Merge pull request #566 from dbarry9/2026.01.13_hp-presets

framework: half-precision presets

Author: Treece-Burgess

src/papiStdEventDefs.h

@@ -166,6 +166,8 @@ enum
 	PAPI_VEC_SP_idx,		 /* Single precision vector/SIMD instructions */
 	PAPI_VEC_DP_idx,		 /* Double precision vector/SIMD instructions */
 	PAPI_REF_CYC_idx,		 /* Reference clock cycles */
+	PAPI_HP_OPS_idx,		 /* Floating point operations executed; optimized to count scaled half precision vector operations */
+	PAPI_VEC_HP_idx,		 /* Half precision vector/SIMD instructions */
 	PAPI_END_idx			 /*This should always be last! */
 };
 
@@ -277,6 +279,8 @@ enum
 #define PAPI_VEC_SP  (PAPI_VEC_SP_idx  | PAPI_PRESET_MASK)	/* Single precision vector/SIMD instructions */
 #define PAPI_VEC_DP  (PAPI_VEC_DP_idx  | PAPI_PRESET_MASK)	/* Double precision vector/SIMD instructions */
 #define PAPI_REF_CYC (PAPI_REF_CYC_idx  | PAPI_PRESET_MASK)	/* Reference clock cycles */
+#define PAPI_HP_OPS  (PAPI_HP_OPS_idx  | PAPI_PRESET_MASK)	/* Floating point operations executed; optimized to count scaled half precision vector operations */
+#define PAPI_VEC_HP  (PAPI_VEC_HP_idx  | PAPI_PRESET_MASK)	/* Half precision vector/SIMD instructions */
 
 #define PAPI_END     (PAPI_END_idx  | PAPI_PRESET_MASK)	/*This should always be last! */
 

src/papi_common_strings.h

@@ -544,8 +544,16 @@ hwi_presets_t _papi_hwi_presets[PAPI_MAX_PRESET_EVENTS] = {
 	  "Reference clock cycles", 0,
 	  0, PAPI_PRESET_BIT_MSC,
       NULL, {0}, {NULL}, {NULL}, {0}, {NULL}, NULL, 0, 0, {NULL}, {NULL}},
-/*108 */ {NULL, NULL, NULL, 0, 0, 0, NULL, {0}, {NULL}, {NULL}, {0}, {NULL}, NULL, 0, 0, {NULL}, {NULL}},
-/*109 */ {NULL, NULL, NULL, 0, 0, 0, NULL, {0}, {NULL}, {NULL}, {0}, {NULL}, NULL, 0, 0, {NULL}, {NULL}},
+/*108 */ {"PAPI_HP_OPS",
+	  "HP operations",
+	  "Floating point operations; optimized to count scaled half precision vector operations", 0,
+	  0, PAPI_PRESET_BIT_INS + PAPI_PRESET_BIT_FP,
+	  NULL, {0}, {NULL}, {NULL}, {0}, {NULL}, NULL, 0, 0, {NULL}, {NULL}},
+/*109 */ {"PAPI_VEC_HP",
+	  "HP Vector/SIMD instr",
+	  "Half precision vector/SIMD instructions", 0,
+	  0, PAPI_PRESET_BIT_INS + PAPI_PRESET_BIT_FP,
+	  NULL, {0}, {NULL}, {NULL}, {0}, {NULL}, NULL, 0, 0, {NULL}, {NULL}},
 /*110 */ {NULL, NULL, NULL, 0, 0, 0, NULL, {0}, {NULL}, {NULL}, {0}, {NULL}, NULL, 0, 0, {NULL}, {NULL}},
 /*111 */ {NULL, NULL, NULL, 0, 0, 0, NULL, {0}, {NULL}, {NULL}, {0}, {NULL}, NULL, 0, 0, {NULL}, {NULL}},
 /*112 */ {NULL, NULL, NULL, 0, 0, 0, NULL, {0}, {NULL}, {NULL}, {0}, {NULL}, NULL, 0, 0, {NULL}, {NULL}},

