Cuda: Statistic Qualifier

icl-utk-edu · Feb 3, 2025 · 3a40e1a · 3a40e1a
1 parent 09fee01
commit 3a40e1a
Show file tree

Hide file tree

Showing 11 changed files with 450 additions and 111 deletions.
diff --git a/src/components/cuda/README_internal.md b/src/components/cuda/README_internal.md
@@ -5,32 +5,44 @@ At current the Cuda component uses bits to create an Event identifier. The follo
 # Event Identifier Encoding Format
 
 ## Unused bits
-As of 09/16/24, there are a total of 34 unused bits. These bits can be used to create a new qualifier or can be used to extended the number of bits for an existing qualifier.
+
+As of 02/02/25, there are a total of 2 unused bits. These bits can be used to create a new qualifier or can be used to extended the number of bits for an existing qualifier.
+
+## STAT
+
+3 bits are allocated for the statistic qualifier. ([0 - 8 stats]).
 
 ## Device
+
 7 bits are allocated for the device which accounts for 128 total devices on a node (e.g. [0 - 127 devices]).
 
 ## Qlmask
-2 bits are allocated for the qualifier mask. 
+
+2 bits are allocated for the qualifier mask.
 
 ## Nameid
+
 21 bits are allocated for the nameid which will roughly account for greater than 2 million Cuda native events per device on a node.
 
 ## Calculations for Bit Masks and Shifts
-| #DEFINE    | Bits |
-| -------- | ------- |
-| EVENTS_WIDTH  | `(sizeof(uint64_t) * 8)`    |
-| DEVICE_WIDTH | `( 7)`   |
-| QLMASK_WIDTH    | `( 2)`   |
-| NAMEID_WIDTH  | `(21)`    |
-| UNUSED_WIDTH   | `(EVENTS_WIDTH - DEVICE_WIDTH - QLMASK_WIDTH - NAMEID_WIDTH)`   |
-| DEVICE_SHIFT  | `(EVENTS_WIDTH - UNUSED_WIDTH - DEVICE_WIDTH)`    |
-| QLMASK_SHIFT | `(DEVICE_SHIFT - QLMASK_WIDTH)`   |
-| NAMEID_SHIFT    | `(QLMASK_SHIFT - NAMEID_WIDTH)`   |
-| DEVICE_MASK  | `((0xFFFFFFFFFFFFFFFF >> (EVENTS_WIDTH - DEVICE_WIDTH)) << DEVICE_SHIFT)`    |
-| QLMASK_MASK | `((0xFFFFFFFFFFFFFFFF >> (EVENTS_WIDTH - QLMASK_WIDTH)) << QLMASK_SHIFT)`   |
-| NAMEID_MASK   | `((0xFFFFFFFFFFFFFFFF >> (EVENTS_WIDTH - NAMEID_WIDTH)) << NAMEID_SHIFT)`   |
-| DEVICE_FLAG  | `DEVICE_FLAG  (0x1)`   |
 
