left to right, the second-to-last is printed from top to bottom, the rest are also printed from top to bottom, with each slice separated from the next by an empty line. One-dimensional arrays are then printed as rows, bidimensionals as matrices and tridimensionals as lists of matrices. >>> >>> a = np.arange(6) # 1d array >>> print(a) [0 1 2 3 4 5] >>> >>> b = np.arange(12).reshape(4,3) # 2d array >>> print(b) [[ 0 1 2] [ 3 4 5] [ 6 7 8] [ 9 10 11]] >>> >>> c = np.arange(24).reshape(2,3,4) # 3d array >>> print(c) [[[ 0 1 2 3] [ 4 5 6 7] [ 8 9 10 11]] [[12 13 14 15] [16 17 18 19] [20 21 22 23]]] See below to get more details on reshape. If an array is too large to be printed, NumPy automatically skips the central part of the array and only prints the corners: >>> >>> print(np.arange(10000)) [ 0 1 2 ... 9997 9998 9999] >>> >>> print(np.arange(10000).reshape(100,100)) [[ 0 1 2 ... 97 98 99] [ 100 101 102 ... 197 198 199] [ 200 201 202 ... 297 298 299] ... [9700 9701 9702 ... 9797 9798 9799] [9800 9801 9802 ... 9897 9898 9899] [9900 9901 9902 ... 9997 9998 9999]] To disable this behaviour and force NumPy to print the entire array, you can change the printing options using set_printoptions. >>> >>> np.set_printoptions(threshold=sys.maxsize) # sys module should be imported Basic Operations Arithmetic operators on arrays apply elementwise. A new array is created and filled with the result. >>> >>> a = np.array( [20,30,40,50] ) >>> b = np.arange( 4 ) >>> b array([0, 1, 2, 3]) >>> c = a-b >>> c array([20, 29, 38, 47]) >>> b**2 array([0, 1, 4, 9]) >>> 10*np.sin(a) array([ 9.12945251, -9.88031624, 7.4511316 , -2.62374854]) >>> a<35 array([ True, True, False, False]) Unlike in many matrix languages, the product operator * operates elementwise in NumPy arrays. The matrix product can be performed using the @ operator (in python >=3.5) or the dot function or method: >>> >>> A = np.array( [[1,1], ... [0,1]] ) >>> B = np.array( [[2,0], ... [3,4]] ) >>> A * B # elementwise product array([[2, 0], [0, 4]]) >>> A @ B # matrix product array([[5, 4], [3, 4]]) >>> A.dot(B) # another matrix product array([[5, 4], [3, 4]]) Some operations, such as += and *=, act in place to modify an existing array rather than create a new one. >>> >>> rg = np.random.default_rng(1) # create instance of default random number generator >>> a = np.ones((2,3), dtype=int) >>> b = rg.random((2,3)) >>> a *= 3 >>> a array([[3, 3, 3], [3, 3, 3]]) >>> b += a >>> b array([[3.51182162, 3.9504637 , 3.14415961], [3.94864945, 3.31183145, 3.42332645]]) >>> a += b # b is not automatically converted to integer type Traceback (most recent call last): ... numpy.core._exceptions.UFuncTypeError: Cannot cast ufunc 'add' output from dtype('float64') to dtype('int64') with casting rule 'same_kind' When operating with arrays of different types, the type of the resulting array corresponds to the more general or precise one (a behavior known as upcasting). >>> >>> a = np.ones(3, dtype=np.int32) >>> b = np.linspace(0,pi,3) >>> b.dtype.name 'float64' >>> c = a+b >>> c array([1. , 2.57079633, 4.14159265]) >>> c.dtype.name 'float64' >>> d = np.exp(c*1j) >>> d array([ 0.54030231+0.84147098j, -0.84147098+0.54030231j, -0.54030231-0.84147098j]) >>> d.dtype.name 'complex128' Many unary operations, such as computing the sum of all the elements in the array, are implemented as methods of the ndarray class. >>> >>> a = rg.random((2,3)) >>> a array([[0.82770259, 0.40919914, 0.54959369], [0.02755911, 0.75351311, 0.53814331]]) >>> a.sum() 3.1057109529998157 >>> a.min() 0.027559113243068367 >>> a.max() 0.8277025938204418 By default, these operations apply to the array as though it were a list of numbers, regardless of its shape. However, by specifying the axis parameter you can apply an operation along the specified axis of an array: >>> >>> b = np.arange(12).reshape(3,4) >>> b array([[ 0, 1, 2, 3], [ 4, 5, 6, 7], [ 8, 9, 10, 11]]) >>> >>> b.sum(axis=0) # sum of each column array([12, 15, 18, 21]) >>> >>> b.min(axis=1) # min of each row array([0, 4, 8]) >>> >>> b.cumsum(axis=1) # cumulative sum along each row array([[ 0, 1, 3, 6], [ 4, 9, 15, 22], [ 8, 17, 27, 38]]) Universal Functions NumPy provides familiar mathematical functions such as sin, cos, and exp. In NumPy, these are called “universal functions”(ufunc). Within NumPy, these functions operate elementwise on an array, producing an array as output. >>> >>> B = np.arange(3) >>> B array([0, 1, 2]) >>> np.exp(B) array([1. , 2.71828183, 7.3890561 ]) >>> np.sqrt(B) array([0. , 1. , 1.41421356]) >>> C = np.array([2., -1., 4.]) >>> np.add(B, C) array([2., 0., 6.]) See also all, any, apply_along_axis, argmax, argmin, argsort, average, bincount, ceil, clip, conj, corrcoef, cov, cross, cumprod, cumsum, diff, dot, floor, inner, invert, lexsort, max, maximum, mean, median, min, minimum, nonzero, outer, prod, re, round, sort, std, sum, trace, transpose, var, vdot, vectorize, where Indexing, Slicing and Iterating One-dimensional arrays can be indexed, sliced and iterated over, much like lists and other Python sequences. >>> >>> a = np.arange(10)**3 >>> a array([ 0, 1, 8, 27, 64, 125, 216, 343, 512, 729]) >>> a[2] 8 >>> a[2:5] array([ 8, 27, 64]) # equivalent to a[0:6:2] = 1000; # from start to position 6, exclusive, set every 2nd element to 1000 >>> a[:6:2] = 1000 >>> a array([1000, 1, 1000, 27, 1000, 125, 216, 343, 512, 729]) >>> a[ : :-1] # reversed a array([ 729, 512, 343, 216, 125, 1000, 27, 1000, 1, 1000]) >>> for i in a: ... print(i**(1/3.)) ... 9.999999999999998 1.0 9.999999999999998 3.0 9.999999999999998 4.999999999999999 5.999999999999999 6.999999999999999 7.999999999999999 8.999999999999998 Multidimensional arrays can have one index per axis. These indices are given in a tuple separated by commas: >>> >>> def f(x,y): ... return 10*x+y ... >>> b = np.fromfunction(f,(5,4),dtype=int) >>> b array([[ 0, 1, 2, 3], [10, 11, 12, 13], [20, 21, 22, 23], [30, 31, 32, 33], [40, 41, 42, 43]]) >>> b[2,3] 23 >>> b[0:5, 1] # each row in the second column of b array([ 1, 11, 21, 31, 41]) >>> b[ : ,1] # equivalent to the previous example array([ 1, 11, 21, 31, 41]) >>> b[1:3, : ] # each column in the second and third row of b array([[10, 11, 12, 13], [20, 21, 22, 23]]) When fewer indices are provided than the number of axes, the missing indices are considered complete slices: >>> >>> b[-1] # the last row. Equivalent to b[-1,:] array([40, 41, 42, 43]) The expression within brackets in b[i] is treated as an i followed by as many instances of : as needed to represent the remaining axes. NumPy also allows you to write this using dots as b[i,...]. The dots (...) represent as many colons as needed to produce a complete indexing tuple. For example, if x is an array with 5 axes, then x[1,2,...] is equivalent to x[1,2,:,:,:], x[...,3] to x[:,:,:,:,3] and x[4,...,5,:] to x[4,:,:,5,:]. >>> >>> c = np.array( [[[ 0, 1, 2], # a 3D array (two stacked 2D arrays) ... [ 10, 12, 13]], ... [[100,101,102], ... [110,112,113]]]) >>> c.shape (2, 2, 3) >>> c[1,...] # same as c[1,:,:] or c[1] array([[100, 101, 102], [110, 112, 113]]) >>> c[...,2] # same as c[:,:,2] array([[ 2, 13], [102, 113]]) Iterating over multidimensional arrays is done with respect to the first axis: >>> >>> for row in b: ... print(row) ... [0 1 2 3] [10 11 12 13] [20 21 22 23] [30 31 32 33] [40 41 42 43] However, if one wants to perform an operation on each element in the array, one can use the flat attribute which is an iterator over all the elements of the array: >>> >>> for element in b.flat: ... print(element) ... 0 1 2 3 10 11 12 13 20 21 22 23 30 31 32 33 40 41 42 43 See also Indexing, Indexing (reference), newaxis, ndenumerate, indices Shape Manipulation Changing the shape of an array An array has a shape given by the number of elements along each axis: >>> >>> a = np.floor(10*rg.random((3,4))) >>> a array([[3., 7., 3., 4.], [1., 4., 2., 2.], [7., 2., 4., 9.]]) >>> a.shape (3, 4) The shape of an array can be changed with various commands. Note that the following three commands all return a modified array, but do not change the original array: >>> >>> a.ravel() # returns the array, flattened array([3., 7., 3., 4., 1., 4., 2., 2., 7., 2., 4., 9.]) >>> a.reshape(6,2) # returns the array with a modified shape array([[3., 7.], [3., 4.], [1., 4.], [2., 2.], [7., 2.], [4., 9.]]) >>> a.T # returns the array, transposed array([[3., 1., 7.], [7., 4., 2.], [3., 2., 4.], [4., 2., 9.]]) >>> a.T.shape (4, 3) >>> a.shape (3, 4) The order of the elements in the array resulting from ravel() is normally “C-style”, that is, the rightmost index “changes the fastest”, so the element after a[0,0] is a[0,1]. If the array is reshaped to some other shape, again the array is treated as “C-style”. NumPy normally creates arrays stored in this order, so ravel() will usually not need to copy its argument, but if the array was made by taking slices of another array or created with unusual options, it may need to be copied. The functions ravel() and reshape() can also be instructed, using an optional argument, to use FORTRAN-style arrays, in which the leftmost