gpu_compute: Support loading BLIT kernels
authorMichael LeBeane <Michael.Lebeane@amd.com>
Thu, 18 Oct 2018 21:45:27 +0000 (17:45 -0400)
committerAnthony Gutierrez <anthony.gutierrez@amd.com>
Fri, 17 Jul 2020 16:13:59 +0000 (16:13 +0000)
commit1d816250f87d950e1dde244c0abc3f846ce37b15
treed3d108ddb29db7b9a21deb42b513bd7548177417
parent72e9324ef0db8a5e5a833207da56b16419618f21
gpu_compute: Support loading BLIT kernels

The BLIT kernels used to implement DMA through the shaders don't fill
out all of the standard fields in an amd_kernel_code_t object.  This
patch modifies the code object parsing logic to support these new
kernels.

BLIT kernels are used in APUs when using ROCm memcopies for certain size
buffers, and are used for dGPUs when the SDMA engines are disabled.

Change-Id: Id4e667474d05e311097dbec443def07dfad14a79
Reviewed-on: https://gem5-review.googlesource.com/c/public/gem5/+/29959
Maintainer: Anthony Gutierrez <anthony.gutierrez@amd.com>
Tested-by: kokoro <noreply+kokoro@google.com>
Reviewed-by: Matt Sinclair <mattdsinclair@gmail.com>
src/gpu-compute/gpu_command_processor.cc
src/gpu-compute/hsa_queue_entry.hh