BCRYPT Kernels: Improved bcrypt performance by 6.5% for high-end NVIDIA GPU devices using CUDA backend

This commit is contained in:
Jens Steube
2021-06-01 22:52:07 +02:00
parent 26fa6f5f32
commit ce8c121b50
7 changed files with 388 additions and 238 deletions
+1 -8
View File
@@ -102,14 +102,7 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
if (device_param->is_cuda == true)
{
if (device_param->kernel_dynamic_local_mem_size_memset % device_param->device_local_mem_size)
{
// this is the case Compute Capability 7.5
// there is also Compute Capability 7.0 which offers a larger dynamic local size access
// however, if it's an exact multiple the driver can optimize this for us more efficient
use_dynamic = true;
}
use_dynamic = true;
}
// this uses some nice feedback effect.
+1 -8
View File
@@ -83,14 +83,7 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
if (device_param->is_cuda == true)
{
if (device_param->kernel_dynamic_local_mem_size_memset % device_param->device_local_mem_size)
{
// this is the case Compute Capability 7.5
// there is also Compute Capability 7.0 which offers a larger dynamic local size access
// however, if it's an exact multiple the driver can optimize this for us more efficient
use_dynamic = true;
}
use_dynamic = true;
}
// this uses some nice feedback effect.
+1 -8
View File
@@ -83,14 +83,7 @@ char *module_jit_build_options (MAYBE_UNUSED const hashconfig_t *hashconfig, MAY
if (device_param->is_cuda == true)
{
if (device_param->kernel_dynamic_local_mem_size_memset % device_param->device_local_mem_size)
{
// this is the case Compute Capability 7.5
// there is also Compute Capability 7.0 which offers a larger dynamic local size access
// however, if it's an exact multiple the driver can optimize this for us more efficient
use_dynamic = true;
}
use_dynamic = true;
}
// this uses some nice feedback effect.