wasm-micro-runtime/core/iwasm/compilation/aot_emit_memory.c

/*
 * Copyright (C) 2019 Intel Corporation. All rights reserved.
 * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 */

#include "aot_emit_memory.h"
#include "aot_compiler.h"
#include "aot_emit_exception.h"
#include "../aot/aot_runtime.h"
#include "aot_intrinsic.h"
#include "aot_emit_control.h"

#define BUILD_IS_NOT_NULL(value, res, name)                                \
    do {                                                                   \
        if (!(res = LLVMBuildIsNotNull(comp_ctx->builder, value, name))) { \
            aot_set_last_error("llvm build is not null failed.");          \
            goto fail;                                                     \
        }                                                                  \
    } while (0)

#define BUILD_BR(llvm_block)                               \
    do {                                                   \
        if (!LLVMBuildBr(comp_ctx->builder, llvm_block)) { \
            aot_set_last_error("llvm build br failed.");   \
            goto fail;                                     \
        }                                                  \
    } while (0)

#define BUILD_COND_BR(value_if, block_then, block_else)               \
    do {                                                              \
        if (!LLVMBuildCondBr(comp_ctx->builder, value_if, block_then, \
                             block_else)) {                           \
            aot_set_last_error("llvm build cond br failed.");         \
            goto fail;                                                \
        }                                                             \
    } while (0)

#define BUILD_TRUNC(value, data_type)                                     \
    do {                                                                  \
        if (!(value = LLVMBuildTrunc(comp_ctx->builder, value, data_type, \
                                     "val_trunc"))) {                     \
            aot_set_last_error("llvm build trunc failed.");               \
            goto fail;                                                    \
        }                                                                 \
    } while (0)

#define BUILD_ICMP(op, left, right, res, name)                                \
    do {                                                                      \
        if (!(res =                                                           \
                  LLVMBuildICmp(comp_ctx->builder, op, left, right, name))) { \
            aot_set_last_error("llvm build icmp failed.");                    \
            goto fail;                                                        \
        }                                                                     \
    } while (0)

#define BUILD_OP(Op, left, right, res, name)                                \
    do {                                                                    \
        if (!(res = LLVMBuild##Op(comp_ctx->builder, left, right, name))) { \
            aot_set_last_error("llvm build " #Op " fail.");                 \
            goto fail;                                                      \
        }                                                                   \
    } while (0)

#define ADD_BASIC_BLOCK(block, name)                                          \
    do {                                                                      \
        if (!(block = LLVMAppendBasicBlockInContext(comp_ctx->context,        \
                                                    func_ctx->func, name))) { \
            aot_set_last_error("llvm add basic block failed.");               \
            goto fail;                                                        \
        }                                                                     \
    } while (0)

#define SET_BUILD_POS(block) LLVMPositionBuilderAtEnd(comp_ctx->builder, block)

static bool
zero_extend_u64(AOTCompContext *comp_ctx, LLVMValueRef *value, const char *name)
{
    if (comp_ctx->pointer_size == sizeof(uint64)) {
        /* zero extend to uint64 if the target is 64-bit */
        *value = LLVMBuildZExt(comp_ctx->builder, *value, I64_TYPE, name);
        if (!*value) {
            aot_set_last_error("llvm build zero extend failed.");
            return false;
        }
    }
    return true;
}

static LLVMValueRef
get_memory_check_bound(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                       uint32 bytes)
{
    LLVMValueRef mem_check_bound = NULL;
    switch (bytes) {
        case 1:
            mem_check_bound = func_ctx->mem_info[0].mem_bound_check_1byte;
            break;
        case 2:
            mem_check_bound = func_ctx->mem_info[0].mem_bound_check_2bytes;
            break;
        case 4:
            mem_check_bound = func_ctx->mem_info[0].mem_bound_check_4bytes;
            break;
        case 8:
            mem_check_bound = func_ctx->mem_info[0].mem_bound_check_8bytes;
            break;
        case 16:
            mem_check_bound = func_ctx->mem_info[0].mem_bound_check_16bytes;
            break;
        default:
            bh_assert(0);
            return NULL;
    }

    if (func_ctx->mem_space_unchanged)
        return mem_check_bound;

    if (!(mem_check_bound = LLVMBuildLoad2(
              comp_ctx->builder,
              (comp_ctx->pointer_size == sizeof(uint64)) ? I64_TYPE : I32_TYPE,
              mem_check_bound, "mem_check_bound"))) {
        aot_set_last_error("llvm build load failed.");
        return NULL;
    }
    return mem_check_bound;
}

#if defined(_WIN32) || defined(_WIN32_)
static inline int
ffs(int n)
{
    int pos = 0;

    if (n == 0)
        return 0;

    while (!(n & 1)) {
        pos++;
        n >>= 1;
    }
    return pos + 1;
}
#endif

static LLVMValueRef
get_memory_curr_page_count(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx);

#if WASM_ENABLE_SHARED_HEAP != 0
uint32
get_module_inst_extra_offset(AOTCompContext *comp_ctx);

#define BUILD_LOAD_PTR(ptr, data_type, res)                           \
    do {                                                              \
        if (!(res = LLVMBuildLoad2(comp_ctx->builder, data_type, ptr, \
                                   "load_value"))) {                  \
            aot_set_last_error("llvm build load failed");             \
            goto fail;                                                \
        }                                                             \
    } while (0)

/* Update last used shared heap info(alloc ptr) in function ctx:
 * 1. shared_heap_start_off 2. shared_heap_end_off 3. shared_heap_base_addr_adj
 */
bool
aot_check_shared_heap_chain_and_update(AOTCompContext *comp_ctx,
                                       AOTFuncContext *func_ctx,
                                       LLVMBasicBlockRef check_succ,
                                       LLVMValueRef start_offset,
                                       LLVMValueRef bytes, bool is_memory64)
{
    LLVMValueRef param_values[7], ret_value, func, value, cmp;
    LLVMTypeRef param_types[7], ret_type, func_type, func_ptr_type;

    param_types[0] = INT8_PTR_TYPE;
    param_types[1] = INTPTR_T_TYPE;
    param_types[2] = SIZE_T_TYPE;
    param_types[3] = INTPTR_T_PTR_TYPE;
    param_types[4] = INTPTR_T_PTR_TYPE;
    param_types[5] = INT8_PTR_TYPE;
    param_types[6] = INT8_TYPE;
    ret_type = INT8_TYPE;

    GET_AOT_FUNCTION(wasm_runtime_check_and_update_last_used_shared_heap, 7);

    /* Call function */
    param_values[0] = func_ctx->aot_inst;
    param_values[1] = start_offset;
    param_values[2] = bytes;
    /* pass alloc ptr */
    param_values[3] = func_ctx->shared_heap_start_off;
    param_values[4] = func_ctx->shared_heap_end_off;
    param_values[5] = func_ctx->shared_heap_base_addr_adj;
    param_values[6] = is_memory64 ? I8_ONE : I8_ZERO;

    if (!(ret_value = LLVMBuildCall2(comp_ctx->builder, func_type, func,
                                     param_values, 7, "call"))) {
        aot_set_last_error("llvm build call failed.");
        goto fail;
    }

    BUILD_ICMP(LLVMIntEQ, ret_value, I8_ZERO, cmp, "shared_heap_oob");
    if (!aot_emit_exception(comp_ctx, func_ctx,
                            EXCE_OUT_OF_BOUNDS_MEMORY_ACCESS, true, cmp,
                            check_succ)) {
        goto fail;
    }

    return true;
fail:
    return false;
}

/*
 * Setup the basic blocks for shared heap and shared chain memory checks.
 *
 * Arguments:
 *   block_curr: The current basic block.
 *   app_addr_in_cache_shared_heap: Output, block for cache shared heap.
 *   app_addr_in_linear_mem: Output, block for linear memory.
 *   app_addr_in_shared_heap_chain: Output, block for shared heap chain
 *                                  (only for shared heap chain).
 *   check_shared_heap_chain: Output, block for checking shared heap chain
 *                            (only for shared heap chain).
 *
 * Topology:
 *   If enable_shared_heap:
 *     block_curr -> app_addr_in_cache_shared_heap
 *                -> app_addr_in_linear_mem
 *   If enable_shared_chain:
 *     block_curr -> app_addr_in_shared_heap_chain
 *                   -> app_addr_in_cache_shared_heap
 *                   -> check_shared_heap_chain
 *                 -> app_addr_in_linear_mem
 */
static bool
setup_shared_heap_blocks(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                         LLVMBasicBlockRef block_curr,
                         LLVMBasicBlockRef *app_addr_in_cache_shared_heap,
                         LLVMBasicBlockRef *app_addr_in_linear_mem,
                         LLVMBasicBlockRef *app_addr_in_shared_heap_chain,
                         LLVMBasicBlockRef *check_shared_heap_chain)
{
    ADD_BASIC_BLOCK(*app_addr_in_cache_shared_heap,
                    "app_addr_in_cache_shared_heap");
    ADD_BASIC_BLOCK(*app_addr_in_linear_mem, "app_addr_in_linear_mem");

    if (comp_ctx->enable_shared_heap) {
        LLVMMoveBasicBlockAfter(*app_addr_in_cache_shared_heap, block_curr);
        LLVMMoveBasicBlockAfter(*app_addr_in_linear_mem,
                                *app_addr_in_cache_shared_heap);
    }
    else if (comp_ctx->enable_shared_chain) {
        ADD_BASIC_BLOCK(*app_addr_in_shared_heap_chain,
                        "app_addr_in_shared_heap_chain");
        ADD_BASIC_BLOCK(*check_shared_heap_chain, "check_shared_heap_chain");
        LLVMMoveBasicBlockAfter(*app_addr_in_shared_heap_chain, block_curr);
        LLVMMoveBasicBlockAfter(*app_addr_in_cache_shared_heap,
                                *app_addr_in_shared_heap_chain);
        LLVMMoveBasicBlockAfter(*check_shared_heap_chain,
                                *app_addr_in_cache_shared_heap);
        LLVMMoveBasicBlockAfter(*app_addr_in_linear_mem,
                                *app_addr_in_cache_shared_heap);
    }

    return true;
fail:
    return false;
}

/*
 * Build a branch to check if start_offset is in the shared heap chain region.
 *
 * Arguments:
 *   start_offset: The offset to check.
 *   app_addr_in_shared_heap_chain: Block to branch if in shared heap chain.
 *   app_addr_in_linear_mem: Block to branch if not in shared heap chain.
 */
static bool
build_check_app_addr_in_shared_heap_chain(
    AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
    LLVMValueRef start_offset, LLVMBasicBlockRef app_addr_in_shared_heap_chain,
    LLVMBasicBlockRef app_addr_in_linear_mem)
{
    LLVMValueRef is_in_shared_heap = NULL;

