Daine McNiven

Hi again @jinz2014, [hipblasDotEx(...)](https://rocm.docs.amd.com/projects/hipBLAS/en/docs-6.2.0/functions.html#hipblasdotex-batched-stridedbatched) should support mixed-precision dot with 32-bit float input and 64-bit double output/compute with the rocBLAS backend now in ROCm 6.2. You can also take a look...

Hi @Epliz, thanks for brining this up. Yes, the disparity between gemm with m == 1/n == 1 and gemv has been brought up in the past as noted by...

Hi @Epliz and @IMbackK, sorry for the delay. Looking at my past notes, it looks like the areas of most concern were where the incx parameter is large (with various...

Hi @IMbackK, Yes it's good to keep this topic up-to-date since it's been delayed for so long, thanks for your reminder. There have been no decisions made on a way...

Hi @IMbackK, I have a pull request open which redirects some calls to rocblas_gemm_ex() to use our internal gemv kernels rather than gemm kernels from Tensile where I found our...

Hi all, I'm sorry again for the extended delay in implementing this. The first-pass implementation has now been merged at https://github.com/ROCm/rocBLAS/commit/1ac1e23057a04ae280a85005f15bd8085bdd11ed and will be included in a future ROCm release....

Hi @IMbackK, Those benchmark calls aren't equivalent. With m == 1, an equivalent call to gemv needs to switch the transpose type, and pass in lda into the incx param...

@amd-garydeng Can you pull in the recent develop commits? I think that should resolve CI failures.

Assigning to Carson for gfx1100 fp32 Tensile tuning.

Hi @JakeSkelton, Would you be able to provide the output of the `./install.sh -i` command? Do you have hipBLAS' dependencies installed on your system (rocBLAS, rocSOLVER, etc.)? Thanks, Daine