Add vavg

README.md

@@ -74,14 +74,6 @@ Vector Half Add Unsigned
 
 Vector Half Sub Unsigned
 
-### vavg.b/h/w/d
-
-Vector Average
-
-### vavg.bu/hu/wu/du
-
-Vector Average Unsigned
-
 ### vavgr.b/h/w/d
 ### vavgr.bu/hu/wu/du
 

code/.gitignore

@@ -4,3 +4,4 @@
 !*.h
 !*.cpp
 !*.sh
+!*.py

code/gen_tb.py

@@ -0,0 +1,32 @@
+widths_all = ["b", "bu", "h", "hu", "w", "wu", "d", "du"]
+
+tb = {
+    "vavg": (widths_all, "v128 a, v128 b")
+}
+
+for name in tb:
+    t = tb[name]
+    widths = t[0]
+    args = t[1]
+
+    for width in widths:
+        inst_name = name + "_" + width
+
+        fuzz_args = 0
+        for arg in args.split(', '):
+            if 'v128' in arg:
+                fuzz_args += 1
+
+        print(f'Saving {inst_name}.cpp')
+        with open(f'{inst_name}.cpp', 'w') as f:
+            print('#include "common.h"', file=f)
+            print('', file=f)
+            print(f'v128 {inst_name}({args}) {{', file=f)
+            print('  v128 dst;', file=f)
+            print(f'#include "{inst_name}.h"', file=f)
+            print('  return dst;', file=f)
+            print('}', file=f)
+            print('', file=f)
+            print('void test() {', file=f)
+            print(f'  FUZZ{fuzz_args}({inst_name});', file=f)
+            print('}', file=f)

code/vavg_b.cpp

@@ -0,0 +1,11 @@
+#include "common.h"
+
+v128 vavg_b(v128 a, v128 b) {
+  v128 dst;
+#include "vavg_b.h"
+  return dst;
+}
+
+void test() {
+  FUZZ2(vavg_b);
+}

code/vavg_b.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 16; i++) {
+  dst.byte[i] = ((s8)a.byte[i] >> 1) + ((s8)b.byte[i] >> 1) + (a.byte[i] & b.byte[i] & 1);
+}

code/vavg_bu.cpp

@@ -0,0 +1,11 @@
+#include "common.h"
+
+v128 vavg_bu(v128 a, v128 b) {
+  v128 dst;
+#include "vavg_bu.h"
+  return dst;
+}
+
+void test() {
+  FUZZ2(vavg_bu);
+}

code/vavg_bu.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 16; i++) {
+  dst.byte[i] = ((u8)a.byte[i] >> 1) + ((u8)b.byte[i] >> 1) + (a.byte[i] & b.byte[i] & 1);
+}

code/vavg_d.cpp

@@ -0,0 +1,11 @@
+#include "common.h"
+
+v128 vavg_d(v128 a, v128 b) {
+  v128 dst;
+#include "vavg_d.h"
+  return dst;
+}
+
+void test() {
+  FUZZ2(vavg_d);
+}

code/vavg_d.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 2; i++) {
+  dst.dword[i] = ((s64)a.dword[i] >> 1) + ((s64)b.dword[i] >> 1) + (a.dword[i] & b.dword[i] & 1);
+}

code/vavg_du.cpp

@@ -0,0 +1,11 @@
+#include "common.h"
+
+v128 vavg_du(v128 a, v128 b) {
+  v128 dst;
+#include "vavg_du.h"
+  return dst;
+}
+
+void test() {
+  FUZZ2(vavg_du);
+}

code/vavg_du.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 2; i++) {
+  dst.dword[i] = ((u64)a.dword[i] >> 1) + ((u64)b.dword[i] >> 1) + (a.dword[i] & b.dword[i] & 1);
+}

code/vavg_h.cpp

@@ -0,0 +1,11 @@
+#include "common.h"
+
+v128 vavg_h(v128 a, v128 b) {
+  v128 dst;
+#include "vavg_h.h"
+  return dst;
+}
+
+void test() {
+  FUZZ2(vavg_h);
+}

code/vavg_h.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 8; i++) {
+  dst.half[i] = ((s16)a.half[i] >> 1) + ((s16)b.half[i] >> 1) + (a.half[i] & b.half[i] & 1);
+}

code/vavg_hu.cpp

@@ -0,0 +1,11 @@
+#include "common.h"
+
+v128 vavg_hu(v128 a, v128 b) {
+  v128 dst;
+#include "vavg_hu.h"
+  return dst;
+}
+
+void test() {
+  FUZZ2(vavg_hu);
+}

code/vavg_hu.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 8; i++) {
+  dst.half[i] = ((u16)a.half[i] >> 1) + ((u16)b.half[i] >> 1) + (a.half[i] & b.half[i] & 1);
+}

code/vavg_w.cpp

@@ -0,0 +1,11 @@
+#include "common.h"
+
+v128 vavg_w(v128 a, v128 b) {
+  v128 dst;
+#include "vavg_w.h"
+  return dst;
+}
+
+void test() {
+  FUZZ2(vavg_w);
+}

code/vavg_w.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 4; i++) {
+  dst.word[i] = ((s32)a.word[i] >> 1) + ((s32)b.word[i] >> 1) + (a.word[i] & b.word[i] & 1);
+}

code/vavg_wu.cpp

@@ -0,0 +1,11 @@
+#include "common.h"
+
+v128 vavg_wu(v128 a, v128 b) {
+  v128 dst;
+#include "vavg_wu.h"
+  return dst;
+}
+
+void test() {
+  FUZZ2(vavg_wu);
+}

code/vavg_wu.h

@@ -0,0 +1,3 @@
+for (int i = 0; i < 4; i++) {
+  dst.word[i] = ((u32)a.word[i] >> 1) + ((u32)b.word[i] >> 1) + (a.word[i] & b.word[i] & 1);
+}

docs/lsx_integer/vaddsub.md

@@ -1,4 +1,4 @@
-# Add/Subtract/Absolute
+# Add, Subtract, Absolute and Average
 
 {{ vadd('b') }}
 {{ vadd('h') }}
@@ -36,4 +36,13 @@
 {{ vaddwev('d', 'wu', 'w') }}
 {{ vaddwev('q', 'd') }}
 {{ vaddwev('q', 'du') }}
-{{ vaddwev('q', 'du', 'd') }}
+{{ vaddwev('q', 'du', 'd') }}
+
+{{ vavg('b') }}
+{{ vavg('bu') }}
+{{ vavg('h') }}
+{{ vavg('hu') }}
+{{ vavg('w') }}
+{{ vavg('wu') }}
+{{ vavg('d') }}
+{{ vavg('du') }}

main.py

@@ -117,6 +117,16 @@ def vaddwev(wide, narrow, narrow2=None):
             desc=f"Add even-positioned {signedness} {narrow_width}-bit elements in `a` and {signedness2} elements in `b`, save the {wide_width}-bit result in `dst`.",
         )
 
+    @env.macro
+    def vavg(name):
+        width = widths[name]
+        signedness = signednesses[name]
+        return instruction(
+            intrinsic=f"__m128i __lsx_vavg_{name} (__m128i a, __m128i b)",
+            instr=f"vavg.{name} vr, vr, vr",
+            desc=f"Compute the average of {signedness} {width}-bit elements in `a` and `b`, save the result in `dst`.",
+        )
+
     @env.macro
     def vshuf_hwd(name):
         width = widths[name]