index changes the fastest. The reshape function returns its argument with a modified shape, whereas the ndarray.resize method modifies the array itself: >>> >>> a array([[3., 7., 3., 4.], [1., 4., 2., 2.], [7., 2., 4., 9.]]) >>> a.resize((2,6)) >>> a array([[3., 7., 3., 4., 1., 4.], [2., 2., 7., 2., 4., 9.]]) If a dimension is given as -1 in a reshaping operation, the other dimensions are automatically calculated: >>> >>> a.reshape(3,-1) array([[3., 7., 3., 4.], [1., 4., 2., 2.], [7., 2., 4., 9.]]) See also ndarray.shape, reshape, resize, ravel Stacking together different arrays Several arrays can be stacked together along different axes: >>> >>> a = np.floor(10*rg.random((2,2))) >>> a array([[9., 7.], [5., 2.]]) >>> b = np.floor(10*rg.random((2,2))) >>> b array([[1., 9.], [5., 1.]]) >>> np.vstack((a,b)) array([[9., 7.], [5., 2.], [1., 9.], [5., 1.]]) >>> np.hstack((a,b)) array([[9., 7., 1., 9.], [5., 2., 5., 1.]]) The function column_stack stacks 1D arrays as columns into a 2D array. It is equivalent to hstack only for 2D arrays: >>> >>> from numpy import newaxis >>> np.column_stack((a,b)) # with 2D arrays array([[9., 7., 1., 9.], [5., 2., 5., 1.]]) >>> a = np.array([4.,2.]) >>> b = np.array([3.,8.]) >>> np.column_stack((a,b)) # returns a 2D array array([[4., 3.], [2., 8.]]) >>> np.hstack((a,b)) # the result is different array([4., 2., 3., 8.]) >>> a[:,newaxis] # view `a` as a 2D column vector array([[4.], [2.]]) >>> np.column_stack((a[:,newaxis],b[:,newaxis])) array([[4., 3.], [2., 8.]]) >>> np.hstack((a[:,newaxis],b[:,newaxis])) # the result is the same array([[4., 3.], [2., 8.]]) On the other hand, the function row_stack is equivalent to vstack for any input arrays. In fact, row_stack is an alias for vstack: >>> >>> np.column_stack is np.hstack False >>> np.row_stack is np.vstack True In general, for arrays with more than two dimensions, hstack stacks along their second axes, vstack stacks along their first axes, and concatenate allows for an optional arguments giving the number of the axis along which the concatenation should happen. Note In complex cases, r_ and c_ are useful for creating arrays by stacking numbers along one axis. They allow the use of range literals (“:”) >>> >>> np.r_[1:4,0,4] array([1, 2, 3, 0, 4]) When used with arrays as arguments, r_ and c_ are similar to vstack and hstack in their default behavior, but allow for an optional argument giving the number of the axis along which to concatenate. See also hstack, vstack, column_stack, concatenate, c_, r_ Splitting one array into several smaller ones Using hsplit, you can split an array along its horizontal axis, either by specifying the number of equally shaped arrays to return, or by specifying the columns after which the division should occur: >>> >>> a = np.floor(10*rg.random((2,12))) >>> a array([[6., 7., 6., 9., 0., 5., 4., 0., 6., 8., 5., 2.], [8., 5., 5., 7., 1., 8., 6., 7., 1., 8., 1., 0.]]) # Split a into 3 >>> np.hsplit(a,3) [array([[6., 7., 6., 9.], [8., 5., 5., 7.]]), array([[0., 5., 4., 0.], [1., 8., 6., 7.]]), array([[6., 8., 5., 2.], [1., 8., 1., 0.]])] # Split a after the third and the fourth column >>> np.hsplit(a,(3,4)) [array([[6., 7., 6.], [8., 5., 5.]]), array([[9.], [7.]]), array([[0., 5., 4., 0., 6., 8., 5., 2.], [1., 8., 6., 7., 1., 8., 1., 0.]])] vsplit splits along the vertical axis, and array_split allows one to specify along which axis to split. Copies and Views When operating and manipulating arrays, their data is sometimes copied into a new array and sometimes not. This is often a source of confusion for beginners. There are three cases: No Copy at All Simple assignments make no copy of objects or their data. >>> >>> a = np.array([[ 0, 1, 2, 3], ... [ 4, 5, 6, 7], ... [ 8, 9, 10, 11]]) >>> b = a # no new object is created >>> b is a # a and b are two names for the same ndarray object True Python passes mutable objects as references, so function calls make no copy. >>> >>> def f(x): ... print(id(x)) ... >>> id(a) # id is a unique identifier of an object 148293216 # may vary >>> f(a) 148293216 # may vary View or Shallow Copy Different array objects can share the same data. The view method creates a new array object that looks at the same data. >>> >>> c = a.view() >>> c is a False >>> c.base is a # c is a view of the data owned by a True >>> c.flags.owndata False >>> >>> c = c.reshape((2, 6)) # a's shape doesn't change >>> a.shape (3, 4) >>> c[0, 4] = 1234 # a's data changes >>> a array([[ 0, 1, 2, 3], [1234, 5, 6, 7], [ 8, 9, 10, 11]]) Slicing an array returns a view of