mirror of
https://github.com/zama-ai/concrete.git
synced 2026-02-08 03:25:05 -05:00
test(compiler): add multi precision tests
This commit is contained in:
aPere3
committed by
Quentin Bourgerie
Quentin Bourgerie
parent
955dea6fcd
commit
0ba154fca8
@@ -255,7 +255,7 @@ $(FIXTURE_CPU_DIR)/%.yaml: tests/end_to_end_fixture/%_gen.py FORCE
|
||||
$(FIXTURE_CPU_DIR)/bug_report.yaml:
|
||||
unzip -o $(FIXTURE_CPU_DIR)/bug_report.zip -d $(FIXTURE_CPU_DIR)
|
||||
|
||||
generate-cpu-tests: $(FIXTURE_CPU_DIR)/end_to_end_leveled.yaml $(FIXTURE_CPU_DIR)/end_to_end_apply_lookup_table.yaml $(FIXTURE_CPU_DIR)/end_to_end_linalg_apply_lookup_table.yaml $(FIXTURE_CPU_DIR)/bug_report.yaml $(FIXTURE_CPU_DIR)/end_to_end_round.yaml
|
||||
generate-cpu-tests: $(FIXTURE_CPU_DIR)/end_to_end_leveled.yaml $(FIXTURE_CPU_DIR)/end_to_end_apply_lookup_table.yaml $(FIXTURE_CPU_DIR)/end_to_end_linalg_apply_lookup_table.yaml $(FIXTURE_CPU_DIR)/bug_report.yaml $(FIXTURE_CPU_DIR)/end_to_end_round.yaml $(FIXTURE_CPU_DIR)/end_to_end_multi_precision.yaml
|
||||
|
||||
SECURITY_TO_TEST=128
|
||||
run-end-to-end-tests: build-end-to-end-tests generate-cpu-tests
|
||||
@@ -266,6 +266,10 @@ run-end-to-end-tests: build-end-to-end-tests generate-cpu-tests
|
||||
$(foreach security,$(SECURITY_TO_TEST),$(BUILD_DIR)/tools/concretelang/tests/end_to_end_tests/end_to_end_test \
|
||||
--backend=cpu --security-level=$(security) --jit $(FIXTURE_CPU_DIR)/*.yaml;)
|
||||
|
||||
run-multi-precision-tests: build-end-to-end-tests generate-cpu-tests
|
||||
$(foreach security,$(SECURITY_TO_TEST),$(BUILD_DIR)/tools/concretelang/tests/end_to_end_tests/end_to_end_test \
|
||||
--backend=cpu --security-level=$(security) --optimizer-strategy=dag-multi --jit $(FIXTURE_CPU_DIR)/end_to_end_multi_precision.yaml;)
|
||||
|
||||
### end-to-end-tests GPU
|
||||
|
||||
FIXTURE_GPU_DIR=tests/end_to_end_fixture/tests_gpu
|
||||
|
||||
@@ -0,0 +1,133 @@
|
||||
import argparse
|
||||
from platform import mac_ver
|
||||
|
||||
import numpy as np
|
||||
|
||||
from end_to_end_linalg_leveled_gen import P_ERROR
|
||||
|
||||
def get_identity_lut(precision):
|
||||
return list(range(2**precision))
|
||||
|
||||
def get_random_of_precision(precision):
|
||||
return np.random.randint(2**precision)
|
||||
|
||||
def get_simple_layered_inputs(precisions):
|
||||
max_result = 2**max(precisions)
|
||||
while True:
|
||||
tentative_inputs = list(map(get_random_of_precision, precisions))
|
||||
if (sum(tentative_inputs) < max_result):
|
||||
return tentative_inputs
|
||||
|
||||
def get_double_layered_inputs(small_precisions, big_precisions):
|
||||
small_max_result = 2**max(small_precisions)
|
||||
while True:
|
||||
tentative_small_inputs = list(map(get_random_of_precision, small_precisions))
|
||||
if (sum(tentative_small_inputs) < small_max_result):
|
||||
break
|
||||
big_max_result = 2**max(big_precisions)
|
||||
while True:
|
||||
tentative_big_inputs = list(map(get_random_of_precision, big_precisions))
|
||||
if (sum(tentative_small_inputs) + sum(tentative_big_inputs) < big_max_result):
|
||||
return tentative_small_inputs + tentative_big_inputs
|
||||
|
||||
def generate_simple_layered(precisions):
|
||||
max_prec = max(precisions)
|
||||
str_name = "_".join(map(str, precisions))
|
||||
arg_names = list(map(lambda x: f"%arg{x}" , range(len(precisions))))
|
||||
arg_types = list(map(lambda x: f"!FHE.eint<{x}>" , precisions))
|
||||
max_type = f"!FHE.eint<{max_prec}>"
|
||||
str_args = ", ".join(map(lambda x: f"{x[0]}: {x[1]}", zip(arg_names, arg_types)))
|
||||
print(f"description: multi_precision_simple_layered_{str_name}")
|
||||
print( "program: |")
|
||||
print(f" func.func @main({str_args}) -> {max_type} {{")
|
||||
print(f" %0 = \"FHE.zero\"() : () -> {max_type}")
|
||||
for i in range(len(precisions)):
|
||||
current_precision = precisions[i]
|
||||
current_type = arg_types[i]
|
||||
current_arg = arg_names[i]
|
||||
identity_lut = get_identity_lut(current_precision)
|
||||
print(f" %lut{i} = arith.constant dense<{identity_lut}> : tensor<{len(identity_lut)}xi64>")
|
