trixirt

Improve finding half.hpp

1

On Fedora half.hpp is installed to the system dir /usr/include. There is no cmake package, so finding the half cmake package will always fail. The source looks for half with...

silence double free warning

6

clang static analysis reports this issue on RHEL open-gpu-kernel-modules/kernel-open/nvidia/nv.c:1392:9: warning: Attempt to free released memory [unix.Malloc] NV_KFREE(nvl->irq_count, nvl->num_intr*sizeof(nv_irq_count_info_t)); ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ This is a false positive, NV_FLAG_USES_MSIX and NV_FLAG_USES_MSI are mutually exclusive....

always clear numPagesAlloc

6

Clang static analysis on RHEL reports this issue addrtree.c:1154:50: warning: The left operand of '/' is a garbage value [core.UndefinedBinaryOperatorResult] *numPagesAlloc += localNumPagesAlloc / 2; ~~~~~~~~~~~~~~~~~~ ^ _pmaAddrtreeScanContiguous() can return...

Introduce NV_KZALLOC macro

2

The pattern of p = kmalloc(...); memset(p, 0, ...); Can be reduced to p = kzalloc(...) So introduce and use NV_KZALLOC, a macro that wraps kzalloc. Signed-off-by: Tom Rix

[Issue]: RFE header only install

7

### Problem Description Building the library takes an extremely long time and only the headers may be needed. Is there or could there be a build option that only installed...

[Issue]: Does not honor cmake BUILD_SHARED

2

### Problem Description All of the libraries are hardcoded to be statically built. ex/ https://github.com/ROCm/composable_kernel/blob/develop/library/src/tensor_operation_instance/gpu/CMakeLists.txt#L312 Fedora requires libraries to be shared and versioned. ### Operating System Fedora Rawhide ### CPU...

[Issue]: RFE cmake BUILD_EXAMPLES

1

### Problem Description The logic to enable examples and testing is confusing https://github.com/ROCm/composable_kernel/blob/develop/CMakeLists.txt#L586 There should be a toplevel cmake parameter for something like BUILD_EXAMPLES. ### Operating System Fedora Rawhide ###...

[Feature]: Remove hipify as a build dependency.

3

### Suggestion Description hipify is only needed to run when the nccl codebase is updated. Can hipify be moved to a maintainers options and a ci system generate/commit the hipified...

FLASH_ATTENTION_TRITON_AMD_ENABLE is used to build the AMD triton backend here https://github.com/Dao-AILab/flash-attention/blob/main/setup.py#L65 Then it is used at runtime to import the correct AMD backend https://github.com/Dao-AILab/flash-attention/blob/main/flash_attn/flash_attn_interface.py#L11 The user of package will not...

[Issue]: Dependency on msgpack even when -DTensile_LIBRARY_FORMAT=yaml

5

### Problem Description tensilelite checks if msgpack can be imported here https://github.com/ROCm/hipBLASLt/blob/develop/tensilelite/Tensile/LibraryIO.py#L63 but there is no check here https://github.com/ROCm/hipBLASLt/blob/develop/tensilelite/Tensile/Ops/ExtOpCreateLibrary.py#L27 For distro's without a python msgpack, like RHEL, this is a...