summaryrefslogtreecommitdiff
path: root/src/database/kernels/xgemm_direct/xgemm_direct_64.hpp
diff options
context:
space:
mode:
authorCedric Nugteren <web@cedricnugteren.nl>2017-09-14 21:27:33 +0200
committerCedric Nugteren <web@cedricnugteren.nl>2017-09-14 21:27:33 +0200
commit0d13d814c21f1fdb3b0781b84bf12914aec5225a (patch)
treef4ed5bfe6b13a5e865b1973ca571df87f1af4bc2 /src/database/kernels/xgemm_direct/xgemm_direct_64.hpp
parent14a61d24258498811ac0dfd11cad6087a231fa2a (diff)
Added architecture layer in the tuning database for better performance on unseen devices
Diffstat (limited to 'src/database/kernels/xgemm_direct/xgemm_direct_64.hpp')
-rw-r--r--src/database/kernels/xgemm_direct/xgemm_direct_64.hpp56
1 files changed, 41 insertions, 15 deletions
diff --git a/src/database/kernels/xgemm_direct/xgemm_direct_64.hpp b/src/database/kernels/xgemm_direct/xgemm_direct_64.hpp
index ac740dae..6c35c38c 100644
--- a/src/database/kernels/xgemm_direct/xgemm_direct_64.hpp
+++ b/src/database/kernels/xgemm_direct/xgemm_direct_64.hpp
@@ -14,33 +14,59 @@ const DatabaseEntry XgemmDirectDouble = {
"XgemmDirect", Precision::kDouble, {"KWID", "MDIMAD", "MDIMCD", "NDIMBD", "NDIMCD", "PADA", "PADB", "VWMD", "VWND", "WGD"}, {
{ // AMD GPUs
kDeviceTypeGPU, "AMD", {
- { "AMD Radeon R9 M370X Compute Engine", { 2, 16, 16, 8, 8, 1, 1, 1, 1, 16 } },
- { "Ellesmere", { 8, 16, 16, 8, 16, 1, 1, 2, 1, 32 } },
- { "Fiji", { 16, 8, 8, 8, 16, 1, 1, 1, 1, 16 } },
- { "Tonga", { 2, 16, 16, 16, 16, 1, 1, 1, 1, 32 } },
- { "default", { 2, 16, 16, 16, 16, 1, 1, 1, 1, 16 } },
+ { "Ellesmere", {
+ { "AMD Radeon RX 480", { 8, 16, 16, 8, 16, 1, 1, 2, 1, 32 } },
+ { "default", { 8, 16, 16, 8, 16, 1, 1, 2, 1, 32 } },
+ } },
+ { "Fiji", {
+ { "AMD Radeon R9 Fury X", { 16, 8, 8, 8, 16, 1, 1, 1, 1, 16 } },
+ { "AMD Radeon R9 M370X Compute Engine", { 2, 16, 16, 8, 8, 1, 1, 1, 1, 16 } },
+ { "default", { 2, 16, 16, 8, 8, 1, 1, 1, 1, 16 } },
+ } },
+ { "Tonga", {
+ { "AMD Radeon R9 380", { 2, 16, 16, 16, 16, 1, 1, 1, 1, 32 } },
+ { "default", { 2, 16, 16, 16, 16, 1, 1, 1, 1, 32 } },
+ } },
+ { "default", {
+ { "default", { 2, 16, 16, 16, 16, 1, 1, 1, 1, 16 } },
+ } },
}
},
{ // Intel CPUs
kDeviceTypeCPU, "Intel", {
- { "Intel(R) Core(TM) i7-2670QM CPU @ 2.20GHz", { 2, 8, 8, 8, 8, 1, 1, 4, 4, 32 } },
- { "Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz", { 8, 8, 8, 8, 8, 0, 0, 1, 4, 32 } },
- { "Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz", { 2, 8, 8, 8, 8, 1, 1, 4, 4, 32 } },
- { "default", { 2, 8, 8, 8, 8, 1, 1, 4, 2, 32 } },
+ { "default", {
+ { "Intel(R) Core(TM) i7-2670QM CPU @ 2.20GHz", { 2, 8, 8, 8, 8, 1, 1, 4, 4, 32 } },
+ { "Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz", { 8, 8, 8, 8, 8, 0, 0, 1, 4, 32 } },
+ { "Intel(R) Core(TM) i7-4790K CPU @ 4.00GHz", { 2, 8, 8, 8, 8, 1, 1, 4, 4, 32 } },
+ { "default", { 2, 8, 8, 8, 8, 1, 1, 4, 2, 32 } },
+ } },
}
},
{ // NVIDIA GPUs
kDeviceTypeGPU, "NVIDIA", {
- { "GeForce GTX 1080", { 2, 16, 16, 8, 8, 1, 1, 1, 2, 16 } },
- { "GeForce GTX 750 Ti", { 2, 8, 8, 8, 8, 1, 1, 2, 4, 32 } },
- { "GeForce GTX TITAN Black", { 8, 16, 16, 16, 8, 1, 0, 1, 1, 16 } },
- { "TITAN X (Pascal)", { 2, 8, 8, 8, 8, 1, 1, 1, 1, 16 } },
- { "default", { 2, 8, 8, 8, 8, 1, 1, 1, 2, 16 } },
+ { "SM3.5", {
+ { "GeForce GTX TITAN Black", { 8, 16, 16, 16, 8, 1, 0, 1, 1, 16 } },
+ { "default", { 8, 16, 16, 16, 8, 1, 0, 1, 1, 16 } },
+ } },
+ { "SM5.0", {
+ { "GeForce GTX 750 Ti", { 2, 8, 8, 8, 8, 1, 1, 2, 4, 32 } },
+ { "default", { 2, 8, 8, 8, 8, 1, 1, 2, 4, 32 } },
+ } },
+ { "SM6.1", {
+ { "GeForce GTX 1080", { 2, 16, 16, 8, 8, 1, 1, 1, 2, 16 } },
+ { "TITAN X (Pascal)", { 2, 8, 8, 8, 8, 1, 1, 1, 1, 16 } },
+ { "default", { 2, 8, 8, 8, 8, 1, 1, 1, 1, 16 } },
+ } },
+ { "default", {
+ { "default", { 2, 8, 8, 8, 8, 1, 1, 1, 2, 16 } },
+ } },
}
},
{ // Default
kDeviceTypeAll, "default", {
- { "default", { 2, 8, 8, 8, 8, 1, 1, 2, 2, 16 } },
+ { "default", {
+ { "default", { 2, 8, 8, 8, 8, 1, 1, 2, 2, 16 } },
+ } },
}
},
}