src/papi_events.csv

@@ -1041,11 +1041,13 @@ PRESET,PAPI_REF_CYC,NOT_DERIVED,UNHALTED_REFERENCE_CYCLES
 # FLOPs
 PRESET,PAPI_DP_OPS,DERIVED_POSTFIX,N0|N1|2|*|+|N2|4|*|+|N3|8|*|+|,FP_ARITH_INST_RETIRED:SCALAR_DOUBLE,FP_ARITH_INST_RETIRED:128B_PACKED_DOUBLE,FP_ARITH_INST_RETIRED:256B_PACKED_DOUBLE,FP_ARITH_INST_RETIRED:512B_PACKED_DOUBLE
 PRESET,PAPI_SP_OPS,DERIVED_POSTFIX,N0|N1|4|*|+|N2|8|*|+|N3|16|*|+|,FP_ARITH_INST_RETIRED:SCALAR_SINGLE,FP_ARITH_INST_RETIRED:128B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:256B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:512B_PACKED_SINGLE
-PRESET,PAPI_FP_OPS,DERIVED_POSTFIX,N0|N1|4|*|+|N2|8|*|+|N3|16|*|+|N4|+|N5|2|*|+|N6|4|*|+|N7|8|*|+|,FP_ARITH_INST_RETIRED:SCALAR_SINGLE,FP_ARITH_INST_RETIRED:128B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:256B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:512B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:SCALAR_DOUBLE,FP_ARITH_INST_RETIRED:128B_PACKED_DOUBLE,FP_ARITH_INST_RETIRED:256B_PACKED_DOUBLE,FP_ARITH_INST_RETIRED:512B_PACKED_DOUBLE
-PRESET,PAPI_FP_INS,DERIVED_POSTFIX,N0|N1|N2|N3|N4|N5|N6|N7|+|+|+|+|+|+|+|,FP_ARITH_INST_RETIRED:SCALAR_SINGLE,FP_ARITH_INST_RETIRED:128B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:256B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:512B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:SCALAR_DOUBLE,FP_ARITH_INST_RETIRED:128B_PACKED_DOUBLE,FP_ARITH_INST_RETIRED:256B_PACKED_DOUBLE,FP_ARITH_INST_RETIRED:512B_PACKED_DOUBLE
+PRESET,PAPI_HP_OPS,DERIVED_POSTFIX,N0|N1|8|*|+|N2|16|*|+|N3|32|*|+|,FP_ARITH_INST_RETIRED2:SCALAR_HALF,FP_ARITH_INST_RETIRED2:128B_PACKED_HALF,FP_ARITH_INST_RETIRED2:256B_PACKED_HALF,FP_ARITH_INST_RETIRED2:512B_PACKED_HALF
+#PRESET,PAPI_FP_OPS,DERIVED_POSTFIX,N0|N1|4|*|+|N2|8|*|+|N3|16|*|+|N4|+|N5|2|*|+|N6|4|*|+|N7|8|*|+|,FP_ARITH_INST_RETIRED:SCALAR_SINGLE,FP_ARITH_INST_RETIRED:128B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:256B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:512B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:SCALAR_DOUBLE,FP_ARITH_INST_RETIRED:128B_PACKED_DOUBLE,FP_ARITH_INST_RETIRED:256B_PACKED_DOUBLE,FP_ARITH_INST_RETIRED:512B_PACKED_DOUBLE
+PRESET,PAPI_FP_INS,DERIVED_POSTFIX,N0|N1|N2|N3|+|+|+|,FP_ARITH_INST_RETIRED:SCALAR,FP_ARITH_INST_RETIRED:VECTOR,FP_ARITH_INST_RETIRED2:SCALAR,FP_ARITH_INST_RETIRED2:VECTOR
 PRESET,PAPI_VEC_DP,DERIVED_POSTFIX,N0|N1|N2|+|+|,FP_ARITH_INST_RETIRED:128B_PACKED_DOUBLE,FP_ARITH_INST_RETIRED:256B_PACKED_DOUBLE,FP_ARITH_INST_RETIRED:512B_PACKED_DOUBLE
 PRESET,PAPI_VEC_SP,DERIVED_POSTFIX,N0|N1|N2|+|+|,FP_ARITH_INST_RETIRED:128B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:256B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:512B_PACKED_SINGLE
-PRESET,PAPI_VEC_INS,DERIVED_POSTFIX,N0|N1|N2|N3|N4|N5|+|+|+|+|+|,FP_ARITH_INST_RETIRED:128B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:256B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:512B_PACKED_SINGLE,FP_ARITH_INST_RETIRED:128B_PACKED_DOUBLE,FP_ARITH_INST_RETIRED:256B_PACKED_DOUBLE,FP_ARITH_INST_RETIRED:512B_PACKED_DOUBLE
+PRESET,PAPI_VEC_HP,DERIVED_POSTFIX,N0|N1|N2|+|+|,FP_ARITH_INST_RETIRED2:128B_PACKED_HALF,FP_ARITH_INST_RETIRED2:256B_PACKED_HALF,FP_ARITH_INST_RETIRED2:512B_PACKED_HALF
+PRESET,PAPI_VEC_INS,DERIVED_ADD,FP_ARITH_INST_RETIRED:VECTOR,FP_ARITH_INST_RETIRED2:VECTOR
 # Branches
 PRESET,PAPI_BR_UCN,DERIVED_SUB,BR_INST_RETIRED:ALL_BRANCHES,BR_INST_RETIRED:COND
 PRESET,PAPI_BR_CN,NOT_DERIVED,BR_INST_RETIRED:COND

src/papi_internal.h

@@ -97,7 +97,7 @@ extern char **_papi_errlist;
 #define NEED_CONTEXT		1
 #define DONT_NEED_CONTEXT 	0
 
-#define PAPI_EVENTS_IN_DERIVED_EVENT	8
+#define PAPI_EVENTS_IN_DERIVED_EVENT	12
 #define PAPI_MAX_COMP_QUALS	8