||||
print(f" %bs{i} = \"FHE.apply_lookup_table\"({current_arg}, %lut{i}): ({current_type}, tensor<{len(identity_lut)}xi64>) -> ({max_type})")
|
||||
print(f" %{i+1} = \"FHE.add_eint\"(%{i}, %bs{i}) : ({max_type}, {max_type}) -> ({max_type})")
|
||||
print(f" return %{i+1} : {max_type}")
|
||||
print(" }")
|
||||
print(f"p-error: {P_ERROR}")
|
||||
print("tests:")
|
||||
for i in range(8):
|
||||
print(" - inputs:")
|
||||
inputs = get_simple_layered_inputs(precisions)
|
||||
for input in inputs:
|
||||
print(f" - scalar: {input}")
|
||||
print(" outputs:")
|
||||
print(f" - scalar: {sum(inputs)}")
|
||||
print("---")
|
||||
|
||||
def generate_double_layered(small_precisions, big_precisions):
|
||||
assert(max(small_precisions) <= min(big_precisions))
|
||||
precisions = small_precisions + big_precisions
|
||||
str_name = "_".join(map(str, precisions))
|
||||
small_max_prec = max(small_precisions)
|
||||
max_prec = max(precisions)
|
||||
arg_names = list(map(lambda x: f"%arg{x}" , range(len(precisions))))
|
||||
arg_types = list(map(lambda x: f"!FHE.eint<{x}>" , precisions))
|
||||
max_type = f"!FHE.eint<{max_prec}>"
|
||||
small_max_type = f"!FHE.eint<{small_max_prec}>"
|
||||
str_args = ", ".join(map(lambda x: f"{x[0]}: {x[1]}", zip(arg_names, arg_types)))
|
||||
print(f"description: multi_precision_double_layered_{str_name}")
|
||||
print( "program: |")
|
||||
print(f" func.func @main({str_args}) -> {max_type} {{")
|
||||
print(f" %0 = \"FHE.zero\"() : () -> {small_max_type}")
|
||||
for i in range(len(small_precisions)):
|
||||
current_precision = precisions[i]
|
||||
current_type = arg_types[i]
|
||||
current_arg = arg_names[i]
|
||||
identity_lut = get_identity_lut(current_precision)
|
||||
print(f" %lut{i} = arith.constant dense<{identity_lut}> : tensor<{len(identity_lut)}xi64>")
|
||||
print(f" %bs{i} = \"FHE.apply_lookup_table\"({current_arg}, %lut{i}): ({current_type}, tensor<{len(identity_lut)}xi64>) -> ({small_max_type})")
|
||||
print(f" %{i+1} = \"FHE.add_eint\"(%{i}, %bs{i}) : ({small_max_type}, {small_max_type}) -> ({small_max_type})")
|
||||
|
||||
identity_lut = get_identity_lut(current_precision)
|
||||
print(f" %lut{i+1} = arith.constant dense<{identity_lut}> : tensor<{len(identity_lut)}xi64>")
|
||||
print(f" %{i+2} = \"FHE.apply_lookup_table\"(%{i+1}, %lut{i+1}): ({small_max_type}, tensor<{len(identity_lut)}xi64>) -> ({max_type})")
|
||||
|
||||
for j in range(len(big_precisions)):
|
||||
current_precision = precisions[i+1+j]
|
||||
current_type = arg_types[i+1+j]
|
||||
current_arg = arg_names[i+1+j]
|
||||
identity_lut = get_identity_lut(current_precision)
|
||||
print(f" %lut{i+2+j} = arith.constant dense<{identity_lut}> : tensor<{len(identity_lut)}xi64>")
|
||||
print(f" %bs{i+2+j} = \"FHE.apply_lookup_table\"({current_arg}, %lut{i+2+j}): ({current_type}, tensor<{len(identity_lut)}xi64>) -> ({max_type})")
|
||||
print(f" %{i+2+j+1} = \"FHE.add_eint\"(%{i+2+j}, %bs{i+2+j}) : ({max_type}, {max_type}) -> ({max_type})")
|
||||
|
||||
print(f" return %{i+2+j+1} : {max_type}")
|
||||
print(" }")
|
||||
print(f"p-error: {P_ERROR}")
|
||||
print("tests:")
|
||||
for i in range(8):
|
||||
print(" - inputs:")
|
||||
inputs = get_double_layered_inputs(small_precisions, big_precisions)
|
||||
for input in inputs:
|
||||
print(f" - scalar: {input}")
|
||||
print(" outputs:")
|
||||
print(f" - scalar: {sum(inputs)}")
|
||||
print("---")
|
||||
|
||||
def generate(args):
|
||||
print("# /!\ DO NOT EDIT MANUALLY THIS FILE MANUALLY")
|
||||
print("# /!\ THIS FILE HAS BEEN GENERATED")
|
||||
np.random.seed(0)
|
||||
generate_simple_layered([3,4])
|
||||
generate_simple_layered([3,4,5])
|
||||
generate_double_layered([3,4],[4,5])
|
||||
generate_double_layered([3,4],[5,6])
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
CLI = argparse.ArgumentParser()
|
||||
CLI.add_argument(
|
||||
"--minimal",
|
||||
help="Specify whether to generate minimal tests only",
|
||||
type=bool,
|
||||
default=False,
|
||||
)
|
||||
generate(CLI.parse_args())
|
||||
Reference in New Issue
Block a user