+| #DEFINE      | Bits                                                                       |
+| ------------ | -------------------------------------------------------------------------- |
+| EVENTS_WIDTH | `(sizeof(uint64_t) * 8)`                                                   |
+| STAT_WIDTH   | `( 3)`                                                                     |
+| DEVICE_WIDTH | `( 7)`                                                                     |
+| QLMASK_WIDTH | `( 2)`                                                                     |
+| NAMEID_WIDTH | `(18)`                                                                     |
+| UNUSED_WIDTH | `(EVENTS_WIDTH - DEVICE_WIDTH - QLMASK_WIDTH - NAMEID_WIDTH - STAT_WIDTH)` |
+| STAT_SHIFT   | `(EVENTS_WIDTH - UNUSED_WIDTH - STAT_WIDTH)`                               |
+| DEVICE_SHIFT | `(EVENTS_WIDTH - UNUSED_WIDTH - STAT_WIDTH - DEVICE_WIDTH)`                |
+| QLMASK_SHIFT | `(DEVICE_SHIFT - QLMASK_WIDTH)`                                            |
+| NAMEID_SHIFT | `(QLMASK_SHIFT - NAMEID_WIDTH)`                                            |
+| STAT_MASK    | `((0xFFFFFFFFFFFFFFFF >> (EVENTS_WIDTH - STAT_WIDTH))   << STAT_SHIFT)`    |
+| DEVICE_MASK  | `((0xFFFFFFFFFFFFFFFF >> (EVENTS_WIDTH - DEVICE_WIDTH)) << DEVICE_SHIFT)`  |
+| QLMASK_MASK  | `((0xFFFFFFFFFFFFFFFF >> (EVENTS_WIDTH - QLMASK_WIDTH)) << QLMASK_SHIFT)`  |
+| NAMEID_MASK  | `((0xFFFFFFFFFFFFFFFF >> (EVENTS_WIDTH - NAMEID_WIDTH)) << NAMEID_SHIFT)`  |
+| STAT_FLAG    | `STAT_FLAG  (0x2)`                                                         |
+| DEVICE_FLAG  | `DEVICE_FLAG  (0x1)`                                                       |
 
 **NOTE**: If adding a new qualifier, you must add it to the table found in the section titled [Calculations for Bit Masks and Shifts](#calculations-for-bit-masks-and-shifts) and account for this addition within `cupti_profiler.c`.
diff --git a/src/components/cuda/cupti_dispatch.c b/src/components/cuda/cupti_dispatch.c
@@ -200,7 +200,7 @@ int cuptid_ctx_destroy(cuptip_control_t *pcupti_ctl)
     return PAPI_ECMP;
 }
 
-int cuptid_evt_enum(uint64_t *event_code, int modifier)
+int cuptid_evt_enum(uint32_t *event_code, int modifier)
 {
     if (cuptic_is_runtime_perfworks_api()) {
 
@@ -218,7 +218,7 @@ int cuptid_evt_enum(uint64_t *event_code, int modifier)
     return PAPI_ECMP;
 }
 
-int cuptid_evt_code_to_descr(uint64_t event_code, char *descr, int len)
+int cuptid_evt_code_to_descr(uint32_t event_code, char *descr, int len)
 {
     if (cuptic_is_runtime_perfworks_api()) {
 
@@ -236,7 +236,7 @@ int cuptid_evt_code_to_descr(uint64_t event_code, char *descr, int len)
     return PAPI_ECMP;
 }
 
-int cuptid_evt_name_to_code(const char *name, uint64_t *event_code)
+int cuptid_evt_name_to_code(const char *name, uint32_t *event_code)
 {
     if (cuptic_is_runtime_perfworks_api()) {
 
@@ -254,7 +254,7 @@ int cuptid_evt_name_to_code(const char *name, uint64_t *event_code)
     return PAPI_ECMP;
 }
 
-int cuptid_evt_code_to_name(uint64_t event_code, char *name, int len)
+int cuptid_evt_code_to_name(uint32_t event_code, char *name, int len)
 {
     if (cuptic_is_runtime_perfworks_api()) {
 
@@ -272,7 +272,7 @@ int cuptid_evt_code_to_name(uint64_t event_code, char *name, int len)
     return PAPI_ECMP;
 }
 
-int cuptid_evt_code_to_info(uint64_t event_code, PAPI_event_info_t *info)
+int cuptid_evt_code_to_info(uint32_t event_code, PAPI_event_info_t *info)
 {
     if (cuptic_is_runtime_perfworks_api()) {
 

diff --git a/src/components/cuda/cupti_dispatch.h b/src/components/cuda/cupti_dispatch.h
@@ -23,14 +23,14 @@ int cuptid_init(void);
 int cuptid_shutdown(void);
 
