MIOpen

MIOpen copied to clipboard

ConvHipImplicitGemmWrwV4R4Xdlops failure due to compiler issue

12

Testing this PR: https://github.com/ROCmSoftwarePlatform/MIOpen/pull/413 This case fails ``` MIOpenDriver convfp16 -n 256 -c 512 -H 8 -W 3 -k 2048 -y 1 -x 1 -p 0 -q 0 -u 1...

asroy

Instructions: How to build in windows?

9

johnnynunez

This PR is for new, upgraded, Op4dTensorGeneric kernel, this is part of porting kernels from OCL to HIP Below is performance (speed-up and drops in performance) comparison between new Op4dTensorGeneric...

novakovicdj

Since this is declared on the SQLiteDB base, the GetCached method is currently unused. Due to returning a & to the map find, and also since SQLiteDB connections cannot be...

BrianHarrisonAMD

Forbid inefficient TensorDescriptor initialization

4

Initializing TensorDescriptor from `std::vector` is very inefficient due to extra checks and multiple intermediate vector, since internally `std::vector` is used. Changed all the initializations to the native `size_t`, removed constructors...

CAHEK7

[CONV] fix naive conv kernel for large tensors

5

for larger tensor I was seeing gdims[0] : 38,654,705,664 globalWorkSizeX : 4,294,967,295 (max allowed by uint32_t) ~~MaxGridDimX : 2,147,483,647~~ gdims[0] was exceeding ~~MaxGridDimX~~ globalWorkSizeX for below driver command. `./bin/MIOpenDriver convbfp16...

bghimireamd

Bumps [rocm-docs-core[api_reference]](https://github.com/ROCm/rocm-docs-core) from 1.21.1 to 1.27.0. Release notes Sourced from rocm-docs-core[api_reference]'s releases. v1.27.0 (2025-10-24) Feat target latest_versions.txt from new_data branch parse modified latest_version.txt add version list to rocm-ds flavor add...

dependabot[bot]

## Proposed changes With this PR MIOpen should be able to use heuristics for the three `conv_hip_implicit_gemm_3d_grouped_*_xdlops`. I have made bunch of changes to how the metadata is structured, we...

amd-bartgips