Script 'mail_helper' called by obssrc
Hello community,
here is the log from the commit of package scs for openSUSE:Factory checked in
at 2021-04-12 17:10:44
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Comparing /work/SRC/openSUSE:Factory/scs (Old)
and /work/SRC/openSUSE:Factory/.scs.new.2401 (New)
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Package is "scs"
Mon Apr 12 17:10:44 2021 rev:2 rq:884647 version:2.1.3
Changes:
--------
--- /work/SRC/openSUSE:Factory/scs/scs.changes 2020-09-25 16:32:27.587939662
+0200
+++ /work/SRC/openSUSE:Factory/.scs.new.2401/scs.changes 2021-04-12
17:10:54.482601073 +0200
@@ -1,0 +2,7 @@
+Mon Apr 12 12:12:03 UTC 2021 - andy great <andythe_gr...@pm.me>
+
+- Update to version 2.1.3.
+ * Improved GPU indirect solver.
+- Stop using %exclude to remove static lib.
+
+-------------------------------------------------------------------
Old:
----
scs-2.1.2.tar.gz
New:
----
scs-2.1.3.tar.gz
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Other differences:
------------------
++++++ scs.spec ++++++
--- /var/tmp/diff_new_pack.HL2UdR/_old 2021-04-12 17:10:54.966601608 +0200
+++ /var/tmp/diff_new_pack.HL2UdR/_new 2021-04-12 17:10:54.970601612 +0200
@@ -1,7 +1,7 @@
#
# spec file for package scs
#
-# Copyright (c) 2020 SUSE LLC
+# Copyright (c) 2021 SUSE LLC
#
# All modifications and additions to the file contributed by third parties
# remain the property of their copyright owners, unless otherwise agreed
@@ -17,7 +17,7 @@
Name: scs
-Version: 2.1.2
+Version: 2.1.3
Release: 0
Summary: Numerical package for solving large-scale convex cone problems
License: MIT
@@ -46,12 +46,12 @@
%install
%make_install INSTALL_LIB_DIR=%{buildroot}%{_libdir}
INSTALL_INC_DIR=%{buildroot}%{_includedir}/%{name}
+rm -r %{buildroot}%{_libdir}/*.a
%files devel
%doc README.md
%license LICENSE.txt
%{_includedir}/%{name}
%{_libdir}/*.so
-%exclude %{_libdir}/*.a
%changelog
++++++ scs-2.1.2.tar.gz -> scs-2.1.3.tar.gz ++++++
diff -urN '--exclude=CVS' '--exclude=.cvsignore' '--exclude=.svn'
'--exclude=.svnignore' old/scs-2.1.2/.bumpversion.cfg
new/scs-2.1.3/.bumpversion.cfg
--- old/scs-2.1.2/.bumpversion.cfg 2020-06-05 11:04:23.000000000 +0200
+++ new/scs-2.1.3/.bumpversion.cfg 2021-04-12 10:43:08.000000000 +0200
@@ -1,5 +1,5 @@
[bumpversion]
-current_version = 2.1.2
+current_version = 2.1.3
[bumpversion:file:README.md]
diff -urN '--exclude=CVS' '--exclude=.cvsignore' '--exclude=.svn'
'--exclude=.svnignore' old/scs-2.1.2/README.md new/scs-2.1.3/README.md
--- old/scs-2.1.2/README.md 2020-06-05 11:04:23.000000000 +0200
+++ new/scs-2.1.3/README.md 2021-04-12 10:43:08.000000000 +0200
@@ -2,7 +2,7 @@
====
[](https://travis-ci.org/cvxgrp/scs)
-[](https://ci.appveyor.com/project/bodono/scs/branch/master)
+<!--[](https://ci.appveyor.com/project/bodono/scs/branch/master)-->
SCS (`splitting conic solver`) is a numerical optimization package for solving
large-scale convex cone problems, based on our paper [Conic Optimization via
@@ -21,7 +21,7 @@
[Convex.jl](https://github.com/JuliaOpt/Convex.jl), and
[Yalmip](https://github.com/johanlofberg/YALMIP).
