SIMD {i8x16,i16x8}.avgr_u instructions

scripts/gen-s-parser.py

@@ -375,6 +375,7 @@
     ("i8x16.min_u",          "makeBinary(s, BinaryOp::MinUVecI8x16)"),
     ("i8x16.max_s",          "makeBinary(s, BinaryOp::MaxSVecI8x16)"),
     ("i8x16.max_u",          "makeBinary(s, BinaryOp::MaxUVecI8x16)"),
+    ("i8x16.avgr_u",         "makeBinary(s, BinaryOp::AvgrUVecI8x16)"),
     ("i16x8.neg",            "makeUnary(s, UnaryOp::NegVecI16x8)"),
     ("i16x8.any_true",       "makeUnary(s, UnaryOp::AnyTrueVecI16x8)"),
     ("i16x8.all_true",       "makeUnary(s, UnaryOp::AllTrueVecI16x8)"),
@@ -392,6 +393,7 @@
     ("i16x8.min_u",          "makeBinary(s, BinaryOp::MinUVecI16x8)"),
     ("i16x8.max_s",          "makeBinary(s, BinaryOp::MaxSVecI16x8)"),
     ("i16x8.max_u",          "makeBinary(s, BinaryOp::MaxUVecI16x8)"),
+    ("i16x8.avgr_u",         "makeBinary(s, BinaryOp::AvgrUVecI16x8)"),
     ("i32x4.neg",            "makeUnary(s, UnaryOp::NegVecI32x4)"),
     ("i32x4.any_true",       "makeUnary(s, UnaryOp::AnyTrueVecI32x4)"),
     ("i32x4.all_true",       "makeUnary(s, UnaryOp::AllTrueVecI32x4)"),

src/binaryen-c.cpp

@@ -780,6 +780,7 @@ BinaryenOp BinaryenMinSVecI8x16(void) { return MinSVecI8x16; }
 BinaryenOp BinaryenMinUVecI8x16(void) { return MinUVecI8x16; }
 BinaryenOp BinaryenMaxSVecI8x16(void) { return MaxSVecI8x16; }
 BinaryenOp BinaryenMaxUVecI8x16(void) { return MaxUVecI8x16; }
+BinaryenOp BinaryenAvgrUVecI8x16(void) { return AvgrUVecI8x16; }
 BinaryenOp BinaryenNegVecI16x8(void) { return NegVecI16x8; }
 BinaryenOp BinaryenAnyTrueVecI16x8(void) { return AnyTrueVecI16x8; }
 BinaryenOp BinaryenAllTrueVecI16x8(void) { return AllTrueVecI16x8; }
@@ -797,6 +798,7 @@ BinaryenOp BinaryenMinSVecI16x8(void) { return MinSVecI16x8; }
 BinaryenOp BinaryenMinUVecI16x8(void) { return MinUVecI16x8; }
 BinaryenOp BinaryenMaxSVecI16x8(void) { return MaxSVecI16x8; }
 BinaryenOp BinaryenMaxUVecI16x8(void) { return MaxUVecI16x8; }
+BinaryenOp BinaryenAvgrUVecI16x8(void) { return AvgrUVecI16x8; }
 BinaryenOp BinaryenNegVecI32x4(void) { return NegVecI32x4; }
 BinaryenOp BinaryenAnyTrueVecI32x4(void) { return AnyTrueVecI32x4; }
 BinaryenOp BinaryenAllTrueVecI32x4(void) { return AllTrueVecI32x4; }

