philipturner / openmm-metal Goto Github PK

OpenMM Metal Plugin

License: MIT License

Shell 0.15% CMake 0.80% C++ 39.09% C 48.80% Metal 11.14% Python 0.02%

openmm-metal's Introduction

OpenMM Metal Plugin

This plugin adds the Metal platform that accelerates OpenMM on Metal 3 GPUs. It supports Apple, AMD, and Intel GPUs running macOS Ventura or higher. The plugin uses Apple's OpenCL compatiblity layer (cl2Metal) to translate OpenCL kernels to AIR.

Vendors:

Apple Mac GPUs are supported in current and future releases.
AMD and Intel GPUs are in long-term service mode. The v1.0.0 tag on GitHub works, but newer versions are failing tests on Intel Macs.

Usage

In Finder, create an empty folder and right-click it. Select New Terminal at Folder from the menu, which launches a Terminal window. Copy and enter the following commands one-by-one.

conda install -c conda-forge openmm
git clone https://github.com/openmm/openmm
cd openmm && git fetch && git checkout 8.1_branch

cd ../
git clone https://github.com/philipturner/openmm-metal
cd openmm-metal

# requires password
bash build.sh --install --quick-tests

Next, you will benchmark OpenCL against Metal. In your originally empty folder, enter openmm/examples. Arrange the files by name and locate benchmark.py. Then, open the file with TextEdit or Xcode. Modify the top of the script as shown below:

from __future__ import print_function
import openmm.app as app
import openmm as mm
import openmm.unit as unit
from datetime import datetime
import argparse
import os

# What you need to add:
mm.Platform.loadPluginsFromDirectory(mm.Platform.getDefaultPluginsDirectory())

# Rest of the code:
def cpuinfo():
    """Return CPU info"""

Press Cmd + S to save. Back at the Terminal window, type the following commands:

cd ../
cd openmm
cd examples
python3 benchmark.py --test apoa1rf --seconds 15 --platform OpenCL
python3 benchmark.py --test apoa1rf --seconds 15 --platform HIP

OpenMM's current energy minimizer hard-codes checks for the CUDA, OpenCL, and HIP platforms. The Metal backend is currently labeled HIP everywhere to bypass this limitation. The plugin name will change to Metal after the next OpenMM version release, which fixes the issue. To prevent source-breaking changes, check for both the HIP and Metal backends in your client code.

Sequential Throughput

Due to its dependency on OpenMM 8.0.0, the Metal plugin can't implement an optimization inside the main code base that speeds up small systems. Therefore, there is an environment variable that can force-disable the usage of nearest neighbor lists inside Metal kernels. Turning off the neighbor list can substantially improve simulation speed for systems with under 3000 atoms.

export OPENMM_METAL_USE_NEIGHBOR_LIST=0 # accepted, force-disables neighor list
export OPENMM_METAL_USE_NEIGHBOR_LIST=1 # accepted, forces usage of neighbor list
export OPENMM_METAL_USE_NEIGHBOR_LIST=2 # runtime crash
unset OPENMM_METAL_USE_NEIGHBOR_LIST # accepted, automatically chooses whether to use the list

Scaling behavior (Water Box, Amber Forcefield, PME):

Atoms	Force No List	Force Use List	Auto
6	1380 ns/day	1120 ns/day	1380 ns/day
21	1260 ns/day	1120 ns/day	1260 ns/day
96	1210 ns/day	1040 ns/day	1210 ns/day
306	1010 ns/day	860 ns/day	1010 ns/day
774	739 ns/day	637 ns/day	739 ns/day
2661	490 ns/day	438 ns/day	490 ns/day
4158	347 ns/day	340 ns/day	340 ns/day

The above table is not a great example of the speedup possible by eliminating the sequential throughput bottleneck. Typically, this will provide an order of magnitude speedup. As in, not 18% speedup, but 18x speedup. Why that is not happening needs to be investigated further.

