diff --git a/COMPILE.TXT b/COMPILE.TXT
index 5887f094..fa2b1d7c 100644
--- a/COMPILE.TXT
+++ b/COMPILE.TXT
@@ -178,8 +178,16 @@ Unicorn requires few dependent packages as followings
 [7] Compile on Windows with MinGW (MSYS2)
 
   To compile with MinGW you need to install MSYS2: https://msys2.github.io/ 
-  Follow the install instructions and don't forget to update the system packages as written in 5 & 6 paragraphs
+  Follow the install instructions and don't forget to update the system packages with:
+
+        $ pacman --needed -Sy bash pacman pacman-mirrors msys2-runtime
+
+     Then close MSYS2, run it again from Start menu and update the rest with:
+
+        $ pacman -Su
   
+  Finally, compile Unicorn with the next steps:
+
         - To compile Windows 32-bit binary with MinGW, run:
                 $ pacman -S make
                 $ pacman -S pkg-config
diff --git a/hook.c b/hook.c
index 31fcaabf..850c9564 100644
--- a/hook.c
+++ b/hook.c
@@ -63,17 +63,17 @@ size_t hook_add(struct uc_struct *uc, int type, uint64_t begin, uint64_t end, vo
                      if (begin > end)
                          uc->hook_insn_idx = i;
                      break;
-            case UC_MEM_READ:
+            case UC_HOOK_MEM_READ:
                      uc->hook_mem_read = true;
                      if (begin > end)
                          uc->hook_read_idx = i;
                      break;
-            case UC_MEM_WRITE:
+            case UC_HOOK_MEM_WRITE:
                      uc->hook_mem_write = true;
                      if (begin > end)
                          uc->hook_write_idx = i;
                      break;
-            case UC_MEM_READ_WRITE:
+            case UC_HOOK_MEM_READ_WRITE:
                      uc->hook_mem_read = true;
                      uc->hook_mem_write = true;
                      if (begin > end) {
@@ -151,12 +151,13 @@ static struct hook_struct *_hook_find(struct uc_struct *uc, int type, uint64_t a
             if (uc->hook_insn_idx)
                 return &uc->hook_callbacks[uc->hook_insn_idx];
             break;
-        case UC_MEM_READ:
+        case UC_HOOK_MEM_READ:
             // already hooked all memory read?
-            if (uc->hook_read_idx)
+            if (uc->hook_read_idx) {
                 return &uc->hook_callbacks[uc->hook_read_idx];
+            }
             break;
-        case UC_MEM_WRITE:
+        case UC_HOOK_MEM_WRITE:
             // already hooked all memory write?
             if (uc->hook_write_idx)
                 return &uc->hook_callbacks[uc->hook_write_idx];
@@ -174,14 +175,14 @@ static struct hook_struct *_hook_find(struct uc_struct *uc, int type, uint64_t a
                              return &uc->hook_callbacks[i];
                      }
                      break;
-            case UC_MEM_READ:
-                     if (uc->hook_callbacks[i].hook_type == UC_MEM_READ || uc->hook_callbacks[i].hook_type == UC_MEM_READ_WRITE) {
+            case UC_HOOK_MEM_READ:
+                     if (uc->hook_callbacks[i].hook_type == UC_HOOK_MEM_READ || uc->hook_callbacks[i].hook_type == UC_HOOK_MEM_READ_WRITE) {
                          if (uc->hook_callbacks[i].begin <= address && address <= uc->hook_callbacks[i].end)
                              return &uc->hook_callbacks[i];
                      }
                      break;
-            case UC_MEM_WRITE:
-                     if (uc->hook_callbacks[i].hook_type == UC_MEM_WRITE || uc->hook_callbacks[i].hook_type == UC_MEM_READ_WRITE) {
+            case UC_HOOK_MEM_WRITE:
+                     if (uc->hook_callbacks[i].hook_type == UC_HOOK_MEM_WRITE || uc->hook_callbacks[i].hook_type == UC_HOOK_MEM_READ_WRITE) {
                          if (uc->hook_callbacks[i].begin <= address && address <= uc->hook_callbacks[i].end)
                              return &uc->hook_callbacks[i];
                      }
diff --git a/include/uc_priv.h b/include/uc_priv.h
index 6e43a930..195d719c 100644
--- a/include/uc_priv.h
+++ b/include/uc_priv.h
@@ -172,6 +172,8 @@ struct uc_struct {
     bool block_full;
     MemoryRegion **mapped_blocks;
     uint32_t mapped_block_count;
+
+    void *qemu_thread_data; // to support cross compile to Windows (qemu-thread-win32.c)
 };
 
 #include "qemu_macro.h"
diff --git a/qemu/cpus.c b/qemu/cpus.c
index ecc66e83..e274fe5a 100644
--- a/qemu/cpus.c
+++ b/qemu/cpus.c
@@ -124,7 +124,7 @@ static void *qemu_tcg_cpu_thread_fn(void *arg)
     struct uc_struct *uc = cpu->uc;
 
     //qemu_tcg_init_cpu_signals();
-    qemu_thread_get_self(cpu->thread);
+    qemu_thread_get_self(uc, cpu->thread);
 
     qemu_mutex_lock(&uc->qemu_global_mutex);
     CPU_FOREACH(cpu) {
@@ -185,7 +185,7 @@ static void qemu_tcg_init_vcpu(CPUState *cpu)
         uc->tcg_halt_cond = cpu->halt_cond;
         snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "CPU %d/TCG",
                 cpu->cpu_index);
-        qemu_thread_create(cpu->thread, thread_name, qemu_tcg_cpu_thread_fn,
+        qemu_thread_create(uc, cpu->thread, thread_name, qemu_tcg_cpu_thread_fn,
                 cpu, QEMU_THREAD_JOINABLE);
 #ifdef _WIN32
         cpu->hThread = qemu_thread_get_handle(cpu->thread);
diff --git a/qemu/include/qemu/thread.h b/qemu/include/qemu/thread.h
index ef2f6962..de5956b2 100644
--- a/qemu/include/qemu/thread.h
+++ b/qemu/include/qemu/thread.h
@@ -52,12 +52,13 @@ void qemu_event_reset(QemuEvent *ev);
 void qemu_event_wait(QemuEvent *ev);
 void qemu_event_destroy(QemuEvent *ev);
 
-void qemu_thread_create(QemuThread *thread, const char *name,
+struct uc_struct;
+void qemu_thread_create(struct uc_struct *uc, QemuThread *thread, const char *name,
                         void *(*start_routine)(void *),
                         void *arg, int mode);
 void *qemu_thread_join(QemuThread *thread);
-void qemu_thread_get_self(QemuThread *thread);
+void qemu_thread_get_self(struct uc_struct *uc, QemuThread *thread);
 bool qemu_thread_is_self(QemuThread *thread);
-void qemu_thread_exit(void *retval);
+void qemu_thread_exit(struct uc_struct *uc, void *retval);
 
 #endif
diff --git a/qemu/softmmu_template.h b/qemu/softmmu_template.h
index 50d6e424..bda85dce 100644
--- a/qemu/softmmu_template.h
+++ b/qemu/softmmu_template.h
@@ -183,7 +183,7 @@ WORD_TYPE helper_le_ld_name(CPUArchState *env, target_ulong addr, int mmu_idx,
 
     // Unicorn: callback on memory read
     if (env->uc->hook_mem_read && READ_ACCESS_TYPE == MMU_DATA_LOAD) {
-        struct hook_struct *trace = hook_find(env->uc, UC_MEM_READ, addr);
+        struct hook_struct *trace = hook_find(env->uc, UC_HOOK_MEM_READ, addr);
         if (trace) {
             ((uc_cb_hookmem_t)trace->callback)(env->uc, UC_MEM_READ,
                     (uint64_t)addr, (int)DATA_SIZE, (int64_t)0, trace->user_data);
@@ -328,7 +328,7 @@ WORD_TYPE helper_be_ld_name(CPUArchState *env, target_ulong addr, int mmu_idx,
 
     // Unicorn: callback on memory read
     if (env->uc->hook_mem_read && READ_ACCESS_TYPE == MMU_DATA_LOAD) {
-        struct hook_struct *trace = hook_find(env->uc, UC_MEM_READ, addr);
+        struct hook_struct *trace = hook_find(env->uc, UC_HOOK_MEM_READ, addr);
         if (trace) {
             ((uc_cb_hookmem_t)trace->callback)(env->uc, UC_MEM_READ,
                     (uint64_t)addr, (int)DATA_SIZE, (int64_t)0, trace->user_data);
@@ -510,8 +510,8 @@ void helper_le_st_name(CPUArchState *env, target_ulong addr, DATA_TYPE val,
     MemoryRegion *mr = memory_mapping(uc, addr);
 
     // Unicorn: callback on memory write
-    if (env->uc->hook_mem_write) {
-        struct hook_struct *trace = hook_find(uc, UC_MEM_WRITE, addr);
+    if (uc->hook_mem_write) {
+        struct hook_struct *trace = hook_find(uc, UC_HOOK_MEM_WRITE, addr);
         if (trace) {
             ((uc_cb_hookmem_t)trace->callback)(uc, UC_MEM_WRITE,
                     (uint64_t)addr, (int)DATA_SIZE, (int64_t)val, trace->user_data);
@@ -649,7 +649,7 @@ void helper_be_st_name(CPUArchState *env, target_ulong addr, DATA_TYPE val,
 
     // Unicorn: callback on memory write
     if (uc->hook_mem_write) {
-        struct hook_struct *trace = hook_find(uc, UC_MEM_WRITE, addr);
+        struct hook_struct *trace = hook_find(uc, UC_HOOK_MEM_WRITE, addr);
         if (trace) {
             ((uc_cb_hookmem_t)trace->callback)(uc, UC_MEM_WRITE,
                     (uint64_t)addr, (int)DATA_SIZE, (int64_t)val, trace->user_data);
diff --git a/qemu/target-i386/translate.c b/qemu/target-i386/translate.c
index 209b1d58..6d93efe2 100644
--- a/qemu/target-i386/translate.c
+++ b/qemu/target-i386/translate.c
@@ -2944,8 +2944,8 @@ typedef void (*SSEFunc_0_epl)(TCGContext *s, TCGv_ptr env, TCGv_ptr reg, TCGv_i6
 typedef void (*SSEFunc_0_epp)(TCGContext *s, TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b);
 typedef void (*SSEFunc_0_eppi)(TCGContext *s, TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
                                TCGv_i32 val);
-typedef void (*SSEFunc_0_ppi)(TCGv_ptr reg_a, TCGv_ptr reg_b, TCGv_i32 val);
-typedef void (*SSEFunc_0_eppt)(TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
+typedef void (*SSEFunc_0_ppi)(TCGContext *s, TCGv_ptr reg_a, TCGv_ptr reg_b, TCGv_i32 val);
+typedef void (*SSEFunc_0_eppt)(TCGContext *s, TCGv_ptr env, TCGv_ptr reg_a, TCGv_ptr reg_b,
                                TCGv val);
 
 #define SSE_SPECIAL ((void *)1)
@@ -4669,7 +4669,7 @@ static void gen_sse(CPUX86State *env, DisasContext *s, int b,
             tcg_gen_addi_ptr(tcg_ctx, cpu_ptr1, cpu_env, op2_offset);
             /* XXX: introduce a new table? */
             sse_fn_ppi = (SSEFunc_0_ppi)sse_fn_epp;
-            sse_fn_ppi(cpu_ptr0, cpu_ptr1, tcg_const_i32(tcg_ctx, val));
+            sse_fn_ppi(tcg_ctx, cpu_ptr0, cpu_ptr1, tcg_const_i32(tcg_ctx, val));
             break;
         case 0xc2:
             /* compare insns */
@@ -4694,7 +4694,7 @@ static void gen_sse(CPUX86State *env, DisasContext *s, int b,
             tcg_gen_addi_ptr(tcg_ctx, cpu_ptr1, cpu_env, op2_offset);
             /* XXX: introduce a new table? */
             sse_fn_eppt = (SSEFunc_0_eppt)sse_fn_epp;
-            sse_fn_eppt(cpu_env, cpu_ptr0, cpu_ptr1, cpu_A0);
+            sse_fn_eppt(tcg_ctx, cpu_env, cpu_ptr0, cpu_ptr1, cpu_A0);
             break;
         default:
             tcg_gen_addi_ptr(tcg_ctx, cpu_ptr0, cpu_env, op1_offset);
@@ -8175,9 +8175,15 @@ static target_ulong disas_insn(CPUX86State *env, DisasContext *s,
         // printf("\n");
         if (changed_cc_op) {
             if (cc_op_dirty)
+#if TCG_TARGET_REG_BITS == 32
+                *(save_opparam_ptr + 16) = s->pc - pc_start;
+            else
+                *(save_opparam_ptr + 14) = s->pc - pc_start;
+#else
                 *(save_opparam_ptr + 12) = s->pc - pc_start;
             else
                 *(save_opparam_ptr + 10) = s->pc - pc_start;
+#endif
         } else {
             *(save_opparam_ptr + 1) = s->pc - pc_start;
         }
diff --git a/qemu/util/qemu-thread-posix.c b/qemu/util/qemu-thread-posix.c
index 1e8f8304..34632bca 100644
--- a/qemu/util/qemu-thread-posix.c
+++ b/qemu/util/qemu-thread-posix.c
@@ -389,7 +389,7 @@ void qemu_event_wait(QemuEvent *ev)
     }
 }
 
