-cl-llvm-stats: Print stats from optimization passes.
-cl-vec={none|loop|slp|all}: Enables Loop and/or SLP Vectorization passes prior to VECZ vectorization.
-cl-precache-local-sizes=<sizes>: Specifies a list of local work group sizes which should be used to pre-cache results of the stages of kernel compilation that require a local size. These stages normally take place when an ND range is enqueued, but by completing them early and storing the results for later lookup (pre-caching) them we can remove all compilation overhead from the enqueue function. A local work group size is expressed in the form of up to three comma separated numbers, multiple sizes can be given in a colon separated list. The same restrictions apply to these sizes as apply to sizes passed to clEnqueueNDRangeKernel, see the spec for that entry point for info on those constraints.

Revision History

Rev	Date	Author	Changes
1	2016/03/11	Ewan Crawford	-g build flag
2	2016/06/25	Ewan Crawford	-S build option
3	2018/03/27	Guillaume Marques	-cl-wfv and -cl-llvm-stats flags
4	2018/03/27	Neil Henning	-cl-dma
5	2019/05/13	Guillaume Marques	-cl-wfv auto option
6	2019/07/21	Amy Worthington	-cl-vec
7	2020/06/17	Stefano Cherubin	-cl-wi-order
8	2020/10/23	Aaron Greig	-cl-precache-local-sizes
9	2021/07/23	Ori Sky Farrell	Remove -cl-wfv option
10	2021/11/12	Kenneth Benzie	Remove -g and -S
11	2022/10/25	Colin Davidson	Remove -cl-dma
12	2020/02/08	Amy Worthington	Remove -cl-wi-order

Rate this Guide

Kernel Debug - cl_codeplay_kernel_debug