Merge branch 'master' into CLBlast-267-convgemm

4 files changed, 13 insertions, 4 deletions

diff --git a/src/database/kernels/xgemm_direct/xgemm_direct_32.hpp b/src/database/kernels/xgemm_direct/xgemm_direct_32.hpp
index b75a75fb..f6ea9523 100644
--- a/src/database/kernels/xgemm_direct/xgemm_direct_32.hpp
+++ b/src/database/kernels/xgemm_direct/xgemm_direct_32.hpp
@@ -55,6 +55,7 @@ const DatabaseEntry XgemmDirectSingle = {
         { "default", {
           { Name{"Intel(R) Core(TM) i7-2670QM CPU @ 2.20GHz         "}, Params{ 2, 8, 8, 8, 8, 0, 0, 1, 8, 64, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i5-4570 CPU @ 3.20GHz           "}, Params{ 8, 16, 16, 16, 16, 0, 0, 1, 1, 64, 0, 0, 0, 0, 0, 0 } },
+          { Name{"Intel(R) Core(TM) i5-4590S CPU @ 3.00GHz          "}, Params{ 8, 8, 8, 8, 8, 0, 0, 8, 4, 64, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz          "}, Params{ 2, 32, 32, 32, 32, 0, 0, 1, 1, 64, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i7 CPU         920  @ 2.67GHz   "}, Params{ 16, 16, 8, 8, 8, 0, 0, 2, 4, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz          "}, Params{ 2, 8, 8, 8, 8, 0, 0, 2, 2, 64, 0, 0, 0, 0, 0, 0 } },
@@ -66,6 +67,7 @@ const DatabaseEntry XgemmDirectSingle = {
     { // Intel GPUs
       kDeviceTypeGPU, "Intel", {
         { "default", {
+          { Name{"Intel(R) HD Graphics 6000 BroadWell U-Processor GT"}, Params{ 2, 16, 16, 8, 8, 1, 1, 2, 1, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) HD Graphics 620                          "}, Params{ 2, 16, 16, 8, 8, 1, 1, 2, 1, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) HD Graphics IvyBridge M GT2              "}, Params{ 8, 16, 8, 16, 16, 1, 0, 2, 2, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) HD Graphics Skylake ULT GT2              "}, Params{ 2, 8, 8, 8, 8, 1, 1, 1, 1, 8, 0, 0, 0, 0, 0, 0 } },
@@ -99,10 +101,11 @@ const DatabaseEntry XgemmDirectSingle = {
           { kDeviceNameDefault                                        , Params{ 2, 8, 8, 32, 8, 1, 1, 2, 1, 32, 0, 0, 0, 0, 0, 0 } },
         } },
         { "SM6.1", {
+          { Name{"GeForce GTX 1070 Ti                               "}, Params{ 2, 16, 8, 8, 8, 1, 1, 1, 2, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"GeForce GTX 1080                                  "}, Params{ 16, 16, 8, 16, 8, 1, 1, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"GeForce GTX 1080 Ti                               "}, Params{ 16, 8, 8, 16, 16, 1, 1, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"TITAN X (Pascal)                                  "}, Params{ 8, 32, 8, 8, 16, 1, 1, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
-          { kDeviceNameDefault                                        , Params{ 2, 16, 16, 8, 8, 1, 1, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
+          { kDeviceNameDefault                                        , Params{ 2, 8, 8, 8, 8, 1, 1, 4, 2, 32, 0, 0, 0, 0, 0, 0 } },
         } },
         { "default", {
           { kDeviceNameDefault                                        , Params{ 2, 8, 8, 16, 16, 1, 1, 4, 2, 32, 0, 0, 0, 0, 0, 0 } },
diff --git a/src/database/kernels/xgemm_direct/xgemm_direct_3232.hpp b/src/database/kernels/xgemm_direct/xgemm_direct_3232.hpp
index fe0fd030..8f24ee7d 100644
--- a/src/database/kernels/xgemm_direct/xgemm_direct_3232.hpp
