summaryrefslogtreecommitdiff
path: root/src/cuda-sim/cuda-sim.h
diff options
context:
space:
mode:
authorMahmoud <[email protected]>2019-07-29 21:18:06 -0400
committerMahmoud <[email protected]>2019-07-29 21:18:06 -0400
commit5875fda72d4402413bc5c04ee5ec15085ff2b90a (patch)
treec920268d899df331e1a5e451a35133eaff7ca341 /src/cuda-sim/cuda-sim.h
parentc05dc90da35fc2bd9dd42da9626bf3a60e2c9e8d (diff)
parent21d937256fbca004c926531cfef1adefcedeef91 (diff)
Merge branch 'dev' of https://github.com/mkhairy/gpgpu-sim-private into dev-purdue-integration-trace
Diffstat (limited to 'src/cuda-sim/cuda-sim.h')
-rw-r--r--src/cuda-sim/cuda-sim.h7
1 files changed, 5 insertions, 2 deletions
diff --git a/src/cuda-sim/cuda-sim.h b/src/cuda-sim/cuda-sim.h
index 958daba..e690356 100644
--- a/src/cuda-sim/cuda-sim.h
+++ b/src/cuda-sim/cuda-sim.h
@@ -32,6 +32,7 @@
#include"../gpgpu-sim/shader.h"
#include <stdlib.h>
#include <map>
+#include <vector>
#include <string>
#include"ptx_sim.h"
@@ -46,6 +47,8 @@ extern int g_debug_thread_uid;
extern void ** g_inst_classification_stat;
extern void ** g_inst_op_classification_stat;
extern int g_ptx_kernel_count; // used for classification stat collection purposes
+extern char *opcode_latency_int, *opcode_latency_fp, *opcode_latency_dp,*opcode_latency_sfu,*opcode_latency_tensor;
+
void ptx_opcocde_latency_options (option_parser_t opp);
extern class kernel_info_t *gpgpu_opencl_ptx_sim_init_grid(class function_info *entry,
@@ -97,7 +100,7 @@ public:
delete[] m_warpAtBarrier;
}
//! executes all warps till completion
- void execute();
+ void execute(int inst_count, unsigned ctaid_cp);
virtual void warp_exit( unsigned warp_id );
virtual bool warp_waiting_at_barrier( unsigned warp_id ) const
{
@@ -107,7 +110,7 @@ public:
private:
void executeWarp(unsigned, bool &, bool &);
//initializes threads in the CTA block which we are executing
- void initializeCTA();
+ void initializeCTA(unsigned ctaid_cp);
virtual void checkExecutionStatusAndUpdate(warp_inst_t &inst, unsigned t, unsigned tid)
{
if(m_thread[tid]==NULL || m_thread[tid]->is_done()){