Implement several unimplemented integer operations and fix casting-related bugs #353
|
@ -0,0 +1,184 @@
|
|||
@extern
|
||||
def output_bool(x: bool):
|
||||
...
|
||||
|
||||
@extern
|
||||
def output_int32(x: int32):
|
||||
...
|
||||
|
||||
@extern
|
||||
def output_int64(x: int64):
|
||||
...
|
||||
|
||||
@extern
|
||||
def output_uint32(x: uint32):
|
||||
...
|
||||
|
||||
@extern
|
||||
def output_uint64(x: uint64):
|
||||
...
|
||||
|
||||
@extern
|
||||
def output_float64(x: float):
|
||||
...
|
||||
|
||||
def u32_min() -> uint32:
|
||||
return uint32(0)
|
||||
|
||||
def u32_max() -> uint32:
|
||||
return ~uint32(0)
|
||||
|
||||
def i32_min() -> int32:
|
||||
return int32(1 << 31)
|
||||
|
||||
def i32_max() -> int32:
|
||||
return int32(~(1 << 31))
|
||||
|
||||
def u64_min() -> uint64:
|
||||
return uint64(0)
|
||||
|
||||
def u64_max() -> uint64:
|
||||
return ~uint64(0)
|
||||
|
||||
def i64_min() -> int64:
|
||||
return int64(1) << int64(63)
|
||||
|
||||
def i64_max() -> int64:
|
||||
return ~(int64(1) << int64(63))
|
||||
|
||||
def test_u32_bnot():
|
||||
output_uint32(~uint32(0))
|
||||
|
||||
def test_u64_bnot():
|
||||
output_uint64(~uint64(0))
|
||||
|
||||
def test_conv_from_i32():
|
||||
for x in [
|
||||
i32_min(),
|
||||
i32_min() + 1,
|
||||
-1,
|
||||
0,
|
||||
1,
|
||||
i32_max() - 1,
|
||||
i32_max()
|
||||
]:
|
||||
output_int64(int64(x))
|
||||
output_uint32(uint32(x))
|
||||
output_uint64(uint64(x))
|
||||
output_float64(float(x))
|
||||
|
||||
def test_conv_from_u32():
|
||||
for x in [
|
||||
u32_min(),
|
||||
u32_min() + uint32(1),
|
||||
u32_max() - uint32(1),
|
||||
u32_max()
|
||||
]:
|
||||
output_uint64(uint64(x))
|
||||
output_int32(int32(x))
|
||||
output_int64(int64(x))
|
||||
output_float64(float(x))
|
||||
|
||||
def test_conv_from_i64():
|
||||
for x in [
|
||||
i64_min(),
|
||||
i64_min() + int64(1),
|
||||
int64(-1),
|
||||
int64(0),
|
||||
int64(1),
|
||||
i64_max() - int64(1),
|
||||
i64_max()
|
||||
]:
|
||||
output_int32(int32(x))
|
||||
output_uint64(uint64(x))
|
||||
output_uint32(uint32(x))
|
||||
output_float64(float(x))
|
||||
|
||||
def test_conv_from_u64():
|
||||
for x in [
|
||||
u64_min(),
|
||||
u64_min() + uint64(1),
|
||||
u64_max() - uint64(1),
|
||||
u64_max()
|
||||
]:
|
||||
output_uint32(uint32(x))
|
||||
output_int64(int64(x))
|
||||
output_int32(int32(x))
|
||||
output_float64(float(x))
|
||||
|
||||
def test_f64toi32():
|
||||
for x in [
|
||||
float(i32_min()) - 1.0,
|
||||
float(i32_min()),
|
||||
float(i32_min()) + 1.0,
|
||||
-1.5,
|
||||
-0.5,
|
||||
0.5,
|
||||
1.5,
|
||||
float(i32_max()) - 1.0,
|
||||
float(i32_max()),
|
||||
float(i32_max()) + 1.0
|
||||
]:
|
||||
output_int32(int32(x))
|
||||
|
||||
def test_f64toi64():
|
||||
for x in [
|
||||
float(i64_min()),
|
||||
float(i64_min()) + 1.0,
|
||||
-1.5,
|
||||
-0.5,
|
||||
0.5,
|
||||
1.5,
|
||||
# 2^53 is the highest integral power-of-two of which uint64 and float have a one-to-one correspondence
|
||||
float(uint64(2) ** uint64(52)) - 1.0,
|
||||
float(uint64(2) ** uint64(52)),
|
||||
float(uint64(2) ** uint64(52)) + 1.0,
|
||||
]:
|
||||
output_int64(int64(x))
|
||||
|
||||
def test_f64tou32():
|
||||
for x in [
|
||||
-1.5,
|
||||
float(u32_min()) - 1.0,
|
||||
-0.5,
|
||||
float(u32_min()),
|
||||
0.5,
|
||||
float(u32_min()) + 1.0,
|
||||
1.5,
|
||||
float(u32_max()) - 1.0,
|
||||
float(u32_max()),
|
||||
float(u32_max()) + 1.0
|
||||
]:
|
||||
output_uint32(uint32(x))
|
||||
|
||||
def test_f64tou64():
|
||||
for x in [
|
||||
-1.5,
|
||||
float(u64_min()) - 1.0,
|
||||
-0.5,
|
||||
float(u64_min()),
|
||||
0.5,
|
||||
float(u64_min()) + 1.0,
|
||||
1.5,
|
||||
# 2^53 is the highest integral power-of-two of which uint64 and float have a one-to-one correspondence
|
||||
float(uint64(2) ** uint64(52)) - 1.0,
|
||||
float(uint64(2) ** uint64(52)),
|
||||
float(uint64(2) ** uint64(52)) + 1.0,
|
||||
]:
|
||||
output_uint64(uint64(x))
|
||||
|
||||
def run() -> int32:
|
||||
test_u32_bnot()
|
||||
test_u64_bnot()
|
||||
|
||||
test_conv_from_i32()
|
||||
test_conv_from_u32()
|
||||
test_conv_from_i64()
|
||||
test_conv_from_u64()
|
||||
|
||||
test_f64toi32()
|
||||
test_f64toi64()
|
||||
test_f64tou32()
|
||||
test_f64tou64()
|
||||
|
||||
return 0
|
Loading…
Reference in New Issue