 /* native event interfaces */
-int cuptid_evt_enum(uint64_t *event_code, int modifier);
-int cuptid_evt_code_to_descr(uint64_t event_code, char *descr, int len);
-int cuptid_evt_name_to_code(const char *name, uint64_t *event_code);
-int cuptid_evt_code_to_name(uint64_t event_code, char *name, int len);
-int cuptid_evt_code_to_info(uint64_t event_code, PAPI_event_info_t *info);
+int cuptid_evt_enum(uint32_t *event_code, int modifier);
+int cuptid_evt_code_to_descr(uint32_t event_code, char *descr, int len);
+int cuptid_evt_name_to_code(const char *name, uint32_t *event_code);
+int cuptid_evt_code_to_name(uint32_t event_code, char *name, int len);
+int cuptid_evt_code_to_info(uint32_t event_code, PAPI_event_info_t *info);
 
 /* profiling context handling interfaces */
-int cuptid_ctx_create(cuptid_info_t thread_info, cuptip_control_t *pcupti_ctl, uint64_t *events_id, int num_events);
+int cuptid_ctx_create(cuptid_info_t thread_info, cuptip_control_t *pcupti_ctl, uint64_t *papi, int num_events);
 int cuptid_ctx_start(cuptip_control_t ctl);
 int cuptid_ctx_read(cuptip_control_t ctl, long long **counters);
 int cuptid_ctx_reset(cuptip_control_t ctl);

diff --git a/src/components/cuda/cupti_events.c b/src/components/cuda/cupti_events.c
@@ -51,27 +51,27 @@ int cuptie_ctx_destroy(cuptie_control_t *pctl)
     return PAPI_ENOIMPL;
 }
 
-int cuptie_evt_enum(uint64_t *event_code, int modifier)
+int cuptie_evt_enum(uint32_t *event_code, int modifier)
 {
     return PAPI_ENOIMPL;
 }
 
-int cuptie_evt_code_to_descr(uint64_t event_code, char *descr, int len) 
+int cuptie_evt_code_to_descr(uint32_t event_code, char *descr, int len) 
 {
     return PAPI_ENOIMPL;
 }
 
-int cuptie_evt_name_to_code(const char *name, uint64_t *event_code)
+int cuptie_evt_name_to_code(const char *name, uint32_t *event_code)
 {
     return PAPI_ENOIMPL;
 }
 
-int cuptie_evt_code_to_name(uint64_t event_code, char *name, int len)
+int cuptie_evt_code_to_name(uint32_t event_code, char *name, int len)
 {
     return PAPI_ENOIMPL;
 }
 
-int cuptie_evt_code_to_info(uint64_t event_code, PAPI_event_info_t *info) 
+int cuptie_evt_code_to_info(uint32_t event_code, PAPI_event_info_t *info) 
 {
     return PAPI_ENOIMPL;
 }

diff --git a/src/components/cuda/cupti_events.h b/src/components/cuda/cupti_events.h
@@ -19,11 +19,11 @@ int cuptie_init(void);
 int cuptie_shutdown(void);
 
 /* native event interfaces */
-int cuptie_evt_enum(uint64_t *event_code, int modifier); 
-int cuptie_evt_code_to_descr(uint64_t event_code, char *descr, int len);
-int cuptie_evt_name_to_code(const char *name, uint64_t *event_code);
-int cuptie_evt_code_to_name(uint64_t event_code, char *name, int len);
-int cuptie_evt_code_to_info(uint64_t event_code, PAPI_event_info_t *info);
+int cuptie_evt_enum(uint32_t *event_code, int modifier); 
+int cuptie_evt_code_to_descr(uint32_t event_code, char *descr, int len);
+int cuptie_evt_name_to_code(const char *name, uint32_t *event_code);
+int cuptie_evt_code_to_name(uint32_t event_code, char *name, int len);
+int cuptie_evt_code_to_info(uint32_t event_code, PAPI_event_info_t *info);
 
 /* profiling context handling interfaces */
 int cuptie_ctx_create(void *thr_info, cuptie_control_t *pctl);