[WASM SIMD] Implement horizontal add for float and integer types.

test/cctest/wasm/test-run-wasm-simd.cc

 // Copyright 2016 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "src/assembler-inl.h"
 #include "src/wasm/wasm-macro-gen.h"
 #include "test/cctest/cctest.h"
 #include "test/cctest/compiler/value-helper.h"
 #include "test/cctest/wasm/wasm-run-utils.h"
 
@@ skipping 569 matching lines @@
 #if V8_TARGET_ARCH_ARM || V8_TARGET_ARCH_MIPS || V8_TARGET_ARCH_MIPS64
 WASM_EXEC_COMPILED_TEST(F32x4RecipRefine) {
   RunF32x4BinOpTest(kExprF32x4RecipRefine, RecipRefine);
 }
 
 WASM_EXEC_COMPILED_TEST(F32x4RecipSqrtRefine) {
   RunF32x4BinOpTest(kExprF32x4RecipSqrtRefine, RecipSqrtRefine);
 }
 #endif  // V8_TARGET_ARCH_ARM || V8_TARGET_ARCH_MIPS || V8_TARGET_ARCH_MIPS64
 
+#if V8_TARGET_ARCH_ARM
+void RunF32x4HorizOpTest(WasmOpcode simd_op, FloatBinOp expected_op) {
+  FLAG_wasm_simd_prototype = true;
+  WasmRunner<int32_t, float, float, float, float> r(kExecuteCompiled);
+  byte a = 0;
+  byte b = 1;
+  byte expected1 = 2;
+  byte expected2 = 3;
+  byte simd0 = r.AllocateLocal(kWasmS128);
+  byte simd1 = r.AllocateLocal(kWasmS128);
+  BUILD(
+      r, WASM_SET_LOCAL(simd0, WASM_SIMD_F32x4_SPLAT(WASM_GET_LOCAL(a))),
+      WASM_SET_LOCAL(simd1, WASM_SIMD_F32x4_SPLAT(WASM_GET_LOCAL(b))),
+      WASM_SET_LOCAL(simd1, WASM_SIMD_BINOP(simd_op, WASM_GET_LOCAL(simd0),
+                                            WASM_GET_LOCAL(simd1))),
+      WASM_SIMD_CHECK_F32x4(simd1, expected1, expected1, expected2, expected2),
+      WASM_RETURN1(WASM_ONE));
+
+  FOR_FLOAT32_INPUTS(i) {
+    if (SkipFPValue(*i)) continue;
+    FOR_FLOAT32_INPUTS(j) {
+      if (SkipFPValue(*j)) continue;
+      float expected1 = expected_op(*i, *i);
+      float expected2 = expected_op(*j, *j);
+      if (SkipFPExpectedValue(expected1) || SkipFPExpectedValue(expected2))
+        continue;
+      CHECK_EQ(1, r.Call(*i, *j, expected1, expected2));
+    }
+  }
+}
+
+WASM_EXEC_COMPILED_TEST(F32x4AddHoriz) {
+  RunF32x4HorizOpTest(kExprF32x4AddHoriz, Add);
+}
+#endif  // V8_TARGET_ARCH_ARM
+
 #if V8_TARGET_ARCH_ARM || SIMD_LOWERING_TARGET || V8_TARGET_ARCH_MIPS || \
     V8_TARGET_ARCH_MIPS64
 void RunF32x4CompareOpTest(WasmOpcode simd_op, FloatCompareOp expected_op) {
   FLAG_wasm_simd_prototype = true;
   WasmRunner<int32_t, float, float, int32_t> r(kExecuteCompiled);
   byte a = 0;
   byte b = 1;
   byte expected = 2;
   byte simd0 = r.AllocateLocal(kWasmS128);
   byte simd1 = r.AllocateLocal(kWasmS128);
@@ skipping 435 matching lines @@
 }
 
 WASM_EXEC_COMPILED_TEST(I32x4MinU) {
   RunI32x4BinOpTest(kExprI32x4MinU, UnsignedMinimum);
 }
 
 WASM_EXEC_COMPILED_TEST(I32x4MaxU) {
   RunI32x4BinOpTest(kExprI32x4MaxU, UnsignedMaximum);
 }
 