it: >>> >>> s = a[ : , 1:3] # spaces added for clarity; could also be written "s = a[:, 1:3]" >>> s[:] = 10 # s[:] is a view of s. Note the difference between s = 10 and s[:] = 10 >>> a array([[ 0, 10, 10, 3], [1234, 10, 10, 7], [ 8, 10, 10, 11]]) Deep Copy The copy method makes a complete copy of the array and its data. >>> >>> d = a.copy() # a new array object with new data is created >>> d is a False >>> d.base is a # d doesn't share anything with a False >>> d[0,0] = 9999 >>> a array([[ 0, 10, 10, 3], [1234, 10, 10, 7], [ 8, 10, 10, 11]]) Sometimes copy should be called after slicing if the original array is not required anymore. For example, suppose a is a huge intermediate result and the final result b only contains a small fraction of a, a deep copy should be made when constructing b with slicing: >>> >>> a = np.arange(int(1e8)) >>> b = a[:100].copy() >>> del a # the memory of ``a`` can be released. If b = a[:100] is used instead, a is referenced by b and will persist in memory even if del a is executed. Functions and Methods Overview Here is a list of some useful NumPy functions and methods names ordered in categories. See Routines for the full list. Array Creation arange, array, copy, empty, empty_like, eye, fromfile, fromfunction, identity, linspace, logspace, mgrid, ogrid, ones, ones_like, r_, zeros, zeros_like Conversions ndarray.astype, atleast_1d, atleast_2d, atleast_3d, mat Manipulations array_split, column_stack, concatenate, diagonal, dsplit, dstack, hsplit, hstack, ndarray.item, newaxis, ravel, repeat, reshape, resize, squeeze, swapaxes, take, transpose, vsplit, vstack Questions all, any, nonzero, where Ordering argmax, argmin, argsort, max, min, ptp, searchsorted, sort Operations choose, compress, cumprod, cumsum, inner, ndarray.fill, imag, prod, put, putmask, real, sum Basic Statistics cov, mean, std, var Basic Linear Algebra cross, dot, outer, linalg.svd, vdot Less Basic Broadcasting rules Broadcasting allows universal functions to deal in a meaningful way with inputs that do not have exactly the same shape. The first rule of broadcasting is that if all input arrays do not have the same number of dimensions, a “1” will be repeatedly prepended to the shapes of the smaller arrays until all the arrays have the same number of dimensions. The second rule of broadcasting ensures that arrays with a size of 1 along a particular dimension act as if they had the size of the array with the largest shape along that dimension. The value of the array element is assumed to be the same along that dimension for the “broadcast” array. After application of the broadcasting rules, the sizes of all arrays must match. More details can be found in Broadcasting. Advanced indexing and index tricks NumPy offers more indexing facilities than regular Python sequences. In addition to indexing by integers and slices, as we saw before, arrays can be indexed by arrays of integers and arrays of booleans. Indexing with Arrays of Indices >>> >>> a = np.arange(12)**2 # the first 12 square numbers >>> i = np.array([1, 1, 3, 8, 5]) # an array of indices >>> a[i] # the elements of a at the positions i array([ 1, 1, 9, 64, 25]) >>> >>> j = np.array([[3, 4], [9, 7]]) # a bidimensional array of indices >>> a[j] # the same shape as j array([[ 9, 16], [81, 49]]) When the indexed array a is multidimensional, a single array of indices refers to the first dimension of a. The following example shows this behavior by converting an image of labels into a color image using a palette. >>> >>> palette = np.array([[0, 0, 0], # black ... [255, 0, 0], # red ... [0, 255, 0], # green ... [0, 0, 255], # blue ... [255, 255, 255]]) # white >>> image = np.array([[0, 1, 2, 0], # each value corresponds to a color in the palette ... [0, 3, 4, 0]]) >>> palette[image] # the (2, 4, 3) color image array([[[ 0, 0, 0], [255, 0, 0], [ 0, 255, 0], [ 0, 0, 0]], [[ 0, 0, 0], [ 0, 0, 255], [255, 255, 255], [ 0, 0, 0]]]) We can also give indexes for more than one dimension. The arrays of indices for each dimension must have the same shape. >>> >>> a = np.arange(12).reshape(3,4) >>> a array([[ 0, 1, 2, 3], [ 4, 5, 6, 7], [ 8, 9, 10, 11]]) >>> i = np.array([[0, 1], # indices for the first dim of a ... [1, 2]]) >>> j = np.array([[2, 1], # indices for the second dim ... [3, 3]]) >>> >>> a[i, j] # i and j must have equal shape array([[ 2, 5], [ 7, 11]]) >>> >>> a[i, 2] array([[ 2, 6], [ 6, 10]]) >>> >>> a[:, j] # i.e., a[ : , j] array([[[ 2, 1], [ 3, 3]], [[ 6, 5], [ 7, 7]], [[10, 9], [11, 11]]]) In Python, arr[i, j] is exactly