+ sljit_si dst_r = FAST_IS_REG(dst) ? dst : TMP_FREG1;
+ sljit_si invert_sign = 1;
+
+ if (src & SLJIT_IMM) {
+ FAIL_IF(load_immediate(compiler, TMP_REG1, srcw ^ 0x80000000));
+ src = TMP_REG1;
+ invert_sign = 0;
+ }
+ else if (!FAST_IS_REG(src)) {
+ FAIL_IF(emit_op_mem2(compiler, WORD_DATA | SIGNED_DATA | LOAD_DATA, TMP_REG1, src, srcw, SLJIT_MEM1(SLJIT_SP), FLOAT_TMP_MEM_OFFSET_LOW));
+ src = TMP_REG1;
+ }
+
+ /* First, a special double floating point value is constructed: (2^53 + (input xor (2^31)))
+ The double precision format has exactly 53 bit precision, so the lower 32 bit represents
+ the lower 32 bit of such value. The result of xor 2^31 is the same as adding 0x80000000
+ to the input, which shifts it into the 0 - 0xffffffff range. To get the converted floating
+ point value, we need to substract 2^53 + 2^31 from the constructed value. */
+ FAIL_IF(push_inst(compiler, ADDIS | D(TMP_REG2) | A(0) | 0x4330));
+ if (invert_sign)
+ FAIL_IF(push_inst(compiler, XORIS | S(src) | A(TMP_REG1) | 0x8000));
+ FAIL_IF(emit_op_mem2(compiler, WORD_DATA, TMP_REG2, SLJIT_MEM1(SLJIT_SP), FLOAT_TMP_MEM_OFFSET_HI, SLJIT_MEM1(SLJIT_SP), FLOAT_TMP_MEM_OFFSET));
+ FAIL_IF(emit_op_mem2(compiler, WORD_DATA, TMP_REG1, SLJIT_MEM1(SLJIT_SP), FLOAT_TMP_MEM_OFFSET_LOW, SLJIT_MEM1(SLJIT_SP), FLOAT_TMP_MEM_OFFSET_HI));
+ FAIL_IF(push_inst(compiler, ADDIS | D(TMP_REG1) | A(0) | 0x8000));
+ FAIL_IF(emit_op_mem2(compiler, DOUBLE_DATA | LOAD_DATA, TMP_FREG1, SLJIT_MEM1(SLJIT_SP), FLOAT_TMP_MEM_OFFSET, SLJIT_MEM1(SLJIT_SP), FLOAT_TMP_MEM_OFFSET_LOW));
+ FAIL_IF(emit_op_mem2(compiler, WORD_DATA, TMP_REG1, SLJIT_MEM1(SLJIT_SP), FLOAT_TMP_MEM_OFFSET_LOW, SLJIT_MEM1(SLJIT_SP), FLOAT_TMP_MEM_OFFSET));
+ FAIL_IF(emit_op_mem2(compiler, DOUBLE_DATA | LOAD_DATA, TMP_FREG2, SLJIT_MEM1(SLJIT_SP), FLOAT_TMP_MEM_OFFSET, SLJIT_MEM1(SLJIT_SP), FLOAT_TMP_MEM_OFFSET_LOW));
+
+ FAIL_IF(push_inst(compiler, FSUB | FD(dst_r) | FA(TMP_FREG1) | FB(TMP_FREG2)));
+
+ if (dst & SLJIT_MEM)
+ return emit_op_mem2(compiler, FLOAT_DATA(op), TMP_FREG1, dst, dstw, 0, 0);
+ if (op & SLJIT_SINGLE_OP)
+ return push_inst(compiler, FRSP | FD(dst_r) | FB(dst_r));
+ return SLJIT_SUCCESS;
+
+#endif
+}
+
+static SLJIT_INLINE sljit_si sljit_emit_fop1_cmp(struct sljit_compiler *compiler, sljit_si op,
+ sljit_si src1, sljit_sw src1w,
+ sljit_si src2, sljit_sw src2w)
+{
+ if (src1 & SLJIT_MEM) {
+ FAIL_IF(emit_op_mem2(compiler, FLOAT_DATA(op) | LOAD_DATA, TMP_FREG1, src1, src1w, src2, src2w));
+ src1 = TMP_FREG1;
+ }
+
+ if (src2 & SLJIT_MEM) {
+ FAIL_IF(emit_op_mem2(compiler, FLOAT_DATA(op) | LOAD_DATA, TMP_FREG2, src2, src2w, 0, 0));
+ src2 = TMP_FREG2;
+ }
+
+ return push_inst(compiler, FCMPU | CRD(4) | FA(src1) | FB(src2));
+}
+
+SLJIT_API_FUNC_ATTRIBUTE sljit_si sljit_emit_fop1(struct sljit_compiler *compiler, sljit_si op,
+ sljit_si dst, sljit_sw dstw,
+ sljit_si src, sljit_sw srcw)
+{
+ sljit_si dst_r;
+
+ CHECK_ERROR();
+ compiler->cache_arg = 0;
+ compiler->cache_argw = 0;
+
+ SLJIT_COMPILE_ASSERT((SLJIT_SINGLE_OP == 0x100) && !(DOUBLE_DATA & 0x4), float_transfer_bit_error);
+ SELECT_FOP1_OPERATION_WITH_CHECKS(compiler, op, dst, dstw, src, srcw);
+
+ if (GET_OPCODE(op) == SLJIT_CONVD_FROMS)
+ op ^= SLJIT_SINGLE_OP;
+
+ dst_r = FAST_IS_REG(dst) ? dst : TMP_FREG1;