Merge branch 'blaslapack_api' into 'master'

Validate first algorithm gemm cblas and blas interfaces.

See merge request solverstack/chameleon!312

cmake_modules/local_subs.py

@@ -81,6 +81,7 @@ subs = {
         ('',                     'ORGLQ',                'ORGLQ',                'UNGLQ',                'UNGLQ'               ),
         ('',                     'SYEV',                 'SYEV',                 'HEEV',                 'HEEV'                ),
         ('',                     'SYG',                  'SYG',                  'HEG',                  'HEG'                 ),
+        ('',                     'const float ',         'const double ',        'const void \*',        'const void \*'       ),
     ]
     + _extra_blas
     + _extra_BLAS

compute/CMakeLists.txt

@@ -264,10 +264,22 @@ precisions_rules_py(CONTROL_SRCS_GENERATED "${ZSRC}"
                     PRECISIONS "${CHAMELEON_PRECISION}"
                     TARGETDIR "control" )
 
+set(LAPACK_API_SRCS_GENERATED "")
+set(ZSRC
+  lapack_api/lapack_zgemm.c
+  )
+precisions_rules_py(LAPACK_API_SRCS_GENERATED "${ZSRC}"
+  PRECISIONS "${CHAMELEON_PRECISION}")
+set(LAPACK_API_SRCS
+  ${LAPACK_API_SRCS_GENERATED}
+  lapack_api/lapack_api_common.c
+  )
+
 set(CHAMELEON_SRCS
     ${CHAMELEON_CONTROL}
     ${CHAMELEON_SRCS_GENERATED}
     ${CONTROL_SRCS_GENERATED}
+    ${LAPACK_API_SRCS}
    )
 
 # Generate the chameleon fortran sources for all possible precisions
@@ -309,7 +321,7 @@ add_dependencies(chameleon
 )
 
 target_include_directories(chameleon PUBLIC
-  $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/include>
+  $<BUILD_INTERFACE:${CMAKE_CURRENT_SOURCE_DIR}/lapack_api>
   $<BUILD_INTERFACE:${CHAMELEON_SOURCE_DIR}/include>
   $<BUILD_INTERFACE:${CHAMELEON_BINARY_DIR}/include>
   $<BUILD_INTERFACE:${CHAMELEON_SOURCE_DIR}/control>
@@ -329,7 +341,9 @@ elseif(CHAMELEON_SCHED_QUARK)
 elseif(CHAMELEON_SCHED_OPENMP)
   target_link_libraries(chameleon PUBLIC chameleon_openmp)
 endif()
-if (NOT CHAMELEON_SIMULATION)
+if (CHAMELEON_SIMULATION)
+  target_include_directories(chameleon PUBLIC $<BUILD_INTERFACE:${CHAMELEON_SOURCE_DIR}/coreblas/include>)
+else()
   target_link_libraries(chameleon PUBLIC coreblas)
 endif()
 target_link_libraries(chameleon PUBLIC hqr)

compute/lapack_api/lapack_api_common.c

+/**
+ *
+ * @file lapack_api_common.c
+ *
+ * @copyright 2022 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
+ *                 Univ. Bordeaux. All rights reserved.
+ *
+ ***
+ *
+ * @brief Chameleon blas/lapack and cblas/lapack api common functions
+ *
+ * @version 1.2.0
+ * @author Mathieu Faverge
+ * @author Florent Pruvost
+ * @date 2022-04-22
+ *
+ */
+#include "lapack_api_common.h"
+
+/**
+ * @brief Convert the input char BLAS trans parameter to a compatible parameter
+ * for the Cblas API.
+ * @param[in] trans The input char BLAS trans parameter
+ * @return The CBLAS equivalent parameter (CblasNoTrans, CblasTrans or
+ * CblasConjTrans).
+ */
+int chameleon_blastocblas_trans(const char* trans)
+{
+    if ( (*trans == 'N') || (*trans == 'n') ) {
+        return CblasNoTrans;
+    } else if ( (*trans == 'T') || (*trans == 't') ) {
+        return CblasTrans;
+    } else if ( (*trans == 'C') || (*trans == 'c') ) {
+        return CblasConjTrans;
+    } else {
+        chameleon_error("chameleon_blastocblas_trans", "illegal value of BLAS transpose parameter");
+        return CHAMELEON_ERR_ILLEGAL_VALUE;
+    }
+}