    /* Use start_offset > func_ctx->shared_heap_head_start_off to test
     * start_off falls in shared heap chain memory region. The shared heap
     * chain oob will be detected in app_addr_in_shared_heap block or
     * aot_check_shared_heap_chain_and_update function
     */
    BUILD_ICMP(LLVMIntUGT, start_offset, func_ctx->shared_heap_head_start_off,
               is_in_shared_heap, "shared_heap_lb_cmp");
    BUILD_COND_BR(is_in_shared_heap, app_addr_in_shared_heap_chain,
                  app_addr_in_linear_mem);

    SET_BUILD_POS(app_addr_in_shared_heap_chain);

    return true;
fail:
    return false;
}

/*
 * Build the conditional branch for cache shared heap or shared heap chain.
 *
 * Arguments:
 *   cmp: The condition for being in cache shared heap.
 *   app_addr_in_cache_shared_heap: Block for cache shared heap.
 *   app_addr_in_linear_mem: Block for linear memory.
 *   check_shared_heap_chain: Block for checking shared heap chain.
 *   bytes: The access size in bytes.
 *   start_offset: The offset to check.
 *   is_memory64: Whether memory is 64-bit.
 */
static bool
build_shared_heap_conditional_branching(
    AOTCompContext *comp_ctx, AOTFuncContext *func_ctx, LLVMValueRef cmp,
    LLVMBasicBlockRef app_addr_in_cache_shared_heap,
    LLVMBasicBlockRef app_addr_in_linear_mem,
    LLVMBasicBlockRef check_shared_heap_chain, LLVMValueRef bytes,
    LLVMValueRef start_offset, bool is_memory64)
{
    if (comp_ctx->enable_shared_heap) {
        BUILD_COND_BR(cmp, app_addr_in_cache_shared_heap,
                      app_addr_in_linear_mem);
    }
    else if (comp_ctx->enable_shared_chain) {
        BUILD_COND_BR(cmp, app_addr_in_cache_shared_heap,
                      check_shared_heap_chain);
        SET_BUILD_POS(check_shared_heap_chain);
        if (!aot_check_shared_heap_chain_and_update(
                comp_ctx, func_ctx, app_addr_in_cache_shared_heap, start_offset,
                bytes, is_memory64))
            goto fail;
    }
    return true;
fail:
    return false;
}

/*
 * Get the native address in the cache shared heap.
 *
 * Arguments:
 *   start_offset: The offset to use for address calculation.
 *   maddr: Output, the native address that in the cache shared heap.
 */
static bool
build_get_maddr_in_cache_shared_heap(AOTCompContext *comp_ctx,
                                     AOTFuncContext *func_ctx,
                                     LLVMValueRef start_offset,
                                     LLVMValueRef *maddr)
{
    LLVMValueRef shared_heap_base_addr_adj;
    /* load the local variable */
    BUILD_LOAD_PTR(func_ctx->shared_heap_base_addr_adj, INT8_PTR_TYPE,
                   shared_heap_base_addr_adj);
    if (!(*maddr = LLVMBuildInBoundsGEP2(
              comp_ctx->builder, INT8_TYPE, shared_heap_base_addr_adj,
              &start_offset, 1, "maddr_cache_shared_heap"))) {
        aot_set_last_error("llvm build inbounds gep failed");
        goto fail;
    }

    return true;
fail:
    return false;
}

/*
 * Check for memory overflow in shared heap for normal memory access.
 *
 * Arguments:
 *   block_curr: The current basic block.
 *   block_maddr_phi: The phi block for memory address.
 *   maddr_phi: The phi node for memory address.
 *   start_offset: The first offset to check.
 *   mem_base_addr: The base address of memory. Only used with segue.
 *   bytes_u32: The access size in bytes.
 *   is_memory64: Whether memory is wasm64 memory.
 *   is_target_64bit: Whether target is 64-bit.
 *   enable_segue: Whether to use segment register addressing.
 */
static bool
aot_check_shared_heap_memory_overflow(
    AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
    LLVMBasicBlockRef block_curr, LLVMBasicBlockRef block_maddr_phi,
    LLVMValueRef maddr_phi, LLVMValueRef start_offset,
    LLVMValueRef mem_base_addr, uint32 bytes_u32, bool is_memory64,
    bool is_target_64bit, bool enable_segue)
{
    LLVMBasicBlockRef app_addr_in_cache_shared_heap, app_addr_in_linear_mem;
    LLVMBasicBlockRef app_addr_in_shared_heap_chain = NULL,
                      check_shared_heap_chain = NULL;
    LLVMValueRef cmp, cmp1, cmp2, shared_heap_start_off, shared_heap_end_off,
        shared_heap_check_bound, maddr = NULL;
    /* On 64/32-bit target, the offset is 64/32-bit */
    LLVMTypeRef offset_type = is_target_64bit ? I64_TYPE : I32_TYPE;
    LLVMValueRef length, bytes;

    if (!setup_shared_heap_blocks(
            comp_ctx, func_ctx, block_curr, &app_addr_in_cache_shared_heap,
            &app_addr_in_linear_mem, &app_addr_in_shared_heap_chain,
            &check_shared_heap_chain))
        goto fail;
    LLVMMoveBasicBlockAfter(block_maddr_phi, app_addr_in_linear_mem);

    /* Early branching when it's not in shared heap chain at all */
    if (comp_ctx->enable_shared_chain
        && !build_check_app_addr_in_shared_heap_chain(
            comp_ctx, func_ctx, start_offset, app_addr_in_shared_heap_chain,
            app_addr_in_linear_mem))
        goto fail;

    /* Load the local variable of the function */
    BUILD_LOAD_PTR(func_ctx->shared_heap_start_off, offset_type,
                   shared_heap_start_off);
    BUILD_LOAD_PTR(func_ctx->shared_heap_end_off, offset_type,
                   shared_heap_end_off);
    /* Check if the app address is in the cache shared heap range.
     * If yes, branch to the cache branch; if not, check the shared heap chain
     */
    BUILD_ICMP(LLVMIntUGE, start_offset, shared_heap_start_off, cmp,
               "cmp_cache_shared_heap_start");
    length =
        is_target_64bit ? I64_CONST(bytes_u32 - 1) : I32_CONST(bytes_u32 - 1);
    CHECK_LLVM_CONST(length);
    BUILD_OP(Sub, shared_heap_end_off, length, shared_heap_check_bound,
             "cache_shared_heap_end_bound");
    BUILD_ICMP(LLVMIntULE, start_offset, shared_heap_check_bound, cmp1,
               "cmp_cache_shared_heap_end");
    BUILD_OP(And, cmp, cmp1, cmp2, "is_in_cache_shared_heap");
    /* Conditional branching based on whether in cached shared heap */
    bytes = is_target_64bit ? I64_CONST(bytes_u32) : I32_CONST(bytes_u32);
    if (!build_shared_heap_conditional_branching(
            comp_ctx, func_ctx, cmp2, app_addr_in_cache_shared_heap,
            app_addr_in_linear_mem, check_shared_heap_chain, bytes,
            start_offset, is_memory64))
        goto fail;

    SET_BUILD_POS(app_addr_in_cache_shared_heap);
    if (!build_get_maddr_in_cache_shared_heap(comp_ctx, func_ctx, start_offset,
                                              &maddr))
        goto fail;

    if (enable_segue) {
        LLVMValueRef mem_base_addr_u64, maddr_u64, offset_to_mem_base;
        if (!(maddr_u64 = LLVMBuildPtrToInt(comp_ctx->builder, maddr, I64_TYPE,
                                            "maddr_u64"))
            || !(mem_base_addr_u64 =
                     LLVMBuildPtrToInt(comp_ctx->builder, mem_base_addr,
                                       I64_TYPE, "mem_base_addr_u64"))) {
            aot_set_last_error("llvm build ptr to int failed");
            goto fail;
        }
        if (!(offset_to_mem_base =
                  LLVMBuildSub(comp_ctx->builder, maddr_u64, mem_base_addr_u64,
                               "offset_to_mem_base"))) {
            aot_set_last_error("llvm build sub failed");
            goto fail;
        }
        if (!(maddr = LLVMBuildIntToPtr(comp_ctx->builder, offset_to_mem_base,
                                        INT8_PTR_TYPE_GS,
                                        "maddr_shared_heap_segue"))) {
            aot_set_last_error("llvm build int to ptr failed.");
            goto fail;
        }
    }

    LLVMAddIncoming(maddr_phi, &maddr, &app_addr_in_cache_shared_heap, 1);
    BUILD_BR(block_maddr_phi);
    SET_BUILD_POS(app_addr_in_linear_mem);

    return true;
fail:
    return false;
}

/*
 * Check for memory overflow in shared heap for bulk memory access.
 *
 * Arguments:
 *   block_curr: The current basic block.
 *   block_maddr_phi: The phi block for memory address.
 *   check_succ: The block to branch to on success.
 *   maddr_phi: The phi node for memory address.
 *   start_offset: The offset to check.
 *   max_addr: The maximum address to check.
 *   bytes: The access size in bytes (LLVMValueRef).
 *   is_memory64: Whether memory is wasm64 memory.
 *   is_target_64bit: Whether target is 64-bit.
 */
static bool
aot_check_bulk_memory_shared_heap_memory_overflow(
    AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
    LLVMBasicBlockRef block_curr, LLVMBasicBlockRef block_maddr_phi,
    LLVMBasicBlockRef check_succ, LLVMValueRef maddr_phi,
    LLVMValueRef start_offset, LLVMValueRef max_addr, LLVMValueRef bytes,
    bool is_memory64, bool is_target_64bit)
{
    LLVMBasicBlockRef app_addr_in_cache_shared_heap, app_addr_in_linear_mem;
    LLVMBasicBlockRef app_addr_in_shared_heap_chain = NULL,
                      check_shared_heap_chain = NULL;
    LLVMValueRef cmp, cmp1, cmp2, shared_heap_start_off, shared_heap_end_off,
        maddr = NULL, max_offset;
    /* On 64/32-bit target, the offset is 64/32-bit */
    LLVMTypeRef offset_type = is_target_64bit ? I64_TYPE : I32_TYPE;

    if (!setup_shared_heap_blocks(
            comp_ctx, func_ctx, block_curr, &app_addr_in_cache_shared_heap,
            &app_addr_in_linear_mem, &app_addr_in_shared_heap_chain,
            &check_shared_heap_chain))
        goto fail;
    LLVMMoveBasicBlockAfter(block_maddr_phi, check_succ);

