Skip to content

Commit

Permalink
Update assignment 5 grading scripts
Browse files Browse the repository at this point in the history
  • Loading branch information
@minseongg
minseongg committed Nov 12, 2024
1 parent a3f5608 commit 48d52ca
Show file tree
Hide file tree
Showing 3 changed files with 123 additions and 73 deletions.
2 changes: 1 addition & 1 deletion scripts/gemmini/unit_test.sh
View file
Original file line number Diff line number Diff line change
Expand Up @@ -39,7 +39,7 @@ pip3 install -r $CURR_DIR/requirements.txt
cd $CURR_DIR/unit_tests/$1
make clean
make WAVES=1 | tee $CURR_DIR/$LOG_FILE
echo "To see the waveform, go to the directory: $CURR_DIR/unit_tests/$1 and check \`$1_top.fst\` file with gtkwave."
echo "To see the waveform, go to the directory: $CURR_DIR/unit_tests/$1 and check \`${TARGET_NAME}_top.fst\` file with gtkwave."

FAILED_COUNT=$(grep -o "failed" $CURR_DIR/$LOG_FILE | wc -l)
if [ $FAILED_COUNT -eq 0 ]; then
Expand Down
74 changes: 38 additions & 36 deletions scripts/gemmini/unit_tests/mesh/test_mesh.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,6 @@
gemmini_unit_tb_dir = os.path.dirname(os.path.dirname(os.path.realpath(__file__)))
sys.path.append(gemmini_unit_tb_dir)

random.seed(0)

# Dataflow discriminant
OS = 0
Expand Down Expand Up @@ -86,7 +85,7 @@ def concatenate_data(data: list, width):
return data_concat


def decopmose_data(data: BinaryValue, width):
def decompose_data(data: BinaryValue, width):
data_list = []
data_len = len(data.binstr)
num_data = data_len // width
Expand All @@ -106,6 +105,9 @@ async def ws_simple(dut):
"""
Test Weight Stationary with ones
"""
random.seed(0)
np.random.seed(0)
np.set_printoptions(linewidth=200)
tb = TB(dut)

await tb.reset()
Expand All @@ -119,10 +121,10 @@ async def ws_simple(dut):
golden_output_data = np.matmul(activation, weight) + bias

cocotb.log.info("== Weight Stationary testcase1 ==")
cocotb.log.info(f"Input data: {activation}")
cocotb.log.info(f"Weight data: {weight}")
cocotb.log.info(f"Bias data: {bias}")
cocotb.log.info(f"Expected Output data: {golden_output_data}")
cocotb.log.info(f"Input data:\n{activation}")
cocotb.log.info(f"Weight data:\n{weight}")
cocotb.log.info(f"Bias data:\n{bias}")
cocotb.log.info(f"Expected output data:\n{golden_output_data}")

output_data = []

Expand Down Expand Up @@ -220,7 +222,7 @@ async def ws_simple(dut):
tb.in_col_ctrl_propagate.value = concatenate_data([REG1] * 16, 1)

if tb.out_col_ctrl_propagate.value.binstr == "1111111111111111":
data_decomposed = decopmose_data(tb.out_col_data_b.value, 20)
data_decomposed = decompose_data(tb.out_col_data_b.value, 20)
output_data.append(data_decomposed)
await RisingEdge(dut.clk)

Expand All @@ -229,11 +231,11 @@ async def ws_simple(dut):

for i in range(24):
if tb.out_col_ctrl_propagate.value.binstr == "1111111111111111":
data_decomposed = decopmose_data(tb.out_col_data_b.value, 20)
data_decomposed = decompose_data(tb.out_col_data_b.value, 20)
output_data.append(data_decomposed)
await RisingEdge(dut.clk)

tb.log.info(f"output_data: {output_data}")
cocotb.log.info(f"Output data (skewed form):\n{np.array(output_data)}")

