OpenCL 2.0 support for Ellesmere (RX580)
sudden6 opened this issue · 9 comments
Hi all,
is there OpenCL 2.0 support for the Ellesmere (RX580) device? Specifically I want to use enqueue_kernel(...) in my OpenCL kernel.
The card is advertised on the AMD Website as "OpenCL 2.0", but clinfo gives me the following (confusing) output:
Device Version OpenCL 1.2
Driver Version 3186.0 (HSA1.1,LC)
Device OpenCL C Version OpenCL C 2.0
Device Type GPU
Device Board Name (AMD) Ellesmere [Radeon RX 470/480/570/570X/580/580X/590]
Additionally when I try to compile an OpenCL kernel with –cl-std=CL2.0 the driver gives me the error code -66, which according to the StreamHPC listing means "the compiler options specified by options are invalid.", but clinfo says OpenCL C 2.0 should be supported (at least in my understanding).
To summarize:
- Can the RX580 hardware support
enqueue_kernel(...)(device-side enqueue)? (I'd guess so) - Will ROCm-OpenCL support
enqueue_kernel(...)on the RX580? - Is there any overview which driver/hardware supports which OpenCL version? I didn't find anything like that, even for amdgpu-pro
- Is there any workaround in the mean-time?
- Which device would I need to buy to get
enqueue_kernel(...)to work?
- RX580 supports devices enqueue and reports OCL2.0 support under Windows. RX580 supported the same functionality(OCL2.0) under Linux in the previous amdgpu-pro releases with PAL backend
- Unfortunately only Gfx9 HW(Vega generation) supports OCL2.0 on ROCm due to the differences in the firmware usage between PAL and ROCm backends. And RX580 is Gfx8 HW
- You could rollback to the old amdgpu-pro driver with OCL based on PAL backend
- You would need a board from Radeon RX Vega series
Thank you for your response!
I will try the amdgpu-pro stack with PAL and see if it works there.
I tried with amdgpu-pro version 20.40 and there's no support for OpenCL 2.0 on my GPU.
When I do amdgpu-pro-install --opencl=pal I get no devices with clinfo and when I try with amdgpu-pro-install --opencl=legacy,pal it detects my GPU, but doesn't show OpenCL 2.0 as supported.
Is there a specific version I need?
Is there a better place to ask about amdgpu-pro problems?
It seems to work in Windows, but if possible I'd prefer to develop on Linux.
There are some complications around Linux support, especially with the older generations.
- Try "export GPU_ENABLE_PAL=1"
- If that won't work, then try just PAL installation: amdgpu-pro-install --opencl=pal and then "export GPU_ENABLE_PAL=1"
I hope that will help.
There are some complications around Linux support, especially with the older generations.
* Try "export GPU_ENABLE_PAL=1" * If that won't work, then try just PAL installation: amdgpu-pro-install --opencl=pal and then "export GPU_ENABLE_PAL=1" I hope that will help.
Tried that with 19.50, up to 20.40. Was close to get it working but it segfaults halfway clinfo output, although recognizes my RX580 as a Device Version OpenCL 2.0. Other drivers (pro ocl legacy/orca and rocm) work fine, but only OpenCL 1.2 so seems we're SOL in regards to OCL2.
Where did you find about the GPU_ENABLE_PAL env var, didnt find any ref to it?
This custom build operates OpenCL fine, I tested with OpenCL-OpenGL interop samples as well. I've absolutely no idea why APT packages don't sport a build with the off-by-default Polaris support enabled.
