GH-111646: Simplify optimizer, by compacting uops when making executor. (GH-111647)

1 files changed, 32 insertions, 3 deletions

diff --git a/Python/optimizer_analysis.c b/Python/optimizer_analysis.c
index 2d177f1..61bda80 100644
--- a/Python/optimizer_analysis.c
+++ b/Python/optimizer_analysis.c
@@ -13,13 +13,42 @@
 #include "pycore_optimizer.h"
 
 
+static void
+remove_unneeded_uops(_PyUOpInstruction *buffer, int buffer_size)
+{
+    // Note that we don't enter stubs, those SET_IPs are needed.
+    int last_set_ip = -1;
+    bool need_ip = true;
+    for (int pc = 0; pc < buffer_size; pc++) {
+        int opcode = buffer[pc].opcode;
+        if (opcode == _SET_IP) {
+            if (!need_ip && last_set_ip >= 0) {
+                buffer[last_set_ip].opcode = NOP;
+            }
+            need_ip = false;
+            last_set_ip = pc;
+        }
+        else if (opcode == _JUMP_TO_TOP || opcode == _EXIT_TRACE) {
+            break;
+        }
+        else {
+            // If opcode has ERROR or DEOPT, set need_ip to true
+            if (_PyOpcode_opcode_metadata[opcode].flags & (HAS_ERROR_FLAG | HAS_DEOPT_FLAG) || opcode == _PUSH_FRAME) {
+                need_ip = true;
+            }
+        }
+    }
+}
+
+
 int
 _Py_uop_analyze_and_optimize(
     PyCodeObject *co,
-    _PyUOpInstruction *trace,
-    int trace_len,
+    _PyUOpInstruction *buffer,
+    int buffer_size,
     int curr_stacklen
 )
 {
-    return trace_len;
+    remove_unneeded_uops(buffer, buffer_size);
+    return 0;
 }