implement extended const expr #4318

kylo5aby · 2025-06-03T05:39:32Z

extended const for interp
extended const for aot
tests
CI

yamt · 2025-06-03T22:41:05Z

core/iwasm/interpreter/wasm_loader.c

+#if WASM_ENABLE_EXTENDED_CONST_EXPR != 0
+            case INIT_EXPR_TYPE_I32_ADD:
+            case INIT_EXPR_TYPE_I32_SUB:
+            case INIT_EXPR_TYPE_I32_MUL:


i guess you can share the logic between i32 and i64.

i guess you can share the logic between i32 and i64.

Resolved. thanks

yamt · 2025-06-03T22:57:34Z

core/iwasm/interpreter/wasm.h

+#if WASM_ENABLE_EXTENDED_CONST_EXPR != 0
+    struct InitializerExpression *l_expr;
+    struct InitializerExpression *r_expr;
+#endif


is there any cases where you need both of value (u) and exprs (l_expr/r_expr)?

is there any cases where you need both of value (u) and exprs (l_expr/r_expr)?

iiuc, for unary init expr, only value (u) is needed, for binary init expr, exprs (l_expr/r_expr) are needed to store sub-exprs in order to compute the expr value at runtime.

can they be a union then?

eg.

union { struct { WASMValue value; } unary; struct { struct InitializerExpression *l_expr; struct InitializerExpression *r_expr; } binary; } u;

can they be a union then?

eg.

union { struct { WASMValue value; } unary; struct { struct InitializerExpression *l_expr; struct InitializerExpression *r_expr; } binary; } u;

I think this is ok, and will save more memories after enable extended const expr, I will update this after finish the aot part, thanks the suggestion.

yamt · 2025-06-03T23:10:36Z

i guess it's safer to have a small limit for the stack height in the loader.

kylo5aby · 2025-06-09T03:58:48Z

i guess it's safer to have a small limit for the stack height in the loader.

I didn't see a limit value specified in spec. and in this discussion, there is a suggestion to limit global initializer go with the function body limit of 7654321 bytes WebAssembly/extended-const#15 (comment), do you have any suggestions?

lum1n0us · 2025-06-08T13:22:01Z

build-scripts/config_common.cmake

@@ -695,11 +709,11 @@ message (
 "       \"Tail call\" via WAMR_BUILD_TAIL_CALL: ${WAMR_BUILD_TAIL_CALL}\n"
 "       \"Threads\" via WAMR_BUILD_SHARED_MEMORY: ${WAMR_BUILD_SHARED_MEMORY}\n"
 "       \"Typed Function References\" via WAMR_BUILD_GC: ${WAMR_BUILD_GC}\n"
+"       \"Extended Constant Expressions\" via WAMR_BUILD_EXTENDED_CONST_EXPR: ${WAMR_BUILD_EXTENDED_CONST_EXPR}\n"


keep alphabeta order

lum1n0us · 2025-06-08T13:24:07Z

core/iwasm/aot/aot_loader.c

-        data_list[i]->offset.init_expr_type = (uint8)init_expr_type;
-        data_list[i]->offset.u.i64 = (int64)init_expr_value;
+        bh_memcpy_s(&data_list[i]->offset, sizeof(InitializerExpression),
+                    &offset_expr, sizeof(InitializerExpression));


better be something like: xxx->offset = offset_expr. Compiler will check the type.

lum1n0us · 2025-06-08T13:25:48Z

wamr-compiler/CMakeLists.txt

+else()
+  add_definitions(-DWASM_ENABLE_EXTENDED_CONST_EXPR=0)
+endif()
+


we might tend to enable it by default in wamrc. Then, wamrc can use a command line option to switch features.

lum1n0us · 2025-06-09T09:08:56Z

core/iwasm/interpreter/wasm_loader.c

+                    value_type = VALUE_TYPE_I64;
+                }
+
+                if (!(pop_const_expr_stack(&const_expr_ctx, &r_flag, value_type,


Please provide a quick introduction to the binary format of a constant expression. It will be helpful.

core/iwasm/interpreter/wasm_loader.c

TianlongLiang · 2025-06-10T03:37:59Z

core/iwasm/aot/aot_loader.c

+#endif
+#if WASM_ENABLE_EXTENDED_CONST_EXPR != 0
+            if (is_expr_binary_op(data_list[i]->offset.init_expr_type)) {
+                destroy_sub_init_expr(&data_list[i]->offset);


Maybe it's a good idea to refactor the destroy_init_expr and merge the code logic of destroy_sub_init_expr into it.

