gpu.h in scs-0.3.0

- old
+ new

@@ -8,53 +8,62 @@
 #include <cublas_v2.h>
 #include <cuda.h>
 #include <cuda_runtime_api.h>
 #include <cusparse.h>
 
-#include "amatrix.h"
 #include "glbopts.h"
 #include "linalg.h"
 #include "linsys.h"
 #include "scs.h"
+#include "scs_matrix.h"
 #include "util.h"
 
-#define CUDA_CHECK_ERR                                                    \
-  do {                                                                    \
-    cudaError_t err = cudaGetLastError();                                 \
-    if (err != cudaSuccess) {                                             \
-      printf("%s:%d:%s\n ERROR_CUDA: %s\n", __FILE__, __LINE__, __func__, \
-             cudaGetErrorString(err));                                    \
-    }                                                                     \
+#define CUDA_CHECK_ERR                                                         \
+  do {                                                                         \
+    cudaDeviceSynchronize();                                                   \
+    cudaError_t err = cudaGetLastError();                                      \
+    if (err != cudaSuccess) {                                                  \
+      scs_printf("%s:%d:%s\n ERROR_CUDA (#): %s\n", __FILE__, __LINE__,        \
+                 __func__, cudaGetErrorString(err));                           \
+    }                                                                          \
   } while (0)
 
-#ifndef EXTRA_VERBOSE
+#if VERBOSITY == 0
 #ifndef SFLOAT
 #define CUBLAS(x) cublasD##x
+#define CUBLASI(x) cublasId##x
 #define CUSPARSE(x) cusparseD##x
 #else
 #define CUBLAS(x) cublasS##x
+#define CUBLASI(x) cublasIs##x
 #define CUSPARSE(x) cusparseS##x
 #endif
 #define CUSPARSE_GEN(x) cusparse##x
 #else
 #ifndef SFLOAT
-#define CUBLAS(x) \
-  CUDA_CHECK_ERR; \
+#define CUBLAS(x)                                                              \
+  CUDA_CHECK_ERR;                                                              \
   cublasD##x
-#define CUSPARSE(x) \
-  CUDA_CHECK_ERR;   \
+#define CUBLASI(x)                                                             \
+  CUDA_CHECK_ERR;                                                              \
+  cublasId##x
+#define CUSPARSE(x)                                                            \
+  CUDA_CHECK_ERR;                                                              \
   cusparseD##x
 #else
-#define CUBLAS(x) \
-  CUDA_CHECK_ERR; \
+#define CUBLAS(x)                                                              \
+  CUDA_CHECK_ERR;                                                              \
   cublasS##x
-#define CUSPARSE(x) \
-  CUDA_CHECK_ERR;   \
+#define CUBLASI(x)                                                             \
+  CUDA_CHECK_ERR;                                                              \
+  cublasIs##x
+#define CUSPARSE(x)                                                            \
+  CUDA_CHECK_ERR;                                                              \
   cusparseS##x
 #endif
-#define CUSPARSE_GEN(x) \
-  CUDA_CHECK_ERR;       \
+#define CUSPARSE_GEN(x)                                                        \
+  CUDA_CHECK_ERR;                                                              \
   cusparse##x
 #endif
 
 #ifndef SFLOAT
 #define SCS_CUDA_FLOAT CUDA_R_64F
@@ -76,28 +85,36 @@
     CSC             CSR             GPU     Mult
     A (m x n)       A' (n x m)      Ag      accum_by_a_trans_gpu
     A'(n x m)       A  (m x n)      Agt     accum_by_a_gpu
 */
 
-/* this struct defines the data matrix A on GPU */
-typedef struct SCS_GPU_A_DATA_MATRIX {
+/* this struct defines the data matrix on GPU */
+typedef struct SCS_GPU_DATA_MATRIX {
   /* A is supplied in column compressed format */
-  scs_float *x; /* A values, size: NNZ A */
-  scs_int *i;   /* A row index, size: NNZ A */
-  scs_int *p;   /* A column pointer, size: n+1 */
+  scs_float *x; /* values, size: NNZ */
+  scs_int *i;   /* row index, size: NNZ */
+  scs_int *p;   /* column pointer, size: n+1 */
   scs_int m, n; /* m rows, n cols */
-  scs_int Annz; /* num non-zeros in A matrix */
+  scs_int nnz;  /* num non-zeros in matrix */
   /* CUDA */
   cusparseSpMatDescr_t descr;
 } ScsGpuMatrix;
 
-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_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(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(accum_by_p_gpu)(const ScsGpuMatrix *P, const cusparseDnVecDescr_t x,
+                         cusparseDnVecDescr_t y,
+                         cusparseHandle_t cusparse_handle, size_t *buffer_size,
+                         void **buffer);
 
 void SCS(free_gpu_matrix)(ScsGpuMatrix *A);
 
 #ifdef __cplusplus
 }