gh-115178: Add Counts of UOp Pairs to pystats

Include/cpython/pystats.h

@@ -100,6 +100,7 @@ typedef struct _gc_stats {
 typedef struct _uop_stats {
     uint64_t execution_count;
     uint64_t miss;
+    uint64_t pair_count[MAX_UOP_ID + 1];
 } UOpStats;
 
 #define _Py_UOP_HIST_SIZE 32

Include/internal/pycore_code.h

@@ -310,6 +310,13 @@ extern int _PyStaticCode_Init(PyCodeObject *co);
 #define GC_STAT_ADD(gen, name, n) do { if (_Py_stats) _Py_stats->gc_stats[(gen)].name += (n); } while (0)
 #define OPT_STAT_INC(name) do { if (_Py_stats) _Py_stats->optimization_stats.name++; } while (0)
 #define UOP_STAT_INC(opname, name) do { if (_Py_stats) { assert(opname < 512); _Py_stats->optimization_stats.opcode[opname].name++; } } while (0)
+#define UOP_PAIR_INC(uopcode, lastuop)                                              \
+    do {                                                                            \
+        if (lastuop && _Py_stats) {                                                 \
+            _Py_stats->optimization_stats.opcode[lastuop].pair_count[uopcode]++;    \
+        }                                                                           \
+        lastuop = uopcode;                                                          \
+    } while (0)
 #define OPT_UNSUPPORTED_OPCODE(opname) do { if (_Py_stats) _Py_stats->optimization_stats.unsupported_opcode[opname]++; } while (0)
 #define OPT_ERROR_IN_OPCODE(opname) do { if (_Py_stats) _Py_stats->optimization_stats.error_in_opcode[opname]++; } while (0)
 #define OPT_HIST(length, name) \
@@ -337,6 +344,7 @@ PyAPI_FUNC(PyObject*) _Py_GetSpecializationStats(void);
 #define GC_STAT_ADD(gen, name, n) ((void)0)
 #define OPT_STAT_INC(name) ((void)0)
 #define UOP_STAT_INC(opname, name) ((void)0)
+#define UOP_PAIR_INC(uopcode, lastuop) ((void)0)
 #define OPT_UNSUPPORTED_OPCODE(opname) ((void)0)
 #define OPT_ERROR_IN_OPCODE(opname) ((void)0)
 #define OPT_HIST(length, name) ((void)0)

Misc/ACKS

@@ -641,6 +641,7 @@ Neil Girdhar
 Matt Giuca
 Andrea Giudiceandrea
 Franz Glasner
+Jeff Glass
 Wim Glenn
 Michael Goderbauer
 Karan Goel

Python/ceval.c

@@ -995,6 +995,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
     ; // dummy statement after a label, before a declaration
     uint16_t uopcode;
 #ifdef Py_STATS
+    int lastuop = 0;
     uint64_t trace_uop_execution_counter = 0;
 #endif
 
@@ -1018,6 +1019,7 @@ _PyEval_EvalFrameDefault(PyThreadState *tstate, _PyInterpreterFrame *frame, int
         next_uop++;
         OPT_STAT_INC(uops_executed);
         UOP_STAT_INC(uopcode, execution_count);
+        UOP_PAIR_INC(uopcode, lastuop);
 #ifdef Py_STATS
         trace_uop_execution_counter++;
 #endif

Python/specialize.c

@@ -11,6 +11,7 @@
 #include "pycore_object.h"
 #include "pycore_opcode_metadata.h" // _PyOpcode_Caches
 #include "pycore_uop_metadata.h"    // _PyOpcode_uop_name
+#include "pycore_uop_ids.h"       // MAX_UOP_ID
 #include "pycore_opcode_utils.h"  // RESUME_AT_FUNC_START
 #include "pycore_pylifecycle.h"   // _PyOS_URandomNonblock()
 #include "pycore_runtime.h"       // _Py_ID()
@@ -269,6 +270,14 @@ print_optimization_stats(FILE *out, OptimizationStats *stats)
         }
     }
 
+    for (int i = 1; i <= MAX_UOP_ID; i++){
+        for (int j = 1; j <= MAX_UOP_ID; j++) {
+            if (stats->opcode[i].pair_count[j]) {
+                fprintf(out, "uop[%s].pair_count[%s] : %" PRIu64 "\n",
+                        _PyOpcode_uop_name[i], _PyOpcode_uop_name[j], stats->opcode[i].pair_count[j]);
+            }
+        }
+    }
     for (int i = 0; i < MAX_UOP_ID; i++) {
         if (stats->error_in_opcode[i]) {
             fprintf(

Tools/scripts/summarize_stats.py

@@ -736,9 +736,9 @@ def execution_count_section() -> Section:
     )
 
 
-def pair_count_section() -> Section:
+def pair_count_section(prefix: str, title=None) -> Section:
     def calc_pair_count_table(stats: Stats) -> Rows:
-        opcode_stats = stats.get_opcode_stats("opcode")
+        opcode_stats = stats.get_opcode_stats(prefix)
         pair_counts = opcode_stats.get_pair_counts()
         total = opcode_stats.get_total_execution_count()
 
@@ -760,7 +760,7 @@ def calc_pair_count_table(stats: Stats) -> Rows:
 
     return Section(
         "Pair counts",
-        "Pair counts for top 100 Tier 1 instructions",
+        f"Pair counts for top 100 {title if title else prefix} pairs",
         [
             Table(
                 ("Pair", "Count:", "Self:", "Cumulative:"),
@@ -1232,6 +1232,7 @@ def iter_optimization_tables(base_stats: Stats, head_stats: Stats | None = None)
                 )
             ],
         )
+        yield pair_count_section(prefix="uop", title="Non-JIT uop")
         yield Section(
             "Unsupported opcodes",
             "",
@@ -1292,7 +1293,7 @@ def calc_rows(stats: Stats) -> Rows:
 
 LAYOUT = [
     execution_count_section(),
-    pair_count_section(),
+    pair_count_section("opcode"),
     pre_succ_pairs_section(),
     specialization_section(),
     specialization_effectiveness_section(),