TianlongLiang · 2025-06-10T03:43:33Z

core/iwasm/aot/aot_runtime.c

+                  || offset_flag == INIT_EXPR_TYPE_FUNCREF_CONST
+                  || offset_flag == INIT_EXPR_TYPE_REFNULL_CONST
+                  || (tbl_inst->is_table64
+                          ? (offset_flag == INIT_EXPR_TYPE_I64_CONST


Maybe we should add a function/macro for this check since it appears several times, and use macro to control it, for it is only valid when extended const proposal is on.

TianlongLiang · 2025-06-10T03:53:21Z

core/iwasm/interpreter/wasm_runtime.c

+        uint8 offset_flag = data_seg->base_offset.init_expr_type;
+        bh_assert(offset_flag == INIT_EXPR_TYPE_GET_GLOBAL
+                  || (memory->is_memory64
+                          ? (offset_flag == INIT_EXPR_TYPE_I64_CONST


Same as aot_runtime.c Maybe we should add a function/macro for this check since it appears several times, and use macro to control it, for it is only valid when extendedd const proposal is on.

Same as aot_runtime.c Maybe we should add a function/macro for this check since it appears several times, and use macro to control it, for it is only valid when extendedd const proposal is on.

resolved.

TianlongLiang · 2025-06-10T03:58:40Z

core/iwasm/common/wasm_loader_common.c

+
+#if WASM_ENABLE_EXTENDED_CONST_EXPR != 0
+void
+destroy_init_expr_recursive(InitializerExpression *expr)


Maybe it's a good idea that we extract the destroy_init_expr in wasm_loader.c and unify them here

Maybe it's a good idea that we extract the destroy_init_expr in wasm_loader.c and unify them here

resolved.

TianlongLiang · 2025-06-10T04:01:10Z

core/iwasm/interpreter/wasm.h

+
+#if WASM_ENABLE_EXTENDED_CONST_EXPR != 0
+void
+destroy_init_expr_recursive(InitializerExpression *expr);


Since those two APIs are only used in the loader, better to put them in wasm_loader_common.h

Since those two APIs are only used in the loader, better to put them in wasm_loader_common.h

resolved. thanks

kylo5aby requested review from loganek, lum1n0us, no1wudi, TianlongLiang, wenyongh, xujuntwt95329 and yamt as code owners June 3, 2025 05:39

kylo5aby changed the title ~~implement extended const expr for interp~~ wip: implement extended const expr Jun 3, 2025

kylo5aby mentioned this pull request Jun 3, 2025

extended-const #4272

Open

kylo5aby force-pushed the extended-const-expr branch 2 times, most recently from 09367e2 to 019a93f Compare June 3, 2025 07:25

yamt reviewed Jun 3, 2025

View reviewed changes

kylo5aby force-pushed the extended-const-expr branch 12 times, most recently from 63c429b to 669c1b8 Compare June 9, 2025 03:51

kylo5aby added 3 commits June 9, 2025 12:00

implement extended const expr for interp

1026241

implement extended const expr for aot

8f08062

enable CI

2233ff4

kylo5aby force-pushed the extended-const-expr branch from 669c1b8 to 2233ff4 Compare June 9, 2025 04:04

kylo5aby changed the title ~~wip: implement extended const expr~~ implement extended const expr Jun 9, 2025

kylo5aby force-pushed the extended-const-expr branch 2 times, most recently from e8e5ba2 to 36a8996 Compare June 9, 2025 09:59

use union to replace single WASMValue

0f475a5

kylo5aby force-pushed the extended-const-expr branch from 36a8996 to 0f475a5 Compare June 9, 2025 10:27

lum1n0us reviewed Jun 9, 2025

View reviewed changes

TianlongLiang reviewed Jun 10, 2025

View reviewed changes

kylo5aby force-pushed the extended-const-expr branch 2 times, most recently from 45c71ed to 1f3d64e Compare June 11, 2025 02:47

resolve comments and enable extended const for mini loader

759d61d

kylo5aby force-pushed the extended-const-expr branch from 1f3d64e to 759d61d Compare June 11, 2025 03:07

implement extended const expr #4318

Are you sure you want to change the base?

implement extended const expr #4318

Conversation

kylo5aby commented Jun 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

yamt Jun 4, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

yamt commented Jun 3, 2025

Uh oh!

kylo5aby commented Jun 9, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

kylo5aby commented Jun 3, 2025 •

edited

Loading

yamt Jun 4, 2025 •

edited

Loading