-The current version is `2.1.2`. If you wish to cite SCS, please use the
+The current version is `2.1.3`. If you wish to cite SCS, please use the
following:
```
@article{ocpb:16,
@@ -37,7 +37,7 @@
}
@misc{scs,
author = {B. O'Donoghue and E. Chu and N. Parikh and S. Boyd},
- title = {{SCS}: Splitting Conic Solver, version 2.1.2},
+ title = {{SCS}: Splitting Conic Solver, version 2.1.3},
howpublished = {\url{https://github.com/cvxgrp/scs}},
month = nov,
year = 2019
@@ -137,7 +137,7 @@
The direct solver uses external numerical linear algebra packages:
* [QDLDL](https://github.com/oxfordcontrol/qdldl)
-* [AMD](http://www.cise.ufl.edu/research/sparse/).
+* [AMD](https://github.com/DrTimothyAldenDavis/SuiteSparse).
### Using SCS in C
Typing `make` at the command line will compile the code and create SCS
libraries
@@ -153,7 +153,7 @@
`libscsindir.ext` (where `.ext` extension is platform dependent) in the same
folder. It will also produce two demo binaries in the `out` folder named
`demo_socp_direct`, and `demo_socp_indirect`. If you have a GPU and have CUDA
-installed, you can also execture `make gpu` to compile SCS to run on the GPU
+installed, you can also execute `make gpu` to compile SCS to run on the GPU
which will create additional libraries and demo binaries in the `out` folder
corresponding to the gpu version. Note that the GPU version requires 32 bit
ints, which can be enforced by compiling with `DLONG=0`.
diff -urN '--exclude=CVS' '--exclude=.cvsignore' '--exclude=.svn'
'--exclude=.svnignore' old/scs-2.1.2/include/glbopts.h
new/scs-2.1.3/include/glbopts.h
--- old/scs-2.1.2/include/glbopts.h 2020-06-05 11:04:23.000000000 +0200
+++ new/scs-2.1.3/include/glbopts.h 2021-04-12 10:43:08.000000000 +0200
@@ -13,7 +13,7 @@
/* SCS VERSION NUMBER ---------------------------------------------- */
#define SCS_VERSION \
- ("2.1.2") /* string literals automatically null-terminated */
+ ("2.1.3") /* string literals automatically null-terminated */
/* SCS returns one of the following integers: */
#define SCS_INFEASIBLE_INACCURATE (-7)
diff -urN '--exclude=CVS' '--exclude=.cvsignore' '--exclude=.svn'
'--exclude=.svnignore' old/scs-2.1.2/linsys/gpu/gpu.c
new/scs-2.1.3/linsys/gpu/gpu.c
--- old/scs-2.1.2/linsys/gpu/gpu.c 2020-06-05 11:04:23.000000000 +0200
+++ new/scs-2.1.3/linsys/gpu/gpu.c 2021-04-12 10:43:08.000000000 +0200
@@ -1,33 +1,72 @@
#include "gpu.h"
-void SCS(_accum_by_atrans_gpu)(const ScsGpuMatrix *Ag, const scs_float *x,
- scs_float *y, cusparseHandle_t cusparse_handle)
{