src/binaryen-c.h

@@ -474,6 +474,7 @@ BINARYEN_API BinaryenOp BinaryenMinSVecI8x16(void);
 BINARYEN_API BinaryenOp BinaryenMinUVecI8x16(void);
 BINARYEN_API BinaryenOp BinaryenMaxSVecI8x16(void);
 BINARYEN_API BinaryenOp BinaryenMaxUVecI8x16(void);
+BINARYEN_API BinaryenOp BinaryenAvgrUVecI8x16(void);
 BINARYEN_API BinaryenOp BinaryenNegVecI16x8(void);
 BINARYEN_API BinaryenOp BinaryenAnyTrueVecI16x8(void);
 BINARYEN_API BinaryenOp BinaryenAllTrueVecI16x8(void);
@@ -491,6 +492,7 @@ BINARYEN_API BinaryenOp BinaryenMinSVecI16x8(void);
 BINARYEN_API BinaryenOp BinaryenMinUVecI16x8(void);
 BINARYEN_API BinaryenOp BinaryenMaxSVecI16x8(void);
 BINARYEN_API BinaryenOp BinaryenMaxUVecI16x8(void);
+BINARYEN_API BinaryenOp BinaryenAvgrUVecI16x8(void);
 BINARYEN_API BinaryenOp BinaryenNegVecI32x4(void);
 BINARYEN_API BinaryenOp BinaryenAnyTrueVecI32x4(void);
 BINARYEN_API BinaryenOp BinaryenAllTrueVecI32x4(void);

src/gen-s-parser.inc

@@ -698,6 +698,9 @@ switch (op[0]) {
               case 'n':
                 if (strcmp(op, "i16x8.any_true") == 0) { return makeUnary(s, UnaryOp::AnyTrueVecI16x8); }
                 goto parse_error;
+              case 'v':
+                if (strcmp(op, "i16x8.avgr_u") == 0) { return makeBinary(s, BinaryOp::AvgrUVecI16x8); }
+                goto parse_error;
               default: goto parse_error;
             }
           }
@@ -2241,6 +2244,9 @@ switch (op[0]) {
               case 'n':
                 if (strcmp(op, "i8x16.any_true") == 0) { return makeUnary(s, UnaryOp::AnyTrueVecI8x16); }
                 goto parse_error;
+              case 'v':
+                if (strcmp(op, "i8x16.avgr_u") == 0) { return makeBinary(s, BinaryOp::AvgrUVecI8x16); }
+                goto parse_error;
               default: goto parse_error;
             }
           }

src/ir/cost.h

@@ -594,6 +594,9 @@ struct CostAnalyzer : public Visitor<CostAnalyzer, Index> {
       case MaxUVecI8x16:
         ret = 1;
         break;
+      case AvgrUVecI8x16:
+        ret = 1;
+        break;
       case AddVecI16x8:
         ret = 1;
         break;
@@ -627,6 +630,9 @@ struct CostAnalyzer : public Visitor<CostAnalyzer, Index> {
       case MaxUVecI16x8:
         ret = 1;
         break;
+      case AvgrUVecI16x8:
+        ret = 1;
+        break;
       case AddVecI32x4:
         ret = 1;
         break;

src/js/binaryen.js-post.js

@@ -359,6 +359,7 @@ Module['MinSVecI8x16'] = Module['_BinaryenMinSVecI8x16']();
 Module['MinUVecI8x16'] = Module['_BinaryenMinUVecI8x16']();
 Module['MaxSVecI8x16'] = Module['_BinaryenMaxSVecI8x16']();
 Module['MaxUVecI8x16'] = Module['_BinaryenMaxUVecI8x16']();
+Module['AvgrUVecI8x16'] = Module['_BinaryenAvgrUVecI8x16']();
 Module['NegVecI16x8'] = Module['_BinaryenNegVecI16x8']();
 Module['AnyTrueVecI16x8'] = Module['_BinaryenAnyTrueVecI16x8']();
 Module['AllTrueVecI16x8'] = Module['_BinaryenAllTrueVecI16x8']();
@@ -376,6 +377,7 @@ Module['MinSVecI16x8'] = Module['_BinaryenMinSVecI16x8']();
 Module['MinUVecI16x8'] = Module['_BinaryenMinUVecI16x8']();
 Module['MaxSVecI16x8'] = Module['_BinaryenMaxSVecI16x8']();
 Module['MaxUVecI16x8'] = Module['_BinaryenMaxUVecI16x8']();
+Module['AvgrUVecI16x8'] = Module['_BinaryenAvgrUVecI16x8']();
 Module['DotSVecI16x8ToVecI32x4'] = Module['_BinaryenDotSVecI16x8ToVecI32x4']();
 Module['NegVecI32x4'] = Module['_BinaryenNegVecI32x4']();
 Module['AnyTrueVecI32x4'] = Module['_BinaryenAnyTrueVecI32x4']();
@@ -1492,6 +1494,9 @@ function wrapModule(module, self) {
     'max_u': function(left, right) {
       return Module['_BinaryenBinary'](module, Module['MaxUVecI8x16'], left, right);
     },
+    'avgr_u': function(left, right) {
+      return Module['_BinaryenBinary'](module, Module['AvgrUVecI8x16'], left, right);
+    },
     'narrow_i16x8_s': function(left, right) {
       return Module['_BinaryenBinary'](module, Module['NarrowSVecI16x8ToVecI8x16'], left, right);
     },
@@ -1594,6 +1599,9 @@ function wrapModule(module, self) {
     'max_u': function(left, right) {
       return Module['_BinaryenBinary'](module, Module['MaxUVecI16x8'], left, right);
     },
+    'avgr_u': function(left, right) {
+      return Module['_BinaryenBinary'](module, Module['AvgrUVecI16x8'], left, right);
+    },
     'narrow_i32x4_s': function(left, right) {
       return Module['_BinaryenBinary'](module, Module['NarrowSVecI32x4ToVecI16x8'], left, right);
     },

src/literal.h

@@ -334,6 +334,7 @@ class Literal {
   Literal minUI8x16(const Literal& other) const;
   Literal maxSI8x16(const Literal& other) const;
   Literal maxUI8x16(const Literal& other) const;
+  Literal avgrUI8x16(const Literal& other) const;
   Literal negI16x8() const;
   Literal anyTrueI16x8() const;
   Literal allTrueI16x8() const;
@@ -351,6 +352,7 @@ class Literal {
   Literal minUI16x8(const Literal& other) const;
   Literal maxSI16x8(const Literal& other) const;
   Literal maxUI16x8(const Literal& other) const;
+  Literal avgrUI16x8(const Literal& other) const;
   Literal negI32x4() const;
   Literal anyTrueI32x4() const;
   Literal allTrueI32x4() const;
@@ -426,6 +428,7 @@ class Literal {
   Literal maxInt(const Literal& other) const;
   Literal minUInt(const Literal& other) const;
   Literal maxUInt(const Literal& other) const;
+  Literal avgrUInt(const Literal& other) const;
 };
 
 } // namespace wasm

src/passes/Print.cpp

@@ -1204,6 +1204,9 @@ struct PrintExpressionContents
       case MaxUVecI8x16:
         o << "i8x16.max_u";
         break;
+      case AvgrUVecI8x16:
+        o << "i8x16.avgr_u";
+        break;
       case AddVecI16x8:
         o << "i16x8.add";
         break;
@@ -1237,6 +1240,9 @@ struct PrintExpressionContents
       case MaxUVecI16x8:
         o << "i16x8.max_u";
         break;
+      case AvgrUVecI16x8:
+        o << "i16x8.avgr_u";
+        break;
       case AddVecI32x4:
         o << "i32x4.add";
         break;

src/wasm-binary.h

@@ -765,6 +765,7 @@ enum ASTNodes {
   I8x16MinU = 0x5f,
   I8x16MaxS = 0x60,
   I8x16MaxU = 0x61,
+  I8x16AvgrU = 0xd9,
   I16x8Neg = 0x62,
   I16x8AnyTrue = 0x63,
   I16x8AllTrue = 0x64,
@@ -782,6 +783,7 @@ enum ASTNodes {
   I16x8MinU = 0x70,
   I16x8MaxS = 0x71,
   I16x8MaxU = 0x72,
+  I16x8AvgrU = 0xda,
   I32x4Neg = 0x73,
   I32x4AnyTrue = 0x74,
   I32x4AllTrue = 0x75,
@@ -795,7 +797,7 @@ enum ASTNodes {
   I32x4MinU = 0x81,
   I32x4MaxS = 0x82,
   I32x4MaxU = 0x83,
-  I32x4DotSVecI16x8 = 0xd9,
+  I32x4DotSVecI16x8 = 0xdb,
   I64x2Neg = 0x84,
   I64x2AnyTrue = 0x85,
   I64x2AllTrue = 0x86,