the same as arr[(i, j)]—so we can put i and j in a tuple and then do the indexing with that. >>> >>> l = (i, j) # equivalent to a[i, j] >>> a[l] array([[ 2, 5], [ 7, 11]]) However, we can not do this by putting i and j into an array, because this array will be interpreted as indexing the first dimension of a. >>> >>> s = np.array([i, j]) # not what we want >>> a[s] Traceback (most recent call last): File "<stdin>", line 1, in <module> IndexError: index 3 is out of bounds for axis 0 with size 3 # same as a[i, j] >>> a[tuple(s)] array([[ 2, 5], [ 7, 11]]) Another common use of indexing with arrays is the search of the maximum value of time-dependent series: >>> >>> time = np.linspace(20, 145, 5) # time scale >>> data = np.sin(np.arange(20)).reshape(5,4) # 4 time-dependent series >>> time array([ 20. , 51.25, 82.5 , 113.75, 145. ]) >>> data array([[ 0. , 0.84147098, 0.90929743, 0.14112001], [-0.7568025 , -0.95892427, -0.2794155 , 0.6569866 ], [ 0.98935825, 0.41211849, -0.54402111, -0.99999021], [-0.53657292, 0.42016704, 0.99060736, 0.65028784], [-0.28790332, -0.96139749, -0.75098725, 0.14987721]]) # index of the maxima for each series >>> ind = data.argmax(axis=0) >>> ind array([2, 0, 3, 1]) # times corresponding to the maxima >>> time_max = time[ind] >>> >>> data_max = data[ind, range(data.shape[1])] # => data[ind[0],0], data[ind[1],1]... >>> time_max array([ 82.5 , 20. , 113.75, 51.25]) >>> data_max array([0.98935825, 0.84147098, 0.99060736, 0.6569866 ]) >>> np.all(data_max == data.max(axis=0)) True You can also use indexing with arrays as a target to assign to: >>> >>> a = np.arange(5) >>> a array([0, 1, 2, 3, 4]) >>> a[[1,3,4]] = 0 >>> a array([0, 0, 2, 0, 0]) However, when the list of indices contains repetitions, the assignment is done several times, leaving behind the last value: >>> >>> a = np.arange(5) >>> a[[0,0,2]]=[1,2,3] >>> a array([2, 1, 3, 3, 4]) This is reasonable enough, but watch out if you want to use Python’s += construct, as it may not do what you expect: >>> >>> a = np.arange(5) >>> a[[0,0,2]]+=1 >>> a array([1, 1, 3, 3, 4]) Even though 0 occurs twice in the list of indices, the 0th element is only incremented once. This is because Python requires “a+=1” to be equivalent to “a = a + 1”. Indexing with Boolean Arrays When we index arrays with arrays of (integer) indices we are providing the list of indices to pick. With boolean indices the approach is different; we explicitly choose which items in the array we want and which ones we don’t. The most natural way one can think of for boolean indexing is to use boolean arrays that have the same shape as the original array: >>> >>> a = np.arange(12).reshape(3,4) >>> b = a > 4 >>> b # b is a boolean with a's shape array([[False, False, False, False], [False, True, True, True], [ True, True, True, True]]) >>> a[b] # 1d array with the selected elements array([ 5, 6, 7, 8, 9, 10, 11]) This property can be very useful in assignments: >>> >>> a[b] = 0 # All elements of 'a' higher than 4 become 0 >>> a array([[0, 1, 2, 3], [4, 0, 0, 0], [0, 0, 0, 0]]) You can look at the following example to see how to use boolean indexing to generate an image of the Mandelbrot set: >>> import numpy as np import matplotlib.pyplot as plt def mandelbrot( h,w, maxit=20 ): """Returns an image of the Mandelbrot fractal of size (h,w).""" y,x = np.ogrid[ -1.4:1.4:h*1j, -2:0.8:w*1j ] c = x+y*1j z = c divtime = maxit + np.zeros(z.shape, dtype=int) for i in range(maxit): z = z**2 + c diverge = z*np.conj(z) > 2**2 # who is diverging div_now = diverge & (divtime==maxit) # who is diverging now divtime[div_now] = i # note when z[diverge] = 2 # avoid diverging too much return divtime plt.imshow(mandelbrot(400,400)) ../_images/quickstart-1.png The second way of indexing with booleans is more similar to integer indexing; for each dimension of the array we give a 1D boolean array selecting the slices we want: >>> >>> a = np.arange(12).reshape(3,4) >>> b1 = np.array([False,True,True]) # first dim selection >>> b2 = np.array([True,False,True,False]) # second dim selection >>> >>> a[b1,:] # selecting rows array([[ 4, 5, 6, 7], [ 8, 9, 10, 11]]) >>> >>> a[b1] # same thing array([[ 4, 5, 6, 7], [ 8, 9, 10, 11]]) >>> >>> a[:,b2] # selecting columns array([[ 0, 2], [ 4, 6], [ 8, 10]]) >>> >>> a[b1,b2] # a weird thing to do array([ 4, 10]) Note that the length of the 1D boolean array must coincide with the length of the dimension (or axis) you want to slice. In the previous example, b1 has length 3 (the number of rows in a), and b2 (of length 4) is suitable