Gpu fix (#199)

* fix gpu * add gpu action * workflow namings
cvxgrp · Jan 11, 2022 · 515c905 · 515c905
1 parent 5a40b44
commit 515c905
Show file tree

Hide file tree

Showing 7 changed files with 26 additions and 9 deletions.
diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
@@ -1,5 +1,5 @@
 ---
-name: Build
+name: Build and Test
 
 on: [push, pull_request]
 

diff --git a/.github/workflows/cmake.yml b/.github/workflows/cmake.yml
@@ -1,5 +1,5 @@
 ---
-name: Cmake
+name: CMake
 
 on: [push, pull_request]
 

diff --git a/.github/workflows/gpu.yml b/.github/workflows/gpu.yml
@@ -0,0 +1,16 @@
+---
+name: GPU
+
+on: [push, pull_request]
+
+jobs:
+  linux:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+      - uses: Jimver/cuda-toolkit@v0.2.5
+      - run: sudo apt-get install libopenblas-dev liblapack-dev
+      - run: make gpu
+      - run: make test_gpu
+      # - run: out/run_tests_gpu_indirect    # gpus not available yet
+
diff --git a/Makefile b/Makefile
@@ -115,7 +115,7 @@ $(LINSYS)/gpu/gpu.o: $(LINSYS)/gpu/gpu.c
 # 	$(CUCC) -c -o $(GPUDIR)/private.o $^ $(CUDAFLAGS)
 
 $(GPUINDIR)/private.o: $(GPUINDIR)/private.c
-	$(CUCC) -c -o $(GPUINDIR)/private.o $^ $(CUDAFLAGS)
+	$(CUCC) -c -o $@ $^ $(CUDAFLAGS)
 
 # $(OUT)/libscsgpudir.$(SHARED): $(SCS_O) $(SCS_OBJECTS) $(GPUDIR)/private.o $(AMD_OBJS) $(LINSYS)/scs_matrix.o $(LINSYS)/gpu/gpu.o
 #	 mkdir -p $(OUT)
@@ -143,7 +143,7 @@ $(OUT)/demo_socp_gpu_indirect: test/random_socp_prob.c $(OUT)/libscsgpuindir.a
 
 .PHONY: clean purge
 clean:
-	@rm -rf $(TARGETS) $(SCS_O) $(SCS_INDIR_O) $(SCS_OBJECTS) $(AMD_OBJS) $(LDL_OBJS) $(LINSYS)/*.o $(DIRSRC)/*.o $(INDIRSRC)/*.o $(GPUDIR)/*.o $(GPUINDIR)/*.o
+	@rm -rf $(TARGETS) $(SCS_O) $(SCS_INDIR_O) $(SCS_OBJECTS) $(AMD_OBJS) $(LDL_OBJS) $(LINSYS)/*.o $(DIRSRC)/*.o $(INDIRSRC)/*.o $(GPUDIR)/*.o $(GPUINDIR)/*.o $(LINSYS)/gpu/*.o
 	@rm -rf $(OUT)/*.dSYM
 	@rm -rf matlab/*.mex*
 	@rm -rf .idea

diff --git a/linsys/gpu/gpu.h b/linsys/gpu/gpu.h
@@ -1,5 +1,5 @@
-#ifndef SCSGPU_H_GUARD
-#define SCSGPU_H_GUARD
+#ifndef SCS_GPU_H_GUARD
+#define SCS_GPU_H_GUARD
 
 #ifdef __cplusplus
 extern "C" {

diff --git a/linsys/gpu/indirect/private.c b/linsys/gpu/indirect/private.c
@@ -233,8 +233,8 @@ ScsLinSysWork *SCS(init_lin_sys_work)(const ScsMatrix *A, const ScsMatrix *P,
   p = (ScsLinSysWork *)scs_calloc(1, sizeof(ScsLinSysWork));
   Ag = (ScsGpuMatrix *)scs_calloc(1, sizeof(ScsGpuMatrix));
 
-  p->inv_r_y = (scs_float *)scs_calloc(A->m * sizeof(scs_float));
-  p->M = (scs_float *)scs_calloc(A->n * sizeof(scs_float));
+  p->inv_r_y = (scs_float *)scs_calloc(A->m, sizeof(scs_float));
+  p->M = (scs_float *)scs_calloc(A->n, sizeof(scs_float));
 
   p->A = A;
   p->P = P;

diff --git a/linsys/gpu/indirect/private.h b/linsys/gpu/indirect/private.h
@@ -15,14 +15,15 @@ struct SCS_LIN_SYS_WORK {
   scs_int n, m; /* linear system dimensions */
   /* reporting */
   scs_int tot_cg_its;
+  scs_float *M; /* preconditioner on cpu */
   /* ALL BELOW HOSTED ON THE GPU */
   scs_float *p;       /* cg iterate, n  */
   scs_float *r;       /* cg residual, n */
   scs_float *Gp;      /* G * p, n */
   scs_float *bg;      /* b, n */
   scs_float *tmp_m;   /* m, used in mat_vec */
   scs_float *z;       /* preconditioned */
-  scs_float *M;       /* preconditioner */
+  scs_float *M_gpu;   /* preconditioner */
   const ScsMatrix *A; /* does *not* own this memory */
   const ScsMatrix *P; /* does *not* own this memory */
   ScsGpuMatrix *Ag;   /* A matrix on GPU */