summaryrefslogtreecommitdiff
path: root/src/database
diff options
context:
space:
mode:
Diffstat (limited to 'src/database')
-rw-r--r--src/database/database.cpp61
-rw-r--r--src/database/database.hpp16
-rw-r--r--src/database/kernels/xgemm.hpp12
-rw-r--r--src/database/kernels/xgemv.hpp200
-rw-r--r--src/database/kernels/xgemv_fast.hpp247
-rw-r--r--src/database/kernels/xgemv_fast_rot.hpp138
6 files changed, 533 insertions, 141 deletions
diff --git a/src/database/database.cpp b/src/database/database.cpp
index 6ec93731..38974b95 100644
--- a/src/database/database.cpp
+++ b/src/database/database.cpp
@@ -17,6 +17,8 @@
#include "database/kernels/xaxpy.hpp"
#include "database/kernels/xdot.hpp"
#include "database/kernels/xgemv.hpp"
+#include "database/kernels/xgemv_fast.hpp"
+#include "database/kernels/xgemv_fast_rot.hpp"
#include "database/kernels/xger.hpp"
#include "database/kernels/xgemm.hpp"
#include "database/kernels/copy.hpp"
@@ -32,8 +34,10 @@ const std::vector<Database::DatabaseEntry> Database::database = {
XaxpyHalf, XaxpySingle, XaxpyDouble, XaxpyComplexSingle, XaxpyComplexDouble,
XdotHalf, XdotSingle, XdotDouble, XdotComplexSingle, XdotComplexDouble,
XgemvHalf, XgemvSingle, XgemvDouble, XgemvComplexSingle, XgemvComplexDouble,
+ XgemvFastHalf, XgemvFastSingle, XgemvFastDouble, XgemvFastComplexSingle, XgemvFastComplexDouble,
+ /* XgemvFastRotHalf, */ XgemvFastRotSingle, XgemvFastRotDouble, XgemvFastRotComplexSingle, XgemvFastRotComplexDouble,
XgerHalf, XgerSingle, XgerDouble, XgerComplexSingle, XgerComplexDouble,
- XgemmHalf, XgemmSingle, XgemmDouble, XgemmComplexSingle, XgemmComplexDouble,
+ /* XgemmHalf, */ XgemmSingle, XgemmDouble, XgemmComplexSingle, XgemmComplexDouble,
CopyHalf, CopySingle, CopyDouble, CopyComplexSingle, CopyComplexDouble,
PadHalf, PadSingle, PadDouble, PadComplexSingle, PadComplexDouble,
TransposeHalf, TransposeSingle, TransposeDouble, TransposeComplexSingle, TransposeComplexDouble,
@@ -42,9 +46,10 @@ const std::vector<Database::DatabaseEntry> Database::database = {
// =================================================================================================
-// Constructor, computing device properties and populating the parameter-vector from the database
+// Constructor, computing device properties and populating the parameter-vector from the database.
+// This takes an optional overlay database in case of custom tuning or custom kernels.
Database::Database(const Queue &queue, const std::vector<std::string> &kernels,
- const Precision precision):
+ const Precision precision, const std::vector<DatabaseEntry> &overlay):
parameters_{} {
// Finds information of the current device
@@ -53,10 +58,26 @@ Database::Database(const Queue &queue, const std::vector<std::string> &kernels,
auto device_vendor = device.Vendor();
auto device_name = device.Name();
+ // Set the short vendor name
+ for (auto &combination : kVendorNames) {
+ if (device_vendor == combination.first) {
+ device_vendor = combination.second;
+ }
+ }
+
// Iterates over all kernels to include, and retrieves the parameters for each of them
for (auto &kernel: kernels) {
- auto search_result = Search(kernel, device_type, device_vendor, device_name, precision);
- parameters_.insert(search_result.begin(), search_result.end());
+ auto search_result = ParametersPtr{};
+
+ for (auto db: { &overlay, &database }) {
+ search_result = Search(kernel, device_type, device_vendor, device_name, precision, *db);
+ if (search_result) {
+ parameters_.insert(search_result->begin(), search_result->end());
+ break;
+ }
+ }
+
+ if (!search_result) { throw std::runtime_error("Database error, could not find a suitable entry"); }
}
}
@@ -73,28 +94,22 @@ std::string Database::GetDefines() const {
// =================================================================================================
-// Searches the database for the right kernel and precision
-Database::Parameters Database::Search(const std::string &this_kernel,
- const std::string &this_type,
- const std::string &this_vendor,
- const std::string &this_device,
- const Precision this_precision) const {
- // Set the short vendor name
- auto this_short_vendor = this_vendor;
- for (auto &combination : kVendorNames) {
- if (this_vendor == combination.first) {
- this_short_vendor = combination.second;
- }
- }
+// Searches a particular database for the right kernel and precision
+Database::ParametersPtr Database::Search(const std::string &this_kernel,
+ const std::string &this_type,
+ const std::string &this_vendor,
+ const std::string &this_device,
+ const Precision this_precision,
+ const std::vector<DatabaseEntry> &this_database) const {
// Selects the right kernel
- for (auto &db: database) {
+ for (auto &db: this_database) {
if (db.kernel == this_kernel && db.precision == this_precision) {
// Searches for the right vendor and device type, or selects the default if unavailable. This
// assumes that the default vendor / device type is last in the database.
for (auto &vendor: db.vendors) {
- if ((vendor.name == this_short_vendor || vendor.name == kDeviceVendorAll) &&
+ if ((vendor.name == this_vendor || vendor.name == kDeviceVendorAll) &&
(vendor.type == this_type || vendor.type == kDeviceTypeAll)) {
// Searches for the right device. If the current device is unavailable, selects the vendor
@@ -104,7 +119,7 @@ Database::Parameters Database::Search(const std::string &this_kernel,
if (device.name == this_device || device.name == "default") {
// Sets the parameters accordingly
- return device.parameters;
+ return &device.parameters;
}
}
}
@@ -112,8 +127,8 @@ Database::Parameters Database::Search(const std::string &this_kernel,
}
}
- // If we reached this point, something is wrong
- throw std::runtime_error("Database error, could not find a suitable entry");
+ // If we reached this point, the entry was not found in this database
+ return nullptr;
}
// =================================================================================================
diff --git a/src/database/database.hpp b/src/database/database.hpp
index 0987cbed..8d6d3863 100644
--- a/src/database/database.hpp
+++ b/src/database/database.hpp
@@ -32,6 +32,7 @@ class Database {
// Type alias for the database parameters
using Parameters = std::unordered_map<std::string,size_t>;
+ using ParametersPtr = const Parameters*;
// Structures for content inside the database
struct DatabaseDevice {
@@ -70,17 +71,19 @@ class Database {
static const DatabaseEntry XaxpyHalf, XaxpySingle, XaxpyDouble, XaxpyComplexSingle, XaxpyComplexDouble;
static const DatabaseEntry XdotHalf, XdotSingle, XdotDouble, XdotComplexSingle, XdotComplexDouble;
static const DatabaseEntry XgemvHalf, XgemvSingle, XgemvDouble, XgemvComplexSingle, XgemvComplexDouble;
+ static const DatabaseEntry XgemvFastHalf, XgemvFastSingle, XgemvFastDouble, XgemvFastComplexSingle, XgemvFastComplexDouble;
+ static const DatabaseEntry /* XgemvFastRotHalf, */ XgemvFastRotSingle, XgemvFastRotDouble, XgemvFastRotComplexSingle, XgemvFastRotComplexDouble;
static const DatabaseEntry XgerHalf, XgerSingle, XgerDouble, XgerComplexSingle, XgerComplexDouble;
- static const DatabaseEntry XgemmHalf, XgemmSingle, XgemmDouble, XgemmComplexSingle, XgemmComplexDouble;
+ static const DatabaseEntry /* XgemmHalf, */ XgemmSingle, XgemmDouble, XgemmComplexSingle, XgemmComplexDouble;
static const DatabaseEntry CopyHalf, CopySingle, CopyDouble, CopyComplexSingle, CopyComplexDouble;
static const DatabaseEntry PadHalf, PadSingle, PadDouble, PadComplexSingle, PadComplexDouble;
static const DatabaseEntry TransposeHalf, TransposeSingle, TransposeDouble, TransposeComplexSingle, TransposeComplexDouble;
static const DatabaseEntry PadtransposeHalf, PadtransposeSingle, PadtransposeDouble, PadtransposeComplexSingle, PadtransposeComplexDouble;
static const std::vector<DatabaseEntry> database;
- // The constructor
+ // The constructor with a user-provided database overlay (potentially an empty vector)
explicit Database(const Queue &queue, const std::vector<std::string> &routines,
- const Precision precision);
+ const Precision precision, const std::vector<DatabaseEntry> &overlay);
// Accessor of values by key
size_t operator[](const std::string key) const { return parameters_.find(key)->second; }
@@ -89,9 +92,10 @@ class Database {
std::string GetDefines() const;
private:
- Parameters Search(const std::string &this_kernel, const std::string &this_type,
- const std::string &this_vendor, const std::string &this_device,
- const Precision this_precision) const;
+ // Search method for a specified database, returning pointer (possibly a nullptr)
+ ParametersPtr Search(const std::string &this_kernel, const std::string &this_type,
+ const std::string &this_vendor, const std::string &this_device,
+ const Precision this_precision, const std::vector<DatabaseEntry> &db) const;
// Found parameters suitable for this device/kernel
Parameters parameters_;
diff --git a/src/database/kernels/xgemm.hpp b/src/database/kernels/xgemm.hpp
index 736f2695..61b7ff05 100644
--- a/src/database/kernels/xgemm.hpp
+++ b/src/database/kernels/xgemm.hpp
@@ -14,18 +14,6 @@
namespace clblast {
// =================================================================================================
-const Database::DatabaseEntry Database::XgemmHalf = {
- "Xgemm", Precision::kHalf, {
- { // Default
- kDeviceTypeAll, "default", {
- { "default", { {"KWG",16}, {"KWI",2}, {"MDIMA",8}, {"MDIMC",8}, {"MWG",32}, {"NDIMB",8}, {"NDIMC",8}, {"NWG",64}, {"SA",0}, {"SB",0}, {"STRM",0}, {"STRN",0}, {"VWM",1}, {"VWN",1} } },
- }
- },
- }
-};
-
-// =================================================================================================
-
const Database::DatabaseEntry Database::XgemmSingle = {
"Xgemm", Precision::kSingle, {
{ // AMD GPUs
diff --git a/src/database/kernels/xgemv.hpp b/src/database/kernels/xgemv.hpp
index 65f4b5c8..6d680b06 100644
--- a/src/database/kernels/xgemv.hpp
+++ b/src/database/kernels/xgemv.hpp
@@ -18,13 +18,13 @@ const Database::DatabaseEntry Database::XgemvHalf = {
"Xgemv", Precision::kHalf, {
{ // Intel GPUs
kDeviceTypeGPU, "Intel", {
- { "Intel(R) HD Graphics Skylake ULT GT2", { {"WGS1",128}, {"WPT1",1}, {"VW2",2}, {"WGS2",128}, {"WPT2",2}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "default", { {"WGS1",128}, {"WPT1",1}, {"VW2",2}, {"WGS2",128}, {"WPT2",2}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "Intel(R) HD Graphics Skylake ULT GT2", { {"WGS1",128}, {"WPT1",1} } },
+ { "default", { {"WGS1",128}, {"WPT1",1} } },
}
},
{ // Default
kDeviceTypeAll, "default", {
- { "default", { {"WGS1",128}, {"WPT1",1}, {"VW2",2}, {"WGS2",128}, {"WPT2",2}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "default", { {"WGS1",128}, {"WPT1",1} } },
}
},
}
@@ -36,57 +36,57 @@ const Database::DatabaseEntry Database::XgemvSingle = {
"Xgemv", Precision::kSingle, {
{ // AMD GPUs
kDeviceTypeGPU, "AMD", {
- { "AMD Radeon R9 M370X Compute Engine", { {"WGS1",128}, {"WPT1",1}, {"VW2",1}, {"WGS2",128}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "Hawaii", { {"WGS1",128}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "Oland", { {"WGS1",128}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",4}, {"WGS3",256}, {"WPT3",4} } },
- { "Pitcairn", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "Tahiti", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "default", { {"WGS1",128}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "AMD Radeon R9 M370X Compute Engine", { {"WGS1",128}, {"WPT1",1} } },
+ { "Hawaii", { {"WGS1",128}, {"WPT1",1} } },
+ { "Oland", { {"WGS1",128}, {"WPT1",1} } },
+ { "Pitcairn", { {"WGS1",256}, {"WPT1",1} } },
+ { "Tahiti", { {"WGS1",256}, {"WPT1",1} } },
+ { "default", { {"WGS1",128}, {"WPT1",1} } },
}
},
{ // Intel CPUs
kDeviceTypeCPU, "Intel", {
- { "Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz", { {"WGS1",64}, {"WPT1",1}, {"VW2",4}, {"WGS2",128}, {"WPT2",4}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz", { {"WGS1",64}, {"WPT1",4}, {"VW2",1}, {"WGS2",64}, {"WPT2",4}, {"VW3",2}, {"WGS3",64}, {"WPT3",4} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",4}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz", { {"WGS1",64}, {"WPT1",1} } },
+ { "Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz", { {"WGS1",64}, {"WPT1",4} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // Intel GPUs
kDeviceTypeGPU, "Intel", {
- { "Intel(R) HD Graphics 530", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "Intel(R) HD Graphics Haswell Ultrabook GT2 Mobile", { {"WGS1",256}, {"WPT1",1}, {"VW2",4}, {"WGS2",128}, {"WPT2",4}, {"VW3",4}, {"WGS3",256}, {"WPT3",4} } },
- { "Intel(R) HD Graphics Skylake ULT GT2", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",4}, {"WGS3",64}, {"WPT3",4} } },
- { "Iris", { {"WGS1",64}, {"WPT1",2}, {"VW2",1}, {"WGS2",128}, {"WPT2",2}, {"VW3",4}, {"WGS3",64}, {"WPT3",8} } },
- { "Iris Pro", { {"WGS1",256}, {"WPT1",2}, {"VW2",1}, {"WGS2",128}, {"WPT2",2}, {"VW3",4}, {"WGS3",64}, {"WPT3",4} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",128}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "Intel(R) HD Graphics 530", { {"WGS1",256}, {"WPT1",1} } },
+ { "Intel(R) HD Graphics Haswell Ultrabook GT2 Mobile", { {"WGS1",256}, {"WPT1",1} } },
+ { "Intel(R) HD Graphics Skylake ULT GT2", { {"WGS1",64}, {"WPT1",1} } },
+ { "Iris", { {"WGS1",64}, {"WPT1",2} } },
+ { "Iris Pro", { {"WGS1",256}, {"WPT1",2} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // Intel accelerators
kDeviceTypeAccelerator, "Intel", {
- { "Intel(R) Many Integrated Core Acceleration Card", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "Intel(R) Many Integrated Core Acceleration Card", { {"WGS1",64}, {"WPT1",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // NVIDIA GPUs
kDeviceTypeGPU, "NVIDIA", {
- { "GRID K520", { {"WGS1",256}, {"WPT1",1}, {"VW2",2}, {"WGS2",256}, {"WPT2",2}, {"VW3",2}, {"WGS3",64}, {"WPT3",2} } },
- { "GeForce GTX 1070", { {"WGS1",128}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",2}, {"WGS3",64}, {"WPT3",2} } },
- { "GeForce GTX 480", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",128}, {"WPT2",1}, {"VW3",2}, {"WGS3",64}, {"WPT3",2} } },
- { "GeForce GTX 670", { {"WGS1",64}, {"WPT1",1}, {"VW2",2}, {"WGS2",256}, {"WPT2",2}, {"VW3",2}, {"WGS3",64}, {"WPT3",2} } },
- { "GeForce GTX 680", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",128}, {"WPT2",1}, {"VW3",2}, {"WGS3",128}, {"WPT3",2} } },
- { "GeForce GTX 750", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",2}, {"WGS3",128}, {"WPT3",2} } },
- { "GeForce GTX 750 Ti", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",4}, {"WGS3",128}, {"WPT3",4} } },
- { "GeForce GTX 980", { {"WGS1",128}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",1}, {"WGS3",128}, {"WPT3",1} } },
- { "GeForce GTX TITAN", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",1}, {"WGS3",256}, {"WPT3",1} } },
- { "GeForce GTX TITAN X", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",128}, {"WPT3",1} } },
- { "Tesla K20m", { {"WGS1",128}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",1}, {"WGS3",256}, {"WPT3",1} } },
- { "Tesla K40m", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",1}, {"WGS3",128}, {"WPT3",1} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "GRID K520", { {"WGS1",256}, {"WPT1",1} } },
+ { "GeForce GTX 1070", { {"WGS1",128}, {"WPT1",1} } },
+ { "GeForce GTX 480", { {"WGS1",64}, {"WPT1",1} } },
+ { "GeForce GTX 670", { {"WGS1",64}, {"WPT1",1} } },
+ { "GeForce GTX 680", { {"WGS1",256}, {"WPT1",1} } },
+ { "GeForce GTX 750", { {"WGS1",256}, {"WPT1",1} } },
+ { "GeForce GTX 750 Ti", { {"WGS1",256}, {"WPT1",1} } },
+ { "GeForce GTX 980", { {"WGS1",128}, {"WPT1",1} } },
+ { "GeForce GTX TITAN", { {"WGS1",256}, {"WPT1",1} } },
+ { "GeForce GTX TITAN X", { {"WGS1",256}, {"WPT1",1} } },
+ { "Tesla K20m", { {"WGS1",128}, {"WPT1",1} } },
+ { "Tesla K40m", { {"WGS1",256}, {"WPT1",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // Default
kDeviceTypeAll, "default", {
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
}
@@ -98,53 +98,53 @@ const Database::DatabaseEntry Database::XgemvComplexSingle = {
"Xgemv", Precision::kComplexSingle, {
{ // AMD GPUs
kDeviceTypeGPU, "AMD", {
- { "AMD Radeon R9 M370X Compute Engine", { {"WGS1",64}, {"WPT1",1}, {"VW2",2}, {"WGS2",256}, {"WPT2",2}, {"VW3",1}, {"WGS3",128}, {"WPT3",1} } },
- { "Hawaii", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "Oland", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",4}, {"WGS3",64}, {"WPT3",4} } },
- { "Pitcairn", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",2}, {"WGS3",64}, {"WPT3",2} } },
- { "Tahiti", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",128}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "AMD Radeon R9 M370X Compute Engine", { {"WGS1",64}, {"WPT1",1} } },
+ { "Hawaii", { {"WGS1",64}, {"WPT1",1} } },
+ { "Oland", { {"WGS1",64}, {"WPT1",1} } },
+ { "Pitcairn", { {"WGS1",64}, {"WPT1",1} } },
+ { "Tahiti", { {"WGS1",64}, {"WPT1",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // Intel CPUs
kDeviceTypeCPU, "Intel", {
- { "Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz", { {"WGS1",128}, {"WPT1",1}, {"VW2",1}, {"WGS2",128}, {"WPT2",2}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz", { {"WGS1",64}, {"WPT1",4}, {"VW2",4}, {"WGS2",64}, {"WPT2",4}, {"VW3",2}, {"WGS3",64}, {"WPT3",2} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",2}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz", { {"WGS1",128}, {"WPT1",1} } },
+ { "Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz", { {"WGS1",64}, {"WPT1",4} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // Intel GPUs
kDeviceTypeGPU, "Intel", {
- { "Intel(R) HD Graphics 530", { {"WGS1",64}, {"WPT1",1}, {"VW2",2}, {"WGS2",128}, {"WPT2",2}, {"VW3",4}, {"WGS3",64}, {"WPT3",4} } },
- { "Intel(R) HD Graphics Haswell Ultrabook GT2 Mobile", { {"WGS1",128}, {"WPT1",1}, {"VW2",2}, {"WGS2",128}, {"WPT2",2}, {"VW3",4}, {"WGS3",128}, {"WPT3",4} } },
- { "Intel(R) HD Graphics Skylake ULT GT2", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",4}, {"WGS3",64}, {"WPT3",4} } },
- { "Iris", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "Iris Pro", { {"WGS1",64}, {"WPT1",1}, {"VW2",4}, {"WGS2",128}, {"WPT2",4}, {"VW3",2}, {"WGS3",64}, {"WPT3",2} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "Intel(R) HD Graphics 530", { {"WGS1",64}, {"WPT1",1} } },
+ { "Intel(R) HD Graphics Haswell Ultrabook GT2 Mobile", { {"WGS1",128}, {"WPT1",1} } },
+ { "Intel(R) HD Graphics Skylake ULT GT2", { {"WGS1",64}, {"WPT1",1} } },
+ { "Iris", { {"WGS1",256}, {"WPT1",1} } },
+ { "Iris Pro", { {"WGS1",64}, {"WPT1",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // Intel accelerators
kDeviceTypeAccelerator, "Intel", {
- { "Intel(R) Many Integrated Core Acceleration Card", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "Intel(R) Many Integrated Core Acceleration Card", { {"WGS1",64}, {"WPT1",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // NVIDIA GPUs
kDeviceTypeGPU, "NVIDIA", {
- { "GRID K520", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "GeForce GTX 1070", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
- { "GeForce GTX 480", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "GeForce GTX 670", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "GeForce GTX 680", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "GRID K520", { {"WGS1",256}, {"WPT1",1} } },
+ { "GeForce GTX 1070", { {"WGS1",64}, {"WPT1",1} } },
+ { "GeForce GTX 480", { {"WGS1",64}, {"WPT1",1} } },
+ { "GeForce GTX 670", { {"WGS1",64}, {"WPT1",1} } },
+ { "GeForce GTX 680", { {"WGS1",64}, {"WPT1",1} } },
{ "GeForce GTX 750", { {"WGS1",128}, {"WPT1",1} } },
- { "GeForce GTX 750 Ti", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "GeForce GTX 750 Ti", { {"WGS1",64}, {"WPT1",1} } },
{ "GeForce GTX TITAN", { {"WGS1",256}, {"WPT1",1} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // Default
kDeviceTypeAll, "default", {
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
}
@@ -156,47 +156,47 @@ const Database::DatabaseEntry Database::XgemvDouble = {
"Xgemv", Precision::kDouble, {
{ // AMD GPUs
kDeviceTypeGPU, "AMD", {
- { "AMD Radeon R9 M370X Compute Engine", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",1}, {"WGS3",128}, {"WPT3",1} } },
- { "Hawaii", { {"WGS1",128}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "Oland", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",4}, {"WGS3",256}, {"WPT3",4} } },
- { "Pitcairn", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",2}, {"WGS3",64}, {"WPT3",2} } },
- { "Tahiti", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",2}, {"WGS3",64}, {"WPT3",2} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "AMD Radeon R9 M370X Compute Engine", { {"WGS1",64}, {"WPT1",1} } },
+ { "Hawaii", { {"WGS1",128}, {"WPT1",1} } },
+ { "Oland", { {"WGS1",256}, {"WPT1",1} } },