-void qemu_thread_create(QemuThread *thread, const char *name,
+void qemu_thread_create(struct uc_struct *uc, QemuThread *thread, const char *name,
                        void *(*start_routine)(void*),
                        void *arg, int mode)
 {
@@ -426,7 +426,7 @@ void qemu_thread_create(QemuThread *thread, const char *name,
     pthread_attr_destroy(&attr);
 }
 
-void qemu_thread_get_self(QemuThread *thread)
+void qemu_thread_get_self(struct uc_struct *uc, QemuThread *thread)
 {
     thread->thread = pthread_self();
 }
@@ -436,7 +436,7 @@ bool qemu_thread_is_self(QemuThread *thread)
    return pthread_equal(pthread_self(), thread->thread);
 }
 
-void qemu_thread_exit(void *retval)
+void qemu_thread_exit(struct uc_struct *uc, void *retval)
 {
     pthread_exit(retval);
 }
diff --git a/qemu/util/qemu-thread-win32.c b/qemu/util/qemu-thread-win32.c
index 004f17bf..3f7f7015 100644
--- a/qemu/util/qemu-thread-win32.c
+++ b/qemu/util/qemu-thread-win32.c
@@ -16,6 +16,8 @@
 #include <assert.h>
 #include <limits.h>
 
+#include "uc_priv.h"
+
 
 static void error_exit(int err, const char *msg)
 {
@@ -264,10 +266,9 @@ struct QemuThreadData {
     bool              exited;
     void             *ret;
     CRITICAL_SECTION  cs;
+    struct uc_struct *uc;
 };
 
-static __thread QemuThreadData *qemu_thread_data;
-
 static unsigned __stdcall win32_start_routine(void *arg)
 {
     QemuThreadData *data = (QemuThreadData *) arg;
@@ -278,14 +279,13 @@ static unsigned __stdcall win32_start_routine(void *arg)
         g_free(data);
         data = NULL;
     }
-    qemu_thread_data = data;
-    qemu_thread_exit(start_routine(thread_arg));
+    qemu_thread_exit(data->uc, start_routine(thread_arg));
     abort();
 }
 
-void qemu_thread_exit(void *arg)
+void qemu_thread_exit(struct uc_struct *uc, void *arg)
 {
-    QemuThreadData *data = qemu_thread_data;
+    QemuThreadData *data = uc->qemu_thread_data;
 
     if (data) {
         assert(data->mode != QEMU_THREAD_DETACHED);
@@ -326,7 +326,7 @@ void *qemu_thread_join(QemuThread *thread)
     return ret;
 }
 
-void qemu_thread_create(QemuThread *thread, const char *name,
+void qemu_thread_create(struct uc_struct *uc, QemuThread *thread, const char *name,
                        void *(*start_routine)(void *),
                        void *arg, int mode)
 {
@@ -338,6 +338,9 @@ void qemu_thread_create(QemuThread *thread, const char *name,
     data->arg = arg;
     data->mode = mode;
     data->exited = false;
+    data->uc = uc;
+
+    uc->qemu_thread_data = data;
 
     if (data->mode != QEMU_THREAD_DETACHED) {
         InitializeCriticalSection(&data->cs);
@@ -352,9 +355,9 @@ void qemu_thread_create(QemuThread *thread, const char *name,
     thread->data = (mode == QEMU_THREAD_DETACHED) ? NULL : data;
 }
 
