Skip to content

Commit 977ecc0

Browse files
authored
FPGA: Downscale the QRD sample (#2332)
The sample used to have a default size of 256x256 for both Stratix10 and Agilex devices. This large size made the compiles very long for little benefit. In addition, not all Agilex devices will fit such large QRDs.
1 parent b5d1fce commit 977ecc0

File tree

2 files changed

+12
-12
lines changed

2 files changed

+12
-12
lines changed

DirectProgramming/C++SYCL_FPGA/ReferenceDesigns/qrd/CMakeLists.txt

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -84,17 +84,17 @@ if(DEVICE_FLAG MATCHES "A10")
8484
set(SEED "-Xsseed=1")
8585
elseif(DEVICE_FLAG MATCHES "S10")
8686
# S10 parameters
87-
set(ROWS_COMPONENT 256)
88-
set(COLS_COMPONENT 256)
87+
set(ROWS_COMPONENT 128)
88+
set(COLS_COMPONENT 128)
8989
set(COMPLEX 1)
90-
set(FIXED_ITERATIONS 110)
90+
set(FIXED_ITERATIONS 85)
9191
set(CLOCK_TARGET "-Xsclock=480MHz")
9292
set(SEED "-Xsseed=3")
9393
elseif(DEVICE_FLAG MATCHES "Agilex7")
9494
# Agilex 7 parameters
95-
set(ROWS_COMPONENT 256)
96-
set(COLS_COMPONENT 256)
97-
set(FIXED_ITERATIONS 110)
95+
set(ROWS_COMPONENT 128)
96+
set(COLS_COMPONENT 128)
97+
set(FIXED_ITERATIONS 85)
9898
set(COMPLEX 1)
9999
set(CLOCK_TARGET "-Xsclock=600MHz")
100100
set(SEED "-Xsseed=5")

DirectProgramming/C++SYCL_FPGA/ReferenceDesigns/qrd/README.md

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -66,7 +66,7 @@ Performance results are based on testing as of August 30, 2023.
6666
6767
| Device | Throughput
6868
|:--- |:---
69-
| Terasic’s DE10-Agilex Development Board | 15k matrices/s for complex matrices of size 256 * 256
69+
| Intel® FPGA SmartNIC N6001-PL | 50k matrices/s for complex matrices of size 128 * 128
7070

7171

7272
## Key Implementation Details
@@ -305,14 +305,14 @@ You can perform the QR decomposition of the set of matrices repeatedly. This ste
305305
306306
## Example Output
307307
308-
Example output when running on **Terasic’s DE10-Agilex Development Board** for the decomposition of 8 matrices 819200 times (each matrix consisting of 256x256 complex numbers).
308+
Example output when running on **Intel® FPGA SmartNIC N6001-PL** for the decomposition of 8 matrices 819200 times (each matrix consisting of 256x256 complex numbers).
309309
310310
```
311-
Running on device: de10_agilex : Agilex Reference Platform (aclde10_agilex0)
312-
Generating 8 random complex matrices of size 256x256
311+
Running on device: ofs_n6001 : Intel OFS Platform (ofs_ee00000)
312+
Generating 8 random complex matrices of size 128x128
313313
Running QR decomposition of 8 matrices 819200 times
314-
Total duration: 427.655 s
315-
Throughput: 15.3245k matrices/s
314+
Total duration: 130.636 s
315+
Throughput: 50.1669k matrices/s
316316
Verifying results...
317317
PASSED
318318
```

0 commit comments

Comments
 (0)