IP-Enabled C/C++ Based High Level Synthesis: A Step towards Better Designer Productivity and Design Performance
Table 7
Mapping of some representative in-built C functions.
Design
XC4VFX12FF668-10
I/P Data
LUT
FF
DSP
SRL
BRAM
CLK (ns)
Latency/throughput
IP core
sqrt (v-HLS)
32 bit integer
3105
4308
0
135
0
5.4191
752/75
No
sqrt (proposed)
32 bit integer
358
363
0
0
0
5.3231
173/1
CORDIC
div (v-HLS)
32 bit integer
4342
4105
0
136
0
7.5724
342/34
No
div (proposed) -D1
32 bit integer
1230
1095
13
262/ 1230
0
5.6744
36/5
Divider5
div (proposed) -D2
32 bit integer
1224
1011
13
262/ 1230
0
6.7694
34/5
Divider5
div (proposed) -D3
32 bit integer
1150
937
13
234/ 1150
0
6.3364
30/5
Divider5
div (proposed) -D4
32 bit integer
2279
6720
0
72/ 2279
0
6.8204
68/1
Divider5
div (proposed) -D5
32 bit integer
1284
3244
0
68/ 1284
0
5.6074
36/1
Divider6
div (proposed) -D6
32 bit integer
1253
1971
0
4/ 1253
0
7.885
37/2
Divider6
abs (v-HLS)
16 bit integer
51
19
0
0
0
2.462
1/1
No
abs (proposed)
16 bit integer
32
32
0
0
0
3.121
1/1
Yes
Clock period constraint of 4.5 ns with input jitter of 0.56 ns (clock uncertainty). Worst case and best case latencies. Maximum latency. Clock period constraint of 8 ns with input jitter of 1 ns (clock uncertainty). CORDIC configuration of integer quotient and fractional remainder with maximum pipelining. CORDIC configuration of integer quotient and integer remainder with maximum pipelining.