[mono][interp] Lower div.un to shr.un.imm (#83498)

* [mono][interp] Remove redundant opcode

Generate sub opcode instead which is detected by other optimizations.

* [mono][interp] Lower div.un to shr.un

Author: BrzVlad

src/mono/mono/mini/interp/interp.c

@@ -5963,11 +5963,6 @@ MINT_IN_CASE(MINT_BRTRUE_I8_SP) ZEROP_SP(gint64, !=); MINT_IN_BREAK;
 			ip += 5;
 			MINT_IN_BREAK;
 		}
-		MINT_IN_CASE(MINT_INTRINS_UNSAFE_BYTE_OFFSET) {
-			LOCAL_VAR (ip [1], mono_u) = LOCAL_VAR (ip [3], guint8*) - LOCAL_VAR (ip [2], guint8*);
-			ip += 4;
-			MINT_IN_BREAK;
-		}
 		MINT_IN_CASE(MINT_INTRINS_RUNTIMEHELPERS_OBJECT_HAS_COMPONENT_SIZE) {
 			MonoObject *obj = LOCAL_VAR (ip [2], MonoObject*);
 			LOCAL_VAR (ip [1], gint32) = (obj->vtable->flags & MONO_VT_FLAG_ARRAY_OR_STRING) != 0;

src/mono/mono/mini/interp/jiterpreter.c

@@ -666,7 +666,6 @@ jiterp_should_abort_trace (InterpInst *ins, gboolean *inside_branch_block)
 		case MINT_GETITEM_SPAN:
 		case MINT_GETITEM_LOCALSPAN:
 		case MINT_INTRINS_SPAN_CTOR:
-		case MINT_INTRINS_UNSAFE_BYTE_OFFSET:
 		case MINT_INTRINS_GET_TYPE:
 		case MINT_INTRINS_MEMORYMARSHAL_GETARRAYDATAREF:
 		case MINT_CASTCLASS:

src/mono/mono/mini/interp/mintops.def

@@ -824,7 +824,6 @@ OPDEF(MINT_INTRINS_GET_HASHCODE, "intrins_get_hashcode", 3, 1, 1, MintOpNoArgs)
 OPDEF(MINT_INTRINS_TRY_GET_HASHCODE, "intrins_try_get_hashcode", 3, 1, 1, MintOpNoArgs)
 OPDEF(MINT_INTRINS_GET_TYPE, "intrins_get_type", 3, 1, 1, MintOpNoArgs)
 OPDEF(MINT_INTRINS_SPAN_CTOR, "intrins_span_ctor", 4, 1, 2, MintOpNoArgs)
-OPDEF(MINT_INTRINS_UNSAFE_BYTE_OFFSET, "intrins_unsafe_byte_offset", 4, 1, 2, MintOpNoArgs)
 OPDEF(MINT_INTRINS_RUNTIMEHELPERS_OBJECT_HAS_COMPONENT_SIZE, "intrins_runtimehelpers_object_has_component_size", 3, 1, 1, MintOpNoArgs)
 OPDEF(MINT_INTRINS_CLEAR_WITH_REFERENCES, "intrin_clear_with_references", 3, 0, 2, MintOpNoArgs)
 OPDEF(MINT_INTRINS_MARVIN_BLOCK, "intrins_marvin_block", 3, 0, 2, MintOpNoArgs)

src/mono/mono/mini/interp/transform.c

@@ -2119,7 +2119,17 @@ interp_handle_intrinsics (TransformData *td, MonoMethod *target_method, MonoClas
 		} else if (!strcmp (tm, "AreSame")) {
 			*op = MINT_CEQ_P;
 		} else if (!strcmp (tm, "ByteOffset")) {
-			*op = MINT_INTRINS_UNSAFE_BYTE_OFFSET;
+#if SIZEOF_VOID_P == 4
+			interp_add_ins (td, MINT_SUB_I4);
+#else
+			interp_add_ins (td, MINT_SUB_I8);
+#endif
+			td->sp -= 2;
+			interp_ins_set_sregs2 (td->last_ins, td->sp [1].local, td->sp [0].local);
+			push_simple_type (td, STACK_TYPE_I);
+			interp_ins_set_dreg (td->last_ins, td->sp [-1].local);
+			td->ip += 5;
+			return TRUE;
 		} else if (!strcmp (tm, "Unbox")) {
 			MonoGenericContext *ctx = mono_method_get_context (target_method);
 			g_assert (ctx);
@@ -9873,6 +9883,39 @@ interp_super_instructions (TransformData *td)
 						dump_interp_inst (new_inst);
 					}
 				}
+			} else if (opcode == MINT_DIV_UN_I4 || opcode == MINT_DIV_UN_I8) {
+				// ldc + div.un -> shr.imm
+				int sreg_imm = ins->sregs [1];
+				InterpInst *def = td->locals [sreg_imm].def;
+				if (def != NULL && td->local_ref_count [sreg_imm] == 1) {
+					int power2 = -1;
+					if (MINT_IS_LDC_I4 (def->opcode)) {
+						guint32 ct = interp_get_const_from_ldc_i4 (def);
+						power2 = mono_is_power_of_two ((guint32)ct);
+					} else if (MINT_IS_LDC_I8 (def->opcode)) {
+						guint64 ct = interp_get_const_from_ldc_i8 (def);
+						if (ct < G_MAXUINT32)
+							power2 = mono_is_power_of_two ((guint32)ct);
+					}
+					if (power2 > 0) {
+						InterpInst *new_inst;
+						if (opcode == MINT_DIV_UN_I4)
+							new_inst = interp_insert_ins (td, ins, MINT_SHR_UN_I4_IMM);
+						else
+							new_inst = interp_insert_ins (td, ins, MINT_SHR_UN_I8_IMM);
+						new_inst->dreg = ins->dreg;
+						new_inst->sregs [0] = ins->sregs [0];
+						new_inst->data [0] = power2;
+
+						interp_clear_ins (def);
+						interp_clear_ins (ins);
+						local_ref_count [sreg_imm]--;
+						if (td->verbose_level) {
+							g_print ("lower div.un: ");
+							dump_interp_inst (new_inst);
+						}
+					}
+				}
 			} else if (MINT_IS_LDIND_INT (opcode)) {
 				int sreg_base = ins->sregs [0];
 				InterpInst *def = td->locals [sreg_base].def;

src/mono/wasm/runtime/jiterpreter-trace-generator.ts

@@ -613,13 +613,6 @@ export function generate_wasm_body (
                 builder.callImport("ldtsflda");
                 break;
             }
-            case MintOpcode.MINT_INTRINS_UNSAFE_BYTE_OFFSET:
-                builder.local("pLocals");
-                append_ldloc(builder, getArgU16(ip, 3), WasmOpcode.i32_load);
-                append_ldloc(builder, getArgU16(ip, 2), WasmOpcode.i32_load);
-                builder.appendU8(WasmOpcode.i32_sub);
-                append_stloc_tail(builder, getArgU16(ip, 1), WasmOpcode.i32_store);
-                break;
             case MintOpcode.MINT_INTRINS_GET_TYPE:
                 builder.block();
                 // dest, src