Change the compiler's big integer library to use libTomMath

This now replaces Bill's crappy big int implementation

1 files changed, 54 insertions, 20 deletions

diff --git a/src/llvm_backend.cpp b/src/llvm_backend.cpp
index c351876d4..fa45583d6 100644
--- a/src/llvm_backend.cpp
+++ b/src/llvm_backend.cpp
@@ -6558,36 +6558,70 @@ lbValue lb_const_value(lbModule *m, Type *type, ExactValue value, bool allow_loc
 
 	case ExactValue_Integer:
 		if (is_type_pointer(type)) {
-			unsigned len = cast(unsigned)value.value_integer.len;
 			LLVMTypeRef t = lb_type(m, original_type);
-			if (len == 0) {
+			if (mp_iszero(&value.value_integer)) {
 				res.value = LLVMConstNull(t);
 			} else {
-				LLVMValueRef i = LLVMConstIntOfArbitraryPrecision(lb_type(m, t_uintptr), len, big_int_ptr(&value.value_integer));
+				unsigned len = cast(unsigned)value.value_integer.used;
+				u64 v = mp_get_u64(&value.value_integer);
+				LLVMValueRef i = LLVMConstInt(lb_type(m, t_uintptr), cast(unsigned long long)v, false);
 				res.value = LLVMConstIntToPtr(i, t);
 			}
 		} else {
-			unsigned len = cast(unsigned)value.value_integer.len;
-			if (len == 0) {
+			if (mp_iszero(&value.value_integer)) {
 				res.value = LLVMConstNull(lb_type(m, original_type));
 			} else {
-				u64 *words = big_int_ptr(&value.value_integer);
-				if (is_type_different_to_arch_endianness(type)) {
-					// NOTE(bill): Swap byte order for different endianness
-					i64 sz = type_size_of(type);
-					isize byte_len = gb_size_of(u64)*len;
-					u8 *old_bytes = cast(u8 *)words;
-					// TODO(bill): Use a different allocator here for a temporary allocation
-					u8 *new_bytes = cast(u8 *)gb_alloc_align(permanent_allocator(), byte_len, gb_align_of(u64));
-					for (i64 i = 0; i < sz; i++) {
-						new_bytes[i] = old_bytes[sz-1-i];
-					}
-					words = cast(u64 *)new_bytes;
-				}
-				res.value = LLVMConstIntOfArbitraryPrecision(lb_type(m, original_type), len, words);
-				if (value.value_integer.neg) {
+				mp_int *a = &value.value_integer;
+
+				u8 *rop = nullptr;
+				size_t max_count = 0;
+				size_t written = 0;
+				size_t size = 1;
+				size_t nails = 0;
+				mp_endian endian = MP_NATIVE_ENDIAN;
+				if (is_type_endian_little(type)) {
+					endian = MP_LITTLE_ENDIAN;
+				} else if (is_type_endian_big(type)) {
+					endian = MP_BIG_ENDIAN;
+				}
+
+				max_count = mp_pack_count(a, nails, size);
+				rop = cast(u8 *)gb_alloc_align(permanent_allocator(), max_count, gb_align_of(u64));
+				mp_err err = mp_pack(rop, max_count, &written,
+				                     MP_LSB_FIRST,
+				                     size, endian, nails,
+				                     &value.value_integer);
+				GB_ASSERT(err == MP_OKAY);
+
+				res.value = LLVMConstIntOfArbitraryPrecision(lb_type(m, original_type), cast(unsigned)((written+7)/8), cast(u64 *)rop);
+				if (value.value_integer.sign) {
 					res.value = LLVMConstNeg(res.value);
 				}
+
+				// size_t written = 0;
+				// size_t max_bytes = (value.value_integer.used*gb_size_of(mp_digit)+7)&~7;
+				// u8 *buf = cast(u8 *)gb_alloc_align(permanent_allocator(), max_bytes, gb_align_of(u64));
+				// mp_to_ubin(&value.value_integer, buf, max_bytes, &written);
+
+				// gb_printf_err("%tu %tu", written, max_bytes);
+				// for (size_t i = 0; i < written; i++) {
+				// 	gb_printf_err("%02x", buf[i]);
+				// }
+				// gb_printf_err("\n");
+				// gb_exit(1);
+
+				// if (is_type_different_to_arch_endianness(type)) {
+				// 	u8 *old_bytes = buf;
+				// 	u8 *new_bytes = cast(u8 *)gb_alloc_align(permanent_allocator(), max_bytes, gb_align_of(u64));
+				// 	for (size_t i = 0; i < written; i++) {
+				// 		new_bytes[i] = old_bytes[written-1-i];
+				// 	}
+				// 	buf = new_bytes;
+				// }
+				// res.value = LLVMConstIntOfArbitraryPrecision(lb_type(m, original_type), cast(unsigned)((written+7)/8), cast(u64 *)buf);
+				// if (value.value_integer.sign) {
+				// 	res.value = LLVMConstNeg(res.value);
+				// }
 			}
 		}
 		return res;