Try   HackMD

Assignment2: Complete Applications

contributed by: <JordyMalone> Github

Part A: Mathematical Functions

In this section, we will implement fundamental matrix operations commonly used in neural networks. We will create functions for the following:

  1. Dot Product: This function will calculate the dot product of two vectors.
  2. Matrix Multiplication: This function will handle the multiplication of two matrices.
  3. Element-wise ReLU (Rectified Linear Unit): This function will apply the ReLU activation function element-wise, replacing negative values with 0.
  4. Argmax: This function will find the index of the maximum value within an array.

Task 1: ReLU

In relu.s, implement the ReLU function, which applies the transformation:

ReLU(a)=max(a,0)
Purpose : Each element of the input array will be individually processed by setting negative values to 0. Since the matrix is stored as a 1D row-major vector, this function operates directly on the flattened array.

Implementation

  1. Compute the Address of the Current Element
    • Determine the memory address of the current element in the array.
  2. Check and Update Negative Values
    • If the current element is negative, set its value to 0 to handle non-positive numbers.
  3. Loop Until Completion
    • Repeat the above steps until all elements in the array have been processed.
.globl relu

.text
# ==============================================================================
# FUNCTION: Array ReLU Activation
#
# Applies ReLU (Rectified Linear Unit) operation in-place:
# For each element x in array: x = max(0, x)
#
# Arguments:
#   a0: Pointer to integer array to be modified
#   a1: Number of elements in array
#
# Returns:
#   None - Original array is modified directly
#
# Validation:
#   Requires non-empty array (length ≥ 1)
#   Terminates (code 36) if validation fails
#
# Example:
#   Input:  [-2, 0, 3, -1, 5]
#   Result: [ 0, 0, 3,  0, 5]
# ==============================================================================
relu:
    li t0, 1             
    blt a1, t0, error     
    li t1, 0                    # Initialize loop variable

loop_start:
    # TODO: Add your own implementation
    bge t1, a1, DONE            # if t1 >= a1, then DONE
    slli t2, t1, 2              # t2 = t1 * 4
    add t2, a0, t2              # t2 = base_addr + offset

    lw t3, 0(t2)                # Load arr[t1] into t3
    blt t3, x0 SET_ZERO         # if t3 < 0, then set zero

    addi t1, t1, 1              # else add t1 to next index
    j loop_start

SET_ZERO:
    li t3, 0

NEXT:
    sw t3, 0(t2)                # set zero then store back into arr
    addi t1, t1, 1              # next index

    j loop_start

error:
    li a0, 36          
    j exit          

DONE:
    ret

Task 2: Argmax

Purpose : In argmax.s, implement the argmax function, which returns the index of the largest element in a given vector. If multiple elements share the largest value, return the smallest index. This function operates on 1D vectors.

Implementation

  1. Compute the Current Element Address
    • Determine the memory address of the current element in the array.
  2. Compare with the Maximum Element
    • Check if the current element is greater than the current maximum element.
  3. Update Maximum Element and Index
    • If the condition is true, update both the maximum element and the index position to reflect the new maximum.
  4. Repeat the Process
    • Continue the loop until all elements in the array have been processed.
.globl argmax

.text
# =================================================================
# FUNCTION: Maximum Element First Index Finder
#
# Scans an integer array to find its maximum value and returns the
# position of its first occurrence. In cases where multiple elements
# share the maximum value, returns the smallest index.
#
# Arguments:
#   a0 (int *): Pointer to the first element of the array
#   a1 (int):  Number of elements in the array
#
# Returns:
#   a0 (int):  Position of the first maximum element (0-based index)
#
# Preconditions:
#   - Array must contain at least one element
#
# Error Cases:
#   - Terminates program with exit code 36 if array length < 1
# =================================================================
argmax:
    li t6, 1
    blt a1, t6, handle_error

    lw t0, 0(a0)

    li t1, 0                    # Assume arr[0] is the biggest element
    li t2, 1                    # Initialize the loop variable (start from element 2)
loop_start:
    # TODO: Add your own implementation
    bge t2, a1, DONE            # if t2 >= a1, then DONE

    slli t3, t2, 2              # t3 = t2 * 4
    add t3, a0, t3              # t3 = base_addr + offset
    lw t4, 0(t3)                # Load t3 into t4

    ble t4, t0, NEXT_LOOP       # if t4 <= t0, then NEXT_LOOP

    mv t0, t4                   # Update the value
    mv t1, t2                   # Update the index

NEXT_LOOP:
    addi t2, t2, 1              # Next loop
    j loop_start

handle_error:
    li a0, 36
    j exit

DONE:
    mv a0, t1
    ret

Task 3.1: Dot Product

Purpose : In dot.s, implement the dot product function, defined as:

dot(a,b)=∑i=0n−1(ai⋅bi)

The function dot calculates the product of two integer arrays with specified strides, without using mul instruction.

