Skip to content

GitLab

Commit 183db78a authored by Umang Yadav's avatar Umang Yadav
Browse files

refactor add some comments

parent 439ea40d
Hide whitespace changes
Inline Side-by-side
Showing with 39 additions and 57 deletions
src/include/migraphx/migraphx_f8_impl.hpp
View file @ 183db78a
......@@ -85,15 +85,18 @@ constexpr uint8_t cast_to_f8(T _x, bool stoch, uint32_t rng)
bias = 15;
}
uint32_t signed_inf = (sign << 7) + (((1 << we) - 1) << wm);
uint32_t signed_max = (sign << 7) + ((((1 << we) - 1) << wm) + ((1 << wm) - 1));
uint32_t signed_inf = (sign << 7) + (((1 << we) - 1) << wm);
uint32_t signed_all_ones = (sign << 7) + ((((1 << we) - 1) << wm) + ((1 << wm) - 1));
// Calcualte maximum singed value FLT_MAX, FLT_MIN
uint32_t signed_max = signed_all_ones;
if(not negative_zero_nan)
{
signed_max = (wm == 2) ? (signed_max - 4) : (signed_max - 1);
}
// Deal with inf and NaNs
if(negative_zero_nan)
if(negative_zero_nan) // For the FNUZ cases, it is simple just return NaNs
{
if(sizeof(T) == 4)
{
......@@ -114,27 +117,8 @@ constexpr uint8_t cast_to_f8(T _x, bool stoch, uint32_t rng)
{
nan_mantissa |= (nan_mantissa << 1);
}
// TODO: abstract duplicate branches
if(sizeof(T) == 4 and ((x & 0x7F800000) == 0x7F800000))
{
// infinity
if(mantissa == 0)
{
if(sign == 0)
{
return (wm == 2) ? 0x7B : 0x7E;
}
else
{
return (wm == 2) ? 0xFB : 0xFE;
}
}
else
{ // NaNs
return signed_inf + nan_mantissa;
}
}
else if(sizeof(T) == 2 and ((x & 0x7C00) == 0x7C00))
if((sizeof(T) == 4 and ((x & 0x7F800000) == 0x7F800000)) or
(sizeof(T) == 2 and ((x & 0x7C00) == 0x7C00)))
{
// infinity
if(mantissa == 0)
......@@ -160,7 +144,7 @@ constexpr uint8_t cast_to_f8(T _x, bool stoch, uint32_t rng)
// handle negative zero
if((sizeof(T) == 4 and x == 0x80000000) or (sizeof(T) == 2 and x == 0x8000))
{
if(negative_zero_nan)
if(negative_zero_nan) // For FNUZ types neg zero is just positive zero
{
return 0;
}
......@@ -170,29 +154,29 @@ constexpr uint8_t cast_to_f8(T _x, bool stoch, uint32_t rng)
}
}
// First need to check if it is normal or denorm as there is a difference of implict 1
// Then need to adjust the exponent to align with the F8 exponent, in the meanwhile, shift
// The mantissa. Then for stochastic rounding, add rng to mantissa and truncate. And for
// RNE, no need to add rng. Then probably need to check whether there is carry and adjust
// exponent and mantissa again
/* First need to check if it is normal or denorm as there is a difference of implict 1
Then need to adjust the exponent to align with the F8 exponent, in the meanwhile, shift
The mantissa. Then for stochastic rounding, add rng to mantissa and truncate. And for
RNE, no need to add rng. Then probably need to check whether there is carry and adjust
exponent and mantissa again*/
// For IEEE bias mode, the bias is 2^(k-1) -1 where k is the width of exponent bits
const int f8_bias = (1 << (we - 1)) - 1 + (negative_zero_nan ? 1 : 0);
const int f8_denormal_act_exponent = 1 - f8_bias; // actual exponent of f8 denormal
// act_exponent is the actual exponent of fp32/fp16 (after subtracting bias)
// f8_exponent is the converted f8 exponent with bias encoding
// exponent_diff is the diff between fp32/fp16 exponent and f8 exponent,
// the difference needs to be adjusted and mantissa shifted
/* act_exponent is the actual exponent of fp32/fp16 (after subtracting bias)
f8_exponent is the converted f8 exponent with bias encoding
exponent_diff is the diff between fp32/fp16 exponent and f8 exponent,
the difference needs to be adjusted and mantissa shifted*/
int act_exponent, f8_exponent, exponent_diff;
if(exponent == 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 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 */