    /* Early branching when it's not in shared heap chain at all */
    if (comp_ctx->enable_shared_chain
        && !build_check_app_addr_in_shared_heap_chain(
            comp_ctx, func_ctx, start_offset, app_addr_in_shared_heap_chain,
            app_addr_in_linear_mem))
        goto fail;

    /* Load the local variable of the function */
    BUILD_LOAD_PTR(func_ctx->shared_heap_start_off, offset_type,
                   shared_heap_start_off);
    BUILD_LOAD_PTR(func_ctx->shared_heap_end_off, offset_type,
                   shared_heap_end_off);
    /* Check if the app address is in the cache shared heap range.
     * If yes, branch to the cache branch; if not, check the shared heap chain
     */
    BUILD_ICMP(LLVMIntUGE, start_offset, shared_heap_start_off, cmp,
               "cmp_cache_shared_heap_start");
    BUILD_OP(Add, max_addr, is_target_64bit ? I64_NEG_ONE : I32_NEG_ONE,
             max_offset, "max_offset");
    BUILD_ICMP(LLVMIntULE, max_offset, shared_heap_end_off, cmp1,
               "cmp_cache_shared_heap_end");
    BUILD_OP(And, cmp, cmp1, cmp2, "is_in_cache_shared_heap");
    /* Conditional branching based on whether in cached shared heap */
    if (!build_shared_heap_conditional_branching(
            comp_ctx, func_ctx, cmp2, app_addr_in_cache_shared_heap,
            app_addr_in_linear_mem, check_shared_heap_chain, bytes,
            start_offset, is_memory64))
        goto fail;

    SET_BUILD_POS(app_addr_in_cache_shared_heap);
    if (!build_get_maddr_in_cache_shared_heap(comp_ctx, func_ctx, start_offset,
                                              &maddr))
        goto fail;

    LLVMAddIncoming(maddr_phi, &maddr, &app_addr_in_cache_shared_heap, 1);
    BUILD_BR(block_maddr_phi);
    SET_BUILD_POS(app_addr_in_linear_mem);

    return true;
fail:
    return false;
}
#endif

LLVMValueRef
aot_check_memory_overflow(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                          mem_offset_t offset, uint32 bytes, bool enable_segue,
                          unsigned int *alignp)
{
    LLVMValueRef offset_const =
        MEMORY64_COND_VALUE(I64_CONST(offset), I32_CONST(offset));
    LLVMValueRef addr, maddr, offset1, cmp1, cmp;
    LLVMValueRef mem_base_addr, mem_check_bound;
    LLVMBasicBlockRef block_curr = LLVMGetInsertBlock(comp_ctx->builder);
    LLVMBasicBlockRef check_succ;
    AOTValue *aot_value_top;
    uint32 local_idx_of_aot_value = 0;
    uint64 const_value;
    bool is_target_64bit, is_local_of_aot_value = false;
    bool is_const = false;
#if WASM_ENABLE_SHARED_MEMORY != 0
    bool is_shared_memory =
        comp_ctx->comp_data->memories[0].flags & SHARED_MEMORY_FLAG;
#endif
#if WASM_ENABLE_MEMORY64 == 0
    bool is_memory64 = false;
#else
    bool is_memory64 = IS_MEMORY64;
#endif
#if WASM_ENABLE_SHARED_HEAP != 0
    LLVMValueRef maddr_phi = NULL;
    LLVMBasicBlockRef block_maddr_phi = NULL;
#endif

    is_target_64bit = (comp_ctx->pointer_size == sizeof(uint64)) ? true : false;

    if (comp_ctx->is_indirect_mode
        && aot_intrinsic_check_capability(
            comp_ctx, MEMORY64_COND_VALUE("i64.const", "i32.const"))) {
        WASMValue wasm_value;
#if WASM_ENABLE_MEMORY64 != 0
        if (IS_MEMORY64) {
            wasm_value.i64 = offset;
        }
        else
#endif
        {
            wasm_value.i32 = (int32)offset;
        }
        offset_const = aot_load_const_from_table(
            comp_ctx, func_ctx->native_symbol, &wasm_value,
            MEMORY64_COND_VALUE(VALUE_TYPE_I64, VALUE_TYPE_I32));
        if (!offset_const) {
            return NULL;
        }
    }
    else {
        CHECK_LLVM_CONST(offset_const);
    }

    /* Get memory base address and memory data size */
    if (func_ctx->mem_space_unchanged
#if WASM_ENABLE_SHARED_MEMORY != 0
        || is_shared_memory
#endif
    ) {
        mem_base_addr = func_ctx->mem_info[0].mem_base_addr;
    }
    else {
        if (!(mem_base_addr = LLVMBuildLoad2(
                  comp_ctx->builder, OPQ_PTR_TYPE,
                  func_ctx->mem_info[0].mem_base_addr, "mem_base"))) {
            aot_set_last_error("llvm build load failed.");
            goto fail;
        }
    }

    aot_value_top =
        func_ctx->block_stack.block_list_end->value_stack.value_list_end;
    if (aot_value_top) {
        /* aot_value_top is freed in the following POP_I32(addr),
           so save its fields here for further use */
        is_local_of_aot_value = aot_value_top->is_local;
        is_const = aot_value_top->is_const;
        local_idx_of_aot_value = aot_value_top->local_idx;
        const_value = aot_value_top->const_value;
    }

    POP_MEM_OFFSET(addr);

    /*
     * Note: not throw the integer-overflow-exception here since it must
     * have been thrown when converting float to integer before
     */
    /* return address directly if constant offset and inside memory space */
    if (LLVMIsEfficientConstInt(addr) || is_const) {
        uint64 value;
        if (LLVMIsEfficientConstInt(addr)) {
            value = (uint64)LLVMConstIntGetZExtValue(addr);
        }
        else {
            value = const_value;
        }
        uint64 mem_offset = value + (uint64)offset;
        uint32 num_bytes_per_page =
            comp_ctx->comp_data->memories[0].num_bytes_per_page;
        uint32 init_page_count =
            comp_ctx->comp_data->memories[0].init_page_count;
        uint64 mem_data_size = (uint64)num_bytes_per_page * init_page_count;

        if (alignp != NULL) {
            /*
             * A note about max_align below:
             * the assumption here is the base address of a linear memory
             * has the natural alignment. for platforms using mmap, it can
             * be even larger. for now, use a conservative value.
             */
            const unsigned int max_align = 8;
            int shift = ffs((int)(unsigned int)mem_offset);
            if (shift == 0) {
                *alignp = max_align;
            }
            else {
                unsigned int align = 1 << (shift - 1);
                if (align > max_align) {
                    align = max_align;
                }
                *alignp = align;
            }
        }
        if (mem_offset + bytes <= mem_data_size) {
            /* inside memory space */
            if (comp_ctx->pointer_size == sizeof(uint64))
                offset1 = I64_CONST(mem_offset);
            else
                offset1 = I32_CONST((uint32)mem_offset);
            CHECK_LLVM_CONST(offset1);
            if (!enable_segue) {
                if (!(maddr = LLVMBuildInBoundsGEP2(comp_ctx->builder,
                                                    INT8_TYPE, mem_base_addr,
                                                    &offset1, 1, "maddr"))) {
                    aot_set_last_error("llvm build add failed.");
                    goto fail;
                }
            }
            else {
                if (!(maddr = LLVMBuildIntToPtr(comp_ctx->builder, offset1,
                                                INT8_PTR_TYPE_GS, "maddr"))) {
                    aot_set_last_error("llvm build IntToPtr failed.");
                    goto fail;
                }
            }
            return maddr;
        }
    }
    else if (alignp != NULL) {
        *alignp = 1;
    }

    /* The overflow check needs to be done under following conditions:
     * 1. In 64-bit target, offset and addr will be extended to 64-bit
     *    1.1 offset + addr can overflow when it's memory64
     *    1.2 no overflow when it's memory32
     * 2. In 32-bit target, offset and addr will be 32-bit
     *    2.1 offset + addr can overflow when it's memory32
     */
    if (is_target_64bit) {
        if (!(offset_const = LLVMBuildZExt(comp_ctx->builder, offset_const,
                                           I64_TYPE, "offset_i64"))
            || !(addr = LLVMBuildZExt(comp_ctx->builder, addr, I64_TYPE,
                                      "addr_i64"))) {
            aot_set_last_error("llvm build zero extend failed.");
            goto fail;
        }
    }

    /* offset1 = offset + addr; */
    BUILD_OP(Add, offset_const, addr, offset1, "offset1");

    /* 1.1 offset + addr can overflow when it's memory64
     * 2.1 Or when it's on 32-bit platform */
    if (is_memory64 || !is_target_64bit) {
        /* Check whether integer overflow occurs in offset + addr */
        LLVMBasicBlockRef check_integer_overflow_end;
        ADD_BASIC_BLOCK(check_integer_overflow_end,
                        "check_integer_overflow_end");
        LLVMMoveBasicBlockAfter(check_integer_overflow_end, block_curr);

        BUILD_ICMP(LLVMIntULT, offset1, offset_const, cmp1, "cmp1");
        if (!aot_emit_exception(comp_ctx, func_ctx,
                                EXCE_OUT_OF_BOUNDS_MEMORY_ACCESS, true, cmp1,
                                check_integer_overflow_end)) {
            goto fail;
        }
        SET_BUILD_POS(check_integer_overflow_end);
        block_curr = check_integer_overflow_end;
    }

