Miroirs/db48x
1
0
Fork 0
mirror of https://gitlab.com/c3d/db48x.git synced 2024-09-29 05:36:58 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

doc: Update performance numbers for 1M loops

Browse source 
Ran 1M loops with hardware-accelerated, updated the numbers.

Signed-off-by: Christophe de Dinechin <christophe@dinechin.org>
This commit is contained in:
Christophe de Dinechin 2024-03-06 07:27:46 +01:00
parent 97cc7f90bd
commit 18352159b0
1 changed files with 22 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

24
doc/6-Performance.md
View file

@ -98,6 +98,9 @@ VP = Variable Precision
ID = Intel Decimal Library ID = Intel Decimal Library
HW = Hardware-accelerated (`float` or `double` types) HW = Hardware-accelerated (`float` or `double` types)
### Variable Precision vs. Intel Decimal
For 100000 loops, we see that the variable-precision implementation at 24-digit For 100000 loops, we see that the variable-precision implementation at 24-digit
is roughly 10 times slower than the fixed precision implementation at 34 digits is roughly 10 times slower than the fixed precision implementation at 34 digits
(128 bits). (128 bits).
@ -108,7 +111,8 @@ is roughly 10 times slower than the fixed precision implementation at 34 digits
| 0.5.2 (ID) | 215421 | 143412 | | 0.5.2 (ID) | 215421 | 143412 |
For 1000 loops, comparing variable-precision decimal with the earlier Intel decimal For 1000 loops, comparing variable-precision decimal with the earlier Intel
decimal
| Version | DM32 ms | DM42 ms | | Version | DM32 ms | DM42 ms |
|--------------|---------|---------| |--------------|---------|---------|
@ -117,6 +121,9 @@ For 1000 loops, comparing variable-precision decimal with the earlier Intel deci
| 0.6.4 (VP6) | 6905 | 5623 | | 0.6.4 (VP6) | 6905 | 5623 |
| 0.5.2 (ID) | 2154 | 1434 | | 0.5.2 (ID) | 2154 | 1434 |
### 1000 loops in various implementations
Time in millisecond for 1000 loops: Time in millisecond for 1000 loops:
| DM32 Version | HW7 | HW16 | VP6 | VP12 | VP24 | VP36 | | DM32 Version | HW7 | HW16 | VP6 | VP12 | VP24 | VP36 |
@ -133,10 +140,23 @@ Time in millisecond for 1000 loops:
| 0.6.0 (Note) | | | | | 17685 | | | 0.6.0 (Note) | | | | | 17685 | |
| 0.5.2 (ID) | 1434 | | | | | | | 0.5.2 (ID) | 1434 | | | | | |
Note: Results for 0.6.0 with variable precision are rtificially good because Note: Results for 0.6.0 with variable precision are artificially good because
intermediate computations were not made with increased precision. intermediate computations were not made with increased precision.
### 1M loops and iPhone results
1 million loops (tests performed with 0.7.1 while on battery):
| Version | Time (ms) | Result |
|----------------|-----------|---------------------------------------------|
| DM32 HW7 | 1748791 | 1'384'348.25 |
| DM32 HW16 | 2188113 | 1'395'612.15872'53834'6 |
| DM42 HW7 | 605102 | 1'384'348.25 |
| DM42 HW16 | 806730 | 1'395'612.15872'53834'6 |
## Drawing `sin X` with `FunctionPlot` ## Drawing `sin X` with `FunctionPlot`
| Configuration | DM32 ms | DM42 ms | | Configuration | DM32 ms | DM42 ms |