tccgen: fix casts using code from upstream

We detected issues tcc-mob built with us was failing to compile code
like:

    unsigned long x = 0xcafecafe;

It was applying sign extension, producing a `x == 0xffffffffcafecafe`.
Instead, when built with GCC tcc-mob produced the proper
`x == 0xcafecafe`. This is because our casts are faulty in RISC-V
because it always sign-extends and we need to clear those high bits
after the sign-extension.

Proper compilation should produce this after the value is loaded:

    slli reg, reg, 0x20 // shift left 32 times
    srli reg, reg, 0x20 // and back

Meaning the highest 32 bits are cleared.

Our casts didn't detect the need for this, and many appearances of
integer constants in tcc-mob were miscompiled, producing future errors
in the chain: `load`, code generation, elf...

This commit backports the `gen_cast` function from tcc-mob and
manipulates the code around it to make it fit in our internals. It's not
superclean, but it should work.

* arm64-gen.c (gen_cvt_csti): New function for char to int conversion.
* i386-gen.c (gen_cvt_csti): Likewise.
* x86_64-gen.c (gen_cvt_csti): Likewise.
(gen_cvt_sxtw): New function for word sign-extension.
* tcc.h: Add them to headers.
* tccgen.c (gen_cast): Update from upstream.
(btype_size): Add function.

arm64-gen.c

@@ -1663,6 +1663,16 @@ ST_FUNC void gen_cvt_sxtw(void)
     o(0x93407c00 | r | r << 5); // sxtw x(r),w(r)
 }
 
+/* char/short to int conversion */
+ST_FUNC void gen_cvt_csti(int t)
+{
+    int r = intr(gv(RC_INT));
+    o(0x13001c00
+        | ((t & VT_BTYPE) == VT_SHORT) << 13
+        | (uint32_t)!!(t & VT_UNSIGNED) << 30
+        | r | r << 5); // [su]xt[bh] w(r),w(r)
+}
+
 ST_FUNC void gen_cvt_itof(int t)
 {
     if (t == VT_LDOUBLE) {

i386-gen.c

@@ -1066,6 +1066,19 @@ ST_FUNC void gen_cvt_ftof(int t)
     gv(RC_FLOAT);
 }
 
+/* char/short to int conversion */
+ST_FUNC void gen_cvt_csti(int t)
+{
+    int r, sz, xl;
+    r = gv(RC_INT);
+    sz = !(t & VT_UNSIGNED);
+    xl = (t & VT_BTYPE) == VT_SHORT;
+    o(0xc0b60f /* mov[sz] %a[xl], %eax */
+        | (sz << 3 | xl) << 8
+        | (r << 3 | r) << 16
+        );
+}
+
 /* computed goto support */
 ST_FUNC void ggoto(void)
 {

tcc.h

@@ -1564,6 +1564,7 @@ ST_FUNC void gen_le16(int c);
 ST_FUNC void gen_le32(int c);
 ST_FUNC void gen_addr32(int r, Sym *sym, long c);
 ST_FUNC void gen_addrpc32(int r, Sym *sym, long c);
+ST_FUNC void gen_cvt_csti(int t);
 #endif
 
 #ifdef CONFIG_TCC_BCHECK
@@ -1573,6 +1574,7 @@ ST_FUNC void gen_bounded_ptr_deref(void);
 
 /* ------------ x86_64-gen.c ------------ */
 #ifdef TCC_TARGET_X86_64
+ST_FUNC void gen_cvt_sxtw(void);
 ST_FUNC void gen_addr64(int r, Sym *sym, int64_t c);
 ST_FUNC void gen_opl(int op);
 #endif
@@ -1589,6 +1591,7 @@ ST_FUNC void gen_cvt_itof1(int t);
 /* ------------ arm64-gen.c ------------ */
 #ifdef TCC_TARGET_ARM64
 ST_FUNC void gen_cvt_sxtw(void);
+ST_FUNC void gen_cvt_csti(int t);
 ST_FUNC void gen_opl(int op);
 ST_FUNC void gfunc_return(CType *func_type);
 ST_FUNC void gen_va_start(void);