gh-104584: Move super-instruction special-casing to generator (#106500)

Instead of special-casing specific instructions,
we add a few more special values to the 'size' field of expansions,
so in the future we can automatically handle
additional super-instructions in the generator.

2 files changed, 25 insertions, 37 deletions

diff --git a/Python/opcode_metadata.h b/Python/opcode_metadata.h
index ac86a4a..d29f721 100644
--- a/Python/opcode_metadata.h
+++ b/Python/opcode_metadata.h
@@ -934,6 +934,12 @@ struct opcode_macro_expansion {
     struct { int16_t uop; int8_t size; int8_t offset; } uops[8];
 };
 
+#define OPARG_FULL 0
+#define OPARG_CACHE_1 1
+#define OPARG_CACHE_2 2
+#define OPARG_CACHE_4 4
+#define OPARG_TOP 5
+#define OPARG_BOTTOM 6
 
 #define OPCODE_METADATA_FMT(OP) (_PyOpcode_opcode_metadata[(OP)].instr_format)
 #define SAME_OPCODE_METADATA(OP1, OP2) \
@@ -1165,8 +1171,11 @@ const struct opcode_macro_expansion _PyOpcode_macro_expansion[256] = {
     [LOAD_FAST_CHECK] = { .nuops = 1, .uops = { { LOAD_FAST_CHECK, 0, 0 } } },
     [LOAD_FAST] = { .nuops = 1, .uops = { { LOAD_FAST, 0, 0 } } },
     [LOAD_FAST_AND_CLEAR] = { .nuops = 1, .uops = { { LOAD_FAST_AND_CLEAR, 0, 0 } } },
+    [LOAD_FAST_LOAD_FAST] = { .nuops = 2, .uops = { { LOAD_FAST, 5, 0 }, { LOAD_FAST, 6, 0 } } },
     [LOAD_CONST] = { .nuops = 1, .uops = { { LOAD_CONST, 0, 0 } } },
     [STORE_FAST] = { .nuops = 1, .uops = { { STORE_FAST, 0, 0 } } },
+    [STORE_FAST_LOAD_FAST] = { .nuops = 2, .uops = { { STORE_FAST, 5, 0 }, { LOAD_FAST, 6, 0 } } },
+    [STORE_FAST_STORE_FAST] = { .nuops = 2, .uops = { { STORE_FAST, 5, 0 }, { STORE_FAST, 6, 0 } } },
     [POP_TOP] = { .nuops = 1, .uops = { { POP_TOP, 0, 0 } } },
     [PUSH_NULL] = { .nuops = 1, .uops = { { PUSH_NULL, 0, 0 } } },
     [END_FOR] = { .nuops = 2, .uops = { { POP_TOP, 0, 0 }, { POP_TOP, 0, 0 } } },
diff --git a/Python/optimizer.c b/Python/optimizer.c
index db117bb..2870f2f 100644
--- a/Python/optimizer.c
+++ b/Python/optimizer.c
@@ -411,44 +411,15 @@ translate_bytecode_to_trace(
     for (;;) {
         ADD_TO_TRACE(SAVE_IP, (int)(instr - (_Py_CODEUNIT *)code->co_code_adaptive));
         int opcode = instr->op.code;
-        uint64_t operand = instr->op.arg;
+        int oparg = instr->op.arg;
         int extras = 0;
         while (opcode == EXTENDED_ARG) {
             instr++;
             extras += 1;
             opcode = instr->op.code;
-            operand = (operand << 8) | instr->op.arg;
+            oparg = (oparg << 8) | instr->op.arg;
         }
         switch (opcode) {
-            case LOAD_FAST_LOAD_FAST:
-            case STORE_FAST_LOAD_FAST:
-            case STORE_FAST_STORE_FAST:
-            {
-                // Reserve space for two uops (+ SAVE_IP + EXIT_TRACE)
-                if (trace_length + 4 > max_length) {
-                    DPRINTF(1, "Ran out of space for LOAD_FAST_LOAD_FAST\n");
-                    goto done;
-                }
-                uint64_t oparg1 = operand >> 4;
-                uint64_t oparg2 = operand & 15;
-                switch (opcode) {
-                    case LOAD_FAST_LOAD_FAST:
-                        ADD_TO_TRACE(LOAD_FAST, oparg1);
-                        ADD_TO_TRACE(LOAD_FAST, oparg2);
-                        break;
-                    case STORE_FAST_LOAD_FAST:
-                        ADD_TO_TRACE(STORE_FAST, oparg1);
-                        ADD_TO_TRACE(LOAD_FAST, oparg2);
-                        break;
-                    case STORE_FAST_STORE_FAST:
-                        ADD_TO_TRACE(STORE_FAST, oparg1);
-                        ADD_TO_TRACE(STORE_FAST, oparg2);
-                        break;
-                    default:
-                        Py_FatalError("Missing case");
-                }
-                break;
-            }
             default:
             {
                 const struct opcode_macro_expansion *expansion = &_PyOpcode_macro_expansion[opcode];
@@ -462,9 +433,11 @@ translate_bytecode_to_trace(
                         goto done;
                     }
                     for (int i = 0; i < nuops; i++) {
+                        uint64_t operand;
                         int offset = expansion->uops[i].offset;
                         switch (expansion->uops[i].size) {
-                            case 0:
+                            case OPARG_FULL:
+                                operand = oparg;
                                 if (extras && OPCODE_HAS_JUMP(opcode)) {
                                     if (opcode == JUMP_BACKWARD_NO_INTERRUPT) {
                                         operand -= extras;
@@ -475,19 +448,25 @@ translate_bytecode_to_trace(
                                     }
                                 }
                                 break;
-                            case 1:
+                            case OPARG_CACHE_1:
                                 operand = read_u16(&instr[offset].cache);
                                 break;
-                            case 2:
+                            case OPARG_CACHE_2:
                                 operand = read_u32(&instr[offset].cache);
                                 break;
-                            case 4:
+                            case OPARG_CACHE_4:
                                 operand = read_u64(&instr[offset].cache);
                                 break;
+                            case OPARG_TOP:  // First half of super-instr
+                                operand = oparg >> 4;
+                                break;
+                            case OPARG_BOTTOM:  // Second half of super-instr
+                                operand = oparg & 0xF;
+                                break;
                             default:
                                 fprintf(stderr,
-                                        "opcode=%d, operand=%" PRIu64 "; nuops=%d, i=%d; size=%d, offset=%d\n",
-                                        opcode, operand, nuops, i,
+                                        "opcode=%d, oparg=%d; nuops=%d, i=%d; size=%d, offset=%d\n",
+                                        opcode, oparg, nuops, i,
                                         expansion->uops[i].size,
                                         expansion->uops[i].offset);
                                 Py_FatalError("garbled expansion");