to index the 2nd axis (columns) of a. The ix_() function The ix_ function can be used to combine different vectors so as to obtain the result for each n-uplet. For example, if you want to compute all the a+b*c for all the triplets taken from each of the vectors a, b and c: >>> >>> a = np.array([2,3,4,5]) >>> b = np.array([8,5,4]) >>> c = np.array([5,4,6,8,3]) >>> ax,bx,cx = np.ix_(a,b,c) >>> ax array([[[2]], [[3]], [[4]], [[5]]]) >>> bx array([[[8], [5], [4]]]) >>> cx array([[[5, 4, 6, 8, 3]]]) >>> ax.shape, bx.shape, cx.shape ((4, 1, 1), (1, 3, 1), (1, 1, 5)) >>> result = ax+bx*cx >>> result array([[[42, 34, 50, 66, 26], [27, 22, 32, 42, 17], [22, 18, 26, 34, 14]], [[43, 35, 51, 67, 27], [28, 23, 33, 43, 18], [23, 19, 27, 35, 15]], [[44, 36, 52, 68, 28], [29, 24, 34, 44, 19], [24, 20, 28, 36, 16]], [[45, 37, 53, 69, 29], [30, 25, 35, 45, 20], [25, 21, 29, 37, 17]]]) >>> result[3,2,4] 17 >>> a[3]+b[2]*c[4] 17 You could also implement the reduce as follows: >>> >>> def ufunc_reduce(ufct, *vectors): ... vs = np.ix_(*vectors) ... r = ufct.identity ... for v in vs: ... r = ufct(r,v) ... return r and then use it as: >>> >>> ufunc_reduce(np.add,a,b,c) array([[[15, 14, 16, 18, 13], [12, 11, 13, 15, 10], [11, 10, 12, 14, 9]], [[16, 15, 17, 19, 14], [13, 12, 14, 16, 11], [12, 11, 13, 15, 10]], [[17, 16, 18, 20, 15], [14, 13, 15, 17, 12], [13, 12, 14, 16, 11]], [[18, 17, 19, 21, 16], [15, 14, 16, 18, 13], [14, 13, 15, 17, 12]]]) The advantage of this version of reduce compared to the normal ufunc.reduce is that it makes use of the Broadcasting Rules in order to avoid creating an argument array the size of the output times the number of vectors. Indexing with strings See Structured arrays. Linear Algebra Work in progress. Basic linear algebra to be included here. Simple Array Operations See linalg.py in numpy folder for more. >>> >>> import numpy as np >>> a = np.array([[1.0, 2.0], [3.0, 4.0]]) >>> print(a) [[1. 2.] [3. 4.]] >>> a.transpose() array([[1., 3.], [2., 4.]]) >>> np.linalg.inv(a) array([[-2. , 1. ], [ 1.5, -0.5]]) >>> u = np.eye(2) # unit 2x2 matrix; "eye" represents "I" >>> u array([[1., 0.], [0., 1.]]) >>> j = np.array([[0.0, -1.0], [1.0, 0.0]]) >>> j @ j # matrix product array([[-1., 0.], [ 0., -1.]]) >>> np.trace(u) # trace 2.0 >>> y = np.array([[5.], [7.]]) >>> np.linalg.solve(a, y) array([[-3.], [ 4.]]) >>> np.linalg.eig(j) (array([0.+1.j, 0.-1.j]), array([[0.70710678+0.j , 0.70710678-0.j ], [0. -0.70710678j, 0. +0.70710678j]])) Parameters: square matrix Returns The eigenvalues, each repeated according to its multiplicity. The normalized (unit "length") eigenvectors, such that the column ``v[:,i]`` is the eigenvector corresponding to the eigenvalue ``w[i]`` . Tricks and Tips Here we give a list of short and useful tips. “Automatic” Reshaping To change the dimensions of an array, you can omit one of the sizes which will then be deduced automatically: >>> >>> a = np.arange(30) >>> b = a.reshape((2, -1, 3)) # -1 means "whatever is needed" >>> b.shape (2, 5, 3) >>> b array([[[ 0, 1, 2], [ 3, 4, 5], [ 6, 7, 8], [ 9, 10, 11], [12, 13, 14]], [[15, 16, 17], [18, 19, 20], [21, 22, 23], [24, 25, 26], [27, 28, 29]]]) Vector Stacking How do we construct a 2D array from a list of equally-sized row vectors? In MATLAB this is quite easy: if x and y are two vectors of the same length you only need do m=[x;y]. In NumPy this works via the functions column_stack, dstack, hstack and vstack, depending on the dimension in which the stacking is to be done. For example: >>> >>> x = np.arange(0,10,2) >>> y = np.arange(5) >>> m = np.vstack([x,y]) >>> m array([[0, 2, 4, 6, 8], [0, 1, 2, 3, 4]]) >>> xy = np.hstack([x,y]) >>> xy array([0, 2, 4, 6, 8, 0, 1, 2, 3, 4]) The logic behind those functions in more than two dimensions can be strange. See also NumPy for Matlab users Histograms The NumPy histogram function applied to an array returns a pair of vectors: the histogram of the array and a vector of the bin edges. Beware: matplotlib also has a function to build histograms (called hist, as in Matlab) that differs from the one in NumPy. The main difference is that pylab.hist plots the histogram automatically, while numpy.histogram only generates the data. >>> import numpy as np rg = np.random.default_rng(1) import matplotlib.pyplot as plt # Build a vector of 10000 normal deviates with variance 0.5^2 and mean 2 mu, sigma = 2, 0.5 v = rg.normal(mu,sigma,10000) # Plot a normalized histogram with 50 bins plt.hist(v, bins=50, density=1) # matplotlib version (plot) # Compute the histogram with numpy and then plot it (n, bins) = np.histogram(v, bins=50, density=True) # NumPy version (no plot) plt.plot(.5*(bins[1:]+bins[:-1]), n) ../_images/quickstart-2.png Further reading The Python tutorial NumPy Reference SciPy Tutorial SciPy Lecture Notes A matlab, R, IDL, NumPy/SciPy dictionary © Copyright 2008-2020, The SciPy community. Last updated on Jun 29, 2020. Created using Sphinx 2.4.4.