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
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;
exponent_diff = f8_denormal_act_exponent -
act_exponent; // actual exponent is exponent-bias+1 as it is denormal
......@@ -203,10 +187,10 @@ this case, the fp16 mantissa should be shift left by 1 */
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
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 */
For example fp8 nanoo 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 */
exponent_diff = f8_denormal_act_exponent - act_exponent;
}
else
......@@ -221,10 +205,10 @@ this case, the fp16 mantissa should be shift left by 1 */
bool midpoint = (mantissa & ((1 << (mfmt - wm + exponent_diff)) - 1)) ==
(1 << (mfmt - wm + exponent_diff - 1));
/* This part is a bit tricky. The judgment of whether it is a tie needs to be done before we
shift right as shift right could rip off some residual part and make something not midpoint look
like midpoint. For example, the fp16 number 0x1002 (0 00100 0000000010), it is larger than
midpoint, but after shift right by 4 bits, it would look like midpoint.
*/
shift right as shift right could rip off some residual part and make something not midpoint look
like midpoint. For example, the fp16 number 0x1002 (0 00100 0000000010), it is larger than
midpoint, but after shift right by 4 bits, it would look like midpoint.
*/
if(exponent_diff > 0)
mantissa >>= exponent_diff;
......@@ -262,7 +246,6 @@ this case, the fp16 mantissa should be shift left by 1 */
// above range: quantize to maximum possible float of the same sign
const int max_exp = (1 << we) - (negative_zero_nan ? 1 : 2);
// TODO: this is ugly, need better way to handle out of range values
if(f8_exponent > max_exp)
{
if(clip)
......@@ -271,15 +254,11 @@ this case, the fp16 mantissa should be shift left by 1 */
}
else
{
// https://onnx.ai/onnx/technical/float8.html#cast
if(negative_zero_nan)
{
return 0x80;
}
else
{
uint32_t tmp_signed_max = (sign << 7) + ((((1 << we) - 1) << wm) + ((1 << wm) - 1));
return (wm == 2) ? signed_inf : tmp_signed_max;
}
return (wm == 2) ? signed_inf : signed_all_ones;
}
}
......@@ -300,7 +279,7 @@ constexpr T cast_from_f8(uint8_t x)
uint32_t ifNegInf = 0xFF800000;
uint32_t ifNaN = 0x7F800001;
uint32_t ifNeg0 = 0x80000000;
// TODO: need to change T for half but right now it would never called with half
fInf = detail::bit_cast<float>(ifInf);
fNegInf = detail::bit_cast<float>(ifNegInf);
fNaN = detail::bit_cast<float>(ifNaN);
......
src/include/migraphx/migraphx_float8.hpp
View file @ 183db78a
......@@ -29,7 +29,9 @@
#pragma clang diagnostic ignored "-Wc++20-extensions"
#endif // __clang__
// We are clipping in down conversion by default
// We are clipping/saturation in down conversion by default. Unclipped version is not tested and
// shouldn't be used without having enough tests.
// logic is based on clipping table from here : https://onnx.ai/onnx/technical/float8.html#cast
#define MIGRAPHX_F8_DOWNCAST_CLIPPING 1
#include <cmath>
......@@ -254,7 +256,8 @@ MIGRAPHX_FP8_BINARY_OP(*, migraphx_fp8::float8<T>)
MIGRAPHX_FP8_BINARY_OP(-, migraphx_fp8::float8<T>)
MIGRAPHX_FP8_BINARY_OP(/, migraphx_fp8::float8<T>)
MIGRAPHX_FP8_BINARY_OP(+, migraphx_fp8::float8<T>)
// TODO: Comparison ops shouldn't convert to float, maybe need to take care of rounding effects.
// TODO: Comparison ops shouldn't convert to float, need to check if need to take care of rounding
// effects.
MIGRAPHX_FP8_BINARY_OP(==, bool)
MIGRAPHX_FP8_BINARY_OP(>=, bool)
MIGRAPHX_FP8_BINARY_OP(<=, bool)
......
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