Profiling

The Metal plugin can use the OpenCL queue profiling API to extract performance data about kernels. This API prevents batching of commands into command buffers, and reports the GPU start and end time of each buffer. The results are written to stdout in the JSON format used by https://ui.perfetto.dev.

export OPENMM_METAL_PROFILE_KERNELS=0 # accepted, does not profile
export OPENMM_METAL_PROFILE_KERNELS=1 # accepted, prints data to the console
export OPENMM_METAL_PROFILE_KERNELS=2 # runtime crash
unset OPENMM_METAL_PROFILE_KERNELS # accepted, does not profile

Reducing Energy

By default, energy summation is serialized among a single threadgroup. The reduceEnergy kernel consumes a significant proportion of execution time for small systems. You can make reduction occur across more than one threadgroup with the following variable.

export OPENMM_METAL_REDUCE_ENERGY_THREADGROUPS=1 # accepted, 1 threadgroup used
export OPENMM_METAL_REDUCE_ENERGY_THREADGROUPS=3 # accepted, 3 threadgroups used
export OPENMM_METAL_REDUCE_ENERGY_THREADGROUPS=-1 # runtime crash
unset OPENMM_METAL_REDUCE_ENERGY_THREADGROUPS # accepted, 1024 threadgroups used

Scaling behavior (Water Box, Amber Forcefield, No Cutoff):

top of table shows number of threadgroups
cells show time to finish the reduceEnergy kernel

Atoms	OpenMM OpenCL	Metal TG=1	Metal TG=2	Metal TG=4	Metal TG=8
96	~65 µs	9 µs	7 µs	6 µs	6 µs
306	~65 µs	9 µs	7 µs	-	-
774	~65 µs	9 µs	7 µs	-	-
2661	~65 µs	9 µs	7 µs	-	-
4158	~65 µs	9 µs	7 µs	6 µs	6 µs

You can increase the precision of energy measurements by setting METAL_REDUCE_ENERGY_THREADGROUPS to a very large number. The GPU splits the energy into a large number of partial sums, which the CPU casts from FP32 -> FP64 and accumulates in mixed precision. An improvement in energy conservation has not yet been observed, because the tested system was too small to occupy many threadgroups. It is theorized to take effect with very large systems.

A recent study proved there is a ~4x reduction in standard deviation of singlepoint energies, provided a massive number of threadgroups is used. In addition the latency appeared to remain at 6 µs in Perfetto. Even with 1024 threadgroups dispatched, the execution time of reduceEnergy was about 1/1000 of the time spent computing forces. In light of this study, the default parameter for OPENMM_METAL_REDUCE_ENERGY_THREADGROUPS was changed from 1 to 1024.

https://gist.github.com/philipturner/20df5482f75b07d91466b972a1a46cd5

Scaling

At the several million atom range, OpenMM starts to experience $O(n^2)$ scaling. The impact of this scaling is relatively minor for the Metal platform, as Apple GPUs calculate the $O(n^2)$ part much faster than CUDA GPUs. The "large blocks" algorithm delays the onset of $O(n^2)$ scaling. It provides a net speedup for most systems regardless of scale, but especially at 1,000,000+ atoms.

export OPENMM_METAL_USE_LARGE_BLOCKS=0 # accepted, force-disables large blocks
export OPENMM_METAL_USE_LARGE_BLOCKS=1 # accepted, forces usage of large blocks
export OPENMM_METAL_USE_LARGE_BLOCKS=2 # runtime crash
unset OPENMM_METAL_USE_LARGE_BLOCKS # accepted, uses large blocks

Scaling behavior (Water Box, Amber Forcefield, PME):

s/100K/500 means "seconds per 100,000 atoms per 500 steps"

Water Box Sizes	Atoms	s/100K/500 (32-blocks)	s/100K/500 (1024-blocks)
5.0 nm	12255	4.29	4.21
10.0 nm	98880	2.02	2.02
15.0 nm	335025	1.95	1.90
20.0 nm	792450	2.03	1.92
25.0 nm	1550160	2.35	2.09
30.0 nm	2682600	2.61	2.32

Testing

Quick tests:

