ROCm Performance Primitives supported functionalities and variants

ROCm Performance Primitives supported functionalities and variants#

The following tables show the CPU and GPU support for ROCm Performance Primitives (RPP) functionalities and variants.

CPU support is also referred to as HOST support, and GPU support is provided through the HIP backend.

The functionalities are grouped to match the RPP API headers described in the API reference. For call syntax, see Run an RPP tensor augmentation.

Color augmentations#

Type

CPU

GPU

brightness

✅

✅

gamma correction

✅

✅

blend

✅

✅

hue

✅

✅

saturation

✅

✅

color twist

✅

✅

color jitter

✅

❌

color cast

✅

✅

exposure

✅

✅

contrast

✅

✅

lut

✅

✅

color temperature

✅

✅

histogram equalize

✅

✅

Effects augmentations#

Type

CPU

GPU

gridmask

✅

✅

spatter

✅

✅

salt and pepper noise

✅

✅

shot noise

✅

✅

gaussian noise

✅

✅

non-linear blend

✅

✅

water

✅

✅

ricap

✅

✅

vignette

✅

✅

jitter

✅

✅

erase

✅

✅

random erase

✅

✅

glitch

✅

✅

rain

✅

✅

pixelate

✅

✅

fog

✅

✅

posterize

✅

✅

solarize

✅

✅

snow

✅

✅

channel dropout

✅

✅

cutout dropout

✅

✅

grid dropout

✅

✅

coarse dropout

✅

✅

Geometric augmentations#

Type

CPU

GPU

crop

✅

✅

crop mirror normalize

✅

✅

crop and patch

✅

✅

flip

✅

✅

resize

✅

✅

resize mirror normalize

✅

✅

resize crop mirror

✅

✅

rotate

✅

✅

warp affine

✅

✅

warp perspective

✅

✅

lens correction

✅

✅

fisheye

✅

✅

phase

✅

✅

slice

✅

✅

remap

✅

✅

transpose

✅

✅

concat

✅

✅

jpeg compression distortion

✅

✅

Morphological operations#

Type

CPU

GPU

erode

✅

✅

dilate

✅

✅

Filter augmentations#

Type

CPU

GPU

box filter

✅

✅

median filter

✅

✅

gaussian filter

✅

✅

sobel filter

✅

✅

emboss

✅

✅

Arithmetic operations#

Type

CPU

GPU

add scalar

✅

✅

subtract scalar

✅

✅

multiply scalar

✅

✅

fused multiply add scalar

✅

✅

magnitude

✅

✅

log

✅

✅

log1p

✅

✅

tensor add

✅

✅

tensor subtract

✅

✅

tensor multiply

✅

✅

tensor divide

✅

✅

Statistical operations#

Type

CPU

GPU

tensor sum

✅

✅

tensor min

✅

✅

tensor max

✅

✅

tensor mean

✅

✅

tensor stddev

✅

✅

normalize

✅

✅

threshold

✅

✅

Bitwise operations#

Type

CPU

GPU

bitwise AND

✅

✅

bitwise OR

✅

✅

bitwise XOR

✅

✅

bitwise NOT

✅

✅

tensor AND tensor

✅

✅

tensor OR tensor

✅

✅

tensor XOR tensor

✅

✅

Data exchange operations#

Type

CPU

GPU

copy

✅

✅

channel permute

✅

✅

color to greyscale

✅

✅

YUV to RGB

❌

✅

YUV to RGB (cubic vertical upsampling)

❌

✅

YUV to RGB (linear vertical upsampling)

❌

✅

Audio augmentations#

Type

CPU

GPU

non-silent region detection

✅

✅

to decibels

✅

✅

pre-emphasis filter

✅

✅

down mixing

✅

✅

spectrogram

✅

✅

mel filter bank

✅

✅

resample

✅

✅

audio tensor add tensor

✅

✅

audio tensor multiply scalar

✅

✅

3D image (voxel) augmentations#

Type

CPU

GPU

flip (voxel)

✅

✅

gaussian noise (voxel)

✅

✅

add scalar

✅

✅

subtract scalar

✅

✅

multiply scalar

✅

✅

fused multiply add scalar

✅

✅

slice

✅

✅

normalize

✅

✅
Contents