Resolve the segmentation fault occurring in the pw float implementation #6130

A-006 · 2025-04-08T09:31:02Z

Linked Issue

Fix #6111

Unit Tests and/or Case Tests for My Changes

Added unit tests for the pw_basis float type.
Included integration tests for the pw_basis float type.

What's Changed?

Fixed a bug in the runtime handling of the pw_basis float type.
Enabled GPU support for pw_basis without requiring the compilation flag ENABLE_FFTW_FLOAT.
ATTENTION!!! In order to reduce the test time of the GPU/CPU,i just add the PW_CG PW_CG_GPU float type to test the float version FFT can be normally used.

source/module_base/test/math_chebyshev_test.cpp

tests/integrate/102_PW_BPCG_GPU_float/README

tests/integrate/102_PW_CG_float/README

source/module_esolver/esolver_fp.cpp

source/module_hamilt_pw/hamilt_pwdft/test/CMakeLists.txt

tests/integrate/102_PW_DS_davsubspace_float/README

tests/integrate/102_PW_DA_davidson_float/threshold

tests/integrate/102_PW_DA_davidson_float/README

tests/integrate/102_PW_DA_davidson_GPU_float/README

tests/integrate/102_PW_CG_float/threshold

kogareru1z · 2025-04-23T08:51:35Z

Could you please fix the single-precision calculation issue for the LCAO basis? When I tested with ABACUS 3.9.0.3, I encountered the same problem as with PW, but the error does not occur in version 3.10.

A-006 · 2025-04-23T14:19:26Z

Could you please fix the single-precision calculation issue for the LCAO basis? When I tested with ABACUS 3.9.0.3, I encountered the same problem as with PW, but the error does not occur in version 3.10.

Alright, I'll try to address this issue later. However, I recall that the LCAO basis does not support a single version; I will verify this information

A-006 · 2025-04-24T07:54:11Z

Could you please fix the single-precision calculation issue for the LCAO basis? When I tested with ABACUS 3.9.0.3, I encountered the same problem as with PW, but the error does not occur in version 3.10.

Currently, the LCAO does not support a single version. Could you please inform me whether you intend to use the GPU LCAO single? Actually, this feature will be implemented by my partner. After the pull request (PR) is merged, you will be able to set the GPU LCAO to single; however, in reality, it will run as GPU LCAO double.

kogareru1z · 2025-04-24T08:07:17Z

Could you please fix the single-precision calculation issue for the LCAO basis? When I tested with ABACUS 3.9.0.3, I encountered the same problem as with PW, but the error does not occur in version 3.10.

Currently, the LCAO does not support a single version. Could you please inform me whether you intend to use the GPU LCAO single? Actually, this feature will be implemented by my partner. After the pull request (PR) is merged, you will be able to set the GPU LCAO to single; however, in reality, it will run as GPU LCAO double.

I tested version 3.10 in single precision, but it still runs in double precision. If that’s the case, I don’t need this feature right now. Thank you.

source/module_base/test/math_chebyshev_test.cpp

source/module_basis/module_pw/module_fft/fft_cpu.cpp

source/module_esolver/esolver_fp.cpp

tests/integrate/102_PW_BPCG_GPU_float/README

mohanchen · 2025-04-29T11:27:24Z

tests/integrate/102_PW_BPCG_float/README

tests/integrate/102_PW_CG_GPU_float/README

mohanchen · 2025-04-29T11:27:51Z

tests/integrate/102_PW_CG_float/README

The README has already been resolved.

mohanchen · 2025-05-10T09:01:49Z

source/module_basis/module_pw/pw_gatherscatter.h

@@ -98,7 +98,7 @@ void PW_Basis::gatherp_scatters(std::complex<T>* in, std::complex<T>* out) const
 template <typename T>
 void PW_Basis::gathers_scatterp(std::complex<T>* in, std::complex<T>* out) const
 {
-    //ModuleBase::timer::tick(this->classname, "gathers_scatterp");
+    ModuleBase::timer::tick(this->classname, "gathers_scatterp");


will it cost a lot of counting numbers?

The time ticker has been deleted.

mohanchen · 2025-05-10T09:02:36Z

source/module_hamilt_pw/hamilt_pwdft/test/CMakeLists.txt

@@ -2,6 +2,7 @@ remove_definitions(-D__DEEPKS)
 remove_definitions(-D__CUDA)
 remove_definitions(-D__ROCM)
 remove_definitions(-D__EXX)
+remove_definitions(-DUSE_PAW)


we will delete PAW

The command has been deleted.

mohanchen · 2025-05-10T09:03:57Z

tests/integrate/102_PW_CG/README

don't write README in this way, need to discuss

The README has been rewritten.

