Skip to content

GitLab

Commit 269ce6d1 authored by Umang Yadav's avatar Umang Yadav
Browse files

Bugfixes and additional tests

parent 26956f1d
Hide whitespace changes
Inline Side-by-side
Showing with 97 additions and 13 deletions
src/include/migraphx/float8_impl.hpp
View file @ 269ce6d1
......@@ -134,15 +134,15 @@ constexpr uint8_t cast_to_f8(T f_x, bool stoch = false, uint32_t rng = 0)
int f8_exponent = 0;
int exponent_diff = 0;
if(exponent == 0)
if(exponent == 0 and mantissa != 0)
{ // fp32/fp16 is in denormal.
/* fp32 denormal is below 2^-127 so it is usually not a concern here, we mostly concern fp16
here. In this case, f8 is usually in denormal. But there could be exceptions. fp16 denormal
has exponent bias 15 while bf8 with NANOO has exponent bias 16. It means that there are some
numbers in fp16 denormal but they are bf8 (NANOO) normals - smallest bf8 (NANOO) normal is
has exponent bias 15 while bf8 with FNUZ has exponent bias 16. It means that there are some
numbers in fp16 denormal but they are bf8 (FNUZ) normals - smallest bf8 (FNUZ) normal is
2^-15. fp16 numbers where exponent==0 (actual exponent -14) and highest bit of mantissa is 1
are bf8 (NANOO) normal. In this case, the fp16 mantissa should be shift left by 1 */
act_exponent = exponent - bias + 1;
are bf8 (FNUZ) normal. In this case, the fp16 mantissa should be shift left by 1 */
act_exponent = 1 - bias;
exponent_diff = f8_denormal_act_exponent -
act_exponent; // actual exponent is exponent-bias+1 as it is denormal
}
......@@ -152,10 +152,10 @@ constexpr uint8_t cast_to_f8(T f_x, bool stoch = false, uint32_t rng = 0)
if(act_exponent <= f8_denormal_act_exponent)
{
/* This is the case where fp32/fp16 is normal but it is in f8 denormal range.
For example fp8 nanoo mode, denormal exponent is -7, but if the fp32/fp16
For example fp8 FNUZ mode, denormal exponent is -7, but if the fp32/fp16
actual exponent is -7, it is actually larger due to the implict 1,
Therefore it needs to be adjust to -6 and mantissa shift right by 1.
So for fp32/fp16, exponent -8 is the cut point to convert to fp8 nanoo */
So for fp32/fp16, exponent -8 is the cut point to convert to fp8 FNUZ */
exponent_diff = f8_denormal_act_exponent - act_exponent;
}
else
......@@ -204,7 +204,8 @@ constexpr uint8_t cast_to_f8(T f_x, bool stoch = false, uint32_t rng = 0)
mantissa >>= (mfmt - Wm);
// above range: quantize to maximum possible float of the same sign
const int max_exp = (1 << We) - (NegativeZeroNan ? 1 : 2);
// for e5m2 case, max_exp is 14, since exp = 15 is reserved for Infs and Nans
const int max_exp = (1 << We) - ((NegativeZeroNan or Wm == 3) ? 1 : 2);
if(f8_exponent > max_exp)
{
if(Clip)
......
src/py/migraphx_py.cpp
View file @ 269ce6d1
......@@ -150,7 +150,7 @@ struct npy_format_descriptor<migraphx::fp8::fp8e4m3fnuz>
static std::string format()
{
// following: https://docs.python.org/3/library/struct.html#format-characters
return "z";
return "B";
}
static constexpr auto name() { return _("fp8e4m3fnuz"); }
};
......