+++ b/src/database/kernels/xgemm_direct/xgemm_direct_3232.hpp
@@ -50,6 +50,7 @@ const DatabaseEntry XgemmDirectComplexSingle = {
         { "default", {
           { Name{"Intel(R) Core(TM) i7-2670QM CPU @ 2.20GHz         "}, Params{ 2, 8, 8, 8, 8, 0, 0, 4, 4, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i5-4570 CPU @ 3.20GHz           "}, Params{ 2, 16, 8, 16, 8, 0, 0, 2, 1, 32, 0, 0, 0, 0, 0, 0 } },
+          { Name{"Intel(R) Core(TM) i5-4590S CPU @ 3.00GHz          "}, Params{ 16, 32, 32, 8, 8, 1, 0, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz          "}, Params{ 8, 8, 8, 8, 8, 0, 0, 1, 1, 8, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i7 CPU         920  @ 2.67GHz   "}, Params{ 2, 16, 16, 8, 8, 1, 1, 1, 4, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz          "}, Params{ 2, 8, 8, 16, 8, 1, 1, 2, 1, 32, 0, 0, 0, 0, 0, 0 } },
@@ -92,6 +93,7 @@ const DatabaseEntry XgemmDirectComplexSingle = {
           { kDeviceNameDefault                                        , Params{ 2, 16, 16, 8, 8, 1, 1, 1, 4, 32, 0, 0, 0, 0, 0, 0 } },
         } },
         { "SM6.1", {
+          { Name{"GeForce GTX 1070 Ti                               "}, Params{ 2, 8, 8, 16, 8, 1, 1, 1, 1, 16, 0, 0, 0, 0, 0, 0 } },
           { Name{"GeForce GTX 1080                                  "}, Params{ 8, 8, 16, 16, 8, 1, 1, 2, 2, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"GeForce GTX 1080 Ti                               "}, Params{ 2, 16, 8, 16, 8, 1, 1, 1, 1, 16, 0, 0, 0, 0, 0, 0 } },
           { Name{"TITAN X (Pascal)                                  "}, Params{ 2, 16, 16, 8, 8, 1, 1, 1, 1, 16, 0, 0, 0, 0, 0, 0 } },
diff --git a/src/database/kernels/xgemm_direct/xgemm_direct_64.hpp b/src/database/kernels/xgemm_direct/xgemm_direct_64.hpp
index 6f2e0f1c..9d563e78 100644
--- a/src/database/kernels/xgemm_direct/xgemm_direct_64.hpp
+++ b/src/database/kernels/xgemm_direct/xgemm_direct_64.hpp
@@ -46,6 +46,7 @@ const DatabaseEntry XgemmDirectDouble = {
         { "default", {
           { Name{"Intel(R) Core(TM) i7-2670QM CPU @ 2.20GHz         "}, Params{ 2, 8, 8, 8, 8, 1, 1, 4, 4, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i5-4570 CPU @ 3.20GHz           "}, Params{ 2, 8, 8, 8, 8, 1, 1, 4, 4, 32, 0, 0, 0, 0, 0, 0 } },
+          { Name{"Intel(R) Core(TM) i5-4590S CPU @ 3.00GHz          "}, Params{ 8, 8, 8, 32, 8, 0, 1, 2, 2, 64, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz          "}, Params{ 2, 8, 8, 8, 8, 1, 1, 1, 1, 8, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i7 CPU         920  @ 2.67GHz   "}, Params{ 8, 8, 8, 8, 8, 0, 0, 1, 4, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz          "}, Params{ 2, 8, 8, 8, 8, 1, 1, 4, 4, 32, 0, 0, 0, 0, 0, 0 } },
@@ -86,6 +87,7 @@ const DatabaseEntry XgemmDirectDouble = {
           { kDeviceNameDefault                                        , Params{ 2, 16, 16, 8, 8, 1, 1, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
         } },
         { "SM6.1", {
+          { Name{"GeForce GTX 1070 Ti                               "}, Params{ 2, 8, 8, 16, 8, 1, 1, 1, 1, 16, 0, 0, 0, 0, 0, 0 } },
           { Name{"GeForce GTX 1080                                  "}, Params{ 2, 16, 16, 8, 8, 1, 1, 1, 2, 16, 0, 0, 0, 0, 0, 0 } },
           { Name{"GeForce GTX 1080 Ti                               "}, Params{ 2, 16, 8, 8, 8, 1, 1, 1, 1, 16, 0, 0, 0, 0, 0, 0 } },