+#if V8_TARGET_ARCH_ARM
+void RunI32x4HorizOpTest(WasmOpcode simd_op, Int32BinOp expected_op) {
+  FLAG_wasm_simd_prototype = true;
+  WasmRunner<int32_t, int32_t, int32_t, int32_t, int32_t> r(kExecuteCompiled);
+  byte a = 0;
+  byte b = 1;
+  byte expected1 = 2;
+  byte expected2 = 3;
+  byte simd0 = r.AllocateLocal(kWasmS128);
+  byte simd1 = r.AllocateLocal(kWasmS128);
+  BUILD(r, WASM_SET_LOCAL(simd0, WASM_SIMD_I32x4_SPLAT(WASM_GET_LOCAL(a))),
+        WASM_SET_LOCAL(simd1, WASM_SIMD_I32x4_SPLAT(WASM_GET_LOCAL(b))),
+        WASM_SET_LOCAL(simd1, WASM_SIMD_BINOP(simd_op, WASM_GET_LOCAL(simd0),
+                                              WASM_GET_LOCAL(simd1))),
+        WASM_SIMD_CHECK4(I32x4, simd1, I32, expected1, expected1, expected2,
+                         expected2),
+        WASM_RETURN1(WASM_ONE));
+
+  FOR_INT32_INPUTS(i) {
+    FOR_INT32_INPUTS(j) {
+      int32_t expected1 = expected_op(*i, *i);
+      int32_t expected2 = expected_op(*j, *j);
+      CHECK_EQ(1, r.Call(*i, *j, expected1, expected2));
+    }
+  }
+}
+
+WASM_EXEC_COMPILED_TEST(I32x4AddHoriz) {
+  RunI32x4HorizOpTest(kExprI32x4AddHoriz, Add);
+}
+#endif  // V8_TARGET_ARCH_ARM
+
 void RunI32x4CompareOpTest(WasmOpcode simd_op, Int32CompareOp expected_op) {
   FLAG_wasm_simd_prototype = true;
   WasmRunner<int32_t, int32_t, int32_t, int32_t> r(kExecuteCompiled);
   byte a = 0;
   byte b = 1;
   byte expected = 2;
   byte simd0 = r.AllocateLocal(kWasmS128);
   byte simd1 = r.AllocateLocal(kWasmS128);
   BUILD(r, WASM_SET_LOCAL(simd0, WASM_SIMD_I32x4_SPLAT(WASM_GET_LOCAL(a))),
         WASM_SET_LOCAL(simd1, WASM_SIMD_I32x4_SPLAT(WASM_GET_LOCAL(b))),
@@ skipping 203 matching lines @@
 }
 
 WASM_EXEC_COMPILED_TEST(I16x8MinU) {
   RunI16x8BinOpTest(kExprI16x8MinU, UnsignedMinimum);
 }
 
 WASM_EXEC_COMPILED_TEST(I16x8MaxU) {
   RunI16x8BinOpTest(kExprI16x8MaxU, UnsignedMaximum);
 }
 
+#if V8_TARGET_ARCH_ARM
+void RunI16x8HorizOpTest(WasmOpcode simd_op, Int16BinOp expected_op) {
+  FLAG_wasm_simd_prototype = true;
+  WasmRunner<int32_t, int32_t, int32_t, int32_t, int32_t> r(kExecuteCompiled);
+  byte a = 0;
+  byte b = 1;
+  byte expected1 = 2;
+  byte expected2 = 3;
+  byte simd0 = r.AllocateLocal(kWasmS128);
+  byte simd1 = r.AllocateLocal(kWasmS128);
+  BUILD(r, WASM_SET_LOCAL(simd0, WASM_SIMD_I16x8_SPLAT(WASM_GET_LOCAL(a))),
+        WASM_SET_LOCAL(simd1, WASM_SIMD_I16x8_SPLAT(WASM_GET_LOCAL(b))),
+        WASM_SET_LOCAL(simd1, WASM_SIMD_BINOP(simd_op, WASM_GET_LOCAL(simd0),
+                                              WASM_GET_LOCAL(simd1))),
+        WASM_SIMD_CHECK8(I16x8, simd1, I32, expected1, expected1, expected1,
+                         expected1, expected2, expected2, expected2, expected2),
+        WASM_RETURN1(WASM_ONE));
+
+  FOR_INT16_INPUTS(i) {
+    FOR_INT16_INPUTS(j) {
+      int32_t expected1 = expected_op(*i, *i);
+      int32_t expected2 = expected_op(*j, *j);
+      CHECK_EQ(1, r.Call(*i, *j, expected1, expected2));
+    }
+  }
+}
+
+WASM_EXEC_COMPILED_TEST(I16x8AddHoriz) {
+  RunI16x8HorizOpTest(kExprI16x8AddHoriz, Add);
+}
+#endif  // V8_TARGET_ARCH_ARM
+
 void RunI16x8CompareOpTest(WasmOpcode simd_op, Int16CompareOp expected_op) {
   FLAG_wasm_simd_prototype = true;
   WasmRunner<int32_t, int32_t, int32_t, int32_t> r(kExecuteCompiled);
   byte a = 0;
   byte b = 1;
   byte expected = 2;
   byte simd0 = r.AllocateLocal(kWasmS128);
   byte simd1 = r.AllocateLocal(kWasmS128);
   BUILD(r, WASM_SET_LOCAL(simd0, WASM_SIMD_I16x8_SPLAT(WASM_GET_LOCAL(a))),
         WASM_SET_LOCAL(simd1, WASM_SIMD_I16x8_SPLAT(WASM_GET_LOCAL(b))),
@@ skipping 173 matching lines @@
 }
 
