Systemverilog code

SIMD (single instruction, multiple data) instructions are a common microprocessor architecture feature that support parallel operations using “sub-word” parallelism. The standard architecture is extended to provide very large registers which can contain multiple operands allowing the ALU to perform the same operation in parallel. Intel released several generations of SIMD instructions and correspondingly longer registers for the x86 architecture since their introduction (as the MMX extension for multimedia applications) to the Pentium in 1997. The most recent generation, AVX-512, in 2017, supports 32 512-bit ZMM registers.

Write SystemVerilog code to declare a new type (using typedef and a union) for an AVX512 register that can be accessed as byte, word, doubleword, quadword, single

precision floating point, or double precision floating point values. Use the

SystemVerilog types (including two-state types where possible) that most closely match

these types. Put this code in a package called SIMD (call the file simd.sv).

To make effective use of parallel ALU operations on multiple operands, the architecture must

also support instructions to move data between these registers and memory.

Create a module that imports this package and declares an unpacked array of 32 ZMM

registers (called ZMM) using the typedef, and a dynamic array of bytes called M used to model

a memory.

Assume a testbench uses an unpacked array of 32 short unsigned integer operands as shown

below.

shortint unsigned operands[32], results[32];

Write a void function CopyToZMM that will copy all 32 shortints to ZMM register i from

operands. Pass operands and the ZMM array by reference. Include it in your SIMD

package.

Write a void function CopyFromZMM to copy the contents of ZMM register i into

results. Pass results and the ZMM array by reference. Include it in your SIMD

package.

Assuming the architecture is little-endian, write a void function LoadZMM that will load

ZMM register i with 32 unsigned short integers from memory beginning at address. The

shortint located at M[address] should be stored in the most significant two bytes of the

ZMM register i , with the shortint located at M[address + 62] stored in the least

significant two bytes of the ZMM register. The memory array should be passed by

reference. Include it in your SIMD package.

Write the corresponding void function StoreZMM to store ZMM register i as 32 unsigned

short integers into memory beginning at address. The memory array should be passed by

reference. Include it in your SIMD package.

Create a file called stream.sv to import the SIMD package and define a module top() that

demonstrates that your solutions are correct. The ZMM registers and the memory should be

defined in top().

What does the string method s.substr(i,j) return if the starting index of the substring is

< 0 or the ending index is > s.len-1?

Create a package called strings in a file strings.sv. Write a string function splice that

takes three arguments: a string s, a starting index int i, and an ending index int j, and if i

and j are both valid indexes it returns the string that remains if the substring

s.substr(i,j) were chopped out of s. It returns the empty string otherwise.