src/wasm-interpreter.h

@@ -765,6 +765,8 @@ class ExpressionRunner : public OverriddenVisitor<SubType, Flow> {
         return left.maxSI8x16(right);
       case MaxUVecI8x16:
         return left.maxUI8x16(right);
+      case AvgrUVecI8x16:
+        return left.avgrUI8x16(right);
       case AddVecI16x8:
         return left.addI16x8(right);
       case AddSatSVecI16x8:
@@ -787,6 +789,8 @@ class ExpressionRunner : public OverriddenVisitor<SubType, Flow> {
         return left.maxSI16x8(right);
       case MaxUVecI16x8:
         return left.maxUI16x8(right);
+      case AvgrUVecI16x8:
+        return left.avgrUI16x8(right);
       case AddVecI32x4:
         return left.addI32x4(right);
       case SubVecI32x4:

src/wasm.h

@@ -365,6 +365,7 @@ enum BinaryOp {
   MinUVecI8x16,
   MaxSVecI8x16,
   MaxUVecI8x16,
+  AvgrUVecI8x16,
   AddVecI16x8,
   AddSatSVecI16x8,
   AddSatUVecI16x8,
@@ -376,6 +377,7 @@ enum BinaryOp {
   MinUVecI16x8,
   MaxSVecI16x8,
   MaxUVecI16x8,
+  AvgrUVecI16x8,
   AddVecI32x4,
   SubVecI32x4,
   MulVecI32x4,

src/wasm/literal.cpp

@@ -860,6 +860,10 @@ Literal Literal::maxUInt(const Literal& other) const {
   return uint32_t(geti32()) > uint32_t(other.geti32()) ? *this : other;
 }
 
+Literal Literal::avgrUInt(const Literal& other) const {
+  return Literal((geti32() + other.geti32() + 1) / 2);
+}
+
 Literal Literal::and_(const Literal& other) const {
   switch (type) {
     case Type::i32:
@@ -1729,6 +1733,9 @@ Literal Literal::maxSI8x16(const Literal& other) const {
 Literal Literal::maxUI8x16(const Literal& other) const {
   return binary<16, &Literal::getLanesUI8x16, &Literal::maxInt>(*this, other);
 }
+Literal Literal::avgrUI8x16(const Literal& other) const {
+  return binary<16, &Literal::getLanesUI8x16, &Literal::avgrUInt>(*this, other);
+}
 Literal Literal::addI16x8(const Literal& other) const {
   return binary<8, &Literal::getLanesUI16x8, &Literal::add>(*this, other);
 }
@@ -1766,6 +1773,9 @@ Literal Literal::maxSI16x8(const Literal& other) const {
 Literal Literal::maxUI16x8(const Literal& other) const {
   return binary<8, &Literal::getLanesUI16x8, &Literal::maxInt>(*this, other);
 }
+Literal Literal::avgrUI16x8(const Literal& other) const {
+  return binary<8, &Literal::getLanesUI16x8, &Literal::avgrUInt>(*this, other);
+}
 Literal Literal::addI32x4(const Literal& other) const {
   return binary<4, &Literal::getLanesI32x4, &Literal::add>(*this, other);
 }

src/wasm/wasm-binary.cpp

@@ -3589,6 +3589,10 @@ bool WasmBinaryBuilder::maybeVisitSIMDBinary(Expression*& out, uint32_t code) {
       curr = allocator.alloc<Binary>();
       curr->op = MaxUVecI8x16;
       break;
+    case BinaryConsts::I8x16AvgrU:
+      curr = allocator.alloc<Binary>();
+      curr->op = AvgrUVecI8x16;
+      break;
     case BinaryConsts::I16x8Add:
       curr = allocator.alloc<Binary>();
       curr->op = AddVecI16x8;
@@ -3633,6 +3637,10 @@ bool WasmBinaryBuilder::maybeVisitSIMDBinary(Expression*& out, uint32_t code) {
       curr = allocator.alloc<Binary>();
       curr->op = MaxUVecI16x8;
       break;
+    case BinaryConsts::I16x8AvgrU:
+      curr = allocator.alloc<Binary>();
+      curr->op = AvgrUVecI16x8;
+      break;
     case BinaryConsts::I32x4Add:
       curr = allocator.alloc<Binary>();
       curr->op = AddVecI32x4;

