diff options
| author | Birte Kristina Friesel <birte.friesel@uos.de> | 2024-07-25 15:42:29 +0200 |
|---|---|---|
| committer | Birte Kristina Friesel <birte.friesel@uos.de> | 2024-07-25 15:42:50 +0200 |
| commit | b471467e738151e643be84d8e3757553aba524d0 (patch) | |
| tree | 9c7cc79c51b5d734141548104a113104106ccd1f /TRNS | |
| parent | b3d5b55cc00cb7977adfbde6c03599a345357767 (diff) | |
TRNS: fix write1 documentation
Diffstat (limited to 'TRNS')
| -rwxr-xr-x | TRNS/dimes-hetsim-hbm.sh | 4 | ||||
| -rwxr-xr-x | TRNS/dimes-hetsim-nmc.sh | 4 |
2 files changed, 4 insertions, 4 deletions
diff --git a/TRNS/dimes-hetsim-hbm.sh b/TRNS/dimes-hetsim-hbm.sh index 6793f8b..e2efaee 100755 --- a/TRNS/dimes-hetsim-hbm.sh +++ b/TRNS/dimes-hetsim-hbm.sh @@ -9,9 +9,9 @@ fn=log/$(hostname)/dimes-hetsim-hbm # # Input: (M_ * m) × (N_ * n) matrix # Output: (N_* n) × (M_ * m) matrix -# Step 1: transpose (M_ * m) × N_ matrix that consits of tiles of size n +# Step 1: transpose (M_ * m) × N_ matrix that consists of tiles of size n # CPU version: explicit -# DPU version: implicit (appropriate write operations to DPUs) +# DPU version: implicit (M_ * m write operations of #DPUs * n elements to DPUs) # Step 2: transpose m × n matrix; this happens N_ * M_ times. # DPU version: Each tasklet transposes a single m × n matrix / tile. # (16 × 8 tile takes up 1 KiB WRAM) diff --git a/TRNS/dimes-hetsim-nmc.sh b/TRNS/dimes-hetsim-nmc.sh index fb2640c..ef32b86 100755 --- a/TRNS/dimes-hetsim-nmc.sh +++ b/TRNS/dimes-hetsim-nmc.sh @@ -9,9 +9,9 @@ source /opt/upmem/upmem-2024.1.0-Linux-x86_64/upmem_env.sh # # Input: (M_ * m) × (N_ * n) matrix # Output: (N_* n) × (M_ * m) matrix -# Step 1: transpose (M_ * m) × N_ matrix that consits of tiles of size n +# Step 1: transpose (M_ * m) × N_ matrix that consists of tiles of size n # CPU version: explicit -# DPU version: implicit (appropriate write operations to DPUs) +# DPU version: implicit (M_ * m write operations of #DPUs * n elements to DPUs) # Step 2: transpose m × n matrix; this happens N_ * M_ times. # DPU version: Each tasklet transposes a single m × n matrix / tile. # (16 × 8 tile takes up 1 KiB WRAM) |