compute/lapack_api/lapack_api_common.h

+/**
+ *
+ * @file lapack_api_common.h
+ *
+ * @copyright 2022 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria, Univ. Bordeaux.
+ *                 All rights reserved.
+ *
+ ***
+ *
+ * @brief Chameleon blas/lapack and cblas/lapack api common internal functions
+ *
+ * @version 1.2.0
+ * @author Mathieu Faverge
+ * @author Florent Pruvost
+ * @date 2022-04-22
+ *
+ */
+#ifndef _lapack_api_common_h_
+#define _lapack_api_common_h_
+
+#include "chameleon.h"
+#include "chameleon/mangling.h"
+#include "control/auxiliary.h"
+
+/**
+ *
+ * @defgroup CHAMELEON_LAPACK_API
+ * @brief Linear algebra routines exposed to users. LAPACK matrix data storage
+ *
+ */
+
+int chameleon_blastocblas_trans(const char* value);
+
+#endif /* _lapack_api_common_h_ */

compute/lapack_api/lapack_zgemm.c

+/**
+ *
+ * @file lapack_zgemm.c
+ *
+ * @copyright 2022 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
+ *                 Univ. Bordeaux. All rights reserved.
+ *
+ ***
+ *
+ * @brief Chameleon blas and cblas api for gemm
+ *
+ * @version 1.2.0
+ * @author Mathieu Faverge
+ * @author Florent Pruvost
+ * @date 2022-04-22
+ * @precisions normal z -> s d c
+ *
+ */
+
+#include "lapack_api_common.h"
+
+/* Fortran BLAS interface */
+
+#define CHAMELEON_blas_zgemm CHAMELEON_GLOBAL( chameleon_blas_zgemm, CHAMELEON_BLAS_ZGEMM )
+void CHAMELEON_blas_zgemm ( const char* transa, const char* transb,
+                            const int* m, const int* n, const int* k,
+                            const CHAMELEON_Complex64_t* alpha, const CHAMELEON_Complex64_t* a, const int* lda,
+                                                                const CHAMELEON_Complex64_t* b, const int* ldb,
+                            const CHAMELEON_Complex64_t* beta,  CHAMELEON_Complex64_t* c, const int* ldc )
+{
+    CHAMELEON_cblas_zgemm( CblasColMajor,
+                           chameleon_blastocblas_trans(transa),
+                           chameleon_blastocblas_trans(transb),
+                           *m, *n, *k,
+                           CBLAS_SADDR(*alpha), a, *lda,
+                           b, *ldb,
+                           CBLAS_SADDR(*beta), c, *ldc );
+}
+
+/* C CBLAS interface */
+
+/**
+ ********************************************************************************
+ *
+ * @ingroup CHAMELEON_LAPACK_API
+ *
+ *  CHAMELEON_cblas_zgemm - Performs one of the matrix-matrix operations
+ *
+ *    \f[ C = \alpha [op( A )\times op( B )] + \beta C \f],
+ *
+ *  where op( X ) is one of
+ *
+ *    op( X ) = X  or op( X ) = X' or op( X ) = conjg( X' )
+ *
+ *  alpha and beta are scalars, and A, B and C  are matrices, with op( A )
+ *  an m by k matrix, op( B ) a k by n matrix and C an m by n matrix.
+ *
+ *******************************************************************************
+ *
+ * @param[in] transA
+ *          Specifies whether the matrix A is transposed, not transposed or conjugate transposed:
+ *          = ChamNoTrans:   A is not transposed;
+ *          = ChamTrans:     A is transposed;
+ *          = ChamConjTrans: A is conjugate transposed.
+ *
+ * @param[in] transB
+ *          Specifies whether the matrix B is transposed, not transposed or conjugate transposed:
+ *          = ChamNoTrans:   B is not transposed;
