Skip to content

Test: CI build optimizations (CUDA compute capabilities and non-curses mode from cl/945255455)#4142

Draft
vkarampudi wants to merge 94 commits into
tensorflow:masterfrom
vkarampudi:test-gpu-docker-fixes
Draft

Test: CI build optimizations (CUDA compute capabilities and non-curses mode from cl/945255455)#4142
vkarampudi wants to merge 94 commits into
tensorflow:masterfrom
vkarampudi:test-gpu-docker-fixes

Conversation

@vkarampudi

@vkarampudi vkarampudi commented Jul 9, 2026

Copy link
Copy Markdown

No description provided.

@vkarampudi vkarampudi changed the title Test: CI build optimizations (CUDA compute capabilities and non-curse… Test: CI build optimizations (CUDA compute capabilities and non-curses mode from cl/945255455) Jul 9, 2026
vkarampudi added 28 commits July 9, 2026 22:37
…es to eliminate 5hr GPU build and 2MiB log clipping
… resolve NCCL 2.29 typeof and uint4 __stwt errors under Clang
…runtime_wrapper.h via rules_ml_toolchain patch_cmds to avoid un-sandboxed header file not found errors
- Align WORKSPACE with opensource_only/root.workspace.
- Patch XLA to fix dso_loader missing dependency declarations.
- Patch XLA to fix CUB scan policy compile error.

TAG=agy
CONV=748d06f3-9c06-4d77-9f81-b365b06d1b3a
…tch Dockerfile.

TAG=agy
CONV=748d06f3-9c06-4d77-9f81-b365b06d1b3a
…E_WRAPPERS_STWT.

TAG=agy
CONV=748d06f3-9c06-4d77-9f81-b365b06d1b3a
TAG=agy
CONV=748d06f3-9c06-4d77-9f81-b365b06d1b3a
…ules errors.

TAG=agy
CONV=748d06f3-9c06-4d77-9f81-b365b06d1b3a
TAG=agy
CONV=748d06f3-9c06-4d77-9f81-b365b06d1b3a
…lation units (.cu, .cu.cc, _cuda.cc) to preserve native Clang 18 __bf16 support
…nt16_t for GCC 10 while preserving Clang 18 native support
…-D__bf16=uint16_t through NVCC to GCC host compiler passes
…ild:cuda to define __bf16 globally for all CUDA compilation passes
…ches .cu, .cu.cc, and _cuda.cc files properly
… to typedef __bf16 as uint16_t for GCC 10 while preserving Clang 18 native __bf16 support
…b/gcc/x86_64-linux-gnu/10/include to satisfy Bazel toolchain builtin include checks
… to satisfy system header undeclared inclusion rules and compiler flag parsing
…lib/gcc/x86_64-linux-gnu/10/include/ and pre-include via cuda_bf16_fix.h
…asses to resolve cuda_bf16_fix.h from /usr/include
…64-linux-gnu/10/include/ and remove -isystem/usr/include to avoid sandbox root checks
…r macro across all C++ and CUDA compilation passes
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant