Add vext2xv testcases, try to fix xvshuf

code/gen_tb.py

@@ -5,6 +5,7 @@
 widths_fp = ["s", "d"]
 widths_all = ["b", "bu", "h", "hu", "w", "wu", "d", "du"]
 widths_vexth = ["h_b", "hu_bu", "w_h", "wu_hu", "d_w", "du_wu", "q_d", "qu_du"]
+widths_vext2xv = ["h_b", "hu_bu", "w_b", "wu_bu", "w_h", "wu_hu", "d_b", "du_bu", "d_h", "du_hu", "d_w", "du_wu"]
 widths_vsllwil = ["h_b", "hu_bu", "w_h", "wu_hu", "d_w", "du_wu"]
 widths_vsrln = ["b_h", "h_w", "w_d"]
 widths_vssrln = ["b_h", "bu_h", "h_w", "hu_w", "w_d", "wu_d"]
@@ -67,7 +68,7 @@
         "v128 a, v128 b, int imm",
         [0, 3, 7, 15, 16, 32, 64, 128, 255],
     ),
-    "vext2xv": (widths_vsllwil, "v128 a"),
+    "vext2xv": (widths_vext2xv, "v128 a"),
     "vfrstp": (widths_vfrstp, "v128 a, v128 b, v128 c"),
     "vfrstpi": (widths_vfrstp, "v128 a, v128 b, int imm", [0, 4, 31]),
     "vhaddw": (widths_vexth, "v128 a, v128 b"),

code/vext2xv_d_b.cpp

@@ -0,0 +1,9 @@
+#include "common.h"
+
+v256 vext2xv_d_b(v256 a) {
+  v256 dst;
+#include "vext2xv_d_b.h"
+  return dst;
+}
+
+void test() { XFUZZ1(vext2xv_d_b); }

code/vext2xv_d_h.cpp

@@ -0,0 +1,9 @@
+#include "common.h"
+
+v256 vext2xv_d_h(v256 a) {
+  v256 dst;
+#include "vext2xv_d_h.h"
+  return dst;
+}
+
+void test() { XFUZZ1(vext2xv_d_h); }

code/vext2xv_du_bu.cpp

@@ -0,0 +1,9 @@
+#include "common.h"
+
+v256 vext2xv_du_bu(v256 a) {
+  v256 dst;
+#include "vext2xv_du_bu.h"
+  return dst;
+}
+
+void test() { XFUZZ1(vext2xv_du_bu); }

code/vext2xv_du_hu.cpp

@@ -0,0 +1,9 @@
+#include "common.h"
+
+v256 vext2xv_du_hu(v256 a) {
+  v256 dst;
+#include "vext2xv_du_hu.h"
+  return dst;
+}
+
+void test() { XFUZZ1(vext2xv_du_hu); }

code/vext2xv_w_b.cpp

@@ -0,0 +1,9 @@
+#include "common.h"
+
+v256 vext2xv_w_b(v256 a) {
+  v256 dst;
+#include "vext2xv_w_b.h"
+  return dst;
+}
+
+void test() { XFUZZ1(vext2xv_w_b); }

code/vext2xv_wu_bu.cpp

@@ -0,0 +1,9 @@
+#include "common.h"
+
+v256 vext2xv_wu_bu(v256 a) {
+  v256 dst;
+#include "vext2xv_wu_bu.h"
+  return dst;
+}
+
+void test() { XFUZZ1(vext2xv_wu_bu); }

code/vshuf_h.h

@@ -7,4 +7,4 @@ for (int i = 0; i < 8; i++) {
   } else {
     dst.half[i] = b.half[a.half[i] % 8];
   }
-}
+}

code/xvshuf_b.h

@@ -1,7 +1,9 @@
 for (int i = 0; i < 32; i++) {
-  if ((c.byte[i] % 64) < 32) {
-    dst.byte[i] = b.byte[c.byte[i] % 32];
+  if (c.byte[i] >= 64) {
+    dst.byte[i] = 0;
+  } else if ((c.byte[i] % 32) < 16) {
+    dst.byte[i] = b.byte[(c.byte[i] % 32) + ((i >= 16) ? 16 : 0)];
   } else {
-    dst.byte[i] = a.byte[c.byte[i] % 32];
+    dst.byte[i] = a.byte[(c.byte[i] % 32) + ((i >= 16) ? 0 : -16)];
   }
-}
+}

code/xvshuf_h.h

@@ -1,7 +1,9 @@
 for (int i = 0; i < 16; i++) {
-  if ((a.half[i] % 32) < 16) {
-    dst.half[i] = c.half[a.half[i] % 16];
+  if ((c.half[i] % 256) >= 64) {
+    dst.half[i] = 0;
+  } else if ((c.half[i] % 16) < 8) {
+    dst.half[i] = b.half[(c.half[i] % 16) + ((i >= 8) ? 8 : 0)];
   } else {
-    dst.half[i] = b.half[a.half[i] % 16];
+    dst.half[i] = a.half[(c.half[i] % 16) + ((i >= 8) ? 0 : -8)];
   }
-}
+}