# Check output
for i in range(16):
Expand All @@ -246,6 +248,9 @@ async def ws_random(dut):
"""
Test Weight Stationary with random inputs
"""
random.seed(0)
np.random.seed(0)
np.set_printoptions(linewidth=200)
tb = TB(dut)

await tb.reset()
Expand All @@ -259,10 +264,10 @@ async def ws_random(dut):
golden_output_data = np.matmul(activation, weight) + bias

cocotb.log.info("== Weight Stationary testcase1 ==")
cocotb.log.info(f"Input data: {activation}")
cocotb.log.info(f"Weight data: {weight}")
cocotb.log.info(f"Bias data: {bias}")
cocotb.log.info(f"Expected Output data: {golden_output_data}")
cocotb.log.info(f"Input data:\n{activation}")
cocotb.log.info(f"Weight data:\n{weight}")
cocotb.log.info(f"Bias data:\n{bias}")
cocotb.log.info(f"Expected output data:\n{golden_output_data}")

output_data = []

Expand Down Expand Up @@ -360,7 +365,7 @@ async def ws_random(dut):
tb.in_col_ctrl_propagate.value = concatenate_data([REG1] * 16, 1)

if tb.out_col_ctrl_propagate.value.binstr == "1111111111111111":
data_decomposed = decopmose_data(tb.out_col_data_b.value, 20)
data_decomposed = decompose_data(tb.out_col_data_b.value, 20)
output_data.append(data_decomposed)
await RisingEdge(dut.clk)

Expand All @@ -369,11 +374,11 @@ async def ws_random(dut):

for i in range(24):
if tb.out_col_ctrl_propagate.value.binstr == "1111111111111111":
data_decomposed = decopmose_data(tb.out_col_data_b.value, 20)
data_decomposed = decompose_data(tb.out_col_data_b.value, 20)
output_data.append(data_decomposed)
await RisingEdge(dut.clk)

tb.log.info(f"output_data: {output_data}")
cocotb.log.info(f"Output data (skewed form):\n{np.array(output_data)}")

# Check output
for i in range(16):
Expand All @@ -386,6 +391,9 @@ async def os_simple(dut):
"""
Test Output Stationary with ones
"""
random.seed(0)
np.random.seed(0)
np.set_printoptions(linewidth=200)
tb = TB(dut)

await tb.reset()
Expand All @@ -398,10 +406,9 @@ async def os_simple(dut):
golden_output_data = np.matmul(activation, weight)

cocotb.log.info("== Output Stationary testcase1 ==")

cocotb.log.info(f"Input data: {activation}")
cocotb.log.info(f"Weight data: {weight}")
cocotb.log.info(f"Expected Output data: {golden_output_data}")
cocotb.log.info(f"Input data:\n{activation}")
cocotb.log.info(f"Weight data:\n{weight}")
cocotb.log.info(f"Expected output data:\n{golden_output_data}")

output_data = []

Expand Down Expand Up @@ -443,8 +450,6 @@ async def os_simple(dut):
col_data[j] = weight[i - j][j]
propagate[j] = REG1 if j <= i else REG2

# tb.log.info(f"i: {i}, row_data: {row_data}, col_data: {col_data}, propagete: {propagate}")

tb.in_row_data_valids.value = concatenate_data(data_valids, 1)
tb.in_col_data_valids.value = concatenate_data(data_valids, 1)
tb.in_row_data_a.value = concatenate_data(row_data, 8)
Expand All @@ -465,8 +470,6 @@ async def os_simple(dut):
col_data[j] = weight[16 - (j - i)][j] if j > i else 0
propagate[j] = REG1 if j > i else REG2