+ { "Pitcairn", { {"WGS1",256}, {"WPT1",1} } },
+ { "Tahiti", { {"WGS1",256}, {"WPT1",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // Intel CPUs
kDeviceTypeCPU, "Intel", {
- { "Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz", { {"WGS1",64}, {"WPT1",2}, {"VW2",4}, {"WGS2",128}, {"WPT2",4}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz", { {"WGS1",64}, {"WPT1",4}, {"VW2",1}, {"WGS2",64}, {"WPT2",4}, {"VW3",1}, {"WGS3",64}, {"WPT3",2} } },
- { "default", { {"WGS1",64}, {"WPT1",2}, {"VW2",1}, {"WGS2",64}, {"WPT2",4}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz", { {"WGS1",64}, {"WPT1",2} } },
+ { "Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz", { {"WGS1",64}, {"WPT1",4} } },
+ { "default", { {"WGS1",64}, {"WPT1",2} } },
}
},
{ // Intel accelerators
kDeviceTypeAccelerator, "Intel", {
- { "Intel(R) Many Integrated Core Acceleration Card", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "Intel(R) Many Integrated Core Acceleration Card", { {"WGS1",64}, {"WPT1",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // NVIDIA GPUs
kDeviceTypeGPU, "NVIDIA", {
- { "GRID K520", { {"WGS1",128}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "GeForce GTX 1070", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",1}, {"WGS3",128}, {"WPT3",1} } },
- { "GeForce GTX 480", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "GeForce GTX 670", { {"WGS1",128}, {"WPT1",1}, {"VW2",1}, {"WGS2",128}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "GeForce GTX 680", { {"WGS1",128}, {"WPT1",1}, {"VW2",1}, {"WGS2",128}, {"WPT2",1}, {"VW3",2}, {"WGS3",128}, {"WPT3",2} } },
- { "GeForce GTX 750", { {"WGS1",64}, {"WPT1",1}, {"VW2",2}, {"WGS2",256}, {"WPT2",2}, {"VW3",2}, {"WGS3",64}, {"WPT3",2} } },
- { "GeForce GTX 750 Ti", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",2}, {"WGS3",256}, {"WPT3",2} } },
- { "GeForce GTX 980", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "GeForce GTX TITAN", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",1}, {"WGS3",256}, {"WPT3",1} } },
- { "GeForce GTX TITAN X", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",128}, {"WPT2",1}, {"VW3",1}, {"WGS3",128}, {"WPT3",1} } },
- { "Tesla K20m", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",128}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "Tesla K40m", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",1}, {"WGS3",256}, {"WPT3",1} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "GRID K520", { {"WGS1",128}, {"WPT1",1} } },
+ { "GeForce GTX 1070", { {"WGS1",64}, {"WPT1",1} } },
+ { "GeForce GTX 480", { {"WGS1",256}, {"WPT1",1} } },
+ { "GeForce GTX 670", { {"WGS1",128}, {"WPT1",1} } },
+ { "GeForce GTX 680", { {"WGS1",128}, {"WPT1",1} } },
+ { "GeForce GTX 750", { {"WGS1",64}, {"WPT1",1} } },
+ { "GeForce GTX 750 Ti", { {"WGS1",64}, {"WPT1",1} } },
+ { "GeForce GTX 980", { {"WGS1",64}, {"WPT1",1} } },
+ { "GeForce GTX TITAN", { {"WGS1",256}, {"WPT1",1} } },
+ { "GeForce GTX TITAN X", { {"WGS1",64}, {"WPT1",1} } },
+ { "Tesla K20m", { {"WGS1",256}, {"WPT1",1} } },
+ { "Tesla K40m", { {"WGS1",256}, {"WPT1",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // Default
kDeviceTypeAll, "default", {
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
}
@@ -208,38 +208,38 @@ const Database::DatabaseEntry Database::XgemvComplexDouble = {
"Xgemv", Precision::kComplexDouble, {
{ // AMD GPUs
kDeviceTypeGPU, "AMD", {
- { "AMD Radeon R9 M370X Compute Engine", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",1}, {"WGS3",128}, {"WPT3",1} } },
- { "Hawaii", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "Oland", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",256}, {"WPT2",1}, {"VW3",1}, {"WGS3",256}, {"WPT3",1} } },
- { "Pitcairn", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "Tahiti", { {"WGS1",256}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "AMD Radeon R9 M370X Compute Engine", { {"WGS1",64}, {"WPT1",1} } },
+ { "Hawaii", { {"WGS1",64}, {"WPT1",1} } },
+ { "Oland", { {"WGS1",256}, {"WPT1",1} } },
+ { "Pitcairn", { {"WGS1",256}, {"WPT1",1} } },
+ { "Tahiti", { {"WGS1",256}, {"WPT1",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // Intel CPUs
kDeviceTypeCPU, "Intel", {
- { "Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz", { {"WGS1",64}, {"WPT1",1}, {"VW2",2}, {"WGS2",64}, {"WPT2",4}, {"VW3",1}, {"WGS3",256}, {"WPT3",1} } },
- { "Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz", { {"WGS1",64}, {"WPT1",4}, {"VW2",4}, {"WGS2",64}, {"WPT2",4}, {"VW3",2}, {"WGS3",256}, {"WPT3",2} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",2}, {"WGS2",64}, {"WPT2",4}, {"VW3",1}, {"WGS3",256}, {"WPT3",1} } },
+ { "Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz", { {"WGS1",64}, {"WPT1",1} } },
+ { "Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz", { {"WGS1",64}, {"WPT1",4} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // Intel accelerators
kDeviceTypeAccelerator, "Intel", {
- { "Intel(R) Many Integrated Core Acceleration Card", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "Intel(R) Many Integrated Core Acceleration Card", { {"WGS1",64}, {"WPT1",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // NVIDIA GPUs
kDeviceTypeGPU, "NVIDIA", {
- { "GRID K520", { {"WGS1",128}, {"WPT1",1}, {"VW2",1}, {"WGS2",128}, {"WPT2",1}, {"VW3",1}, {"WGS3",256}, {"WPT3",1} } },
- { "GeForce GTX 480", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "GeForce GTX 670", { {"WGS1",128}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "GRID K520", { {"WGS1",128}, {"WPT1",1} } },
+ { "GeForce GTX 480", { {"WGS1",64}, {"WPT1",1} } },
+ { "GeForce GTX 670", { {"WGS1",128}, {"WPT1",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
{ // Default
kDeviceTypeAll, "default", {
- { "default", { {"WGS1",64}, {"WPT1",1}, {"VW2",1}, {"WGS2",64}, {"WPT2",1}, {"VW3",1}, {"WGS3",64}, {"WPT3",1} } },
+ { "default", { {"WGS1",64}, {"WPT1",1} } },
}
},
}
diff --git a/src/database/kernels/xgemv_fast.hpp b/src/database/kernels/xgemv_fast.hpp
new file mode 100644
index 00000000..65b15030
--- /dev/null
+++ b/src/database/kernels/xgemv_fast.hpp
@@ -0,0 +1,247 @@
+
+// =================================================================================================
+// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This
+// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max-
+// width of 100 characters per line.
+//
+// Author(s):
+// Database generator <database.py>
+//
+// This file populates the database with best-found tuning parameters for the 'Xgemv_Fast' kernels.
+//
+// =================================================================================================
+
+namespace clblast {
+// =================================================================================================
+
+const Database::DatabaseEntry Database::XgemvFastHalf = {
+ "XgemvFast", Precision::kHalf, {
+ { // Intel GPUs
+ kDeviceTypeGPU, "Intel", {
+ { "Intel(R) HD Graphics Skylake ULT GT2", { {"VW2",2}, {"WGS2",128}, {"WPT2",2} } },
+ { "default", { {"VW2",2}, {"WGS2",128}, {"WPT2",2} } },
+ }
+ },
+ { // Default
+ kDeviceTypeAll, "default", {
+ { "default", { {"VW2",2}, {"WGS2",128}, {"WPT2",2} } },
+ }
+ },
+ }
+};
+
+// =================================================================================================
+
+const Database::DatabaseEntry Database::XgemvFastSingle = {
+ "XgemvFast", Precision::kSingle, {
+ { // AMD GPUs
+ kDeviceTypeGPU, "AMD", {
+ { "AMD Radeon R9 M370X Compute Engine", { {"VW2",1}, {"WGS2",128}, {"WPT2",1} } },
+ { "Hawaii", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "Oland", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "Pitcairn", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "Tahiti", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ { // Intel CPUs
+ kDeviceTypeCPU, "Intel", {
+ { "Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz", { {"VW2",4}, {"WGS2",128}, {"WPT2",4} } },
+ { "Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz", { {"VW2",1}, {"WGS2",64}, {"WPT2",4} } },
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",4} } },
+ }
+ },
+ { // Intel GPUs
+ kDeviceTypeGPU, "Intel", {
+ { "Intel(R) HD Graphics 530", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "Intel(R) HD Graphics Haswell Ultrabook GT2 Mobile", { {"VW2",4}, {"WGS2",128}, {"WPT2",4} } },
+ { "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",128}, {"WPT2",1} } },
+ }
+ },
+ { // Intel accelerators
+ kDeviceTypeAccelerator, "Intel", {
+ { "Intel(R) Many Integrated Core Acceleration Card", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ { // NVIDIA GPUs
+ kDeviceTypeGPU, "NVIDIA", {
+ { "GRID K520", { {"VW2",2}, {"WGS2",256}, {"WPT2",2} } },
+ { "GeForce GTX 1070", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "GeForce GTX 480", { {"VW2",1}, {"WGS2",128}, {"WPT2",1} } },
+ { "GeForce GTX 670", { {"VW2",2}, {"WGS2",256}, {"WPT2",2} } },
+ { "GeForce GTX 680", { {"VW2",1}, {"WGS2",128}, {"WPT2",1} } },
+ { "GeForce GTX 750", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "GeForce GTX 750 Ti", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "GeForce GTX 980", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "GeForce GTX TITAN", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "GeForce GTX TITAN X", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "Tesla K20m", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "Tesla K40m", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ { // Default
+ kDeviceTypeAll, "default", {
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ }
+};
+
+// =================================================================================================
+
+const Database::DatabaseEntry Database::XgemvFastComplexSingle = {
+ "XgemvFast", Precision::kComplexSingle, {
+ { // AMD GPUs
+ kDeviceTypeGPU, "AMD", {
+ { "AMD Radeon R9 M370X Compute Engine", { {"VW2",2}, {"WGS2",256}, {"WPT2",2} } },
+ { "Hawaii", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "Oland", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "Pitcairn", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "Tahiti", { {"VW2",1}, {"WGS2",128}, {"WPT2",1} } },
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ { // Intel CPUs
+ kDeviceTypeCPU, "Intel", {
+ { "Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz", { {"VW2",1}, {"WGS2",128}, {"WPT2",2} } },
+ { "Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz", { {"VW2",4}, {"WGS2",64}, {"WPT2",4} } },
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",2} } },
+ }
+ },
+ { // Intel GPUs
+ kDeviceTypeGPU, "Intel", {
+ { "Intel(R) HD Graphics 530", { {"VW2",2}, {"WGS2",128}, {"WPT2",2} } },
+ { "Intel(R) HD Graphics Haswell Ultrabook GT2 Mobile", { {"VW2",2}, {"WGS2",128}, {"WPT2",2} } },
+ { "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",64}, {"WPT2",1} } },
+ }
+ },
+ { // Intel accelerators
+ kDeviceTypeAccelerator, "Intel", {
+ { "Intel(R) Many Integrated Core Acceleration Card", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ { // NVIDIA GPUs
+ kDeviceTypeGPU, "NVIDIA", {
+ { "GRID K520", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "GeForce GTX 1070", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "GeForce GTX 480", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "GeForce GTX 670", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "GeForce GTX 680", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "GeForce GTX 750 Ti", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ { // Default
+ kDeviceTypeAll, "default", {
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ }
+};
+
+// =================================================================================================
+
+const Database::DatabaseEntry Database::XgemvFastDouble = {
+ "XgemvFast", Precision::kDouble, {
+ { // AMD GPUs
+ kDeviceTypeGPU, "AMD", {
+ { "AMD Radeon R9 M370X Compute Engine", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "Hawaii", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "Oland", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "Pitcairn", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "Tahiti", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ { // Intel CPUs
+ kDeviceTypeCPU, "Intel", {
+ { "Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz", { {"VW2",4}, {"WGS2",128}, {"WPT2",4} } },
+ { "Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz", { {"VW2",1}, {"WGS2",64}, {"WPT2",4} } },
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",4} } },
+ }
+ },
+ { // Intel accelerators
+ kDeviceTypeAccelerator, "Intel", {
+ { "Intel(R) Many Integrated Core Acceleration Card", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ { // NVIDIA GPUs
+ kDeviceTypeGPU, "NVIDIA", {
+ { "GRID K520", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "GeForce GTX 1070", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "GeForce GTX 480", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "GeForce GTX 670", { {"VW2",1}, {"WGS2",128}, {"WPT2",1} } },
+ { "GeForce GTX 680", { {"VW2",1}, {"WGS2",128}, {"WPT2",1} } },
+ { "GeForce GTX 750", { {"VW2",2}, {"WGS2",256}, {"WPT2",2} } },
+ { "GeForce GTX 750 Ti", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "GeForce GTX 980", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "GeForce GTX TITAN", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "GeForce GTX TITAN X", { {"VW2",1}, {"WGS2",128}, {"WPT2",1} } },
+ { "Tesla K20m", { {"VW2",1}, {"WGS2",128}, {"WPT2",1} } },
+ { "Tesla K40m", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ { // Default
+ kDeviceTypeAll, "default", {
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ }
+};
+
+// =================================================================================================
+
+const Database::DatabaseEntry Database::XgemvFastComplexDouble = {
+ "XgemvFast", Precision::kComplexDouble, {
+ { // AMD GPUs
+ kDeviceTypeGPU, "AMD", {
+ { "AMD Radeon R9 M370X Compute Engine", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "Hawaii", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "Oland", { {"VW2",1}, {"WGS2",256}, {"WPT2",1} } },
+ { "Pitcairn", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "Tahiti", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ { // Intel CPUs
+ kDeviceTypeCPU, "Intel", {
+ { "Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz", { {"VW2",2}, {"WGS2",64}, {"WPT2",4} } },
+ { "Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz", { {"VW2",4}, {"WGS2",64}, {"WPT2",4} } },
+ { "default", { {"VW2",2}, {"WGS2",64}, {"WPT2",4} } },
+ }
+ },
+ { // Intel accelerators
+ kDeviceTypeAccelerator, "Intel", {
+ { "Intel(R) Many Integrated Core Acceleration Card", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ { // NVIDIA GPUs
+ kDeviceTypeGPU, "NVIDIA", {
+ { "GRID K520", { {"VW2",1}, {"WGS2",128}, {"WPT2",1} } },
+ { "GeForce GTX 480", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "GeForce GTX 670", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ { // Default
+ kDeviceTypeAll, "default", {
+ { "default", { {"VW2",1}, {"WGS2",64}, {"WPT2",1} } },
+ }
+ },
+ }
+};
+
+// =================================================================================================
+} // namespace clblast
diff --git a/src/database/kernels/xgemv_fast_rot.hpp b/src/database/kernels/xgemv_fast_rot.hpp
new file mode 100644
index 00000000..9822fb20
--- /dev/null
+++ b/src/database/kernels/xgemv_fast_rot.hpp
@@ -0,0 +1,138 @@
+
+// =================================================================================================
+// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This
+// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max-
+// width of 100 characters per line.
+//
+// Author(s):
+// Database generator <database.py>
+//
+// This file populates the database with best-found tuning parameters for the 'Xgemv_Fast_Rot' kernels.