test/fp8e4m3fn.cpp
View file @ 269ce6d1
......@@ -117,6 +117,27 @@ TEST_CASE(test_fp8_cast_to_float)
})});
}
TEST_CASE(test_fp8_cast_from_float)
{
std::unordered_map<float, uint8_t> test_vals = {
{{512, 0x7e}, {-512, 0xfe}, {448, 0x7e}, {-448, 0xfe},
{256, 0x78}, {-256, 0xf8}, {240, 0x77}, {-240, 0xf7},
{1e-07, 0x0}, {1e+07, 0x7e}, {1, 0x38}, {-1, 0xb8},
{0.1, 0x1d}, {0.11, 0x1e}, {0.111, 0x1e}, {0.1111, 0x1e},
{-0.1, 0x9d}, {-0.11, 0x9e}, {-0.111, 0x9e}, {-0.1111, 0x9e},
{0.2, 0x25}, {2, 0x40}, {20, 0x5a}, {200, 0x74},
{-0.2, 0xa5}, {-2, 0xc0}, {-20, 0xda}, {-200, 0xf4},
{0.5, 0x30}, {-0.5, 0xb0}, {1.17549e-38, 0x0}, {1.4013e-45, 0x0},
{0.0078125, 0x4}, {-0.0078125, 0x84}, {0.000976562, 0x0}, {-0.000976562, 0x80},
{0.000488281, 0x0}, {-0.000488281, 0x80}}};
EXPECT(bool{std::all_of(test_vals.begin(), test_vals.end(), [](const auto sample) {
return migraphx::float_equal(
migraphx::fp8::fp8e4m3fn(sample.first),
migraphx::fp8::fp8e4m3fn(sample.second, migraphx::fp8::fp8e4m3fn::from_bits()));
})});
}
TEST_CASE(test_positive_zero)
{
float zero = 0.0;
......@@ -241,7 +262,7 @@ TEST_CASE(test_binary_ops)
auto f = migraphx::fp8::fp8e4m3fn(-10.0);
EXPECT(bool{e > f});
EXPECT(bool{f < e});
EXPECT(bool(f <= e));
EXPECT(bool{f <= e});
EXPECT(bool{e >= f});
EXPECT(bool{e <= e});
EXPECT(bool{f >= f});
......
test/fp8e4m3fnuz.cpp
View file @ 269ce6d1
......@@ -138,6 +138,34 @@ TEST_CASE(test_fp8_cast_to_float)
})});
}
TEST_CASE(test_fp8_cast_from_float)
{
std::unordered_map<float, uint8_t> test_vals = {{256, 0x7f}, {-256, 0xff},
{240, 0x7f}, {-240, 0xff},
{1e-07, 0x0}, {1e+07, 0x7f},
{1, 0x40}, {-1, 0xc0},
{0.1, 0x25}, {0.11, 0x26},
{0.111, 0x26}, {0.1111, 0x26},
{-0.1, 0xa5}, {-0.11, 0xa6},
{-0.111, 0xa6}, {-0.1111, 0xa6},
{0.2, 0x2d}, {2, 0x48},
{20, 0x62}, {200, 0x7c},
{-0.2, 0xad}, {-2, 0xc8},
{-20, 0xe2}, {-200, 0xfc},
{0.5, 0x38}, {-0.5, 0xb8},
{1.17549e-38, 0x0}, {1.4013e-45, 0x0},
{0.00390625, 0x4}, {-0.00390625, 0x84},
{0.00195312, 0x2}, {-0.00195312, 0x82},
{0.000976562, 0x1}, {-0.000976562, 0x81},
{0.000488281, 0x0}, {-0.000488281, 0x0}};
EXPECT(bool{std::all_of(test_vals.begin(), test_vals.end(), [](const auto sample) {
return migraphx::float_equal(
migraphx::fp8::fp8e4m3fnuz(sample.first),
migraphx::fp8::fp8e4m3fnuz(sample.second, migraphx::fp8::fp8e4m3fnuz::from_bits()));
})});
}
TEST_CASE(test_positive_zero)
{
float zero = 0.0;
......@@ -256,7 +284,7 @@ TEST_CASE(test_binary_ops)
auto f = migraphx::fp8::fp8e4m3fnuz(-10.0);
EXPECT(bool{e > f});
EXPECT(bool{f < e});
EXPECT(bool(f <= e));
EXPECT(bool{f <= e});
EXPECT(bool{e >= f});
EXPECT(bool{e <= e});
EXPECT(bool{f >= f});
......