Implementation

  1. Initialization
    First, we initialize the skip distances for the first and second arrays, multiplying them by 4 (equivalent to a left shift by 2 bits) to convert element strides to byte offsets, as each integer occupies 4 bytes by default.
  2. Main Loop
    We load the current elements, the multiplicand (t3) and the multiplier (t4), into the T registers. The register t2 is used to store the product sum.
  3. Multiplication Loop
    This loop executes the multiplication using bit manipulation and addition.

  • LSB Check of the Multiplier (t4)

    • The instruction andi t5, t4, 1 isolates the least significant bit of t4.
    • If the LSB is 1, it means that the current bit position contributes to the final product.

  • Conditional Addition

    • The instruction beqz t5, skip_add skips the addition if LSB is 0.
    • If LSB is 1, the instruction add t2, t2, t3 adds the current value of the multiplicand(t3) to the product(t2).

  • Shift Operations

    • The instruction slli t3, t3, 1 left shifts the multiplicand by 1 bit, effectively multiply it by 2. This prepares the multiplicand for the next bit position of the multiplier.
    • The instruction srli t4, t4, 1 right shifts the multiplier by 1 bit, effectively dividing it by 2. This moves to the next bit of the multiplier.

  • Loop Continuation

    • The instruction bnez t4, mul_loop continues the loop as long as the multiplier(t4) is not zero.
.globl dot

.text
# =======================================================
# FUNCTION: Strided Dot Product Calculator
#
# Calculates sum(arr0[i * stride0] * arr1[i * stride1])
# where i ranges from 0 to (element_count - 1)
#
# Args:
#   a0 (int *): Pointer to first input array
#   a1 (int *): Pointer to second input array
#   a2 (int):   Number of elements to process
#   a3 (int):   Skip distance in first array
#   a4 (int):   Skip distance in second array
#
# Returns:
#   a0 (int):   Resulting dot product value
#
# Preconditions:
#   - Element count must be positive (>= 1)
#   - Both strides must be positive (>= 1)
#
# Error Handling:
#   - Exits with code 36 if element count < 1
#   - Exits with code 37 if any stride < 1
# =======================================================
dot:
    li t0, 1
    blt a2, t0, error_terminate  
    blt a3, t0, error_terminate   
    blt a4, t0, error_terminate  

    li t0, 0                            # Represent sum
    li t1, 0                            # Represent count

    bge t1, a2, loop_end
    slli a3, a3, 2                      # stride1 * 4 for later move index
    slli a4, a4, 2                      # stride2 * 4 for later move index

loop_start:
    # TODO: Add your own implementation
    addi t1, t1 1
    lw t3, 0(a0)                        # Load multiplicand into t3
    lw t4, 0(a1)                        # Load multiplier into t4
    
    # mul t2, t3, t4
    li t2, 0                            # Represent product

mul_loop:
    andi t5, t4, 1                      # Check multiplier LSB is 1 or not
    beqz t5, skip_add
    add t2, t2, t3                      # Add multiplicand into t2

skip_add:
    slli t3, t3, 1
    srli t4, t4, 1
    bnez t4, mul_loop

    add t0, t0, t2                      # Add product (t3 * t4) into sum
    add a0, a0, a3                      # Move to next index
    add a1, a1, a4
    blt t1, a2, loop_start              # if count < number of elements, then loop_start

loop_end:
    mv a0, t0
    jr ra

error_terminate:
    blt a2, t0, set_error_36
    li a0, 37
    j exit

set_error_36:
    li a0, 36
    j exit

Task 3.2: Matrix Multiplication

Purpose : In matmul.s, implement matrix multiplication, where:

C[i][j]=dot(A[i],B[:,j])

The purpose of inner_loop_end is after completing the computation for all columns of M1. We have to adjust the M0's row position.