+void SCS(_accum_by_atrans_gpu)(const ScsGpuMatrix *Ag, const
cusparseDnVecDescr_t x,
+ cusparseDnVecDescr_t y, cusparseHandle_t
cusparse_handle,
+ size_t *buffer_size, void **buffer) {
/* y += A'*x
x and y MUST be on GPU already
*/
const scs_float onef = 1.0;
- CUSPARSE(csrmv)
- (cusparse_handle, CUSPARSE_OPERATION_NON_TRANSPOSE, Ag->n, Ag->m, Ag->Annz,
- &onef, Ag->descr, Ag->x, Ag->p, Ag->i, x, &onef, y);
+ size_t new_buffer_size = 0;
+
+ CUSPARSE_GEN(SpMV_bufferSize)
+ (cusparse_handle, CUSPARSE_OPERATION_NON_TRANSPOSE,
+ &onef, Ag->descr, x, &onef, y,
+ SCS_CUDA_FLOAT, SCS_CSRMV_ALG,
+ &new_buffer_size);
+
+ if (new_buffer_size > *buffer_size) {
+ if (*buffer != SCS_NULL) {
+ cudaFree(*buffer);
+ }
+ cudaMalloc(buffer, *buffer_size);
+ *buffer_size = new_buffer_size;
+ }
+
+ CUSPARSE_GEN(SpMV)
+ (cusparse_handle, CUSPARSE_OPERATION_NON_TRANSPOSE,
+ &onef, Ag->descr, x, &onef, y,
+ SCS_CUDA_FLOAT, SCS_CSRMV_ALG,
+ buffer);
}
-void SCS(_accum_by_a_gpu)(const ScsGpuMatrix *Ag, const scs_float *x,
- scs_float *y, cusparseHandle_t cusparse_handle) {
+void SCS(_accum_by_a_gpu)(const ScsGpuMatrix *Ag, const cusparseDnVecDescr_t x,
+ cusparseDnVecDescr_t y, cusparseHandle_t
cusparse_handle,
+ size_t *buffer_size, void **buffer) {
/* y += A*x
x and y MUST be on GPU already
*/
const scs_float onef = 1.0;
+ size_t new_buffer_size = 0;
+
/* The A matrix idx pointers must be ORDERED */
- CUSPARSE(csrmv)
- (cusparse_handle, CUSPARSE_OPERATION_TRANSPOSE, Ag->n, Ag->m, Ag->Annz,
&onef,
- Ag->descr, Ag->x, Ag->p, Ag->i, x, &onef, y);
+
+ CUSPARSE_GEN(SpMV_bufferSize)
+ (cusparse_handle, CUSPARSE_OPERATION_TRANSPOSE,
+ &onef, Ag->descr, x, &onef, y,
+ SCS_CUDA_FLOAT, SCS_CSRMV_ALG,
+ &new_buffer_size);
+
+ if (new_buffer_size > *buffer_size) {
+ if (*buffer != SCS_NULL) {
+ cudaFree(*buffer);
+ }
+ cudaMalloc(buffer, *buffer_size);
+ *buffer_size = new_buffer_size;
+ }
+
+ CUSPARSE_GEN(SpMV)
+ (cusparse_handle, CUSPARSE_OPERATION_TRANSPOSE,
+ &onef, Ag->descr, x, &onef, y,
+ SCS_CUDA_FLOAT, SCS_CSRMV_ALG,
+ buffer);
}
void SCS(free_gpu_matrix)(ScsGpuMatrix *A) {
cudaFree(A->x);
cudaFree(A->i);
cudaFree(A->p);
- cusparseDestroyMatDescr(A->descr);
+ cusparseDestroySpMat(A->descr);
}
void SCS(normalize_a)(ScsMatrix *A, const ScsSettings *stgs, const ScsCone *k,
diff -urN '--exclude=CVS' '--exclude=.cvsignore' '--exclude=.svn'
'--exclude=.svnignore' old/scs-2.1.2/linsys/gpu/gpu.h
new/scs-2.1.3/linsys/gpu/gpu.h
--- old/scs-2.1.2/linsys/gpu/gpu.h 2020-06-05 11:04:23.000000000 +0200
+++ new/scs-2.1.3/linsys/gpu/gpu.h 2021-04-12 10:43:08.000000000 +0200