+//
+// =================================================================================================
+
+namespace clblast {
+// =================================================================================================
+
+const Database::DatabaseEntry Database::XgemvFastRotSingle = {
+ "XgemvFastRot", Precision::kSingle, {
+ { // AMD GPUs
+ kDeviceTypeGPU, "AMD", {
+ { "AMD Radeon R9 M370X Compute Engine", { {"VW3",8}, {"WGS3",64}, {"WPT3",32} } },
+ { "default", { {"VW3",8}, {"WGS3",64}, {"WPT3",32} } },
+ }
+ },
+ { // Intel CPUs
+ kDeviceTypeCPU, "Intel", {
+ { "Intel(R) Core(TM) i7-5930K CPU @ 3.50GHz", { {"VW3",8}, {"WGS3",16}, {"WPT3",8} } },
+ { "default", { {"VW3",8}, {"WGS3",16}, {"WPT3",8} } },
+ }
+ },
+ { // Intel GPUs
+ kDeviceTypeGPU, "Intel", {
+ { "Intel(R) HD Graphics Skylake ULT GT2", { {"VW3",4}, {"WGS3",128}, {"WPT3",16} } },
+ { "Iris Pro", { {"VW3",4}, {"WGS3",32}, {"WPT3",16} } },
+ { "default", { {"VW3",4}, {"WGS3",32}, {"WPT3",16} } },
+ }
+ },
+ { // NVIDIA GPUs
+ kDeviceTypeGPU, "NVIDIA", {
+ { "GeForce GTX TITAN", { {"VW3",1}, {"WGS3",16}, {"WPT3",16} } },
+ { "default", { {"VW3",1}, {"WGS3",16}, {"WPT3",16} } },
+ }
+ },
+ { // Default
+ kDeviceTypeAll, "default", {
+ { "default", { {"VW3",1}, {"WGS3",16}, {"WPT3",8} } },
+ }
+ },
+ }
+};
+
+// =================================================================================================
+
+const Database::DatabaseEntry Database::XgemvFastRotComplexSingle = {
+ "XgemvFastRot", Precision::kComplexSingle, {
+ { // AMD GPUs
+ kDeviceTypeGPU, "AMD", {
+ { "AMD Radeon R9 M370X Compute Engine", { {"VW3",8}, {"WGS3",16}, {"WPT3",16} } },
+ { "default", { {"VW3",8}, {"WGS3",16}, {"WPT3",16} } },
+ }
+ },
+ { // Intel CPUs
+ kDeviceTypeCPU, "Intel", {
+ { "Intel(R) Core(TM) i7-5930K CPU @ 3.50GHz", { {"VW3",4}, {"WGS3",16}, {"WPT3",16} } },
+ { "default", { {"VW3",4}, {"WGS3",16}, {"WPT3",16} } },
+ }
+ },
+ { // Intel GPUs
+ kDeviceTypeGPU, "Intel", {
+ { "Intel(R) HD Graphics Skylake ULT GT2", { {"VW3",2}, {"WGS3",32}, {"WPT3",16} } },
+ { "Iris Pro", { {"VW3",4}, {"WGS3",16}, {"WPT3",16} } },
+ { "default", { {"VW3",2}, {"WGS3",16}, {"WPT3",16} } },
+ }
+ },
+ { // Default
+ kDeviceTypeAll, "default", {
+ { "default", { {"VW3",2}, {"WGS3",16}, {"WPT3",16} } },
+ }
+ },
+ }
+};
+
+// =================================================================================================
+
+const Database::DatabaseEntry Database::XgemvFastRotDouble = {
+ "XgemvFastRot", Precision::kDouble, {
+ { // AMD GPUs
+ kDeviceTypeGPU, "AMD", {
+ { "AMD Radeon R9 M370X Compute Engine", { {"VW3",4}, {"WGS3",16}, {"WPT3",16} } },
+ { "default", { {"VW3",4}, {"WGS3",16}, {"WPT3",16} } },
+ }
+ },
+ { // Intel CPUs
+ kDeviceTypeCPU, "Intel", {
+ { "Intel(R) Core(TM) i7-5930K CPU @ 3.50GHz", { {"VW3",8}, {"WGS3",16}, {"WPT3",8} } },
+ { "default", { {"VW3",8}, {"WGS3",16}, {"WPT3",8} } },
+ }
+ },
+ { // NVIDIA GPUs
+ kDeviceTypeGPU, "NVIDIA", {
+ { "GeForce GTX TITAN", { {"VW3",1}, {"WGS3",16}, {"WPT3",16} } },
+ { "default", { {"VW3",1}, {"WGS3",16}, {"WPT3",16} } },
+ }
+ },
+ { // Default
+ kDeviceTypeAll, "default", {
+ { "default", { {"VW3",1}, {"WGS3",16}, {"WPT3",8} } },
+ }
+ },
+ }
+};
+
+// =================================================================================================
+
+const Database::DatabaseEntry Database::XgemvFastRotComplexDouble = {
+ "XgemvFastRot", Precision::kComplexDouble, {
+ { // AMD GPUs
+ kDeviceTypeGPU, "AMD", {
+ { "AMD Radeon R9 M370X Compute Engine", { {"VW3",4}, {"WGS3",32}, {"WPT3",16} } },
+ { "default", { {"VW3",4}, {"WGS3",32}, {"WPT3",16} } },
+ }
+ },
+ { // Intel CPUs
+ kDeviceTypeCPU, "Intel", {
+ { "Intel(R) Core(TM) i7-5930K CPU @ 3.50GHz", { {"VW3",8}, {"WGS3",16}, {"WPT3",16} } },
+ { "default", { {"VW3",8}, {"WGS3",16}, {"WPT3",16} } },
+ }
+ },
+ { // Default
+ kDeviceTypeAll, "default", {
+ { "default", { {"VW3",4}, {"WGS3",16}, {"WPT3",16} } },
+ }
+ },
+ }
+};
+
+// =================================================================================================
+} // namespace clblast
generated by cgit v1.2.3 (git 2.39.1) at 2024-08-01 08:40:00 +0000