Implementation

  1. Inner Loop End
    • After completing the first row across all columns, adjust the row position to move to the next row.
  2. Outer Loop End
    • Restore the stack to maintain the proper state for subsequent operations.
.globl matmul

.text
# =======================================================
# FUNCTION: Matrix Multiplication Implementation
#
# Performs operation: D = M0 × M1
# Where:
#   - M0 is a (rows0 × cols0) matrix
#   - M1 is a (rows1 × cols1) matrix
#   - D is a (rows0 × cols1) result matrix
#
# Arguments:
#   First Matrix (M0):
#     a0: Memory address of first element
#     a1: Row count
#     a2: Column count
#
#   Second Matrix (M1):
#     a3: Memory address of first element
#     a4: Row count
#     a5: Column count
#
#   Output Matrix (D):
#     a6: Memory address for result storage
#
# Validation (in sequence):
#   1. Validates M0: Ensures positive dimensions
#   2. Validates M1: Ensures positive dimensions
#   3. Validates multiplication compatibility: M0_cols = M1_rows
#   All failures trigger program exit with code 38
#
# Output:
#   None explicit - Result matrix D populated in-place
# =======================================================
matmul:
    # Error checks
    li t0 1
    blt a1, t0, error
    blt a2, t0, error
    blt a4, t0, error
    blt a5, t0, error
    bne a2, a4, error

    # Prologue
    addi sp, sp, -28
    sw ra, 0(sp)
    
    sw s0, 4(sp)
    sw s1, 8(sp)
    sw s2, 12(sp)
    sw s3, 16(sp)
    sw s4, 20(sp)
    sw s5, 24(sp)
    
    li s0, 0 # outer loop counter
    li s1, 0 # inner loop counter
    mv s2, a6 # incrementing result matrix pointer
    mv s3, a0 # incrementing matrix A pointer, increments durring outer loop
    mv s4, a3 # incrementing matrix B pointer, increments during inner loop 
    
outer_loop_start:
    #s0 is going to be the loop counter for the rows in A
    li s1, 0                    # Reset the col loop
    mv s4, a3
    blt s0, a1, inner_loop_start

    j outer_loop_end
    
inner_loop_start:
# HELPER FUNCTION: Dot product of 2 int arrays
# Arguments:
#   a0 (int*) is the pointer to the start of arr0
#   a1 (int*) is the pointer to the start of arr1
#   a2 (int)  is the number of elements to use = number of columns of A, or number of rows of B
#   a3 (int)  is the stride of arr0 = for A, stride = 1
#   a4 (int)  is the stride of arr1 = for B, stride = len(rows) - 1
# Returns:
#   a0 (int)  is the dot product of arr0 and arr1
    beq s1, a5, inner_loop_end

    addi sp, sp, -24
    sw a0, 0(sp)
    sw a1, 4(sp)
    sw a2, 8(sp)
    sw a3, 12(sp)
    sw a4, 16(sp)
    sw a5, 20(sp)
    
    mv a0, s3 # setting pointer for matrix A into the correct argument value
    mv a1, s4 # setting pointer for Matrix B into the correct argument value
    mv a2, a2 # setting the number of elements to use to the columns of A
    li a3, 1 # stride for matrix A
    mv a4, a5 # stride for matrix B
    
    jal dot
    
    mv t0, a0 # storing result of the dot product into t0
    
    lw a0, 0(sp)
    lw a1, 4(sp)
    lw a2, 8(sp)
    lw a3, 12(sp)
    lw a4, 16(sp)
    lw a5, 20(sp)
    addi sp, sp, 24
    
    sw t0, 0(s2)
    addi s2, s2, 4 # Incrememtning pointer for result matrix
    
    li t1, 4
    add s4, s4, t1 # incrememtning the column on Matrix B
    
    addi s1, s1, 1                  # add col index
    j inner_loop_start
    
inner_loop_end:
    # TODO: Add your own implementation
# =======================================================
    # After finish the col loop, adjust row position to M0
    addi s0, s0, 1                  # add row loop
    li t1, 4
    
    # mul t0, a2, t1                # t0 = col0 * 4 (offset) 
    mv t2, a2
    li t0, 0

mul_loop:
    beqz t1, mul_done
    andi t3, t1, 1
    beqz t3, skip_add
    add t0, t0, t2

skip_add:
    slli t2, t2, 1
    srli t1, t1, 1
    j mul_loop

mul_done:
    add s3, s3, t0                  # Move s3 to the next row of M0

    j outer_loop_start

outer_loop_end:
    # Epilogue
    lw ra, 0(sp)                    # Restore saved registers
    lw s0, 4(sp)   
    lw s1, 8(sp)
    lw s2, 12(sp)
    lw s3, 16(sp)
    lw s4, 20(sp)
    lw s5, 24(sp)
    addi sp, sp, 28                 # Deallocate the stack space

    ret
# =======================================================
error:
    li a0, 38
    j exit

Part B: File Operations and Main

Purpose : This section focuses on reading and writing matrices to files and building the main function to perform digit classification using the pretrained MNIST weights.