lolin32 / ESP32Arduino core for ESP32 WiFi chip Build Status Need help or have a question? Join the chat at https://gitter.im/espressif/arduino-esp32 Development Status Installation Instructions: Using Arduino IDE Windows Mac OS Debian/Ubuntu Decoding Exceptions Using PlatformIO Using as ESP-IDF component ESP32Dev Board PINMAP Development Status Most of the framework is implemented. Most noticable is the missing analogWrite. While analogWrite is on it's way, there are a few other options that you can use: 16 channels LEDC which is PWM 8 channels SigmaDelta which uses SigmaDelta modulation 2 channels DAC which gives real analog output Installation Instructions Using through Arduino IDE ###Instructions for Windows Instructions for Mac Install latest Arduino IDE from arduino.cc Open Terminal and execute the following command (copy->paste and hit enter): mkdir -p ~/Documents/Arduino/hardware/espressif && \ cd ~/Documents/Arduino/hardware/espressif && \ git clone https://github.com/espressif/arduino-esp32.git esp32 && \ cd esp32/tools/ && \ python get.py Restart Arduino IDE Instructions for Debian/Ubuntu Linux Install latest Arduino IDE from arduino.cc Open Terminal and execute the following command (copy->paste and hit enter): sudo usermod -a -G dialout $USER && \ sudo apt-get install git && \ wget https://bootstrap.pypa.io/get-pip.py && \ sudo python get-pip.py && \ sudo pip install pyserial && \ mkdir -p ~/Arduino/hardware/espressif && \ cd ~/Arduino/hardware/espressif && \ git clone https://github.com/espressif/arduino-esp32.git esp32 && \ cd esp32/tools/ && \ python get.py Restart Arduino IDE Decoding exceptions You can use EspExceptionDecoder to get meaningful call trace. Using PlatformIO PlatformIO is an open source ecosystem for IoT development with cross platform build system, library manager and full support for Espressif ESP32 development. It works on the popular host OS: Mac OS X, Windows, Linux 32/64, Linux ARM (like Raspberry Pi, BeagleBone, CubieBoard). What is PlatformIO? PlatformIO IDE Quick Start with PlatformIO IDE or PlatformIO Core Integration with Cloud and Standalone IDEs - Cloud9, Codeanywehre, Eclipse Che (Codenvy), Atom, CLion, Eclipse, Emacs, NetBeans, Qt Creator, Sublime Text, VIM and Visual Studio Project Examples Using "Stage" (Git) version of Arduino Core Building with make makeEspArduino is a generic makefile for any ESP8266/ESP32 Arduino project. Using make instead of the Arduino IDE makes it easier to do automated and production builds. Using as ESP-IDF component Download and install esp-idf Create blank idf project (from one of the examples) in the project folder, create a folder called components and clone this repository inside mkdir -p components && \ cd components && \ git clone https://github.com/espressif/arduino-esp32.git arduino && \ cd .. && \ make menuconfig make menuconfig has some Arduino options "Autostart Arduino setup and loop on boot" If you enable this options, your main.cpp should be formated like any other sketch //file: main.cpp #include "Arduino.h" void setup(){ Serial.begin(115200); } void loop(){ Serial.println("loop"); delay(1000); } Else you need to implement app_main() and call initArduino(); in it. Keep in mind that setup() and loop() will not be called in this case. If you plan to base your code on examples provided in esp-idf, please make sure move the app_main() function in main.cpp from the files in the example. //file: main.cpp #include "Arduino.h" extern "C" void app_main() { initArduino(); pinMode(4, OUTPUT); digitalWrite(4, HIGH); //do your own thing } "Disable mutex locks for HAL" If enabled, there will be no protection on the drivers from concurently accessing them from another thread/interrupt/core "Autoconnect WiFi on boot" If enabled, WiFi will start with the last known configuration Else it will wait for WiFi.begin make flash monitor will build, upload and open serial monitor to your board ESP32Dev Board PINMAP Pin Functions Hint Sometimes to program ESP32 via serial you must keep GPIO0 LOW during the programming process
nickjones / EtmImplements packet decode for an ARM ETMv4 implementation of instruction trace.