	Apple	AMD	Intel
Last tested version	1.1.0-dev	1.0.0	1.0.0

Test	Apple FP32	AMD FP32	Intel FP32
CMAPTorsion	✅	✅	✅
CMAPMotionRemover	✅	✅	✅
CMMotionRemover	✅	✅	✅
Checkpoints	✅	✅	✅
CompoundIntegrator	✅	✅	✅
CustomAngleForce	✅	✅	✅
CustomBondForce	✅	✅	✅
CustomCVForce	✅	✅	✅
CustomCentroidBondForce	✅	✅	✅
CustomCompoundBondForce	✅	✅	✅
CustomExternalForce	✅	✅	✅
CustomGBForce	✅	✅	✅
CustomHbondForce	✅	✅	✅
CustomTorsionForce	✅	✅	✅
DeviceQuery	✅	✅	✅
FFT	✅	✅	❌
GBSAOBCForce	✅	✅	✅
GayBerneForce	✅	✅	❌
HarmonicAngleForce	✅	✅	✅
HarmonicBondForce	✅	✅	✅
MultipleForces	✅	✅	✅
PeriodicTorsionForce	✅	✅	✅
RBTorsionForce	✅	✅	✅
RMSDForce	✅	✅	✅
Random	✅	✅	✅
Settle	✅	✅	✅
Sort	✅	✅	✅
VariableVerlet	✅	✅	✅
AmoebaExtrapolatedPolarization	✅	✅	❌
AmoebaGeneralizedKirkwoodForce	✅	✅	❌
AmoebaMultipoleForce	✅	✅	❌
AmoebaTorsionTorsionForce	✅	✅	✅
HippoNonbondedForce	✅	✅	❌
WcaDispersionForce	✅	✅	✅
DrudeForce	✅	✅	✅

Long tests:

	Apple	AMD	Intel
Last tested version	1.1.0-dev	1.0.0	1.0.0

Test	Apple FP32	AMD FP32	Intel FP32
AndersenThermostat	✅	✅	✅
CustomManyParticleForce	✅	✅	✅
CustomNonbondedForce	✅	✅	✅
DispersionPME	✅	✅	❌
Ewald	✅	✅	❌
LangevinIntegrator	✅	✅	✅
LangevinMiddleIntegrator	✅	✅	✅
LocalEnergyMinimizer	✅	❌	✅
MonteCarloFlexibleBarostat	✅	✅	✅
NonbondedForce	✅	✅	❌
VerletIntegrator	✅	✅	✅
VirtualSites	✅	✅	✅
DrudeNoseHoover	✅	✅	✅

Very long tests:

	Apple	AMD	Intel
Last tested version	1.1.0-dev	n/a	n/a

Test	Apple FP32	AMD FP32	Intel FP32
BrownianIntegrator	✅	-	-
CustomIntegrator	✅	-	-
MonteCarloAnisotropicBarostat	✅	-	-
MonteCarloBarostat	✅	-	-
NoseHooverIntegrator	✅	-	-
VariableLangevinIntegrator	✅	-	-
DrudeLangevinIntegrator	✅	-	-
DrudeSCFIntegrator	✅	-	-

Roadmap

Releases:

v1.0.0
- OpenMM 8.0.0
- Initial release, bringing proper support for Mac GPUs
v1.1.0
- OpenMM 8.1.0
- AMD and Intel go into long-term service
- Remove "HIP" platform name workaround, introduce "Metal"
- Remove double and mixed precision
v1.2.0
- OpenMM 8.2.0
- openmm/openmm#4346
- openmm/openmm#4348

License

The Metal Platform uses OpenMM API under the terms of the MIT License. A copy of this license may be found in the accompanying file MIT.txt.

The Metal Platform is based on the OpenCL Platform of OpenMM under the terms of the GNU Lesser General Public License. A copy of this license may be found in the accompanying file LGPL.txt. It in turn incorporates the terms of the GNU General Public License, which may be found in the accompanying file GPL.txt.

The Metal Platform uses VkFFT by Dmitrii Tolmachev under the terms of the MIT License. A copy of this license may be found in the accompanying file MIT-VkFFT.txt.

openmm-metal's People

Contributors

Stargazers

Watchers

Forkers

bdenhollander ssghost

openmm-metal's Issues

OpenMM Update added a new common kernel type

New class CalcATMForceKernel and ATMForce are not included in the metal kernel factory. Log file included
log.txt

AMD GPU Testing

I am starting this thread to discuss testing this plugin on Mac AMD GPUs. I only own an Apple and Intel GPU, but it would be nice to run tests on AMD before each release.

cl2Metal Error

I installed openmm-metal today on my new MacBook Pro, which has a M2 Max and Ventura 13.2.1. Unfortunately, 7 of the unit tests failed. I attached the ctest log below.