Task 1: Read Matrix

In read_matrix.s, implement the function to read a binary matrix from a file and load it into memory. If any file operation fails, exit with the following codes:

  • 48: Malloc error
  • 50: fopen error
  • 51: fread error
  • 52: fclose error
.globl read_matrix

.text
# ==============================================================================
# FUNCTION: Binary Matrix File Reader
#
# Loads matrix data from a binary file into dynamically allocated memory.
# Matrix dimensions are read from file header and stored at provided addresses.
#
# Binary File Format:
#   Header (8 bytes):
#     - Bytes 0-3: Number of rows (int32)
#     - Bytes 4-7: Number of columns (int32)
#   Data:
#     - Subsequent 4-byte blocks: Matrix elements
#     - Stored in row-major order: [row0|row1|row2|...]
#
# Arguments:
#   Input:
#     a0: Pointer to filename string
#     a1: Address to write row count
#     a2: Address to write column count
#
#   Output:
#     a0: Base address of loaded matrix
#
# Error Handling:
#   Program terminates with:
#   - Code 26: Dynamic memory allocation failed
#   - Code 27: File access error (open/EOF)
#   - Code 28: File closure error
#   - Code 29: Data read error
#
# Memory Note:
#   Caller is responsible for freeing returned matrix pointer
# ==============================================================================
read_matrix:
    
    # Prologue
    addi sp, sp, -40
    sw ra, 0(sp)
    sw s0, 4(sp)
    sw s1, 8(sp)
    sw s2, 12(sp)
    sw s3, 16(sp)
    sw s4, 20(sp)

    mv s3, a1         # save and copy rows
    mv s4, a2         # save and copy cols

    li a1, 0

    jal fopen

    li t0, -1
    beq a0, t0, fopen_error   # fopen didn't work

    mv s0, a0        # file

    # read rows n columns
    mv a0, s0
    addi a1, sp, 28  # a1 is a buffer

    li a2, 8         # look at 2 numbers

    jal fread

    li t0, 8
    bne a0, t0, fread_error

    lw t1, 28(sp)    # opening to save num rows
    lw t2, 32(sp)    # opening to save num cols

    sw t1, 0(s3)     # saves num rows
    sw t2, 0(s4)     # saves num cols

    # mul s1, t1, t2   # s1 is number of elements
    # FIXME: Replace 'mul' with your own implementation
# ==============================================================================
    mv s1, x0                   # Initializing the result s1 to 0
    mv t0, t2                   # copy multiplier to t0

mul_loop:
    beq t0, x0, mul_done        # if t0 == 0, then mul_done
    addi t0, t0, -1             # t0 -= 1
    add s1, s1, t1              # Add row once into s1
    j mul_loop

mul_done:
# ==============================================================================
    slli t3, s1, 2
    sw t3, 24(sp)    # size in bytes

    lw a0, 24(sp)    # a0 = size in bytes

    jal malloc

    beq a0, x0, malloc_error

    # set up file, buffer and bytes to read
    mv s2, a0        # matrix
    mv a0, s0
    mv a1, s2
    lw a2, 24(sp)

    jal fread

    lw t3, 24(sp)
    bne a0, t3, fread_error

    mv a0, s0

    jal fclose

    li t0, -1

    beq a0, t0, fclose_error

    mv a0, s2

    # Epilogue
    lw ra, 0(sp)
    lw s0, 4(sp)
    lw s1, 8(sp)
    lw s2, 12(sp)
    lw s3, 16(sp)
    lw s4, 20(sp)
    addi sp, sp, 40

    jr ra

malloc_error:
    li a0, 26
    j error_exit

fopen_error:
    li a0, 27
    j error_exit

fread_error:
    li a0, 29
    j error_exit

fclose_error:
    li a0, 28
    j error_exit

error_exit:
    lw ra, 0(sp)
    lw s0, 4(sp)
    lw s1, 8(sp)
    lw s2, 12(sp)
    lw s3, 16(sp)
    lw s4, 20(sp)
    addi sp, sp, 40
    j exit

Task 2: Write Matrix

In write_matrix.s, implement the function to write a matrix to a binary file. Use the following exit codes for errors:

  • 53: fopen error
  • 54: fwrite error
  • 55: fclose error
.globl write_matrix

.text
# ==============================================================================
# FUNCTION: Write a matrix of integers to a binary file
# FILE FORMAT:
#   - The first 8 bytes store two 4-byte integers representing the number of 
#     rows and columns, respectively.
#   - Each subsequent 4-byte segment represents a matrix element, stored in 
#     row-major order.
#
# Arguments:
#   a0 (char *) - Pointer to a string representing the filename.
#   a1 (int *)  - Pointer to the matrix's starting location in memory.
#   a2 (int)    - Number of rows in the matrix.
#   a3 (int)    - Number of columns in the matrix.
#
# Returns:
#   None
#
# Exceptions:
#   - Terminates with error code 27 on `fopen` error or end-of-file (EOF).
#   - Terminates with error code 28 on `fclose` error or EOF.
#   - Terminates with error code 30 on `fwrite` error or EOF.
# ==============================================================================
write_matrix:
    # Prologue
    addi sp, sp, -44
    sw ra, 0(sp)
    sw s0, 4(sp)
    sw s1, 8(sp)
    sw s2, 12(sp)
    sw s3, 16(sp)
    sw s4, 20(sp)