-void qemu_thread_get_self(QemuThread *thread)
+void qemu_thread_get_self(struct uc_struct *uc, QemuThread *thread)
 {
-    thread->data = qemu_thread_data;
+    thread->data = uc->qemu_thread_data;
     thread->tid = GetCurrentThreadId();
 }
 
diff --git a/regress/rep_movsb.c b/regress/rep_movsb.c
index ab49175b..c9fb6f15 100644
--- a/regress/rep_movsb.c
+++ b/regress/rep_movsb.c
@@ -128,7 +128,7 @@ int main(int argc, char **argv, char **envp)
       printf("ok %d - Program written to memory\n", log_num++);
    }
 
-   if (uc_hook_add(handle, &trace2, UC_HOOK_CODE, hook_code, NULL, 1, 0) != UC_ERR_OK) {
+   if (uc_hook_add(handle, &trace2, UC_HOOK_CODE, hook_code, NULL, (uint64_t)1, (uint64_t)0) != UC_ERR_OK) {
       printf("not ok %d - Failed to install UC_HOOK_CODE handler\n", log_num++);
       return 5;
    }
@@ -137,7 +137,7 @@ int main(int argc, char **argv, char **envp)
    }
 
    // intercept memory write events only, NOT read events
-   if (uc_hook_add(handle, &trace1, UC_HOOK_MEM_WRITE, hook_mem_write, NULL) != UC_ERR_OK) {
+   if (uc_hook_add(handle, &trace1, UC_HOOK_MEM_WRITE, hook_mem_write, NULL, (uint64_t)1, (uint64_t)0) != UC_ERR_OK) {
       printf("not ok %d - Failed to install UC_HOOK_MEM_WRITE handler\n", log_num++);
       return 6;
    }
diff --git a/uc.c b/uc.c
index 0492ba6e..efe02622 100644
--- a/uc.c
+++ b/uc.c
@@ -414,7 +414,7 @@ static void *_timeout_fn(void *arg)
 static void enable_emu_timer(struct uc_struct *uc, uint64_t timeout)
 {
     uc->timeout = timeout;
-    qemu_thread_create(&uc->timer, "timeout", _timeout_fn,
+    qemu_thread_create(uc, &uc->timer, "timeout", _timeout_fn,
             uc, QEMU_THREAD_JOINABLE);
 }
 
@@ -530,7 +530,7 @@ static int _hook_code(struct uc_struct *uc, int type, uint64_t begin, uint64_t e
 }
 
 
-static uc_err _hook_mem_access(struct uc_struct *uc, uc_mem_type type,
+static uc_err _hook_mem_access(struct uc_struct *uc, uc_hook_t type,
         uint64_t begin, uint64_t end,
         void *callback, void *user_data, uc_hook_h *hh)
 {
@@ -718,16 +718,17 @@ uc_err uc_hook_add(struct uc_struct *uc, uc_hook_h *hh, uc_hook_t type, void *ca
         case UC_HOOK_MEM_READ:
             begin = va_arg(valist, uint64_t);
             end = va_arg(valist, uint64_t);
-            ret = _hook_mem_access(uc, UC_MEM_READ, begin, end, callback, user_data, hh);
+            ret = _hook_mem_access(uc, UC_HOOK_MEM_READ, begin, end, callback, user_data, hh);
             break;
         case UC_HOOK_MEM_WRITE:
             begin = va_arg(valist, uint64_t);
             end = va_arg(valist, uint64_t);
-            ret = _hook_mem_access(uc, UC_MEM_WRITE, begin, end, callback, user_data, hh);
+            ret = _hook_mem_access(uc, UC_HOOK_MEM_WRITE, begin, end, callback, user_data, hh);
+            break;
         case UC_HOOK_MEM_READ_WRITE:
             begin = va_arg(valist, uint64_t);
             end = va_arg(valist, uint64_t);
-            ret = _hook_mem_access(uc, UC_MEM_READ_WRITE, begin, end, callback, user_data, hh);
+            ret = _hook_mem_access(uc, UC_HOOK_MEM_READ_WRITE, begin, end, callback, user_data, hh);
             break;
     }