search

ARM-software / ComputeLibrary

The Compute Library is a set of computer vision and machine learning functions optimised for both Arm CPUs and GPUs using SIMD technologies.
2.87k stars 782 forks source link

NEGEMMLowpMatrixMultiplyCore: why QASYMM8 sources are not supported for F32 output #1122

Closed eshoguli closed 3 months ago

eshoguli commented 4 months ago

In accordance with documentation NEGEMMLowpMatrixMultiplyCore F32 dequantised output is supported only for QASYMM8_SIGNED values in src0 and scr1:

src0 src1 src2 dst
QASYMM8_SIGNED QASYMM8_SIGNED F32 F32

Why F32 is not supported for QASYMM8 sources? Are you going to support QASYMM8 sources in the future? Thanks!

morgolock commented 3 months ago

Hi @eshoguli

The following patch adds support for 

Valid data type configurations:
     * |src0           |src1               |src2     |dst            |
     * |:--------------|:------------------|:--------|:--------------|
     * |QASYMM8        |QASYMM8_SIGNED     |F32      |F32            |

Could you please confirm this is what you need?

eshoguli commented 3 months ago

It works from https://review.mlplatform.org/ml/ComputeLibrary, thanks!