    # save arguments
    mv s1, a1        # s1 = matrix pointer
    mv s2, a2        # s2 = number of rows
    mv s3, a3        # s3 = number of columns

    li a1, 1

    jal fopen

    li t0, -1
    beq a0, t0, fopen_error   # fopen didn't work

    mv s0, a0        # file descriptor

    # Write number of rows and columns to file
    sw s2, 24(sp)    # number of rows
    sw s3, 28(sp)    # number of columns

    mv a0, s0
    addi a1, sp, 24  # buffer with rows and columns
    li a2, 2         # number of elements to write
    li a3, 4         # size of each element

    jal fwrite

    li t0, 2
    bne a0, t0, fwrite_error

    # mul s4, s2, s3   # s4 = total elements
    # FIXME: Replace 'mul' with your own implementation
# ==============================================================================
    addi sp, sp, -8
    sw t0, 0(sp)
    sw t1, 4(sp)

    mv t0, x0                   # Initializing the result t0 to 0
    mv t1, s3                   # copy multiplier to t2

mul_loop:
    beq t1, x0, mul_done        # if t1 == 0, then mul_done
    addi t1, t1, -1             # t1 -= 1
    add t0, t0, s2              # Add row once into t0
    j mul_loop

mul_done:
    mv s4, t0
    lw t0, 0(sp)
    lw t1, 4(sp)
    addi sp, sp 8
# ==============================================================================
    # write matrix data to file
    mv a0, s0
    mv a1, s1        # matrix data pointer
    mv a2, s4        # number of elements to write
    li a3, 4         # size of each element

    jal fwrite

    bne a0, s4, fwrite_error

    mv a0, s0

    jal fclose

    li t0, -1
    beq a0, t0, fclose_error

    # Epilogue
    lw ra, 0(sp)
    lw s0, 4(sp)
    lw s1, 8(sp)
    lw s2, 12(sp)
    lw s3, 16(sp)
    lw s4, 20(sp)
    addi sp, sp, 44

    jr ra

fopen_error:
    li a0, 27
    j error_exit

fwrite_error:
    li a0, 30
    j error_exit

fclose_error:
    li a0, 28
    j error_exit

error_exit:
    lw ra, 0(sp)
    lw s0, 4(sp)
    lw s1, 8(sp)
    lw s2, 12(sp)
    lw s3, 16(sp)
    lw s4, 20(sp)
    addi sp, sp, 44
    j exit

Task 3: Classification

In classify.s, bring everything together to classify an input using two weight matrices and the ReLU and ArgMax functions.

.globl classify

.text
# =====================================
# NEURAL NETWORK CLASSIFIER
# =====================================
# Description:
#   Command line program for matrix-based classification
#
# Command Line Arguments:
#   1. M0_PATH      - First matrix file location
#   2. M1_PATH      - Second matrix file location
#   3. INPUT_PATH   - Input matrix file location
#   4. OUTPUT_PATH  - Output file destination
#
# Register Usage:
#   a0 (int)        - Input: Argument count
#                   - Output: Classification result
#   a1 (char **)    - Input: Argument vector
#   a2 (int)        - Input: Silent mode flag
#                     (0 = verbose, 1 = silent)
#
# Error Codes:
#   31 - Invalid argument count
#   26 - Memory allocation failure
#
# Usage Example:
#   main.s <M0_PATH> <M1_PATH> <INPUT_PATH> <OUTPUT_PATH>
# =====================================
classify:
    # Error handling
    li t0, 5
    blt a0, t0, error_args
    
    # Prolouge
    addi sp, sp, -48
    
    sw ra, 0(sp)
    
    sw s0, 4(sp) # m0 matrix
    sw s1, 8(sp) # m1 matrix
    sw s2, 12(sp) # input matrix
    
    sw s3, 16(sp) # m0 matrix rows
    sw s4, 20(sp) # m0 matrix cols
    
    sw s5, 24(sp) # m1 matrix rows
    sw s6, 28(sp) # m1 matrix cols
     
    sw s7, 32(sp) # input matrix rows
    sw s8, 36(sp) # input matrix cols
    sw s9, 40(sp) # h
    sw s10, 44(sp) # o
    
