From 37a709bb69fcd9f2b0fe53a189e92e548164cc4b Mon Sep 17 00:00:00 2001 From: Mahmoud Date: Wed, 5 Jun 2019 19:18:29 -0400 Subject: adding new cuda 9 APIs to run the deepbench workloads --- src/abstract_hardware_model.h | 1 + 1 file changed, 1 insertion(+) (limited to 'src/abstract_hardware_model.h') diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h index 22ef509..64bbaa2 100644 --- a/src/abstract_hardware_model.h +++ b/src/abstract_hardware_model.h @@ -212,6 +212,7 @@ public: // m_num_cores_running=0; // m_param_mem=NULL; // } + kernel_info_t( dim3 gridDim, dim3 blockDim, class function_info *entry); kernel_info_t( dim3 gridDim, dim3 blockDim, class function_info *entry, std::map nameToCudaArray, std::map nameToTextureInfo); ~kernel_info_t(); -- cgit v1.3 From ba7145d18dd65391e9510664d26c963fcaab20ee Mon Sep 17 00:00:00 2001 From: "Mahmoud Khairy A. Abdallah" Date: Sun, 14 Jul 2019 10:58:39 -0400 Subject: fixing the volta local memory size per thread --- src/abstract_hardware_model.h | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'src/abstract_hardware_model.h') diff --git a/src/abstract_hardware_model.h b/src/abstract_hardware_model.h index 64bbaa2..27a1ba6 100644 --- a/src/abstract_hardware_model.h +++ b/src/abstract_hardware_model.h @@ -438,9 +438,9 @@ protected: #define GLOBAL_HEAP_START 0xC0000000 // start allocating from this address (lower values used for allocating globals in .ptx file) -#define SHARED_MEM_SIZE_MAX (64*1024) -#define LOCAL_MEM_SIZE_MAX (8*1024) -#define MAX_STREAMING_MULTIPROCESSORS 64 +#define SHARED_MEM_SIZE_MAX (96*1024) +#define LOCAL_MEM_SIZE_MAX (16*1024) +#define MAX_STREAMING_MULTIPROCESSORS 80 //scale it to Volta #define MAX_THREAD_PER_SM 2048 #define MAX_WARP_PER_SM 64 #define TOTAL_LOCAL_MEM_PER_SM (MAX_THREAD_PER_SM*LOCAL_MEM_SIZE_MAX) -- cgit v1.3