test/fp8e5m2.cpp
View file @ 269ce6d1
......@@ -314,6 +314,17 @@ TEST_CASE(test_fp8_cast_to_float)
})});
}
TEST_CASE(test_fp8_cast_from_float)
{
std::unordered_map<float, uint8_t> test_vals = {};
EXPECT(bool{std::all_of(test_vals.begin(), test_vals.end(), [](const auto sample) {
return migraphx::float_equal(
migraphx::fp8::fp8e5m2(sample.first),
migraphx::fp8::fp8e5m2(sample.second, migraphx::fp8::fp8e5m2::from_bits()));
})});
}
TEST_CASE(test_positive_zero)
{
float zero = 0.0;
......@@ -438,7 +449,7 @@ TEST_CASE(test_binary_ops)
auto f = migraphx::fp8::fp8e5m2(-10.0);
EXPECT(bool{e > f});
EXPECT(bool{f < e});
EXPECT(bool(f <= e));
EXPECT(bool{f <= e});
EXPECT(bool{e >= f});
EXPECT(bool{e <= e});
EXPECT(bool{f >= f});
......
test/fp8e5m2fnuz.cpp
View file @ 269ce6d1
......@@ -308,6 +308,29 @@ TEST_CASE(test_fp8_cast_to_float)
})});
}
TEST_CASE(test_fp8_cast_from_float)
{
std::unordered_map<float, uint8_t> test_vals = {
{57344, 0x7f}, {-57344, 0xff}, {60000, 0x7f}, {-60000, 0xff},
{448, 0x63}, {-448, 0xe3}, {256, 0x60}, {-256, 0xe0},
{240, 0x60}, {-240, 0xe0}, {3.05176e-05, 0x4}, {-3.05176e-05, 0x84},
{1.52588e-05, 0x2}, {-1.52588e-05, 0x82}, {7.62939e-06, 0x1}, {-7.62939e-06, 0x81},
{3.81469e-06, 0x0}, {-3.81469e-06, 0x0}, {1e+07, 0x7f}, {1, 0x40},
{-1, 0xc0}, {0.1, 0x32}, {0.11, 0x33}, {0.111, 0x33},
{0.1111, 0x33}, {-0.1, 0xb2}, {-0.11, 0xb3}, {-0.111, 0xb3},
{-0.1111, 0xb3}, {0.2, 0x36}, {2, 0x44}, {20, 0x51},
{200, 0x5e}, {-0.2, 0xb6}, {-2, 0xc4}, {-20, 0xd1},
{-200, 0xde}, {0.5, 0x3c}, {-0.5, 0xbc}, {1.17549e-38, 0x0},
{1.4013e-45, 0x0},
};
EXPECT(bool{std::all_of(test_vals.begin(), test_vals.end(), [](const auto sample) {
return migraphx::float_equal(
migraphx::fp8::fp8e5m2fnuz(sample.first),
migraphx::fp8::fp8e5m2fnuz(sample.second, migraphx::fp8::fp8e5m2fnuz::from_bits()));
})});
}
TEST_CASE(test_positive_zero)
{
float zero = 0.0;
......@@ -426,7 +449,7 @@ TEST_CASE(test_binary_ops)
auto f = migraphx::fp8::fp8e5m2fnuz(-10.0);
EXPECT(bool{e > f});
EXPECT(bool{f < e});
EXPECT(bool(f <= e));
EXPECT(bool{f <= e});
EXPECT(bool{e >= f});
EXPECT(bool{e <= e});
EXPECT(bool{f >= f});
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment