Merge branch 'xgemm_tuner_exhaustive' into development

6 files changed, 71 insertions, 36 deletions

diff --git a/src/database/kernels/copy.hpp b/src/database/kernels/copy.hpp
index dc2011fd..a6b7dfe8 100644
--- a/src/database/kernels/copy.hpp
+++ b/src/database/kernels/copy.hpp
@@ -87,7 +87,7 @@ const Database::DatabaseEntry Database::CopySingle = {
         { "GeForce GTX 750 Ti",                              { {"COPY_DIMX",32}, {"COPY_DIMY",8}, {"COPY_VW",2}, {"COPY_WPT",1} } },
         { "GeForce GTX 980",                                 { {"COPY_DIMX",32}, {"COPY_DIMY",16}, {"COPY_VW",1}, {"COPY_WPT",1} } },
         { "GeForce GTX TITAN",                               { {"COPY_DIMX",32}, {"COPY_DIMY",8}, {"COPY_VW",2}, {"COPY_WPT",4} } },
-        { "GeForce GTX TITAN X",                             { {"COPY_DIMX",32}, {"COPY_DIMY",8}, {"COPY_VW",2}, {"COPY_WPT",1} } },
+        { "GeForce GTX TITAN X",                             { {"COPY_DIMX",32}, {"COPY_DIMY",8}, {"COPY_VW",1}, {"COPY_WPT",2} } },
         { "Tesla K20m",                                      { {"COPY_DIMX",8}, {"COPY_DIMY",8}, {"COPY_VW",4}, {"COPY_WPT",4} } },
         { "Tesla K40m",                                      { {"COPY_DIMX",8}, {"COPY_DIMY",8}, {"COPY_VW",4}, {"COPY_WPT",2} } },
         { "default",                                         { {"COPY_DIMX",8}, {"COPY_DIMY",16}, {"COPY_VW",4}, {"COPY_WPT",1} } },
diff --git a/src/database/kernels/xaxpy.hpp b/src/database/kernels/xaxpy.hpp
index 60fa7555..6e84ca5a 100644
--- a/src/database/kernels/xaxpy.hpp
+++ b/src/database/kernels/xaxpy.hpp
@@ -79,10 +79,10 @@ const Database::DatabaseEntry Database::XaxpySingle = {
     { // NVIDIA GPUs
       kDeviceTypeGPU, "NVIDIA", {
         { "GRID K520",                                       { {"VW",2}, {"WGS",64}, {"WPT",1} } },
-        { "GeForce GTX 1070",                                { {"VW",2}, {"WGS",64}, {"WPT",1} } },
-        { "GeForce GTX 480",                                 { {"VW",4}, {"WGS",64}, {"WPT",1} } },
+        { "GeForce GTX 1070",                                { {"VW",1}, {"WGS",64}, {"WPT",4} } },
+        { "GeForce GTX 480",                                 { {"VW",2}, {"WGS",128}, {"WPT",1} } },
         { "GeForce GTX 670",                                 { {"VW",2}, {"WGS",64}, {"WPT",1} } },
-        { "GeForce GTX 680",                                 { {"VW",2}, {"WGS",64}, {"WPT",1} } },
+        { "GeForce GTX 680",                                 { {"VW",1}, {"WGS",128}, {"WPT",1} } },
         { "GeForce GTX 750",                                 { {"VW",1}, {"WGS",64}, {"WPT",1} } },
         { "GeForce GTX 750 Ti",                              { {"VW",1}, {"WGS",1024}, {"WPT",1} } },
         { "GeForce GTX 980",                                 { {"VW",1}, {"WGS",1024}, {"WPT",1} } },
@@ -209,7 +209,7 @@ const Database::DatabaseEntry Database::XaxpyDouble = {
       kDeviceTypeGPU, "NVIDIA", {
         { "GRID K520",                                       { {"VW",1}, {"WGS",64}, {"WPT",1} } },
         { "GeForce GTX 1070",                                { {"VW",1}, {"WGS",64}, {"WPT",8} } },
-        { "GeForce GTX 480",                                 { {"VW",2}, {"WGS",64}, {"WPT",1} } },
+        { "GeForce GTX 480",                                 { {"VW",1}, {"WGS",128}, {"WPT",1} } },
         { "GeForce GTX 670",                                 { {"VW",1}, {"WGS",64}, {"WPT",1} } },
         { "GeForce GTX 680",                                 { {"VW",1}, {"WGS",64}, {"WPT",1} } },
         { "GeForce GTX 750",                                 { {"VW",1}, {"WGS",128}, {"WPT",1} } },
diff --git a/src/database/kernels/xgemm.hpp b/src/database/kernels/xgemm.hpp
index c960592d..cc81cf6a 100644
--- a/src/database/kernels/xgemm.hpp
+++ b/src/database/kernels/xgemm.hpp
@@ -158,7 +158,7 @@ const Database::DatabaseEntry Database::XgemmDouble = {
   "Xgemm", Precision::kDouble, {
     { // AMD GPUs
       kDeviceTypeGPU, "AMD", {
-        { "AMD Radeon R9 M370X Compute Engine",              { {"KWG",32}, {"KWI",2}, {"MDIMA",32}, {"MDIMC",32}, {"MWG",64}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",64}, {"SA",0}, {"SB",0}, {"STRM",1}, {"STRN",1}, {"VWM",2}, {"VWN",8} } },
+        { "AMD Radeon R9 M370X Compute Engine",              { {"KWG",32}, {"KWI",2}, {"MDIMA",16}, {"MDIMC",16}, {"MWG",64}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",32}, {"SA",0}, {"SB",0}, {"STRM",0}, {"STRN",0}, {"VWM",4}, {"VWN",4} } },
         { "Hawaii",                                          { {"KWG",16}, {"KWI",8}, {"MDIMA",32}, {"MDIMC",8}, {"MWG",128}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",32}, {"SA",0}, {"SB",1}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",4} } },
         { "Oland",                                           { {"KWG",16}, {"KWI",2}, {"MDIMA",8}, {"MDIMC",16}, {"MWG",64}, {"NDIMB",16}, {"NDIMC",8}, {"NWG",16}, {"SA",0}, {"SB",0}, {"STRM",1}, {"STRN",1}, {"VWM",1}, {"VWN",1} } },
         { "Pitcairn",                                        { {"KWG",32}, {"KWI",2}, {"MDIMA",32}, {"MDIMC",16}, {"MWG",64}, {"NDIMB",8}, {"NDIMC",16}, {"NWG",32}, {"SA",0}, {"SB",0}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",2} } },
diff --git a/src/database/kernels/xgemv.hpp b/src/database/kernels/xgemv.hpp
index 7e8e64e3..03e84525 100644
--- a/src/database/kernels/xgemv.hpp
+++ b/src/database/kernels/xgemv.hpp
@@ -88,7 +88,7 @@ const Database::DatabaseEntry Database::XgemvSingle = {
     },
     { // Default
       kDeviceTypeAll, "default", {
-        { "default",                                         { {"WGS1",64}, {"WPT1",4} } },
+        { "default",                                         { {"WGS1",64}, {"WPT1",1} } },
       }
     },
   }
diff --git a/src/database/kernels/xgemv_fast.hpp b/src/database/kernels/xgemv_fast.hpp
index f5e3e630..c12fcdca 100644
--- a/src/database/kernels/xgemv_fast.hpp
+++ b/src/database/kernels/xgemv_fast.hpp
@@ -60,7 +60,7 @@ const Database::DatabaseEntry Database::XgemvFastSingle = {
         { "Intel(R) HD Graphics Skylake ULT GT2",            { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
         { "Iris",                                            { {"VW2",1}, {"WGS2",128}, {"WPT2",2} } },
         { "Iris Pro",                                        { {"VW2",1}, {"WGS2",128}, {"WPT2",2} } },
-        { "default",                                         { {"VW2",1}, {"WGS2",32}, {"WPT2",1} } },
+        { "default",                                         { {"VW2",2}, {"WGS2",64}, {"WPT2",2} } },
       }
     },
     { // Intel accelerators
@@ -88,7 +88,7 @@ const Database::DatabaseEntry Database::XgemvFastSingle = {
     },
     { // Default
       kDeviceTypeAll, "default", {
-        { "default",                                         { {"VW2",4}, {"WGS2",128}, {"WPT2",4} } },
+        { "default",                                         { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
       }
     },
   }
@@ -123,7 +123,7 @@ const Database::DatabaseEntry Database::XgemvFastComplexSingle = {
         { "Intel(R) HD Graphics Skylake ULT GT2",            { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
         { "Iris",                                            { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
         { "Iris Pro",                                        { {"VW2",4}, {"WGS2",128}, {"WPT2",4} } },
-        { "default",                                         { {"VW2",1}, {"WGS2",32}, {"WPT2",1} } },
+        { "default",                                         { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
       }
     },
     { // Intel accelerators
@@ -145,7 +145,7 @@ const Database::DatabaseEntry Database::XgemvFastComplexSingle = {
     },
     { // Default
       kDeviceTypeAll, "default", {
-        { "default",                                         { {"VW2",1}, {"WGS2",32}, {"WPT2",1} } },
+        { "default",                                         { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
       }
     },
   }
diff --git a/src/tuning/kernels/xgemm.cpp b/src/tuning/kernels/xgemm.cpp
index eb7c8a66..7c9ac76a 100644
--- a/src/tuning/kernels/xgemm.cpp
+++ b/src/tuning/kernels/xgemm.cpp
@@ -7,7 +7,9 @@
 // Author(s):
 //   Cedric Nugteren <www.cedricnugteren.nl>
 //
-// This file uses the CLTune auto-tuner to tune the xgemm OpenCL kernels.
+// This file uses the CLTune auto-tuner to tune the xgemm OpenCL kernels. There are two variations:
+// - V==1: This tests some limited set of tuning parameters exhaustively.
+// - V==2: This tests a much larger set of tuning parameters by randomly sampling a subset.
 //
 // =================================================================================================
 
@@ -21,12 +23,12 @@ namespace clblast {
 // =================================================================================================
 
 // See comment at top of file for a description of the class
-template <typename T>
+template <typename T, int V>
 class TuneXgemm {
  public:
 
   // The representative kernel and the source code
-  static std::string KernelFamily() { return "xgemm"; }
+  static std::string KernelFamily() { return (V==1) ? "xgemm_1" : "xgemm_2"; }
   static std::string KernelName() { return "Xgemm"; }
   static std::string GetSources() {
     return
@@ -48,7 +50,7 @@ class TuneXgemm {
   static size_t DefaultM() { return 1024; }
   static size_t DefaultN() { return 1024; }
   static size_t DefaultK() { return 1024; }
-  static double DefaultFraction() { return 256.0; }
+  static double DefaultFraction() { return (V==1) ? 1.0 : 512.0; } // test all or sample randomly
 
   // Describes how to obtain the sizes of the buffers
   static size_t GetSizeX(const Arguments<T> &) { return 1; } // N/A for this kernel
@@ -60,20 +62,38 @@ class TuneXgemm {
 
   // Sets the tuning parameters and their possible values
   static void SetParameters(cltune::Tuner &tuner, const size_t id) {
-    tuner.AddParameter(id, "MWG", {16, 32, 64, 128});
-    tuner.AddParameter(id, "NWG", {16, 32, 64, 128});
-    tuner.AddParameter(id, "KWG", {16, 32});
-    tuner.AddParameter(id, "MDIMC", {8, 16, 32});
-    tuner.AddParameter(id, "NDIMC", {8, 16, 32});
-    tuner.AddParameter(id, "MDIMA", {8, 16, 32});
-    tuner.AddParameter(id, "NDIMB", {8, 16, 32});
-    tuner.AddParameter(id, "KWI", {2});
-    tuner.AddParameter(id, "VWM", {1, 2, 4});
-    tuner.AddParameter(id, "VWN", {1, 2, 4});
-    tuner.AddParameter(id, "STRM", {0, 1});
-    tuner.AddParameter(id, "STRN", {0, 1});
-    tuner.AddParameter(id, "SA", {0, 1});
-    tuner.AddParameter(id, "SB", {0, 1});
+    if (V==1) { // limited subset of tuning parameters - but explorable exhaustively
+      tuner.AddParameter(id, "MWG", {16, 32, 64});
+      tuner.AddParameter(id, "NWG", {16, 32, 64});
+      tuner.AddParameter(id, "KWG", {32});
+      tuner.AddParameter(id, "MDIMC", {8, 16, 32});
+      tuner.AddParameter(id, "NDIMC", {8, 16, 32});
+      tuner.AddParameter(id, "MDIMA", {8, 16, 32});
+      tuner.AddParameter(id, "NDIMB", {8, 16, 32});
+      tuner.AddParameter(id, "KWI", {2});
+      tuner.AddParameter(id, "VWM", {1, 2, 4});
+      tuner.AddParameter(id, "VWN", {1, 2, 4});
+      tuner.AddParameter(id, "STRM", {0});
+      tuner.AddParameter(id, "STRN", {0});
+      tuner.AddParameter(id, "SA", {0, 1});
+      tuner.AddParameter(id, "SB", {0, 1});
+    } // a lot more tuning parameters - has to be sampled randomly, too much to test all
+    else {
+      tuner.AddParameter(id, "MWG", {16, 32, 64, 128});
+      tuner.AddParameter(id, "NWG", {16, 32, 64, 128});
+      tuner.AddParameter(id, "KWG", {16, 32});
+      tuner.AddParameter(id, "MDIMC", {8, 16, 32});
+      tuner.AddParameter(id, "NDIMC", {8, 16, 32});
+      tuner.AddParameter(id, "MDIMA", {8, 16, 32});
+      tuner.AddParameter(id, "NDIMB", {8, 16, 32});
+      tuner.AddParameter(id, "KWI", {2});
+      tuner.AddParameter(id, "VWM", {1, 2, 4, 8});
+      tuner.AddParameter(id, "VWN", {1, 2, 4, 8});
+      tuner.AddParameter(id, "STRM", {0, 1});
+      tuner.AddParameter(id, "STRN", {0, 1});
+      tuner.AddParameter(id, "SA", {0, 1});
+      tuner.AddParameter(id, "SB", {0, 1});
+    }
   }
 
   // Sets the constraints
@@ -92,6 +112,14 @@ class TuneXgemm {
     // KWG has to be a multiple of KDIMA = ((MDIMC*NDIMC)/(MDIMA)) and KDIMB = (...)
     tuner.AddConstraint(id, MultipleOfXMulYDivZ, {"KWG", "MDIMC", "NDIMC", "MDIMA"});
     tuner.AddConstraint(id, MultipleOfXMulYDivZ, {"KWG", "MDIMC", "NDIMC", "NDIMB"});
+
+    // Extra constraints for variation 1 to limit the set of options significantly
+    if (V==1) {
+      auto IsEqual = [] (std::vector<size_t> v) { return v[0] == v[1]; };
+      tuner.AddConstraint(id, IsEqual, {"MDIMC", "MDIMA"});
+      tuner.AddConstraint(id, IsEqual, {"NDIMC", "NDIMB"});
+      tuner.AddConstraint(id, IsEqual, {"SA", "SB"});
+    }
   }
 
   // Sets the local memory size
@@ -145,15 +173,22 @@ class TuneXgemm {
 using float2 = clblast::float2;
 using double2 = clblast::double2;
 
-// Main function (not within the clblast namespace)
-int main(int argc, char *argv[]) {
+// Function to tune a specific variation V (not within the clblast namespace)
+template <int V>
+void StartVariation(int argc, char *argv[]) {
   switch(clblast::GetPrecision(argc, argv)) {
-    case clblast::Precision::kHalf: clblast::Tuner<clblast::TuneXgemm<half>, half>(argc, argv); break;
-    case clblast::Precision::kSingle: clblast::Tuner<clblast::TuneXgemm<float>, float>(argc, argv); break;
-    case clblast::Precision::kDouble: clblast::Tuner<clblast::TuneXgemm<double>, double>(argc, argv); break;
-    case clblast::Precision::kComplexSingle: clblast::Tuner<clblast::TuneXgemm<float2>, float2>(argc, argv); break;
-    case clblast::Precision::kComplexDouble: clblast::Tuner<clblast::TuneXgemm<double2>, double2>(argc, argv); break;
+    case clblast::Precision::kHalf: clblast::Tuner<clblast::TuneXgemm<half,V>, half>(argc, argv); break;
+    case clblast::Precision::kSingle: clblast::Tuner<clblast::TuneXgemm<float,V>, float>(argc, argv); break;
+    case clblast::Precision::kDouble: clblast::Tuner<clblast::TuneXgemm<double,V>, double>(argc, argv); break;
+    case clblast::Precision::kComplexSingle: clblast::Tuner<clblast::TuneXgemm<float2,V>, float2>(argc, argv); break;
+    case clblast::Precision::kComplexDouble: clblast::Tuner<clblast::TuneXgemm<double2,V>, double2>(argc, argv); break;
   }
+}
+
+// Main function (not within the clblast namespace)
+int main(int argc, char *argv[]) {
+  StartVariation<1>(argc, argv);
+  StartVariation<2>(argc, argv);
   return 0;
 }