dengtao1st / Svc80BreakPointThe svc80BreakPoint is an automated lldb script designed to trace and detect anti-debugging and jailbreak attempts in iOS apps that directly use the assembly instruction svc 0x80.
OCEANOFANYTHINGOFFICIAL / Writewrl## Writeurl - an online collaborative editor ### Writeurl is a collaborative editor Writeurl is a client server system. The frontend code is written in pure javascript using no frameworks. The backend is a node.js application. The client and server communicate through a WebSocket connection. The client stores local changes in the browser's local storage. The editor can be used in offline mode. Changes are always uploaded to the server when a connection is available. Writeurl documents are identified by their (write)url: www.writeurl.com/text/id/read-password/write-password There is a read only version with a (read)url of the form www.writeurl.com/text/id/read-password This url structure makes it easy to share documents. No user registration is needed. #### Writeurl as an online service Writeurl is available as an online service at www.writeurl.com The code running the online service is the same as in this git repo. ###Local installation Writeurl can be installed and run locally as an alternative to using the online service. #### Installation instructions ##### Dependencies The only required dependeny is node.js and the modules in package.json. It is recommended to use node.js version 8. ##### Clone the repo ``` git clone https://github.com/morten-krogh/writeurl.git ``` ##### Install the node js modules ``` npm install ``` ##### Build the browser code Go to the writeurl directory. Use the build script. ``` bash build.sh browser ``` Now the browser code is available in the directory `build/release/browser/` ##### Configuring the server The node.js server code is located in the directory `server-nodejs-express` The server needs a configuration file in the YAML format. An example file is ```server-nodejs-express/config.yaml``` ``` port: 9000 release: public: /Users/mkrogh/writeurl/build/release/browser debug: host: debug.writeurl.localhost public: /Users/mkrogh/writeurl/build/debug/browser documents: path: /Users/mkrogh/writeurl-test-dir/doc_dir publish: public: /Users/mkrogh/writeurl-test-dir/publish_dir # Pino logger # Output goes to stdout. logger: # levels: silent, fatal, error, warn, info, debug, trace level: trace ``` `port` is the port at which the server listens. `release` and `debug` are the directories built above containing the browser code. `documents` is the directory where all the writeurl documents will be stored. The server uses files to store the documents (one subdirectory per document). `publish` is the directory where the published (html) versions of the documents will be stored. `logger.level` is the logging level. Logging goes to standard output. ##### Starting the server In the directory `server-nodejs-express` type ```node writeurl-server.js config.yaml ``` ##### Start typing Go to localhost:9000 in the browser and start typing. ### Example production installation For production, it is recommended to use a reverse proxy with TLS, and to daemonize the Writeurl server. For daemonization, one can use a node.js process manager or a system daemon such as systemd on Linux. The online Writeurl service uses nginx as a reverse proxy and systemd under Linux for daemonization. ##### Example nginx configuration An example nginx.conf file is located at https://github.com/morten-krogh/writeurl/blob/master/documentation/nginx.conf ##### Example systemd unit file ``` [Unit] Description = writeurl server After = network.target [Service] Type = simple User = www ExecStart = /home/www/.nvm/versions/node/v8.9.4/bin/node /home/www/writeurl/server-nodejs-express/writeurl-server.js /home/www/writeurl/server-nodejs-express/config-debian.yaml Restart = on-failure [Install] WantedBy = multi-user.target ``` ### Backup The server can be backed up by just backing up the files in the `documents` and `publish` directories specified in the config.yaml file. Any type of file backup can be used, e.g. periodic rsync. The backup script can be used while the server is running as long as the backup script does not change any files. The server can be restarted from a backup by just placing the backup directories in the place pointed to by `documents` and `publish` in the config file. ### Embedding Writeurl can be embedded as described in https://github.com/morten-krogh/writeurl/blob/master/html/embed/index.html This page is available on the online service as well https://www.writeurl.com/embed ### Contributions and issues Bugs and feature requests are appreciated and can be opened as Github issues. We also welcome code contributions as pull requests.
antonblanchard / Qtrace ToolsTools to read and write qtrace instruction traces
sawansib / ARMTracerARM tracer to generate a compressed trace file that contains all the instructions executed with various information.
jeads-sec / Etherannotate XenEtherAnnotate Xen Ether Modification - Adds a feature to Ether that pulls register values and potential string values at each instruction during an instruction trace.