#if WASM_ENABLE_SHARED_HEAP != 0
    if (comp_ctx->enable_shared_heap
        || comp_ctx->enable_shared_chain /* TODO: && mem_idx == 0 */) {
        ADD_BASIC_BLOCK(block_maddr_phi, "maddr_phi");
        SET_BUILD_POS(block_maddr_phi);
        if (!(maddr_phi =
                  LLVMBuildPhi(comp_ctx->builder,
                               enable_segue ? INT8_PTR_TYPE_GS : INT8_PTR_TYPE,
                               "maddr_phi"))) {
            aot_set_last_error("llvm build phi failed");
            goto fail;
        }
        SET_BUILD_POS(block_curr);

        if (!aot_check_shared_heap_memory_overflow(
                comp_ctx, func_ctx, block_curr, block_maddr_phi, maddr_phi,
                offset1, mem_base_addr, bytes, is_memory64, is_target_64bit,
                enable_segue)) {
            goto fail;
        }
    }
#endif

    if (comp_ctx->enable_bound_check
        && !(is_local_of_aot_value
             && aot_checked_addr_list_find(func_ctx, local_idx_of_aot_value,
                                           offset, bytes))) {
        uint32 init_page_count =
            comp_ctx->comp_data->memories[0].init_page_count;
        if (init_page_count == 0) {
            LLVMValueRef mem_size;

            if (!(mem_size = get_memory_curr_page_count(comp_ctx, func_ctx))) {
                goto fail;
            }
            BUILD_ICMP(LLVMIntEQ, mem_size,
                       MEMORY64_COND_VALUE(I64_ZERO, I32_ZERO), cmp, "is_zero");
            ADD_BASIC_BLOCK(check_succ, "check_mem_size_succ");
            LLVMMoveBasicBlockAfter(check_succ, block_curr);
            if (!aot_emit_exception(comp_ctx, func_ctx,
                                    EXCE_OUT_OF_BOUNDS_MEMORY_ACCESS, true, cmp,
                                    check_succ)) {
                goto fail;
            }

            SET_BUILD_POS(check_succ);
            block_curr = check_succ;
        }

        if (!(mem_check_bound =
                  get_memory_check_bound(comp_ctx, func_ctx, bytes))) {
            goto fail;
        }

        BUILD_ICMP(LLVMIntUGT, offset1, mem_check_bound, cmp, "cmp");

        /* Add basic blocks */
        ADD_BASIC_BLOCK(check_succ, "check_succ");
        LLVMMoveBasicBlockAfter(check_succ, block_curr);

        if (!aot_emit_exception(comp_ctx, func_ctx,
                                EXCE_OUT_OF_BOUNDS_MEMORY_ACCESS, true, cmp,
                                check_succ)) {
            goto fail;
        }

        SET_BUILD_POS(check_succ);

        if (is_local_of_aot_value) {
            if (!aot_checked_addr_list_add(func_ctx, local_idx_of_aot_value,
                                           offset, bytes))
                goto fail;
        }
    }

    if (!enable_segue) {
        /* maddr = mem_base_addr + offset1 */
        if (!(maddr =
                  LLVMBuildInBoundsGEP2(comp_ctx->builder, INT8_TYPE,
                                        mem_base_addr, &offset1, 1, "maddr"))) {
            aot_set_last_error("llvm build add failed.");
            goto fail;
        }
    }
    else {
        LLVMValueRef maddr_base;

        if (!(maddr_base = LLVMBuildIntToPtr(comp_ctx->builder, addr,
                                             INT8_PTR_TYPE_GS, "maddr_base"))) {
            aot_set_last_error("llvm build int to ptr failed.");
            goto fail;
        }
        if (!(maddr = LLVMBuildInBoundsGEP2(comp_ctx->builder, INT8_TYPE,
                                            maddr_base, &offset_const, 1,
                                            "maddr"))) {
            aot_set_last_error("llvm build inboundgep failed.");
            goto fail;
        }
    }

#if WASM_ENABLE_SHARED_HEAP != 0
    if (comp_ctx->enable_shared_heap
        || comp_ctx->enable_shared_chain /* TODO: && mem_idx == 0 */) {
        block_curr = LLVMGetInsertBlock(comp_ctx->builder);
        LLVMAddIncoming(maddr_phi, &maddr, &block_curr, 1);
        if (!LLVMBuildBr(comp_ctx->builder, block_maddr_phi)) {
            aot_set_last_error("llvm build br failed");
            goto fail;
        }
        SET_BUILD_POS(block_maddr_phi);
        return maddr_phi;
    }
    else
#endif
        return maddr;
fail:
    return NULL;
}

#define BUILD_PTR_CAST(ptr_type)                                           \
    do {                                                                   \
        if (!(maddr = LLVMBuildBitCast(comp_ctx->builder, maddr, ptr_type, \
                                       "data_ptr"))) {                     \
            aot_set_last_error("llvm build bit cast failed.");             \
            goto fail;                                                     \
        }                                                                  \
    } while (0)

#define BUILD_LOAD(data_type)                                             \
    do {                                                                  \
        if (!(value = LLVMBuildLoad2(comp_ctx->builder, data_type, maddr, \
                                     "data"))) {                          \
            aot_set_last_error("llvm build load failed.");                \
            goto fail;                                                    \
        }                                                                 \
        LLVMSetAlignment(value, known_align);                             \
    } while (0)

#define BUILD_STORE()                                                   \
    do {                                                                \
        LLVMValueRef res;                                               \
        if (!(res = LLVMBuildStore(comp_ctx->builder, value, maddr))) { \
            aot_set_last_error("llvm build store failed.");             \
            goto fail;                                                  \
        }                                                               \
        LLVMSetAlignment(res, known_align);                             \
    } while (0)

#define BUILD_SIGN_EXT(dst_type)                                        \
    do {                                                                \
        if (!(value = LLVMBuildSExt(comp_ctx->builder, value, dst_type, \
                                    "data_s_ext"))) {                   \
            aot_set_last_error("llvm build sign ext failed.");          \
            goto fail;                                                  \
        }                                                               \
    } while (0)

#define BUILD_ZERO_EXT(dst_type)                                        \
    do {                                                                \
        if (!(value = LLVMBuildZExt(comp_ctx->builder, value, dst_type, \
                                    "data_z_ext"))) {                   \
            aot_set_last_error("llvm build zero ext failed.");          \
            goto fail;                                                  \
        }                                                               \
    } while (0)

#if WASM_ENABLE_SHARED_MEMORY != 0 || WASM_ENABLE_STRINGREF != 0
bool
check_memory_alignment(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                       LLVMValueRef addr, uint32 align)
{
    LLVMBasicBlockRef block_curr = LLVMGetInsertBlock(comp_ctx->builder);
    LLVMBasicBlockRef check_align_succ;
    LLVMValueRef align_mask = I32_CONST(((uint32)1 << align) - 1);
    LLVMValueRef res;

    CHECK_LLVM_CONST(align_mask);

    /* Convert pointer to int */
    if (!(addr = LLVMBuildPtrToInt(comp_ctx->builder, addr, I32_TYPE,
                                   "address"))) {
        aot_set_last_error("llvm build ptr to int failed.");
        goto fail;
    }

    /* The memory address should be aligned */
    BUILD_OP(And, addr, align_mask, res, "and");
    BUILD_ICMP(LLVMIntNE, res, I32_ZERO, res, "cmp");

    /* Add basic blocks */
    ADD_BASIC_BLOCK(check_align_succ, "check_align_succ");
    LLVMMoveBasicBlockAfter(check_align_succ, block_curr);

    if (!aot_emit_exception(comp_ctx, func_ctx, EXCE_UNALIGNED_ATOMIC, true,
                            res, check_align_succ)) {
        goto fail;
    }

    SET_BUILD_POS(check_align_succ);

    return true;
fail:
    return false;
}
#endif /* WASM_ENABLE_SHARED_MEMORY != 0 || WASM_ENABLE_STRINGREF != 0 */

#if WASM_ENABLE_SHARED_MEMORY != 0
#define BUILD_ATOMIC_LOAD(align, data_type)                                \
    do {                                                                   \
        if (!(check_memory_alignment(comp_ctx, func_ctx, maddr, align))) { \
            goto fail;                                                     \
        }                                                                  \
        if (!(value = LLVMBuildLoad2(comp_ctx->builder, data_type, maddr,  \
                                     "data"))) {                           \
            aot_set_last_error("llvm build load failed.");                 \
            goto fail;                                                     \
        }                                                                  \
        LLVMSetAlignment(value, 1 << align);                               \
        LLVMSetVolatile(value, true);                                      \
        LLVMSetOrdering(value, LLVMAtomicOrderingSequentiallyConsistent);  \
    } while (0)

#define BUILD_ATOMIC_STORE(align)                                          \
    do {                                                                   \
        LLVMValueRef res;                                                  \
        if (!(check_memory_alignment(comp_ctx, func_ctx, maddr, align))) { \
            goto fail;                                                     \
        }                                                                  \
        if (!(res = LLVMBuildStore(comp_ctx->builder, value, maddr))) {    \
            aot_set_last_error("llvm build store failed.");                \
            goto fail;                                                     \
        }                                                                  \
        LLVMSetAlignment(res, 1 << align);                                 \
        LLVMSetVolatile(res, true);                                        \
        LLVMSetOrdering(res, LLVMAtomicOrderingSequentiallyConsistent);    \
    } while (0)
#endif

bool
aot_compile_op_i32_load(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                        uint32 align, mem_offset_t offset, uint32 bytes,
                        bool sign, bool atomic)
{
    LLVMValueRef maddr, value = NULL;
    LLVMTypeRef data_type;
    bool enable_segue = comp_ctx->enable_segue_i32_load;

    unsigned int known_align;
    if (!(maddr = aot_check_memory_overflow(comp_ctx, func_ctx, offset, bytes,
                                            enable_segue, &known_align)))
        return false;

    switch (bytes) {
        case 4:
            if (!enable_segue)
                BUILD_PTR_CAST(INT32_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT32_PTR_TYPE_GS);
#if WASM_ENABLE_SHARED_MEMORY != 0
            if (atomic)
                BUILD_ATOMIC_LOAD(align, I32_TYPE);
            else
#endif
                BUILD_LOAD(I32_TYPE);
            break;
        case 2:
        case 1:
            if (bytes == 2) {
                if (!enable_segue)
                    BUILD_PTR_CAST(INT16_PTR_TYPE);
                else
                    BUILD_PTR_CAST(INT16_PTR_TYPE_GS);
                data_type = INT16_TYPE;
            }
            else {
                if (!enable_segue)
                    BUILD_PTR_CAST(INT8_PTR_TYPE);
                else
                    BUILD_PTR_CAST(INT8_PTR_TYPE_GS);
                data_type = INT8_TYPE;
            }