@@ -34,6 +34,7 @@
#define CUBLAS(x) cublasS##x
#define CUSPARSE(x) cusparseS##x
#endif
+#define CUSPARSE_GEN(x) cusparse##x
#else
#ifndef SFLOAT
#define CUBLAS(x) \
@@ -50,8 +51,26 @@
CUDA_CHECK_ERR; \
cusparseS##x
#endif
+#define CUSPARSE_GEN(x) \
+ CUDA_CHECK_ERR; \
+ cusparse##x
#endif
+#ifndef SFLOAT
+#define SCS_CUDA_FLOAT CUDA_R_64F
+#else
+#define SCS_CUDA_FLOAT CUDA_R_32F
+#endif
+
+#ifndef DLONG
+#define SCS_CUSPARSE_INDEX CUSPARSE_INDEX_32I
+#else
+#define SCS_CUSPARSE_INDEX CUSPARSE_INDEX_64I
+#endif
+
+#define SCS_CSRMV_ALG CUSPARSE_CSRMV_ALG1
+#define SCS_CSR2CSC_ALG CUSPARSE_CSR2CSC_ALG1
+
/*
CUDA matrix routines only for CSR, not CSC matrices:
CSC CSR GPU Mult
@@ -68,14 +87,16 @@
scs_int m, n; /* m rows, n cols */
scs_int Annz; /* num non-zeros in A matrix */
/* CUDA */
- cusparseMatDescr_t descr;
+ cusparseSpMatDescr_t descr;
} ScsGpuMatrix;
-void SCS(_accum_by_atrans_gpu)(const ScsGpuMatrix *A, const scs_float *x,
- scs_float *y, cusparseHandle_t cusparse_handle);
-
-void SCS(_accum_by_a_gpu)(const ScsGpuMatrix *A, const scs_float *x,
- scs_float *y, cusparseHandle_t cusparse_handle);
+void SCS(_accum_by_atrans_gpu)(const ScsGpuMatrix *A, const
cusparseDnVecDescr_t x,
+ cusparseDnVecDescr_t y, cusparseHandle_t
cusparse_handle,
+ size_t *buffer_size, void **buffer);
+
+void SCS(_accum_by_a_gpu)(const ScsGpuMatrix *A, const cusparseDnVecDescr_t x,
+ cusparseDnVecDescr_t y, cusparseHandle_t
cusparse_handle,
+ size_t *buffer_size, void **buffer);
void SCS(free_gpu_matrix)(ScsGpuMatrix *A);
diff -urN '--exclude=CVS' '--exclude=.cvsignore' '--exclude=.svn'
'--exclude=.svnignore' old/scs-2.1.2/linsys/gpu/indirect/private.c
new/scs-2.1.3/linsys/gpu/indirect/private.c
--- old/scs-2.1.2/linsys/gpu/indirect/private.c 2020-06-05 11:04:23.000000000
+0200
+++ new/scs-2.1.3/linsys/gpu/indirect/private.c 2021-04-12 10:43:08.000000000
+0200
@@ -10,7 +10,14 @@
scs_float *v_n = p->r;
cudaMemcpy(v_m, x, A->m * sizeof(scs_float), cudaMemcpyHostToDevice);
cudaMemcpy(v_n, y, A->n * sizeof(scs_float), cudaMemcpyHostToDevice);
- SCS(_accum_by_atrans_gpu)(p->Ag, v_m, v_n, p->cusparse_handle);
+
+ cusparseDnVecSetValues(p->dn_vec_m, (void *) v_m);
+ cusparseDnVecSetValues(p->dn_vec_n, (void *) v_n);
+ SCS(_accum_by_atrans_gpu)(
+ p->Ag, p->dn_vec_m, p->dn_vec_n, p->cusparse_handle,
+ &p->buffer_size, &p->buffer
+ );
+
cudaMemcpy(y, v_n, A->n * sizeof(scs_float), cudaMemcpyDeviceToHost);
}
@@ -21,11 +28,21 @@
scs_float *v_n = p->r;
cudaMemcpy(v_n, x, A->n * sizeof(scs_float), cudaMemcpyHostToDevice);
