v8x16.swizzle (#2368)

As specified at
https://github.com/WebAssembly/simd/blob/master/proposals/simd/SIMD.md#swizzling-using-variable-indices.

15 files changed, 56 insertions, 5 deletions

diff --git a/src/binaryen-c.cpp b/src/binaryen-c.cpp
index a4b3e9b5b..7b4ee2bc5 100644
--- a/src/binaryen-c.cpp
+++ b/src/binaryen-c.cpp
@@ -942,6 +942,7 @@ BinaryenOp BinaryenWidenLowUVecI16x8ToVecI32x4(void) {
 BinaryenOp BinaryenWidenHighUVecI16x8ToVecI32x4(void) {
   return WidenHighUVecI16x8ToVecI32x4;
 }
+BinaryenOp BinaryenSwizzleVec8x16(void) { return SwizzleVec8x16; }
 
 BinaryenExpressionRef BinaryenBlock(BinaryenModuleRef module,
                                     const char* name,
diff --git a/src/binaryen-c.h b/src/binaryen-c.h
index a31df8cae..3d15a7efe 100644
--- a/src/binaryen-c.h
+++ b/src/binaryen-c.h
@@ -546,6 +546,7 @@ BINARYEN_API BinaryenOp BinaryenWidenLowSVecI16x8ToVecI32x4(void);
 BINARYEN_API BinaryenOp BinaryenWidenHighSVecI16x8ToVecI32x4(void);
 BINARYEN_API BinaryenOp BinaryenWidenLowUVecI16x8ToVecI32x4(void);
 BINARYEN_API BinaryenOp BinaryenWidenHighUVecI16x8ToVecI32x4(void);
+BINARYEN_API BinaryenOp BinaryenSwizzleVec8x16(void);
 
 typedef void* BinaryenExpressionRef;
 
diff --git a/src/gen-s-parser.inc b/src/gen-s-parser.inc
index bab468e2a..7f1958582 100644
--- a/src/gen-s-parser.inc
+++ b/src/gen-s-parser.inc
@@ -2509,9 +2509,17 @@ switch (op[0]) {
           case 'l':
             if (strcmp(op, "v8x16.load_splat") == 0) { return makeSIMDLoad(s, SIMDLoadOp::LoadSplatVec8x16); }
             goto parse_error;
-          case 's':
-            if (strcmp(op, "v8x16.shuffle") == 0) { return makeSIMDShuffle(s); }
-            goto parse_error;
+          case 's': {
+            switch (op[7]) {
+              case 'h':
+                if (strcmp(op, "v8x16.shuffle") == 0) { return makeSIMDShuffle(s); }
+                goto parse_error;
+              case 'w':
+                if (strcmp(op, "v8x16.swizzle") == 0) { return makeBinary(s, BinaryOp::SwizzleVec8x16); }
+                goto parse_error;
+              default: goto parse_error;
+            }
+          }
           default: goto parse_error;
         }
       }
diff --git a/src/ir/cost.h b/src/ir/cost.h
index a4d9a0d40..a3d9f8584 100644
--- a/src/ir/cost.h
+++ b/src/ir/cost.h
@@ -666,6 +666,9 @@ struct CostAnalyzer : public Visitor<CostAnalyzer, Index> {
       case NarrowUVecI32x4ToVecI16x8:
         ret = 1;
         break;
+      case SwizzleVec8x16:
+        ret = 1;
+        break;
       case InvalidBinary:
         WASM_UNREACHABLE();
     }
diff --git a/src/js/binaryen.js-post.js b/src/js/binaryen.js-post.js
index 8769e1c9e..b11d2769b 100644
--- a/src/js/binaryen.js-post.js
+++ b/src/js/binaryen.js-post.js
@@ -418,6 +418,7 @@ Module['WidenLowSVecI16x8ToVecI32x4'] = Module['_BinaryenWidenLowSVecI16x8ToVecI
 Module['WidenHighSVecI16x8ToVecI32x4'] = Module['_BinaryenWidenHighSVecI16x8ToVecI32x4']();
 Module['WidenLowUVecI16x8ToVecI32x4'] = Module['_BinaryenWidenLowUVecI16x8ToVecI32x4']();
 Module['WidenHighUVecI16x8ToVecI32x4'] = Module['_BinaryenWidenHighUVecI16x8ToVecI32x4']();
+Module['SwizzleVec8x16'] = Module['_BinaryenSwizzleVec8x16']();
 
 // The size of a single literal in memory as used in Const creation,
 // which is a little different: we don't want users to need to make
@@ -1845,6 +1846,9 @@ function wrapModule(module, self) {
         return Module['_BinaryenSIMDShuffle'](module, left, right, i8sToStack(mask));
       });
     },
+    'swizzle': function(left, right) {
+      return Module['_BinaryenBinary'](module, Module['SwizzleVec8x16'], left, right);
+    },
     'load_splat': function(offset, align, ptr) {
       return Module['_BinaryenSIMDLoad'](module, Module['LoadSplatVec8x16'], offset, align, ptr);
     },
diff --git a/src/literal.h b/src/literal.h
index 98bb1aa7b..19db6a1d6 100644
--- a/src/literal.h
+++ b/src/literal.h
@@ -398,6 +398,7 @@ public:
   Literal widenHighSToVecI32x4() const;
   Literal widenLowUToVecI32x4() const;
   Literal widenHighUToVecI32x4() const;
+  Literal swizzleVec8x16(const Literal& other) const;
 
 private:
   Literal addSatSI8(const Literal& other) const;
diff --git a/src/passes/Print.cpp b/src/passes/Print.cpp
index b856b2000..a5b5e3c2b 100644
--- a/src/passes/Print.cpp
+++ b/src/passes/Print.cpp
@@ -1243,6 +1243,10 @@ struct PrintExpressionContents
         o << "i16x8.narrow_i32x4_u";
         break;
 