#if WASM_ENABLE_SHARED_MEMORY != 0
            if (atomic) {
                BUILD_ATOMIC_LOAD(align, data_type);
                BUILD_ZERO_EXT(I32_TYPE);
            }
            else
#endif
            {
                BUILD_LOAD(data_type);
                if (sign)
                    BUILD_SIGN_EXT(I32_TYPE);
                else
                    BUILD_ZERO_EXT(I32_TYPE);
            }
            break;
        default:
            bh_assert(0);
            break;
    }

    PUSH_I32(value);
    (void)data_type;
    return true;
fail:
    return false;
}

bool
aot_compile_op_i64_load(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                        uint32 align, mem_offset_t offset, uint32 bytes,
                        bool sign, bool atomic)
{
    LLVMValueRef maddr, value = NULL;
    LLVMTypeRef data_type;
    bool enable_segue = comp_ctx->enable_segue_i64_load;

    unsigned int known_align;
    if (!(maddr = aot_check_memory_overflow(comp_ctx, func_ctx, offset, bytes,
                                            enable_segue, &known_align)))
        return false;

    switch (bytes) {
        case 8:
            if (!enable_segue)
                BUILD_PTR_CAST(INT64_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT64_PTR_TYPE_GS);
#if WASM_ENABLE_SHARED_MEMORY != 0
            if (atomic)
                BUILD_ATOMIC_LOAD(align, I64_TYPE);
            else
#endif
                BUILD_LOAD(I64_TYPE);
            break;
        case 4:
        case 2:
        case 1:
            if (bytes == 4) {
                if (!enable_segue)
                    BUILD_PTR_CAST(INT32_PTR_TYPE);
                else
                    BUILD_PTR_CAST(INT32_PTR_TYPE_GS);
                data_type = I32_TYPE;
            }
            else if (bytes == 2) {
                if (!enable_segue)
                    BUILD_PTR_CAST(INT16_PTR_TYPE);
                else
                    BUILD_PTR_CAST(INT16_PTR_TYPE_GS);
                data_type = INT16_TYPE;
            }
            else {
                if (!enable_segue)
                    BUILD_PTR_CAST(INT8_PTR_TYPE);
                else
                    BUILD_PTR_CAST(INT8_PTR_TYPE_GS);
                data_type = INT8_TYPE;
            }

#if WASM_ENABLE_SHARED_MEMORY != 0
            if (atomic) {
                BUILD_ATOMIC_LOAD(align, data_type);
                BUILD_ZERO_EXT(I64_TYPE);
            }
            else
#endif
            {
                BUILD_LOAD(data_type);
                if (sign)
                    BUILD_SIGN_EXT(I64_TYPE);
                else
                    BUILD_ZERO_EXT(I64_TYPE);
            }
            break;
        default:
            bh_assert(0);
            break;
    }

    PUSH_I64(value);
    (void)data_type;
    return true;
fail:
    return false;
}

bool
aot_compile_op_f32_load(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                        uint32 align, mem_offset_t offset)
{
    LLVMValueRef maddr, value;
    bool enable_segue = comp_ctx->enable_segue_f32_load;

    unsigned int known_align;
    if (!(maddr = aot_check_memory_overflow(comp_ctx, func_ctx, offset, 4,
                                            enable_segue, &known_align)))
        return false;

    if (!enable_segue)
        BUILD_PTR_CAST(F32_PTR_TYPE);
    else
        BUILD_PTR_CAST(F32_PTR_TYPE_GS);
    BUILD_LOAD(F32_TYPE);

    PUSH_F32(value);
    return true;
fail:
    return false;
}

bool
aot_compile_op_f64_load(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                        uint32 align, mem_offset_t offset)
{
    LLVMValueRef maddr, value;
    bool enable_segue = comp_ctx->enable_segue_f64_load;

    unsigned int known_align;
    if (!(maddr = aot_check_memory_overflow(comp_ctx, func_ctx, offset, 8,
                                            enable_segue, &known_align)))
        return false;

    if (!enable_segue)
        BUILD_PTR_CAST(F64_PTR_TYPE);
    else
        BUILD_PTR_CAST(F64_PTR_TYPE_GS);
    BUILD_LOAD(F64_TYPE);

    PUSH_F64(value);
    return true;
fail:
    return false;
}

bool
aot_compile_op_i32_store(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                         uint32 align, mem_offset_t offset, uint32 bytes,
                         bool atomic)
{
    LLVMValueRef maddr, value;
    bool enable_segue = comp_ctx->enable_segue_i32_store;

    POP_I32(value);

    unsigned int known_align;
    if (!(maddr = aot_check_memory_overflow(comp_ctx, func_ctx, offset, bytes,
                                            enable_segue, &known_align)))
        return false;

    switch (bytes) {
        case 4:
            if (!enable_segue)
                BUILD_PTR_CAST(INT32_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT32_PTR_TYPE_GS);
            break;
        case 2:
            if (!enable_segue)
                BUILD_PTR_CAST(INT16_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT16_PTR_TYPE_GS);
            BUILD_TRUNC(value, INT16_TYPE);
            break;
        case 1:
            if (!enable_segue)
                BUILD_PTR_CAST(INT8_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT8_PTR_TYPE_GS);
            BUILD_TRUNC(value, INT8_TYPE);
            break;
        default:
            bh_assert(0);
            break;
    }

#if WASM_ENABLE_SHARED_MEMORY != 0
    if (atomic)
        BUILD_ATOMIC_STORE(align);
    else
#endif
        BUILD_STORE();
    return true;
fail:
    return false;
}

bool
aot_compile_op_i64_store(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                         uint32 align, mem_offset_t offset, uint32 bytes,
                         bool atomic)
{
    LLVMValueRef maddr, value;
    bool enable_segue = comp_ctx->enable_segue_i64_store;

    POP_I64(value);

    unsigned int known_align;
    if (!(maddr = aot_check_memory_overflow(comp_ctx, func_ctx, offset, bytes,
                                            enable_segue, &known_align)))
        return false;

    switch (bytes) {
        case 8:
            if (!enable_segue)
                BUILD_PTR_CAST(INT64_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT64_PTR_TYPE_GS);
            break;
        case 4:
            if (!enable_segue)
                BUILD_PTR_CAST(INT32_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT32_PTR_TYPE_GS);
            BUILD_TRUNC(value, I32_TYPE);
            break;
        case 2:
            if (!enable_segue)
                BUILD_PTR_CAST(INT16_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT16_PTR_TYPE_GS);
            BUILD_TRUNC(value, INT16_TYPE);
            break;
        case 1:
            if (!enable_segue)
                BUILD_PTR_CAST(INT8_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT8_PTR_TYPE_GS);
            BUILD_TRUNC(value, INT8_TYPE);
            break;
        default:
            bh_assert(0);
            break;
    }

#if WASM_ENABLE_SHARED_MEMORY != 0
    if (atomic)
        BUILD_ATOMIC_STORE(align);
    else
#endif
        BUILD_STORE();
    return true;
fail:
    return false;
}

bool
aot_compile_op_f32_store(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                         uint32 align, mem_offset_t offset)
{
    LLVMValueRef maddr, value;
    bool enable_segue = comp_ctx->enable_segue_f32_store;

    POP_F32(value);

    unsigned int known_align;
    if (!(maddr = aot_check_memory_overflow(comp_ctx, func_ctx, offset, 4,
                                            enable_segue, &known_align)))
        return false;

    if (!enable_segue)
        BUILD_PTR_CAST(F32_PTR_TYPE);
    else
        BUILD_PTR_CAST(F32_PTR_TYPE_GS);
    BUILD_STORE();
    return true;
fail:
    return false;
}

bool
aot_compile_op_f64_store(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                         uint32 align, mem_offset_t offset)
{
    LLVMValueRef maddr, value;
    bool enable_segue = comp_ctx->enable_segue_f64_store;

    POP_F64(value);

    unsigned int known_align;
    if (!(maddr = aot_check_memory_overflow(comp_ctx, func_ctx, offset, 8,
                                            enable_segue, &known_align)))
        return false;

    if (!enable_segue)
        BUILD_PTR_CAST(F64_PTR_TYPE);
    else
        BUILD_PTR_CAST(F64_PTR_TYPE_GS);
    BUILD_STORE();
    return true;
fail:
    return false;
}

static LLVMValueRef
get_memory_curr_page_count(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
    LLVMValueRef mem_size;

    if (func_ctx->mem_space_unchanged) {
        mem_size = func_ctx->mem_info[0].mem_cur_page_count_addr;
    }
    else {
        if (!(mem_size = LLVMBuildLoad2(
                  comp_ctx->builder, I32_TYPE,
                  func_ctx->mem_info[0].mem_cur_page_count_addr, "mem_size"))) {
            aot_set_last_error("llvm build load failed.");
            goto fail;
        }
    }

    return LLVMBuildIntCast(comp_ctx->builder, mem_size,
                            MEMORY64_COND_VALUE(I64_TYPE, I32_TYPE), "");
fail:
    return NULL;
}

bool
aot_compile_op_memory_size(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
    LLVMValueRef mem_size = get_memory_curr_page_count(comp_ctx, func_ctx);

    if (mem_size)
        PUSH_PAGE_COUNT(mem_size);
    return mem_size ? true : false;
fail:
    return false;
}

bool
aot_compile_op_memory_grow(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
    LLVMValueRef mem_size = get_memory_curr_page_count(comp_ctx, func_ctx);
    LLVMValueRef delta, param_values[2], ret_value, func, value;
    LLVMTypeRef param_types[2], ret_type, func_type, func_ptr_type;
    int32 func_index;
#if WASM_ENABLE_MEMORY64 != 0
    LLVMValueRef u32_max, u32_cmp_result;
#endif

    if (!mem_size)
        return false;

    POP_PAGE_COUNT(delta);

