Add vavgr

README.md

@@ -74,9 +74,6 @@ Vector Half Add Unsigned
 
 Vector Half Sub Unsigned
 
-### vavgr.b/h/w/d
-### vavgr.bu/hu/wu/du
-
 ### vmax.b/h/w/d
 
 Vector Maximum

code/gen_impl.py

@@ -33,4 +33,8 @@
     with open(f"vavg_{width}.h", 'w') as f:
         print(f'for (int i = 0;i < {128 // w};i++) {{', file=f)
         print(f'  dst.{m}[i] = (({sign}{w})a.{m}[i] >> 1) + (({sign}{w})b.{m}[i] >> 1) + (a.{m}[i] & b.{m}[i] & 1);', file=f)
+        print(f'}}', file=f)
+    with open(f"vavgr_{width}.h", 'w') as f:
+        print(f'for (int i = 0;i < {128 // w};i++) {{', file=f)
+        print(f'  dst.{m}[i] = (({sign}{w})a.{m}[i] >> 1) + (({sign}{w})b.{m}[i] >> 1) + ((a.{m}[i] | b.{m}[i]) & 1);', file=f)
         print(f'}}', file=f)

code/gen_tb.py

@@ -1,7 +1,8 @@
 widths_all = ["b", "bu", "h", "hu", "w", "wu", "d", "du"]
 
 tb = {
-    "vavg": (widths_all, "v128 a, v128 b")
+    "vavg": (widths_all, "v128 a, v128 b"),
+    "vavgr": (widths_all, "v128 a, v128 b")
 }
 
 for name in tb:

code/vavgr_b.cpp

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

code/vavgr_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/vavgr_bu.cpp

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

code/vavgr_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/vavgr_d.cpp

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

code/vavgr_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/vavgr_du.cpp

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

code/vavgr_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/vavgr_h.cpp

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

code/vavgr_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/vavgr_hu.cpp

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

code/vavgr_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/vavgr_w.cpp

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

code/vavgr_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/vavgr_wu.cpp

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

code/vavgr_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

@@ -46,3 +46,12 @@
 {{ vavg('wu') }}
 {{ vavg('d') }}
 {{ vavg('du') }}
+
+{{ vavgr('b') }}
+{{ vavgr('bu') }}
+{{ vavgr('h') }}
+{{ vavgr('hu') }}
+{{ vavgr('w') }}
+{{ vavgr('wu') }}
+{{ vavgr('d') }}
+{{ vavgr('du') }}

main.py

@@ -124,7 +124,17 @@ def vavg(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`.",
+            desc=f"Compute the average (rounded towards negative infinity) of {signedness} {width}-bit elements in `a` and `b`, save the result in `dst`.",
+        )
+
+    @env.macro
+    def vavgr(name):
+        width = widths[name]
+        signedness = signednesses[name]
+        return instruction(
+            intrinsic=f"__m128i __lsx_vavgr_{name} (__m128i a, __m128i b)",
+            instr=f"vavgr.{name} vr, vr, vr",
+            desc=f"Compute the average (rounded towards positive infinity) of {signedness} {width}-bit elements in `a` and `b`, save the result in `dst`.",
         )
 
     @env.macro