+ *          = ChamTrans:     B is transposed;
+ *          = ChamConjTrans: B is conjugate transposed.
+ *
+ * @param[in] M
+ *          M specifies the number of rows of the matrix op( A ) and of the matrix C. M >= 0.
+ *
+ * @param[in] N
+ *          N specifies the number of columns of the matrix op( B ) and of the matrix C. N >= 0.
+ *
+ * @param[in] K
+ *          K specifies the number of columns of the matrix op( A ) and the number of rows of
+ *          the matrix op( B ). K >= 0.
+ *
+ * @param[in] alpha
+ *          alpha specifies the scalar alpha
+ *
+ * @param[in] A
+ *          A is a LDA-by-ka matrix, where ka is K when  transA = ChamNoTrans,
+ *          and is  M  otherwise.
+ *
+ * @param[in] LDA
+ *          The leading dimension of the array A. LDA >= max(1,M).
+ *
+ * @param[in] B
+ *          B is a LDB-by-kb matrix, where kb is N when  transB = ChamNoTrans,
+ *          and is  K  otherwise.
+ *
+ * @param[in] LDB
+ *          The leading dimension of the array B. LDB >= max(1,N).
+ *
+ * @param[in] beta
+ *          beta specifies the scalar beta
+ *
+ * @param[in,out] C
+ *          C is a LDC-by-N matrix.
+ *          On exit, the array is overwritten by the M by N matrix ( alpha*op( A )*op( B ) + beta*C )
+ *
+ * @param[in] LDC
+ *          The leading dimension of the array C. LDC >= max(1,M).
+ *
+ *******************************************************************************
+ *
+ * @retval CHAMELEON_SUCCESS successful exit
+ *
+ *******************************************************************************
+ *
+ * @sa CHAMELEON_cblas_zgemm
+ * @sa CHAMELEON_cblas_cgemm
+ * @sa CHAMELEON_cblas_dgemm
+ * @sa CHAMELEON_cblas_sgemm
+ *
+ */
+void CHAMELEON_cblas_zgemm( const CBLAS_ORDER Order, const CBLAS_TRANSPOSE TransA, const CBLAS_TRANSPOSE TransB,
+                            const int M, const int N, const int K,
+                            const void *alpha, const CHAMELEON_Complex64_t *A, const int lda,
+                                               const CHAMELEON_Complex64_t *B, const int ldb,
+                            const void *beta,        CHAMELEON_Complex64_t *C, const int ldc )
+{
+    if (Order != CblasColMajor){
+        chameleon_error("CHAMELEON_cblas_zgemm", "illegal value of order");
+    }
+
+#if defined(PRECISION_z) || defined(PRECISION_c)
+    CHAMELEON_Complex64_t alphac = *(CHAMELEON_Complex64_t *)alpha;
+    CHAMELEON_Complex64_t betac = *(CHAMELEON_Complex64_t *)beta;
+#else
+    CHAMELEON_Complex64_t alphac = alpha;
+    CHAMELEON_Complex64_t betac = beta;
+#endif
+
+    CHAMELEON_zgemm( (cham_trans_t)TransA, (cham_trans_t)TransB, M, N, K,
+                     alphac, (CHAMELEON_Complex64_t *)A, lda,
+                     (CHAMELEON_Complex64_t *)B, ldb,
+                     betac, (CHAMELEON_Complex64_t *)C, ldc );
+}

coreblas/compute/core_dzasum.c

@@ -20,7 +20,6 @@
  * @precisions normal z -> c d s
  *
  */
-#include "coreblas/cblas.h"
 #include <math.h>
 #include "coreblas.h"
 

coreblas/compute/core_zgessm.c

@@ -24,7 +24,6 @@
  * @precisions normal z -> c d s
  *
  */
-#include "coreblas/cblas.h"
 #include "coreblas/lapacke.h"
 #include "coreblas.h"
 

coreblas/compute/core_zlascal.c

@@ -20,7 +20,6 @@
  *
  */
 #include "coreblas.h"
-#include "coreblas/cblas.h"
 #include <math.h>
 
