illwieckz wrote:
The `/usr/include/hsa/hsa.h` file is provided by the stock Ubuntu
`libhsa-runtime-dev` package.
The `/opt/rocm-6.1.2/include/hsa/hsa.h` file is provided by the AMD
`hsa-rocr-dev` package.
The AMD `rocm-hip-runtime` package depends on both `hsa-rocr-dev` and
`libhsa-runtime-dev
illwieckz wrote:
> I don't think AMD handles the actual packaging.
They do: https://repo.radeon.com
It happens that Ubuntu provides an old 5.2.3 `libhsa-runtime-dev` and it seems
to be a dependency of some package provided by the AMD `repo.radeon.com` own
repository providing ROCm 6.1.2…
> I
@@ -394,7 +400,7 @@ createFlowFunction(const
BinaryFunction::BasicBlockOrderType &BlockOrder) {
void matchWeightsByHashes(BinaryContext &BC,
const BinaryFunction::BasicBlockOrderType
&BlockOrder,
const yaml::bolt::BinaryFunc
@@ -394,7 +400,7 @@ createFlowFunction(const
BinaryFunction::BasicBlockOrderType &BlockOrder) {
void matchWeightsByHashes(BinaryContext &BC,
const BinaryFunction::BasicBlockOrderType
&BlockOrder,
const yaml::bolt::BinaryFunc
@@ -575,10 +583,16 @@ void preprocessUnreachableBlocks(FlowFunction &Func) {
/// Decide if stale profile matching can be applied for a given function.
/// Currently we skip inference for (very) large instances and for instances
/// having "unexpected" control flow (e.g., having
@@ -394,7 +400,7 @@ createFlowFunction(const
BinaryFunction::BasicBlockOrderType &BlockOrder) {
void matchWeightsByHashes(BinaryContext &BC,
const BinaryFunction::BasicBlockOrderType
&BlockOrder,
const yaml::bolt::BinaryFunc
jhuber6 wrote:
> This doesn't make sense, the ROCm packages may be just badly made by AMD.
>
> When I temporarily remove `/usr/include/hsa` without uninstalling
> `libhsa-runtime-dev` **AND** while using
> `-DLIBOMPTARGET_FORCE_DLOPEN_LIBHSA=ON`, I can build LLVM18 without patch.
>
> With `-D