+      case SwizzleVec8x16:
+        o << "v8x16.swizzle";
+        break;
+
       case InvalidBinary:
         WASM_UNREACHABLE();
     }
diff --git a/src/tools/fuzzing.h b/src/tools/fuzzing.h
index 45ddc59b1..c4070427c 100644
--- a/src/tools/fuzzing.h
+++ b/src/tools/fuzzing.h
@@ -2174,7 +2174,8 @@ private:
                                  NarrowSVecI16x8ToVecI8x16,
                                  NarrowUVecI16x8ToVecI8x16,
                                  NarrowSVecI32x4ToVecI16x8,
-                                 NarrowUVecI32x4ToVecI16x8),
+                                 NarrowUVecI32x4ToVecI16x8,
+                                 SwizzleVec8x16),
                             make(v128),
                             make(v128)});
       }
diff --git a/src/wasm-binary.h b/src/wasm-binary.h
index f299099b2..104e7b208 100644
--- a/src/wasm-binary.h
+++ b/src/wasm-binary.h
@@ -879,6 +879,7 @@ enum ASTNodes {
   I32x4LoadExtUVec16x4 = 0xd5,
   I64x2LoadExtSVec32x2 = 0xd6,
   I64x2LoadExtUVec32x2 = 0xd7,
+  V8x16Swizzle = 0xc0,
 
   // bulk memory opcodes
 
diff --git a/src/wasm-interpreter.h b/src/wasm-interpreter.h
index 8c429caa9..5995808a5 100644
--- a/src/wasm-interpreter.h
+++ b/src/wasm-interpreter.h
@@ -816,6 +816,9 @@ public:
       case NarrowUVecI32x4ToVecI16x8:
         return left.narrowUToVecI16x8(right);
 
+      case SwizzleVec8x16:
+        return left.swizzleVec8x16(right);
+
       case InvalidBinary:
         WASM_UNREACHABLE();
     }
diff --git a/src/wasm.h b/src/wasm.h
index 8006bd319..d4b53f5bb 100644
--- a/src/wasm.h
+++ b/src/wasm.h
@@ -392,6 +392,9 @@ enum BinaryOp {
   NarrowSVecI32x4ToVecI16x8,
   NarrowUVecI32x4ToVecI16x8,
 
+  // SIMD Swizzle
+  SwizzleVec8x16,
+
   InvalidBinary
 };
 
diff --git a/src/wasm/literal.cpp b/src/wasm/literal.cpp
index b382d1a3d..7f0c9fa6b 100644
--- a/src/wasm/literal.cpp
+++ b/src/wasm/literal.cpp
@@ -1867,4 +1867,15 @@ Literal Literal::widenHighUToVecI32x4() const {
   return widen<4, &Literal::getLanesUI16x8, LaneOrder::High>(*this);
 }
 
+Literal Literal::swizzleVec8x16(const Literal& other) const {
+  auto lanes = getLanesUI8x16();
+  auto indices = other.getLanesUI8x16();
+  LaneArray<16> result;
+  for (size_t i = 0; i < 16; ++i) {
+    size_t index = indices[i].geti32();
+    result[i] = index >= 16 ? Literal(int32_t(0)) : lanes[index];
+  }
+  return Literal(result);
+}
+
 } // namespace wasm
diff --git a/src/wasm/wasm-binary.cpp b/src/wasm/wasm-binary.cpp
index 1606d74a1..8446f3c17 100644
--- a/src/wasm/wasm-binary.cpp
+++ b/src/wasm/wasm-binary.cpp
@@ -3963,6 +3963,10 @@ bool WasmBinaryBuilder::maybeVisitSIMDBinary(Expression*& out, uint32_t code) {
       curr = allocator.alloc<Binary>();
       curr->op = NarrowUVecI32x4ToVecI16x8;
       break;
+    case BinaryConsts::V8x16Swizzle:
+      curr = allocator.alloc<Binary>();
+      curr->op = SwizzleVec8x16;
+      break;
     default:
       return false;
   }
diff --git a/src/wasm/wasm-stack.cpp b/src/wasm/wasm-stack.cpp
index 8b7ccfb2e..949bf1364 100644
--- a/src/wasm/wasm-stack.cpp
+++ b/src/wasm/wasm-stack.cpp
@@ -1484,6 +1484,11 @@ void BinaryInstWriter::visitBinary(Binary* curr) {
         << U32LEB(BinaryConsts::I16x8NarrowUI32x4);
       break;
 
+    case SwizzleVec8x16:
+      o << int8_t(BinaryConsts::SIMDPrefix)
+        << U32LEB(BinaryConsts::V8x16Swizzle);
+      break;
+
     case InvalidBinary:
       WASM_UNREACHABLE();
   }
diff --git a/src/wasm/wasm-validator.cpp b/src/wasm/wasm-validator.cpp
index 1c30d0d2a..278ec4769 100644
--- a/src/wasm/wasm-validator.cpp
+++ b/src/wasm/wasm-validator.cpp
@@ -1366,7 +1366,8 @@ void FunctionValidator::visitBinary(Binary* curr) {
     case NarrowSVecI16x8ToVecI8x16:
     case NarrowUVecI16x8ToVecI8x16:
     case NarrowSVecI32x4ToVecI16x8:
-    case NarrowUVecI32x4ToVecI16x8: {
+    case NarrowUVecI32x4ToVecI16x8:
+    case SwizzleVec8x16: {
       shouldBeEqualOrFirstIsUnreachable(
         curr->left->type, v128, curr, "v128 op");
       shouldBeEqualOrFirstIsUnreachable(