Added convgemm to the CLBlast database, added initial parameters for Skylake GPU

CMakeLists.txt

@@ -212,11 +212,10 @@ endif()
 
 # Sets the supported routines and the used kernels. New routines and kernels should be added here.
 set(KERNELS copy_fast copy_pad transpose_fast transpose_pad xaxpy xdot xger
-            xgemm xgemm_direct xgemv invert)
-set(KERNELS_EXTRA xconvgemm)  # kernels for which not to include a tuner in 'all tuners' target
+            xgemm xgemm_direct xgemv invert xconvgemm)
 set(DATABASES copy pad padtranspose transpose xaxpy xdot
               xgemm xgemm_direct xgemv xgemv_fast xgemv_fast_rot xger invert
-              gemm_routine trsv_routine)
+              gemm_routine trsv_routine xconvgemm)
 set(ROUTINE_TUNERS xgemm xtrsv)
 set(LEVEL1_ROUTINES xswap xscal xcopy xaxpy xdot xdotu xdotc xnrm2 xasum xamax)
 set(LEVEL2_ROUTINES xgemv xgbmv xhemv xhbmv xhpmv xsymv xsbmv xspmv xtrmv xtbmv xtpmv xtrsv
@@ -435,7 +434,7 @@ if(TUNERS)
   endif()
 
   # Adds tuning executables
-  set(ALLKERNELS ${KERNELS} ${KERNELS_EXTRA})
+  set(ALLKERNELS ${KERNELS})
   foreach(KERNEL ${ALLKERNELS})
     add_executable(clblast_tuner_${KERNEL} ${TUNERS_COMMON} src/tuning/kernels/${KERNEL}.cpp)
     target_link_libraries(clblast_tuner_${KERNEL} ${API_LIBRARIES})

src/database/apple_cpu_fallback.hpp

@@ -49,6 +49,9 @@ const DatabaseEntry XgemmApple = {
 const DatabaseEntry XgemmDirectApple = {
   "XgemmDirect", Precision::kAny, {"KWID", "MDIMAD", "MDIMCD", "NDIMBD", "NDIMCD", "PADA", "PADB", "VWMD", "VWND", "WGD"}, { {  kDeviceTypeAll, "default", { { "default", { { kDeviceNameDefault, Params{ 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0 } } } } } } }
 };
+const DatabaseEntry XconvgemmApple = {
+    "Xconvgemm", Precision::kAny, {"KWID", "MDIMAD", "MDIMCD", "NDIMBD", "NDIMCD", "PADA", "PADB", "VWMD", "VWND", "WGD"}, { {  kDeviceTypeAll, "default", { { "default", { { kDeviceNameDefault, Params{ 1, 1, 1, 1, 1, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0 } } } } } } }
+};
 const DatabaseEntry CopyApple = {
   "Copy", Precision::kAny, {"COPY_DIMX", "COPY_DIMY", "COPY_VW", "COPY_WPT"}, { {  kDeviceTypeAll, "default", { { "default", { { kDeviceNameDefault, Params{ 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 } } } } } } }
 };

src/database/database.cpp

@@ -25,6 +25,7 @@
 #include "database/kernels/xger/xger.hpp"
 #include "database/kernels/xgemm/xgemm.hpp"
 #include "database/kernels/xgemm_direct/xgemm_direct.hpp"
+#include "database/kernels/xconvgemm/xconvgemm.hpp"
 #include "database/kernels/copy/copy.hpp"
 #include "database/kernels/pad/pad.hpp"
 #include "database/kernels/transpose/transpose.hpp"
@@ -43,7 +44,7 @@ std::vector<database::DatabaseEntry> Database::database = std::vector<database::
 const std::vector<database::DatabaseEntry> Database::apple_cpu_fallback = std::vector<database::DatabaseEntry>{
   database::XaxpyApple, database::XdotApple,
   database::XgemvApple, database::XgemvFastApple, database::XgemvFastRotApple, database::XgerApple, database::XtrsvApple,
-  database::XgemmApple, database::XgemmDirectApple,
+  database::XgemmApple, database::XgemmDirectApple, database::XconvgemmApple,
   database::CopyApple, database::PadApple, database::TransposeApple, database::PadtransposeApple,
   database::InvertApple,
   database::TrsvRoutineApple
@@ -71,6 +72,7 @@ Database::Database(const Device &device, const std::string &kernel_name,
         database::XgerHalf, database::XgerSingle, database::XgerDouble, database::XgerComplexSingle, database::XgerComplexDouble,
         database::XgemmHalf, database::XgemmSingle, database::XgemmDouble, database::XgemmComplexSingle, database::XgemmComplexDouble,
         database::XgemmDirectHalf, database::XgemmDirectSingle, database::XgemmDirectDouble, database::XgemmDirectComplexSingle, database::XgemmDirectComplexDouble,
+        database::XconvgemmHalf, database::XconvgemmSingle, database::XconvgemmDouble, database::XconvgemmComplexSingle, database::XconvgemmComplexDouble,
         database::CopyHalf, database::CopySingle, database::CopyDouble, database::CopyComplexSingle, database::CopyComplexDouble,
         database::PadHalf, database::PadSingle, database::PadDouble, database::PadComplexSingle, database::PadComplexDouble,
         database::TransposeHalf, database::TransposeSingle, database::TransposeDouble, database::TransposeComplexSingle, database::TransposeComplexDouble,