    /* TODO: multi-memory aot_enlarge_memory_with_idx() */
    /* Function type of aot_enlarge_memory() */
    param_types[0] = INT8_PTR_TYPE;
    param_types[1] = I32_TYPE;
    ret_type = INT8_TYPE;

    if (!(func_type = LLVMFunctionType(ret_type, param_types, 2, false))) {
        aot_set_last_error("llvm add function type failed.");
        return false;
    }

    if (comp_ctx->is_jit_mode) {
        /* JIT mode, call the function directly */
        if (!(func_ptr_type = LLVMPointerType(func_type, 0))) {
            aot_set_last_error("llvm add pointer type failed.");
            return false;
        }
        if (!(value = I64_CONST((uint64)(uintptr_t)wasm_enlarge_memory))
            || !(func = LLVMConstIntToPtr(value, func_ptr_type))) {
            aot_set_last_error("create LLVM value failed.");
            return false;
        }
    }
    else if (comp_ctx->is_indirect_mode) {
        if (!(func_ptr_type = LLVMPointerType(func_type, 0))) {
            aot_set_last_error("create LLVM function type failed.");
            return false;
        }
        func_index =
            aot_get_native_symbol_index(comp_ctx, "aot_enlarge_memory");
        if (func_index < 0) {
            return false;
        }
        if (!(func = aot_get_func_from_table(comp_ctx, func_ctx->native_symbol,
                                             func_ptr_type, func_index))) {
            return false;
        }
    }
    else {
        char *func_name = "aot_enlarge_memory";
        /* AOT mode, declare the function */
        if (!(func = LLVMGetNamedFunction(func_ctx->module, func_name))
            && !(func =
                     LLVMAddFunction(func_ctx->module, func_name, func_type))) {
            aot_set_last_error("llvm add function failed.");
            return false;
        }
    }

    /* Call function aot_enlarge_memory() */
    param_values[0] = func_ctx->aot_inst;
    param_values[1] = LLVMBuildTrunc(comp_ctx->builder, delta, I32_TYPE, "");
    if (!(ret_value = LLVMBuildCall2(comp_ctx->builder, func_type, func,
                                     param_values, 2, "call"))) {
        aot_set_last_error("llvm build call failed.");
        return false;
    }

    BUILD_ICMP(LLVMIntUGT, ret_value, I8_ZERO, ret_value, "mem_grow_ret");
#if WASM_ENABLE_MEMORY64 != 0
    if (IS_MEMORY64) {
        if (!(u32_max = I64_CONST(UINT32_MAX))) {
            aot_set_last_error("llvm build const failed");
            return false;
        }
        BUILD_ICMP(LLVMIntULE, delta, u32_max, u32_cmp_result, "page_size_cmp");
        BUILD_OP(And, ret_value, u32_cmp_result, ret_value, "and");
    }
#endif

    /* ret_value = ret_value == true ? pre_page_count : -1 */
    if (!(ret_value = LLVMBuildSelect(
              comp_ctx->builder, ret_value, mem_size,
              MEMORY64_COND_VALUE(I64_NEG_ONE, I32_NEG_ONE), "mem_grow_ret"))) {
        aot_set_last_error("llvm build select failed.");
        return false;
    }

    PUSH_PAGE_COUNT(ret_value);
    return true;
fail:
    return false;
}

#if WASM_ENABLE_BULK_MEMORY_OPT != 0 || WASM_ENABLE_STRINGREF != 0
LLVMValueRef
check_bulk_memory_overflow(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                           LLVMValueRef offset, LLVMValueRef bytes)
{
    LLVMValueRef maddr, max_addr, cmp, cmp1, offset1;
    LLVMValueRef mem_base_addr;
    LLVMBasicBlockRef block_curr = LLVMGetInsertBlock(comp_ctx->builder);
    LLVMBasicBlockRef check_succ;
    LLVMValueRef mem_size;
    bool is_target_64bit;
#if WASM_ENABLE_MEMORY64 == 0
    bool is_memory64 = false;
#else
    bool is_memory64 = IS_MEMORY64;
#endif
#if WASM_ENABLE_SHARED_HEAP != 0
    LLVMValueRef maddr_phi = NULL;
    LLVMBasicBlockRef block_maddr_phi = NULL;
#endif

    is_target_64bit = (comp_ctx->pointer_size == sizeof(uint64)) ? true : false;

    /* Get memory base address and memory data size */
#if WASM_ENABLE_SHARED_MEMORY != 0
    bool is_shared_memory = comp_ctx->comp_data->memories[0].flags & 0x02;

    if (func_ctx->mem_space_unchanged || is_shared_memory) {
#else
    if (func_ctx->mem_space_unchanged) {
#endif
        mem_base_addr = func_ctx->mem_info[0].mem_base_addr;
    }
    else {
        if (!(mem_base_addr = LLVMBuildLoad2(
                  comp_ctx->builder, OPQ_PTR_TYPE,
                  func_ctx->mem_info[0].mem_base_addr, "mem_base"))) {
            aot_set_last_error("llvm build load failed.");
            goto fail;
        }
    }

    /*
     * Note: not throw the integer-overflow-exception here since it must
     * have been thrown when converting float to integer before
     */
    /* return address directly if constant offset and inside memory space */
    if (LLVMIsEfficientConstInt(offset) && LLVMIsEfficientConstInt(bytes)) {
        uint64 mem_offset = (uint64)LLVMConstIntGetZExtValue(offset);
        uint64 mem_len = (uint64)LLVMConstIntGetZExtValue(bytes);
        uint32 num_bytes_per_page =
            comp_ctx->comp_data->memories[0].num_bytes_per_page;
        uint32 init_page_count =
            comp_ctx->comp_data->memories[0].init_page_count;
        uint64 mem_data_size = (uint64)num_bytes_per_page * init_page_count;
        if (mem_data_size > 0 && mem_offset + mem_len <= mem_data_size) {
            /* inside memory space */
            /* maddr = mem_base_addr + moffset */
            /* Perform zero extension in advance to avoid LLVMBuildInBoundsGEP2
             * interpreting a negative address due to sign extension when
             * mem_offset >= 2GiB */
            if (comp_ctx->pointer_size == sizeof(uint64)) {
                offset1 = I64_CONST(mem_offset);
            }
            else {
                offset1 = I32_CONST((uint32)mem_offset);
            }
            CHECK_LLVM_CONST(offset1);
            if (!(maddr = LLVMBuildInBoundsGEP2(comp_ctx->builder, INT8_TYPE,
                                                mem_base_addr, &offset1, 1,
                                                "maddr"))) {
                aot_set_last_error("llvm build add failed.");
                goto fail;
            }
            return maddr;
        }
    }

    if (func_ctx->mem_space_unchanged) {
        mem_size = func_ctx->mem_info[0].mem_data_size_addr;
    }
    else {
        if (!(mem_size = LLVMBuildLoad2(
                  comp_ctx->builder, I64_TYPE,
                  func_ctx->mem_info[0].mem_data_size_addr, "mem_size"))) {
            aot_set_last_error("llvm build load failed.");
            goto fail;
        }
    }

    ADD_BASIC_BLOCK(check_succ, "check_succ");
    LLVMMoveBasicBlockAfter(check_succ, block_curr);

    /* Same logic with aot_check_memory_overflow, offset and bytes are 32/64
     * bits on 32/64 bits platform */
    if (is_target_64bit) {
        offset =
            LLVMBuildZExt(comp_ctx->builder, offset, I64_TYPE, "extend_offset");
        bytes = LLVMBuildZExt(comp_ctx->builder, bytes, I64_TYPE, "extend_len");
        if (!offset || !bytes) {
            aot_set_last_error("llvm build zext failed.");
            goto fail;
        }
    }

    BUILD_OP(Add, offset, bytes, max_addr, "max_addr");

    /* Check overflow when it's memory64 or it's on 32 bits platform */
    if (is_memory64 || !is_target_64bit) {
        /* Check whether integer overflow occurs in offset + bytes */
        LLVMBasicBlockRef check_integer_overflow_end;
        ADD_BASIC_BLOCK(check_integer_overflow_end,
                        "check_integer_overflow_end");
        LLVMMoveBasicBlockAfter(check_integer_overflow_end, block_curr);

        /* offset + bytes can overflow yet is valid(for example, 0xffffffff, 1),
         * allow it to be 0(either 0, 0 or overflow and valid) */
        BUILD_ICMP(LLVMIntULT, max_addr, offset, cmp, "cmp");
        BUILD_ICMP(LLVMIntNE, max_addr, is_target_64bit ? I64_ZERO : I32_ZERO,
                   cmp1, "cmp1");
        BUILD_OP(And, cmp, cmp1, cmp, "overflow");
        if (!aot_emit_exception(comp_ctx, func_ctx,
                                EXCE_OUT_OF_BOUNDS_MEMORY_ACCESS, true, cmp,
                                check_integer_overflow_end)) {
            goto fail;
        }
        SET_BUILD_POS(check_integer_overflow_end);
        block_curr = check_integer_overflow_end;
    }

#if WASM_ENABLE_SHARED_HEAP != 0
    if (comp_ctx->enable_shared_heap
        || comp_ctx->enable_shared_chain /* TODO: && mem_idx == 0 */) {
        ADD_BASIC_BLOCK(block_maddr_phi, "maddr_phi");
        SET_BUILD_POS(block_maddr_phi);
        if (!(maddr_phi = LLVMBuildPhi(comp_ctx->builder, INT8_PTR_TYPE,
                                       "maddr_phi"))) {
            aot_set_last_error("llvm build phi failed");
            goto fail;
        }
        SET_BUILD_POS(block_curr);

        if (!aot_check_bulk_memory_shared_heap_memory_overflow(
                comp_ctx, func_ctx, block_curr, block_maddr_phi, check_succ,
                maddr_phi, offset, max_addr, bytes, is_memory64,
                is_target_64bit)) {
            goto fail;
        }
    }
#endif

    /* mem_size is always 64-bit, extend max_addr on 32 bits platform */
    if (!is_target_64bit
        && !(max_addr = LLVMBuildZExt(comp_ctx->builder, max_addr, I64_TYPE,
                                      "extend_max_addr"))) {
        aot_set_last_error("llvm build zext failed.");
        goto fail;
    }
    BUILD_ICMP(LLVMIntUGT, max_addr, mem_size, cmp, "cmp_max_mem_addr");

    if (!aot_emit_exception(comp_ctx, func_ctx,
                            EXCE_OUT_OF_BOUNDS_MEMORY_ACCESS, true, cmp,
                            check_succ)) {
        goto fail;
    }

    /* maddr = mem_base_addr + offset */
    if (!(maddr = LLVMBuildInBoundsGEP2(comp_ctx->builder, INT8_TYPE,
                                        mem_base_addr, &offset, 1, "maddr"))) {
        aot_set_last_error("llvm build add failed.");
        goto fail;
    }

