Skip to content
Automatic functional verification tool for assembly implementation of cryptographic algorithms
Python Shell Dockerfile
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.


CASM_Verify is a tool that automatically checks the equivalence of highly optimized assembly implementation of cryptographic algorithms against a reference implementation. The reference implementation can be another assembly implementation, or an implementation written in our DSL.

You can read more about our approach in the following paper:

Automatic Equivalence Checking for Assembly Implementations of Cryptography Libraries


CASM_Verify requires python3 and the z3 bindings for python3. In ubuntu, the requirements can be installed using the following commands:

$ sudo apt-get install python3 python3-pip
$ sudo python3 -m pip install z3-solver

In macOS, use homebrew instead of apt-get:

$ brew install python3
$ sudo python3 -m pip install z3-solver

To install z3 manually instead of using python package manager, follow the instruction in the link provided: z3


Installation using Docker

The simplest way to install CASM_Verify is to use the Docker Makefile provided in the repo. The makefile automatically installs all pre-requisite softwares.

  1. Install Docker: Docker provides easy to follow documentation on how to install Docker on various platforms. You can find the instruction for your specific OS by going to, on the left sidebar, go to "Get Docker" -> "Docker CE" and click the OS for your machine.

  2. Build Docker image: You can create docker image for CASM_Verify by typing the following command in the terminal,

docker build -t casmverify
  1. Run casmverify image: To run the docker image, type,
docker run -it casmverify

Manual Installation

To manually install CASM-Verify, install the prerequisites and clone this repository:

git clone

Now you're ready to use CASM-Verify.

CASM-Verify Benchmarks.

We provide a number of examples (used in our paper) in the "test" folder. We also provide a bash script that automatically runs the verification of these examples:



For the rest of the readme, we will be using the example in test/sha2rnd unless otherwise specified.

To verify the assembly implementation(test/sha2rnd/asm) using CASM-Verify, use the following command:

python3 --pre test/sha2rnd/pre --post test/sha2rnd/post --p1 test/sha2rnd/dsl --p1lang dsl --p2 test/sha2rnd/asm --p2lang asm --mem-model 32

There are seven important parameters:

  1. --p1: specifies the file path to the reference implementation.
  2. --p1lang: specifies whether the reference implementation is written in x86_64 assembly (asm) or in our DSL (dsl).
  3. --p2: specifies the file path to the target implementation.
  4. --p2lang: specifies whether the target implementation is written in x86_64 assembly (asm) or in our DSL (dsl).
  5. --pre: File containing the precondition that specifies the program state at the beginning of p1 and p2.
  6. --post: File containing the postcondition that specifies which variables have to be equivalent for p1 and p2 to be equivalent.
  7. --mem-model: specifies how to model the memory. We currently either model the memory as an array of 8-bit values (8) or as an array of 32-bit values (32) for implementations that read/write word-sized values. This parameter is used for translating assembly implementation to our internal DSL.

The above example (test/sha2rnd) verifies an assembly implementation against the reference implementation written in our DSL, using a 32-bit value memory model.

We also provide an example that verifies an assembly implementation (ChaCha20 in x86_64 with SSE) against another assembly implementation (ChaCha20 in x86_64) in test/ChaCha20_naive_to_ssse:

python3 --pre test/ChaCha20_naive_to_ssse/pre --post test/ChaCha20_naive_to_ssse/post --p1 test/ChaCha20_naive_to_ssse/p1 --p1lang asm --p2 test/ChaCha20_naive_to_ssse/p2 --p2lang asm --mem-model 32

and an example that uses an 8-bit value memory model in test/AES_decrypt:

python3 --pre test/AES_decrypt/pre --post test/AES_decrypt/post --p1 test/AES_decrypt/dsl --p1lang dsl --p2 test/AES_decrypt/asm --p2lang asm --mem_model 8


Assembly Implementations

Click to see details

CASM-Verify accepts AT&T syntax of assembly instructions. Although CASM-Verify can accept a wide variety of instructions, it does not reason about all instructions (notably, jump instructions and labels). Whenever CASM-Verify encounters an instruction it cannot reason about, it will notify the user of the instruction.

Domain Specific Language (DSL)

Click to see details

We provide an imperative C-like DSL to write the reference implementation, precondition, and the postcondition. We are actively refining our DSL to be more user friendly. To illustrate the features of our DSL, here is a code snippet of test/sha2rnd/dsl:

function SIGMA0(a) {
return (a >>> 2) ^ (a >>> 13) ^ (a >>> 22);
function CH(x, y, z) {
return (x & y) ^ (!x & z);

for (i from 0 to 1) {
	w[i] = m[i];
	sigma1 = SIGMA1(e);
	ch = CH(e,f,g);
	t1 = h + sigma1 + ch + k[i] + w[i];

Variables (i.e. a): All variables are implicitly declared and they are integers. By default, they are 32-bit integers. If you need a different sized integers, use the format var:size, i.e. x:64 is a 64-bit integer named x. Constants are defaulted to 32-bit integers. Similar to variables, you can declare different sizes of constants, i.e. 1553:40 is a 40-bit integer representing the value 1553. Arrays (i.e. w[i]): Like variables, arrays use 32-bit indices and hold 32-bit values by default. If you need different sized arrays, use the format name:s1[index:s2], i.e. mem:8[addr:64] is an array named mem, which uses 64-bit indices and returns 8-bit values. Loops: Our DSL only allows a fixed-iteration for loop. CASM-verify internally unroll the for loop. Function: our DSL supports mathematical functions. CASM-verify internally inlines the function.


Click to see details

CASM-Verify requires information about the initial program states of p1 and p2 through Precondition. Because of possible ambiguity between variable names in p1 and p2, all variables in Precondition and Postcondition must be prepended with "P1." if the variable is from p1 and "P2." if the variable is from p2. For example, precondition test/sha2rnd/pre,

@Data{P2.rbp:64, P2.rbp:64 ~ P2.rbp:64 + 7:64};

P1.a == P2.eax; 
P1.b == P2.ebx;
P2.mem[P2.rbp:64] == 0xe49b69c1;
P1.m[0] == P2.mem[P2.rsi:64];

specifies that the variable a from p1 and register eax from p2 are equivalent. Additionally,

@Data: Describes the data regions in the memory used by assembly implementations.
1) Argument #1: Which register points to the region. (i.e. P2.rbp:64 says this data region is pointed by the register rbp)
2) Argument #2: What is the range of the regions? (i.e. The size of the region is 8 bytes starting from rbp)

mem[]: Array "mem" is registered to be used for the memory model for the assembly implementations. To describe the content of the memory, use mem.


Click to see details

The postcondition specifies which variables should be equivalent after executing p1 and p2:

P1.a == P2.r10d;
P1.b == P2.r11d;

The postcondition can compare between values in arrays as well, similar to precondition.

You can’t perform that action at this time.