ctest --output-on-failure
Test project /Users/andrewstokely/projects/github/vantai/openmm-metal/build
Start 1: TestMetalCMAPTorsionForceSingle
1/34 Test #1: TestMetalCMAPTorsionForceSingle ................. Passed 1.00 sec
Start 2: TestMetalCMMotionRemoverSingle
2/34 Test #2: TestMetalCMMotionRemoverSingle .................. Passed 1.36 sec
Start 3: TestMetalCheckpointsSingle
3/34 Test #3: TestMetalCheckpointsSingle ......................***Failed 0.31 sec
UNSUPPORTED (log once): buildComputeProgram: cl2Metal failed
Metal internal error: [CL_DEVICE_NOT_AVAILABLE] : OpenCL Error : Error: Build Program driver returned (10014)
Metal internal error: OpenCL Warning : clBuildProgram failed: could not build program for 0x1027f00 (Apple M2 Max) (err:-2)
Metal internal error: [CL_BUILD_ERROR] : OpenCL Build Error : Compiler build log:
program_source:575:26: warning: unsupported OpenCL extension 'cl_khr_byte_addressable_store' - ignoring
#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
^
program_source:773:45: error: call to 'vloada_half3' is ambiguous
real3 largeSize = real3(vloada_half3(largeBlockIndex, largeBlockBoundingBox));
^~~~~~~~~~~~
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12475:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __constant half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12489:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __global half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12495:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __local half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12501:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __private half *p);
^
program_source:721:57: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int block1 = startBlockIndex+warpIndex; block1 < startBlockIndex+numBlocks; block1 += totalWarps) {
~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~
program_source:946:45: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:962:38: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~
program_source:991:33: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:993:35: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int j = 0; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~

exception: Error compiling kernel: program_source:575:26: warning: unsupported OpenCL extension 'cl_khr_byte_addressable_store' - ignoring
#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
^
program_source:773:45: error: call to 'vloada_half3' is ambiguous
real3 largeSize = real3(vloada_half3(largeBlockIndex, largeBlockBoundingBox));
^~~~~~~~~~~~
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12475:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __constant half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12489:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __global half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12495:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __local half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12501:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __private half *p);
^
program_source:721:57: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int block1 = startBlockIndex+warpIndex; block1 < startBlockIndex+numBlocks; block1 += totalWarps) {
~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~
program_source:946:45: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:962:38: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~
program_source:991:33: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:993:35: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int j = 0; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~

  Start  4: TestMetalCompoundIntegratorSingle

4/34 Test #4: TestMetalCompoundIntegratorSingle ............... Passed 2.05 sec
Start 5: TestMetalCustomAngleForceSingle
5/34 Test #5: TestMetalCustomAngleForceSingle ................. Passed 0.81 sec
Start 6: TestMetalCustomBondForceSingle
6/34 Test #6: TestMetalCustomBondForceSingle .................. Passed 0.46 sec
Start 7: TestMetalCustomCVForceSingle
7/34 Test #7: TestMetalCustomCVForceSingle .................... Passed 1.05 sec
Start 8: TestMetalCustomCentroidBondForceSingle
8/34 Test #8: TestMetalCustomCentroidBondForceSingle .......... Passed 0.59 sec
Start 9: TestMetalCustomCompoundBondForceSingle
9/34 Test #9: TestMetalCustomCompoundBondForceSingle .......... Passed 3.12 sec
Start 10: TestMetalCustomExternalForceSingle
10/34 Test #10: TestMetalCustomExternalForceSingle .............. Passed 0.57 sec
Start 11: TestMetalCustomGBForceSingle
11/34 Test #11: TestMetalCustomGBForceSingle ....................***Failed 0.87 sec
UNSUPPORTED (log once): buildComputeProgram: cl2Metal failed
Metal internal error: [CL_DEVICE_NOT_AVAILABLE] : OpenCL Error : Error: Build Program driver returned (10014)
Metal internal error: OpenCL Warning : clBuildProgram failed: could not build program for 0x1027f00 (Apple M2 Max) (err:-2)
Metal internal error: [CL_BUILD_ERROR] : OpenCL Build Error : Compiler build log:
program_source:574:26: warning: unsupported OpenCL extension 'cl_khr_byte_addressable_store' - ignoring
#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
^
program_source:772:45: error: call to 'vloada_half3' is ambiguous
real3 largeSize = real3(vloada_half3(largeBlockIndex, largeBlockBoundingBox));
^~~~~~~~~~~~
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12475:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __constant half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12489:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __global half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12495:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __local half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12501:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __private half *p);
^
program_source:720:57: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int block1 = startBlockIndex+warpIndex; block1 < startBlockIndex+numBlocks; block1 += totalWarps) {
~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~
program_source:945:45: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:961:38: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~
program_source:990:33: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:992:35: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int j = 0; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~