 WASM_EXEC_COMPILED_TEST(I8x16MinU) {
   RunI8x16BinOpTest(kExprI8x16MinU, UnsignedMinimum);
 }
 
 WASM_EXEC_COMPILED_TEST(I8x16MaxU) {
   RunI8x16BinOpTest(kExprI8x16MaxU, UnsignedMaximum);
 }
 
+#if V8_TARGET_ARCH_ARM
+void RunI8x16HorizOpTest(WasmOpcode simd_op, Int8BinOp expected_op) {
+  FLAG_wasm_simd_prototype = true;
+  WasmRunner<int32_t, int32_t, int32_t, int32_t, int32_t> r(kExecuteCompiled);
+  byte a = 0;
+  byte b = 1;
+  byte expected1 = 2;
+  byte expected2 = 3;
+  byte simd0 = r.AllocateLocal(kWasmS128);
+  byte simd1 = r.AllocateLocal(kWasmS128);
+  BUILD(r, WASM_SET_LOCAL(simd0, WASM_SIMD_I8x16_SPLAT(WASM_GET_LOCAL(a))),
+        WASM_SET_LOCAL(simd1, WASM_SIMD_I8x16_SPLAT(WASM_GET_LOCAL(b))),
+        WASM_SET_LOCAL(simd1, WASM_SIMD_BINOP(simd_op, WASM_GET_LOCAL(simd0),
+                                              WASM_GET_LOCAL(simd1))),
+        WASM_SIMD_CHECK16(I8x16, simd1, I32, expected1, expected1, expected1,
+                          expected1, expected1, expected1, expected1, expected1,
+                          expected2, expected2, expected2, expected2, expected2,
+                          expected2, expected2, expected2),
+        WASM_RETURN1(WASM_ONE));
+
+  FOR_INT8_INPUTS(i) {
+    FOR_INT8_INPUTS(j) {
+      int32_t expected1 = expected_op(*i, *i);
+      int32_t expected2 = expected_op(*j, *j);
+      CHECK_EQ(1, r.Call(*i, *j, expected1, expected2));
+    }
+  }
+}
+
+WASM_EXEC_COMPILED_TEST(I8x16AddHoriz) {
+  RunI8x16HorizOpTest(kExprI8x16AddHoriz, Add);
+}
+#endif  // V8_TARGET_ARCH_ARM
+
 void RunI8x16CompareOpTest(WasmOpcode simd_op, Int8CompareOp expected_op) {
   FLAG_wasm_simd_prototype = true;
   WasmRunner<int32_t, int32_t, int32_t, int32_t> r(kExecuteCompiled);
   byte a = 0;
   byte b = 1;
   byte expected = 2;
   byte simd0 = r.AllocateLocal(kWasmS128);
   byte simd1 = r.AllocateLocal(kWasmS128);
   BUILD(r, WASM_SET_LOCAL(simd0, WASM_SIMD_I8x16_SPLAT(WASM_GET_LOCAL(a))),
         WASM_SET_LOCAL(simd1, WASM_SIMD_I8x16_SPLAT(WASM_GET_LOCAL(b))),
@@ skipping 751 matching lines @@
         WASM_SIMD_I32x4_EXTRACT_LANE(
             0, WASM_LOAD_MEM(MachineType::Simd128(), WASM_ZERO)));
 
   FOR_INT32_INPUTS(i) {
     int32_t expected = *i;
     r.module().WriteMemory(&memory[0], expected);
     CHECK_EQ(expected, r.Call());
   }
 }
 #endif  // V8_TARGET_ARCH_ARM || SIMD_LOWERING_TARGET