cudaMemcpy(v_m, y, A->m * sizeof(scs_float), cudaMemcpyHostToDevice);
+
+ cusparseDnVecSetValues(p->dn_vec_m, (void *) v_m);
+ cusparseDnVecSetValues(p->dn_vec_n, (void *) v_n);
#if GPU_TRANSPOSE_MAT > 0
- SCS(_accum_by_atrans_gpu)(p->Agt, v_n, v_m, p->cusparse_handle);
+ SCS(_accum_by_atrans_gpu)(
+ p->Agt, p->dn_vec_n, p->dn_vec_m, p->cusparse_handle,
+ &p->buffer_size, &p->buffer
+ );
#else
- SCS(_accum_by_a_gpu)(p->Ag, v_n, v_m, p->cusparse_handle);
+ SCS(_accum_by_a_gpu)(
+ p->Ag, p->dn_vec_n, p->dn_vec_m, p->cusparse_handle,
+ &p->buffer_size, &p->buffer
+ );
#endif
+
cudaMemcpy(y, v_m, A->m * sizeof(scs_float), cudaMemcpyDeviceToHost);
}
@@ -64,6 +81,11 @@
SCS(free_gpu_matrix)(p->Agt);
scs_free(p->Agt);
}
+ if (p->buffer != SCS_NULL) {
+ cudaFree(p->buffer);
+ }
+ cusparseDestroyDnVec(p->dn_vec_m);
+ cusparseDestroyDnVec(p->dn_vec_n);
cusparseDestroy(p->cusparse_handle);
cublasDestroy(p->cublas_handle);
/* Don't reset because it interferes with other GPU programs. */
@@ -78,9 +100,30 @@
/* x and y MUST already be loaded to GPU */
scs_float *tmp_m = p->tmp_m; /* temp memory */
cudaMemset(tmp_m, 0, A->m * sizeof(scs_float));
- SCS(_accum_by_a_gpu)(A, x, tmp_m, p->cusparse_handle);
+
+ cusparseDnVecSetValues(p->dn_vec_m, (void *) tmp_m);
+ cusparseDnVecSetValues(p->dn_vec_n, (void *) x);
+#if GPU_TRANSPOSE_MAT > 0
+ SCS(_accum_by_atrans_gpu)(
+ p->Agt, p->dn_vec_n, p->dn_vec_m, p->cusparse_handle,
+ &p->buffer_size, &p->buffer
+ );
+#else
+ SCS(_accum_by_a_gpu)(
+ A, p->dn_vec_n, p->dn_vec_m, p->cusparse_handle,
+ &p->buffer_size, &p->buffer
+ );
+#endif
+
cudaMemset(y, 0, A->n * sizeof(scs_float));
- SCS(_accum_by_atrans_gpu)(A, tmp_m, y, p->cusparse_handle);
+
+ cusparseDnVecSetValues(p->dn_vec_m, (void *) tmp_m);
+ cusparseDnVecSetValues(p->dn_vec_n, (void *) y);
+ SCS(_accum_by_atrans_gpu)(
+ A, p->dn_vec_m, p->dn_vec_n, p->cusparse_handle,
+ &p->buffer_size, &p->buffer
+ );
+
CUBLAS(axpy)(p->cublas_handle, A->n, &(s->rho_x), x, 1, y, 1);
}
@@ -112,6 +155,14 @@
cudaError_t err;
ScsLinSysWork *p = (ScsLinSysWork *)scs_calloc(1, sizeof(ScsLinSysWork));
ScsGpuMatrix *Ag = (ScsGpuMatrix *)scs_malloc(sizeof(ScsGpuMatrix));
+
+ /* Used for initializing dense vectors */
+ scs_float *tmp_null_n = SCS_NULL;
+ scs_float *tmp_null_m = SCS_NULL;
+
+#if GPU_TRANSPOSE_MAT > 0
+ size_t new_buffer_size = 0;
+#endif
p->cublas_handle = 0;
p->cusparse_handle = 0;
@@ -119,6 +170,9 @@
p->total_solve_time = 0;
p->tot_cg_its = 0;
+ p->buffer_size = 0;
+ p->buffer = SCS_NULL;
+