exception: Error compiling kernel: program_source:574:26: warning: unsupported OpenCL extension 'cl_khr_byte_addressable_store' - ignoring
#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
^
program_source:772:45: error: call to 'vloada_half3' is ambiguous
real3 largeSize = real3(vloada_half3(largeBlockIndex, largeBlockBoundingBox));
^~~~~~~~~~~~
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12475:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __constant half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12489:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __global half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12495:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __local half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12501:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __private half *p);
^
program_source:720:57: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int block1 = startBlockIndex+warpIndex; block1 < startBlockIndex+numBlocks; block1 += totalWarps) {
~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~
program_source:945:45: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:961:38: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~
program_source:990:33: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:992:35: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int j = 0; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~

  Start 12: TestMetalCustomHbondForceSingle

12/34 Test #12: TestMetalCustomHbondForceSingle ................. Passed 0.83 sec
Start 13: TestMetalCustomTorsionForceSingle
13/34 Test #13: TestMetalCustomTorsionForceSingle ............... Passed 0.76 sec
Start 14: TestMetalDeviceQuerySingle
14/34 Test #14: TestMetalDeviceQuerySingle ...................... Passed 0.03 sec
Start 15: TestMetalFFTSingle
15/34 Test #15: TestMetalFFTSingle .............................. Passed 1.27 sec
Start 16: TestMetalGBSAOBCForceSingle
16/34 Test #16: TestMetalGBSAOBCForceSingle .....................***Failed 0.62 sec
UNSUPPORTED (log once): buildComputeProgram: cl2Metal failed
Metal internal error: [CL_DEVICE_NOT_AVAILABLE] : OpenCL Error : Error: Build Program driver returned (10014)
Metal internal error: OpenCL Warning : clBuildProgram failed: could not build program for 0x1027f00 (Apple M2 Max) (err:-2)
Metal internal error: [CL_BUILD_ERROR] : OpenCL Build Error : Compiler build log:
program_source:574:26: warning: unsupported OpenCL extension 'cl_khr_byte_addressable_store' - ignoring
#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
^
program_source:772:45: error: call to 'vloada_half3' is ambiguous
real3 largeSize = real3(vloada_half3(largeBlockIndex, largeBlockBoundingBox));
^~~~~~~~~~~~
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12475:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __constant half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12489:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __global half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12495:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __local half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12501:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __private half *p);
^
program_source:720:57: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int block1 = startBlockIndex+warpIndex; block1 < startBlockIndex+numBlocks; block1 += totalWarps) {
~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~
program_source:945:45: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:961:38: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~
program_source:990:33: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:992:35: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int j = 0; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~

  Start 17: TestMetalGayBerneForceSingle

17/34 Test #17: TestMetalGayBerneForceSingle .................... Passed 1.48 sec
Start 18: TestMetalHarmonicAngleForceSingle
18/34 Test #18: TestMetalHarmonicAngleForceSingle ............... Passed 0.40 sec
Start 19: TestMetalHarmonicBondForceSingle
19/34 Test #19: TestMetalHarmonicBondForceSingle ................ Passed 0.46 sec
Start 20: TestMetalMultipleForcesSingle
20/34 Test #20: TestMetalMultipleForcesSingle ................... Passed 0.29 sec
Start 21: TestMetalPeriodicTorsionForceSingle
21/34 Test #21: TestMetalPeriodicTorsionForceSingle ............. Passed 0.55 sec
Start 22: TestMetalRBTorsionForceSingle
22/34 Test #22: TestMetalRBTorsionForceSingle ................... Passed 0.72 sec
Start 23: TestMetalRMSDForceSingle
23/34 Test #23: TestMetalRMSDForceSingle ........................ Passed 0.38 sec
Start 24: TestMetalRandomSingle
24/34 Test #24: TestMetalRandomSingle ........................... Passed 0.49 sec
Start 25: TestMetalSettleSingle
25/34 Test #25: TestMetalSettleSingle ........................... Passed 0.86 sec
Start 26: TestMetalSortSingle
26/34 Test #26: TestMetalSortSingle ............................. Passed 0.38 sec
Start 27: TestMetalVariableVerletIntegratorSingle
27/34 Test #27: TestMetalVariableVerletIntegratorSingle .........***Failed 4.38 sec
UNSUPPORTED (log once): buildComputeProgram: cl2Metal failed
Metal internal error: [CL_DEVICE_NOT_AVAILABLE] : OpenCL Error : Error: Build Program driver returned (10014)
Metal internal error: OpenCL Warning : clBuildProgram failed: could not build program for 0x1027f00 (Apple M2 Max) (err:-2)
Metal internal error: [CL_BUILD_ERROR] : OpenCL Build Error : Compiler build log:
program_source:575:26: warning: unsupported OpenCL extension 'cl_khr_byte_addressable_store' - ignoring
#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
^
program_source:773:45: error: call to 'vloada_half3' is ambiguous
real3 largeSize = real3(vloada_half3(largeBlockIndex, largeBlockBoundingBox));
^~~~~~~~~~~~
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12475:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __constant half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12489:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __global half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12495:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __local half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12501:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __private half *p);
^
program_source:721:57: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int block1 = startBlockIndex+warpIndex; block1 < startBlockIndex+numBlocks; block1 += totalWarps) {
~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~
program_source:946:45: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:962:38: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~
program_source:991:33: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:993:35: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int j = 0; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~