    # Read pretrained m0
    
    addi sp, sp, -12
    
    sw a0, 0(sp)
    sw a1, 4(sp)
    sw a2, 8(sp)
    
    li a0, 4
    jal malloc # malloc 4 bytes for an integer, rows
    beq a0, x0, error_malloc
    mv s3, a0 # save m0 rows pointer for later
    
    li a0, 4
    jal malloc # malloc 4 bytes for an integer, cols
    beq a0, x0, error_malloc
    mv s4, a0 # save m0 cols pointer for later
    
    lw a1, 4(sp) # restores the argument pointer
    
    lw a0, 4(a1) # set argument 1 for the read_matrix function  
    mv a1, s3 # set argument 2 for the read_matrix function
    mv a2, s4 # set argument 3 for the read_matrix function
    
    jal read_matrix
    
    mv s0, a0 # setting s0 to the m0, aka the return value of read_matrix
    
    lw a0, 0(sp)
    lw a1, 4(sp)
    lw a2, 8(sp)
    
    addi sp, sp, 12
    # Read pretrained m1
    
    addi sp, sp, -12
    
    sw a0, 0(sp)
    sw a1, 4(sp)
    sw a2, 8(sp)
    
    li a0, 4
    jal malloc # malloc 4 bytes for an integer, rows
    beq a0, x0, error_malloc
    mv s5, a0 # save m1 rows pointer for later
    
    li a0, 4
    jal malloc # malloc 4 bytes for an integer, cols
    beq a0, x0, error_malloc
    mv s6, a0 # save m1 cols pointer for later
    
    lw a1, 4(sp) # restores the argument pointer
    
    lw a0, 8(a1) # set argument 1 for the read_matrix function  
    mv a1, s5 # set argument 2 for the read_matrix function
    mv a2, s6 # set argument 3 for the read_matrix function
    
    jal read_matrix
    
    mv s1, a0 # setting s1 to the m1, aka the return value of read_matrix
    
    lw a0, 0(sp)
    lw a1, 4(sp)
    lw a2, 8(sp)
    
    addi sp, sp, 12

    # Read input matrix
    
    addi sp, sp, -12
    
    sw a0, 0(sp)
    sw a1, 4(sp)
    sw a2, 8(sp)
    
    li a0, 4
    jal malloc # malloc 4 bytes for an integer, rows
    beq a0, x0, error_malloc
    mv s7, a0 # save input rows pointer for later
    
    li a0, 4
    jal malloc # malloc 4 bytes for an integer, cols
    beq a0, x0, error_malloc
    mv s8, a0 # save input cols pointer for later
    
    lw a1, 4(sp) # restores the argument pointer
    
    lw a0, 12(a1) # set argument 1 for the read_matrix function  
    mv a1, s7 # set argument 2 for the read_matrix function
    mv a2, s8 # set argument 3 for the read_matrix function
    
    jal read_matrix
    
    mv s2, a0 # setting s2 to the input matrix, aka the return value of read_matrix
    
    lw a0, 0(sp)
    lw a1, 4(sp)
    lw a2, 8(sp)
    
    addi sp, sp, 12

    # Compute h = matmul(m0, input)
    addi sp, sp, -28
    
    sw a0, 0(sp)
    sw a1, 4(sp)
    sw a2, 8(sp)
    sw a3, 12(sp)
    sw a4, 16(sp)
    sw a5, 20(sp)
    sw a6, 24(sp)
    
    lw t0, 0(s3)                    # load m0 rows
    lw t1, 0(s8)                    # load input cols
    # mul a0, t0, t1 # FIXME: Replace 'mul' with your own implementation
# =======================================================
    li a0, 0                        # Clear a0 (result)
    mv t2, t1                       # Copy multiplier into t1

mul_loop1:
    beqz t2, mul_done1              # if t2 = 0, then mul_done
    add a0, a0, t0                  # add multiplicand into a0
    addi t2, t2, -1
    j mul_loop1

mul_done1:
# =======================================================
    slli a0, a0, 2
    jal malloc 
    beq a0, x0, error_malloc
    mv s9, a0 # move h to s9
    
    mv a6, a0 # h 
    
    mv a0, s0 # move m0 array to first arg
    lw a1, 0(s3) # move m0 rows to second arg
    lw a2, 0(s4) # move m0 cols to third arg
    
    mv a3, s2 # move input array to fourth arg
    lw a4, 0(s7) # move input rows to fifth arg
    lw a5, 0(s8) # move input cols to sixth arg
    
    jal matmul
    
    lw a0, 0(sp)
    lw a1, 4(sp)
    lw a2, 8(sp)
    lw a3, 12(sp)
    lw a4, 16(sp)
    lw a5, 20(sp)
    lw a6, 24(sp)
    