mohanchen · 2025-05-10T09:04:08Z

source/module_lr/esolver_lrtd_lcao.cpp

@@ -257,7 +257,11 @@ LR::ESolver_LR<T, TR>::ESolver_LR(ModuleESolver::ESolver_KS_LCAO<T, TR>&& ks_sol
    this->gint_->reset_DMRGint(1);

    // move pw basis
-    delete this->pw_rho;    // newed in ESolver_FP::ESolver_FP
+    if (this->pw_rho_flag)


need to discuss

mohanchen · 2025-05-10T09:04:28Z

source/module_hamilt_pw/hamilt_pwdft/test/CMakeLists.txt

@@ -26,4 +27,31 @@ AddTest(
 	TARGET radial_proj_test
 	LIBS parameter  base device ${math_libs}
 	SOURCES radial_proj_test.cpp ../radial_proj.cpp
+)
+
+AddTest(


what's the aim of this test?

The error with the float type originated from the structure_factor function, which did not have a test for float inputs. We have now added the test to prevent this issue from occurring again.

mohanchen · 2025-05-10T09:05:24Z

source/module_esolver/esolver_fp.cpp

+        delete this->pw_rho;
+        this->pw_rho_flag = false;
+    }
+    if ( PARAM.globalv.double_grid)


extra blank found

The blank has been deleted.

A-006 and others added 13 commits April 8, 2025 16:44

add unit test

7b855e6

add intergrate test

4c779e8

fix process

f9e7710

modify jd

a8d72af

update bug

5faf27e

set fftw float

4773008

add the float BPCG

503bf58

add float test

4b5df98

fix compile bug

16cb172

fix error

f5c1fc1

fix the compile test

677e5d6

Merge branch 'develop' into fft_float2

98decc4

Merge branch 'develop' into fft_float2

f632326

mohanchen reviewed Apr 14, 2025

View reviewed changes

A-006 added 8 commits April 18, 2025 14:19

add

300713c

remove the test file

1dbacf8

change the file

f565945

revert bug

e1601ee

set the float type

f6fd16d

Merge branch 'develop' into fft_float2

bed7852

reset the FFT_MEASURE

80344ac

update unittest

c60bf81

mohanchen reviewed Apr 21, 2025

View reviewed changes

mohanchen added the Refactor Refactor ABACUS codes label Apr 21, 2025

A-006 and others added 5 commits April 22, 2025 17:00

change readme

ed18346

update threashold

1f66367

Merge branch 'develop' into fft_float2

4c63669

use the test file

7553e06

fix unresonable comments

385b010

A-006 and others added 5 commits April 27, 2025 21:38

update eslover before all runners

2e13c7f

Merge branch 'develop' into fft_float2

2bf18b9

fix compile bug

a224da7

fix bug

59b73f5

Merge branch 'develop' into fft_float2

f750e10

mohanchen reviewed Apr 29, 2025

View reviewed changes

A-006 added 6 commits May 6, 2025 21:47

update README

d193075

change chebyshev MPI part

a9b53a1

Merge branch 'develop' into fft_float2

5c156e5

add new test

d5084f6

delete old test

aa443f1

remove old tests

2d2a550

mohanchen reviewed May 10, 2025

View reviewed changes

A-006 and others added 6 commits May 13, 2025 17:16

add change

b1f144e

Merge branch 'develop' into fft_float2

c60d13f

update tick

df3c712

add back marco

ca1b0d9

update change

5ca14cd

Merge branch 'develop' into fft_float2

be074ab

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Resolve the segmentation fault occurring in the pw float implementation #6130

Resolve the segmentation fault occurring in the pw float implementation #6130

A-006 commented Apr 8, 2025 •

edited

Loading

kogareru1z commented Apr 23, 2025

A-006 commented Apr 23, 2025 •

edited

Loading

A-006 commented Apr 24, 2025

kogareru1z commented Apr 24, 2025

mohanchen Apr 29, 2025

mohanchen Apr 29, 2025

A-006 May 6, 2025

mohanchen May 10, 2025

A-006 May 13, 2025

mohanchen May 10, 2025

A-006 May 13, 2025

mohanchen May 10, 2025

A-006 May 13, 2025

mohanchen May 10, 2025

mohanchen May 10, 2025

A-006 May 13, 2025

mohanchen May 10, 2025

A-006 May 13, 2025

Resolve the segmentation fault occurring in the pw float implementation #6130

Are you sure you want to change the base?

Resolve the segmentation fault occurring in the pw float implementation #6130

Conversation

A-006 commented Apr 8, 2025 • edited Loading

Linked Issue

Unit Tests and/or Case Tests for My Changes

What's Changed?

kogareru1z commented Apr 23, 2025

A-006 commented Apr 23, 2025 • edited Loading

A-006 commented Apr 24, 2025

kogareru1z commented Apr 24, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

A-006 commented Apr 8, 2025 •

edited

Loading

A-006 commented Apr 23, 2025 •

edited

Loading