summaryrefslogtreecommitdiff
path: root/src/database/kernels/xgemm
diff options
context:
space:
mode:
authorCedric Nugteren <web@cedricnugteren.nl>2021-08-19 20:31:52 +0200
committerCedric Nugteren <web@cedricnugteren.nl>2021-08-19 20:31:52 +0200
commitadb4b02982bb5d7b20600587af4e8bafb99bd5e7 (patch)
tree6e39aca1bbf2310fd0728d3f3981a9b3289bd647 /src/database/kernels/xgemm
parentdea3b5fadbd533616aac9ba48fb991bf85f1e32b (diff)
Add tuning results for NVIDIA Tesla T4
Diffstat (limited to 'src/database/kernels/xgemm')
-rw-r--r--src/database/kernels/xgemm/xgemm_32.hpp1
-rw-r--r--src/database/kernels/xgemm/xgemm_3232.hpp3
-rw-r--r--src/database/kernels/xgemm/xgemm_64.hpp3
-rw-r--r--src/database/kernels/xgemm/xgemm_6464.hpp3
4 files changed, 7 insertions, 3 deletions
diff --git a/src/database/kernels/xgemm/xgemm_32.hpp b/src/database/kernels/xgemm/xgemm_32.hpp
index d71bf11f..983c5f42 100644
--- a/src/database/kernels/xgemm/xgemm_32.hpp
+++ b/src/database/kernels/xgemm/xgemm_32.hpp
@@ -166,6 +166,7 @@ const DatabaseEntry XgemmSingle = {
{ "SM7.5", {
{ Name{"Quadro T2000 "}, Params{ 0, 1, 16, 2, 16, 16, 128, 32, 16, 128, 1, 1, 1, 1, 1, 2 } },
{ Name{"TITAN RTX "}, Params{ 0, 1, 32, 2, 8, 32, 128, 16, 8, 128, 1, 1, 1, 1, 2, 2 } },
+ { Name{"Tesla T4 "}, Params{ 1, 8, 1, 1, 16, 16, 128, 4, 4, 64, 0, 0, 0, 0, 2, 2 } },
{ kDeviceNameDefault , Params{ 0, 1, 16, 2, 16, 16, 128, 32, 16, 128, 1, 1, 1, 1, 1, 2 } },
} },
{ "SM8.0", {
diff --git a/src/database/kernels/xgemm/xgemm_3232.hpp b/src/database/kernels/xgemm/xgemm_3232.hpp
index 49a55a60..7d85e6f1 100644
--- a/src/database/kernels/xgemm/xgemm_3232.hpp
+++ b/src/database/kernels/xgemm/xgemm_3232.hpp
@@ -159,6 +159,7 @@ const DatabaseEntry XgemmComplexSingle = {
{ "SM7.5", {
{ Name{"Quadro T2000 "}, Params{ 0, 1, 32, 2, 16, 8, 64, 8, 32, 128, 0, 0, 1, 0, 2, 2 } },
{ Name{"TITAN RTX "}, Params{ 0, 1, 32, 2, 8, 8, 16, 16, 16, 64, 1, 0, 0, 1, 2, 2 } },
+ { Name{"Tesla T4 "}, Params{ 1, 4, 1, 1, 8, 8, 32, 16, 16, 128, 0, 0, 0, 0, 4, 2 } },
{ kDeviceNameDefault , Params{ 0, 1, 32, 2, 8, 8, 16, 16, 16, 64, 1, 0, 0, 1, 2, 2 } },
} },
{ "SM8.0", {
@@ -173,7 +174,7 @@ const DatabaseEntry XgemmComplexSingle = {
{ // Default
kDeviceTypeAll, "default", {
{ "default", {
- { kDeviceNameDefault , Params{ 0, 1, 32, 2, 16, 16, 32, 8, 8, 32, 1, 1, 0, 0, 1, 4 } },
+ { kDeviceNameDefault , Params{ 0, 1, 32, 2, 32, 32, 32, 8, 8, 32, 1, 1, 0, 0, 1, 4 } },
} },
}
},
diff --git a/src/database/kernels/xgemm/xgemm_64.hpp b/src/database/kernels/xgemm/xgemm_64.hpp
index 69123aef..139d642d 100644
--- a/src/database/kernels/xgemm/xgemm_64.hpp
+++ b/src/database/kernels/xgemm/xgemm_64.hpp
@@ -144,7 +144,8 @@ const DatabaseEntry XgemmDouble = {
{ "SM7.5", {
{ Name{"Quadro T2000 "}, Params{ 0, 1, 32, 2, 8, 16, 128, 8, 8, 64, 0, 0, 1, 1, 4, 4 } },
{ Name{"TITAN RTX "}, Params{ 1, 2, 1, 1, 16, 16, 32, 8, 8, 16, 0, 0, 0, 0, 2, 1 } },
- { kDeviceNameDefault , Params{ 1, 1, 1, 1, 16, 16, 16, 16, 16, 32, 0, 0, 0, 0, 1, 1 } },
+ { Name{"Tesla T4 "}, Params{ 1, 16, 1, 1, 8, 8, 32, 16, 16, 64, 0, 0, 0, 0, 2, 1 } },
+ { kDeviceNameDefault , Params{ 1, 2, 1, 1, 16, 16, 32, 8, 8, 16, 0, 0, 0, 0, 2, 1 } },
} },
{ "SM8.0", {
{ Name{"A100-PCIE-40GB "}, Params{ 0, 1, 32, 2, 32, 32, 64, 8, 8, 32, 1, 1, 0, 0, 2, 2 } },
diff --git a/src/database/kernels/xgemm/xgemm_6464.hpp b/src/database/kernels/xgemm/xgemm_6464.hpp
index 9a12329c..c1b0c22a 100644
--- a/src/database/kernels/xgemm/xgemm_6464.hpp
+++ b/src/database/kernels/xgemm/xgemm_6464.hpp
@@ -143,7 +143,8 @@ const DatabaseEntry XgemmComplexDouble = {
{ "SM7.5", {
{ Name{"Quadro T2000 "}, Params{ 0, 1, 32, 2, 8, 8, 32, 8, 8, 64, 0, 0, 0, 0, 4, 2 } },
{ Name{"TITAN RTX "}, Params{ 0, 1, 32, 2, 8, 8, 16, 16, 16, 16, 0, 0, 0, 0, 1, 1 } },
- { kDeviceNameDefault , Params{ 0, 1, 32, 2, 8, 8, 16, 16, 16, 16, 0, 0, 0, 0, 1, 1 } },
+ { Name{"Tesla T4 "}, Params{ 0, 1, 32, 2, 16, 16, 16, 8, 8, 16, 0, 0, 0, 0, 1, 2 } },
+ { kDeviceNameDefault , Params{ 0, 1, 32, 2, 16, 16, 16, 8, 8, 16, 0, 0, 0, 0, 1, 1 } },
} },
{ "SM8.0", {
{ Name{"A100-PCIE-40GB "}, Params{ 0, 1, 16, 2, 8, 8, 16, 32, 16, 32, 1, 1, 1, 1, 2, 1 } },