github/concrete
1
1
Fork 0
mirror of https://github.com/zama-ai/concrete.git synced 2026-02-09 03:55:04 -05:00
Code Issues Packages Projects Releases Wiki Activity

feat: add support for 8-bit integers

Browse Source
This commit is contained in:
Umut
2022-03-28 10:45:06 +03:00
parent a9d2733230
commit d19514bb97
4 changed files with 49 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
concrete/common/mlir/utils.py
View File

@@ -17,8 +17,8 @@ from ..operator_graph import OPGraph
from ..representation import intermediate
from ..representation.intermediate import Conv2D, IntermediateNode
# TODO: should come from compiler, through an API, #402
ACCEPTABLE_MAXIMAL_BITWIDTH_FROM_CONCRETE_LIB = 7
# TODO: should be removed as the supported bit-width is now dynamic
ACCEPTABLE_MAXIMAL_BITWIDTH_FROM_CONCRETE_LIB = 8
def check_node_compatibility_with_mlir(

2
docs/user/howto/reduce_needed_precision.md
View File

@@ -38,7 +38,7 @@ Traceback (most recent call last):
update_bit_width_for_mlir(op_graph)
File "/home/default/Documents/Projects/Zama/hdk/concrete/common/mlir/utils.py", line 204, in update_bit_width_for_mlir
raise RuntimeError(
RuntimeError: max_bit_width of some nodes is too high for the current version of the compiler (maximum must be 7) which is not compatible with:
RuntimeError: max_bit_width of some nodes is too high for the current version of the compiler (maximum must be 8) which is not compatible with:
%0 = x # EncryptedScalar<uint3>
%1 = 42 # ClearScalar<uint6>

26
tests/common/mlir/test_conversion_helpers.py
View File

@@ -34,24 +34,6 @@ def test_integer_to_mlir_type(integer, is_encrypted, expected_mlir_type_str):
assert str(integer_to_mlir_type(ctx, integer, is_encrypted)) == expected_mlir_type_str
@pytest.mark.parametrize(
"integer,is_encrypted,expected_error_message",
[
pytest.param(SignedInteger(32), True, "can't create eint with the given width"),
pytest.param(UnsignedInteger(32), True, "can't create eint with the given width"),
],
)
def test_fail_integer_to_mlir_type(integer, is_encrypted, expected_error_message):
"""Test function for failed integer to MLIR type conversion."""
with pytest.raises(ValueError) as excinfo:
with Context() as ctx, Location.unknown():
concretelang.register_dialects(ctx)
integer_to_mlir_type(ctx, integer, is_encrypted)
assert str(excinfo.value) == expected_error_message
@pytest.mark.parametrize(
"value,expected_mlir_type_str",
[
@@ -92,14 +74,6 @@ def test_value_to_mlir_type(value, expected_mlir_type_str):
EncryptedTensor(Float(32), shape=(2, 3)),
"EncryptedTensor<float32, shape=(2, 3)> is not supported for MLIR conversion",
),
pytest.param(
EncryptedScalar(UnsignedInteger(32)),
"EncryptedScalar<uint32> is not supported for MLIR conversion",
),
pytest.param(
EncryptedTensor(UnsignedInteger(32), shape=(2, 3)),
"EncryptedTensor<uint32, shape=(2, 3)> is not supported for MLIR conversion",
),
],
)
def test_fail_value_to_mlir_type(value, expected_error_message):