           { Name{"TITAN X (Pascal)                                  "}, Params{ 2, 8, 8, 8, 8, 1, 1, 1, 1, 16, 0, 0, 0, 0, 0, 0 } },
@@ -99,7 +101,7 @@ const DatabaseEntry XgemmDirectDouble = {
     { // Default
       kDeviceTypeAll, "default", {
         { "default", {
-          { kDeviceNameDefault                                        , Params{ 2, 8, 8, 8, 8, 1, 1, 2, 2, 16, 0, 0, 0, 0, 0, 0 } },
+          { kDeviceNameDefault                                        , Params{ 2, 16, 16, 8, 8, 1, 1, 2, 4, 32, 0, 0, 0, 0, 0, 0 } },
         } },
       }
     },
diff --git a/src/database/kernels/xgemm_direct/xgemm_direct_6464.hpp b/src/database/kernels/xgemm_direct/xgemm_direct_6464.hpp
index e2aa9f47..af5eeeb5 100644
--- a/src/database/kernels/xgemm_direct/xgemm_direct_6464.hpp
+++ b/src/database/kernels/xgemm_direct/xgemm_direct_6464.hpp
@@ -46,11 +46,12 @@ const DatabaseEntry XgemmDirectComplexDouble = {
         { "default", {
           { Name{"Intel(R) Core(TM) i7-2670QM CPU @ 2.20GHz         "}, Params{ 2, 8, 8, 32, 8, 0, 0, 1, 1, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i5-4570 CPU @ 3.20GHz           "}, Params{ 2, 16, 16, 8, 8, 0, 0, 1, 4, 32, 0, 0, 0, 0, 0, 0 } },
+          { Name{"Intel(R) Core(TM) i5-4590S CPU @ 3.00GHz          "}, Params{ 2, 16, 16, 8, 8, 1, 1, 1, 4, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i5-6200U CPU @ 2.30GHz          "}, Params{ 2, 8, 8, 8, 8, 0, 0, 2, 2, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i7 CPU         920  @ 2.67GHz   "}, Params{ 8, 16, 16, 8, 8, 0, 0, 2, 1, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz          "}, Params{ 8, 16, 8, 8, 8, 0, 0, 2, 2, 32, 0, 0, 0, 0, 0, 0 } },
           { Name{"Intel(R) Core(TM) i7-6770HQ CPU @ 2.60GHz         "}, Params{ 2, 32, 8, 8, 8, 0, 0, 1, 4, 32, 0, 0, 0, 0, 0, 0 } },
-          { kDeviceNameDefault                                        , Params{ 2, 8, 8, 8, 8, 1, 1, 1, 2, 16, 0, 0, 0, 0, 0, 0 } },
+          { kDeviceNameDefault                                        , Params{ 2, 8, 8, 8, 8, 1, 1, 2, 2, 16, 0, 0, 0, 0, 0, 0 } },
         } },
       }
     },
@@ -86,6 +87,7 @@ const DatabaseEntry XgemmDirectComplexDouble = {
           { kDeviceNameDefault                                        , Params{ 2, 8, 8, 8, 8, 1, 1, 2, 1, 16, 0, 0, 0, 0, 0, 0 } },
         } },
         { "SM6.1", {
+          { Name{"GeForce GTX 1070 Ti                               "}, Params{ 2, 16, 16, 16, 16, 1, 1, 1, 1, 16, 0, 0, 0, 0, 0, 0 } },
           { Name{"GeForce GTX 1080                                  "}, Params{ 2, 16, 16, 8, 8, 1, 1, 1, 1, 16, 0, 0, 0, 0, 0, 0 } },
           { Name{"GeForce GTX 1080 Ti                               "}, Params{ 2, 16, 16, 8, 8, 1, 1, 1, 1, 16, 0, 0, 0, 0, 0, 0 } },
           { Name{"TITAN X (Pascal)                                  "}, Params{ 2, 16, 16, 8, 8, 1, 1, 1, 2, 16, 0, 0, 0, 0, 0, 0 } },
@@ -99,7 +101,7 @@ const DatabaseEntry XgemmDirectComplexDouble = {
     { // Default
       kDeviceTypeAll, "default", {
         { "default", {
-          { kDeviceNameDefault                                        , Params{ 2, 8, 8, 8, 8, 1, 1, 1, 1, 16, 0, 0, 0, 0, 0, 0 } },
+          { kDeviceNameDefault                                        , Params{ 2, 8, 8, 8, 8, 1, 1, 2, 1, 16, 0, 0, 0, 0, 0, 0 } },
         } },
       }
     },