/* Get handle to the CUBLAS context */
cublasCreate(&p->cublas_handle);
@@ -130,9 +184,7 @@
Ag->Annz = A->p[A->n];
Ag->descr = 0;
/* Matrix description */
- cusparseCreateMatDescr(&Ag->descr);
- cusparseSetMatType(Ag->descr, CUSPARSE_MATRIX_TYPE_GENERAL);
- cusparseSetMatIndexBase(Ag->descr, CUSPARSE_INDEX_BASE_ZERO);
+
p->Ag = Ag;
p->Agt = SCS_NULL;
@@ -155,6 +207,18 @@
cudaMemcpy(Ag->x, A->x, (A->p[A->n]) * sizeof(scs_float),
cudaMemcpyHostToDevice);
+ cusparseCreateCsr
+ (&Ag->descr, Ag->n, Ag->m, Ag->Annz, Ag->p, Ag->i, Ag->x,
+ SCS_CUSPARSE_INDEX, SCS_CUSPARSE_INDEX,
+ CUSPARSE_INDEX_BASE_ZERO, SCS_CUDA_FLOAT);
+
+ cudaMalloc((void **)&tmp_null_n, A->n * sizeof(scs_float));
+ cudaMalloc((void **)&tmp_null_m, A->m * sizeof(scs_float));
+ cusparseCreateDnVec(&p->dn_vec_n, Ag->n, tmp_null_n, SCS_CUDA_FLOAT);
+ cusparseCreateDnVec(&p->dn_vec_m, Ag->m, tmp_null_m, SCS_CUDA_FLOAT);
+ cudaFree(tmp_null_n);
+ cudaFree(tmp_null_m);
+
get_preconditioner(A, stgs, p);
#if GPU_TRANSPOSE_MAT > 0
@@ -164,18 +228,41 @@
p->Agt->Annz = A->p[A->n];
p->Agt->descr = 0;
/* Matrix description */
- cusparseCreateMatDescr(&p->Agt->descr);
- cusparseSetMatType(p->Agt->descr, CUSPARSE_MATRIX_TYPE_GENERAL);
- cusparseSetMatIndexBase(p->Agt->descr, CUSPARSE_INDEX_BASE_ZERO);
cudaMalloc((void **)&p->Agt->i, (A->p[A->n]) * sizeof(scs_int));
cudaMalloc((void **)&p->Agt->p, (A->m + 1) * sizeof(scs_int));
cudaMalloc((void **)&p->Agt->x, (A->p[A->n]) * sizeof(scs_float));
/* transpose Ag into Agt for faster multiplies */
/* TODO: memory intensive, could perform transpose in CPU and copy to GPU */
- CUSPARSE(csr2csc)
- (p->cusparse_handle, A->n, A->m, A->p[A->n], Ag->x, Ag->p, Ag->i, p->Agt->x,
- p->Agt->i, p->Agt->p, CUSPARSE_ACTION_NUMERIC, CUSPARSE_INDEX_BASE_ZERO);
+ cusparseCsr2cscEx2_bufferSize
+ (p->cusparse_handle, A->n, A->m, A->p[A->n],
+ Ag->x, Ag->p, Ag->i,
+ p->Agt->x, p->Agt->p, p->Agt->i,
+ SCS_CUDA_FLOAT, CUSPARSE_ACTION_NUMERIC,
+ CUSPARSE_INDEX_BASE_ZERO, SCS_CSR2CSC_ALG,
+ &new_buffer_size);
+
+ if (new_buffer_size > p->buffer_size) {
+ if (p->buffer != SCS_NULL) {
+ cudaFree(p->buffer);
+ }
+ cudaMalloc(&p->buffer, new_buffer_size);
+ p->buffer_size = new_buffer_size;
+ }
+
+ cusparseCsr2cscEx2
+ (p->cusparse_handle, A->n, A->m, A->p[A->n],
+ Ag->x, Ag->p, Ag->i,
+ p->Agt->x, p->Agt->p, p->Agt->i,
+ SCS_CUDA_FLOAT, CUSPARSE_ACTION_NUMERIC,
+ CUSPARSE_INDEX_BASE_ZERO, SCS_CSR2CSC_ALG,
+ p->buffer);
+
+ cusparseCreateCsr
+ (&p->Agt->descr, p->Agt->n, p->Agt->m, p->Agt->Annz,
+ p->Agt->p, p->Agt->i, p->Agt->x,