55
tests/numpy/test_compile.py
View File

@@ -181,6 +181,39 @@ def random_lut_7b(x):
return table[x]
def random_lut_8b(x):
"""8-bit random table lookup"""
# fmt: off
table = LookupTable(
[
212, 38, 101, 238, 40, 147, 221, 83, 237, 177, 130, 149, 173,
122, 217, 118, 77, 109, 180, 223, 84, 39, 83, 161, 44, 177,
95, 28, 9, 171, 163, 114, 178, 50, 57, 142, 6, 78, 153,
184, 170, 40, 94, 45, 237, 50, 220, 45, 97, 212, 174, 42,
190, 93, 153, 212, 71, 153, 253, 207, 24, 171, 179, 86, 5,
99, 92, 217, 134, 151, 154, 17, 95, 238, 194, 21, 73, 243,
158, 242, 216, 176, 251, 46, 23, 20, 18, 107, 49, 87, 149,
150, 96, 96, 245, 164, 56, 67, 63, 209, 70, 145, 95, 55,
52, 34, 13, 71, 149, 86, 254, 50, 85, 95, 124, 179, 190,
108, 137, 116, 245, 207, 83, 131, 128, 39, 232, 138, 198, 164,
218, 19, 170, 134, 237, 165, 231, 178, 170, 65, 7, 225, 160,
152, 33, 114, 0, 82, 232, 68, 3, 212, 107, 16, 140, 66,
177, 102, 48, 234, 204, 164, 54, 23, 120, 164, 172, 5, 107,
50, 210, 139, 228, 94, 188, 169, 90, 113, 77, 252, 235, 104,
124, 70, 65, 186, 29, 59, 113, 147, 71, 17, 65, 255, 15,
88, 24, 229, 74, 16, 90, 122, 146, 176, 33, 250, 175, 170,
75, 81, 62, 120, 127, 109, 89, 37, 98, 152, 157, 18, 166,
121, 248, 19, 243, 159, 161, 133, 166, 154, 126, 175, 65, 8,
203, 252, 82, 145, 189, 143, 123, 245, 104, 197, 16, 222, 236,
24, 139, 132, 32, 242, 94, 117, 231, 98
]
)
# fmt: on
return table[x]
def multi_lut(x):
"""2-bit multi table lookup"""
@@ -771,6 +804,7 @@ def test_compile_function_multiple_outputs(
pytest.param(lambda x: -x + 50, ((0, 20),), ["x"]),
pytest.param(lambda x: numpy.dot(x, 2), ((0, 20),), ["x"]),
pytest.param(lambda x: numpy.dot(2, x), ((0, 20),), ["x"]),
pytest.param(lambda x: x + 128, ((0, 127),), ["x"]),
],
)
def test_compile_and_run_correctness(
@@ -1867,6 +1901,7 @@ def test_compile_and_run_matmul_correctness(
pytest.param(identity_lut_generator(5), (5,), ["x"], id="identity function (5-bit)"),
pytest.param(identity_lut_generator(6), (6,), ["x"], id="identity function (6-bit)"),
pytest.param(identity_lut_generator(7), (7,), ["x"], id="identity function (7-bit)"),
pytest.param(identity_lut_generator(8), (8,), ["x"], id="identity function (8-bit)"),
pytest.param(random_lut_1b, (1,), ["x"], id="random function (1-bit)"),
pytest.param(random_lut_2b, (2,), ["x"], id="random function (2-bit)"),
pytest.param(random_lut_3b, (3,), ["x"], id="random function (3-bit)"),
@@ -1874,6 +1909,7 @@ def test_compile_and_run_matmul_correctness(
pytest.param(random_lut_5b, (5,), ["x"], id="random function (5-bit)"),
pytest.param(random_lut_6b, (6,), ["x"], id="random function (6-bit)"),
pytest.param(random_lut_7b, (7,), ["x"], id="random function (7-bit)"),
pytest.param(random_lut_8b, (8,), ["x"], id="random function (8-bit)"),
pytest.param(small_fused_table, (5,), ["x"], id="small fused table (5-bits)"),
],
)
@@ -1918,15 +1954,15 @@ def test_compile_and_run_lut_correctness(
"function,table,bit_width",
[
pytest.param(*negative_identity_smaller_lut_generator(n), n, id=f"smaller ({n}-bit)")
for n in range(1, 8)
for n in range(1, 9)
]
+ [
pytest.param(*negative_identity_lut_generator(n), n, id=f"normal ({n}-bit)")
for n in range(1, 8)
for n in range(1, 9)
]
+ [
pytest.param(*negative_identity_bigger_lut_generator(n), n, id=f"bigger ({n}-bit)")
for n in range(1, 7)
for n in range(1, 8)
]
+ [
pytest.param(*weird_lut(3), 3, id="weird"),
@@ -2391,7 +2427,7 @@ def test_compile_too_high_bitwidth(default_compilation_configuration):
}
# A bit too much
input_ranges = [(0, 100), (0, 28)]
input_ranges = [(100, 200), (100, 200)]
with pytest.raises(RuntimeError) as excinfo:
compile_numpy_function(
@@ -2405,12 +2441,12 @@ def test_compile_too_high_bitwidth(default_compilation_configuration):
str(excinfo.value)
== """
max_bit_width of some nodes is too high for the current version of the compiler (maximum must be 7) which is not compatible with:
max_bit_width of some nodes is too high for the current version of the compiler (maximum must be 8) which is not compatible with:
%0 = x # EncryptedScalar<uint7>
%1 = y # EncryptedScalar<uint5>
%2 = add(%0, %1) # EncryptedScalar<uint8>
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 8 bits is not supported for the time being
%0 = x # EncryptedScalar<uint8>
%1 = y # EncryptedScalar<uint8>
%2 = add(%0, %1) # EncryptedScalar<uint9>
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 9 bits is not supported for the time being
return %2
""".strip() # noqa: E501 # pylint: disable=line-too-long
@@ -2518,6 +2554,7 @@ def test_wrong_inputs(default_compilation_configuration):
"function,input_ranges,list_of_arg_names",
[
pytest.param(lambda x: (x + (-27)) + 32, ((0, 10),), ["x"]),
pytest.param(lambda x: (x + (-128)) + 373, ((0, 10),), ["x"]),
pytest.param(lambda x: ((-3) * x) + (100 - (x + 1)), ((0, 10),), ["x"]),
pytest.param(
lambda x, y: (-1) * x + (-2) * y + 40,