Skip to content
New issue

Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.

#

By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? # to your account

Jump to bottom

Simd for complex numbers (r,r,r, … i,i,i….) #8

Closed
roiser opened this issue Aug 12, 2020 · 1 comment
Closed

Simd for complex numbers (r,r,r, … i,i,i….) #8

roiser opened this issue Aug 12, 2020 · 1 comment
Assignees
@valassi
Labels
idea Possible new development (may need further discussion)

Comments

@roiser
Copy link
Member

roiser commented Aug 12, 2020

Try to rearrange the real and imaginary parts of the complex numbers in SIMD format, check performance
@roiser roiser added the idea Possible new development (may need further discussion) label Aug 12, 2020
@valassi valassi self-assigned this Dec 9, 2020
@valassi
Copy link
Member

valassi commented Dec 9, 2020 •
edited
Loading

This is addressed in the context of issue #71 and PR #72.

The RRRRIIII memory layout is implemented in

madgraph4gpu/epoch1/cuda/ee_mumu/src/mgOnGpuVectors.h

Line 52 in 5e90bc8

fptype_v m_real, m_imag; // RRRRIIII

Closing.

@valassi valassi closed this as completed Dec 9, 2020
valassi added a commit to valassi/madgraph4gpu that referenced this issue Apr 23, 2021
@valassi
[klas2ep12] Ep1 temporarely bypass protections to allow clang vector …
69529f3 
…builds.

The build fails on clang10 at compilation time

clang++: /build/gcc/build/contrib/clang-10.0.0/src/clang/10.0.0/tools/clang/lib/CodeGen/CGExpr.cpp:596: clang::CodeGen::RValue clang::CodeGen::CodeGenFunction::EmitReferenceBindingToExpr(const clang::Expr*): Assertion `LV.isSimple()' failed.
Stack dump:
0.      Program arguments: /cvmfs/sft.cern.ch/lcg/releases/clang/10.0.0-62e61/x86_64-centos7/bin/clang++ -O3 -std=c++17 -I. -I../../src -I../../../../../tools -DUSE_NVTX -Wall -Wshadow -Wextra -fopenmp -ffast-math -march=skylake-avx512 -mprefer-vector-width=256 -I/usr/local/cuda-11.0/include/ -c CPPProcess.cc -o CPPProcess.o
1.      <eof> parser at end of file
2.      Per-file LLVM IR generation
3.      ../../src/mgOnGpuVectors.h:59:16: Generating code for declaration 'mgOnGpu::cxtype_v::operator[]'
 #0 0x0000000001af5f9a llvm::sys::PrintStackTrace(llvm::raw_ostream&) (/cvmfs/sft.cern.ch/lcg/releases/clang/10.0.0-62e61/x86_64-centos7/bin/clang+++0x1af5f9a)
 #1 0x0000000001af3d54 llvm::sys::RunSignalHandlers() (/cvmfs/sft.cern.ch/lcg/releases/clang/10.0.0-62e61/x86_64-centos7/bin/clang+++0x1af3d54)
 #2 0x0000000001af3fa9 llvm::sys::CleanupOnSignal(unsigned long) (/cvmfs/sft.cern.ch/lcg/releases/clang/10.0.0-62e61/x86_64-centos7/bin/clang+++0x1af3fa9)
 madgraph5#3 0x0000000001a6ed08 CrashRecoverySignalHandler(int) (/cvmfs/sft.cern.ch/lcg/releases/clang/10.0.0-62e61/x86_64-centos7/bin/clang+++0x1a6ed08)
 madgraph5#4 0x00007fd31c178630 __restore_rt (/lib64/libpthread.so.0+0xf630)
 madgraph5#5 0x00007fd31ac8c3d7 raise (/lib64/libc.so.6+0x363d7)
 madgraph5#6 0x00007fd31ac8dac8 abort (/lib64/libc.so.6+0x37ac8)
 madgraph5#7 0x00007fd31ac851a6 __assert_fail_base (/lib64/libc.so.6+0x2f1a6)
 madgraph5#8 0x00007fd31ac85252 (/lib64/libc.so.6+0x2f252)
 madgraph5#9 0x000000000203a042 clang::CodeGen::CodeGenFunction::EmitReferenceBindingToExpr(clang::Expr const*) (/cvmfs/sft.cern.ch/lcg/releases/clang/10.0.0-62e61/x86_64-centos7/bin/clang+++0x203a042)
valassi added a commit to valassi/madgraph4gpu that referenced this issue Feb 23, 2022
@valassi
[shared] add debug printout - flagAbnormalMEs actually works! but #na…
b2a54e8 
…ns is different for fcheck

> ./fcheck.exe  2048 64 10
 GPUBLOCKS=          2048
 GPUTHREADS=           64
 NITERATIONS=          10
WARNING! Instantiate host Bridge (nevt=131072)
INFO: The application is built for skylake-avx512 (AVX512VL) and the host supports it
WARNING! Instantiate host Sampler (nevt=131072)
Iteration #1
Iteration #2
Iteration madgraph5#3
Iteration madgraph5#4
Iteration madgraph5#5
Iteration madgraph5#6
Iteration madgraph5#7
Iteration madgraph5#8
Iteration madgraph5#9
WARNING! flagging abnormal ME for ievt=111162
Iteration madgraph5#10
 Average Matrix Element:   1.3716954486179133E-002
 Abnormal MEs:           1

> ./check.exe -p  2048 64 10 | grep FLOAT
FP precision                = FLOAT (NaN/abnormal=2, zero=0)

