Some tasks requires FP64 support on GPU. But there is no way to prevent sending those WUs to a GPU which only supports FP32.
From error log:
1 error detected in the compilation of "/tmp/OCLUf91im.cl".
Frontend phase failed compilation.
OpenCL device has no FP64 support
Nevertheless the Einstein app tries to run this miscompiled code and fails again with a lot of errors.
Copyright © 2024 Einstein@Home. All rights reserved.
Double precision is only
)
Double precision is only needed at the end of a task for a short time. The test is done at the beginning to determine if this can be done on the GPU or CPU. The app should fall back to CPU processing of this part of the computation. As soon as Bernd is back from holidays he will surely look into this.
Hi! I also have similar
)
Hi! I also have similar issues. My AMD HD5670 [*] does not hardware support double (fp64), neither "softfloat" when using mesa opencl driver (fglrx driver is unsupported since around 2015).
The error I get is message [**], in link https://einsteinathome.org/task/640711447
[*]01:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Redwood XT [Radeon HD 5670/5690/5730]
[**]
<core_client_version>7.6.31</core_client_version>
09:51:09 (7796): [normal]: Start of BOINC application '../../projects/einstein.phys.uwm.edu/hsgamma_FGRPB1G_1.18_x86_64-pc-linux-gnu__FGRPopencl1K-ati'.
09:51:09 (7796): [debug]: 1e+16 fp, 3.1e+09 fp/s, 3385555 s, 940h25m54s50
command line: ../../projects/einstein.phys.uwm.edu/hsgamma_FGRPB1G_1.18_x86_64-pc-linux-gnu__FGRPopencl1K-ati --inputfile ../../projects/einstein.phys.uwm.edu/LATeah0027L.dat --alpha 4.42281478648 --delta -0.0345027837249 --skyRadius 2.152570e-06 --ldiBins 15 --f0start 1092.0 --f0Band 8.0 --firstSkyPoint 0 --numSkyPoints 1 --f1dot -1e-13 --f1dotBand 1e-13 --df1dot 3.344368011e-15 --ephemdir ../../projects/einstein.phys.uwm.edu/JPLEPH --Tcoh 2097152.0 --toplist 10 --cohFollow 10 --numCells 1 --useWeights 1 --Srefinement 1 --CohSkyRef 1 --cohfullskybox 1 --mmfu 0.1 --reftime 56100 --model 0 --f0orbit 0.005 --mismatch 0.1 --demodbinary 1 --BinaryPointFile ../../projects/einstein.phys.uwm.edu/templates_LATeah0027L_1100_10129105.dat --debug 1 --device 0 -o LATeah0027L_1100.0_0_0.0_10129105_1_0.out
output files: 'LATeah0027L_1100.0_0_0.0_10129105_1_0.out' '../../projects/einstein.phys.uwm.edu/LATeah0027L_1100.0_0_0.0_10129105_1_0' 'LATeah0027L_1100.0_0_0.0_10129105_1_0.out.cohfu' '../../projects/einstein.phys.uwm.edu/LATeah0027L_1100.0_0_0.0_10129105_1_1'
09:51:09 (7796): [debug]: Flags: X64 SSE SSE2 GNUC X86 GNUX86
09:51:09 (7796): [debug]: glibc version/release: 2.23/stable
09:51:09 (7796): [debug]: Set up communication with graphics process.
boinc_get_opencl_ids returned [0x1f19e78 , 0x7f21424cbec0]
Using OpenCL platform provided by: Mesa
Using OpenCL device "AMD REDWOOD (DRM 2.49.0 / 4.11.0-041100rc6-generic, LLVM 4.0.0)" by: AMD
Max allocation limit: 751619276
Global mem size: 1073741824
OpenCL compiling FAILED! : -11 . Error message: input.cl:7:26: error: unsupported OpenCL extension 'cl_khr_fp64' - ignoring
input.cl:10:30: error: unknown type name 'double2'; did you mean 'double'?
input.cl:10:30: error: use of type 'double' requires cl_khr_fp64 extension to be enabled
OpenCL device has no FP64 support
LLVM ERROR: Cannot select: 0x21a7168: i32,ch = AtomicCmpSwap<Volatile LDST4[%1405(addrspace=1)]> 0x205bba0, 0x2020760, 0x2001ce8, 0x20023d0
0x2020760: i32,ch = CopyFromReg 0x205bba0, Register:i32 %vreg200
0x201ff40: i32 = Register %vreg200
0x2001ce8: i32,ch = CopyFromReg 0x205bba0, Register:i32 %vreg202
0x20b2c38: i32 = Register %vreg202
0x20023d0: i32 = bitcast 0x21b4af8
0x21b4af8: f32 = fadd 0x20004e8, 0x201f548
0x20004e8: f32,ch = CopyFromReg 0x205bba0, Register:f32 %vreg194
0x20b2a30: f32 = Register %vreg194
0x201f548: f32 = bitcast 0x2001ce8
0x2001ce8: i32,ch = CopyFromReg 0x205bba0, Register:i32 %vreg202
0x20b2c38: i32 = Register %vreg202
In function: kernel_ts_2_phase_diff_sorted
</stderr_txt>
]]>
I get this problem too on my
)
I get this problem too on my HD 6950 running kernel 4.10.0.30-generic, Ubuntu 16.04.3 with Mesa and OpenCl using "radeon" driver. Almost the exact quote from "FAILED!: 011" through the next several lines.
Task: https://einsteinathome.org/task/670857689
Northern Islands are not well
)
Northern Islands are not well supported running OpenCL on Ubuntu 16.04
You may wish to look over this thread
edit: I have looked over at ROCm and it appears PCIe Gen2 cards are not supported, so that option should be discarded.