Intel Agilex User Manual page 45

3. Intel Agilex Variable Precision DSP Blocks Operational Modes
UG-20213 | 2019.04.02
Table 18. Differences between Flushed, Extended, and Bfloat Formats
Features
Input format
(sign.exponent.mantissa)
FP16 operation format
(sign.exponent.mantissa)
Input width
Minimum representable
exponent
FP16 Subnormal
Exception flags
Invalid flag behavior
Rounding
3.2.2.2. Sum of Two FP16 Multiplication Mode
This mode performs a summation of two half-precision multiplication and provide a
single-precision result:
fp32_result = (fp16_mult_top_a*fp16_mult_top_b) +
(fp16_mult_bot_a*fp16_mult_bot_b)
The following are exception flags supported in flushed and bfloat16 formats:
•
fp16_mult_top_invalid
•
fp16_mult_top_inexact
•
fp16_mult_top_overflow
•
fp16_mult_top_underflow
Send Feedback
Flushed
1.5.10
1.5.10
16 bit
5'h01 - 5'h0f = -14
No support for subnormal.
Subnormal result is flushed
to zero.
Overflow, underflow,
inexact, and invalid
Asserted when there is an
ill-defined operation
Round to nearest even
(RNE)
Extended
1.5.10
1.8.10
16 bit
8'h01 - 8'h7f = -126
Subnormal results can be
represented as normal
numbers
Infinite, zero, inexact, and
invalid
Asserted when there is an
ill-defined operation or a
qNaN input
RNE:
• if both FP16 operands
are normal numbers
• if one of the FP16
operands is a subnormal
number and mantissa
product is ≥ 1
• if one of the FP16
operands is a subnormal
number and mantissa
product =
"0.1111111111|
1xxxxxxxxx"
• when using adder/
subtractor operations
Round to zero(RZ)
• if both FP16 operands
are subnormal numbers
• if one of the FP16
operands is a subnormal
number and mantissa
product is ≤ 1
® ™
Intel Agilex Variable Precision DSP Blocks User Guide
Bfloat16/Bfloat 16+
1.8.7 or 1.8.10 (Bfloat16+)
1.8.10
16 or 19 bit (Bfloat16+)
8'h01 - 8'h7f = -126
No support for subnormal.
Subnormal result is flushed
to zero.
Overflow, underflow,
inexact, and invalid
Asserted when there is an
ill-defined operation
RZ
45