David Dixon

@aviallon we've had some changes that improve memory usage over the last month or so. All changes should be in develop.

@trixirt I see this in the docs: Joblib has no mandatory dependencies besides Python (supported versions are 3.8+ I suspect there is an older version of python on RHEL8 presumably...

@mysoreanoop try unbundling first. ``` > clang-offload-bunder --type=o --input=my.co --list hipv4-amdgcn-amd-amdhsa-unknown-gfx90a host-x86_64-unknown-linux-gnu- > clang-offload-bunder --targets=hipv4-amdgcn-amd-amdhsa-unknown-gfx90a --type=o --input=my.co --output=temp.o --unbundle > nm -D temp.o 0000000000017740 R Cijk_Alik_Bljk_BBS_BH_Bias_HAS_SAV_UserArgs_MT96x224x32_MI16x16x1_SN_LDSB1_AFC1_AFEM1_AFEM1_ASEM1_CLR1_CADS0_DTVA0_EPS0_GRPM1_GRVWA4_GRVWB4_GSUAMB_ISA90a_IU1_K1_LBSPPA128_LBSPPB128_LBSPPM0_LPA4_LPB4_LPM0_LRVW4_LWPMn1_MIAV0_MIWT3_7_MO40_NTn1_NTA0_NTB0_NTC0_NTD0_NTM0_NEPBS2_NLCA1_NLCB1_ONLL1_PGR2_PLR1_PKA1_SIA3_SS1_SPO1_SRVW0_SSO0_SVW1_TLDS1_ULSGRO0_USL1_UIOFGRO0_USFGROn1_VSn1_VWA1_VWB1_WSGRA1_WSGRB1_WS64_WG32_8_1.kd ``` I haven't tried...

@mysoreanoop please replace .dat with .co in the title. The .dat file is a metadata file that uses the msgpack format. It doesn't have any assembly. The .co file is...

> I rebuilt hipblaslt with --keep-build-tmp, and I do see that the new build directory (46 GB) is larger than without that flag, built for gfx90a (and 942), I don't...

> A low level related question: > > 1. When I invoke hipblaslt-bench with some parameters, would different waves have potentially different execution traces? Cause the disassembly file that I...