 /**

coreblas/compute/core_zpamm.c

@@ -23,7 +23,6 @@
  * @precisions normal z -> c d s
  *
  */
-#include "coreblas/cblas.h"
 #include "coreblas/lapacke.h"
 #include "coreblas.h"
 

coreblas/compute/core_zparfb.c

@@ -18,7 +18,6 @@
  * @precisions normal z -> c d s
  *
  */
-#include "coreblas/cblas.h"
 #include "coreblas/lapacke.h"
 #include "coreblas.h"
 

coreblas/compute/core_zpemv.c

@@ -23,7 +23,6 @@
  * @precisions normal z -> c d s
  *
  */
-#include "coreblas/cblas.h"
 #include "coreblas/lapacke.h"
 #include "coreblas.h"
 

coreblas/compute/core_zssssm.c

@@ -24,7 +24,6 @@
  * @precisions normal z -> c d s
  *
  */
-#include "coreblas/cblas.h"
 #include "coreblas.h"
 
 /**

coreblas/compute/core_ztrasm.c

@@ -20,7 +20,6 @@
  * @precisions normal z -> c d s
  *
  */
-#include "coreblas/cblas.h"
 #include <math.h>
 #include "coreblas.h"
 

coreblas/compute/core_ztstrf.c

@@ -25,7 +25,6 @@
  *
  */
 #include "coreblas.h"
-#include "coreblas/cblas.h"
 #include <math.h>
 
 /**

coreblas/include/coreblas.h

@@ -29,14 +29,7 @@
 #include <string.h>
 #include <assert.h>
 
-/**
- *  CBLAS requires for scalar arguments to be passed
- *        by address rather than by value
- */
-#ifndef CBLAS_SADDR
-#define CBLAS_SADDR( _val_ ) &(_val_)
-#endif
-#include "coreblas/cblas.h"
+#include "coreblas/cblas_wrapper.h"
 
 /**
  * CHAMELEON types and constants
@@ -68,14 +61,6 @@ END_C_DECLS
         assert(0);                                                      \
     } while(0)
 
-/**
- * CBlas enum
- */
-#define CBLAS_TRANSPOSE enum CBLAS_TRANSPOSE
-#define CBLAS_UPLO      enum CBLAS_UPLO
-#define CBLAS_DIAG      enum CBLAS_DIAG
-#define CBLAS_SIDE      enum CBLAS_SIDE
-
 /**
  *  LAPACK Constants
  */

coreblas/include/coreblas/cblas_wrapper.h

+/**
+ *
+ * @file cblas_wrapper.h
+ *
+ * @copyright 2009-2014 The University of Tennessee and The University of
+ *                      Tennessee Research Foundation. All rights reserved.
+ * @copyright 2012-2022 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
+ *                      Univ. Bordeaux. All rights reserved.
+ *
+ ***
+ *
+ * @brief Chameleon cblas header wrapper
+ *
+ * @version 1.2.0
+ * @author Cedric Castagnede
+ * @author Florent Pruvost
+ * @author Mathieu Faverge
+ * @date 2022-02-22
+ *
+ */
+#ifndef _cblas_wrapper_h_
+#define _cblas_wrapper_h_
+
+/**
+ *  CBLAS requires for scalar arguments to be passed
+ *        by address rather than by value
+ */
+#ifndef CBLAS_SADDR
+#define CBLAS_SADDR( _val_ ) &(_val_)
+#endif
+#include "coreblas/cblas.h"
+
+/**
+ * CBlas enum
+ */
+#define CBLAS_ORDER     enum CBLAS_ORDER
+#define CBLAS_TRANSPOSE enum CBLAS_TRANSPOSE
+#define CBLAS_UPLO      enum CBLAS_UPLO
+#define CBLAS_DIAG      enum CBLAS_DIAG
+#define CBLAS_SIDE      enum CBLAS_SIDE
+
+#endif /* _cblas_wrapper_h_ */

