[MIGraphx EP] Sync AMD changes upstream #25338

Closed

TedThemistokleous wants to merge 33 commits intomicrosoft:mainfrom

TedThemistokleous:migraphx_ep_sync

Commits on Jul 8, 2025

Add groupnorm contrib operator
TedThemistokleous
committed
Add support for NhwcConv Contrib Operator
TedThemistokleous
committed
Gate fp8e5m2fnuz_type to ROCm 6.4 builds
TedThemistokleous
committed
inline

apwojcik
authored and
TedThemistokleous
committed
check symbol (#84)

apwojcik
authored and
TedThemistokleous
committed
Rename HIPPinnedAllocator to MIGraphXPinnedAllocator (#83)

apwojcik
authored and
TedThemistokleous
committed
Fix build error CK_BUFFER_RESOURCE_3RD_DWORD (#97)

authored and
TedThemistokleous
committed
Add Attention and SkipLayerNormalization Contrib op support

Ted Themistokleous
authored and
TedThemistokleous
committed
Align path for fused_ops cpp extension to resolve discrapency between platforms.

amd-mipantic
authored and
TedThemistokleous
committed
Update patch for gfx 1151 targets
TedThemistokleous
committed
Move provider option names to header file (#86)

apwojcik
authored and
TedThemistokleous
committed
Rename HIPPinnedAllocator to MIGraphXPinnedAllocator (#83)

apwojcik
authored and
TedThemistokleous
committed

Commits on Jul 9, 2025

Support for saving and loading mxr files in MGX EP (#105)

committed
Fix pybind input for calibration table loading
TedThemistokleous
committed
Adding hash input names and shapes to get unique graphx hash (#112)

authored and
TedThemistokleous
committed
Add null check for unique graphx hash (#121)

pnikolic-amd
authored and
TedThemistokleous
committed
Align path for fused_ops cpp extension to resolve discrapency between platforms.

amd-mipantic
authored and
TedThemistokleous
committed
Add in BF16 datatype and quantization support to MIGraphX EP (#117)
TedThemistokleous
committed
ROCm: add python arg to pass DCMAKE_HIP_ARCHITECTURES

jagadish-amd
authored and
TedThemistokleous
committed
Use SHARED lib not MODULE for MIGraphX EP (#134)

apwojcik
authored and
TedThemistokleous
committed
Correct tidy warnings on MIGraphXAllocator (#134)
TedThemistokleous
committed
Change from 'bool' to 'int' for C interface (#143)

apwojcik
authored and
TedThemistokleous
committed
Changing ifdef order for MGX and Dml

Uros Petkovic
authored and
TedThemistokleous
committed
Default submission to compute Q

hisham-hchowdhu
authored and
TedThemistokleous
committed
Clang Tidy suggestions for the MIGraphX execution provider classes (#142)

apwojcik
authored and
TedThemistokleous
committed
Simlify MIGraphX EP CMake (#146)

apwojcik
authored and
TedThemistokleous
committed
Changing ifdef order for MGX and Dml

Uros Petkovic
authored and
TedThemistokleous
committed
Change to OrtDevice::DeviceId (#147)

apwojcik
authored and
TedThemistokleous
committed
Lintrunner pass applied to patches
Ted Themistokleous
committed

Commits on Jul 10, 2025

Use #ifdef instead of #if (#149)

apwojcik
authored and
TedThemistokleous
committed
use MIGraphX ONNXRT DLL location to search for MIGX and HIP DLLs (#150)

apwojcik
authored and
TedThemistokleous
committed
change to INFO to reduce logs produced by default (#151)

apwojcik
authored and
TedThemistokleous
committed

Commits on Jul 11, 2025

Implement Clang-Tidy suggestions for MIGraphX provider factory (#148)

apwojcik
authored and
TedThemistokleous
committed