src/wasm/wasm-stack.cpp

@@ -1401,6 +1401,9 @@ void BinaryInstWriter::visitBinary(Binary* curr) {
     case MaxUVecI8x16:
       o << int8_t(BinaryConsts::SIMDPrefix) << U32LEB(BinaryConsts::I8x16MaxU);
       break;
+    case AvgrUVecI8x16:
+      o << int8_t(BinaryConsts::SIMDPrefix) << U32LEB(BinaryConsts::I8x16AvgrU);
+      break;
     case AddVecI16x8:
       o << int8_t(BinaryConsts::SIMDPrefix) << U32LEB(BinaryConsts::I16x8Add);
       break;
@@ -1438,6 +1441,9 @@ void BinaryInstWriter::visitBinary(Binary* curr) {
     case MaxUVecI16x8:
       o << int8_t(BinaryConsts::SIMDPrefix) << U32LEB(BinaryConsts::I16x8MaxU);
       break;
+    case AvgrUVecI16x8:
+      o << int8_t(BinaryConsts::SIMDPrefix) << U32LEB(BinaryConsts::I16x8AvgrU);
+      break;
     case AddVecI32x4:
       o << int8_t(BinaryConsts::SIMDPrefix) << U32LEB(BinaryConsts::I32x4Add);
       break;

src/wasm/wasm-validator.cpp

@@ -1344,6 +1344,7 @@ void FunctionValidator::visitBinary(Binary* curr) {
     case MinUVecI8x16:
     case MaxSVecI8x16:
     case MaxUVecI8x16:
+    case AvgrUVecI8x16:
     case AddVecI16x8:
     case AddSatSVecI16x8:
     case AddSatUVecI16x8:
@@ -1355,6 +1356,7 @@ void FunctionValidator::visitBinary(Binary* curr) {
     case MinUVecI16x8:
     case MaxSVecI16x8:
     case MaxUVecI16x8:
+    case AvgrUVecI16x8:
     case AddVecI32x4:
     case SubVecI32x4:
     case MulVecI32x4:

test/binaryen.js/kitchen-sink.js

@@ -355,6 +355,7 @@ function test_core() {
     module.i8x16.min_u(module.v128.const(v128_bytes), module.v128.const(v128_bytes)),
     module.i8x16.max_s(module.v128.const(v128_bytes), module.v128.const(v128_bytes)),
     module.i8x16.max_u(module.v128.const(v128_bytes), module.v128.const(v128_bytes)),
+    module.i8x16.avgr_u(module.v128.const(v128_bytes), module.v128.const(v128_bytes)),
     module.i16x8.add(module.v128.const(v128_bytes), module.v128.const(v128_bytes)),
     module.i16x8.add_saturate_s(module.v128.const(v128_bytes), module.v128.const(v128_bytes)),
     module.i16x8.add_saturate_u(module.v128.const(v128_bytes), module.v128.const(v128_bytes)),
@@ -366,6 +367,7 @@ function test_core() {
     module.i16x8.min_u(module.v128.const(v128_bytes), module.v128.const(v128_bytes)),
     module.i16x8.max_s(module.v128.const(v128_bytes), module.v128.const(v128_bytes)),
     module.i16x8.max_u(module.v128.const(v128_bytes), module.v128.const(v128_bytes)),
+    module.i16x8.avgr_u(module.v128.const(v128_bytes), module.v128.const(v128_bytes)),
     module.i32x4.add(module.v128.const(v128_bytes), module.v128.const(v128_bytes)),
     module.i32x4.sub(module.v128.const(v128_bytes), module.v128.const(v128_bytes)),
     module.i32x4.mul(module.v128.const(v128_bytes), module.v128.const(v128_bytes)),