  Start 28: TestMetalAmoebaExtrapolatedPolarizationSingle

28/34 Test #28: TestMetalAmoebaExtrapolatedPolarizationSingle ...***Failed 1.57 sec
UNSUPPORTED (log once): buildComputeProgram: cl2Metal failed
Metal internal error: [CL_DEVICE_NOT_AVAILABLE] : OpenCL Error : Error: Build Program driver returned (10014)
Metal internal error: OpenCL Warning : clBuildProgram failed: could not build program for 0x1027f00 (Apple M2 Max) (err:-2)
Metal internal error: [CL_BUILD_ERROR] : OpenCL Build Error : Compiler build log:
program_source:592:26: warning: unsupported OpenCL extension 'cl_khr_byte_addressable_store' - ignoring
#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
^
program_source:790:45: error: call to 'vloada_half3' is ambiguous
real3 largeSize = real3(vloada_half3(largeBlockIndex, largeBlockBoundingBox));
^~~~~~~~~~~~
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12475:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __constant half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12489:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __global half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12495:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __local half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12501:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __private half *p);
^
program_source:738:57: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int block1 = startBlockIndex+warpIndex; block1 < startBlockIndex+numBlocks; block1 += totalWarps) {
~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~
program_source:963:45: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:979:38: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~
program_source:1008:33: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:1010:35: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int j = 0; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~

exception: Error compiling kernel: program_source:592:26: warning: unsupported OpenCL extension 'cl_khr_byte_addressable_store' - ignoring
#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
^
program_source:790:45: error: call to 'vloada_half3' is ambiguous
real3 largeSize = real3(vloada_half3(largeBlockIndex, largeBlockBoundingBox));
^~~~~~~~~~~~
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12475:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __constant half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12489:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __global half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12495:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __local half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12501:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __private half *p);
^
program_source:738:57: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int block1 = startBlockIndex+warpIndex; block1 < startBlockIndex+numBlocks; block1 += totalWarps) {
~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~
program_source:963:45: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:979:38: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~
program_source:1008:33: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:1010:35: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int j = 0; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~

FAIL - ERROR. Test failed.

  Start 29: TestMetalAmoebaGeneralizedKirkwoodForceSingle

29/34 Test #29: TestMetalAmoebaGeneralizedKirkwoodForceSingle ... Passed 7.81 sec
Start 30: TestMetalAmoebaMultipoleForceSingle
30/34 Test #30: TestMetalAmoebaMultipoleForceSingle .............***Failed 1.95 sec
UNSUPPORTED (log once): buildComputeProgram: cl2Metal failed
Metal internal error: [CL_DEVICE_NOT_AVAILABLE] : OpenCL Error : Error: Build Program driver returned (10014)
Metal internal error: OpenCL Warning : clBuildProgram failed: could not build program for 0x1027f00 (Apple M2 Max) (err:-2)
Metal internal error: [CL_BUILD_ERROR] : OpenCL Build Error : Compiler build log:
program_source:575:26: warning: unsupported OpenCL extension 'cl_khr_byte_addressable_store' - ignoring
#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
^
program_source:773:45: error: call to 'vloada_half3' is ambiguous
real3 largeSize = real3(vloada_half3(largeBlockIndex, largeBlockBoundingBox));
^~~~~~~~~~~~
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12475:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __constant half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12489:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __global half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12495:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __local half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12501:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __private half *p);
^
program_source:721:57: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int block1 = startBlockIndex+warpIndex; block1 < startBlockIndex+numBlocks; block1 += totalWarps) {
~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~
program_source:946:45: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:962:38: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~
program_source:991:33: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:993:35: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int j = 0; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~

FAIL - ERROR. Test failed.

