From de7711971b41d9d2c0ff604cd0eff3fa526530f5 Mon Sep 17 00:00:00 2001
From: Jiajie Chen <c@jia.je>
Date: Thu, 14 Dec 2023 13:13:04 +0800
Subject: [PATCH] Add vext2xv testcases, try to fix xvshuf

---
 code/gen_tb.py         |  3 ++-
 code/vext2xv_d_b.cpp   |  9 +++++++++
 code/vext2xv_d_h.cpp   |  9 +++++++++
 code/vext2xv_du_bu.cpp |  9 +++++++++
 code/vext2xv_du_hu.cpp |  9 +++++++++
 code/vext2xv_w_b.cpp   |  9 +++++++++
 code/vext2xv_wu_bu.cpp |  9 +++++++++
 code/vshuf_h.h         |  2 +-
 code/xvshuf_b.h        | 10 ++++++----
 code/xvshuf_h.h        | 10 ++++++----
 10 files changed, 69 insertions(+), 10 deletions(-)
 create mode 100644 code/vext2xv_d_b.cpp
 create mode 100644 code/vext2xv_d_h.cpp
 create mode 100644 code/vext2xv_du_bu.cpp
 create mode 100644 code/vext2xv_du_hu.cpp
 create mode 100644 code/vext2xv_w_b.cpp
 create mode 100644 code/vext2xv_wu_bu.cpp

diff --git a/code/gen_tb.py b/code/gen_tb.py
index b1f501d0..d572749d 100644
--- a/code/gen_tb.py
+++ b/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"),
diff --git a/code/vext2xv_d_b.cpp b/code/vext2xv_d_b.cpp
new file mode 100644
index 00000000..5ad738f8
--- /dev/null
+++ 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); }
diff --git a/code/vext2xv_d_h.cpp b/code/vext2xv_d_h.cpp
new file mode 100644
index 00000000..6b28ec00
--- /dev/null
+++ 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); }
diff --git a/code/vext2xv_du_bu.cpp b/code/vext2xv_du_bu.cpp
new file mode 100644
index 00000000..001f595e
--- /dev/null
+++ b/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); }
diff --git a/code/vext2xv_du_hu.cpp b/code/vext2xv_du_hu.cpp
new file mode 100644
index 00000000..2b7b42ab
--- /dev/null
+++ b/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); }
diff --git a/code/vext2xv_w_b.cpp b/code/vext2xv_w_b.cpp
new file mode 100644
index 00000000..45a76a78
--- /dev/null
+++ b/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); }
diff --git a/code/vext2xv_wu_bu.cpp b/code/vext2xv_wu_bu.cpp
new file mode 100644
index 00000000..7dd67937
--- /dev/null
+++ 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); }
diff --git a/code/vshuf_h.h b/code/vshuf_h.h
index 9b6d7dc5..bea8c20e 100644
--- a/code/vshuf_h.h
+++ b/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];
   }
-}
\ No newline at end of file
+}
diff --git a/code/xvshuf_b.h b/code/xvshuf_b.h
index ed70425d..7befd03c 100644
--- a/code/xvshuf_b.h
+++ b/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)];
   }
-}
\ No newline at end of file
+}
diff --git a/code/xvshuf_h.h b/code/xvshuf_h.h
index 0a337092..88c5ab55 100644
--- a/code/xvshuf_h.h
+++ b/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)];
   }
-}
\ No newline at end of file
+}