+ SCS_CUSPARSE_INDEX, SCS_CUSPARSE_INDEX,
+ CUSPARSE_INDEX_BASE_ZERO, SCS_CUDA_FLOAT);
#endif
err = cudaGetLastError();
@@ -285,11 +372,32 @@
SCS(tic)(&linsys_timer);
/* all on GPU */
cudaMemcpy(bg, b, (Ag->n + Ag->m) * sizeof(scs_float),
cudaMemcpyHostToDevice);
- SCS(_accum_by_atrans_gpu)(Ag, &(bg[Ag->n]), bg, p->cusparse_handle);
+
+ cusparseDnVecSetValues(p->dn_vec_m, (void *) &(bg[Ag->n]));
+ cusparseDnVecSetValues(p->dn_vec_n, (void *) bg);
+ SCS(_accum_by_atrans_gpu)(
+ Ag, p->dn_vec_m, p->dn_vec_n, p->cusparse_handle,
+ &p->buffer_size, &p->buffer
+ );
+
/* solves (I+A'A)x = b, s warm start, solution stored in b */
cg_its = pcg(p->Ag, stgs, p, s, bg, Ag->n, MAX(cg_tol, CG_BEST_TOL));
CUBLAS(scal)(p->cublas_handle, Ag->m, &neg_onef, &(bg[Ag->n]), 1);
- SCS(_accum_by_a_gpu)(Ag, bg, &(bg[Ag->n]), p->cusparse_handle);
+
+ cusparseDnVecSetValues(p->dn_vec_m, (void *) &(bg[Ag->n]));
+ cusparseDnVecSetValues(p->dn_vec_n, (void *) bg);
+#if GPU_TRANSPOSE_MAT > 0
+ SCS(_accum_by_atrans_gpu)(
+ p->Agt, p->dn_vec_n, p->dn_vec_m, p->cusparse_handle,
+ &p->buffer_size, &p->buffer
+ );
+#else
+ SCS(_accum_by_a_gpu)(
+ Ag, p->dn_vec_n, p->dn_vec_m, p->cusparse_handle,
+ &p->buffer_size, &p->buffer
+ );
+#endif
+
cudaMemcpy(b, bg, (Ag->n + Ag->m) * sizeof(scs_float),
cudaMemcpyDeviceToHost);
if (iter >= 0) {
diff -urN '--exclude=CVS' '--exclude=.cvsignore' '--exclude=.svn'
'--exclude=.svnignore' old/scs-2.1.2/linsys/gpu/indirect/private.h
new/scs-2.1.3/linsys/gpu/indirect/private.h
--- old/scs-2.1.2/linsys/gpu/indirect/private.h 2020-06-05 11:04:23.000000000
+0200
+++ new/scs-2.1.3/linsys/gpu/indirect/private.h 2021-04-12 10:43:08.000000000
+0200
@@ -28,6 +28,11 @@
/* CUDA */
cublasHandle_t cublas_handle;
cusparseHandle_t cusparse_handle;
+ /* CUSPARSE */
+ size_t buffer_size;
+ void *buffer;
+ cusparseDnVecDescr_t dn_vec_m; /* Dense vector of length m */
+ cusparseDnVecDescr_t dn_vec_n; /* Dense vector of length n */
};
#ifdef __cplusplus
diff -urN '--exclude=CVS' '--exclude=.cvsignore' '--exclude=.svn'
'--exclude=.svnignore' old/scs-2.1.2/src/rw.c new/scs-2.1.3/src/rw.c
--- old/scs-2.1.2/src/rw.c 2020-06-05 11:04:23.000000000 +0200
+++ new/scs-2.1.3/src/rw.c 2021-04-12 10:43:08.000000000 +0200
@@ -37,7 +37,7 @@
fread(&(k->ep), sizeof(scs_int), 1, fin);
fread(&(k->ed), sizeof(scs_int), 1, fin);
fread(&(k->psize), sizeof(scs_int), 1, fin);
- k->p = scs_calloc(k->psize, sizeof(scs_int));
+ k->p = scs_calloc(k->psize, sizeof(scs_float));
fread(k->p, sizeof(scs_float), k->psize, fin);
return k;
}