    addi sp, sp, 28
    
    # Compute h = relu(h)
    addi sp, sp, -8
    
    sw a0, 0(sp)
    sw a1, 4(sp)
    
    mv a0, s9 # move h to the first argument
    lw t0, 0(s3)
    lw t1, 0(s8)
    # mul a1, t0, t1 # length of h array and set it as second argument
# =======================================================
    # FIXME: Replace 'mul' with your own implementation
    li a1, 0                        # Set a1 into 0
    mv t2, t1

mul_loop2:
    beqz t2, mul_done2
    add a1, a1, t0
    addi t2, t2, -1
    j mul_loop2

mul_done2:
# =======================================================
    jal relu                        # relu will not call another func so no ra saved
    
    lw a0, 0(sp)                    
    lw a1, 4(sp)                    
    
    addi sp, sp, 8
    
    # Compute o = matmul(m1, h)
    addi sp, sp, -28
    
    sw a0, 0(sp)
    sw a1, 4(sp)
    sw a2, 8(sp)
    sw a3, 12(sp)
    sw a4, 16(sp)
    sw a5, 20(sp)
    sw a6, 24(sp)
    
    lw t0, 0(s3)
    lw t1, 0(s6)
    # mul a0, t0, t1 # FIXME: Replace 'mul' with your own implementation
# =======================================================
    li a0, 0
    mv t2, t1

mul_loop3:
    beqz t2, mul_done3
    add a0, a0, t0
    addi t2, t2, -1
    j mul_loop3

mul_done3:
# =======================================================
    slli a0, a0, 2
    jal malloc 
    beq a0, x0, error_malloc
    mv s10, a0 # move o to s10
    
    mv a6, a0 # o
    
    mv a0, s1 # move m1 array to first arg
    lw a1, 0(s5) # move m1 rows to second arg
    lw a2, 0(s6) # move m1 cols to third arg
    
    mv a3, s9 # move h array to fourth arg
    lw a4, 0(s3) # move h rows to fifth arg
    lw a5, 0(s8) # move h cols to sixth arg
    
    jal matmul
    
    lw a0, 0(sp)
    lw a1, 4(sp)
    lw a2, 8(sp)
    lw a3, 12(sp)
    lw a4, 16(sp)
    lw a5, 20(sp)
    lw a6, 24(sp)
    
    addi sp, sp, 28
    
    # Write output matrix o
    addi sp, sp, -16
    
    sw a0, 0(sp)
    sw a1, 4(sp)
    sw a2, 8(sp)
    sw a3, 12(sp)
    
    lw a0, 16(a1) # load filename string into first arg
    mv a1, s10 # load array into second arg
    lw a2, 0(s5) # load number of rows into fourth arg
    lw a3, 0(s8) # load number of cols into third arg
    
    jal write_matrix
    
    lw a0, 0(sp)
    lw a1, 4(sp)
    lw a2, 8(sp)
    lw a3, 12(sp)
    
    addi sp, sp, 16
    
    # Compute and return argmax(o)
    addi sp, sp, -12
    
    sw a0, 0(sp)
    sw a1, 4(sp)
    sw a2, 8(sp)
    
    mv a0, s10 # load o array into first arg
    lw t0, 0(s3)
    lw t1, 0(s6)
    # mul a1, t0, t1 # load length of array into second arg
# =======================================================
    # FIXME: Replace 'mul' with your own implementation
    li a1, 0
    mv t2, t1

mul_loop4:
    beqz t2, mul_done4
    add a1, a1, t0
    addi t2, t2, -1
    j mul_loop4

mul_done4:
# =======================================================
    
    jal argmax
    
    mv t0, a0 # move return value of argmax into t0
    
    lw a0, 0(sp)
    lw a1, 4(sp)
    lw a2, 8(sp)
    
    addi sp, sp 12
    
    mv a0, t0

    # If enabled, print argmax(o) and newline
    bne a2, x0, epilouge
    
    addi sp, sp, -4
    sw a0, 0(sp)
    
    jal print_int
    li a0, '\n'
    jal print_char
    
    lw a0, 0(sp)
    addi sp, sp, 4
    
    # Epilouge
epilouge:
    addi sp, sp, -4
    sw a0, 0(sp)
    
    mv a0, s0
    jal free
    
    mv a0, s1
    jal free
    
    mv a0, s2
    jal free
    
    mv a0, s3
    jal free
    
    mv a0, s4
    jal free
    
    mv a0, s5
    jal free
    
    mv a0, s6
    jal free
    
    mv a0, s7
    jal free
    
    mv a0, s8
    jal free
    
    mv a0, s9
    jal free
    
    mv a0, s10
    jal free
    
    lw a0, 0(sp)
    addi sp, sp, 4

    lw ra, 0(sp)
    