# tb.log.info(f"i: {i}, row_data: {row_data}, col_data: {col_data}, propagete: {propagate}")

tb.in_row_data_valids.value = concatenate_data(data_valids, 1)
tb.in_col_data_valids.value = concatenate_data(data_valids, 1)
tb.in_row_data_a.value = concatenate_data(row_data, 8)
Expand All @@ -481,20 +484,24 @@ async def os_simple(dut):
tb.in_col_ctrl_propagate.value = concatenate_data([REG2] * 16, 1)

for i in range(64):
output_data.append(decopmose_data(tb.out_col_data_d.value, 20))
output_data.append(decompose_data(tb.out_col_data_d.value, 20))
await RisingEdge(dut.clk)
output_data = output_data[2:]
tb.log.info(f"output_data: {output_data}")
cocotb.log.info(f"Output data (skewed form):\n{np.array(output_data)}")

for i in range(16):
for j in range(16):
assert golden_output_data[i][j] == output_data[15 - i + j][j]


@cocotb.test(timeout_time=10, timeout_unit="ms")
async def os_random(dut):
"""
Test Output Stationary with random numbers
"""
random.seed(0)
np.random.seed(0)
np.set_printoptions(linewidth=200)
tb = TB(dut)

await tb.reset()
Expand All @@ -507,10 +514,9 @@ async def os_random(dut):
golden_output_data = np.matmul(activation, weight)

cocotb.log.info("== Output Stationary testcase1 ==")

cocotb.log.info(f"Input data: {activation}")
cocotb.log.info(f"Weight data: {weight}")
cocotb.log.info(f"Expected Output data: {golden_output_data}")
cocotb.log.info(f"Input data:\n{activation}")
cocotb.log.info(f"Weight data:\n{weight}")
cocotb.log.info(f"Expected output data:\n{golden_output_data}")

output_data = []

Expand Down Expand Up @@ -552,8 +558,6 @@ async def os_random(dut):
col_data[j] = weight[i - j][j]
propagate[j] = REG1 if j <= i else REG2

# tb.log.info(f"i: {i}, row_data: {row_data}, col_data: {col_data}, propagete: {propagate}")

tb.in_row_data_valids.value = concatenate_data(data_valids, 1)
tb.in_col_data_valids.value = concatenate_data(data_valids, 1)
tb.in_row_data_a.value = concatenate_data(row_data, 8)
Expand All @@ -574,8 +578,6 @@ async def os_random(dut):
col_data[j] = weight[16 - (j - i)][j] if j > i else 0
propagate[j] = REG1 if j > i else REG2

# tb.log.info(f"i: {i}, row_data: {row_data}, col_data: {col_data}, propagete: {propagate}")

tb.in_row_data_valids.value = concatenate_data(data_valids, 1)
tb.in_col_data_valids.value = concatenate_data(data_valids, 1)
tb.in_row_data_a.value = concatenate_data(row_data, 8)
Expand All @@ -590,10 +592,10 @@ async def os_random(dut):
tb.in_col_ctrl_propagate.value = concatenate_data([REG2] * 16, 1)

for i in range(64):
output_data.append(decopmose_data(tb.out_col_data_d.value, 20))
output_data.append(decompose_data(tb.out_col_data_d.value, 20))
await RisingEdge(dut.clk)
output_data = output_data[2:]
tb.log.info(f"output_data: {output_data}")
cocotb.log.info(f"Output data (skewed form):\n{np.array(output_data)}")

for i in range(16):
for j in range(16):
Expand Down
120 changes: 84 additions & 36 deletions scripts/gemmini/unit_tests/transposer/test_transposer.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@
import random

import numpy as np
from itertools import product

import cocotb
from cocotb.clock import Clock
Expand All @@ -10,7 +11,7 @@
from cocotb.binary import BinaryValue


random.seed(0)
DIM = 16


class TB(object):
Expand Down Expand Up @@ -40,58 +41,105 @@ async def reset(self):
def concatenate_data(data: list, width):
data_concat = 0
bitmask = (1 << width) - 1

for i in range(len(data)):
data_concat |= (int(data[i]) << (width * i)) & (bitmask << (width * i))

return data_concat


def decopmose_data(data: BinaryValue, width):
def decompose_data(data: BinaryValue, width):
data_list = []
data_len = len(data.binstr)
num_data = data_len // width

for i in reversed(range(num_data)):
data_list.append(
BinaryValue(data.binstr[i * width : (i + 1) * width]).signed_integer
)
data_list.append(BinaryValue(data.binstr[i * width : (i + 1) * width]).signed_integer)

return data_list


@cocotb.test(timeout_time=10, timeout_unit="ms")
async def transpose_test1(dut):
async def transpose_test(dut):
"""
Weight Stationary testcase1
Testcase.
"""
random.seed(0)
np.set_printoptions(linewidth=200)
tb = TB(dut)

await tb.reset()
await ClockCycles(dut.clk, 10)

