Merge pull request #345 from CNugteren/convolution-fixes-and-tuner

Convolution with single kernel

1 files changed, 35 insertions, 0 deletions

diff --git a/src/database/kernels/xconvgemm/xconvgemm_32.hpp b/src/database/kernels/xconvgemm/xconvgemm_32.hpp
new file mode 100644
index 00000000..216a9f42
--- /dev/null
+++ b/src/database/kernels/xconvgemm/xconvgemm_32.hpp
@@ -0,0 +1,35 @@
+
+// =================================================================================================
+// 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) Gen9 HD Graphics NEO                     "}, Params{ 1, 16, 32, 8, 8, 0, 0, 1, 4, 32, 0, 0, 0, 0, 0, 0 } },
+          { 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, 8, 16, 16, 8, 0, 0, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
+        } },
+      }
+    },
+    { // Default
+      kDeviceTypeAll, "default", {
+        { "default", {
+          { kDeviceNameDefault                                        , Params{ 1, 8, 16, 16, 8, 0, 0, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
+        } },
+      }
+    },
+  }
+};
+
+} // namespace database
+} // namespace clblast