I imagine that this is because momenta in Fortran get translated from float to double and back to float, while in c++ they stay in float?
valassi added a commit to valassi/madgraph4gpu that referenced this issue Jun 14, 2022
@valassi
[fvsc] in ggttmad, add old fashioned debug printouts
5a36385 
./cmadevent_cudacpp < /tmp/avalassi/input_ggtt_cpp | grep DEBUG_ | sort | uniq -c
  16416  DEBUG_SMATRIX1 #1
 262656  DEBUG_SMATRIX1 #1a
      1  DEBUG_SMATRIX1 #2
  16416  DEBUG_SMATRIX1 madgraph5#4
     25  DEBUG_SMATRIX1 #4a
      1  DEBUG_SMATRIX1 #4b
  16416  DEBUG_SMATRIX1 madgraph5#7
  16416  DEBUG_SMATRIX1 madgraph5#8
valassi pushed a commit to valassi/madgraph4gpu that referenced this issue Jan 25, 2024
@Jooorgen
Merge pull request madgraph5#8 from Jooorgen/gpu_abstraction
7dbb79b 
Merge latest changes in gpu_abstraction branch to master branch with even newer CI tests
valassi added a commit to valassi/madgraph4gpu that referenced this issue May 17, 2024
@valassi
[jtmk2] rerun 30 tmad tests on itscrd90 - one new FPE issue madgraph5…
7f1ee3b 
…#845 in log_gqttq_mad_f_inl0_hrd0.txt, the rest as expected

STARTED  AT Thu May 16 01:24:16 AM CEST 2024
(SM tests)
ENDED(1) AT Thu May 16 05:58:45 AM CEST 2024 [Status=0]
(BSM tests)
ENDED(1) AT Thu May 16 06:07:42 AM CEST 2024 [Status=0]

24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_eemumu_mad/log_eemumu_mad_d_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_eemumu_mad/log_eemumu_mad_f_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_eemumu_mad/log_eemumu_mad_m_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_ggttggg_mad/log_ggttggg_mad_d_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_ggttggg_mad/log_ggttggg_mad_f_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_ggttggg_mad/log_ggttggg_mad_m_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_ggttgg_mad/log_ggttgg_mad_d_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_ggttgg_mad/log_ggttgg_mad_f_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_ggttgg_mad/log_ggttgg_mad_m_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_ggttg_mad/log_ggttg_mad_d_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_ggttg_mad/log_ggttg_mad_f_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_ggttg_mad/log_ggttg_mad_m_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_ggtt_mad/log_ggtt_mad_d_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_ggtt_mad/log_ggtt_mad_f_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_ggtt_mad/log_ggtt_mad_m_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_gqttq_mad/log_gqttq_mad_d_inl0_hrd0.txt
18 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_gqttq_mad/log_gqttq_mad_f_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_gqttq_mad/log_gqttq_mad_m_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_heftggbb_mad/log_heftggbb_mad_d_inl0_hrd0.txt
1 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_heftggbb_mad/log_heftggbb_mad_f_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_heftggbb_mad/log_heftggbb_mad_m_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_smeftggtttt_mad/log_smeftggtttt_mad_d_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_smeftggtttt_mad/log_smeftggtttt_mad_f_inl0_hrd0.txt
24 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_smeftggtttt_mad/log_smeftggtttt_mad_m_inl0_hrd0.txt
0 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_susyggt1t1_mad/log_susyggt1t1_mad_d_inl0_hrd0.txt
0 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_susyggt1t1_mad/log_susyggt1t1_mad_f_inl0_hrd0.txt
0 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_susyggt1t1_mad/log_susyggt1t1_mad_m_inl0_hrd0.txt
0 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_susyggtt_mad/log_susyggtt_mad_d_inl0_hrd0.txt
0 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_susyggtt_mad/log_susyggtt_mad_f_inl0_hrd0.txt
0 /data/avalassi/GPU2023/madgraph4gpuX/epochX/cudacpp/tmad/logs_susyggtt_mad/log_susyggtt_mad_m_inl0_hrd0.txt

The new issue madgraph5#845 is the following
+Program received signal SIGFPE: Floating-point exception - erroneous arithmetic operation.
+
+Backtrace for this error:
+#0  0x7f2a1a623860 in ???
+#1  0x7f2a1a622a05 in ???
+#2  0x7f2a1a254def in ???
+madgraph5#3  0x7f2a1ae20acc in ???
+madgraph5#4  0x7f2a1acc4575 in ???
+madgraph5#5  0x7f2a1ae1d4c9 in ???
+madgraph5#6  0x7f2a1ae2570d in ???
+madgraph5#7  0x7f2a1ae2afa1 in ???
+madgraph5#8  0x43008b in ???
+madgraph5#9  0x431c10 in ???
+madgraph5#10  0x432d47 in ???
+madgraph5#11  0x433b1e in ???
+madgraph5#12  0x44a921 in ???
+madgraph5#13  0x42ebbf in ???
+madgraph5#14  0x40371e in ???
+madgraph5#15  0x7f2a1a23feaf in ???
+madgraph5#16  0x7f2a1a23ff5f in ???
+madgraph5#17  0x403844 in ???
+madgraph5#18  0xffffffffffffffff in ???
+./madX.sh: line 379: 3004240 Floating point exception(core dumped) $timecmd $cmd < ${tmpin} > ${tmp}
+ERROR! ' ./build.512z_f_inl0_hrd0/madevent_cpp < /tmp/avalassi/input_gqttq_x10_cudacpp > /tmp/avalassi/output_gqttq_x10_cudacpp' failed
# for free to join this conversation on GitHub. Already have an account? # to comment
Assignees

@valassi valassi

Labels
idea Possible new development (may need further discussion)
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@roiser @valassi