src/database/kernels/xconvgemm/xconvgemm.cpp

@@ -0,0 +1,15 @@
+
+// =================================================================================================
+// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. It
+// is auto-generated by the 'scripts/database/database.py' Python script.
+//
+// This file populates the database with best-found tuning parameters for the 'Xconvgemm' kernels.
+//
+// =================================================================================================
+
+#include "database/kernels/xconvgemm/xconvgemm.hpp"
+#include "database/kernels/xconvgemm/xconvgemm_16.hpp"
+#include "database/kernels/xconvgemm/xconvgemm_32.hpp"
+#include "database/kernels/xconvgemm/xconvgemm_3232.hpp"
+#include "database/kernels/xconvgemm/xconvgemm_64.hpp"
+#include "database/kernels/xconvgemm/xconvgemm_6464.hpp"

src/database/kernels/xconvgemm/xconvgemm.hpp

@@ -0,0 +1,22 @@
+
+// =================================================================================================
+// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. It
+// is auto-generated by the 'scripts/database/database.py' Python script.
+//
+// This file populates the database with best-found tuning parameters for the 'Xconvgemm' kernels.
+//
+// =================================================================================================
+
+#include "database/database_structure.hpp"
+
+namespace clblast {
+namespace database {
+
+extern const DatabaseEntry XconvgemmHalf;
+extern const DatabaseEntry XconvgemmSingle;
+extern const DatabaseEntry XconvgemmComplexSingle;
+extern const DatabaseEntry XconvgemmDouble;
+extern const DatabaseEntry XconvgemmComplexDouble;
+
+} // namespace database
+} // namespace clblast

src/database/kernels/xconvgemm/xconvgemm_16.hpp

@@ -0,0 +1,26 @@
+
+// =================================================================================================
+// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. It
+// is auto-generated by the 'scripts/database/database.py' Python script.
+//
+// This file populates the database with best-found tuning parameters for the 'Xconvgemm16' kernels.
+//
+// =================================================================================================
+
+namespace clblast {
+namespace database {
+
+const DatabaseEntry XconvgemmHalf = {
+  "Xconvgemm", Precision::kHalf, {"KWID", "MDIMAD", "MDIMCD", "NDIMBD", "NDIMCD", "PADA", "PADB", "VWMD", "VWND", "WGD"}, {
+    { // Default
+      kDeviceTypeAll, "default", {
+        { "default", {
+          { kDeviceNameDefault                                        , Params{ 1, 16, 8, 8, 16, 0, 0, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
+        } },
+      }
+    },
+  }
+};
+
+} // namespace database
+} // namespace clblast