#if WASM_ENABLE_SHARED_HEAP != 0
    if (comp_ctx->enable_shared_heap
        || comp_ctx->enable_shared_chain /* TODO: && mem_idx == 0 */) {
        block_curr = LLVMGetInsertBlock(comp_ctx->builder);
        LLVMAddIncoming(maddr_phi, &maddr, &block_curr, 1);
        if (!LLVMBuildBr(comp_ctx->builder, block_maddr_phi)) {
            aot_set_last_error("llvm build br failed");
            goto fail;
        }
        LLVMPositionBuilderAtEnd(comp_ctx->builder, block_maddr_phi);
        return maddr_phi;
    }
    else
#endif
        return maddr;
fail:
    return NULL;
}
#endif /* end of WASM_ENABLE_BULK_MEMORY != 0 || WASM_ENABLE_STRINGREF != 0 */

#if WASM_ENABLE_BULK_MEMORY != 0
bool
aot_compile_op_memory_init(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                           uint32 seg_index)
{
    LLVMValueRef seg, offset, dst, len, param_values[5], ret_value, func, value;
    LLVMTypeRef param_types[5], ret_type, func_type, func_ptr_type;
    AOTFuncType *aot_func_type = func_ctx->aot_func->func_type;
    LLVMBasicBlockRef block_curr = LLVMGetInsertBlock(comp_ctx->builder);
    LLVMBasicBlockRef mem_init_fail, init_success;

    seg = I32_CONST(seg_index);

    POP_I32(len);
    POP_I32(offset);
    POP_MEM_OFFSET(dst);

    if (!zero_extend_u64(comp_ctx, &dst, "dst64")) {
        return false;
    }

    param_types[0] = INT8_PTR_TYPE;
    param_types[1] = I32_TYPE;
    param_types[2] = I32_TYPE;
    param_types[3] = I32_TYPE;
    param_types[4] = SIZE_T_TYPE;
    ret_type = INT8_TYPE;

    if (comp_ctx->is_jit_mode)
        GET_AOT_FUNCTION(llvm_jit_memory_init, 5);
    else
        GET_AOT_FUNCTION(aot_memory_init, 5);

    /* Call function aot_memory_init() */
    param_values[0] = func_ctx->aot_inst;
    param_values[1] = seg;
    param_values[2] = offset;
    param_values[3] = len;
    param_values[4] = dst;
    if (!(ret_value = LLVMBuildCall2(comp_ctx->builder, func_type, func,
                                     param_values, 5, "call"))) {
        aot_set_last_error("llvm build call failed.");
        return false;
    }

    BUILD_ICMP(LLVMIntUGT, ret_value, I8_ZERO, ret_value, "mem_init_ret");

    ADD_BASIC_BLOCK(mem_init_fail, "mem_init_fail");
    ADD_BASIC_BLOCK(init_success, "init_success");

    LLVMMoveBasicBlockAfter(mem_init_fail, block_curr);
    LLVMMoveBasicBlockAfter(init_success, block_curr);

    if (!LLVMBuildCondBr(comp_ctx->builder, ret_value, init_success,
                         mem_init_fail)) {
        aot_set_last_error("llvm build cond br failed.");
        goto fail;
    }

    /* If memory.init failed, return this function
       so the runtime can catch the exception */
    LLVMPositionBuilderAtEnd(comp_ctx->builder, mem_init_fail);
    if (!aot_build_zero_function_ret(comp_ctx, func_ctx, aot_func_type)) {
        goto fail;
    }

    LLVMPositionBuilderAtEnd(comp_ctx->builder, init_success);

    return true;
fail:
    return false;
}

bool
aot_compile_op_data_drop(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                         uint32 seg_index)
{
    LLVMValueRef seg, param_values[2], ret_value, func, value;
    LLVMTypeRef param_types[2], ret_type, func_type, func_ptr_type;

    seg = I32_CONST(seg_index);
    CHECK_LLVM_CONST(seg);

    param_types[0] = INT8_PTR_TYPE;
    param_types[1] = I32_TYPE;
    ret_type = INT8_TYPE;

    if (comp_ctx->is_jit_mode)
        GET_AOT_FUNCTION(llvm_jit_data_drop, 2);
    else
        GET_AOT_FUNCTION(aot_data_drop, 2);

    /* Call function aot_data_drop() */
    param_values[0] = func_ctx->aot_inst;
    param_values[1] = seg;
    if (!(ret_value = LLVMBuildCall2(comp_ctx->builder, func_type, func,
                                     param_values, 2, "call"))) {
        aot_set_last_error("llvm build call failed.");
        return false;
    }

    return true;
fail:
    return false;
}
#endif /* end of WASM_ENABLE_BULK_MEMORY */

#if WASM_ENABLE_BULK_MEMORY_OPT != 0
bool
aot_compile_op_memory_copy(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
    LLVMValueRef src, dst, src_addr, dst_addr, len, res;
    bool call_aot_memmove = false;

    POP_MEM_OFFSET(len);
    POP_MEM_OFFSET(src);
    POP_MEM_OFFSET(dst);

    if (!(src_addr = check_bulk_memory_overflow(comp_ctx, func_ctx, src, len)))
        return false;

    if (!(dst_addr = check_bulk_memory_overflow(comp_ctx, func_ctx, dst, len)))
        return false;

    if (!zero_extend_u64(comp_ctx, &len, "len64")) {
        return false;
    }

    call_aot_memmove = comp_ctx->is_indirect_mode || comp_ctx->is_jit_mode;
    if (call_aot_memmove) {
        LLVMTypeRef param_types[3], ret_type, func_type, func_ptr_type;
        LLVMValueRef func, params[3];

        param_types[0] = INT8_PTR_TYPE;
        param_types[1] = INT8_PTR_TYPE;
        param_types[2] = SIZE_T_TYPE;
        ret_type = INT8_PTR_TYPE;

        if (!(func_type = LLVMFunctionType(ret_type, param_types, 3, false))) {
            aot_set_last_error("create LLVM function type failed.");
            return false;
        }

        if (!(func_ptr_type = LLVMPointerType(func_type, 0))) {
            aot_set_last_error("create LLVM function pointer type failed.");
            return false;
        }

        if (comp_ctx->is_jit_mode) {
            if (!(func = I64_CONST((uint64)(uintptr_t)aot_memmove))
                || !(func = LLVMConstIntToPtr(func, func_ptr_type))) {
                aot_set_last_error("create LLVM value failed.");
                return false;
            }
        }
        else {
            int32 func_index;
            func_index = aot_get_native_symbol_index(comp_ctx, "memmove");
            if (func_index < 0) {
                return false;
            }
            if (!(func =
                      aot_get_func_from_table(comp_ctx, func_ctx->native_symbol,
                                              func_ptr_type, func_index))) {
                return false;
            }
        }

        params[0] = dst_addr;
        params[1] = src_addr;
        params[2] = len;
        if (!(res = LLVMBuildCall2(comp_ctx->builder, func_type, func, params,
                                   3, "call_memmove"))) {
            aot_set_last_error("llvm build memmove failed.");
            return false;
        }
    }
    else {
        if (!(res = LLVMBuildMemMove(comp_ctx->builder, dst_addr, 1, src_addr,
                                     1, len))) {
            aot_set_last_error("llvm build memmove failed.");
            return false;
        }
    }

    return true;
fail:
    return false;
}

static void *
jit_memset(void *s, int c, size_t n)
{
    return memset(s, c, n);
}

bool
aot_compile_op_memory_fill(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
    LLVMValueRef val, dst, dst_addr, len, res;
    LLVMTypeRef param_types[3], ret_type, func_type, func_ptr_type;
    LLVMValueRef func, params[3];

    POP_MEM_OFFSET(len);
    POP_I32(val);
    POP_MEM_OFFSET(dst);

    if (!(dst_addr = check_bulk_memory_overflow(comp_ctx, func_ctx, dst, len)))
        return false;

    if (!zero_extend_u64(comp_ctx, &len, "len64")) {
        return false;
    }

    param_types[0] = INT8_PTR_TYPE;
    param_types[1] = I32_TYPE;
    param_types[2] = SIZE_T_TYPE;
    ret_type = INT8_PTR_TYPE;

    if (!(func_type = LLVMFunctionType(ret_type, param_types, 3, false))) {
        aot_set_last_error("create LLVM function type failed.");
        return false;
    }

    if (!(func_ptr_type = LLVMPointerType(func_type, 0))) {
        aot_set_last_error("create LLVM function pointer type failed.");
        return false;
    }

    if (comp_ctx->is_jit_mode) {
        if (!(func = I64_CONST((uint64)(uintptr_t)jit_memset))
            || !(func = LLVMConstIntToPtr(func, func_ptr_type))) {
            aot_set_last_error("create LLVM value failed.");
            return false;
        }
    }
    else if (comp_ctx->is_indirect_mode) {
        int32 func_index;
        func_index = aot_get_native_symbol_index(comp_ctx, "memset");
        if (func_index < 0) {
            return false;
        }
        if (!(func = aot_get_func_from_table(comp_ctx, func_ctx->native_symbol,
                                             func_ptr_type, func_index))) {
            return false;
        }
    }
    else {
        if (!(func = LLVMGetNamedFunction(func_ctx->module, "memset"))
            && !(func =
                     LLVMAddFunction(func_ctx->module, "memset", func_type))) {
            aot_set_last_error("llvm add function failed.");
            return false;
        }
    }

    params[0] = dst_addr;
    params[1] = val;
    params[2] = len;
    if (!(res = LLVMBuildCall2(comp_ctx->builder, func_type, func, params, 3,
                               "call_memset"))) {
        aot_set_last_error("llvm build memset failed.");
        return false;
    }

    return true;
fail:
    return false;
}
#endif /* end of WASM_ENABLE_BULK_MEMORY_OPT */

