TTMath broken on Windows: GCC ASM stack corruption, VCC refuses inline ASM, string conversion

[mratsim]

From @coffeepots

VCC doesn't support inline asm: https://www.ttmath.org/forum/how_to_compile_with_visual_studio_2010_64-bit

GCC 6.3 on windows has a stack corruption issue:

Before #11 it was this:

#0  ttmath::UInt<8ul>::Pow (this=0x65fd70, pow=...) at c:/Status/nim-ttmath/src/headers/ttmathuint.h:2418
#1  0x0000000000401603 in pow (a=0x40a040 <a>, b=0x40a060 <b>) at c:\Status\nim-ttmath\src\headers\test.cpp:12
#2  0x0000000000401722 in main () at c:\Status\nim-ttmath\src\headers\test.cpp:21

After it is now this:

#0  ttmath::UInt<8ul>::Pow (this=0x65fd70, pow=...) at c:/Status/nim-ttmath/src/headers/ttmathuint.h:2419
#1  0x0000000000401603 in pow (a=0x40a040 <a>, b=0x40a060 <b>) at c:\Status\nim-ttmath\src\headers\test.cpp:12
#2  0x0000000000200b91 in ?? ()

2418-2419 corresponds to the Rcr2_one call in the following part

	uint Pow(UInt<value_size> pow)
	{
		if(pow.IsZero() && IsZero())
			// we don't define zero^zero
			return 2;

		UInt<value_size> start(*this);
		UInt<value_size> result;
		result.SetOne();
		uint c = 0;

		while( !c )
		{
			if( pow.table[0] & 1 )
				c += result.Mul(start);

			pow.Rcr2_one(0);
			if( pow.IsZero() )
				break;

			c += start.Mul(start);
		}

		*this = result;

		TTMATH_LOGC("UInt::Pow(UInt<>)", c)

	return (c==0)? 0 : 1;
	}

	template<uint value_size>
	uint UInt<value_size>::Rcr2_one(uint c)
	{
	sint b = value_size;
	uint * p1 = table;
	

		#ifndef __GNUC__
			c = ttmath_rcr_x64(p1,b,c);
		#endif


		#ifdef __GNUC__
		uint dummy;

		__asm__  __volatile__(

			"negq %%rax						\n"   // CF=1 if rax!=0 , CF=0 if rax==0

		"1:									\n"
			"rcrq $1, -8(%%rbx, %%rcx, 8)	\n"

			"decq %%rcx						\n"
		"jnz 1b								\n"

			"adcq %%rcx, %%rcx				\n"

			: "=c" (c), "=a" (dummy)
			: "0" (b),  "1" (c), "b" (p1)
			: "cc", "memory" );

		#endif

		TTMATH_LOGC("UInt::Rcr2_one", c)

[mratsim]

A simple fix would be to deactivate ASM on Windows. In any case we target mobile and restricted devices (ARM and MIPS) and no ASM is available for those.

when defined(windows):
  {.passC: "-DTTMATH_NOASM".}

[mratsim]

Adding to the list of complaints: string conversion doesn't work on Windows: https://ci.appveyor.com/project/jarradh/nim-ttmath/build/22/job/yjwudram3m15we4f#L1378

[Suite] ttmath
    test1.nim(9, 15): Check failed: $(a + b) == "28497092031222200795837"
    $(a + b) was 25873220285
    test1.nim(10, 12): Check failed: a + b == "28497092031222200795837".i256
    a + b was 25873220285
    "28497092031222200795837".i256 was 103416509
    test1.nim(12, 12): Check failed: a + b < "28497092031222200795838".i256
    a + b was 25873220285
    "28497092031222200795838".i256 was 103416510
    test1.nim(13, 9): Check failed: -a == "-12345678910111213141516".i256
    -a was 12309411850224945125033585585165461537827125511483805057073903131095380176159220
    "-12345678910111213141516".i256 was 848018956
  [FAILED] Ints
    test1.nim(20, 15): Check failed: $(a + b) == "28497092031222200795837"
    $(a + b) was 25873220285
    test1.nim(21, 12): Check failed: a + b == "28497092031222200795837".u256
    a + b was 25873220285
    "28497092031222200795837".u256 was 103416509
    test1.nim(23, 12): Check failed: a + b < "28497092031222200795838".u256
    a + b was 25873220285
    "28497092031222200795838".u256 was 103416510
  [FAILED] UInts
[Suite] Testing conversion functions: Hex, Bytes, Endianness
  [OK] hex -> uint256
    test1.nim(35, 26): Check failed: SECPK1_N.toHex == SECPK1_N_HEX
    SECPK1_N.toHex was d0364141
    SECPK1_N_HEX was fffffffffffffffffffffffffffffffebaaedce6af48a03bbfd25e8cd0364141
  [FAILED] uint256 -> hex
    test1.nim(38, 43): Check failed: readUIntBE[256](SECPK1_N_BYTES) == SECPK1_N
    readUIntBE[256](SECPK1_N_BYTES) was 103443464614620605021057432097659110162469981302141632742275809279
    SECPK1_N was 3493216577
  [FAILED] hex -> big-endian array -> uint256
    test1.nim(41, 34): Check failed: SECPK1_N.toByteArrayBE == SECPK1_N_BYTES
    SECPK1_N.toByteArrayBE was [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 208, 54, 65, 65]
    SECPK1_N_BYTES was [255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 254, 186, 174, 220, 230, 175, 72, 160, 59, 191, 210, 94, 140, 208, 54, 65, 65]
  [FAILED] uint256 -> big-endian array -> hex
[Suite] Confirming consistency: hex vs decimal conversion
  [OK] Alice signature
  [OK] Bob signature
  [OK] Eve signature

[mratsim]

Fixed by @coffeepots 38b1ca4