include/CMakeLists.txt

@@ -30,10 +30,11 @@
 # ----------------------------------------------------------
 set(CHAMELEON_HDRS_GENERATED "")
 set(ZHDR
-    chameleon/chameleon_z.h
-    chameleon/chameleon_zc.h
-    chameleon/tasks_z.h
-    chameleon/tasks_zc.h
+  chameleon/chameleon_z.h
+  chameleon/chameleon_zc.h
+  chameleon/chameleon_zlapack.h
+  chameleon/tasks_z.h
+  chameleon/tasks_zc.h
 )
 
 precisions_rules_py(CHAMELEON_HDRS_GENERATED "${ZHDR}"

include/chameleon.h

@@ -68,6 +68,11 @@ typedef struct gepdf_info_s {
 #include "chameleon/chameleon_zc.h"
 #include "chameleon/chameleon_ds.h"
 
+#include "chameleon/chameleon_zlapack.h"
+#include "chameleon/chameleon_clapack.h"
+#include "chameleon/chameleon_dlapack.h"
+#include "chameleon/chameleon_slapack.h"
+
 BEGIN_C_DECLS
 
 /* ****************************************************************************

include/chameleon/chameleon_zlapack.h

+/**
+ *
+ * @file chameleon_zlapack.h
+ *
+ * @copyright 2022-2022 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
+ *                      Univ. Bordeaux. All rights reserved.
+ *
+ ***
+ *
+ * @brief Chameleon blas/lapack and cblas/lapack api functions
+ *
+ * @version 1.2.0
+ * @author Mathieu Faverge
+ * @author Florent Pruvost
+ * @date 2022-04-26
+ * @precisions normal z -> c d s
+ *
+ */
+#ifndef _chameleon_zlapack_h_
+#define _chameleon_zlapack_h_
+
+#include "coreblas/cblas_wrapper.h"
+
+BEGIN_C_DECLS
+
+/**
+ *  Declarations of math functions (LAPACK layout, Cblas/Lapacke interface) - alphabetical order
+ */
+void CHAMELEON_cblas_zgemm( const CBLAS_ORDER Order, const CBLAS_TRANSPOSE TransA, const CBLAS_TRANSPOSE TransB,
+                            const int M, const int N, const int K,
+                            const void *alpha, const CHAMELEON_Complex64_t *A, const int lda,
+                                               const CHAMELEON_Complex64_t *B, const int ldb,
+                            const void *beta,        CHAMELEON_Complex64_t *C, const int ldc );
+
+END_C_DECLS
+
+#endif /* _chameleon_zlapack_h_ */

testing/chameleon_ztesting.c

@@ -48,7 +48,7 @@ parameter_t parameters[] = {
     { "async",    "Switch to the Async interface",                        's', PARAM_OPTION, 0, 0, TestValInt, {0}, NULL, pread_int, sprint_int },
     { "splitsub", "Split the task submission and execution stages",       'S', PARAM_OPTION, 0, 0, TestValInt, {0}, NULL, pread_int, sprint_int },
     { "generic",  "Switch to the non optimized generic algorithms",       -35, PARAM_OPTION, 0, 0, TestValInt, {0}, NULL, pread_int, sprint_int },
-    { "api",      "Select the API to test (0: Descriptors, 1: Standard)", -36, PARAM_OPTION, 1, 3, TestValInt, {0}, NULL, pread_int, sprint_int },
+    { "api",      "Select the API to test (0: Descriptors, 1: Standard, 2: Lapack)", -36, PARAM_OPTION, 1, 3, TestValInt, {0}, NULL, pread_int, sprint_int },
 #endif
 
     { NULL, "Machine parameters", 0, PARAM_OPTION, 0, 0, 0, {0}, NULL, NULL, NULL },