Ubuntu 20.04, ROCm 5.0 works like this
mate@GL702ZC:~$ sudo amdgpu-install --usecase=graphics,hiplibsdk,opencl --no-dkms --no-32
mate@GL702ZC:~$ wget https://github.com/xuhuisheng/rocm-gfx803/releases/download/rocm500/rocm-opencl_2.0.0-local_amd64.deb
mate@GL702ZC:~$ sudo dpkg -i ./rocm-opencl_2.0.0-local_amd64.deb
mate@GL702ZC:~$ /opt/rocm/bin/rocm-smi
======================= ROCm System Management Interface =======================
================================= Concise Info =================================
GPU Temp AvgPwr SCLK MCLK Fan Perf PwrCap VRAM% GPU%
0 65.0c 17.011W 608Mhz 2000Mhz 0% auto 68.0W 19% 0%
================================================================================
WARNING: One or more commands failed
============================= End of ROCm SMI Log ==============================
mate@GL702ZC:~$ /opt/rocm/bin/rocminfo
ROCk module is loaded
=====================
HSA System Attributes
=====================
Runtime Version: 1.1
System Timestamp Freq.: 1000.000000MHz
Sig. Max Wait Duration: 18446744073709551615 (0xFFFFFFFFFFFFFFFF) (timestamp count)
Machine Model: LARGE
System Endianness: LITTLE
==========
HSA Agents
==========
*******
Agent 1
*******
Name: AMD Ryzen 7 1700 Eight-Core Processor
Uuid: CPU-XX
Marketing Name: AMD Ryzen 7 1700 Eight-Core Processor
Vendor Name: CPU
Feature: None specified
Profile: FULL_PROFILE
Float Round Mode: NEAR
Max Queue Number: 0(0x0)
Queue Min Size: 0(0x0)
Queue Max Size: 0(0x0)
Queue Type: MULTI
Node: 0
Device Type: CPU
Cache Info:
L1: 32768(0x8000) KB
Chip ID: 0(0x0)
Cacheline Size: 64(0x40)
Max Clock Freq. (MHz): 3000
BDFID: 0
Internal Node ID: 0
Compute Unit: 16
SIMDs per CU: 0
Shader Engines: 0
Shader Arrs. per Eng.: 0
WatchPts on Addr. Ranges:1
Features: None
Pool Info:
Pool 1
Segment: GLOBAL; FLAGS: FINE GRAINED
Size: 32805700(0x1f49344) KB
Allocatable: TRUE
Alloc Granule: 4KB
Alloc Alignment: 4KB
Accessible by all: TRUE
Pool 2
Segment: GLOBAL; FLAGS: KERNARG, FINE GRAINED
Size: 32805700(0x1f49344) KB
Allocatable: TRUE
Alloc Granule: 4KB
Alloc Alignment: 4KB
Accessible by all: TRUE
Pool 3
Segment: GLOBAL; FLAGS: COARSE GRAINED
Size: 32805700(0x1f49344) KB
Allocatable: TRUE
Alloc Granule: 4KB
Alloc Alignment: 4KB
Accessible by all: TRUE
ISA Info:
*******
Agent 2
*******
Name: gfx803
Uuid: GPU-XX
Marketing Name: Radeon RX 580 Series
Vendor Name: AMD
Feature: KERNEL_DISPATCH
Profile: BASE_PROFILE
Float Round Mode: NEAR
Max Queue Number: 128(0x80)
Queue Min Size: 4096(0x1000)
Queue Max Size: 131072(0x20000)
Queue Type: MULTI
Node: 1
Device Type: GPU
Cache Info:
L1: 16(0x10) KB
Chip ID: 26591(0x67df)
Cacheline Size: 64(0x40)
Max Clock Freq. (MHz): 1077
BDFID: 3072
Internal Node ID: 1
Compute Unit: 36
SIMDs per CU: 4
Shader Engines: 4
Shader Arrs. per Eng.: 1
WatchPts on Addr. Ranges:4
Features: KERNEL_DISPATCH
Fast F16 Operation: TRUE
Wavefront Size: 64(0x40)
Workgroup Max Size: 1024(0x400)
Workgroup Max Size per Dimension:
x 1024(0x400)
y 1024(0x400)
z 1024(0x400)
Max Waves Per CU: 40(0x28)
Max Work-item Per CU: 2560(0xa00)
Grid Max Size: 4294967295(0xffffffff)
Grid Max Size per Dimension:
x 4294967295(0xffffffff)
y 4294967295(0xffffffff)
z 4294967295(0xffffffff)
Max fbarriers/Workgrp: 32
Pool Info:
Pool 1
Segment: GLOBAL; FLAGS: COARSE GRAINED
Size: 4194304(0x400000) KB
Allocatable: TRUE
Alloc Granule: 4KB
Alloc Alignment: 4KB
Accessible by all: FALSE
Pool 2
Segment: GROUP
Size: 64(0x40) KB
Allocatable: FALSE
Alloc Granule: 0KB
Alloc Alignment: 0KB
Accessible by all: FALSE
ISA Info:
ISA 1
Name: amdgcn-amd-amdhsa--gfx803
Machine Models: HSA_MACHINE_MODEL_LARGE
Profiles: HSA_PROFILE_BASE
Default Rounding Mode: NEAR
Default Rounding Mode: NEAR
Fast f16: TRUE
Workgroup Max Size: 1024(0x400)
Workgroup Max Size per Dimension:
x 1024(0x400)
y 1024(0x400)
z 1024(0x400)
Grid Max Size: 4294967295(0xffffffff)
Grid Max Size per Dimension:
x 4294967295(0xffffffff)
y 4294967295(0xffffffff)
z 4294967295(0xffffffff)
FBarrier Max Size: 32
*** Done ***
mate@GL702ZC:~$ /opt/rocm/opencl/bin/clinfo
Number of platforms: 1
Platform Profile: FULL_PROFILE
Platform Version: OpenCL 2.2 AMD-APP.dbg (3406.0)
Platform Name: AMD Accelerated Parallel Processing
Platform Vendor: Advanced Micro Devices, Inc.
