Fix the bug when using FADDP with Capstone version > 4

librz/analysis/arch/x86/il_fp_ops.inc

@@ -391,7 +391,7 @@ IL_LIFTER(fabs) {
 		} \
 \
 		RzILOpPure *result = x86_il_##op##_with_rmode(x86_il_get_st_reg(X86_REG_ST0), x86_il_get_st_reg(dest_reg)); \
-		return x86_il_set_st_reg(dest_reg, result, RZ_FLOAT_IEEE754_BIN_80); \
+		return SEQ2(x86_il_set_st_reg(dest_reg, result, RZ_FLOAT_IEEE754_BIN_80), x86_il_st_pop()); \
 	} while (0)
 
 #define FLOATING_INT_ARITHMETIC_IL(op) \
@@ -406,16 +406,15 @@ IL_LIFTER(fabs) {
  * Add floating point values
  */
 IL_LIFTER(fadd) {
-	FLOATING_ARITHMETIC_IL(fadd);
-}
-
-/**
- * FADDP
- * (the corresponding enum is PFADD for Capstone version > 4)
- * Add ST(0) to ST(i) and pop the stack
- */
-IL_LIFTER(faddp) {
-	FLOATING_ARITHMETIC_POP_IL(fadd);
+	/* Have a unified IL lifter for FADD and FADDP since Capstone has removed the
+	 * distinction after version 4, which I think is a terrible thing. */
+	if (ins->structure->opcode[0] == 0xde) {
+		// FADDP
+		FLOATING_ARITHMETIC_POP_IL(fadd);
+	} else {
+		// FADD
+		FLOATING_ARITHMETIC_IL(fadd);
+	}
 }
 
 /**

librz/analysis/arch/x86/x86_il.c

@@ -288,15 +288,8 @@ x86_il_ins x86_ins[X86_INS_ENDING] = {
 	[X86_INS_FBSTP] = x86_il_fbstp,
 	[X86_INS_FABS] = x86_il_fabs,
 	[X86_INS_FADD] = x86_il_fadd,
-#if CS_API_MAJOR > 4
-	/* This feels wrong (since PFADD is a 3DNow instruction from what I
-	 * understand after going through the source code), but aquynh said this is
-	 * correct and this is what radare2 also uses.
-	 * See https://github.com/capstone-engine/capstone/issues/1456#issuecomment-482620580
-	 * and https://github.com/radareorg/radare2/blob/7fddeb97096e5d4db977dcd7d4f84db148eba595/libr/arch/p/x86/plugin_cs.c#L2158 */
-	[X86_INS_PFADD] = x86_il_faddp,
-#else
-	[X86_INS_FADDP] = x86_il_faddp,
+#if CS_API_MAJOR <= 4
+	[X86_INS_FADDP] = x86_il_fadd,
 #endif
 	[X86_INS_FIADD] = x86_il_fiadd,
 	[X86_INS_FMUL] = x86_il_fmul,