# Generate inputs randomly
input_data = [[random.randint(-128, 127) for _ in range(16)] for _ in range(16)]

cocotb.log.info("== Tranaposer testcase1 ==")
cocotb.log.info(f"Input data: {input_data}")

output_data = np.transpose(input_data)
cocotb.log.info(f"Expected Output data: {output_data}")

for i in range(16):
in_row_data = concatenate_data(input_data[i], 8)
tb.in_row_data.value = in_row_data
tb.in_row_valid.value = 1
await RisingEdge(dut.clk)
for i in range(16):
tb.in_row_valid.value = 0
tb.in_row_data.value = 0
await RisingEdge(dut.clk)

tb.in_row_valid.value = 1
tb.in_row_data.value = 0

await FallingEdge(dut.clk)

out_col_data = np.array(decopmose_data(tb.out_col_data.value, 8))
assert np.array_equal(out_col_data, output_data[i])
cocotb.log.info(f"Output data: {out_col_data}")

await RisingEdge(dut.clk)
# Generate random matrices
A = np.array([[random.randint(-128, 127) for _ in range(DIM)] for _ in range(DIM)])
B = np.array([[random.randint(-128, 127) for _ in range(DIM)] for _ in range(DIM)])

A_t = np.transpose(A)
B_t = np.transpose(B)

cocotb.log.info("== Transposer testcase ==")

cocotb.log.info(f"A (input 1):\n{A}")
cocotb.log.info(f"A transposed (expected output 1):\n{A_t}")

cocotb.log.info(f"B (input 2):\n{B}")
cocotb.log.info(f"B transposed (expected output 2):\n{B_t}")

for (i, (delay1, delay2, delay3)) in enumerate(product([False, True], repeat=3)):
cocotb.log.info(f"Config {i}:")
if i == 0:
cocotb.log.info("No delay")
else:
if delay1:
cocotb.log.info("Add a delay for inputting A")
if delay2:
cocotb.log.info("Add a delay for outputting A and inputting B")
if delay3:
cocotb.log.info("Add a delay for outputting B")

A_o, B_o = [], []

for i in range(DIM):
# Add a delay for inputting A
if delay1:
tb.in_row_valid.value = 0
await RisingEdge(dut.clk)

tb.in_row_valid.value = 1
tb.in_row_data.value = concatenate_data(A[i], 8)
await RisingEdge(dut.clk)

for i in range(DIM):
# Add a delay for outputting A and inputting B
if delay2:
tb.in_row_valid.value = 0
await RisingEdge(dut.clk)

tb.in_row_valid.value = 1
tb.in_row_data.value = concatenate_data(B[i], 8)
await FallingEdge(dut.clk)

A_o.append(decompose_data(tb.out_col_data.value, 8))
await RisingEdge(dut.clk)

A_o = np.array(A_o)
cocotb.log.info(f"A transposed (output):\n{A_o}")
assert np.array_equal(A_o, A_t)

for i in range(DIM):
# Add a delay for outputting B
if delay3:
tb.in_row_valid.value = 0
await RisingEdge(dut.clk)

tb.in_row_valid.value = 1
tb.in_row_data.value = 0
await FallingEdge(dut.clk)

B_o.append(decompose_data(tb.out_col_data.value, 8))
await RisingEdge(dut.clk)

B_o = np.array(B_o)
cocotb.log.info(f"B transposed (output):\n{B_o}")
assert np.array_equal(B_o, B_t)

0 comments on commit 48d52ca

Please sign in to comment.