  Start 31: TestMetalAmoebaTorsionTorsionForceSingle

31/34 Test #31: TestMetalAmoebaTorsionTorsionForceSingle ........ Passed 0.45 sec
Start 32: TestMetalHippoNonbondedForceSingle
32/34 Test #32: TestMetalHippoNonbondedForceSingle ..............***Failed 2.66 sec
UNSUPPORTED (log once): buildComputeProgram: cl2Metal failed
Metal internal error: [CL_DEVICE_NOT_AVAILABLE] : OpenCL Error : Error: Build Program driver returned (10014)
Metal internal error: OpenCL Warning : clBuildProgram failed: could not build program for 0x1027f00 (Apple M2 Max) (err:-2)
Metal internal error: [CL_BUILD_ERROR] : OpenCL Build Error : Compiler build log:
program_source:575:26: warning: unsupported OpenCL extension 'cl_khr_byte_addressable_store' - ignoring
#pragma OPENCL EXTENSION cl_khr_byte_addressable_store : enable
^
program_source:773:45: error: call to 'vloada_half3' is ambiguous
real3 largeSize = real3(vloada_half3(largeBlockIndex, largeBlockBoundingBox));
^~~~~~~~~~~~
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12475:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __constant half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12489:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __global half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12495:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __local half *p);
^
/System/Library/PrivateFrameworks/GPUCompiler.framework/Versions/31001/Libraries/lib/clang/31001.669/include/opencl-c.h:12501:15: note: candidate function
float3 __ovld vloada_half3(size_t offset, const __private half *p);
^
program_source:721:57: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int block1 = startBlockIndex+warpIndex; block1 < startBlockIndex+numBlocks; block1 += totalWarps) {
~~~~~~ ^ ~~~~~~~~~~~~~~~~~~~~~~~~~
program_source:946:45: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:962:38: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~
program_source:991:33: warning: comparison of integers of different signs: 'const int' and 'unsigned int'
if (indexInWarp < tilesToStore)
~~~~~~~~~~~ ^ ~~~~~~~~~~~~
program_source:993:35: warning: comparison of integers of different signs: 'int' and 'unsigned int'
for (int j = 0; j < tilesToStore; j++)
~ ^ ~~~~~~~~~~~~

  Start 33: TestMetalWcaDispersionForceSingle

33/34 Test #33: TestMetalWcaDispersionForceSingle ............... Passed 0.32 sec
Start 34: TestMetalDrudeForceSingle
34/34 Test #34: TestMetalDrudeForceSingle ....................... Passed 0.84 sec

79% tests passed, 7 tests failed out of 34

Total Test time (real) = 41.69 sec

The following tests FAILED:
3 - TestMetalCheckpointsSingle (Failed)
11 - TestMetalCustomGBForceSingle (Failed)
16 - TestMetalGBSAOBCForceSingle (Failed)
27 - TestMetalVariableVerletIntegratorSingle (Failed)
28 - TestMetalAmoebaExtrapolatedPolarizationSingle (Failed)
30 - TestMetalAmoebaMultipoleForceSingle (Failed)
32 - TestMetalHippoNonbondedForceSingle (Failed)
Errors while running CTest

Recommend Projects

React

A declarative, efficient, and flexible JavaScript library for building user interfaces.
Vue.js

🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
Typescript

TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
TensorFlow

An Open Source Machine Learning Framework for Everyone
Django

The Web framework for perfectionists with deadlines.
Laravel

A PHP framework for web artisans
D3

Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

javascript

JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
web

Some thing interesting about web. New door for the world.
server

A server is a program made to process requests and deliver data to clients.
Machine learning

Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Visualization

Some thing interesting about visualization, use data art
Game

Some thing interesting about game, make everyone happy.

Recommend Org

Facebook

We are working to build community through open source technology. NB: members must have two-factor auth.
Microsoft

Open source projects and samples from Microsoft.
Google

Google ❤️ Open Source for everyone.
Alibaba

Alibaba Open Source for everyone
D3

Data-Driven Documents codes.
Tencent

China tencent open source team.