src/database/kernels/xconvgemm/xconvgemm_32.hpp

@@ -0,0 +1,34 @@
+
+// =================================================================================================
+// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. It
+// is auto-generated by the 'scripts/database/database.py' Python script.
+//
+// This file populates the database with best-found tuning parameters for the 'Xconvgemm32' kernels.
+//
+// =================================================================================================
+
+namespace clblast {
+namespace database {
+
+const DatabaseEntry XconvgemmSingle = {
+  "Xconvgemm", Precision::kSingle, {"KWID", "MDIMAD", "MDIMCD", "NDIMBD", "NDIMCD", "PADA", "PADB", "VWMD", "VWND", "WGD"}, {
+    { // Intel GPUs
+      kDeviceTypeGPU, "Intel", {
+        { "default", {
+          { Name{"Intel(R) HD Graphics Skylake ULT GT2              "}, Params{ 1, 16, 8, 8, 16, 0, 0, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
+          { kDeviceNameDefault                                        , Params{ 1, 16, 8, 8, 16, 0, 0, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
+        } },
+      }
+    },
+    { // Default
+      kDeviceTypeAll, "default", {
+        { "default", {
+          { kDeviceNameDefault                                        , Params{ 1, 16, 8, 8, 16, 0, 0, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
+        } },
+      }
+    },
+  }
+};
+
+} // namespace database
+} // namespace clblast

src/database/kernels/xconvgemm/xconvgemm_3232.hpp

@@ -0,0 +1,26 @@
+
+// =================================================================================================
+// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. It
+// is auto-generated by the 'scripts/database/database.py' Python script.
+//
+// This file populates the database with best-found tuning parameters for the 'Xconvgemm3232' kernels.
+//
+// =================================================================================================
+
+namespace clblast {
+namespace database {
+
+const DatabaseEntry XconvgemmComplexSingle = {
+  "Xconvgemm", Precision::kComplexSingle, {"KWID", "MDIMAD", "MDIMCD", "NDIMBD", "NDIMCD", "PADA", "PADB", "VWMD", "VWND", "WGD"}, {
+    { // Default
+      kDeviceTypeAll, "default", {
+        { "default", {
+          { kDeviceNameDefault                                        , Params{ 1, 16, 8, 8, 16, 0, 0, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
+        } },
+      }
+    },
+  }
+};
+
+} // namespace database
+} // namespace clblast

src/database/kernels/xconvgemm/xconvgemm_64.hpp

@@ -0,0 +1,26 @@
+
+// =================================================================================================
+// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. It
+// is auto-generated by the 'scripts/database/database.py' Python script.
+//
+// This file populates the database with best-found tuning parameters for the 'Xconvgemm64' kernels.
+//
+// =================================================================================================
+
+namespace clblast {
+namespace database {
+
+const DatabaseEntry XconvgemmDouble = {
+  "Xconvgemm", Precision::kDouble, {"KWID", "MDIMAD", "MDIMCD", "NDIMBD", "NDIMCD", "PADA", "PADB", "VWMD", "VWND", "WGD"}, {
+    { // Default
+      kDeviceTypeAll, "default", {
+        { "default", {
+          { kDeviceNameDefault                                        , Params{ 1, 16, 8, 8, 16, 0, 0, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
+        } },
+      }
+    },
+  }
+};
+
+} // namespace database
+} // namespace clblast

src/database/kernels/xconvgemm/xconvgemm_6464.hpp

@@ -0,0 +1,26 @@
+
+// =================================================================================================
+// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. It
+// is auto-generated by the 'scripts/database/database.py' Python script.
+//
+// This file populates the database with best-found tuning parameters for the 'Xconvgemm6464' kernels.
+//
+// =================================================================================================
+
+namespace clblast {
+namespace database {
+
+const DatabaseEntry XconvgemmComplexDouble = {
+  "Xconvgemm", Precision::kComplexDouble, {"KWID", "MDIMAD", "MDIMCD", "NDIMBD", "NDIMCD", "PADA", "PADB", "VWMD", "VWND", "WGD"}, {
+    { // Default
+      kDeviceTypeAll, "default", {
+        { "default", {
+          { kDeviceNameDefault                                        , Params{ 1, 16, 8, 8, 16, 0, 0, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
+        } },
+      }
+    },
+  }
+};
+
+} // namespace database
+} // namespace clblast

src/routines/levelx/xconvgemm.cpp

@@ -25,7 +25,7 @@ namespace clblast {
 template <typename T>
 Xconvgemm<T>::Xconvgemm(Queue &queue, EventPointer event, const std::string &name,
                         const ConvGemmMethod method):
-    Routine(queue, event, name, {"XgemmDirect"},
+    Routine(queue, event, name, {"Xconvgemm"},
         PrecisionValue<T>(), {}, {
             (method == ConvGemmMethod::kWithIm2Col) ? "#define CONVGEMM_WITH_IM2COL\n" : "",
             #include "../../kernels/level3/level3.opencl"