    lw s0, 4(sp) # m0 matrix
    lw s1, 8(sp) # m1 matrix
    lw s2, 12(sp) # input matrix
    
    lw s3, 16(sp) 
    lw s4, 20(sp)
    
    lw s5, 24(sp)
    lw s6, 28(sp)
    
    lw s7, 32(sp)
    lw s8, 36(sp)
    
    lw s9, 40(sp) # h
    lw s10, 44(sp) # o
    
    addi sp, sp, 48
    
    jr ra

error_args:
    li a0, 31
    j exit

error_malloc:
    li a0, 26
    j exit

Result

test_abs_minus_one (__main__.TestAbs) ... ok
test_abs_one (__main__.TestAbs) ... ok
test_abs_zero (__main__.TestAbs) ... ok
test_argmax_invalid_n (__main__.TestArgmax) ... ok
test_argmax_length_1 (__main__.TestArgmax) ... ok
test_argmax_standard (__main__.TestArgmax) ... ok
test_chain_1 (__main__.TestChain) ... ok
test_classify_1_silent (__main__.TestClassify) ... ok
test_classify_2_print (__main__.TestClassify) ... ok
test_classify_3_print (__main__.TestClassify) ... ok
test_classify_fail_malloc (__main__.TestClassify) ... ok
test_classify_not_enough_args (__main__.TestClassify) ... ok
test_dot_length_1 (__main__.TestDot) ... ok
test_dot_length_error (__main__.TestDot) ... ok
test_dot_length_error2 (__main__.TestDot) ... ok
test_dot_standard (__main__.TestDot) ... ok
test_dot_stride (__main__.TestDot) ... ok
test_dot_stride_error1 (__main__.TestDot) ... ok
test_dot_stride_error2 (__main__.TestDot) ... ok
test_matmul_incorrect_check (__main__.TestMatmul) ... ok
test_matmul_length_1 (__main__.TestMatmul) ... ok
test_matmul_negative_dim_m0_x (__main__.TestMatmul) ... ok
test_matmul_negative_dim_m0_y (__main__.TestMatmul) ... ok
test_matmul_negative_dim_m1_x (__main__.TestMatmul) ... ok
test_matmul_negative_dim_m1_y (__main__.TestMatmul) ... ok
test_matmul_nonsquare_1 (__main__.TestMatmul) ... ok
test_matmul_nonsquare_2 (__main__.TestMatmul) ... ok
test_matmul_nonsquare_outer_dims (__main__.TestMatmul) ... ok
test_matmul_square (__main__.TestMatmul) ... ok
test_matmul_unmatched_dims (__main__.TestMatmul) ... ok
test_matmul_zero_dim_m0 (__main__.TestMatmul) ... ok
test_matmul_zero_dim_m1 (__main__.TestMatmul) ... ok
test_read_1 (__main__.TestReadMatrix) ... ok
test_read_2 (__main__.TestReadMatrix) ... ok
test_read_3 (__main__.TestReadMatrix) ... ok
test_read_fail_fclose (__main__.TestReadMatrix) ... ok
test_read_fail_fopen (__main__.TestReadMatrix) ... ok
test_read_fail_fread (__main__.TestReadMatrix) ... ok
test_read_fail_malloc (__main__.TestReadMatrix) ... ok
test_relu_invalid_n (__main__.TestRelu) ... ok
test_relu_length_1 (__main__.TestRelu) ... ok
test_relu_standard (__main__.TestRelu) ... ok
test_write_1 (__main__.TestWriteMatrix) ... ok
test_write_fail_fclose (__main__.TestWriteMatrix) ... ok
test_write_fail_fopen (__main__.TestWriteMatrix) ... ok
test_write_fail_fwrite (__main__.TestWriteMatrix) ... ok

----------------------------------------------------------------------
Ran 46 tests in 21.026s

OK

Reference

Last changed by 
 
JordyMalone
學習的過程總是充滿挑戰與成長
0
44

Read more

2025q1 Homework1 (lab0)

2025q1 Homework1 (lab0)

Apr 1, 2025
2025q1 Homework2 (quiz1+2)

2025q1 Homework2 (quiz1+2)

Mar 19, 2025
2025q1 Homework1 (ideas)

2025q1 Homework1 (ideas)

Mar 4, 2025
Assignment1: RISC-V Assembly and Instruction Pipeline

Assignment1: RISC-V Assembly and Instruction Pipeline

Nov 2, 2024
Read more from JordyMalone
Published on HackMD