Platform Extensions: cl_khr_icd cl_amd_event_callback
Platform Name: AMD Accelerated Parallel Processing
Number of devices: 1
Device Type: CL_DEVICE_TYPE_GPU
Vendor ID: 1002h
Board name: Radeon RX 580 Series
Device Topology: PCI[ B#12, D#0, F#0 ]
Max compute units: 36
Max work items dimensions: 3
Max work items[0]: 1024
Max work items[1]: 1024
Max work items[2]: 1024
Max work group size: 256
Preferred vector width char: 4
Preferred vector width short: 2
Preferred vector width int: 1
Preferred vector width long: 1
Preferred vector width float: 1
Preferred vector width double: 1
Native vector width char: 4
Native vector width short: 2
Native vector width int: 1
Native vector width long: 1
Native vector width float: 1
Native vector width double: 1
Max clock frequency: 1077Mhz
Address bits: 64
Max memory allocation: 3650722200
Image support: Yes
Max number of images read arguments: 128
Max number of images write arguments: 8
Max image 2D width: 16384
Max image 2D height: 16384
Max image 3D width: 16384
Max image 3D height: 16384
Max image 3D depth: 8192
Max samplers within kernel: 26591
Max size of kernel argument: 1024
Alignment (bits) of base address: 1024
Minimum alignment (bytes) for any datatype: 128
Single precision floating point capability
Denorms: No
Quiet NaNs: Yes
Round to nearest even: Yes
Round to zero: Yes
Round to +ve and infinity: Yes
IEEE754-2008 fused multiply-add: Yes
Cache type: Read/Write
Cache line size: 64
Cache size: 16384
Global memory size: 4294967296
Constant buffer size: 3650722200
Max number of constant args: 8
Local memory type: Scratchpad
Local memory size: 65536
Max pipe arguments: 16
Max pipe active reservations: 16
Max pipe packet size: 3650722200
Max global variable size: 3650722200
Max global variable preferred total size: 4294967296
Max read/write image args: 64
Max on device events: 1024
Queue on device max size: 8388608
Max on device queues: 1
Queue on device preferred size: 262144
SVM capabilities:
Coarse grain buffer: Yes
Fine grain buffer: Yes
Fine grain system: No
Atomics: No
Preferred platform atomic alignment: 0
Preferred global atomic alignment: 0
Preferred local atomic alignment: 0
Kernel Preferred work group size multiple: 64
Error correction support: 0
Unified memory for Host and Device: 0
Profiling timer resolution: 1
Device endianess: Little
Available: Yes
Compiler available: Yes
Execution capabilities:
Execute OpenCL kernels: Yes
Execute native function: No
Queue on Host properties:
Out-of-Order: No
Profiling : Yes
Queue on Device properties:
Out-of-Order: Yes
Profiling : Yes
Platform ID: 0x7f3cc2c2bc80
Name: gfx803
Vendor: Advanced Micro Devices, Inc.
Device OpenCL C version: OpenCL C 2.0
Driver version: 3406.0 (HSA1.1,LC)
Profile: FULL_PROFILE
Version: OpenCL 1.2
Extensions: cl_khr_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_fp16 cl_khr_gl_sharing cl_amd_device_attribute_query cl_amd_media_ops cl_amd_media_ops2 cl_khr_image2d_from_buffer cl_khr_subgroups cl_khr_depth_images cl_amd_copy_buffer_p2p cl_amd_assembly_program
I've absolutely no idea why APT packages don't sport a build with the off-by-default Polaris support enabled.
Usually distro packager try to follow upstream as they have not enough resources supporting otherwise untested configurations. Also there are some open rocBLAS bugs for gfx803 (aka "Polaris") which won't get fixed (at least not by AMD):
Granted, these bugs don't affect OpenCL but you might get the idea why distro packagers are not too keen on enabling otherwise unsupported configurations.
That being said at least in Fedora we could enable OpenCL for Polaris (once we figure out how to package it in a sane way) if we had trusted users who can run test builds and have some test suites to ensure everything keeps working. You could consider joining the debian-ai mailing list to provide feedback once they finished packaging OpenCL for debian.
This "rant" wasn't aimed at Debian/Canonical or other distro maintainers. AMD offers their own APT repo which is their recommended way of installing their software. Note that they don't endorse using AUR or other 3rd party packaging efforts but documentation suggests using their repo on the supported platforms. Period.
gfx803 is still in a "partial support" state, but it isn't clear whether partial means things may not work as intended or if we disable entire APIs due to lack of testing/assets towards bug fixes.
All I'm saying is that this "partial support" technique doesn't seem to do much service to users. It's weird how AMD says they don't test the OpenCL runtime on Polaris.
Having that said I can run occasional tests, but I don't have the bandwidth to do much more than that.