#if WASM_ENABLE_SHARED_MEMORY != 0
bool
aot_compile_op_atomic_rmw(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                          uint8 atomic_op, uint8 op_type, uint32 align,
                          mem_offset_t offset, uint32 bytes)
{
    LLVMValueRef maddr, value, result;
    bool enable_segue = (op_type == VALUE_TYPE_I32)
                            ? comp_ctx->enable_segue_i32_load
                                  && comp_ctx->enable_segue_i32_store
                            : comp_ctx->enable_segue_i64_load
                                  && comp_ctx->enable_segue_i64_store;

    if (op_type == VALUE_TYPE_I32)
        POP_I32(value);
    else
        POP_I64(value);

    if (!(maddr = aot_check_memory_overflow(comp_ctx, func_ctx, offset, bytes,
                                            enable_segue, NULL)))
        return false;

    if (!check_memory_alignment(comp_ctx, func_ctx, maddr, align))
        return false;

    switch (bytes) {
        case 8:
            if (!enable_segue)
                BUILD_PTR_CAST(INT64_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT64_PTR_TYPE_GS);
            break;
        case 4:
            if (!enable_segue)
                BUILD_PTR_CAST(INT32_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT32_PTR_TYPE_GS);
            if (op_type == VALUE_TYPE_I64)
                BUILD_TRUNC(value, I32_TYPE);
            break;
        case 2:
            if (!enable_segue)
                BUILD_PTR_CAST(INT16_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT16_PTR_TYPE_GS);
            BUILD_TRUNC(value, INT16_TYPE);
            break;
        case 1:
            if (!enable_segue)
                BUILD_PTR_CAST(INT8_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT8_PTR_TYPE_GS);
            BUILD_TRUNC(value, INT8_TYPE);
            break;
        default:
            bh_assert(0);
            break;
    }

    if (!(result = LLVMBuildAtomicRMW(
              comp_ctx->builder, atomic_op, maddr, value,
              LLVMAtomicOrderingSequentiallyConsistent, false))) {
        goto fail;
    }

    LLVMSetVolatile(result, true);

    if (op_type == VALUE_TYPE_I32) {
        if (!(result = LLVMBuildZExt(comp_ctx->builder, result, I32_TYPE,
                                     "result_i32"))) {
            goto fail;
        }
        PUSH_I32(result);
    }
    else {
        if (!(result = LLVMBuildZExt(comp_ctx->builder, result, I64_TYPE,
                                     "result_i64"))) {
            goto fail;
        }
        PUSH_I64(result);
    }

    return true;
fail:
    return false;
}

bool
aot_compile_op_atomic_cmpxchg(AOTCompContext *comp_ctx,
                              AOTFuncContext *func_ctx, uint8 op_type,
                              uint32 align, mem_offset_t offset, uint32 bytes)
{
    LLVMValueRef maddr, value, expect, result;
    bool enable_segue = (op_type == VALUE_TYPE_I32)
                            ? comp_ctx->enable_segue_i32_load
                                  && comp_ctx->enable_segue_i32_store
                            : comp_ctx->enable_segue_i64_load
                                  && comp_ctx->enable_segue_i64_store;

    if (op_type == VALUE_TYPE_I32) {
        POP_I32(value);
        POP_I32(expect);
    }
    else {
        POP_I64(value);
        POP_I64(expect);
    }

    if (!(maddr = aot_check_memory_overflow(comp_ctx, func_ctx, offset, bytes,
                                            enable_segue, NULL)))
        return false;

    if (!check_memory_alignment(comp_ctx, func_ctx, maddr, align))
        return false;

    switch (bytes) {
        case 8:
            if (!enable_segue)
                BUILD_PTR_CAST(INT64_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT64_PTR_TYPE_GS);
            break;
        case 4:
            if (!enable_segue)
                BUILD_PTR_CAST(INT32_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT32_PTR_TYPE_GS);
            if (op_type == VALUE_TYPE_I64) {
                BUILD_TRUNC(value, I32_TYPE);
                BUILD_TRUNC(expect, I32_TYPE);
            }
            break;
        case 2:
            if (!enable_segue)
                BUILD_PTR_CAST(INT16_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT16_PTR_TYPE_GS);
            BUILD_TRUNC(value, INT16_TYPE);
            BUILD_TRUNC(expect, INT16_TYPE);
            break;
        case 1:
            if (!enable_segue)
                BUILD_PTR_CAST(INT8_PTR_TYPE);
            else
                BUILD_PTR_CAST(INT8_PTR_TYPE_GS);
            BUILD_TRUNC(value, INT8_TYPE);
            BUILD_TRUNC(expect, INT8_TYPE);
            break;
        default:
            bh_assert(0);
            break;
    }

    if (!(result = LLVMBuildAtomicCmpXchg(
              comp_ctx->builder, maddr, expect, value,
              LLVMAtomicOrderingSequentiallyConsistent,
              LLVMAtomicOrderingSequentiallyConsistent, false))) {
        goto fail;
    }

    LLVMSetVolatile(result, true);

    /* CmpXchg return {i32, i1} structure,
       we need to extract the previous_value from the structure */
    if (!(result = LLVMBuildExtractValue(comp_ctx->builder, result, 0,
                                         "previous_value"))) {
        goto fail;
    }

    if (op_type == VALUE_TYPE_I32) {
        if (LLVMTypeOf(result) != I32_TYPE) {
            if (!(result = LLVMBuildZExt(comp_ctx->builder, result, I32_TYPE,
                                         "result_i32"))) {
                goto fail;
            }
        }
        PUSH_I32(result);
    }
    else {
        if (LLVMTypeOf(result) != I64_TYPE) {
            if (!(result = LLVMBuildZExt(comp_ctx->builder, result, I64_TYPE,
                                         "result_i64"))) {
                goto fail;
            }
        }
        PUSH_I64(result);
    }

    return true;
fail:
    return false;
}

bool
aot_compile_op_atomic_wait(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx,
                           uint8 op_type, uint32 align, mem_offset_t offset,
                           uint32 bytes)
{
    LLVMValueRef maddr, value, timeout, expect, cmp;
    LLVMValueRef param_values[5], ret_value, func, is_wait64;
    LLVMTypeRef param_types[5], ret_type, func_type, func_ptr_type;
    LLVMBasicBlockRef wait_fail, wait_success;
    LLVMBasicBlockRef block_curr = LLVMGetInsertBlock(comp_ctx->builder);
    AOTFuncType *aot_func_type = func_ctx->aot_func->func_type;

    POP_I64(timeout);
    if (op_type == VALUE_TYPE_I32) {
        POP_I32(expect);
        is_wait64 = I8_CONST(false);
        if (!(expect = LLVMBuildZExt(comp_ctx->builder, expect, I64_TYPE,
                                     "expect_i64"))) {
            goto fail;
        }
    }
    else {
        POP_I64(expect);
        is_wait64 = I8_CONST(true);
    }

    CHECK_LLVM_CONST(is_wait64);

    if (!(maddr = aot_check_memory_overflow(comp_ctx, func_ctx, offset, bytes,
                                            false, NULL)))
        return false;

    if (!check_memory_alignment(comp_ctx, func_ctx, maddr, align))
        return false;

    param_types[0] = INT8_PTR_TYPE;
    param_types[1] = INT8_PTR_TYPE;
    param_types[2] = I64_TYPE;
    param_types[3] = I64_TYPE;
    param_types[4] = INT8_TYPE;
    ret_type = I32_TYPE;

    GET_AOT_FUNCTION(wasm_runtime_atomic_wait, 5);

    /* Call function wasm_runtime_atomic_wait() */
    param_values[0] = func_ctx->aot_inst;
    param_values[1] = maddr;
    param_values[2] = expect;
    param_values[3] = timeout;
    param_values[4] = is_wait64;
    if (!(ret_value = LLVMBuildCall2(comp_ctx->builder, func_type, func,
                                     param_values, 5, "call"))) {
        aot_set_last_error("llvm build call failed.");
        return false;
    }

    BUILD_ICMP(LLVMIntNE, ret_value, I32_NEG_ONE, cmp, "atomic_wait_ret");

    ADD_BASIC_BLOCK(wait_fail, "atomic_wait_fail");
    ADD_BASIC_BLOCK(wait_success, "wait_success");

    LLVMMoveBasicBlockAfter(wait_fail, block_curr);
    LLVMMoveBasicBlockAfter(wait_success, block_curr);

    if (!LLVMBuildCondBr(comp_ctx->builder, cmp, wait_success, wait_fail)) {
        aot_set_last_error("llvm build cond br failed.");
        goto fail;
    }

    /* If atomic wait failed, return this function
       so the runtime can catch the exception */
    LLVMPositionBuilderAtEnd(comp_ctx->builder, wait_fail);
    if (!aot_build_zero_function_ret(comp_ctx, func_ctx, aot_func_type)) {
        goto fail;
    }

    LLVMPositionBuilderAtEnd(comp_ctx->builder, wait_success);

    PUSH_I32(ret_value);

    /* Insert suspend check point */
    if (comp_ctx->enable_thread_mgr) {
        if (!check_suspend_flags(comp_ctx, func_ctx, false))
            return false;
    }

    return true;
fail:
    return false;
}

bool
aot_compiler_op_atomic_notify(AOTCompContext *comp_ctx,
                              AOTFuncContext *func_ctx, uint32 align,
                              mem_offset_t offset, uint32 bytes)
{
    LLVMValueRef maddr, value, count;
    LLVMValueRef param_values[3], ret_value, func;
    LLVMTypeRef param_types[3], ret_type, func_type, func_ptr_type;

    POP_I32(count);

    if (!(maddr = aot_check_memory_overflow(comp_ctx, func_ctx, offset, bytes,
                                            false, NULL)))
        return false;

    if (!check_memory_alignment(comp_ctx, func_ctx, maddr, align))
        return false;

    param_types[0] = INT8_PTR_TYPE;
    param_types[1] = INT8_PTR_TYPE;
    param_types[2] = I32_TYPE;
    ret_type = I32_TYPE;

    GET_AOT_FUNCTION(wasm_runtime_atomic_notify, 3);

    /* Call function wasm_runtime_atomic_notify() */
    param_values[0] = func_ctx->aot_inst;
    param_values[1] = maddr;
    param_values[2] = count;
    if (!(ret_value = LLVMBuildCall2(comp_ctx->builder, func_type, func,
                                     param_values, 3, "call"))) {
        aot_set_last_error("llvm build call failed.");
        return false;
    }

    PUSH_I32(ret_value);

    return true;
fail:
    return false;
}

bool
aot_compiler_op_atomic_fence(AOTCompContext *comp_ctx, AOTFuncContext *func_ctx)
{
    return LLVMBuildFence(comp_ctx->builder,
                          LLVMAtomicOrderingSequentiallyConsistent, false, "")
               ? true
               : false;
}

#endif /* end of WASM_ENABLE_SHARED_MEMORY */