Mihai Preda
Mihai Preda
Using Linux kernel 5.13-rc4, ROCm 4.2.0 OpenCL only (without dkms), Radeon VII, I see this segfault when compiling with ROCm 4.2.0. Everything works correctly with ROCm 3.3.0. ``` Using host...
ROCm 2.2 OpenCL: when clBuildProgram() is invoked with an invalid path in -save-temps , it terminates the process abruptly after displaying the message: LLVM ERROR: IO failure on output stream:...
ROCm 3.1, Radeon VII, looking at the generated code I often see this block: ``` s_waitcnt lgkmcnt(0) s_barrier s_waitcnt lgkmcnt(0) ``` It seems that the second s_waitcnt is not needed,...
### Problem Description On Ubuntu 22.04 w. linux kernel 6.7.5, Radeon Pro VII. The behavior of the OpenCL compilation flag -save-temps seems altered on ROCm 6.1 vs. all previous ROCm...
On Ubuntu 22.04, kernel 6.7.9, ROCm 6.1.0 (RC), Radeon Pro VII. In brief: when creating a second command-queue (that is not even used at all) one thread starts eating 100%...
A function such as: double sum2(double x, double y) { return 2 * (x + y); } could be compiled to a single VOP3 GCN instructions such as: ``` v_add_f64...
Consider the following example, which implements a complex multiply-add function named "mad" with a signature that clashes with the builtin OpenCL math function mad(gentype, gentype, gentype) (where gentype includes double2)....
Is there a way, in the OpenCL source, to identify whether it's the ROCm OpenCL that is compiling the code (as opposed to, e.g. amdgpu-pro or something else). In particular,...